Nephrol Dial Transplant (2014) 29: 823–832doi: 10.1093/ndt/gft531Advance Access publication 20 January 2014

Original Article

High-performance information search filters for acutekidney injury content in PubMed, Ovid Medline and Embase

Ainslie M. Hildebrand1,2, Arthur V. Iansavichus2, R. Brian Haynes3, Nancy L. Wilczynski3,

Ravindra L. Mehta4, Chirag R. Parikh5 and Amit X. Garg1,2,3,6

1Division of Nephrology, Western University, London, ON, Canada, 2Kidney Clinical Research Unit, London Health Sciences Centre, London,

ON, Canada, 3Department of Epidemiology and Biostatistics, McMaster University, Hamilton, ON, Canada, 4Division of Nephrology,

University of California, San Diego, CA, USA, 5Section of Nephrology, Yale University School of Medicine, VA-CT Healthcare System, and

Program of Applied Translational Research, New Haven, CT, USA and and 6Department of Epidemiology and Biostatistics, Western University,

London, ON, Canada

Correspondence and offprint requests to: Ainslie M. Hildebrand; E-mail: ainslie.hildebrand@londonhospitals.ca

ABSTRACT

Background. We frequently fail to identify articles relevant tothe subject of acute kidney injury (AKI) when searching thelarge bibliographic databases such as PubMed, Ovid Medlineor Embase. To address this issue, we used computer auto-mation to create information search filters to better identifyarticles relevant to AKI in these databases.Methods. We first manually reviewed a sample of 22 992 full-text articles and used prespecified criteria to determinewhether each article contained AKI content or not. In the de-velopment phase (two-thirds of the sample), we developedand tested the performance of >1.3-million unique filters.Filters with high sensitivity and high specificity for the identi-fication of AKI articles were then retested in the validationphase (remaining third of the sample).Results. We succeeded in developing and validating high-performance AKI search filters for each bibliographic databasewith sensitivities and specificities in excess of 90%. Filters opti-mized for sensitivity reached at least 97.2% sensitivity, andfilters optimized for specificity reached at least 99.5% speci-ficity. The filters were complex; for example one PubMed filterincluded >140 terms used in combination, including ‘acutekidney injury’, ‘tubular necrosis’, ‘azotemia’ and ‘ischemicinjury’. In proof-of-concept searches, physicians found morearticles relevant to topics in AKI with the use of the filters.Conclusions. PubMed, Ovid Medline and Embase can befiltered for articles relevant to AKI in a reliable manner. These

high-performance information filters are now available onlineand can be used to better identify AKI content in large biblio-graphic databases.

Keywords: acute kidney injury, Embase, information retrie-val, medical informatics, Medline

INTRODUCTION

Current, high-quality information is a prerequisite for evi-dence-based healthcare [1]. Although much of this infor-mation exists in bibliographic databases such as PubMed,searching these large databases to inform patient care oftenmeans sifting through several thousand non-relevant citationsto uncover a select few articles of interest. Without sufficienttime and skill, physicians and even those who perform sys-tematic reviews may also miss articles relevant to their search[2, 3]. Better methods are needed for identifying literature inbibliographic databases for content areas in which new evi-dence is constantly emerging, and where point-of-care accessto this information is imperative. Such is the case for the topicof acute kidney injury (AKI), a field that has undergone con-siderable changes in concept and nomenclature in recent years(and most recently redefined in international clinical practiceguidelines) [4]. Routinely used search terms such as ‘acuterenal failure’, ‘acute tubular necrosis’ and even ‘acute kidneyinjury’, when used alone, are poorly equipped to deal withthese new definitions, lacking sensitivity for the retrieval of

© The Author 2014. Published by Oxford University Presson behalf of ERA-EDTA. All rights reserved.

823

at Ondokuz M

ayis University on N

ovember 7, 2014

http://ndt.oxfordjournals.org/D

ownloaded from

AKI-relevant content in PubMed and other databases. A sol-ution to the problem is to optimize bibliographic databasesearching for AKI content through the use of search filters.

Search filters are objectively derived using computer auto-mation and use various search terms in combination to restricta bibliographic database such as PubMed to a certain type ofarticle [5, 6]. Methods-based filters, such as those incorporatedinto the PubMed interface in the ‘clinical queries’ section(which use up to eight terms in combination), have beenhighly successful in retrieving articles of diagnosis, etiology,therapy, prognosis and clinical prediction guides filtered forhigh methodological merit [7–10]. Topic-based search filtersoffer a unique opportunity to further focus the search in onecontent area. With relevant literature in nephrology now dis-persed across over 400 journals, search filters such as thosedeveloped to retrieve articles relevant specifically to dialysis,kidney transplantation and glomerular disease are becomingincreasingly important and have proven to maximize the re-trieval of relevant articles, minimize non-relevant articles andincrease the overall precision of each search [11–16]. Thesefilters have used several terms in combination (sometimes upto 65 terms) and account for different ways the same conceptcan be expressed, British and American English spelling, andterm-entering variability (such as a free text word, a medicalsubject heading or a truncated word).

A search filter for AKI would have the same functionality.For example, if users wanted to determine the evidence for therole of N-acetylcysteine in the prevention of contrast-inducednephropathy, they could simply type in the term ‘acetylcys-teine’ and select the AKI filter to perform the search within asubset of articles in an online database that has been prese-lected as relevant to AKI. In this case, the filter acts as an opti-mized substitute for the many phrases commonly used toindex and describe AKI, such as ‘acute renal failure’, ‘acutetubular necrosis’ and ‘contrast nephropathy’, which alone lacksufficient sensitivity to retrieve all relevant articles.

In this report, we describe how we used computer auto-mation to develop high-performance information searchfilters to identify articles relevant to AKI in PubMed, OvidMedline and Ovid Embase bibliographic databases. Thesehigh-performance AKI search filters proved valid whensearching a separate set of articles, and we illustrate howsearches were improved when the filter was used in someproof-of-concept physician searches.

MATERIALS AND METHODS

We used a diagnostic test assessment framework to developand validate search filters for AKI. For the purpose of thisstudy, AKI was defined as a sudden loss of kidney function asdefined by the AKIN or RIFLE criteria or terms commonlyused to describe this sudden loss of kidney function [17, 18].

Sample of articles

We first established the reference standard by manualreview of a subset of full-text articles published in 39 journalsbetween 2004 and 2008. To develop this collection of journals,

we adopted a similar strategy for sampling as published inprior search filter studies. This approach has resulted in filtersthat generalize well over publication years and journal types[16, 19, 20]. We first compiled a list of 466 journals from a listof journals that had published at least one article relevant torenal care from 1961 to 2005 [11]. We then ranked these jour-nals according to the number of articles with relevant infor-mation and selected the top 20 journals available to uselectronically. In addition to this, we selected 19 more journalsat random from the remaining 446 journals. We then ran-domly divided these 39 journals into development and vali-dation sets at a ratio of 2 to 1, respectively.

