When a Randomised Controlled Trial is Needed to Assess Drug Safety

COMMENTARY Drug Safety 13 (I): 15-24, 1995 o 1 1 4-59 1 6(95(OOO7-{)() 15($0500(0

© Adis International Limited. All rights reserved.

When a Randomised Controlled Trial is Needed to Assess Drug Safety The Case of Paediatric Ibuprofen

Allen A. Mitchell and Samuel M. Lesko Slone Epidemiology Unit, School of Public Health, Boston University School of Medicine, Brookline, Massachusetts, USA

Contents Summary , , , , , , , , , , , , , , , 1, Sources of Drug Safety Information

1 ,1 Pre marketing Studies , 1.2 Spontaneous Reports 1,3 Observational Studies

2, Issues Related to Paediatric Ibuprofen 2,1 Adverse Reactions , , , , , 2,2 How Is Paediatric Ibuprofen Used? , 2,3 Physician Survey """,""

3, The Advantages Of A Randomised Controlled Trial 3,1 Randomisation , 3,2 Comparison Treatment '" 3,3 Blinding, " """'" 3,4 Study Population and Setting 3,5 Exposure of Interest, 3,6 Outcomes of Interest, , , , 3,7 Sample Size Considerations

4, Feasibility And Design , , 4,1 Risk Estimation , , , , 4.2 Advisory Committee 4,3 Enrolled Sample,

5, Limitations , 6, Conclusion , , , , , ,

15 16 16 16 17 17 17 18 18 19 19 19 19 20 20 20 20 21 21 22 22 22 22

Summary Drugs are frequently made available for use before risks of rare but serious reactions have been identified and quantified. While this situation may be acceptable for drugs used to treat serious conditions, greater information on safety is needed for drugs used to treat less serious conditions, and particularly those medications available without prescription. Spontaneous reports and observational studies can provide useful data in most instances, but nonrandomised studies are inadequate in the presence of confounding by indication (i.e. when patients treated with a drug differ in their underlying risk of adverse outcome from patients given alternate treatments, independent of the effect of the drug). Such is the case in the US with regard to the use of paediatric ibuprofen as an antipyretic. In this

16 Mitchell & Lesko

setting, a rigorous and large randomised controlled trial is needed to provide valid and statistically stable risk estimates. A trial of this kind is a feasible way to develop clinically meaningful data on safety with respect to rare but serious adverse reactions.

The practice of rational drug therapy requires knowledge of not only the benefits of individual therapeutic agents, but also their risks. Although it would be ideal to know the nature and frequency of all adverse effects of a drug prior to its widespread use, we recognise that this ideal cannot be achieved. In theory, our demand for information on safety varies according to certain concerns. For example, we are willing to accept the possibility of relatively high rates of serious reactions for drugs used to treat life-threatening conditions, such as cancer or AIDS, but such risks are unacceptable for drugs used to treat less serious conditions, such as mild anxiety or fever. In practice, however, drugs may become available before their risks are well described; this may occur because of unusual demand for new therapeutic agents (e.g. for the treatment of AIDS) or because the assessment of safety is constrained by ethical or practical limitations (e.g. fetal hazards from drugs given to pregnant women; large sample sizes needed to detect rare reactions).

Unfortunately, instances in which serious adverse reactions were discovered only after a drug had come into widespread use[I-7) reinforce the need to obtain the best possible information on the safety of a drug prior to its use by large populations. This issue is particularly relevant in the context of recent efforts in the US to switch drugs from prescription to over-the-counter (OTe) availability. Because OTe use does not, by definition, require a physician's involvement, drugs made available in this way are typically intended to treat relatively benign and selflimiting conditions and are assumed to meet much more stringent standards of safety than prescription drugs. The occurrence of rare but serious reactions to such drugs can seriously affect the risk-benefit ratio, and it is therefore important to identify these risks prior to OTe use.

We recently dealt with the question of how to identify rare but serious reactions to paediatric

© Adls Internatlonal Limited. All rights reserved.

ibuprofen. In doing so, we assessed available sources of drug safety information, considered issues related to this particular drug, and then designed and implemented a study to provide valid and meaningful estimates of the risks of rare but serious reactions following use of ibuprofen for the treatment of fever in children. The findings of the study[8) provide insight into those risks. The rationale for our approach has wider implications, and is described in this report.

