I

Division of Clinical Pharmacology Department of Medicine and Health Sciences

Linköping University Sweden

Drug interaction surveillance

using individual case safety reports

Johanna Strandell

Linköping 2011

II

© Johanna Strandell, 2011

Division of Clinical Pharmacology Department of Medicine and Health Sciences Faculty of Health Sciences Linköping University SE-581 85 Linköping Sweden ISBN: 978-91-7393-106-9 ISSN: 0345-0082 Linköping University Medical Dissertations No. 1252

Printed in Sweden by Liu-Tryck, Linköping, 2011

III

Abstract

Background: Drug interactions resulting in adverse drug reactions (ADRs) represent a major health problem both for individuals and society in general. Post-marketing pharmacovigilance reporting databases with compiled individual case safety reports (ICSRs) have been shown to be particularly useful in the detection of novel drug - ADR combinations, though these reports have not been fully used to detect adverse drug interactions.

Aim: To explore the potential to identify drug interactions using ICSRs and to develop a method to facilitate the detection of adverse drug interaction signals in the WHO Global ICSR Database, VigiBase.

Methods: All six studies included in this thesis are based on ICSRs available in VigiBase. Two studies aimed to characterise drug interactions reported in VigiBase. In the first study we examined if contraindicated drug combinations (given in a reference source of drug interactions) were reported on the individual reports in the database, and in the second study we examined the scientific literature for interaction mechanisms for drug combinations most frequently co-reported as interacting in VigiBase. Two studies were case series analyses where the individual reports were manually reviewed. The two remaining studies aimed to develop a method to facilitate detection of novel adverse drug interactions in VigiBase. One examined what information (referred to as indicators) was reported on ICSRs in VigiBase before the interactions became listed in the literature. In the second methodological study, logistic regression was used to set the relative weights of the indicators to form triage algorithms. Three algorithms (one completely data driven, one semi-automated and one based on clinical knowledge) based on pharmacological and reported clinical information and the relative reporting rate of an ADR with a drug combination were developed. The algorithms were then evaluated against a set of 100 randomly selected case series with potential adverse drug interactions. The algorithm’s performances were then evaluated among DDAs with high coefficients.

Results: Drug interactions classified as contraindicated are reported on the individual reports in VigiBase, although they are not necessarily recognised as interactions when reported. The majority (113/123) of drug combinations suspected for being responsible for an ADR were established drug interactions in the literature. Of the 113 drug interactions 46 (41%) were identified as purely pharmacodynamic; 28 (25%) as pharmacokinetic; 18 (16%)

were a mix of both types and for 21 (19%) the mechanism have not yet been identified. Suspicions of a drug interaction explicitly noted by the reporter are much more common for known adverse drug interactions than for drugs not known to interact. The clinical evaluation of the triage algorithms showed that 20 were already known in the literature, 30 were classified as signals and 50 as not signals. The performance of the semi-automated and the clinical algorithm were comparable. In the end the clinical algorithm was chosen. At a relevant level, 38% were of the adverse drug interactions were already known in the literature and of the remaining 80% were classified as signals for this algorithm.

Conclusions: This thesis demonstrated that drug interactions can be identified in large post-marketing pharmacovigilance reporting databases. As both pharmacokinetic and pharmacodynamic interactions were reported on ICSRs the surveillance system should aim to detect both. The proposed triage algorithm had a high performance in comparison to the disproportionality measure alone.

Key words: adverse drug reactions, adverse drug interaction surveillance, drug interactions, individual case safety reports, postmarketing pharmacovigilance, signal detection

IV

V

Populärvetenskaplig sammanfattning

Läkemedelsinteraktioner som resulterar i biverkningar är ett betydande hälsoproblem för såväl enskilda individer som för samhället i stort. Misstänkta läkemedelsinteraktioner nämns sällan i enskilda biverkningsrapporter, vilket medför att oönskade interaktioner kan vara svåra att upptäcka. Samtidigt är det sedan länge känt att enskilda biverkningsrapporter med fördel kan användas för att identifiera signaler om nya läkemedelsbiverkningar.

Syftet med denna avhandling är dels att undersöka möjligheten att identifiera läkemedelsinteraktioner i WHO:s globala biverkningsdatabas, VigiBase, dels att utveckla en metod som underlättar upptäckten av okända läkemedelsinteraktioner.

De studier som ingår i denna avhandling är samtliga baserade på biverkningsrapporter som förekommer i databasen VigiBase. Studierna undersöker bland annat huruvida läkemedelsinteraktioner omnämns i de enskilda rapporterna samt hur identifieringen av nya oönskade läkemedelsinteraktioner i VigiBase kan underlättas.

Resultatet från denna avhandling visar att läkemedelsinteraktioner klassificerade som kontraindicerade, förekommer i rapporterna i VigiBase även om de inte alltid anges som interagerande. Farmakodynamiska och farmakokinetiska mekanismer är involverade i de läkemedelskombinationer som misstänks vara orsaken till att en läkemedelsbiverkning uppkommer. Baserat på dessa resultat så utvecklades en strategi för att hitta nya okända läkemedelsinteraktioner i VigiBase.

Denna avhandling visar på möjligheten att identifiera okända interaktioner med hjälp av biverkningsrapporter. Eftersom sjukvården rapporterar biverkningar som ett resultat av både farmakokinetiska och farmakodynamiska interaktioner bör övervakningssystemet sträva efter att upptäcka båda dessa typer. Den presenterade strategin är mer effektiv än nuvarande metoder, vilket är lovande i arbetet med att identifiera problematiska läkemedelskombinationer.

Nyckelord: biverkningar, monitorering av läkemedelsinteraktioner, läkemedelsinteraktioner, individuella biverkningsrapporter, farmakovigilans, signaldetektion

VI

VII

List of Papers

This thesis is based on the following papers that will be referred to according to their Roman numerals:

I. Strandell J, Bate A, Lindquist M, Edwards IR; Swedish, Finnish, Interaction X-referencing Drug-drug Interaction Database (SFINX Group). Drug-drug interactions - a preventable patient safety issue? Br J Clin Pharmacol 2008; 65(1):144-6.

II. Strandell J, Bate A, Hägg S, Edwards IR. Rhabdomyolysis a result of azithromycin and statins: an unrecognized interaction. Br J Clin Pharmacol 2009;68(3):427-34.

III. Strandell J, Wahlin S. Pharmacodynamic and pharmacokinetic drug interactions reported to VigiBase, the WHO Global Individual Case Safety Report Database. Eur J Clin Pharmacol 2011;67(6):633-41.

IV. Strandell J, Caster O, Bate A, Norén GN, Edwards IR. Reporting patterns indicative of adverse drug interactions – a systematic evaluation. Drug Saf 2011;34(3):253-266.

V. Strandell J, Norén GN, Hägg S. Key Elements in Adverse Drug Interaction Safety Signals. Submitted to Drug Safety.

VI. Strandell J, Caster O, Hopstadius J, Edwards IR, Norén GN. Triage algorithms for early discovery of adverse drug interactions. Submitted to Drug Safety.

The published papers are reprinted with permission of the copyright holders.

VIII

IX

Abbreviations and Acronyms

ADR Adverse Drug Reaction AERS US Food and Drug Administration’s Adverse Event Reporting System ASA Acetylsalicylic acid ATC Anatomical Therapeutic Chemical Classification CYP Cytochrome P450 DDA Drug-Drug-Adverse Drug Reaction DIPS Drug interaction probability scale HCP Health Care Professional IC Information Component ICSR Individual Case Safety Report INR International Normalized Ratio LHR Longitudinal health care records MAH Marketing Authorisation Holder MAO Monoamine oxidase MedDRA Medical Dictionary for Regulatory Activities MPA Swedish Medical Product Agency NSAID Non-Steroidal Anti-Inflammatory Drug OTC Over the counter PD Pharmacodynamic PK Pharmacokinetic PMS Post-Marketing Surveillance RCT Randomised Clinical Trial SPC Summary of Product Characteristics SWEDIS Swedish Drug Information System UMC Uppsala Monitoring Centre US FDA US Food and Drug Administration VigiBase WHO Global Individual Case Safety Report Database WHO World Health Organisation WHO-ART WHO Adverse Reaction Terminology WHO DDE WHO Drug Dictionary Enhanced Ω Omega, a three-way measure of disproportionality

X

XI

Definitions and terminology

ADR (WHO) A response to a drug which is noxious and unintended, and which occurs at doses normally used in man for the prophylaxis, diagnosis, or therapy of disease, or for the modification of physiological function.[1]

ADR(Edwards and Aronson) An appreciably harmful or unpleasant reaction, resulting from

an intervention related to the use of a medicinal product, which predicts hazard from future administration and warrants prevention or specific treatment, or alteration of the dosage regimen, or withdrawal of the product.[2]

Drug Any substance or combination of substances presented as

having properties for treating or preventing disease in human beings; or any substance or combination of substances which may be used in or administered to human beings either with a view to restoring, correcting or modifying physiological functions by exerting a pharmacological, immunological or metabolic action, or to making a medical diagnosis.[3]

Drug Interaction The effects of one drug are changed in the presence of

another drug, herbal medicine, food, drink or by some environmental chemical agent.[4]

Pharmacovigilance The science and activities relating to the detection,

assessment, understanding and prevention of adverse effects or any other drug-related problems.[5]

Signal (WHO) Reported information on a possible causal relationship between an adverse event and a drug, the relationship being unknown, or incompletely documented previously. Note: A signal is an evaluated combination which is considered important to investigate further. A signal may refer to new information on an already known combination. Usually more than a single report is required to generate a signal, depending upon the seriousness of the event and the quality of the information.[1]

Signal (CIOMS VIII) Information that arises from one or multiple sources (including observations and experiments), which suggests a new potentially causal association, or a new aspect of a known association, between an intervention and an event or a set of related events, either adverse or beneficial, that is judged to be of sufficient likelihood to justify verificatory action.[6]

XII

XIII

Contents

BACKGROUND ............................................................................................................................... 1

DRUG SAFETY ............................................................................................................................................................. 2 Historic background ............................................................................................................................................. 2 Adverse drug reactions ........................................................................................................................................ 3 Pharmacovigilance ................................................................................................................................................ 4 Individual case safety reports .............................................................................................................................. 5 WHO Programme for International Drug Monitoring ................................................................................. 6 Signal detection ..................................................................................................................................................... 7

DRUG INTERACTIONS ................................................................................................................................................ 8 Historic background ............................................................................................................................................. 9 Pharmacology ........................................................................................................................................................ 9 Epidemiology ....................................................................................................................................................... 10 Drug interaction surveillance ............................................................................................................................ 12 Drug interaction surveillance using individual case safety reports ............................................................. 13

AIMS OF THIS THESIS ................................................................................................................. 15

MATERIAL AND METHODS ....................................................................................................... 16

DATA SOURCES ......................................................................................................................................................... 17 VigiBase ................................................................................................................................................................ 17 Reference sources of drug interactions ........................................................................................................... 21

METHODS................................................................................................................................................................... 23 Causality assessment method ............................................................................................................................ 23 Statistical methods .............................................................................................................................................. 23

RESULTS ....................................................................................................................................... 25

CHARACTERISTICS OF SUSPECTED/POTENTIAL DRUG INTERACTIONS .......................................................... 26 SIGNAL DETECTION AND INFORMATION STRENGTHENING CAUSALITY ....................................................... 27 REPORTING RATE OF ADVERSE DRUG INTERACTIONS ...................................................................................... 29 INDICATORS OF ADVERSE DRUG INTERACTIONS ............................................................................................... 30 TRIAGE ALGORITHMS FOR DETECTION OF ADVERSE DRUG INTERACTIONS ................................................ 31

DISCUSSION ................................................................................................................................. 33

CONCLUSIONS ............................................................................................................................ 37

FUTURE PERSPECTIVES ............................................................................................................................................ 38

ACKNOWLEDGEMENTS ............................................................................................................ 39

REFERENCES .............................................................................................................................. 40

XIV

1

Background

Drugs have been used successfully to treat and prevent illnesses for long time and have revolutionised health care. Today drug treatment is the most important intervention for curing diseases and maintaining mankind’s well being. The number of drugs used per individual has successively increased over time. For many individuals the numerous drugs used are necessary, with undisputable benefits, though for some patients multiple drug therapies (polypharmacy) are a result of irrational and excessive drug use. No drug is absolutely free from harmful effects and polypharmacy increases the risk of reactions related to drug use, adverse drug reactions (ADRs), and ADRs as consequence of drug-drug interactions (adverse drug interactions).[7-10]

This thesis focuses on early detection and surveillance of drug interactions in post-marketing pharmacovigilance reporting databases. For the early detection of novel ADRs related to single drugs in large post-marketing pharmacovigilance reporting databases computerised screening including measures of disproportionality and selection strategies have been recognised as being essential.[11-13] However, for drug interactions there are no systems in place currently (at least not published) that apply computerised methods that incorporates other information than measures of disproportionality, to detect adverse drug interaction signals.

2

Drug Safety

Historic background

Over the past centuries many drug related problems have been discovered and some have changed our view on drugs. Two major drug disasters, sulfanliamide and thalidomide (described in more detail below), have played a key role for the awareness of ADRs as a real threat and have had a major impact on social guidelines and drug regulation. All drug related incidents have contributed to the development and processes of drug surveillance, though a more recent example, cerivastatin, is described to show that there is still need for development of pharmacovigilance world-wide, including surveillance of drug interactions.

Sulfanliamide

The liquid form of sulfanliamide (a sulfonamide indicated for streptococcal infections) entered the drug market in 1937 in United States (US).[14] The drug had previously been distributed in tablet and powder form without serious reactions. The liquid formulation of sulfanilamide was reported to have caused deaths in more than 100 people in United States in 1937. The reason for problems with the liquid form and not tablets were that sulfanilamide was dissolved in diethylene glycol (an antifreeze agent used in windscreen washer fluid) which is deadly poison. The disaster led to the passage of the 1938 Food, Drug, and Cosmetic Act, which dramatically increased the US Food and Drug Administration's (FDAs) authority to regulate drugs.