Article review

For each journal in the development and validation set, wemanually reviewed all full-text articles indexed in PubMed, OvidMedline and Embase in 2006 and randomly selected anadditional 500 full-text articles per year for the remaining yearsbetween 2004 and 2008 to review. These 22 992 articles includedoriginal investigations, reviews, letters and editorials. We deriveda standardized checklist of qualifications and terms to classifyarticles as relevant to AKI from a review of nephrology textbooksand the medical subject heading (MeSH) thesaurus (Supplemen-tary Appendix B). Three readers used this checklist to determine,whether the full. All reviewers were calibrated against a nephrol-ogist in their application of checklist criteria using two test setsof 100 articles (agreement beyond chance, κ = 0.91).

Filters

Using computer automation, we developed unique filtersfor PubMed, Ovid Medline and Embase. We first identifiedthe search terms used in filter development from the followingsources: US National Library of Medicine (NLM) MeSH the-saurus using Medline MeSH browser [21], Medline permutedindex [22], Emtree thesaurus [23], SNOMED clinical terms,nephrology textbooks [24], clinical practice guidelines [25,26], systematic reviews [27–31], website glossaries and clini-cian and librarian opinions. Examples of these terms included‘acute renal failure’, ‘acute dialysis’, ‘rapidly progressive glo-merulonephritis’ and ‘rhabdomyolysis’. We selected MeSHterms with or without major focus and with or withoutadditional subheadings or explosion capability. Major focusrefers to records in which an index term has been tagged asthe major topic of the article. Entering the exploded MeSHterm ‘acute kidney injury’ means ‘acute tubular necrosis’ isalso automatically included in the search. We consideredfree text words as full and truncated terms and accounted forboth American and British English spelling. The inclusion ofmultiple endings was achieved through the use of the $ symbol(i.e. nephrotox$). Terms could appear anywhere in a citationbut not solely in the journal name. We repeated the sameprocess for Embase using Emtree index terms to replace theMeSH terms in PubMed and Ovid Medline.

We automated the process of combining and testing thefilters by using a computer-implemented algorithm. The com-puter combined single-term filters into multiple-term filtersby selectively using the Boolean operators ‘OR’, ‘AND’ and‘NOT’ to maximize sensitivity and specificity. In the

ORIG

INALARTIC

LE

824 A. M. Hildebrand et al.

at Ondokuz M

ayis University on N

ovember 7, 2014

http://ndt.oxfordjournals.org/D

ownloaded from

development set, we then compared the retrieval performanceof various filters (made up of individual and combinations ofsearch terms) with the reference standard from manual review.

Statistical analysis

For each filter, we constructed a two-by-two contingencytable and assessed filter performance by calculating sensitivity,specificity, precision and accuracy, similar to evaluation of adiagnostic test (Table 1). We then selected filters from the de-velopment phase that demonstrated high performance ineither sensitivity or specificity without compromising pre-cision and retested them in the validation set of articles.

Proof-of-concept searches

To illustrate the potential effectiveness of validated filters inPubMed, we selected five independent nephrologists from a di-rectory of Canadian nephrologists provided by the Royal Collegeof Physicians and Surgeons of Canada to provide a search queryfor a unique predetermined clinical question. We formulated fiveclinical questions, each which could be answered by a recent cor-responding systematic review [27–31]. These systematic reviewswere then used as a reference source for relevant articles on thegiven topic. For example the question ‘Is acetylcysteine effectivein preventing contrast-related nephropathy?’ was framed tomatch a systematic review of 15 articles on prevention of contrastnephropathy in adults by Nallamothu et al. [27]. Throughmeans of an electronic survey, we asked each nephrologist to for-mulate a search strategy that they would use in PubMed for thegiven clinical question without knowledge of the search filter inuse. We then applied these searches to the PubMed databasewith and without the filters validated as part of this study. Searchdates were restricted to the date on which the review wasupdated. In each case, we noted the number of relevant articlesidentified in searches with and without the validated filters, com-pared with the reference standard, which in this case was the setof relevant articles as determined by each systematic review.

RESULTS

Sample of articles

We used 22 992 full-text articles from 39 journals (Sup-plementary Appendix A). In total, 21 300 articles contributed

to the PubMed set, while 21 280 and 22 158 articles contribu-ted to the Ovid Medline and Embase set, respectively. We as-signed 14 619 articles to the development set and 8373 articlesto the validation set. Of the 22 992 full-text articles included, atotal of 386 (1.7%) contained AKI content, 286 of these articleswere in the development set and the remaining 100 articleswere in the validation set.

High-performance AKI search filters

Using computer automation in the development set ofarticles, we tested a total of 1 370 506 different multiple-termfilters across all three databases. These filters included combi-nations of index terms specific to each database to help ident-ify AKI content. The filters that operated best were complex;for example one PubMed filter included over 140 terms usedin combination, including ‘acute kidney injury’, ‘tubular ne-crosis’, ‘azotemia’ and ‘ischemic injury’. Our best performingfilters for PubMed, Ovid Medline and Embase are shown inTable 2, categorized by high sensitivity and high specificity.High-performance filters in the development set achieved89.9–98.5% sensitivity, 94.3–99.5% specificity, 29.8–81.8%precision and 94.3–99.3% accuracy. Filters optimized for sen-sitivity achieved 97.2–98.5% sensitivity, and filters optimizedfor specificity achieved 99.5% specificity in the developmentset (Table 2).

The performance of these filters was consistent in the vali-dation set. Filters in the validation set achieved 82.5–96.1%sensitivity, 94.7–99.2% specificity and 94.7–99.0% accuracy;however, the precision dropped to 19.8–57.4%. Filters opti-mized for sensitivity achieved 94.6–96.1% sensitivity, andfilters optimized for specificity achieved 99.1–99.2% specificityin the validation set (Table 2).

To put the performance of these multiple-term filters intocontext, we also compared these results to 261 255 simplersingle-term filters. Single-term filters with the highest sensi-tivity achieved 87.4–94.9% sensitivity, 79.7–84.8% specificityand 80.1–84.9% accuracy; however, precision was low at 9.9–11.8% and terms appeared less relevant. For example thesingle-term filter with the highest sensitivity in PubMed was‘nephropathy’. Single-term filters with the highest specificityachieved 99.6–99.9% specificity, 74.2–92.5% precision and98.4–98.8% accuracy; however, sensitivity was low at 50.6–50.7% (when restricted to at least 50% sensitivity). In terms ofachieving an optimal balance of sensitivity and specificity, ourbest performing multiple-term filters outperformed the single-term filters.