1. Sources of Drug Safety Information

Information on drug safety derives from a number of sources.[9)

1.1 Premarketing Studies

These are designed to assure that a drug is effective for a particular indication and does not pose major threats to health. These are usually clinical trials and involve relatively few patients who meet selected criteria. Premarketing studies may be adequate in size to detect common adverse effects, but they cannot estimate the risk of rare reactions, whether minor or serious.

1.2 Spontaneous Reports

Information on the safety of marketed drugs is frequently provided by spontaneous reports of adverse reactions submitted to manufacturers, regulatory agencies, or published in the medical literature, and such reports have been very useful in identifying rare but serious reactions to a number of medications. [10) The major limitation of spontaneous reports is the absence of information on the number of exposed patients, making it difficult to estimate rates of adverse events. Even if this number were identifiable, only a small (but unknown) proportion of adverse effects are reported, leading to underestimation of the true rates of adversity.

Drug Safely 13 (l) 1995

Randomised Controlled Trials in Assessing Drug Safety

Furthermore, spontaneous reports are subject to both bias and misclassification; that is, events not causally related to the drug may be described while true adverse reactions may not be recognised. This phenomenon is complicated by relatively aggressive reporting by physicians and others when a drug is newly introduced, followed by diminished reporting once use becomes established. It is often difficult to distinguish whether a reported event represents a newly identified drug effect or something else - either a condition associated with the illness for which the drug was given or a chance event unrelated to the drug or the illness.

Spontaneous reports are most useful when they yield a cluster of similar events shortly following exposure to a drug used to treat a condition which itself is not associated with such events (e.g. fatal cholestatic jaundice among patients treated with benoxaprofen for arthritis[lll). However, spontaneous reports are of questionable value when the reported events may be related either to the drug or the underlying condition (e.g. myocardial infarction among patients being treated with an antihypertensive medication; complications of febrile illnesses among patients treated with an antipyretic).

1.3 ObseNational Studies

Observational studies (typically, cohort or casecontrol designs) provide considerable information on drug safety. Their respective strengths and weaknesses have been reviewed elsewhere,ll2l but a major concern is the availability of information on potential biases and the rigor with which such information is considered. Some biases, such as selection or information bias, usually can be avoided or minimised by careful design; others, such as confounding, can (and must) be taken into account by proper analysis of the data. For example, in assessing the safety of phenylpropanolamine in relation to cerebrovascular disease, one must know and take into account bodyweight, since obesity increases the likelihood of exposure to the drug and at the same time increases the likelihood of stroke.

© Adis International limited. All rights reserved.

17

While many concerns about confounding can be met in the design or analysis of a study, there are situations where confounding cannot be adequately controlled. For example, a confounder may be incompletely or inadequately measured (e.g. education as a measure of socioeconomic status; temperature as an index of severity of an illness). Of particular concern is 'confounding by indication', which occurs when patients who receive a given treatment differ in their risk of adverse outcome from those receiving alternative treatments, independent of the treatment received.l 13 ,14l For example, patients with metastatic cancer who are treated with a new chemotherapeutic agent are, because of their advanced stage of illness, more likely to die or experience nonfatal adverse events than patients with less advanced forms of the same cancer, a risk that is unrelated to the new agent with which they are treated. In a situation where confounding by indication is likely to exist, observational studies cannot generally provide valid estimates of risk.

2. Issues Related to Paediatric Ibuprofen

Ibuprofen is a nonsteroidal anti-inflammatory drug (NSAID) that has been widely used in adult popUlations, first as a prescription drug and then as an OTC drug. Just as ibuprofen was first made available for use in adults by prescription only, so too was ibuprofen for use in the paediatric population. Based on clinical trials in children, in 1989 the US Food and Drug Administration made a paediatric preparation of ibuprofen available, by prescription only. When over 2 million prescriptions had been written for paediatric ibuprofen, some proposed that the medication be switched to OTC availability for the treatment of fever in children. The major concern related to such a switch was the safety of the drug.

2.1 Adverse Reactions

When ibuprofen was first made available for adult use, relatively common adverse effects were well described, particularly among those who had taken it for long periods of time or at high doses.