Thalidomide

In 1961 it was reported by William McBride that women who ingested thalidomide, a non-barbiturate hypnotic agent indicated for nausea (morning sickness), gave birth to children with skeletal malformations (phocomelia).[15] More than 10 000 children worldwide (46 countries) have been affected by this very rare form of limb reduction. Thalidomide was withdrawn from the global market during the early 1960s. Following the thalidomide disaster the need of closer drug monitoring to detect novel adverse reactions was recognised and the incident led to the systematic collection of suspected reports of adverse drug reactions on national and global level. It also led to the development of other registers of birth defects for example.

Cerivastatin

In August 2001, cerivastatin, a HMG-CoA-reductase inhibitor (referred to as statin), was voluntarily withdrawn by the market authorisation holder (MAH) from the global market because of fatal reports of rhabdomyolysis.[16] The risk of rhabdomyolysis, a rare and dose dependent reaction was therefore higher in patients using high dosages (0.8 mg/day), or other drugs that potentially could increase the plasma concentration of cerivastatin such as gemfibrozil. Even though the drug combination was contraindicated,[17] 12 out of 31 deceased cerivastatin patients in United States had received gemfibrozil concurrently.[16]

3

Adverse drug reactions

An ADR is defined as “a response to a drug which is noxious and unintended, and which occurs at doses normally used in man for the prophylaxis, diagnosis, or therapy of disease, or for the modification of physiological function” by the World Health Organisation (WHO).[1] This definition has been widely used, although it does not include errors in drug use or intoxications, or reactions related to contaminants or inactive ingredients. Therefore, Edwards and Aronson definition of an adverse reaction is used in this thesis: “an appreciably harmful or unpleasant reaction, resulting from an intervention related to the use of a medicinal product, which predicts hazard from future administration and warrants prevention or specific treatment, or alteration of the dosage regimen, or withdrawal of the product”.[2] Furthermore, adverse reaction is synonymously used with ADR in this thesis.

Incidence of adverse drug reactions

ADRs have been reported to account for 2.4% - 13.8% of all hospital admissions[18-22] and been described as the fourth to seventh leading cause of death in Sweden and the United States.[18,23] Drug-related hospital admissions are reported to lead to a mean length of hospital stay of 6-13 days,[19,21-22] which is longer than for typical medical admission[19,22] and therefore also more expensive.[22] In Germany, drug-related hospital admissions were estimated to cost an average of 3700 Euros per stay with annual costs of 400 million Euros.[22] A single drug-related hospital admission was calculated to cost 2200 Euros in Sweden in 2002.[19]

Risk factors of adverse drug reactions

Factors that may influence the risk of experiencing an ADRs are drug dosage, drug formulation, pharmacological properties of the drug, phenotype of the user affecting the pharmacokinetics and pharmacodynamics of the drug, use of multiple drugs, and drug-drug interactions.[24] Furthermore, females have been reported to experience more ADRs than males.[19,21,25] Specific groups such as elderly, elderly with cognitive impairment and individuals with specific diseases such as renal failure are also more likely to experience ADRs.[24,26]

Type of adverse drug reactions

ADRs are commonly divided into type A and type B reactions.[2,24] Type A reactions are characterised as being an augmented pharmacologic effect of the drug. These effects are dose dependent and common (representing approximately 80% of all ADRs).[18,22] Type A reactions are in theory preventable, as they can be predicted from the pharmacological properties of the drug. It has been suggested that 18-73% of these are preventable.[20,27-29] Type B reactions are unexpected as they are not related to the pharmacological properties of the drug. Type B reactions are often serious, occurring in a minority of patients and are often allergic or idiosyncratic reactions.[30] In addition to type A and B reactions, ADRs have been further categorised as C (chronic) that are relatively uncommon and related to the cumulative dose (for example analgesic nephropathy), D (delayed reactions) are

4

uncommon and usually dose-related, occurring after some time of usage (for example teratogenous effects), E (end of use) occur shortly after the drug is withdrawn (for example opiate withdrawal syndrome), F (unexpected failure of therapy) reactions, are common, dose-related and often caused by drug interactions (for example inadequate effect of oral contraceptives during concurrent use of enzyme inducers).[2]

Therapeutic ineffectiveness was not included in WHOs definitions of an ADR,[1-2] although the lack of effect is reported as one of the most common drug related problems.[31] Therapeutic ineffectiveness of a medicinal product may be the result of pharmaceutical defects such as substandard and counterfeit drugs, resistance, inappropriate use, tolerance or drug interactions.[32]

Adverse drug reactions and drugs involved in drug related admissions

Among the most commonly observed ADRs for patients with drug related admissions are gastrointestinal complications including gastrointestinal bleedings, central nervous system complications, cardiovascular disorders and hemorrhages.[19,21] Non-steroidal anti-inflammatory drugs (NSAIDs), antithrombotic drugs, sedatives, cardiovascular agents including cardiac stimulants and antiarrhythmics are some of the pharmacological classes responsible for these drug-related admissions.[19,21,27]

Pharmacovigilance

Pharmacovigilance is usually described as ‘the science and activities relating to the

detection, assessment, understanding and prevention of adverse effects or any other

drug-related problem’.[5] This scope involves both pre and post marketing activities with the primary endpoint to improve patient care and safety.

Before marketing a drug general pharmacology, efficacy and safety are tested. Pre-marketing studies include animal experiments (studying acute toxicity, dose dependence, carcinogenicity and mutagenicity/teratogenicity), and three phases of clinical testing on humans. Phase I is based on a small group of healthy volunteers to gather preliminary data, whereas phase II includes patients to study efficacy, dosage recommendations, and collect early safety data. In phase III, most often undertaken as Randomised Clinical Trials (RCT), a group of patients are randomly assigned to the drug of interest, or to placebo or a comparator. Because of the limited and restricted study populations exhibited for a short study period in RCTs, only common ADRs and ADRs occurring in the recent time frame are detected. ADRs occurring more rarely, after a long time, or in populations previously excluded in RCTs (such as children, elderly, pregnant women or patients with co-morbid conditions) will therefore not be known at the time of marketing. Since the drug’s usage may evolve during the drug’s life-time, post marketing studies (experimental studies (continuous RCTs) or non-experimental pharmacoepidemiological studies, case reports (one patient), case series (a collection of patients); case control studies; cohort studies and meta-analyses) will be essential to detect drug related problems including new ADRs, frequency of ADRs, and to identify risk factors for developing ADRs for drugs after approval.

5

Individual case safety reports

Following the thalidomide disaster world-wide national systems (referred post-marketing pharmacovigilance reporting databases in this thesis) were set up to collect reports of suspected ADRs. These reports are referred to as spontaneous reports or individual case safety reports (ICSRs), and synonymously referred to as reports in this thesis.

ICSRs represent alerts of clinical concerns of drug related problems, occurring in individuals in the real-world use of drugs. These reports have large population coverage including patients with certain pre-dispositions, and patient groups that are excluded from clinical trials such as pregnant women or children. ICSRs have been described as a cornerstone in the early detection of the previously novel ADRs, post-marketing[33] and they are particularly useful in the detection of rare ADRs. Furthermore, the post-marketing pharmacovigilance reporting system is an inexpensive source of ADR information and is simple to manage.

National databases

During 1961-1965 the first countries Australia, Italy, Netherlands, New Zealand, Sweden, United Kingdom and USA, started to systematically collect national reports including suspected ADRs.[34] The reports are assessed, and stored in the individual countries databases. The collection, assessment and regulatory action are often maintained by the regulatory authority within that country. However in some countries such as Netherlands is data collection separate from regulation. The national reports and systems vary between countries in other aspects. For instance the reporting requirements vary for individual countries for example in terms of who is allowed to/should report (reporter). Some countries only allow reports from physicians while others allow reports from all types of health care professionals (physicians, pharmacists, nurses, and dentists) and/or reports from non health care professionals such as consumers and lawyers. The nature of reports in a national database may also vary if the reports are submitted via pharmaceutical companies, or directly reported to the authority. There are also variations between national databases in terms of drugs that are included in these collections. Some countries include all drugs (traditional drugs, herbals and vaccines) in the same database, while others separate vaccines, traditional drugs and herbals. Furthermore, some of the national centres focus the analyses on aggregated data, while countries perform manual causality assessments (assessments examining the relationship between a drug and an adverse reaction) of all reports received. There is a range of operational causality algorithms (Naranjo’s Scale,[35] WHO causality criteria,[36] French algorithm[37]) available which results in variations of causality outcome. There are also variations between national databases in terms of the level of suspicion that the drug caused the ADR. The majority of databases include reports concerning ADRs where the drug is believed to have caused the unpleasant reaction. While some national databases include reports of adverse events[38] which are an adverse outcome that occurs when a patient is taking a drug, though the adverse event is not necessarily related to drug use.[2]

6

International databases

There are three large international post-marketing pharmacovigilance reporting databases. The EudraVigilance (the European Medical Agency’s database) including 4 million reports (March 2011) with reports from countries within European Unions,[39] the US Food and Drug Administration’s Adverse Event Reporting System (AERS) including more than 4 million public reports (August 2011)[38] of which around one third are foreign reports submitted from companies around the world, and the WHO Global ICSR Database, VigiBase containing more than 6.6 million reports (August 2011) from 106 countries worldwide.[40] For country members of the European Union (EU) reporting to EudraVigilance is mandatory, while the reporting to VigiBase is not regulated by law for the countries which are members of the WHO Programme for International Monitoring (see section WHO Programme for International Monitoring).

When interpreting data from international databases one should consider that there might be national variations in terms of who is allowed to report, the type of reports, language (the national reports are usually provided in the official language/s of the country), causality outcome, as well as the level of suspicion that the drug caused the ADR (adverse reaction vs. adverse event). For the latter, reactions reported on ICSRs are referred to as ADRs for simplicity in this thesis.

Limitations

ICSRs and the system maintaining these reports have well recognised disadvantages. The dataset is often heterogeneous (for the reasons mentioned above) and these reports sometimes lack clinical information which can make the causality assessment difficult. Furthermore, underreporting together with increased reporting of known associations (sometimes referred to as reporting bias), are two fundamental challenges to effective ADR surveillance.[41-42] Another fundamental problem for collections of ADR reports is the presence of duplicate reports that may lead to inflated disproportionality measures[43] or distort the manual analysis. Because of the influence of different sources of biases ICSRs are primarily useful for hypothesis generation, in contrast to case-control, or cohort designs that are used to test hypothesis.

WHO Programme for International Drug Monitoring

In 1968, ten countries with developed national reporting systems from Europe, North America and Oceania1 agreed to compile their national data in an international database with the intention to detect rare and serious ADRs as early as possible in an international perspective.[44] This international collaboration was initiated by the WHO and later formed the WHO Programme for International Drug Monitoring. Since 1978 the WHO Collaborating Centre for International Drug Monitoring has been responsible for the Programme. The centre is localised in Uppsala, Sweden and operates under the name Uppsala Monitoring Centre (UMC). On behalf of the WHO

1 Australia, Canada, Czechoslovakia, Germany (the Federal Republic of Germany), Ireland, Netherlands, New Zealand, Sweden, United Sates of America and United Kingdom.

7

Programme UMC is responsible for maintaining and analysing the global database, WHO Global ICSR Database, VigiBase. As of August 2011, 106 countries were full members and forwarded reports to VigiBase. Figure 1 shows countries members of the WHO Programme for International Drug Monitoring (August 2011).

Figure 1. August 2011, WHO Programme for International Drug Monitoring member countries: full member

states (dark blue) and associated member countries (light blue). Associated member countries have not yet

successfully transmitted ICSRs to the international database and fulfilled other requirements.

Signal detection

The early discovery of ADRs for which the causality related to drug use has previously not been established is referred to as signal detection. It should be stressed that a signal in this thesis is a clinically evaluated association (drug-ADR, drug-drug or drug-drug-ADR), which is considered important, though it is preliminary and therefore needs further investigation. With the exception for study VI (used CIOMS definition)[6] WHO’s signal definition “reported information on a possible causal relationship between an adverse event and a drug, the relationship being unknown or incompletely documented previously”[1] has been used in this thesis.

Signal detection of single drug-ADRs in VigiBase

Data is collected in national as well as the international databases to detect drug related problems. For this purpose some national centres perform case- by- case analysis while others focus the analyses on aggregated data. The latter approach is

8

often done in large national databases and also in VigiBase. Because of the large amount of data more advanced methods to facilitate detection of ADRs is needed.

Historically all drug-ADR combinations reported during the past quarter were reviewed.[13] For obvious reasons this approach was not efficient, and to improve the systematic detection of new signals in VigiBase quantitative signal detection was implemented in 1998.[45] A measure of disproportionality referred to as the Information Component (IC)[11] is used in VigiBase. The measure indicates how frequently a single drug-ADR combination is reported in relation to the background of the dataset. The IC is based on the observed and the expected reporting of a drug-ADR pair. The IC measure reflects the relative reporting rate of the drug-ADR in the database and a positive IC value indicates that a particular drug-ADR pair is reported more often than expected, based on all the reports in the database. The higher the value of the IC, the more the combination stands out from the background. The IC025

is the lower limit of IC’s 95% credibility interval and IC025 used as the threshold in the routine screening of VigiBase.

After implementation of the IC measure, the systematic surveillance of single drug-ADR combinations was even further improved by introducing selection strategies (Triage algorithms) in 2001.[46] The current routine signal detection process in VigiBase involves systematic screening of drug-ADR combinations listed on at least one report entered into the database during the last quarter and having an IC025

above zero. After the initial screening are two Triage algorithms applied.[47] These triage algorithms contain pre-defined criteria that need to be fulfilled on the case series level. One of the algorithms examines drug-ADR combinations reported from at least two countries and involving new drugs (defined as drugs entered into WHO DD during the past five years), and serious terms (defined as critical terms according to WHO-ART). A second algorithm focuses on drug-ADR combinations reported from at least two countries and where the IC has increased with one unit (the ratio of the observed to the expected number of reports has been doubled) since the last quarter. After the systematic screening, literature sources[48-50] are reviewed to assess if the ADR has been described for the drug of interest. If not, the individual VigiBase reports are then examined for the relationship between a drug and an adverse reaction. If more information is required for a thorough case analysis, the original files (that generally provide more detailed information in the form of narrative text for the individual report) are requested from the national authority and reviewed. Topics that fall under the signal definition are then presented in the SIGNAL document, which is circulated to the national centres.