Proof-of-concept searches

We selected five systematic reviews to identify relevantarticles for five AKI questions, each of which was posed to anindependent nephrologist [27–31]. The number of includedstudies per systematic review ranged from 10 to 51. The searchstrategies for PubMed provided by the nephrologists to ident-ify these articles included: ‘Acute Kidney Injury’ [MeSH]AND acetylcysteine’, ‘dopamine AND acute renal failure’, ‘fe-noldopam AND acute kidney injury’, ‘biocompatible mem-branes AND survival AND ARF’ and ‘atrial natriuretic peptideAND acute kidney injury’.

Table 1. Two-by-two contingency table comparing filter to ‘referencestandard’

Filter (consisting of singleor combined terms)

Manual review of each article

Articles relevant toacute kidney injury

Articles not relevantto acute kidney injury

Article identified a bArticle not identified c d

Sensitivity = a/(a + c): proportion of all articles with information on AKI in the referenceset that are retrieved by the filter (also called recall in information retrieval studies).Specificity = d/(b + d): proportion of all articles without information on AKI in thereference set that are correctly not retrieved by the filter.Precision = a/(a + b): proportion of all articles retrieved by the filter with information onAKI (also referred to as positive predictive value in diagnostic test terminology).Accuracy = (a + d)/(a + b + c + d): proportion of all articles dealt with correctly by filter.

ORIG

INALARTIC

LE

H i g h - p e r f o r m a n c e s e a r c h fi l t e r s f o r a c u t e k i d n e y i n j u r y 825

at Ondokuz M

ayis University on N

ovember 7, 2014

http://ndt.oxfordjournals.org/D

ownloaded from

Table 2. AKI search filters for PubMed, Ovid Medline and Embase optimized for high sensitivity and high specificity

Set Sensitivity (%)(95% ConfidenceInterval)

Specificity (%)(95% ConfidenceInterval)

Precision (%)(95% ConfidenceInterval)

Accuracy (%)(95% ConfidenceInterval)

PubMed filtersa

High-sensitivity filter (acute kidney[tw] OR acute renal[tw] OR acute nephr*[tw] OR acute glomer*[tw]OR acute dialysis[tw] OR acute tubul*[tw] OR "Acute Kidney Injury"[mh] ORkidney injur*[tiab] OR renal injur*[tiab] OR "Kidney Diseases/chemicallyinduced"[mh] OR tubular injury[tiab] OR tubular necrosis*[tiab] OR tubulardamage*[tiab] OR tubule damage*[tiab] OR nephrotox*[tiab] OR "Nephritis,Interstitial"[mh:noexp] OR tubulointerstitial nephritis[tiab] OR interstitialnephritis[tiab] OR kidney ischemi*[tiab] OR kidney ischaemi*[tiab] OR renalischemi*[tiab] OR renal ischaemi*[tiab] OR induced kidney[tiab] OR induced renal[tiab] OR hemolytic uremi*[tiab] OR haemolytic uraemi*[tiab] OR "Hemolytic-Uremic Syndrome"[majr:noexp] OR aki[tiab] OR oliguri*[tw] OR anuri*[tw] ORanti-glomerular[tw] OR antiglomerular[tw] OR "Kidney Cortex Necrosis"[mh:noexp] OR pre-renal[tiab] OR prerenal[tiab] OR anti-gbm[tiab] OR obstructedkidney*[tiab] OR renal obstruction[tiab] OR obstructive nephropathy[tiab] ORobstructive uropathy[tiab] OR hepatorenal syndrome[tw] OR "Hemorrhagic Feverwith Renal Syndrome"[majr:noexp] OR thrombotic thrombocytopeni*[tiab] ORthrombotic microangiopathy[tiab] OR "Acidosis/chemically induced"[mh] ORrenal hypoperfusion[tiab] OR (worsening[tiab] AND renal[tiab]) OR improvedrenal function[tiab] OR (impair*[tiab] AND renal function[tiab]) OR azotemi*[tw]OR azotaemi*[tw] OR (renal[tiab] AND thrombosis[tiab]) OR (("ReperfusionInjury"[mh:noexp] OR ischemic injury[tiab] OR ischemia injury[tiab] ORischaemic injury[tiab] OR ischaemia injury[tiab] OR ischemic reperfusion[tiab]OR ischemia reperfusion[tiab] OR ischaemic reperfusion[tiab] OR ischaemiareperfusion[tiab] OR critical care[tw] OR critically ill[tw] OR (critical*[tw] ANDillness[tw]) OR sepsis[tw] or septic[tw] OR intensive care[tw] OR icu[tiab] ORtubular cell*[tiab] OR rhabdomyolysis[tw] OR thrombocytopeni*[tiab] OR life-threatening[tw] OR vasculit*[tw] OR polyarteritis[tw] OR cardiogenic shock[tiab]OR multiorgan dysfunction[tw] OR multi-organ dysfunction[tw] OR multipleorgan dysfunction[tw] OR multiple organ failure[tw] OR multiorgan failure[tw]OR multi-organ failure[tw] OR polyangiitis[tw] OR (wegener*[tw] ANDgranulomatosis[tw]) OR "Blood Urea Nitrogen"[mh:noexp]) AND (kidney[tw] ORrenal[tw] OR dialysis[tw] OR uremi*[tiab] OR uraemi*[tiab] OR dehydrat*[tw] ORcreatinin*[tw])) OR (nephropath*[tw] AND (contrast medi*[tw] OR contrastinduced[tw] OR contrast agent*[tw] OR radiocontrast*[tw] OR iodinated[tw] ORcrystal*[tw] OR cast[tw])) OR ((glomerulonephritis[tw] OR nephrit*[tiab]) AND(crescentic[tw] OR anca*[tiab] OR rapidly progressive[tiab] OR acute[tiab])) OR(("Kidney Diseases"[mh:noexp] OR renal insufficienc*[tw] OR renal failure[tw] ORrenal function[tiab] OR renal impairment[tiab] OR glomerular filtration rate[tiab]OR ischemia-reperfusion injury[tiab]) AND ("Cardiovascular SurgicalProcedures"[majr] OR "Cardiovascular Diseases"[mh:noexp] OR "CardiovascularSystem/surgery"[majr] OR cardiac surg*[tw] OR (cardiopulmonary[tiab]) OR"Ischemia"[mh:noexp] OR "diagnostic imaging"[majr] OR "Contrast Media"[majr:noexp] OR "chemically induced"[sh] OR revers*[tiab] OR microangiopath*[tiab]OR cirrhosis[ti] OR "Substance-Related Disorders"[mh] OR "NeurologicManifestations"[mh] OR preoperative*[tiab] OR pre-operative*[tiab] OR