Drug Safety 13 (1) 1995

18

These included gastrointestinal complaints (e.g. nausea and abdominal pain), dizziness and rash. As more experience with this and other NSAIDs began to accumulate, additional reactions were documented. These appeared to occur only rarely, and included major upper gastrointestinal bleeding,[IS) acute renal failure and anaphylaxis.[16)

Once available OTC, paediatric ibuprofen would become more widely used for the treatment of fever, which is typically a minor and self-limiting condition. Because it is an antipyretic, we should require greater assurance about the safety of this drug than we demand for an agent used to treat more serious conditions. (The larger issue of whether antipyretics should be used in children does not bear directly on the approach to assessing the safety of this or other drugs.) As for any drug, we accept the occurrence of minor reactions (e.g. nausea and dizziness), but we would not accept the common occurrence of moderate to severe adverse reactions (e.g. a 10% risk of gastrointestinal bleeding for a typical episode of use).

The occurrence of rare but serious reactions (e.g. anaphylaxis in 1 per 1000 exposed children) would have important implications for an OTC drug used to treat febrile children, yet meaningful information on such risks is not available for paediatric ibuprofen. Because we cannot assume that children will have the same adverse reactions to ibuprofen as adults (e.g. major gastrointestinal bleeding, acute renal failure and anaphylaxis), and because children may experience adverse reactions not observed in adults,D7) we need information on the risks of all rare but serious reactions that might affect children treated with this drug. Although the need for such data is clear, the method by which they can be obtained is less clear.

A reaction that occurs only rarely is unlikely to be discovered in the relatively small premarketing clinical trials used to assess the effectiveness and general safety of a medication. A trial involving 100 children, for example, can reasonably be expected to identify adverse reactions (such as epigastric pain) that affect 3 to 5 % of patients, but not adverse reactions that affect 1 in 1000 children.

© Adis Intematlonol Limited, All rights reserved,

Mitchell & Lesko

Thus, the experience typically provided by premarketing studies is inadequate to provide risk estimates for rare reactions, and this is the case for paediatric ibuprofen. Such information generally derives from experience gathered after a drug becomes available for general use.

2.2 How Is Paediatric Ibuprofen Used?

The vast majority of febrile children in the US receive no antipyretics or are treated with paracetamol (acetaminophen), which has been available for decades without prescription and is the most commonly used antipyretic in children. Because paediatric ibuprofen requires a prescription, its use requires contact with a physician. For the treatment of fever in children, therefore, the drug would most likely be used in hospitalised patients and among those with a febrile illness severe enough to warrant a visit (or phone call) to the doctor.

The use of paediatric ibuprofen would therefore be reserved for children whose disease is relatively severe, or whose fever is high or unresponsive to paracetamol. Such children may experience relatively high rates of serious sequelae related to the febrile illnesses themselves (or perhaps to their treatment with drugs other than antipyretics). In this setting, higher rates of adverse events might be expected among ibuprofen-treated children than among children treated with paracetamol or no antipyretics, but those rates would be unrelated to ibuprofen itself. This phenomenon is an example of confounding by indication.l13,14)

2.3 Physician SUNey

To test the hypothesis that children treated with ibuprofen differ in the nature of their febrile illnesses from children treated with alternative therapies, we surveyed physicians' practices with regard to their treatment of fever in children (see appendix A). The findings support our a priori hypothesis that practitioners consider the prescription drug paediatric ibuprofen a second-line antipyretic, reserving it to treat febrile illnesses associated with higher or more resistant or persistent fevers than

Drug Sofely 13 (1) 1995


those they treat with paracetamol (or no antipyretics).

Severe illness or its treatment with drugs other than antipyretics may be associated with various adverse events independent of the effects of a given antipyretic, and such confounding by indication must be considered in the assessment of the safety of paediatric ibuprofen.