Drug Interactions

A general description of a drug - drug interaction is when “the effects of one drug

are changed in the presence of another drug, herbal medicine, food, drink or by some

environmental chemical agent”.[4] The effects of the drug combination may be:

synergistic or additive; antagonistic or reduced; or altered or idiosyncratic, and it may result in beneficial effects or adverse reactions.

9

This thesis focuses on drug interactions which have negative effects: adverse reactions or failure of the therapeutic effects in humans. Subsequently interactions as a result of pharmaceutical incompatibility are not covered. Furthermore, within the scope of this thesis a drug interaction refers to the combination of two drugs, and an adverse drug interaction is a drug-drug combination resulting in an ADR or therapeutic failure of either drug.

In the general description of a drug interaction is a drug interaction with additive pharmacodynamic effects excluded as the outcome of the two drugs is not more than a direct result of their individual effects.[4,51] However, the risk of an ADR during concurrent use of two drugs with additive effect may be synergistic. Since our intention is to detect drug combinations that are of particular concern in health care, are drug combinations with additive pharmacodynamic effects, but with synergistic risks of ADRs during concurrent use included in our concept of drug interactions and adverse drug interactions.

Historic background

The first reports on drug interactions in the literature concern the potential to enhance or reduce the drug/s effect,[52] though it was not until the 1960s that reports of clinically significant drug interactions began to appear in the literature. Among the first clinically important interactions discovered were hypertensive crises in patients who had taken concurrently certain cheeses and were using monoamine oxidase (MAO) inhibitors.[53-56] During the 1960s the first drug interaction tabulations appeared in journals and methods for systematically addressing drug interactions when dispensing or prescribing were developed, including checklists of factors to be asked, wall-chart systems, and monographs of drug interactions.[57] In 1970 the regulatory agencies began to require pharmaceutical companies to issue annual reviews of drug interactions in the national formularies.[56]

Pharmacology

Drug interaction mechanisms are categorised into two main groups, pharmacokinetic and pharmacodynamic, depending on the principles that determine the drug’ behaviour in the human body.[4,51,58]

Pharmacokinetic interactions include mechanisms where the absorption, distribution, metabolism or excretion of one drug is altered by a second drug, and results in changes in the drug concentration.[4,51,58] A large proportion of potentially clinically significant drug interactions are reported to occur by alterations in the drug metabolism through inhibition and induction of enzymes and drug transport proteins in the liver.[59] The outcome of changed metabolism depends on the drug, for instance inhibition of an active drug can lead to rises in the concentration to toxic levels, while for a pro-drug that is activated via the enzyme inhibition can lead to reduced efficacy.[60] Among the most important enzymes involved in the metabolism are cytochrome P450 enzymes (CYP). They are responsible for the metabolism in approximately 50% of drugs used clinically. CYP3A4 is by far the most abundant isoform accountable for most cytochrome P450-related metabolism of all marketed

10

drugs. Other isoforms often involved in drug metabolism are CYP1A2, CYP2C9, 2C19, 2D6 and CYP2E1. Inhibition of CYP enzymes is more common than induction.

Amongst other pharmacokinetic mechanisms are absorption which may be affected via changes in the gastrointestinal pH or motility, damage in the gastrointestinal tract, alterations in intestinal flora, and drug binding in the gastrointestinal tract; [4,51,60-61] distribution that may be changed via displacement of plasma proteins; and excretion that primarily occurs in the kidney[58] and involves changes in the urinary pH or renal blood flow including passive tubular re-absorption, glomerular filtration and active tubular secretion.[4]

Pharmacodynamic interactions include mechanisms where the effect of one drug is altered by a second drug at its site of action without changes in the drug concentration.[4,51,61] These interactions can result in antagonistic, synergistic or additive effects.

Time course

A drug interaction can occur within a couple minutes while others can take several weeks to develop.[58] Although the time course of drug interactions may be relatively consistent within a group of patients, there may be significant variations between individuals. There are numerous factors explaining variations in the time course of drug interactions e.g. half-life time, dosages, route of administration and whether the active substance is the metabolite or the parent drug. For instance long half-life time of the drug inducing the interaction means that it takes a longer period to reach steady-state; whereas high doses of the affected drug during administration of another drug that inhibits its elimination and intravenous administration result in a shorter time period before the upper end of the therapeutic range is reached. The time course varies also according to mechanism. For instance enzyme induction occurs gradually and it can take several days up to weeks for the affected drug to accumulate toxic levels.[4,58] In contrast, enzyme inhibition of CYP enzymes is rapid in development and occurs within 2-3 days. Enzyme inhibition also dissipates more rapidly than enzyme induction.[4,58] Interactions via renal excretion are similar to enzyme inhibition as they often fairly rapid when occurring and dissipating and the elimination is usually back to normal after 2-3 half-lives after the drug inducing the interaction is discontinued. Pharmacodynamic interactions have in general a rapid onset, though there are some more complex pharmacodynamic interactions which are developed during a longer period.[58]

Epidemiology

Incidence of potential drug interactions

In primary health care, 4% to 70% of the patients are exposed to potential drug interactions, of these are 1% to 26% considered clinically relevant.[10,62-68] However, these figures may be under-estimated since traditional drugs provided for self-medication and herbals also increase the risk of drug interactions, but are usually not included in dispensed prescriptions.[69]

11

Incidence of adverse drug interactions

The magnitude of ADRs related to drug interactions in the existing literature is inconsistent, reflecting the variety of settings, populations studied and the methodology used and definitions applied. For instance results from a post-marketing pharmacovigilance reporting database suggest that 22% of patients exposed to a potential drug interaction experienced an associated ADR,[70] while in specific hospital settings drug interactions have been reported to cause from 1-21% of all ADRs.[20,65]

Pharmacodynamic and pharmacokinetic adverse drug interactions

Few studies have examined drug interaction categories responsible for causing ADRs. In a review of drug related reactions occurring during hospital stay, the majority were pharmacodynamic (91.7%), pharmacokinetic (5.3%) and had both pharmacodynamic-pharmacokinetic mechanisms (3%).[71] In another small study investigating ADRs leading to hospital admissions, all drug interactions assessed as responsible for the ADR were pharmacocodynamic.[68]

Risk factors of adverse drug interactions

An increased number of dispensed drugs have been reported to increase the risk of experiencing adverse reactions related to drug interactions.[62] Other risk factors for drug interactions related to the drug involved are high doses, route of administration, long time drug therapies, drugs with self-induced or saturable metabolism, substances with identical or similar pharmacological profile and drugs with steep dose-response curves for which moderate changes in plasma concentration may lead to significant increases in the drug effect.[58,72] Furthermore substances with narrow therapeutic windows are more likely, in comparison to those with broad therapeutic window, to be involved in adverse drug interactions.[58] In addition, for many new drugs the risk of adverse drug interactions is increased as they have complex mechanisms of action and multiple effects.[72]

Patients with particular risks of experiencing an ADR as a result of a drug interaction are elderly (as they are often exposed to multiple drugs, have underlying diseases and impaired homeostatic systems), individuals with hepatic or renal disease, patients in intensive care (not only due to the number of drugs, but also because of impaired homeostatic mechanisms), patients who undergo complicated surgical procedures, transplant recipients, patients with more than one prescribing doctor. Furthermore, there is also a risk of overdoses and augmented toxic effects in patients whereas the CYP system is induced via genetically and/or environmental factors such as chronic use of alcohol or nicotine.[24,73]

Problematic drug interactions in clinical practice

Drugs that have been reported to be involved in potentially serious drug interactions are cardiovascular agents (including enalapril, digoxin, ramipril, furosemide and

12

spironolactone),[62,70] anti-inflammatoric drugs (acetylsalicylic acid and other NSAIDs (diclofenac, naproxen, ibuprofen)) and anticoagulants (such as warfarin).[25,62,74] Anticoagulants and antiplatelets have been reported as responsible for the greatest number of fatal and serious reactions.[70]

Drug interaction surveillance

It is well established that polypharmacy increases the risk of adverse reactions related to drug use, and ADRs as consequence of drug interactions. However, polypharmacy including potentially interacting drug combinations will not lead to an adverse outcome in every patient. The clinical impact of an adverse drug interaction on the population level does not only depend on the seriousness of harm, but also on the risk that the adverse drug interaction actually occurs, which is dependent on to what extent the two drugs are co-prescribed, the existence of risk factors, and the incidence of the adverse reaction. The risk of serious drug interactions is of concern for the patient, the physician, the regulatory authority and society at large. It is also a concern for the pharmaceutical companies marketing the drug since great economic value may be at stake. Table I shows some examples of drugs that have been withdrawn because of serious ADRs partly due to drug interactions. Since drug’s usage may evolve during the drug’s life-time, new indications and new patient groups will be exposed, post-marketing studies are essential in the process of identifying new potential drug interactions.

Clinical surveillance

Many drug interactions are predictable as they are related to the pharmacokinetic and pharmacodynamic effects of the drugs. To prevent unnecessary interactions information should be available for the prescriber so that he or she can take action to minimise the risk of adverse reactions or therapeutic failure by using an alternative drug, making dose adjustment, or to monitor the patient. However, one problem regarding information available for drug interactions today is the overload of information and it can be difficult to retrieve, sort and incorporate all available information in clinical decision making.[75] To improve the practitioners’ prescribing habits computerised decision systems that prompt the physicians when prescribing have been integrated in dispensing software, and these systems’ (when sending relevant alerts) have been reported as successful.[76]

13

Table I. Examples of drugs that have been withdrawn worldwide or in some part of the world

because of serious ADRs partly related to drug interactions

Year

approved

Year

withdrawn

Affected drug (ATC class*) Drug/s inducing

the interaction

ADRs

1984 1999 Astemizole (other antihistamines for systemic use)

CYP3A4 inhibitors Torsade de

pointes

1985 1997 Terfenadine (other

antihistamines for systemic

use)

CYP3A4 inhibitors Torsades de

Pointes and

cardiac

arrhythmias

1993 2000 Cisapride (propulsive) CYP3A4 inhibitors Torsade de

pointes

1997 2001 Cerivastatin (HMG CoA

reductase inhibitor)

Gemfibrozil and

other fibrates

Rhabdomyolysis

1997 1998 Mibefradil** (other calcium

channel blockers with mainly

vascular effects)

CYP3A substances Torsade de

pointes,

rhabdomyolysis

1997 2010 Sibutramine (centrally acting antiobesity product)

CYP3A4 inhibitors.

MAO inhibitors

and other centrally

active drugs.

Cardiovascular

effects and

strokes.

CNS active drugs

increase the risk

for serotonin

syndrome.

*Pharmacological subgroup

**Mibefradil (CYP 3A4 inhibitor) was withdrawn for its plausibility to affect other agents

Drug interaction surveillance using individual case safety reports

Even though post-marketing pharmacovigilance reporting system’s have been shown to be particularly useful in detection of novel drug - ADRs combinations,[33] ICSRs have not been fully used to detect adverse drug interactions. One explanation could be that it can be difficult to interpret whether an ADR arise from a single drug or a combination of two or more drugs in individual patients, and particularly in patients with underlying risk factors such as multiple drugs and other diseases.

Among published drug interaction signals the majority have been generated from regular case-by-case analysis performed by national or regional centres worldwide. For instance, the Swedish post-marketing pharmacovigilance reporting system has provided a range of signals about drug interactions with St John’s wort (hypericum perforatum), between warfarin and tramadol, between warfarin and noscapine, and between warfarin and tetracyclines.[77-81] The Swedish examples show that post-marketing pharmacovigilance reporting system’s can be a valuable source to detect clinically relevant drug interactions. However this approach is not feasible in a large database such as VigiBase where additional methods to facilitate the detection of potential adverse drug interactions are needed.

14

In the beginning of this project (2008) we found that a disproportionality measure Omega (Ω) (see section Statistical methods Omega (Ω)) used to detect adverse drug interactions in VigiBase[82] highlighted several false positive adverse drug interactions due to clusters of reports.[83] We also had examples of signals, found by manual analysis (for instance warfarin - noscapine and bleeding[80]), that were not highlighted with the measure of disproportionality. At that point the need for efficient algorithms for detection of adverse drug interactions were clear, thus such a method could not rely on Ω025 alone, as in the single drug – ADR surveillance in VigiBase (see section Signal detection of single drug-ADRs in VigiBase). We therefore we hypothesised that adverse drug interaction surveillance would be much more effective if clinical information and the disproportionate reporting for adverse drug interactions were combined.

15

Aims of this thesis

The overall purpose of this thesis was to explore the potential to identify drug interactions using ICSRs and to develop a method to facilitate the detection of adverse drug interactions signals in VigiBase. Specific objectives were:

I To establish in what form known drug interactions are identified on ICSRs and to determine whether these can give insight into the inappropriate co-prescribing of drugs.

II To examine if the case reports of the potential association of azithromycin - statins and rhabdomyolysis is suggestive of a drug interaction and how the reporting of this potential interaction has changed over time.

III To explore those drug combinations most frequently co-reported as interacting in VigiBase, and categorise them with respect to the drug interaction mechanisms.

IV To systematically examine a set of indicators’ (information supportive of drug interactions) propensity of highlighting suspected adverse drug interactions in the time period before the interaction became known in the literature.

V To identify what reported information may support the identification of a drug interaction safety signal, and to what extent this information is available in structured format.

VI To design triage algorithms for adverse drug interaction surveillance in VigiBase, and to evaluate the algorithms prospectively relative to clinical assessment.

16

Material and Methods

All studies in this thesis were based on ICSRs in VigiBase. Studies II and V also included original reports from each respective country. Studies I, IV and VI used information from drug interaction databases (Swedish, Finnish, INteraction X-referencing drug-drug interaction database (SFINX database)[75] and Stockley’s Drug Interactions[4].) Studies II, IV and VI applied a measure of three-way disproportionality referred to as Ω (Omega).[82] In study V an operational algorithm for causality assessment of drug interactions (Drug Interaction Probability Scale (DIPS))[84] was applied. The sources and methods used are described in more detail further down in this chapter.

Below are short summaries of data and methods used in each study (I-VI).

I An explorative study where all drug combinations classified as ‘established’ and ‘clinically important’ drug interactions (given as D4) in the SFINX database[75] were examined for their co-reporting on reports in Vigibase.