Development 97.2 (95.3–99.0) 94.8 (94.5–95.2) 31.5 (28.6–34.4) 94.9 (94.5–95.3)Validation 96.1 (92.4–99.8) 95.0 (94.6–95.5) 20.0 (16.5–23.6) 95.1 (94.6–95.5)

ORIGINAL ARTICLE

826A.M

.Hild

ebrandet

al.

at Ondokuz Mayis University on November 7, 2014 http://ndt.oxfordjournals.org/ Downloaded from

postoperative*[tiab] OR post-operative*[tiab])) OR ((injury[tw] OR ischemi*[tw]OR ischaemi*[tw] OR reperfusion[tw] OR contrast medi*[tw]) AND (renal tubul*[tiab] OR tubular[tiab])))

High-specificity filter (("Acute Kidney Injury"[mh] OR acute kidney[tiab] OR acute renal[tiab] OR acutenephr*[ti] OR acute tubular[ti] OR acute dialys*[ti] OR kidney ischemi*[ti] ORrenal ischemi*[ti] OR kidney ischaemi*[ti] OR renal ischaemi*[ti] OR inducedkidney injury[tiab] OR induced renal injury[tiab] OR acute nephr*[tiab] OR pre-renal[tiab] OR prerenal[tiab] OR "Nephritis,Interstitial/chemically induced"[majr:noexp] OR "Hemorrhagic Fever with Renal Syndrome"[majr:noexp] OR"Oliguria"[mh:noexp] OR ((crescent*[ti] OR progressive[ti] OR anca*[ti] OR acute[ti]) AND (glomerul*[ti] OR nephrit*[ti])) OR ((arf[tiab] OR aki[tiab]) AND(renal[tiab] OR kidney[tiab])) OR (nephropath*[tw] AND (contrast medi*[tw] ORcontrast induced[tw] OR contrast agent*[tw] OR radiocontrast*[tw] OR iodinated[tw] OR crystal*[tw] OR cast[tw])) OR ((nephrotox*[ti] OR (renal[ti] AND toxi*[ti]) OR renal tubul*[ti]) AND ("chemically induced"[sh] OR contrast medi*[tiab]OR induced[tw])) OR ((kidney ischemi*[tiab] OR renal ischemi*[tiab] OR kidneyischaemi*[tiab] OR renal ischaemi*[tiab] OR "Kidney Tubules, Proximal"[mh:noexp] OR uremi*[ti] OR uraemi*[ti] OR renal inflammation[tiab]) AND("Reperfusion Injury"[majr:noexp] OR ischaemic reperfusion[tiab] OR ischemicreperfusion[tiab] OR ischaemia reperfusion[tiab] OR ischemia reperfusion[tiab]OR injury[ti] OR acute[tiab])) OR ((hemolytic uremi*[ti] OR haemolytic uraemi*[ti] OR thrombotic thrombocytopeni*[tiab] OR thrombotic microangiopathy[tiab])AND (kidney[tw] OR renal[tw] OR acute[tw])) OR ((induced kidney[tiab] ORinduced renal[tiab]) AND (nephrotox*[tw] OR contrast medi*[tw] OR reperfusion[tw] OR perfusion[tw])) OR (((tubulointerstitial[ti] OR interstitial[ti] OR anti-glomerular[ti] OR antiglomerular[ti]) AND (glomerul*[ti] OR nephrit*[ti])) AND(acute[tw] OR crescentic[tw] OR atypical[tw] OR progressive[tw])) OR ((intensivecare[tw] OR "Neurologic Manifestations"[majr] OR "Substance-RelatedDisorders"[majr] OR cardiac surg*[tw]) AND renal failure[tiab]) OR ((acute[tiab]OR nephrotox*[tiab] OR toxicity[tiab] OR ureteral obstruction[tw]) AND (kidneyinjur*[tiab] OR renal injur*[tiab]))) NOT (allograft*[ti] OR glomerulosclerosis[ti]))

Development 89.9 (86.5–94.0) 99.5 (99.4–99.6) 81.4 (77.3–85.5) 99.3 (99.1–99.4)Validation 82.5 (75.2–89.9) 99.1 (99.0–99.4) 57.4 (49.5–65.4) 99.0 (98.8–99.2)

Ovid Medline filtersb

High-sensitivity filter ((acute adj2 (kidney or renal or nephr$ or glomer$ or h?emodialy$ or dialysis)).mpOR exp Acute Kidney Injury/ OR ((kidney or renal) adj injur$).tw OR exp KidneyDiseases/ci OR (tubul$ adj (injury or necrosis or damage)).tw OR nephrotox$.twOR Nephritis, Interstitial/ OR ((tubulointerstitial or interstitial) adj nephr$).tw OR((kidney$ or renal) adj isch?emi$).tw OR (induced adj (kidney or renal)).tw OR (h?emolytic ur?emi$).tw OR *Hemolytic-Uremic Syndrome/ OR aki.tw OR oliguri$.mp or anuri$.mp OR anti-glomerular.mp OR antiglomerular.mp OR KidneyCortex Necrosis/ OR pre-renal.tw or prerenal.tw OR anti-gbm.tw OR (obstruct$adj2 (kidney$ or nephropath$ or renal or uropathy)).tw OR hepatorenal syndrome.mp OR *Hemorrhagic Fever with Renal Syndrome/ OR (thrombotic adj(thrombocytopeni$ or microangiopathy)).tw OR exp Acidosis/ci OR renalhypoperfusion.tw OR (worsening and renal).tw OR ((improved or recover$ orimpair$) adj2 renal function).tw OR azot?emi$.mp OR (renal adj2 thrombosis).twOR ((Reperfusion Injury/ OR (isch?emi$ adj (reperfusion or injury)).tw OR(critical$ adj (care or ill$ or patient$)).mp OR sepsis.mp OR septic.mp ORintensive care.mp OR icu.tw OR tubular cell$.tw OR rhabdomyolysis.mp ORthrombocytopeni$.tw OR life-threatening.mp OR vasculit$.mp OR polyarteritis.mpOR ((multi$ organ or multiorgan) adj (failure or dysfunction)).mp OR cardiogenic

Development 97.2 (95.3–99.0) 94.9 (94.5–95.4) 31.9 (29.0–34.6) 95.0 (94.6–95.3)Validation 96.1 (92.4–99.8) 95.0 (94.6–95.5) 20.0 (16.5–23.6) 95.1 (94.6–95.5)

Continued

ORIGINALARTICLE

High-perform

ancesearch

filt

ersforacutekidneyinju

ry

827

at Ondokuz Mayis University on November 7, 2014 http://ndt.oxfordjournals.org/ Downloaded from

Table 2. Continued

Set Sensitivity (%)(95% ConfidenceInterval)

Specificity (%)(95% ConfidenceInterval)

Precision (%)(95% ConfidenceInterval)