3. The Advantages Of A Randomised Controlled Trial

For the evaluation of drug effects in human populations, the randomised controlled trial (RCT) is generally considered to be the 'gold standard'. It is typically used to assess efficacy because it assures that the observed benefits of the experimental therapy are due to the drug and not to differences in the nature of the illness or other factors. The RCT also represents a 'gold standard' for the assessment of safety, since it assures, with appropriate consideration given to the role of chance, that observed differences in rates of adverse events are not related to the patient's illness, but rather to the medication itself. RCTs of ibuprofen have provided information on common adverse effects of therapy, usually as part of studies designed to assess efficacy.f18-201 In contrast, they are not generally considered feasible for the identification of rare reactions because of the very large sample sizes that are required. However, when assessment of safety is complicated by the possibility of confounding by indication, the appropriateness and feasibility of an RCT must be considered.

3.1 Randomisation

Confounding by indication seriously complicates interpretation of spontaneous reports related to an antipyretic, since there is no satisfactory way to distinguish adverse events that are due to the drug itself from those due to other aspects of the febrile illnesses for which the drug was prescribed. Similarly, observational studies cannot provide valid estimates of adverse reactions attributable to ibuprofen because they cannot assure that children treated with ibuprofen and those treated with an

© Adis International Limited. All rights reserved.

19

alternate therapy had underlying illnesses of equivalent nature and severity. Attempts to control for severity of illness in the analysis are constrained by our inability to identify factors (many of which are subjective) that relate both to the severity of a child's illness and the likelihood of the physician prescribing ibuprofen. For these reasons, observational studies cannot assure that any risk differences that might be observed are due to ibuprofen and not to the differences in the children's febrile illnesses.

The only way to eliminate the possibility of confounding by indication is to assure that children who receive ibuprofen and those who do not receive the drug have the same baseline risk of experiencing a subsequent adverse event. Given that physicians use paediatric ibuprofen differently from other antipyretics, the only design that can provide unbiased data is a study in which children with febrile illnesses are randomly assigned to treatment with either ibuprofen or an alternative therapy.

3.2 Comparison Treatment

An RCT involves a comparison of a treatment against a control, which may be an alternative treatment or placebo. For paediatric ibuprofen, the comparison treatment should be the most widely used OTC antipyretic, paracetamol, since parents who would choose OTC ibuprofen (were it to become available) are likely to be the ones who currently use paracetamol. From a practical standpoint, physicians and parents can be expected to be more willing to participate in a study if they know the child will receive an efficacious agent (irrespective of study arm), rather than a placebo.

3.3 Blinding

To eliminate the possibility that adverse outcomes will be detected differentially among ibuprofenand paracetamol-exposed children (i.e. detection bias), the study must be double-blinded (i.e. neither the physician nor the parent/patient know which treatment is given). If physicians know which medication a patient has been given, the information they provide to parents may differ accordingly. For


20

example, general concerns about the safety of the relatively new drug (ibuprofen) may prompt physicians to advise parents of children given this drug to be particularly vigilant for signs of possible adverse effects (e.g. gastrointestinal symptoms). Furthermore, physicians may more aggressively manage and evaluate events that occur in ibuprofentreated children (e.g. a child with dark-coloured stools may be hospitalised and evaluated for gastrointestinal bleeding if treated with ibuprofen but not with paracetamol). Children and their parents must also be blind with respect to exposure, since parents who know that their children were treated with the 'new study drug' (ibuprofen) rather than the 'standard fever drug' (paracetamol) may monitor them more carefully and more likely bring them to the doctor's office or emergency department for any problem they observe.

3.4 Study Population and Setting

While most RCTs are conducted among highly selected patients recruited from specialised settings, such restrictions are not an inherent requirement (or limitation) of this study design. In fact, because concern about ibuprofen relates to its safety in the OTC setting, attempts should be made to approximate that setting by studying ambulatory children with febrile illnesses. Since ibuprofen is currently available by prescription only, the closest approximation to the 'real world' setting of OTC antipyretic use is the office-based physician who treats ambulatory children presenting with a wide variety of febrile illnesses treated in various ways, both recommended (or prescribed) by the physician and initiated by the parent. Issues of compliance and overdose, which are directly relevant to the safety of a drug as it will be used by parents, can also be studied.

3.5 Exposure of Interest

Since the current recommended dose of ibuprofen for the treatment of fever in children is 5 or 10 mg/kg per dose (depending on the child's temperature), the study should assess the safety of both doses; the comparison drug, paracetamol, should

© Adis Interna~onal Umited. All rights reserved.