II A case series study where all reports in VigiBase, and the original files, including azithromycin co-reported with any statin and rhabdomyolysis were reviewed. The reporting over time in VigiBase was investigated by generating Ω values retrospectively for rhabdomyolysis with azithromycin and statins.

III A descriptive study where drug combinations co-reported as interacting in at least 20 reports in VigiBase during the past 20 years were examined. Each drug combination was reviewed in the literature to identify if the drug combination was known to interact and the mechanism of interaction. Report characteristics were also examined.

IV This study examined the reporting patterns for 322 known adverse drug interactions in the time period before the adverse drug interactions become known in the literature. The patterns for these known adverse drug interactions were compared to group of 6440 drug-drug-ADR (DDA) triplets where the drug combinations were not known to interact (the set of interacting and non interacting DDAs was referred to as the reference set). A reference set were created from information in Stockley’s Drug Interaction Alerts. VigiBase reports including known adverse drug interactions and non interacting drug combinations were screened for indicators of drug interactions (for example pharmacological properties such as common CYP metabolism, reported clinical information suggestive of a drug interaction and a positive Ω025 measure indicating an excessive reporting of the ADR and the drug combination) in the time period before the drug interaction became established in the literature. The results for known adverse drug interactions were compared to the results for non interacting drug combinations.

V The reports in VigiBase and original files referred to in three published drug interaction signals were assessed using the causality assessment

tool DIPS.[84] EVigiBase report and its corresponding original fileinformation was specified as being listed in the structured fields, free text and, in total.

VI The reference set generated in model for detection of novelstudy IV individual additional indicators tested previously.of the indicators completely data driven, one semiknowledge) were then evaluated against case series including potential adverse drug interactions. The algorithm’s performances were then evaluated bydrug-drug-ADR (DDA) triplets

Data sources

VigiBase

VigiBase is a vast resource 6.6 million ICSRs reported non-health professionals VigiBase varies greatly between two percent and in total 87%

Figure 2. The proportion of reports in VigiBase per country (including countries contributing with at least 2%)

as of July 2011.

Thailand; 2%

Sweden; 2%Italy; 2%

Netherlands; 2%

Total proportion of reports per country in VigiBase

17

Each DIPS element was evaluated for being aVigiBase report and its corresponding original file. The retrieved case information was specified as being listed in the structured fields, free text

The reference set generated in study IV was used to develop a predictive detection of novel adverse drug interactions in VigiBase.

individual indicators were studied, although additional indicators were introduced by combining the unique indicators tested previously. Logistic regression was used to set the relative weights of the indicators to form triage algorithms. Three algorithms (one completely data driven, one semi-automated and one based on clinical knowledge) were then evaluated against a set of 100 randomly sel

including potential adverse drug interactions. The algorithm’s performances were then evaluated by comparing true positive rates for

ADR (DDA) triplets with high coefficients.

VigiBase is a vast resource of medicine safety information and consists of more than reported from 1968 and onwards by health care professionals and

health professionals from 106 countries worldwide. The volume of reports in between countries and twelve countries contribute with over

87%. See Figure 2.

of reports in VigiBase per country (including countries contributing with at least 2%)

United States; 50%

United Kingdom; 9%

Germany; 6%

Canada; 5%

France; 4%Australia; 4%Spain; 3%

Sweden; 2%

All others; 13%

Total proportion of reports per country in VigiBase

being available on the he retrieved case

information was specified as being listed in the structured fields, free text

develop a predictive drug interactions in VigiBase. In

although in this study introduced by combining the unique indicators regression was used to set the relative weights

to form triage algorithms. Three algorithms (one automated and one based on clinical

a set of 100 randomly selected including potential adverse drug interactions. The algorithm’s

comparing true positive rates for

safety information and consists of more than health care professionals and

volume of reports in contribute with over

of reports in VigiBase per country (including countries contributing with at least 2%)

Kingdom; 9%

Germany; 6%

Total proportion of reports per country in VigiBase

18

An individual report in VigiBase includes at least one drug suspected of causing the adverse reaction, at least one suspected ADR, country of origin and an identification number. Drugs and ADRs listed on the reports are mapped with standard terminologies. Drugs are coded with WHO Drug Dictionary Enhanced and ADRs are mapped with WHO-Adverse Reaction Terminology (ART)[41] or Medical Dictionary for Regulatory Activities (MedDRA).[85] Drugs listed on the individual case reports are assigned to one of the following categories: 1) “suspected” (drugs suspected of causing the reaction, but not explicitly due to a drug interaction) or, 2)”interacting” (if an adverse drug reaction is suspected of being related to a drug interaction between two or more drugs) or 3) “concomitant” (drugs used concurrently but not suspected by the reporter to have caused the ADR). In addition to the mandatory information (drug, ADR, country of origin and identification number), the reports can also include more detailed clinical information such as therapy dates, date of reaction (onset date), dose information, route of administration and information regarding the outcome of drug withdrawal and/or drug re-introduction. In 2003 the first reports in ICH/E2B format was entered in VigiBase. This reporting format allows the reporter/sender (the person at the national or regional centre sending the report) to include more detailed and specific information such as case narrative and diagnostic tests which permits a thorough case analysis.

Drug Interactions in VigiBase

On the 6.6 million case reports in VigiBase 27 million drug-drug-ADR triplets have been reported at least once. These triplets could potentially represent suspected adverse drug interactions. There are three alternatives of how a drug interaction can be recorded on a report:

• two drugs assigned as interacting, • drug interaction terms, or • a case narrative including the word stem ‘interact’ or ‘interact’.

A drug interaction is most explicitly noted when the drugs are listed as interacting. A drug interaction term is in general less informative than when two drugs are assigned as interacting, as the interacting effect cannot be directly related to the specific drugs involved. Since interaction terms were recently (2010) added to the WHO-ART terminology the following MedDRA preferred terms to define drug interactions: Drug interaction, Labelled drug-drug interaction medication error, Inhibitory drug interaction and Potentiating drug interaction. Information regarding a suspected drug interaction is sometimes also available only in the form of free text. To capture this interaction information free text searches including the individual wording ‘interact’ or ‘interact’ were used. Furthermore, a drug interaction ascribed in the free text is more difficult to manage in the systematic screening as some level of text screening needs to be made, and the current method results in some false positives (around 14%) interactions.[86] In this thesis an interaction report is defined to include any of the three elements mentioned above.

In total, approximately 52 000 (0.8%) of VigiBase reports contains any of the three above elements signifying a drug interaction. Figure 3 shows the proportion of

interaction reports per year. It shouldonly include reports having two drugterms have been reported since 2002 and reports including interaction in the free text since 2003. As seen in greater during 1985 and 1986. incorrectly categorised as interacidentified quality issue studies IIIonwards.

Figure 3. Percentage of reports with any indication of a drug interaction

assigned as interacting, or a MedDRA interaction term or a drug interaction noted in the case narrative.

Of all interacting reports, tthe reports, MedDRA interaction term areports. There is an overlap between these variables i.e. some reports include several of these variables. However the majority of reports include just one90%). Figure 4 illustrates the overlap interacting variables on interaction reports entered from January 1990 to October 2009 in VigiBase.

The country distribution for suspected drug interaction reports in VigiBase contrasts general distribution of reports in VigiBaseUSA with 50%, while for drug interactions the distribution is spread between top 10 reporting countries. See Figure one report containing a suspected drug interaction, though the majority reports (84%) has been received from

0,0%

0,5%

1,0%

1,5%

2,0%

2,5%

3,0%

3,5%

4,0%

4,5%

5,0%

Proportion of interacting reports per year

19

interaction reports per year. It should be noted that interaction reports only include reports having two drugs assigned as interacting. MedDRA interaction

reported since 2002 and reports including interaction in the free text since 2003. As seen in Figure 3 the proportion of interaction reports are much greater during 1985 and 1986. During these two years nearly 5000 reports were incorrectly categorised as interacting. To avoid potential biases related to this identified quality issue studies III-VI include reports from 1st of January 1990

of reports with any indication of a drug interaction, including reports having

assigned as interacting, or a MedDRA interaction term or a drug interaction noted in the case narrative.

, two drugs assigned as interacting were reported on 0.5%MedDRA interaction term and Narrative both were listed on 0.2% of the

here is an overlap between these variables i.e. some reports include several of these variables. However the majority of reports include just one

illustrates the overlap interacting variables on interaction reports January 1990 to October 2009 in VigiBase.[86]

istribution for suspected drug interaction reports in VigiBase contrasts general distribution of reports in VigiBase. In the general distribution contributes

, while for drug interactions the distribution is spread between top 10 Figure 2 and Figure 5. 61 countries have submitted a

report containing a suspected drug interaction, though the majority received from the top ten countries.

Proportion of interacting reports per year

be noted that interaction reports up to 2002 assigned as interacting. MedDRA interaction

reported since 2002 and reports including interaction in the free the proportion of interaction reports are much

During these two years nearly 5000 reports were potential biases related to this

of January 1990

reports having two drugs

assigned as interacting, or a MedDRA interaction term or a drug interaction noted in the case narrative.

reported on 0.5% of listed on 0.2% of the

here is an overlap between these variables i.e. some reports include several of these variables. However the majority of reports include just one (around

illustrates the overlap interacting variables on interaction reports

istribution for suspected drug interaction reports in VigiBase contrasts . In the general distribution contributes

, while for drug interactions the distribution is spread between top 10 have submitted at least

report containing a suspected drug interaction, though the majority of such

Figure 4. The overlap (in proportion)

January 1990 to October 2009 in VigiBase

Figure 5. Top 10 countries having reported any indication o

assigned as interacting, MedDRA ADR term interaction or

and 1986 are excluded to avoid an identified quality issue in the categorisation of drugs as

these reports.

Data management

External drug interaction reference sources have been used systematically access the information available in these sources the xml format used. The standard drug and ADR tautomatically link the data in these external sources to VigiBaseliterature references were mapped to their substance name in the WHO Drug

MedDRA term 29.3

GBR 6%

NLD 4%

USA 4%

SWE

Proportion of interaction reports per country excluding reports entered during 1985

20

proportion)between interacting variables on interaction reports entered between

in VigiBase.[86]

op 10 countries having reported any indication of a drug interaction including reports with two

assigned as interacting, MedDRA ADR term interaction or specified in the case narrative.

to avoid an identified quality issue in the categorisation of drugs as

xternal drug interaction reference sources have been used in studiesthe information available in these sources the xml format

ard drug and ADR terminologies used in VigiBase permitted us automatically link the data in these external sources to VigiBase

were mapped to their substance name in the WHO Drug

Narrative20.7

Interacting62.4

MedDRA term 29.3

FRA 15%

ESP 14%

AUS12%

DEU11%

CHE7%CAN

7%

GBR 6%

USA 4%

SWE 4%

All others16%

Proportion of interaction reports per country excluding reports entered during 1985-86

4.9

6.0 3.4

1.9

reports entered between

including reports with two drugs

in the case narrative. Reports from 1985

to avoid an identified quality issue in the categorisation of drugs as interacting among

tudies I, IV and VI. To the information available in these sources the xml format was

permitted us to automatically link the data in these external sources to VigiBase. Drugs within

were mapped to their substance name in the WHO Drug

21

Dictionary Enhanced[41] and adverse reactions in the reference set used in studies IV and VI were mapped to the WHO-ART terminology.

As mentioned above one of the fundamental problems for collections of ICSRs is the presence of duplicate reports. VigiBase is therefore regularly screened for suspected duplicate reports.[43] In the case analysis, study II, suspected duplicates reports were highlighted with the automatic method and excluded from the further analysis when confirmed via clinical evaluation. In the large, systematic screenings of VigiBase, studies IV and VI, suspected duplicates reports were automatically identified and only one report in a group of suspected duplicates were retained in the analysis (the report with the greatest amount of information).

Ethical considerations

As all studies in this thesis are based on ICSRs which are anonymised no ethical approvals were needed.

Reference sources of drug interactions

Swedish, Finnish, INteraction X-referencing drug-drug interaction (SFINX) database (Study I)

SFINX is a drug-drug interaction database containing potential interactions for substances registered in Sweden and Finland.[75] The database is developed by staff at the Department of Clinical Pharmacology, Karolinska Institute, Stockholm, Sweden, Drug Interaction Unit at University Hospital in Turku, Finland, and the Division of Drug Management and Informatics at Stockholm County Council, Stockholm, Sweden. In Sweden, SFINX is integrated in the Janus medical journal system in Stockholm county council, and accessible through the Internet[75,87] and in Finland it is published in the largest medical portal.[75]

At the time of paper I, the SFINX database included around 5500 drug interactions supported by scientific literature (Medline, Drugline, DrugDex, Stockley’s Drug Interactions 6th ed 2002, Hansten and Horns Drug Interactions Analysis and Management, European Public Assessment Report (EPARs), Swedish SPC and Pharmaca Fennica). The drug interactions included were primarily pharmacokinetic, but there were some pharmacodynamic interactions (unless they were apparent from the main pharmacological action of the substances).

The drug interactions are classified according to clinical significance (A-D) and documentation level (0-4). In study I we examined interactions classified as D4, where clinical significance of D represents: “the interaction may result in serious

clinical consequences in terms of adverse drug reactions, therapeutic failure and is

difficult to master by individual dosages. The combination therefore ought to be

avoided” and a documentation level of 4 informs that: “the interaction has been

documented in controlled studies of the relevant patient material”. These drug interactions were referred to as ‘established and clinically important’ in study I and in this thesis.

22

Stockley’s Drug Interactions (Studies IV, VI)

Stockley’s Drug Interactions[4] is well-renowned and the most complete international listing of drug interactions.[88] Stockley’s Drug Interaction Alerts,[89] is derived from the text book/electronic publication. It is a web interface designed to be used for reference at the point of prescribing or dispensing. It categorises and summarises information of more than 40000 clinically evaluated drug–drug, drug-herbal, drug–alcohol and drug–food pairs. Each of the alerts are rated and formed into three separate categories: 1) action (describes whether or not any action needs to be taken to accommodate the interaction and this category ranges from ‘avoid’ to ‘no action needed’); 2) severity (describes the likely effect of an unmanaged interaction on the patient and this category ranges from ‘severe’ to ‘nothing expected’); 3) evidence (describes the weight of evidence behind the interaction and this category ranges from ‘extensive’ to ‘theoretical’). For each interaction a short summary is also provided describing the interaction and indicating whether the drugs can safely be taken together, if they may alter the therapeutic effect of one another, or may result in ADRs. Information in Stockley’s Drug Interaction Alerts was used to create the reference set used in studies IV and VI.