Accuracy (%)(95% ConfidenceInterval)

shock.tw OR Blood Urea Nitrogen/ OR polyangiitis.mp OR wegener$granulomatosis.mp) AND (kidney.mp OR renal.mp OR dialysis.mp OR ur?emi$.twOR dehydrat$.mp OR creatinin$.mp)) OR (nephropath$ AND ((contrast$ adj(medi$ OR induced OR agent$)) OR radiocontrast$ OR iodinated OR crystal$ ORcast)).mp. OR ((glomerulonephritis.mp OR nephrit$.tw) AND (acute.tw ORcrescentic.mp OR anca$.tw OR rapidly progressive.tw)) OR ((Kidney Diseases/ OR(renal adj (insufficienc$ or failure or function or impairment)).mp OR ischemia-reperfusion injury.tw OR glomerular filtration rate.tw) AND (exp *CardiovascularSurgical Procedures/ OR Cardiovascular Diseases/ OR exp *Cardiovascular System/su OR cardiac surg$.mp OR cardiopulmonary.tw OR Ischemia/ OR exp *diagnosticimaging/ OR exp Neurologic Manifestations/ OR *Contrast Media/ ORpreoperative$.tw OR pre-operative$.tw OR postoperative$.tw OR post-operative$.tw OR exp Substance-Related Disorders/ OR microangiopath$.tw OR cirrhosis.tiOR revers$.tw OR ci.fs)) OR ((injury.mp or isch?emi$.mp or reperfusion.mp orcontrast medi$.mp) AND (renal tubul$.tw or tubular.tw)))

High-specificity filter ((exp Acute Kidney Injury/ OR (acute adj2 (kidney or renal)).tw OR (acute adj(kidney or renal or nephr$ or tubular or dialys$)).ti OR ((crescent$ or progressiveor anca$ or acute) and (glomerul$ or nephrit$)).ti OR ((kidney or renal) adj isch?emi$).ti OR *Nephritis,Interstitial/ci OR *Hemorrhagic Fever with RenalSyndrome/ OR (induced adj (kidney injury or renal injury)).tw OR Oliguria/ OR(acute nephr$).tw OR (pre-renal or prerenal).tw OR ((arf OR aki) AND (renal ORkidney)).tw. OR (nephropath$ AND ((contrast$ adj (medi$ OR induced OR agent$)) OR radiocontrast$ OR iodinated OR crystal$ OR cast)).mp. OR (((nephrotox$or (renal and toxi$)).ti OR (renal tubul$).ti) AND (ci.fs OR contrast medi$.tw ORinduced.mp)) OR ((((kidney or renal) adj isch?emi$).tw OR Kidney Tubules,Proximal/ OR ur?emi$.ti OR (renal inflammation).tw) AND (*Reperfusion Injury/OR (isch?emi$ reperfusion).tw OR injury.ti OR acute.tw)) OR (((h?emolyt$ ur?emi$).ti OR (thrombotic adj (thrombocytopeni$ or microangiopathy)).tw) AND(kidney or renal or acute).mp) OR ((induced adj (kidney or renal)).tw AND(nephrotox$ or contrast medi$ or reperfusion or perfusion).mp) OR(((tubulointerstitial or interstitial or anti-glomerular or antiglomerular) and(glomerul$ or nephrit$)).ti AND (acute or crescentic or atypical or progressive).mp) OR (((intensive care).mp OR exp *Neurologic Manifestations/ OR exp*Substance-Related Disorders/ OR (cardiac surg$).mp) AND (renal failure).tw) OR((acute.tw OR nephrotox$.tw OR toxicity.tw OR (ureteral obstruction).mp) AND((kidney or renal) adj injur$).tw)) NOT (allograft$ OR glomerulosclerosis).ti)

Development 89.9 (87.3–93.7) 99.5 (99.3–99.6) 80.9 (76.0–84.3) 99.3 (99.1–99.4)Validation 82.5 (75.2–89.9) 99.2 (99.0–99.4) 57.1 (49.1–65.0) 99.0 (98.8–99.2)

Embase filtersc

High-sensitivity filter ((acute adj2 (kidney OR renal OR nephr$ OR glomer$ OR h?emodial$ ORdialysis)).mp OR ((renal OR kidney OR glomerul$) adj injur$).mp OR (oliguri$OR anuri$).mp OR (tubul$ adj (injur$ OR necrosis OR damage)).mp ORnephrotoxi$.tw OR ((tubulointerstitial OR interstitial) adj nephr$).mp OR ureanitrogen blood level/ OR ((kidney$ OR renal) adj isch?emi$).tw OR (thromboticadj (thrombocytopeni$ OR microangiopathy)).mp OR (h?emolytic$ ur?emi$).twOR (induced adj (kidney OR renal)).tw OR (anti-gbm OR anti-glomerular ORantiglomerular).mp OR azot?emi$.mp OR (obstruct$ adj2 (kidney$ ORnephropath$ OR renal OR uropathy)).mp OR (worsen$ renal).tw OR (renal adj2

Development 98.5 (97.2–99.8) 94.3 (93.9–94.7) 29.8 (27.1–32.5) 94.4 (94.0–94.7)Validation 94.6 (90.5–98.8) 94.7 (94.2–95.2) 19.8 (16.4–23.2) 94.7 (94.2–95.2)

ORIGINAL ARTICLE

828A.M

.Hild

ebrandet

al.

at Ondokuz Mayis University on November 7, 2014 http://ndt.oxfordjournals.org/ Downloaded from

thrombosis).tw OR (((intensive care$).mp OR (sepsis OR septic).mp OR (isch?emi$ adj (reperfusion OR injury)).mp OR reperfusion injury/ OR (critical$ adj (careOR ill$ OR patient$)).mp OR icu.mp OR life-threatening.mp OR drug induceddisease/ OR thrombocytopeni$.tw OR (emergency ward).mp OR rhabdomyolysis.tw OR vasculit$.tw OR (polyangiitis OR wegener$ granulomatosis).mp OR ((multi$ organ OR multiorgan) adj (failure OR dysfunction)).mp OR (cardiogenic shock).mp) AND ((kidney OR renal OR dialysis OR ur?emi$ OR dehydrat$ OR creatinin$).mp)) OR (((renal adj (insufficienc$ OR failure)).mp OR kidney disease/ ORkidney failure/ OR *uremia/) AND (contrast$ medi$.mp OR exp cardiovascularsurgery/ OR exp drug dose/ OR exp cardiovascular procedures/ OR microangiopath$.tw)) OR ((nephropath$.mp OR nephrotox$.mp OR exp *acidosis/) AND((contrast$ adj (medi$ OR induced$ OR agent$)).mp OR to.fs OR iodinated.mpOR (crystal$ OR cast).mp OR radiocontrast$.mp OR exp hantavirus/)) OR((nephritis.mp OR glomerulonephritis.mp) AND (crescent$.mp OR anca$.mp ORrapidly progressive.mp)))