Mitchell & Lesko

be given at the usual recommended dose of 10 to 15 mg/kg per dose. To mimic the 'real world', multiple doses should be given (see Section 4).

3.6 Outcomes of Interest

The prospective nature of an RCT permits study of a wide variety of outcomes. Minor adverse effects are not sufficiently important to warrant study; those that occur commonly have been described in clinical trials designed to assess efficacy,[18-201 and those that occur rarely would be impractical to document and are of little clinical importance. Rather, the outcomes of greatest concern are serious events, which can be defined as clinical conditions that result in a child being admitted to hospital.

Information on hospital admissions among patients in an RCT permits assessment of the risk of adverse effects which are suspected to occur among ibuprofen-treated children (hypothesis testing), as well as identification of adverse effects which may be associated with the drug but were not suspected in advance (hypothesis generation). The former include those rare but serious reactions that have been associated with ibuprofen use in adults: major gastrointestinal bleeding, acute renal failure and anaphylaxis. With regard to hypothesis generation, it is equally important to identify previously unidentified rare but serious reactions which may be due to the use of ibuprofen in children; included among these would be Reye's syndrome, which is strongly associated with exposure to another NSAID, aspirin (acetylsalicylic acid)J21 1

3.7 Sample Size Considerations

On theoretical grounds, it would be desirable to rule out even small increases in risk for any serious adverse event among ibuprofen-exposed children; however, it is not feasible to demonstrate absolute safety. For relatively common events, such as those that occur in about 2% of unexposed patients, small increases in risk (e.g. a relative risk of 2) can be identified with samples of about 1000 exposed and 1000 unexposed children. To detect a doubling in



the risk of an event with a background frequency of 1 per 10 000 requires over 100 000 children exposed to each drug. Similar considerations apply to the identification of absolute risks.

There are no generally accepted standards for what level of risk is compatible with 'safety'. Such judgements are necessarily complex, taking into account the indication for the drug, the nature of the adverse effect, and alternative therapies. An ReT designed to assess safety, like any other study, must balance the ideal and practical; for rare events, it should provide risk estimates from the largest population of participants that realistically can be studied.

4. Feasibility And Design

Based on the considerations described above, we designed and conducted an ReT to assess the risk of rare but serious reactions among febrile children treated with ibuprofen (see Appendix B).[8]

Given that an ReT of the kind described has not previously been conducted, it was difficult to determine how large a study should and could be mounted. We believed that physicians in practice had an interest in participating in office-based research, particularly if it involved illnesses and therapies they routinely encountered, and if it represented rigorous science, not product promotion. However, we also recognised that participation would be enhanced if physicians were comfortable with the clinical and ethical aspects of the protocol and if the study conformed to their office routine and time constraints. We believed that parents would be comfortable with the study design if they knew their child was receiving 1 of 2 medications with which they themselves were familiar, and would be provided sufficient doses to treat a typical febrile course.

To simplify participation, we asked physicianinvestigators to describe the study to the parents of eligible patients and obtain appropriate consent. The physician then dispensed a bottle containing 100ml of 1 of 3 suspensions (ibuprofen 50 mg/5ml, ibuprofen 100 mg/5ml or paracetamol120 mg/5ml), all identical in colour and taste (and distinct from

© Adis International Umited. All rights reseNed.

21

ibuprofen and paracetamol liquids then available in the US). With the exception of a unique numeric code, the bottle labels were identical, with a single dosage regimen (1 dose, based on the child's weight, every 4 to 6 hours, when required) and a toll-free number to call in case of overdose or other need to break the code. Other than restricting antipyretic medication to the study drugs, physicians and parents were free to use additional therapies for the febrile illness, as they wished.

The responsibility for follow-up was ours, and involved a questionnaire mailed to the parent 1 month after enrolment. Information was sought on the current condition of the child, with particular attention to descriptions of hospitalisations or visits to a doctor or emergency room in the month following enrolment. Parents failing to respond to mailed questionnaires were contacted by telephone, and the prescribing doctor was called only in cases in which telephone contact failed.

Based largely on a 'best guess', we estimated that we would be able to recruit 1000 to 2000 officebased physicians, who in tum would enrol a maximum of 75 000 children with febrile illnesses (25000 in each of the 3 treatment arms). We anticipated that accrual of a study population of this size would represent a massive undertaking; at the same time, a sample of 75 000 would provide a study large enough to detect meaningful increases in the risk of rare reactions.