23

Methods

Causality assessment method

Drug interaction probability scale (DIPS) (Study V)

Causality assessment of suspected drug interactions is in general even more difficult than that of single drug-ADR associations due to the complexity of having to consider two drugs in parallel.

In study V we used a causality assessment algorithm for suspected drug interactions, the drug interaction probability scale (DIPS)[84] which in general is intended to serve as a reference for scientific publication of clinical case reports with large amount of clinical information. The assessment includes general information whether there is support in the literature that the drugs of interest may indeed interact, and whether there is a plausible pharmacological basis for the interaction. It also evaluates relevant properties of the clinical case at hand: whether the time course of the ADR is consistent with the suspected drug interaction, whether there is information on the effect on the drug interaction of withdrawing the drug inducing the interaction, and whether the adverse reaction reappeared when drug inducing the interaction was re-administered in the presence of the affected drug.

Statistical methods

The raw number of reports of a single drug-ADR combination or an adverse drug interaction in post-marketing pharmacovigilance reporting databases cannot be interpreted as an estimate of incidence in the population as there is no direct information that describes the usage of drugs, frequency of ADRs, drug-ADR combination or drug interactions in the general population. However, measures of disproportionality can be used to gauge the relative reporting rates within the dataset and to highlight for which drug-ADR combinations or adverse drug interactions the reporting rate is greater than expected in relation to the background. Studies II, IV and VI used the three-way disproportionality measure Omega (Ω) which is described in more detail below.

Logistic regression[90] was used to set the relative weights of the indicators (information supportive of a drug interaction) to form Triage algorithms in study VI.

Omega (Ω) (Studies II, IV, VI)

Omega (Ω) is an observed-to-expected measure of disproportionate reporting for a drug-drug-ADR (DDA) triplet whose ultimate endpoint is to highlight potential adverse drug interaction signals. The measure indicates how frequently a DDA is represented in the dataset in comparison to what is expected based on the relative reporting in the dataset. The observed counts are based on the relative frequency of the ADR with both drugs together (i.e. the DDA). The expected reporting is based on the relative frequencies for the ADR) without the two drugs (f00), 2) with drug1 but not drug2 (f10), and 3) with drug2 but not drug1 (f01).

Figure 6 is a schematic figure of the concept of Ω. It’s an additive model where two drugs causing the same ADR The method is described iindicates that the risk of the ADR is greater if the they are taken one at a time.

Ω log observed 0expected 0

Observed = the relative reporting rate of the ADR with both drugs

Expected = is based on relative frequencies for the ADR 1) without the two drugs (f

but not drug2 (f10), and 3) with drug2 but not drug1 (f

Figure 6. Illustration of the principle for

pie in the middle f11 is the observed relative reporting rate for the DDA that is compared to expected relative

reporting rate that is based on f00,

drug2) in the dataset.

In the formula above 0.5 is given. applied to stabilise the estimated value, in particular when there are few observed reports (i.e. the number of reports for the without the shrinkage 2 observed reports and ratio of 20, but when the shrinkagshrinkage to 200 observed reports observedimpact as it gives a ratio of (20). The logarithm is applilower limit of the 95% credibility interval for further statistical support.

Drug

The relative frequency of the ADR with drug1

without drug2 (f10)

24

is a schematic figure of the concept of Ω. It’s an additive model where two the same ADR add to the risk of the ADR, independently of each other. is described in more detail in Norén et al.[82] Loosely, a

risk of the ADR is greater if the two drugs are used togetherthey are taken one at a time. The formula for Ω is as follows:

0.50.5

the relative reporting rate of the ADR with both drugs

is based on relative frequencies for the ADR 1) without the two drugs (f

), and 3) with drug2 but not drug1 (f01)

the principle for relative ADR frequencies on which the Ω measure is based. The small

is the observed relative reporting rate for the DDA that is compared to expected relative

00, f10 and f01 and the number of reports with the drug combination (

In the formula above 0.5 is given. 0.5 is the shrinkage (a dampening value) which is the estimated value, in particular when there are few observed

the number of reports for the DDA triplet) in the database. For without the shrinkage 2 observed reports and 0.1 expected reports would

he shrinkage is applied the ratio is 4.2. Whileobserved reports observed and 10 expected reports

gives a ratio of 19.1 which is almost the same as without the shrinkageThe logarithm is applied to centralise the quotient of Ω around zero. Usually the

of the 95% credibility interval for Ω, Ω025, is used in order to increase further statistical support.

ADR

Drug2Drug1

without drug

ADR without drug

is a schematic figure of the concept of Ω. It’s an additive model where two add to the risk of the ADR, independently of each other.

Loosely, a positive Ω used together than if

is based on relative frequencies for the ADR 1) without the two drugs (f00), 2) with drug1

Ω measure is based. The small

is the observed relative reporting rate for the DDA that is compared to expected relative

umber of reports with the drug combination (drug1 and

(a dampening value) which is the estimated value, in particular when there are few observed

DDA triplet) in the database. For example, would lead to a

While applying the reports it has limited

which is almost the same as without the shrinkage Ω around zero. Usually the

, is used in order to increase

The relative frequency of the ADR with drug2

without drug1 (f01)

The relative frequency of the ADR with drug1 and drug2 (f11 or observed value)

The relative frequency of the

ADR without drug1

or drug2 (f00)

25

Results

Below are short summaries of results presented in each study (I-VI).

I This study showed that well established drug interactions are reported, though not necessarily recognised as such on the reports in VigiBase. The results from this study illustrated a long-standing international problem of co-medication of contraindicated drugs.

II The case reports of the potential association of azithromycin - statins and rhabdomyolysis was suggestive of a drug interaction. The reporter/sender assessments suspecting a possible interaction and plausible time lapses between the drug interaction and reaction and strengthened the causality. Azithromycin – statins - rhabdomyolysis was reported more often than expected (in comparison to general reporting to VigiBase) from 2000 and onwards as indicated with positive Ω025 values.

III Drug interactions reported on globally collected ICSRs cover both pharmacodynamic, specifically additive pharmacological effects, and pharmacokinetic mechanisms, where the drugs involved have the same metabolic pathway.

IV Information that were reported on the ICSRs for known adverse drug interactions to greater extent than for non interacting drug combinations in the time period before a drug interaction became listed in the literature are: suspicions of a drug interaction by the reporter as noted in a case narrative, as an ADR term, or the assignment of the two drugs as interacting, an increased effect, and a positive Ω025.

V Plausible time lap between the drug interaction and reaction and resolution of the ADR upon withdrawal of the drug inducing the interaction were reported information that strengthens the causality of a drug interaction on individual reports. For two of the three case series provided the original files more complete information. While for one case series there were small differences in the DIPS assessment between VigiBase reports and original files.

VI The clinical evaluation of the 100 randomly selected case series showed that 20% were already known in the literature, 30% were classified as signals and 50% as not signals. Even though the performance of the semi-automated and the clinical algorithm were comparable the clinical algorithm was chosen as it is less complex, uses a broader basis of qualitative information, and promotes CYP mediated interactions to a less extent as well as incorporates Ω025. Among DDA triplets with high coefficients (a true positive rate of ≥0.15): 38% were already known in the literature and 80% of the remaining DDAs were classified as signals worthy of further follow-up.

26

Study I showed that drug interactions are a longstanding (including reports from 1968 an onwards) and international problem (including reports from 50 countries). However, only 14 countries2 had reported an interaction term in the time period before an adverse drug interaction became established in the literature (unpublished data from study IV). Demographics from the descriptive studies (I, II, III and V) are summarised in Table II.

Table II. Summary of results from descriptive studies (I, II, III and V). Rhabdomyolysis as a result of

azithromycin and statins are presented in study II. In study V the series of coagulation disorders due to

concurrent use of omeprazole and coumarine derivates referred is to as series I and changes in international

normalized ratio (INR) related to the concurrent use of glucosamine and warfarin as series II.

Study Total

number

of ICSRs

Countries

reporting

Age distribution Sex

distribution

Reporter

I 9547 50 countries 66% from USA

- Males 56% Females 44%

HCP 50% (Physicians 38% Pharmacists 8% Nurses and dentists 4%)

II 53 5 countries 87% from USA

41-85 years (Median 60 years)

Males 58% Females 42%

HCP 79% (Physicians 51% Pharmacists 19% Nurses and dentists 9%)

III 3766 47 countries 67% from Europe*

≤17: 3% 18-44: 14% 45-64: 21% 65-74: 21% ≥75: 33%

Males 48% Females 52%

HCP 83% (Physicians 78% Pharmacists 4% Nurses and dentists 1%)

V.I 51 8 countries Netherlands 21% Switzerland 17% Germany 16%

38-95 years (Median 72 years)

Males 53% Females 47%

HCP 93% Physicians 74% Pharmacists 17% Nurses and dentists 2%

V.II 33 6 countries USA (33%) AUS (27%) UK (15%)

33-99 (Median 70 years)

Male 61%; Female 39%

HCP 81% (Physicians 59% Pharmacists 11% Nurses and dentists11%)

*In study III proportions for regions ere examined, therefore are data not provided for individual countries

HCP= health care professionals

Characteristics of suspected/potential drug interactions (Studies I, III)

The majority (113/123) of drug combinations co-reported as interacting in study III, were established to interact in the literature. Of the 113 drug interactions 46 (41%) were identified as purely pharmacodynamic; 28 (25%) as pharmacokinetic; and 18 (16 %) were a mix of both types, and for 21 (19%) the mechanism have not yet been identified. In this subset of drug combinations the most common mechanisms were additive pharmacological effects and inhibition of metabolic pathway. Of the

2 Australia, Bulgaria, Canada, France, Germany, Ireland, Netherlands, New Zealand, Norway, Spain, Sweden, Switzerland, United Kingdom, United States.

27

interactions acknowledged to a specific enzyme, CYP3A4 and CYP2C9 were particularly frequent and these accounted for 42% and 24%, respectively. Other enzymes involved were CYP1A2, CYP2B6, CYP2C19, CYP2C8 and epoxide hydrolase.

Despite different study designs in studies I and III, the results from these studies cover a similar set of drugs were identified. A large proportion of the reports involved drugs with a serious ADR profile and drugs with narrow therapeutic range such as anticoagulants including warfarin, and anticonvulsants (carbamazepine and phenytoin). A large proportion of reports also concerned drugs that have been available on the market for more than 10 years.

Table III shows top 15 ADR terms for drug pairs reported ≥20 times as interacting in VigiBase (study III).

Table III. Top 15 ADR terms for drug pairs reported ≥20 times as interacting in VigiBase (study III).

ADR Term No of Reports Critical Term

Prothrombin level decreased 402 x

Drug level increased 305

Haematoma 201

Therapeutic response increased 180

Anaemia 177

Rhabdomyolysis 172 x

Drug interaction 168

Melaena 163 x

Renal failure acute 152 x

Vomiting 130

INR increased 129

Gastrointestinal haemorrhage 127 x

Bradycardia 122

Drug toxicity 119

Nausea 110

INR: International Normalised Ratio.

Signal detection and information strengthening causality (Studies II, IV, V)

In study II a potential adverse drug interaction between azithromycin – statins had been suspected by the reporter/sender in 11 of the 53 reports. The drugs were assigned as interacting in three reports, and the interaction was noted in narrative text in another eight cases. In total, azithromycin and a statin were both reported as suspected or as interacting in 45% of the reports. Furthermore, the long term use of statins with a rapid onset within 10 days of rhabdomyolysis from initiation of azithromycin was also indicative of a drug interaction.[91-92] Since the majority of reported statin doses were within the recommended daily doses[48-49] this suggests that an external factor may have induced the interaction.

28

Similarly to what was reported in study II, study V showed that consistent time lapses between the drug interaction and the ADR strengthened the suspected causality of a drug interaction on the individual report. The strength of this information varied with duration and initiation of the affected drug and the drug inducing the interaction. Case causality was particularly strong when the drug expected to induce the interaction, altered the effect of the affected drug shortly after its initiation. Study V also showed that resolution of the ADR upon withdrawal of the drug inducing the interaction (referred to as positive dechallenge) strengthened the suspected causality of a drug interaction. Figure 7 shows one of the reports included in study V; it exemplifies how re-exposure of glucosamine affected warfarin’s expected outcome.

Figure 7. A sample of an original individual case safety report showing repeatedly raised INR values related to

re-introduction of glucosamine. The case report is reprinted with permission from Health Canada, Canada.

29

Study V also showed that the level of information provided on VigiBase reports and original files varies. For two case series were the information given in the original files more complete. While for the remaining case series there was small differences in the DIPS assessment between VigiBase reports and original files. The variations between the case series were primarily explained by three factors: country submitting the report, reporting format, and when report was submitted.

Reporting rate of adverse drug interactions (Studies II, IV and VI)

Within the scope of this thesis the Ω (Omega) measure has been tested in practice for detection of novel adverse drug interactions. The first prospective screening of VigiBase using the Ω measure was done in July 2008. That screening highlighted an increased relative reporting (Ω025 >0) of rhabdomyolysis under the concurrent use of azithromycin and the individual statin: atorvastatin, lovastatin and simvastatin. This resulted in a study II. To study the reporting over time in VigiBase Ω025 values was retrospectively calculated. The retrospective analysis of rhabdomyolysis with azithromycin – statins showed that the association was reported more often than expected in comparison to general reporting to VigiBase from 2000 and onwards. The results suggest that association could not be explained by high reporting of statins observed since the withdrawal of cerivastatin in 2001.

Study IV was the first large scale evaluation relative to a comprehensive references set that demonstrate the value of the Ω025 to highlight adverse drug interactions. This study showed that positive Ω025 values are much more common for known adverse drug interactions than for drugs not known to interact. Study IV also demonstrated that reporting patterns based on excess reporting rates tend to highlight other adverse drug interactions than those highlighted by detailed clinical information. Although, the results in study VI show that Ω025>0 in isolation has a more limited value than an algorithm based on broader scope of information (see section Triage algorithms for detection of adverse drug interactions).