High-specificity filter (((acute renal failure).mp OR (acute adj2 (kidney or renal or nephr$ or glomer$ ortubular)).ti OR *acute kidney failure/ OR ((arf or aki) and (renal or kidney)).mpOR (nephropath$ and ((contrast$ adj (medi$ or induced$ or agent$)) or iodinatedor radiocontrast$)).mp OR nephrotoxic$.ti OR *rapidly progressiveglomerulonephritis/ OR ((crescent$ or progressive) adj glomerulonephrit$).ti OR*acute kidney tubule necrosis/ OR (induced adj (kidney injury or renal injury)).twOR (acute nephr$).tw OR (pre-renal or prerenal).tw OR ((((tubulointerstitial orinterstitial or anti-glomerular or antiglomerular) and (glomerul$ or nephrit$)).tiOR *interstitial nephritis/) AND (acute or crescentic or atypical or drug-induced).mp) OR (((crescent$ or progressive) adj glomerulonephrit$).mp AND(*glomerulonephritis/ OR (necrotizing or anca$ or vasculitis).mp)) OR (exp*kidney disease/ AND (*contrast medium/ OR *iodinated contrast medium/ ORexp hantavirus/)) OR ((*kidney ischemia/ OR (renal inflammation).mp) AND(injury.ti OR *reperfusion injury/)) OR ((oliguri$.mp OR (intensive care$).mp OR(emergency ward).mp OR (critical$ adj (care or ill$ or patient$)).mp OR (cardiacsurg$).mp OR exp respiratory function disorder/) AND ((renal failure$).tw OR(renal function).ti)) OR (((thrombotic adj (thrombocytopeni$ ormicroangiopathy)).tw OR (h?emolyt$ ur?emi$).ti OR *hemolytic uremicsyndrome/ OR *thrombotic thrombocytopenic purpura/ OR *vasculitis/ OR*rhabdomyolysis/) AND (((kidney or renal) adj failure).mp OR kidney biopsy/ ORacute disease/)) OR ((exp renal replacement therapy/ OR ((kidney or renal$) adjisch?emi$).mp OR apoptosi$.mp OR exp *heart surgery/ OR ((multi$ organ ormultiorgan) adj (failure or dysfunction)).mp) AND ((acute adj2 (kidney or renal$)).mp OR ((kidney or renal) adj injury).mp)) OR (((acute adj (kidney or renal orglomer$ or tubul$)).mp OR (renal AND (toxi$ OR tubul$)).ti) AND ((contrast$medi$).mp OR (induced adj2 (kidney or renal or nephr$)).tw OR *kidney failure/))) NOT allograft$.ti)

Development 90.8 (87.7–93.9) 99.5 (99.4–99.6) 81.8 (77.9–85.7) 99.3 (99.2–99.4)Validation 84.0 (77.1–90.7) 99.1 (98.9–99.3) 57.3 (49.7–64.9) 98.9 (98.7–99.2)

aPubMed fields: *, truncation character; [tw], text word present in title, abstract, or MeSH term; [tiab], term present in title or abstract; [majr:noexp], not exploded and focused MeSH term; [mh:noexp], non-exploded MeSH term; [mh], explodedMeSH term; [majr], exploded and focused MeSH term; [ti], term present in title.bOvid Medline fields: $, truncation character; mp, multiple posting (term appears in title, abstract, or MeSH); tw, text word present in title; /, MeSH character; adj, adjacent operator; *, focused MeSH term; adj2, defined adjacency operator; exp,exploded MeSH term; ?, optional wildcard; ti, term present in title.cEmbase fields: exp, exploded Emtree term; /, Emtree character; adj, adjacent operator; $, truncation character; mp, multiple posting (term appears in title, abstract or Emtree); tw, term present in title or abstract; *, focused Emtree term; adj2, definedadjacency operator; ec, Embase section headings field; ti, term present in title.

ORIGINALARTICLE

High-perform

ancesearch

filt

ersforacutekidneyinju

ry

829

at Ondokuz Mayis University on November 7, 2014 http://ndt.oxfordjournals.org/ Downloaded from

In the five proof-of-concept searches, the increase in thenumber of relevant articles retrieved was between a 33 and233% when a filter-aided search was performed comparedwith when the physician search was conducted alone(Table 3). The retrieval of non-relevant articles per each rel-evant article increased modestly with the filter-aided searchcompared with the physician search alone. As expected, high-performance filters designed to maximize sensitivity retrieveda greater number of non-relevant articles compared with thefilters designed to maximize specificity. While the filtersmissed fewer relevant articles, there was still incomplete retrie-val of relevant articles in four of five cases (between 40 and93% of all relevant articles were retrieved with the filter-aidedsearch).

DISCUSSION

Until recently, there has been no consensus on the clinicaldefinition or diagnostic criteria for AKI. For example there are>35 different definitions in the literature that describe asudden loss of kidney function and phrases such as ‘acuterenal failure’ and ‘acute tubular necrosis’ are used loosely[4, 32]. In the same way that this lack of consensus has led to awide variation in the reported incidence and clinical signifi-cance of an acute impairment of kidney function, it has alsoled to a high degree of variability in indexing of articles about

AKI. Articles that may be of general AKI interest may beindexed with common phrases such as ‘tubular necrosis’ and‘prerenal azotemia’, while others such as ‘acute dialysis’ mayonly imply AKI or make reference to more specific diseasestates such as with ‘hepatorenal syndrome’, ‘rapidly progressiveglomerulonephritis’ or ‘ischemia reperfusion injury’. Searchesmust be able to account for the historic and newer terminologyused to describe AKI [17, 18]. Building on the same concepts,our group has used to create novel high-performance searchfilters for general nephrology, dialysis, renal transplantationand glomerular disease; using computer automation, we suc-cessfully developed search filters that address these challengesand allow users to capture articles relevant to AKI with a highdegree of sensitivity, specificity and precision [12, 13, 15, 16].