4.1 Risk Estimation

Based on comparisons of 25 000 children treated with either dosage of ibuprofen and 25 000 treated with paracetamol, a power of 0.8 and an a (1-tailed) of 0.05, the study could detect a relative risk of 1.8 for an event occurring among 1 per 1000 paracetamol-exposed children; a risk of 4.4 for an event affecting 1 per 10 000; and 20 for an event affecting 1 per 100 000. (It is of interestto note that for an event as rare as Reye's syndrome, the risk would have to be in the range of 200-fold to be detected even in so large a study.) If the rate of a given adverse event did not differ among children exposed to either dosage of ibuprofen, the combined


22

experience of all ibuprofen-exposed children would increase the power of the study and would permit detection of lower relative risks.

Apart from estimating the risk of a given event among ibuprofen-exposed children relative to the risk in those exposed to paracetamol, an RCT also permits estimation of the absolute risks (and corresponding 95% confidence intervals) of adverse events among children exposed to ibuprofen. If, despite the study size, no cases of acute renal failure were observed among the 25 000 exposed to ibuprofen 5 mg/kg, (i.e. an observed risk of a per 25 000), the maximum likely absolute risk would be estimated by the upper 95% confidence bound, which in this case would be approximately 12 per 100 000 children. If no cases of this event were observed among all ibuprofen-exposed children, the upper 95% confidence bound would be approximately 6 per 100 000.

4.2 Advisory Committee

To provide independent guidance to the study, we recruited an Advisory Committee composed of experts in paediatric pharmacology, epidemiology and various subspecialities (see acknowledgments). The Advisory Committee also served as the data monitoring committee.

4.3 Enrolled Sample

The RCT was completed over a period of 28 months.l8] Critical elements of the design were fulfilled: over 84 000 children were enrolled, and follow-up was achieved for over 95% of these; patients had illnesses typical of those in the outpatient environment, and the children were randomly distributed with respect to demographic characteristics, illnesses and other factors. Blinding was maintained, and follow-up was similar across treatment groups.

5. Limitations

Although the study provided risk estimates that were both valid and statistically stable, even so large a study could only provide evidence of relative

© Adis Interna~onal Limited. All rights reserved.

Mitchell & Lesko

safety (for example, it could not rule out a 15-fo1d increase in risk for a reaction with a baseline frequency of 1 per 100 000). It should be recognised, however, that this limitation is a function of the sample size, not study design. While larger samples might be achieved with a different design (such as an observational or other nonrandomised study), the validity of estimates derived from the RCT in this situation is of greater scientific and clinical importance than the increased statistical power that would derive from a larger but less rigorous study.

Apart from the cost of such an effort, it should be recognised that our study was conducted under an unusual set of circumstances: the study drug was available by prescription but not OTC, a 'standard' comparison drug for the same indication was available without prescription, both drugs were familiar to doctors and parents, treatment was relatively brief, and practitioners were sufficiently interested in the study question to participate in large numbers. Whether a study of similar design and size could be carried out in other settings remains unclear.

6. Conclusion

In situations where assessment of a drug's safety is complicated by confounding by indication, nonrandomised studies generally cannot be relied upon to provide valid estimates of the risk of adverse effects. Under selected circumstances, it is possible to conduct a rigorous RCT large enough to provide estimates of the risk of rare but serious reactions that are valid, statistically stable and clinically meaningful. While complex and expensive, such approaches are important to consider in assessing the appropriateness of switching certain drugs from prescription to OTC status.

Appendix A

Physician Survey: We sought to contact approximately 200 physicians, and identified a systematic, nationally representative sample of 401 paediatricians and family practitioners listed in the 1992 directories of the American Academy of Pediatrics and the American Academy of Family

Drug Safely 13 (1) 1995


Physicians, respectively. During the summer of 1992, these physicians were sent a letter indicating that we were conducting a survey of 'current practices used in the management of febrile children'. Using a standardised questionnaire, an interviewer systematically telephoned the targeted offices until contact was made with approximately 200. A total of 207 physicians' offices were successfully contacted (of physicians not reached, 34% had incorrect telephone numbers, 10% were out of the office, and the remainder were unavailable); 44 of the 207 were excluded because they did not treat febrile paediatric patients, leaving a sample of 163 eligible physicians. Of these, 54 (33%) refused participation and 1 could not be interviewed because of language difficulties.