Underreporting of interacting drugs on the individual report

To increase our understanding of how adverse drug interactions are reported in VigiBase a complementary analysis including DDAs highlighted in study I were done. One of the examples was insulin-rosiglitazone-cardiac failure of which the US FDA issued a warning in 2004.[93] Figure 8 shows the absolute number of reports (including all drugs listed on the reports i.e. reported as suspected, interacting or concomitant) of cardiac failure with rosiglitazone, insulin and the combination of the two drugs. Both drugs are individually reported more frequently with cardiac failure than the combination. Since the number of reports with cardiac failure for insulin raised at the same time as rosigliatzone was launched, it may well be that rosiglitazone is co-administered but not listed on insulin and cardiac failure reports. If so, Figure 8 is an example of under-reporting of concurrently used agents. However, this increase could also be related to the enhanced focus and reporting on cardiac ADRs during the past decade, which occasionally would include cardiac failure and also insulin.

30

Figure 8. Reporting trend including absolute numbers for rosiglitazone-cardiac failure, insulin-cardiac failure

and rosiglitazone-insulin- cardiac failure in VigiBase. 21 percent of all reports of insulin and cardiac failure also

had rosiglitazone co-reported.

Indicators of adverse drug interactions (Study IV)

Study IV showed that suspicions of a drug interaction noted by the reporter/sender in a case narrative, as an ADR term, or the assignment of the two drugs as interacting are much more common for known adverse drug interactions than for drugs not known to interact. Excessive co-reporting of an ADR together with two drugs as measured by the Ω025 and the co-reporting of enhanced therapeutic effect (effect increased), were also much more frequent for known adverse drug interactions than for drugs not known to interact. The study also demonstrated that reporting patterns based on detailed clinical information tend to highlight other adverse drug interactions than those with Ω025.

Figure 9 shows the proportion of DDAs occurring with the primary indicators (indicators that may independently drive suspicion of an adverse drug interaction as they provide clinical information to suggest that a suspected drug interaction has occurred) among DDAs (known adverse drug interactions and DDAs where the drug combinations is not known to interact) in the reference set developed in study IV. The ratio between the groups and the numbers behind the proportions are given in text.

31

Figure 9. Shows the proportion (with 95% confidence intervals) of DDAs occurring with the primary indicators,

among DDAs in the reference set developed in study IV. The ratio between the groups and the numbers behind

the proportions are given in text.

Triage algorithms for detection of adverse drug interactions (Study VI)

Three algorithms (a full data driven, a semi-automated (referred to as lean data driven in paper VI) and a clinical algorithm (referred to as lean clinical in paper VI)) were examined. Though in practice the semi-automated and a clinical algorithm were tested, since the full data driven algorithm was excluded as it was believed to be over-adjusted to the reference set. The algorithms used a broader range of information than what is customary in first-pass screening of adverse drug interactions. Detailed reported information such as notes of suspected interactions and indications of altered therapeutic effect, and case series driven by individual case reports with strong support of an adverse drug interaction was utilised. The algorithms also took into account whether the drugs are metabolised via the same cytochrome P450 enzyme, and if so, if their activity may affect the metabolism of the other drug involved.

The clinical evaluation included case series for 100 randomly selected adverse drug interactions. Of these 100 were 20 already known adverse drug interactions in the literature. Of the remaining 80 DDAs, 30 were classified as signals and 50 as not signals. The performance of the semi-automated and the clinical algorithm were comparable for DDAs with high coefficients. Furthermore, both algorithms out perform the Ω025 measure when applied alone. See Figure 10, whereas the algorithms show a higher true positive rate (more to the left) than the Ω025 (represented as the dotted line).

Even though the algorithms’ performances were comparable (see Figure 10) there were clear advantages of choosing the clinical algorithm rather than the semi-

32

automatic: it is less complex, uses a broader basis of qualitative information, and promotes CYP mediated interactions to a less extent as well as incorporates Ω025. To estimate how well the clinical algorithm would work in practice we hypostasized that we will be able to yearly assess around 2000 DDAs for the next 7 years (in total 14000). DDAs with a true positive rate of 0.15 or greater met this threshold. Of the 100 DDAs in the clinical evaluation, 16 cases series had a true positive rate of 0.15 or greater, 6 (38%) were identified already known in the literature and 8 of the remaining 10 (80%) were classified as signals worthy of further follow-up.

Figure 10. Receiver operating curves (ROC curves) comparing the triage algorithms to a pure disproportionality

triage based on Ω025 only, relative to the manual clinical review of the 80 DDAs in the evaluation data set not

found in the literature. The circle corresponds to a threshold of 0 for Ω025. The region to the left of the vertical

line is considered to be the one relevant in practical use. The clinical algorithm is referred to as the lean clinical

and the semi-automated algorithm as the lean data driven in study VI.

33

Discussion

Drug interactions represent an important drug health issue and if known they could at least in theory be avoided. This thesis has demonstrated that post-marketing pharmacovigilance reporting databases are a valuable source for surveillance of drug interactions. The reported results within this thesis showed some important aspects concerning drug interactions. In study I we found a continuing reporting of seemingly well established interactions, these results suggests that individuals regularly take high risk drugs together. The results in study I may also well be a reflection of real life where physicians fail to recognise drug information and continue to prescribe interacting drugs.[63-66,94-96] Furthermore, the results of drug combinations with additive effects in study III could indicate that the pharmacological mechanism is not apparent or understood by the prescriber[97] and perhaps a reflection of the prescribers’ general knowledge of drug interactions. These results suggests that multidisciplinary strategies including increased openness and understanding of ADRs[98] and enhanced pharmacological knowledge among prescribers, as well as enhanced routine prescribing practices are required to reduce future harm and costs for preventable ADRs related to drug interactions.

We found in study III that the interactions were of both pharmacodynamic and pharmacokinetic nature. In contrast to these results, results from hospital settings report that the majority of drug interactions responsible for ADRs are pharmacodynamic.[68,71] Most likely are our results a reflection of different treatment strategies and drug usages in various settings (i.e. hospital versus primary care), wherefore the heterogeneity sometimes regarded as a disadvantage of post-marketing pharmacovigilance reporting system (in comparison to other epidemiological studies), was a clear advantage in study III. It is likely that both pharmacodynamic and pharmacokinetic interactions are responsible for interactions occurring in clinical practise, although the prominence of CYP mediated in study III may have been influenced by selective reporting related to drug withdrawals (see Table I) or the emphasis on these interactions in the literature.

A new drug interaction is often suspected when other substances within the same pharmacological class have been shown to cause an interaction, in particular if the drugs have the same pharmacokinetic properties. However, the azithromycin-statin example in study II shows that potential drug interactions should not be completely disregarded if a drug does not hold the same (or to the same extent) pharmacokinetic properties as the others within the same pharmacological class. There can be more mechanisms involved, even though they are not known or less well understood at the time being. Furthermore the risk of a drug interaction may be increased in particularly vulnerable patients.

Even if the regulatory post-marketing surveillance focuses on new drugs[99] the results in studies I and III and data presented previously by Johansson and Meyboom[100-101] jointly emphasise the importance of continuous monitoring of well established drugs with a serious ADR profile. This is particularly important for drug interactions, as problems related to the concurrent use of new drugs are constantly being discovered.[79-81,102] Since post-marketing pharmacovigilance reporting

34

databases are representative of drug safety problems occurring in the entire population for all drugs during their entire life span these reports will be a good cost-efficient alternative in the surveillance of adverse drug interactions.

For obvious reasons 100% reporting can never be achieved. The reporting of ADRs relies on the recognition of drug induced effects and that the information regarding the ADR is being transferred successfully from the patient via health care professionals, sometimes via the market authorization holder, to the national centres, and then to VigiBase. Some of the reasons for under-reporting[103] could be that the patient may not recognise and report the adverse reaction; or the health care system may fail to recognise the symptoms as a drug induced effect from one single drug or from the combination of two drugs. Under-reporting in terms of drug interactions have additional implications, for instance not all potential drug interactions are recognised as interactions (studies I, II and V) or that not all medicines taken by a patient who has experienced a suspected ADR is reported on the individual reports (illustration of rosiglitazone and insulin in figure 8). These findings were expected as the rarity of drug interactions on individual reports have been described previously by van Puijenbroek,[12] and other studies have shown a much higher rate of ADRs related to drug interactions (around 20%),[20,65,70] than what is reported in VigiBase(0.8%).

Even if a drug interaction is rarely listed on the individual reports in VigiBase the results in study IV showed that when an adverse drug interaction is recognised (by the person providing the information to the national authority or by staff at the national authority) it may be a good indicator of what becomes acknowledged as adverse drug interactions in the future. The strengths and merits of ICSRs are dependent on the quality of information provided. The possibility for thorough case assessments of ICSRs is often influenced by the quality of the documentation of the observations, and the amount of uncertainty in a report is often determined by the absence of data. The absence of data is particularly a problem for adverse drug interactions for which the assessments are in general more complex than the assessment for single drug-ADR associations, as more data is needed to be able to evaluate the reports. The reports in VigiBase are have sometimes been mentioned to be of poor quality, however the results presented within this thesis demonstrates that VigiBase reports contain sufficient qualitative information which can be used to generate drug interaction signals.

The triage algorithm that we propose in study VI incorporates the knowledge that we have gained through the individual projects in this thesis. For instance studies I and II fed ideas on what information on an individual report supports an adverse drug interaction and study III addressed the importance of detecting any drug combination with an increased risk of an ADR during concurrent use independently if the two drugs have additive or synergistic pharmacological effects. Figure 10 shows that the algorithm is superior to Ω025 alone, which probably reflects its broad basis using clinical and pharmacological information as well as the relative reporting rate. In comparison to the first-pass screening of single drug-ADR in VigiBase (around 50%), the drug interaction triage has a lower proportion (38%) of already known associations. It also highlights a greater proportion (80%) of the remaining associations as potential signals. The relative figure for the pair-wise drug-ADR triage

35

strategy is around 20%. These results indicate that the proposed triage algorithm will work more efficiently than the current triage algorithms for pair-wise associations. To our knowledge clinical information listed on the case reports has not yet been used in the automatic detection of ADRs ascribed one single drug and a similar approach has therefore also the potential to improve pair-wise drug-ADR surveillance. Though, it should be stressed that it is not necessarily an advantage with a high proportion of signals. It is more important that signals have clinical value and consist of some level of support. For instance the rising number of statistics of disproportionate reporting (SDR)[6] (sometimes referred to as signals of disproportionality i.e. signals that have not been evaluated clinically) (for single drug-ADRs) in the literature is a problem as it causes an overload of information, which is difficult to retrieve, sort and interpret in clinical decision making. Furthermore, to the best of our current knowledge, the interaction algorithm does not appear to include associations confounded by data quality issues in the top ranked associations as reported for pair-wise disproportionality measures by Bate et al.,[11] and mentioned previously as a potential problem for disproportionality measures used for adverse drug interactions detection (see section Drug interaction surveillance using individual case safety reports).[83] The reason for this is probably that the disproportionality measures exclusively rely on raw numbers of reports, which makes the measures particularly vulnerable to selective reporting.[42] Selective reporting was clearly a risk in the analysis of statins and rhabdomyolysis[104-105] (with azithromycin), because of the world wide withdrawal of cerivastatin in 2001. However the results in study II showed that the reporting of this potential interaction had changed over time and the relative reporting rate was higher than expected already before the statin publicity (2001). This shows that the association could not be explained by the peculiarities of statin reporting observed in recent years.

The results in studies IV and VI clearly shows that the systematic screening benefit when only suspected and interacting drugs are included i.e. concomitant drugs does not improve the systematic screening for potential drug interactions. The descriptive analysis (I, II and V) on the other hand, show that there is a clear risk to miss important cases if concomitant drugs are excluded. Because of the conflicting information, our approach will therefore be to include drugs reported as suspected and interacting in the first-pass screening, while the following case analyses should include all reports with both drugs listed independently of how they are characterised.

The reference literature used within the scope of this thesis have been essential. These sources have had an apparent clinical and practical value on the conducted studies. For example the evaluation of VigiBase has been more efficient. Without the reference literature would the results in this thesis been more limited, and presumably the methodological development wouldn’t had come as far. Since drug interaction reference databases are often differently structured and use different classification systems[4,58,75] the results could potentially vary with the reference source. However, in study I it was verified that the majority of drug interactions defined as contraindicated in SFINX, were also considered as severe in the international comprehensive drug interaction source,[88] Stockley’s Drug

36

Interactions.[4] We can therefore assume that the subset of contraindicated drug combinations is generalisable (these are drug combinations should be avoided in clinical practice) and not specific to the developers of SFINX.

The main advantage of using causality algorithms in general rather than explorative case analysis is to reduce the subjectivity. However, the main limitation with causality methods is that there are no methods general enough to be applicable for all types of problems. Using DIPS (in study V) was an important step towards a better understanding of causality assessment for suspected adverse drug interactions. However, it is clear that DIPS was originally developed for well-documented clinical case reports rather than for those with scarcer information, which sometimes is the issue for ICSRs. To apply DIPS in the context of broad adverse drug interaction surveillance, the threshold for what constitutes a potentially interesting report probably needs to be lowered. For example, de- and rechallenge interventions of the drug believed to have induced the interaction without change to the affected drug can be useful for causality assessment, but are feasible primarily for interactions with limited clinical impact. For the large proportion of ICSRs that refer to serious reactions, it would not be defensible to only withdraw the drug assumed to have induced the interaction, and in practice both drugs are withdrawn simultaneously in order to reduce toxicity as rapidly as possible. DIPS de- and rechallenge interventions are appropriate in terms of the strength of causality but not by factoring in a patient’s safety. Similarly, DIPS places great emphasis on established mechanisms for the suspected interaction, but post-marketing safety surveillance must be able to detect unexpected ADRs, also in the absence of well-understood mechanisms of the interaction.