All filters developed as part of this study achieved a balanceof at least 90% sensitivity and specificity. Our best performinghigh-sensitivity filter was in Embase, which achieved 98.5%sensitivity and 94.3% specificity. The best performing high-specificity filter was also in Embase, which reached 90.8% sen-sitivity and 99.5% specificity. As of September 2013, our high-sensitivity filter reduced the number of records retrieved inPubMed from >22 million citations to just <2 00 000 records;the high-specificity filter reduced this number to ∼68 000records. The effect of this on search results is similar to the in-crease in positive predictive value of a screening test whenapplied to a high-risk population: there is an increase in theprecision of the search (greater proportion of relevant articles

Table 3. Proof-of-concept searches showing the number of relevant articles retrieved with and without AKI filter

Clinical question Number of relevant articles retrieved (n) Number of non-relevant articles retrieved for eachrelevant article (n)

Physiciansearch alone

Physician searchwith the high-sensitivity filter

Physician searchwith the high-specificity filter

Physiciansearch alone

Physician searchwith the high-sensitivity filter

Physician searchwith the high-specificity filter

Is acetylcysteine effective in preventingcontrast-related nephropathy? (15 relevantarticles)

7 14 11 8 22 13

What is the efficacy of low-dose dopamine (<5µg/kg of body weight per minute) comparedwith no therapy in patients with or at risk foracute renal failure? (51 relevant articles)

15 38 23 30 28 22

What is the impact of fenoldopam on acutekidney injury, patient mortality and length ofhospital stay in critically ill patients? (12relevant articles)

5 9 7 19 24 23

Does the use of biocompatible membranes(BCM) confer an advantage in either survivalor recovery of renal function over the use ofbioincompatible membranes (BICM) in adultpatients with acute renal failure (ARF)requiring intermittent hemodialysis? (10relevant articles)

3 6 6 8 10 9

What are the benefits of atrial natriureticpeptide (ANP) in the prevention and treatmentof acute kidney injury (AKI)? (10 relevantarticles)

3 10 4 62 77 73

The search phrases provided by physicians by means of an electronic survey and applied to PubMed were ("Acute Kidney Injury"[MeSH] AND Acetylcysteine), (dopamine AND acuterenal failure), (fenoldopam AND acute kidney injury), (biocompatible membranes AND survival AND ARF) and (atrial natriuretic peptide AND acute kidney injury).

ORIG

INALARTIC

LE

830 A. M. Hildebrand et al.

at Ondokuz M

ayis University on N

ovember 7, 2014

http://ndt.oxfordjournals.org/D

ownloaded from

retrieved), regardless of the search terms entered by the user.This alone has merit to improve the user experience andarticle retrieval, though these filters serve many other pur-poses. These filters act as an optimized substitute for the vari-able terminology used to index articles relevant to AKI andallow users to simplify search strategies, targeting only thecontent of interest within the area of AKI and avoiding overlyspecific AKI terminology. In addition, these filters account forindexing inconsistencies that inexperienced users may not beprepared to deal with. For example many articles in the inten-sive care and surgical literature do not explicitly refer to asudden loss of kidney function in the title, abstract or indexterms and, only upon full-text examination, it is apparent thatthe article is relevant to AKI. Without a more sophisticatedstrategy that incorporates phrases such as ‘post-operative’, ‘cri-tically-ill’ or ‘ischemia reperfusion’ in the search query, sucharticles may be overlooked [33].

While these filters are complex, combining >140 termswith Boolean operators, they are easy to apply and are readyfor use (Supplementary Appendix C). At this time, these filtersmay be copied from Table 2 and pasted directly into the searchquery and saved for future use with user-generated searchterms. In the future, these filters may also be incorporated intothe PubMed and Ovid search engine interfaces to achievemore widespread access. In the meantime, we provide thesefilters online via the following link: http://hiru.mcmaster.ca/hiru/hiru_hedges_nephrology_filters.aspx. Users can selecteither the high-sensitivity (broad) filter or the high-specificity(narrow) filter according to the degree of article retrievalthey deem manageable and the extent to which they placeimportance on retrieving all relevant articles. For busy clini-cians at the point of care, we recommend use of the narrowhigh-specificity filter, while the high-sensitivity filter may bemore appropriate for researchers who are interested in morecomprehensive article retrieval. Although use of the high-sensitivity filter compromises precision to some extent, thismay be offset by using these filters in combination with othermethods based filters already incorporated into the PubMedinterface to retrieve articles by study design, such as those opti-mized to retrieve high-quality studies about treatment (clinicalqueries ‘therapy’ filter) [34]. As we deliberately designed thefilter to retrieve clinical and basic science content, this strategymay also be useful for users interested in only retrieving clini-cal AKI content.

AKI is a global problem, occurring in the community, hos-pital wards, emergency departments and intensive care units.Use of these filters can improve the quality and efficiency ofarticle retrieval for a broad range of physicians managing AKIin these settings, which in turn may translate to more efficientand effective evidence-based decision-making, education andpatient care. However, the effectiveness of these search filtersis subject to some limitations. Most importantly, as withsearches performed without a filter, results remain highly de-pendent on the quality of search terms entered by the user andthe quality and consistency of indexing. This is evident fromthe incomplete retrieval of relevant articles in four of fiveproof-of-concept searches even with the high-sensitivity filterin use. Our filters cannot compensate for incorrect or

incomplete indexing or terms and abbreviations entered byusers that are overly specific, irrelevant, misspelled or inappro-priately combined. Further, as knowledge of the pathogenesisof AKI grows and definitions change, these filters will need tobe updated to incorporate new terminology used to index rel-evant articles. Although our filters retained high sensitivityand specificity in the validation phase, there was a drop in pre-cision that was expected, as it was a smaller database by designwith a lower proportion of relevant articles. Users can expectthis to be more pronounced when applying the filter to thebibliographic database at large, as our collection of journalscontributing to the validation set of articles was deliberatelyenriched with leading clinical nephrology journals. Finally,proof-of-concept searches were used to illustrate the function-ality of our best performing filters with real physician searches.Based on these results, users can expect improved retrieval ofrelevant articles with a modest increase in the number of non-relevant articles, resulting in increased overall precision of thesearch. However, these results are limited by the methods usedto define the reference standard (based on articles used in sys-tematic reviews of variable quality) and should be viewed onlyas illustrative examples.

In conclusion, PubMed, Ovid Medline and Embase can befiltered for articles relevant to AKI in a reliable manner. Ourfilters are available online for use by clinicians and researchers.A future research agenda is to assess the uptake of these filtersamong physicians and measure the impact they have on infor-mation-seeking behavior, knowledge, decision-making andultimately patient care.

SUPPLEMENTARY DATA

Supplementary data are available online at http://ndt.oxfordjournals.org.

ACKNOWLEDGEMENTS

We thank Dr Christopher Lee for his help reviewing articlesand Dr Ann McKibbon and Dr Salimah Shariff for theiradvice on this research. We also thank Mr Nicholas Hobsonand Mr Chris Cotoi who performed the computer program-ming. This project was supported by an operating grant fromthe Canadian Institutes of Health Research (CIHR). A.M.H.was supported by the Clinical Investigator Program at WesternUniversity. A.X.G. was supported by a CIHR Clinician-Scientist Award.

CONFLICT OF INTEREST STATEMENT

We have no competing financial interests to declare. Resultspresented in this paper have not been published previously inwhole or part, except in abstract format.