The final sample of 108 physicians consisted of 61 paediatricians and 47 family practitioners. 73 and 26% were in group and solo practice, respectively; they had been in practice a median of 12 years (10 and 90 percentiles were 2 and 26 years, respectively), and they saw a median of 60 paediatric patients per week (18 and 150).

Respondents were asked about their management of fever in children, including the circumstances under which they treated (e.g. age, height oftemperature), the therapies they used (e.g. paracetamol, ibuprofen, aspirin, tepid baths/sponging), and the reasons for using various antipyretics (e.g. cost, height of temperature).

All physicians indicated that they had used paracetamol in the last year, and 83% reported that they had used ibuprofen. The average minimal age and temperature for which physicians recommended paracetamol was 3.7 months and lOLl OF, respectively, whereas the equivalent values for ibuprofen were 11.4 months and 102.2°F; these differences are statistically significant (both at p <0.0001; Wilcoxon signed-rank test). When asked the open-ended question 'Under what circumstances do you prescribe ibuprofen for the control of fever in children?', 51 % volunteered that they used ibuprofen when paracetamol failed; no physician reported using paracetamol when ibuprofen failed.

© Adis International limited. All rights reserved.

23

Assessment of these results requires consideration of whether they are likely to be representative. By restricting the survey to the first 200 physicians with whom we made contact, we may have excluded those that differed in their treatment of fever. However, the varied reasons for failed contact (e.g. wrong number listed in directory, disconnected phone, repeated busy signal, discontinued practice) make this possibility unlikely. The 33% who refused participation did so without knowledge of the hypothesis being tested and the majority of refusals were from the physician's office staff who stated that the doctor did not participate in surveys. Thus, we believe the practices participating in this survey are likely to be representative.

Appendix B

Study Fiudiugs:[8] Overall, 84 192 eligible children were enrolled, and 277 were lost to follow-up, giving a final sample of 83 915 children. 1 % in each treatment arm were hospitalised, primarily for infectious diseases. Four children had diagnoses of acute, nonmajor gastrointestinal bleeding (2 in each ibuprofen group), and the risk of this event among ibuprofen-treated children was 7.2 per 100000 (95% confidence interval, 2 to 18 per 100 000; not significantly different from the risk among paracetamol-treated children). There were no hospitalisations for acute renal failure, anaphylaxis or Reye's syndrome; the upper 95% confidence bound for the risk of each of these outcomes was 5.4 per 100 000 ibuprofen-treated children. Thus, the risk of hospitalisation for the study outcomes was not increased following short term use of ibuprofen in febrile children. These data, however, provide no information on the risks of less severe outcomes or the risks of prolonged ibuprofen use.

Acknowledgements

This work was supported by McNeil Consumer Products Company, Fort Washington, PA, USA.

We thank the study Advisory Committee for their valuable advice and guidance throughout the study: Ralph E. Kauffman, (Chair), Director, Divisions of Clinical Pharma-


24

cologyffoxicology, Children's Hospital of Michigan; Michael D. Bailie, Dean, University of Illinois College of Medicine at Peoria; William Gerson, paediatrician in private practice, Burlington, VT; Alan Leichtner, Clinical Director, Division of Gastroenterology and Nutrition, Children's Hospital, Boston; Alan Leviton, Director, Neuroepidemiology Unit, Children's Hospital, Boston, and Frederick H. Lovejoy, Jr., Associate Physician-in-Chief, Children's Hospital, Boston. Liaison to the Advisory Committee: Sumner J. Yaffe, Director, Center for Research for Mothers and Children, NlCHD; Anthony R. Temple, Executive Director, Medical, and Barbara H. Korberly, Director of Medical Affairs, McNeil Consumer Products Company.

We also wish to thank Richard Vezina, who coordinated data collection and supervised the field staff, Deborah Mitchell, who conducted the physician survey, and Samuel Shapiro for guidance and review of the manuscript. We are especially grateful to the more than 1700 physician-investigators without whose participation this study would not have been possible.