Finally, because of the influence of different sources of biases the results within the scope of this thesis should be interpreted with caution. The raw number of reports presented within these studies, per drug-drug combination, adverse drug interactions or mechanism for drug interactions, should not be interpreted as an estimate of incidence or frequency outside the database. Similarly, the disproportionality measures (IC and Ω) describe the relative reporting rates within the database and these measures are not representative of the true occurrence of a drug - ADR or an adverse drug interaction. The triage strategy should be used for hypothesis generation of adverse drug interaction signals, and not for signal testing, where other methods are superior.

37

Conclusions

• Drug interactions can be identified in large post-marketing pharmacovigilance reporting databases.

• ADRs related to drug interactions are a global problem. Many of the interactions analysed and assessed within the scope of this thesis are preventable. They include well established interactions with documented risks as well as those that are expected from the drugs pharmacological mechanism.

• Drug interactions reported on globally collected ADR reports covers both pharmacodynamic, specifically additive pharmacological effects, and pharmacokinetic mechanisms primarily accredited inhibition of hepatic CYP enzymes.

• The results also demonstrate important differences in the reporting patterns for adverse drug interactions and those of drugs not known to interact. Suspicions of a drug interaction by the reporter as explicitly noted in a case narrative, as an ADR term, or the assignment of the two drugs as interacting are much more common for known adverse drug interactions than for drugs not known to interact. Also, excessive co-reporting of an ADR together with two drugs as measured by the Ω025 and the co-reporting of enhanced therapeutic effect were also much more frequent for known adverse drug interactions than for non interacting combinations.

• Beyond the reporter’s notification of a suspected drug interaction, positive dechallenge from the drug inducing the interaction and plausible time lapses between the drug interaction and the ADR are clinical information that strengthen the causality of an adverse drug interaction signal.

• A triage algorithm, promoting strong case reports, pharmacological information more than relative reporting rates is developed. The proposed triage algorithm is superior to the disproportionality measure, Ω, alone.

38

Future perspectives

This thesis has shown that the post-marketing pharmacovigilance reporting system is an important source in the future surveillance of adverse drug interactions. This is important as often only combinations of two drugs are examined with respect to interactions, although patients are often treated with several drugs in health care today. However, to improve the international surveillance of drug interactions even further, guidelines to harmonise the reporting of drug interactions are needed.

The algorithm for surveillance of adverse drug interactions will soon be put into routine use at the UMC. This routine screening can hopefully aid on a global level to detect novel adverse drug interactions that are important for public health. However, the first-pass screening could be improved even further if one could automatically identify strong cases including likely time lapses between the drug interactions and the ADR as well as positive dechallenge of the drug inducing the interaction.

The results in study VI clearly demonstrate the value of incorporating clinical information and pharmacological information in first-pass screening, rather than just using measures of disproportionality. The first-pass screening for single drug-ADR combinations or in groups that are particularly vulnerable, such as children, elderly, or patients with decreased renal or hepatic function, will also most likely benefit from this approach. The systematic screening and clinical case analysis can be further improved by having automatic access to clinical factors that are important for ADR outcome: dosage, time to onset,[106-108] therapeutic window, half-life time, receptor activity and specific chemical structures. However, more work needs to be done in this field.

Electronic medical records which have potential for detecting new signals signal have not yet been fully established.[109-110] Though these data sets have clear potential to be used as background information and for signal strengthening for adverse drug interaction surveillance as recently showed by Tatonetti et al.[111]

In Sweden there are exceptional opportunities to follow up patients via different registers because of the personal identification numbers. Since its introduction in 2005 the Prescription register has been much used. This register has to some extent been used for research on drug interactions[62,112-114] although much more knowledge can be gained by linking this register with others (SWEDIS, forensic databases, and hospital discharge registers) and drug-drug interaction databases.

39

Acknowledgements

I wish to express my warm and sincere gratitude to supervisors, colleagues, family, and friends who have helped and supported me during the preparation of this thesis. In particular I would like to thank the following:

Staffan Hägg, my main supervisor, for giving me the possibility to perform my PhD studies under your supervision, for broadening my perspective of my work within clinical pharmacology, and also for all support throughout the whole process!

Niklas Norén, my co-supervisor, for your day to day support, for challenging questions and discussions, and for helping me finalise this thesis.

Andrew Bate, my former supervisor, for your enthusiasm, for believing in me and inspiring me to pursue my PhD studies.

Marie Lindquist and Ralph Edwards for making me realise that pharmacovigilance is a very important field of research and making it possible for me to perform my PhD studies within this area: Marie, for being a role model and always sharing your expertise on all various aspects of VigiBase; Ralph, for your wisdom and constructive feedback.

Kristina Star, for broadening the scope of drug safety in my early days within pharmacovigilance, for many interesting discussions, for being a great friend for many years, and for being there for me when I needed it!

Ola Caster, for statistical support, for your efforts developing models, and thinking of how to present our results most optimally.

Johan Hopstadius, for computational support, for your engagement and encouragement during this work.

All my colleagues at the Uppsala Monitoring Centre for providing an excellent work environment, particularly the members of past Signal Detection and Analysis team who introduced me to pharmacovigilance: Anne Kiuru, Jenny Bate, Monica Plöen and Ronald Meyboom; and my current colleagues at the Research department: Ghazaleh Khodabakhshi, Tomas Bergvall and Kristina Juhlin for showing interest in our research and support on various issues.

Colleagues at Linköping University and the Sahlgrenska Academy in Gothenburg.

The SFINX group for giving us access to your data, and particularly Birgit Eiermann for making it possible. Pharmaceutical Press, for giving us the possibility to use Stockley’s Drug Interactions as reference source.

My family and friends in Falköping, Gothenburg, Skåne, Uppsala and elsewhere. In particular: my parents, for your longstanding support; my brother Fredrik and sister in law Charlotta, for heaps of good fun and making my stay in Gothenburg during the mandatory course very pleasant.

Finally I would like to thank Peter, my love and happiness, for all your love, and for all patient support during this period; and Arvid, my little sunshine, for boosting me with energy!

40

References

1. Edwards IR, Biriell C. Harmonisation in pharmacovigilance. Drug Saf. 1994;10(2):93-102.

2. Edwards IR, Aronson JK. Adverse drug reactions: definitions, diagnosis, and management. Lancet. 2000 Oct 7;356(9237):1255-9.

3. Swedish constitution for drugs (1992:859), 1 §. 1992. 4. Baxter K, editor. Stockley's Drug Interactions. Seventh ed: Pharmaceutical Press;

2006. 5. The importance of pharmacovigilance: safety monitoring of medicinal

productsGeneva: WHO; 2002. 6. Practical Aspects of Signal Detection in Pharmacovigilance. Report of CIOMS

Working Group VIII. 2010. 7. Juurlink DN, Mamdani M, Kopp A, Laupacis A, Redelmeier DA. Drug-drug interactions

among elderly patients hospitalized for drug toxicity. JAMA. 2003 Apr 2;289(13):1652-8.

8. Hamilton RA, Briceland LL, Andritz MH. Frequency of hospitalization after exposure to known drug-drug interactions in a Medicaid population. Pharmacotherapy. 1998 Sep-Oct;18(5):1112-20.

9. Doucet J, Chassagne P, Trivalle C, Landrin I, Pauty MD, Kadri N, et al. Drug-drug interactions related to hospital admissions in older adults: a prospective study of 1000 patients. J Am Geriatr Soc. 1996 Aug;44(8):944-8.

10. McDonnell PJ, Jacobs MR. Hospital admissions resulting from preventable adverse drug reactions. Ann Pharmacother. 2002 Sep;36(9):1331-6.

11. Bate A, Lindquist M, Edwards IR, Olsson S, Orre R, Lansner A, et al. A Bayesian neural network method for adverse drug reaction signal generation. Eur J Clin Pharmacol. 1998;54(4):315-21.

12. Van Puijenbroek EP, Egberts AC, Meyboom RH, Leufkens HG. Signalling possible drug-drug interactions in a spontaneous reporting system: delay of withdrawal bleeding during concomitant use of oral contraceptives and itraconazole. Br J Clin Pharmacol. 1999 Jun;47(6):689-93.

13. Lindquist M. Seeing and Observing in International Pharmacovigilance - Achievements and Prospects in Worldwide Drug Safety. Nijmegen: University of Nijmegen; 2003.

14. Wax PM. Elixirs, diluents, and the passage of the 1938 Federal Food, Drug and Cosmetic Act. Ann Intern Med. 1995 Mar 15;122(6):456-61.

15. McBride W. Thalidomide and congenital abnormalities. Lancet. 1961;278(7216):1358.

16. Furberg CD, Pitt B. Withdrawal of cerivastatin from the world market. Curr Control Trials Cardiovasc Med. 2001;2(5):205-7.

17. Psaty BM, Furberg CD, Ray WA, Weiss NS. Potential for conflict of interest in the evaluation of suspected adverse drug reactions: use of cerivastatin and risk of rhabdomyolysis. JAMA. 2004 Dec 1;292(21):2622-31.

18. Lazarou J, Pomeranz BH, Corey PN. Incidence of adverse drug reactions in hospitalized patients: a meta-analysis of prospective studies. JAMA. 1998 Apr 15;279(15):1200-5.

19. Mjörndal T, Boman MD, Hägg S, Bäckstrom M, Wiholm BE, Wahlin A, et al. Adverse drug reactions as a cause for admissions to a department of internal medicine. Pharmacoepidemiol and Drug Saf. 2002 Jan-Feb;11(1):65-72.

41

20. Pirmohamed M, James S, Meakin S, Green C, Scott AK, Walley TJ, et al. Adverse drug reactions as cause of admission to hospital: prospective analysis of 18 820 patients. BMJ. 2004 Jul 3;329(7456):15-9.

21. Pouyanne P, Haramburu F, Imbs JL, Bégaud B. Admissions to hospital caused by adverse drug reactions: cross sectional incidence study. French Pharmacovigilance Centres. BMJ. 2000 Apr 15;320(7241):1036.

22. Schneeweiss S, Hasford J, Gottler M, Hoffmann A, Riethling AK, Avorn J. Admissions caused by adverse drug events to internal medicine and emergency departments in hospitals: a longitudinal population-based study. Eur J Clin Pharmacol. 2002 Jul;58(4):285-91.

23. Wester K, Jönsson AK, Spigset O, Druid H, Hägg S. Incidence of fatal adverse drug reactions: a population based study. Br J Clin Pharmacol. 2008 Apr;65(4):573-9.

24. Pirmohamed M, Breckenridge AM, Kitteringham NR, Park BK. Adverse drug reactions. BMJ. 1998 Apr 25;316(7140):1295-8.

25. von Euler M, Eliasson E, Öhlen G, Bergman U. Adverse drug reactions causing hospitalization can be monitored from computerized medical records and thereby indicate the quality of drug utilization. Pharmacoepidemiol and Drug Saf. 2006 Mar;15(3):179-84.

26. Davies EC, Green CF, Mottram DR, Pirmohamed M. Adverse drug reactions in hospitals: a narrative review. Curr Drug Saf. 2007 Jan;2(1):79-87.

27. Bates DW, Cullen DJ, Laird N, Petersen LA, Small SD, Servi D, et al. Incidence of adverse drug events and potential adverse drug events. Implications for prevention. ADE Prevention Study Group. JAMA. 1995 Jul 5;274(1):29-34.

28. Kanjanarat P, Winterstein AG, Johns TE, Hatton RC, Gonzalez-Rothi R, Segal R. Nature of preventable adverse drug events in hospitals: a literature review. Am J Health Syst Pharm. 2003 Sep 1;60(17):1750-9.

29. von Laue NC, Schwappach DL, Koeck CM. The epidemiology of preventable adverse drug events: a review of the literature. Wien Klin Wochenschr. 2003 Jul 15;115(12):407-15.

30. Meyboom RH, Egberts AC, Edwards IR, Hekster YA, de Koning FH, Gribnau FW. Principles of signal detection in pharmacovigilance. Drug Saf. 1997 Jun;16(6):355-65.

31. Nelson KM, Talbert RL. Drug-related hospital admissions. Pharmacotherapy. 1996 Jul-Aug;16(4):701-7.

32. Meyboom RH, Lindquist M, Flygare AK, Biriell C, Edwards IR. The value of reporting therapeutic ineffectiveness as an adverse drug reaction. Drug Saf. 2000 Aug;23(2):95-9.

33. Rawlins MD. Spontaneous reporting of adverse drug reactions. I: the data. Br J Clin Pharmacol. 1988 Jul;26(1):1-5.

34. Wiholm BE, Olsson S, Moore N, Waller P. In Spontaneous reporting systems outside the US. Strom, editor. Pharmacoepidemiology. Second edition ed. New York: Churchill Livingstone; 1994. p. 139-55.

35. Naranjo CA, Busto U, Sellers EM, Sandor P, Ruiz I, Roberts EA, et al. A method for estimating the probability of adverse drug reactions. Clin Pharmacol Ther. 1981 Aug;30(2):239-45.

36. Meyboom RH, Hekster YA, Egberts AC, Gribnau FW, Edwards IR. Causal or casual? The role of causality assessment in pharmacovigilance. Drug Saf. 1997 Dec;17(6):374-89.

37. Bégaud B, Evreux JC, Jouglard J, Lagier G. Imputation of the unexpected or toxic effects of drugs. Actualization of the method used in France. Therapie. 1985;40(2):111-8.

42

38. Homepage of U.S. Food and Drug Administration. Accessed 17 August 2011 Available from: http://www.fda.gov.

39. EudraVigilance: Background information (FAQs). Accessed 17 August 2011 Available from: http://eudravigilance.ema.europa.eu/human/EVBackground(FAQ).asp.

40. Edwards IR, Olsson S. In The WHO International Drug Monitoring Programme. Aronson JK, editor. Side Effects of Drugs, Annual 25: Elsevier Science B.V.; 2002. p. 589-98.

41. Lindquist M. VigiBase, the WHO Global ICSR Database System: Basic facts. Drug Inf J. 2008;42(5):409-19.

42. Pariente A, Gregoire F, Fourrier-Reglat A, Haramburu F, Moore N. Impact of safety alerts on measures of disproportionality in spontaneous reporting databases: the notoriety bias. Drug Saf. 2007;30(10):891-8.

43. Norén GN, Orre R, Bate A, Edwards IR. Duplicate detection in adverse drug reaction surveillance. Data Mining and Knowledge Discovery. 2007;14(3):305-28.