ORIG

INALARTIC

LE

H i g h - p e r f o r m a n c e s e a r c h fi l t e r s f o r a c u t e k i d n e y i n j u r y 831

at Ondokuz M

ayis University on N

ovember 7, 2014

http://ndt.oxfordjournals.org/D

ownloaded from

REFERENCES

1. Coumou HCH, Meijman FJ. How do primary care physicians seek answersto clinical questions? A literature review. J Med Libr Assoc 2006; 94: 55–60

2. Ely JW, Osheroff JA, Ebell MH et al. Obstacles to answering doctors’ ques-tions about patient care with evidence: qualitative study. BMJ 2002; 324: 710

3. Davies K. The information-seeking behaviour of doctors: a review of theevidence. Health Info Libr J 2007; 24: 78–94

4. Kidney Disease: Improving Global Outcomes (KDIGO) Acute KidneyInjury Work Group. KDIGO clinical practice guideline for acute kidneyinjury. Kidney Int Suppl 2012; 2: 1–138

5. Haynes RB, Wilczynski N, McKibbon KA et al. Developing optimalsearch strategies for detecting clinically sound studies in MEDLINE. J AmMed Inform Assoc 1994; 1: 447–458

6. Jenkins M. Evaluation of methodological search filters–a review. HealthInfo Libr J 2004; 21: 148–763

7. Haynes RB, Wilczynski NL. Optimal search strategies for retrieving scien-tifically strong studies of diagnosis from Medline: analytical survey. BMJ2004; 328: 1040

8. Wilczynski NL, Haynes RB; Hedges Team. Developing optimal searchstrategies for detecting clinically sound causation studies in MEDLINE.AMIA Annu Symp Proc 2003; 719–723

9. Haynes RB, McKibbon KA, Wilczynski NL et al. Hedges Team. Optimalsearch strategies for retrieving scientifically strong studies of treatmentfromMedline: analytical survey. BMJ 2005; 330: 1179

10. Wong SS, Wilczynski NL, Haynes RB et al. Hedges Team. Developingoptimal search strategies for detecting sound clinical prediction studies inMEDLINE. AMIA Annu Symp Proc 2003; 728–732

11. Garg AX, Iansavichus AV, Kastner M et al. Lost in publication: half of allrenal practice evidence is published in non-renal journals. Kidney Int2006; 70: 1995–2005

12. Iansavichus AV, Haynes RB, Shariff SZ et al. Optimal search filters forrenal information in EMBASE. Am J Kidney Dis 2010; 56: 14–22

13. Hildebrand AM, Iansavichus AV, Lee CWC et al. Glomerular diseasesearch filters for Pubmed, Ovid Medline, and Embase: a development andvalidation study. BMCMed Inform Decis Mak 2012; 12: 49

14. Iansavichus AV, Haynes RB, Lee CWC et al. Dialysis search filters forPubMed, Ovid MEDLINE, and Embase databases. Clin J Am Soc Nephrol2012; 7: 1624–1631

15. Lee CWC, Iansavichus AV, Haynes RB et al. Kidney transplantationsearch filters for PubMed, Ovid Medline, and Embase. Transplantation2012; 93: 460–466

16. Garg AX, Iansavichus AV, Wilczynski NL et al. Filtering Medline for a clini-cal discipline: diagnostic test assessment framework. BMJ 2009; 339: b3435

17. Bellomo R, Ronco C, Kellum JA et al. Acute renal failure—definition,outcome measures, animal models, fluid therapy and information

technology needs: the Second International Consensus Conference of theAcute Dialysis Quality Initiative (ADQI) Group. Crit Care 2004; 8: R204

18. Mehta RL, Kellum JA, Shah SV et al. Acute Kidney Injury Network:report of an initiative to improve outcomes in acute kidney injury. CritCare 2007; 11: R31

19. Wilczynski NL, Haynes RB, Team H. Robustness of empirical search strat-egies for clinical content in MEDLINE. AMIA Annu Symp Proc 2002;904–908

20. Yao X, Wilczynski NL, Walter SD et al. Sample size determination for bib-liographic retrieval studies. BMCMed Inform Decis Mak 2008; 8: 43

21. National Library of Medicine. MeSH Browser.22. National Library of Medicine. Permuted Medical Subject Headings.

Bethesda, MD: US Department of Commerce, National Technical Infor-mation Center, 2003

23. Excerpta Medica. EMTREE Thesaurus; Amsterdam: Exerpta Medica, 200124. Brenner B, Rector F. The Kidney, 5th edn. Philadelphia: WB Saunders,

199625. National Kidney Foundation. K/DOQI clinical practice guidelines for

chronic kidney disease: evaluation, classification, and stratification. Am JKidney Dis 2002; 39: S1–266

26. National Kidney Foundation. NKF-KDOQI Guidelines. www.kidney.org/professionals/kdoqi/guidelines.cfm. (September 2013 accessed).

27. Nallamothu BK, Shojania KG, Saint S et al. Is acetylcysteine effective inpreventing contrast-related nephropathy? A meta-analysis. Am J Med2004; 117: 938–947

28. Friedrich JO, Adhikari N, Herridge MS et al. Meta-analysis: low-dosedopamine increases urine output but does not prevent renal dysfunctionor death. Ann Intern Med 2005; 142: 510–524

29. Landoni G, Biondi-Zoccai GGL, Tumlin JA et al. Beneficial impact offenoldopam in critically ill patients with or at risk for acute renalfailure: a meta-analysis of randomized clinical trials. Am J Kid Dis2007; 49: 56–68

30. Alonso A, Lau J, Jaber BL. Biocompatible hemodialysis membranes foracute renal failure. Cochrane Database Syst Rev 2008; 1: CD005283

31. Nigwekar SU, Navaneethan SD, Parikh CR et al. Atrial natriuretic peptidefor management of acute kidney injury: a systematic review and meta-analysis. Clin J Am Soc Nephrol 2009; 4: 261–272

32. Kellum JA, Levin N, Bouman C et al. Developing a consensus classifi-cation system for acute renal failure. Curr Opin Crit Care 2002; 8:509–514

33. Mangano DT, Tudor IC, Dietzel C. The risk associated with aprotinin incardiac surgery. N Engl J Med 2006; 354: 353–365

34. Shariff SZ, Sontrop JM, Haynes RB et al. Impact of PubMed search filterson the retrieval of evidence by physicians. Can Med Assoc J 2012; 184:E184–E190

Received for publication: 15.11.2013; Accepted in revised form: 18.12.2013

ORIG

INALARTIC

LE

832 A. M. Hildebrand et al.

at Ondokuz M

ayis University on N

ovember 7, 2014

http://ndt.oxfordjournals.org/D

ownloaded from