References I. Bottiger LE, Westerholm G. Drug-induced blood dyscrasias in

Sweden. BMJ 1973; 3: 339-43 2. Bums LE, Hodgman JE, Cass AB. Fatal circulatory collapse in

premature infants receiving chloramphenicol. N Engl J Med 1959; 261: 1318-21

3. Hazards of non-practolol beta blockers [editorial]. BMJ 1977; I: 529-30

4. Lenz W. Thalidomide and congenital abnormalities. Lancet 1961; 1: 45

5. The International Agranulocytosis and Aplastic Anaemia Study. Risks of agranulocytosis and aplastic anemia: a fIrst report of their relation to drug use with special reference to analgesics. JAMA 1986; 256: 1749-57

6. Levy M. Aspirin use in patients with major upper gastrointestinal bleeding and peptic ulcer disease. N Engl J Med 1974; 290: 1158-62

7. Hurwitz ES, Barrett MJ, Bregman D, et al. Public health service study ofReye's syndrome and medications. JAMA 1987; 257: 1905-11

8. Lesko SM, Mitchell AA. An assessment of the safety of pediatric ibuprofen: a practitioner-based randomized clinical trial. JAMA 1995; 273: 929-33

© Adls Intematlonal Llm~ed. All rights reserved.

Mitchell & Lesko

9. Mitchell AA. Adverse drug effects and drug epidemiology. In: Yaffe SJ, Aranda JV, editors. 2nd ed. Pediatric pharmacology: therapeutic principles in practice. Philadelphia: W.B. Saunders Co., 1992: 552-5

10. Faich GA. Adverse drug-reaction monitoring. N Engl J Med 1986;314: 1589-92

II. Taggart HM, Alderdice 1M. Fatal cholestatic jaundice in elderly patients taking benoxaprofen. BMJ 1982; 284: 1372

12. Strom BL. Study designs available for pharmacoepidemiology studies. In: Strom BL, editor. Pharmacoepidemiology. 2nd ed. Chichester UK: John Wiley & Sons, Ltd. 1994: 15-27

13. Slone D, Shapiro S, Miettinen OS, et al. Drug evaluation after marketing. Ann Intern Med 1979; 90: 257-61

14. Miettinen OS. Efficacy of therapeutic practice: will epidemiology provide the answers? In: Melmon KL, editor. Drug therapeutics - concepts for physicians. New York: ElsevierlNorth Holland, 1980: 201-8

15. Kaufman DW, Kelly JP, Sheehan JE, et aI. Nonsteroidal anti-inflammatory drug use in relation to major upper gastrointestinal bleeding. Clin Pharmacol Ther 1993; 53: 485-94

16. 'Anti-inflammatory, antiallergic, and immunologic drugs' in AMA Drug Evaluations Subscription. Chicago, IL: American Medical Association, 1994; I: 10-4

17. Yaffe SJ, Aranda JV. Introduction and historical perspectives. In: Yaffe SJ, ArandaJV, editors. Pediatric pharmacology: therapeutic principles in practice. 2nd ed. Philadelphia: w.B. Saunders Co., 1992: 3-9

18. Walson IT, Galletta G, Braden NJ, et al. Ibuprofen, acetaminophen, and placebo treatment of febrile children. Clin Pharmacol Ther 1989; 46: 9-17

'19. Wilson IT, Brown RD, Keams GL, et aI. Single-dose, placebocontrolled comparative study of ibuprofen and acetaminophen antipyresis in children. J Pediatr 1991; 119: 803-11

20. Bertin L, Pons G, d' Athis P, et al. Randomized, double-blind, multicenter, controlled trial of ibuprofen versus acetaminophen (paracetamol) and placebo for treatment of symptoms of tonsillitis and pharyngitis in children. J Pediatr 1991; 119: 811-4

21. Hurwitz ES, Nelson DB, Davis C, et aI. National surveillance for Reye syndrome: a fIve-year review. Pediatrics 1982; 70: 895-900

Correspondence and reprints: Dr Allen A. Mitchell, Slone Epidemiology Unit, 1371 Beacon Street, Brookline, MA02146, USA.


Documents

When a Randomised Controlled Trial is Needed to Assess Drug Safety