44. Venulet J, Helling-Borda M. WHO's international drug monitoring--the formative years, 1968-1975: preparatory, pilot and early operational phases. Drug Saf. 2010 Jul 1;33(7):e1-e23.

45. Bate A. The Use of a Bayesian Confidence Propagation Neural Network in Pharmacovigilance. Umeå: Umeå University; 2003.

46. Ståhl M, Lindquist M, Edwards IR, Brown EG. Introducing triage logic as a new strategy for the detection of signals in the WHO Drug Monitoring Database. Pharmacoepidemiol and Drug Saf. 2004 Jun;13(6):355-63.

47. Lindquist M. Use of triage strategies in the WHO signal-detection process. Drug Saf. 2007 2007;30(7):635-7.

48. Drugdex, Thompson Micromedex database [online]. Available from: http://www.thomsonhc.com.

49. Electronic Medicine Compendium. Datapharm Communications Ltd. Electronic Medicine Compendium. Electronic version; Available from: http://www.medicines.org.uk/.

50. Martindale, Thompson Micromedex database [online]. Available from: http://www.thomsonhc.com.

51. Pirmohamed M, L'E Orme M. In Drug Interactions of Clinical Importance. Davies DM, Ferner RE, de Glanville H, editors. Davies's Textbook of Adverse Drug Reactions. 5 ed. London: Chapman & Hall; 1998.

52. Salassa RM, Bollman JL, Dry TJ. The effect of para-aminobenzoic acid on the metabolism and excretion of salicylate. J Lab Clin Med. 1948 Nov;33(11):1393-401.

53. Blackwell B. Hypertensive crisis due to monoamine-oxidase inhibitors. Lancet. 1963 Oct 26;2(7313):849-50.

54. Blackwell B, Marley E. Interaction between cheese and monoamine-oxidase inhibitors in rats and cats. Lancet. 1964 Mar 7;1(7332):530-1.

55. Natoff IL. Cheese and monoamine oxidase inhibitors. Interaction in anaesthetised cats. Lancet. 1964 Mar 7;1(7332):532-3.

56. Sjöqvist F, Böttiger Y. Historical perspectives: drug interactions - it all began with cheese. J Intern Med. 2010 Dec;268(6):512-5.

57. Dunphy TW. The pharmacist's role in the prevention of adverse drug interactions. Am J Hosp Pharm. 1969 Jul;26(7):366-77.

58. Hansten P, Horn J. Drug interactions Analysis and Management.St. Louis. United Sates of America: Wolters Kluwer Health; 2010.

59. Faber KN, Muller M, Jansen PL. Drug Transport Proteins in the Liver. Adv Drug Deliv Rev. 2003;55(1):107-24.

43

60. Sjöqvist F. In Interaktion mellan läkemedel. In:Akademi-FASS. Stockholm: Läkemedelsindustriföreningen; 2008.

61. Rang HP, Dale MM, Ritter JM, Flower RJ. Rang and Dale's Pharmacology. 6: Elsevier; 2007.

62. Johnell K, Klarin I. The relationship between number of drugs and potential drug-drug interactions in the elderly: a study of over 600,000 elderly patients from the Swedish Prescribed Drug Register. Drug Saf. 2007;30(10):911-8.

63. Bergendal L, Friberg A, Schaffrath A. Potential drug--drug interactions in 5,125 mostly elderly out-patients in Gothenburg, Sweden. Pharm World Sci. 1995 Sep 22;17(5):152-7.

64. Björkman IK, Fastbom J, Schmidt IK, Bernsten CB. Drug-drug interactions in the elderly. Ann Pharmacother. 2002 Nov;36(11):1675-81.

65. Jankel CA, Speedie SM. Detecting drug interactions: a review of the literature. DICP. 1990 Oct;24(10):982-9.

66. Rosholm JU, Bjerrum L, Hallas J, Worm J, Gram LF. Polypharmacy and the risk of drug-drug interactions among Danish elderly. A prescription database study. Dan Med Bull. 1998 Apr;45(2):210-3.

67. Hohl CM, Dankoff J, Colacone A, Afilalo M. Polypharmacy, adverse drug-related events, and potential adverse drug interactions in elderly patients presenting to an emergency department. Ann Emerg Med. 2001 Dec;38(6):666-71.

68. Stanton LA, Peterson GM, Rumble RH, Cooper GM, Polack AE. Drug-related admissions to an Australian hospital. J Clin Pharm Ther. 1994 Dec;19(6):341-7.

69. Schellander R, Donnerer J. Antidepressants: clinically relevant drug interactions to be considered. Pharmacology. 2010;86(4):203-15.

70. Leone R, Magro L, Moretti U, Cutroneo P, Moschini M, Motola D, et al. Identifying adverse drug reactions associated with drug-drug interactions: data mining of a spontaneous reporting database in Italy. Drug Saf. 2010 Aug 1;33(8):667-75.

71. Davies EC, Green CF, Taylor S, Williamson PR, Mottram DR, Pirmohamed M. Adverse drug reactions in hospital in-patients: a prospective analysis of 3695 patient-episodes. PLoS One. 2009;4 (2):e4439.

72. Baxter K, Lee A, Stockley IH. In Drug-drug Interactions. van Boxtel CJ, Santoso B, Edwards IR, editors. Drug benefits and risks: international textbook of clinical pharmacology. Amsterdam: IOS Press Inc; 2008.

73. Alexanderson B, Evans DA, Sjöqvist F. Steady-state plasma levels of nortriptyline in twins: influence of genetic factors and drug therapy. Br Med J. 1969 Dec 27;4(5686):764-8.

74. Miranda V, Fede A, Nobuo M, Ayres V, Giglio A, Miranda M, et al. Adverse drug reactions and drug interactions as causes of hospital admission in oncology. J Pain Symptom Manage. 2011 Sep;42(3):342-53.

75. Böttiger Y, Laine K, Andersson ML, Korhonen T, Molin B, Övesjo ML, et al. SFINX-a drug-drug interaction database designed for clinical decision support systems. Eur J Clin Pharmacol. 2009 Jun;65(6):627-33.

76. Garg AX, Adhikari NK, McDonald H, Rosas-Arellano MP, Devereaux PJ, Beyene J, et al. Effects of computerized clinical decision support systems on practitioner performance and patient outcomes: a systematic review. JAMA. 2005 Mar 9;293(10):1223-38.

77. Hedenmalm K, Lindh JD, Sawe J, Rane A. Increased liability of tramadol-warfarin interaction in individuals with mutations in the cytochrome P450 2D6 gene. Eur J Clin Pharmacol. 2004 Jul;60(5):369-72.

78. Yue QY, Bergquist C, Gerden B. Safety of St John's wort (Hypericum perforatum). Lancet. 2000 Feb 12;355(9203):576-7.

44

79. Henderson L, Yue QY, Bergquist C, Gerden B, Arlett P. St John's wort (Hypericum perforatum): drug interactions and clinical outcomes. Br J Clin Pharmacol. 2002;54(4):349-56.

80. Ohlsson S, Holm L, Myrberg O, Sundström A, Yue QY. Noscapine may increase the effect of warfarin. Br J Clin Pharmacol. 2008 Feb;65(2):277-8.

81. Johansson M-J, Sundström A, Strandell J, Wallerstedt SM. Samtliga tetracykliner misstänks förstärka warfarins effekt. Poster på svenska läkaresällskapets riksstämma 28 November 2007.

82. Norén GN, Sundberg R, Bate A, Edwards IR. A statistical methodology for drug-drug interaction surveillance. Stat Med. 2008 Jul 20;27(16):3057-70.

83. Strandell J, Norén GN, Bate A, Edwards IR. Drug- Drug- ADR Screening in Spontaneous Reports as a Tool for Detecting Clustered Reporting and Brand Name Confusion. Pharmacoepidemiol and Drug Saf. 2008;17 supplement 1(ICPE abstract no 38):S17.

84. Horn JR, Hansten PD, Chan LN. Proposal for a new tool to evaluate drug interaction cases. Ann Pharmacother. 2007 Apr;41(4):674-80.

85. MedDRA - the Medical Dictionary for Regulatory Activities. The MSSO - Maintenance and Support Services Organization; Accessed the 28th of May 2010. Available from: http://www.meddramsso.com/.

86. Strandell J, Caster O, Norén GN. Free text extraction from case narratives to highlight suspected drug interactions. Drug Saf. 2010;33(10):916-7.

87. Swedish, Finnish, INteraction X-referencing drug-drug interaction database. Department of clinical pharmacology, Karolinska Institute, Sweden; Drug Management and Informatics in Stockholm County Council, Sweden and medbase Ltd, Finland; Accessed: 2011 March 24. Available from: http://www.janusinfo.se/sfinx/interactions/index_menus.jsp.

88. Böttiger Y, Rane A. In Drug Information. van Boxtel CJ, Santoso B, Edwards IR, editors. Drug benefits and risks: international textbook of clinical pharmacology. Amsterdam: IOS Press Inc; 2008.

89. Stockley's Drug Interactions Alerts (xml). Publisher: Royal Pharmaceutical Society. 90. Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models

via coordinate descent. J Stat Softw. 2010;33(1):1-22. 91. Kogan AD OS. Lovastatin-induced acute rhabdomyolysis. Postgrad Med J.

1990;66(774):294-6. 92. Hansen KE HJ, Ferguson EE, Stein JH. Outcomes in 45 patients with statin-associated

myopathy. Arch Intern Med. 2005;165(22):2671-6. 93. Safety Labeling Changes Approved By FDA Center for Drug Evaluation and Research

(CDER). May 2004. Cited in February-July 2007 Available from: http://www.fda.gov/medwatch/SAFETY/2004/may04.htm.

94. Chen YF, Avery AJ, Neil KE, Johnson C, Dewey ME, Stockley IH. Incidence and possible causes of prescribing potentially hazardous/contraindicated drug combinations in general practice. Drug Saf. 2005;28(1):67-80.

95. Glassman PA, Simon B, Belperio P, Lanto A. Improving recognition of drug interactions: benefits and barriers to using automated drug alerts. Med Care. 2002 Dec;40(12):1161-71.

96. Bjerrum L, Andersen M, Petersen G, Kragstrup J. Exposure to potential drug interactions in primary health care. Scand J Prim Health Care. 2003 Sep;21(3):153-8.

97. Griffin J P, D'Arcy P F. A manual of adverse drug Interactions. Fifth Edition. Amsterdam: Elsvier Science B.V; 1997.

98. Ferner RE, Aronson JK. Communicating information about drug safety. BMJ. 2006 Jul 15;333(7559):143-5.

45

99. Edwards BD, Furlan G. How to apply the human factor to periodic safety update reports. Drug Saf. 2010 Oct 1;33(10):811-20.

100. Johansson K, Olsson S, Hellman B, Meyboom RH. An analysis of Vigimed, a global e-mail system for the exchange of pharmacovigilance information. Drug Saf. 2007;30(10):883-9.

101. Meyboom R. Detecting Adverse Drug Reactions, Pharmacovigilance in Netherlands. Nijmegen1998.

102. Yue QY, Strandell J, Myrberg O. Concomitant Use of Glucosamine Potentiates the Effect of Warfarin. Drug Saf. 2006;29(10):911-1010.

103. Gony M, Badie K, Sommet A, Jacquot J, Baudrin D, Gauthier P, et al. Improving adverse drug reaction reporting in hospitals: results of the French Pharmacovigilance in Midi-Pyrenees region (PharmacoMIP) network 2-year pilot study. Drug Saf. 2010 May 1;33(5):409-16.

104. McAdams M, Staffa J, Dal Pan G. Estimating the extent of reporting to FDA: a case study of statin-associated rhabdomyolysis. Pharmacoepidemiol Drug Saf. 2008;17(3):229-39.

105. Pariente A, Gregoire F, Fourrier-Reglat A, Haramburu F, Moore N. Impact of safety alerts on measures of disproportionality in spontaneous reporting databases: the notoriety bias. Drug Saf. 2007;30(10):891-8.

106. Härmark L, Puijenbroek E, Grootheest K. Longitudinal monitoring of the safety of drugs by using a web-based system: the case of pregabalin. Pharmacoepidemiol and Drug Saf. 2011 Jun;20(6):591-7.

107. Maignen F, Hauben M, Tsintis P. Modelling the time to onset of adverse reactions with parametric survival distributions: a potential approach to signal detection and evaluation. Drug Saf. 2010 May 1;33(5):417-34.

108. Khodabakhshi G, Star K, Norén GN, Hägg S. Reported time-to-onset for adverse drug reactions – the impact of duration of treatment and confirmation bias. Manuscript submitted to Drug Safety. 2011.

109. Norén GN, Bate A, Hopstadius J, Star K, Edwards IR, editors. Temporal pattern discovery for trends and transient effects: its application to patient records. ACM SIGKDD international Conference on Knowledge Discovery and Data Mining; 2008 August 24 - 27, 2008; Las Vegas, Nevada, USA. 259: KDD '08. ACM.

110. Star K, Bate A, Meyboom RH, Edwards IR. Pneumonia following antipsychotic prescriptions in electronic health records: a patient safety concern? Br J Gen Pract. 2010 October 2010;60(579):385-94.

111. Tatonetti NP, Denny JC, Murphy SN, Fernald GH, Krishnan G, Castro V, et al. Detecting drug interactions from adverse-event reports: interaction between paroxetine and pravastatin increases blood glucose levels. Clin Pharmacol Ther. 2011 Jul;90(1):133-42.

112. Åstrand E, Åstrand B, Antonov K, Petersson G. Potential drug interactions during a three-decade study period: a cross-sectional study of a prescription register. Eur J Clin Pharmacol. 2007 Sep;63(9):851-9.

113. Haider SI, Johnell K, Weitoft GR, Thorslund M, Fastbom J. The influence of educational level on polypharmacy and inappropriate drug use: a register-based study of more than 600,000 older people. J Am Geriatr Soc. 2009 Jan;57(1):62-9.

114. Ljung R, Lu Y, Lagergren J. High concomitant use of interacting drugs and low use of gastroprotective drugs among NSAID users in an unselected elderly population: a nationwide register-based study. Drugs Aging. 2011 Jun 1;28(6):469-76.