Concepts & Trends

Methodological issues and recommendations for systematic reviews of prognostic studies: an example from cardiovascular disease

Prognostic factors are associated with the risk of future health outcomes in individuals with a particular health condition. The prognostic ability of such factors is increasingly being assessed in both primary research and systematic reviews.
of 10
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  METHODOLOGY Open Access Methodological issues and recommendationsfor systematic reviews of prognostic studies:an example from cardiovascular disease Janine Dretzke 1* , Joie Ensor 1 , Sue Bayliss 1 , James Hodgkinson 2 , Marie Lordkipanidzé 3,4 , Richard D Riley 5 ,David Fitzmaurice 2 and David Moore 1 Abstract Background:  Prognostic factors are associated with the risk of future health outcomes in individuals with a particularhealth condition. The prognostic ability of such factors is increasingly being assessed in both primary research andsystematic reviews. Systematic review methodology in this area is continuing to evolve, reflected in variableapproaches to key methodological aspects. The aim of this article was to (i) explore and compare the methodology of systematic reviews of prognostic factors undertaken for the same clinical question, (ii) to discuss implications for reviewfindings, and (iii) to present recommendations on what might be considered to be  ‘ good practice ’  approaches. Methods:  The sample was comprised of eight systematic reviews addressing the same clinical question, namelywhether  ‘ aspirin resistance ’  (a potential prognostic factor) has prognostic utility relative to future vascular events inpatients on aspirin therapy for secondary prevention. A detailed comparison of methods around study identification,study selection, quality assessment, approaches to analysis, and reporting of findings was undertaken and theimplications discussed. These were summarised into key considerations that may be transferable to future systematicreviews of prognostic factors. Results:  Across systematic reviews addressing the same clinical question, there were considerable differences in thenumbers of studies identified and overlap between included studies, which could only partially be explained bydifferent study eligibility criteria. Incomplete reporting and differences in terminology within primary studies hamperedstudy identification and selection process across reviews. Quality assessment was highly variable and only onesystematic review considered a checklist for studies of prognostic questions. There was inconsistency between reviewsin approaches towards analysis, synthesis, addressing heterogeneity and reporting of results. Conclusions:  Different methodological approaches may ultimately affect the findings and interpretation of systematicreviews of prognostic research, with implications for clinical decision-making. Keywords:  Systematic review methodology, Prognostic utility, Prognostic factor, Aspirin resistance, Cardiovasculardisease, Search strategy, Study selection, Quality assessment, Reporting bias * Correspondence:  1 School of Health and Population Sciences, University of Birmingham,Edgbaston, Birmingham B15 2TT, UK Full list of author information is available at the end of the article © 2014 Dretzke et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the CreativeCommons Attribution License (, which permits unrestricted use, distribution, andreproduction in any medium, provided the srcinal work is properly credited. The Creative Commons Public DomainDedication waiver ( ) applies to the data made available in this article,unless otherwise stated. Dretzke  et al. Systematic Reviews  2014,  3 :140  Background Prognosis research is becoming increasingly important inhealth care, with a greater number of people than ever be-fore living with chronic disease [1]. Prognostic factors re-late to any measures that are associated with the risk of future health outcomes in individuals with a particularhealth condition [2]. Identification of prognostic factorscan be potentially useful for informing a patient ’ s riskprofile and for making therapeutic decisions [2]. Thepoor quality of prognostic research is however welldocumented, with issues relating to study design (e.g.retrospective rather than prospective), reporting (e.g.inconsistent use of nomenclature relating to prognosticresearch) and publication bias (e.g. preferential publicationof articles with positive findings); this in turn impacts onthe quality of systematic reviews and meta-analyses of published prognostic studies, which may be limited orbiased in their conclusions [2,3]. There are no accepted guidelines for undertaking sys-tematic reviews of prognosis, and methodology is stillevolving. Some methodological recommendations havebeen proposed by Hayden et al. [4] based on an evaluationof systematic reviews of low back pain prognosis. Recom-mendations include, amongst others, the assessment of allimportant potential biases and testing of the impact of specific biases on the review conclusions; listing studieswhere eligibility criteria may be unclear; and the use of sub-group and sensitivity analyses to explore sources of heterogeneity. Some recommendations for conductingsystematic reviews of prognostic tests have also been madeby the US Agency for Healthcare Research and Quality (AHRQ) [5] relating to, for example, defining the review question, searching, specific quality criteria to be used andextraction of summary statistics.The Cochrane Prognosis Methods Group [6] acts as arepository for publications of prognosis methodology. Italso offers advice to review authors on incorporatingprognosis information into their reviews and to establishmethods for undertaking systematic reviews and meta-analyses of prognosis studies; a chapter about reviews of prognosis for inclusion into the Cochrane Handbook isplanned but as yet there are no specific guidelines.Guidelines on data extraction and critical appraisal of risk prediction model studies have recently been pub-lished [7], and it is likely that some issues may overlapwith prognostic studies.This article explores the methodology of systematic re- views of prognosis in a cardiovascular area. The clinicalquestion related to the prognostic utility of platelet func-tion tests (PFT) for the detection of   ‘ aspirin resistance ’  inpatients with established cardiovascular or cerebrovascu-lar disease; more specifically, whether insufficient plateletfunction inhibition by aspirin ( ‘ aspirin resistance ’ ) as de-fined using one of several PFTs was associated with anincreased risk of adverse clinical outcomes. A further aimwas to identify whether individual patients at greater riskof future adverse clinical events could be identifiedthrough PFTs. In this example, the prognostic factor( ‘ aspirin resistance ’ ) is defined by the result of a clinical(diagnostic) test result (i.e. an individual is designatedas  ‘ aspirin resistant ’  or  ‘ aspirin sensitive ’  using a PFT),so either  ‘ aspirin resistance ’  or the PFT result could beconsidered to be the prognostic factor, as they are bothdescribing a state of platelet reactivity.A  ‘ systematic review of systematic reviews ’  was under-taken as part of a wider project, which comprised a new systematic review on the same topic and an economicevaluation [8,9]. Both the new systematic review and the previous systematic reviews are used as illustrative exam-ples throughout. The aim of this article was not to critiqueindividual reviews but to explore and compare the differ-ent methodological approaches employed within thesereviews, examine whether different methodological ap-proaches can affect overall findings and conclusionsdrawn, and draw out common methodological consid-erations which may be useful for informing future sys-tematic reviews of prognostic factors. Methods The search for systematic reviews [9] was performed inApril 2012 and updated in May 2014 (see Additional file 1for sample search strategy). Systematic reviews were eli-gible for inclusion if their primary aim was to examine orquantify the potential association between  ‘ aspirin resist-ance ’  (a candidate prognostic factor, which is defined by aPFT) and risk of future cardiovascular events in patientswho were prescribed aspirin therapy. Further inclusioncriteria were as follows: reporting of search strategy and atleast one other methodological component (e.g. details onstudy selection process or quality assessment) and alltypes of PFT eligible (i.e. reviews focussing on only onespecific PFT were excluded). The following componentsof the reviews were then compared and key differencestabulated (where reporting allowed):   Volume of evidence   Search strategy    Eligibility criteria   Quality assessment strategy    Reporting of results. Potential impacts on findings and implications for futuresystematic reviews of prognostic factors were discussed.In a separate exercise, and in order to further exploredifferent approaches to search strategies, different vali-dated prognostic search filters were also added to the ori-ginal search strategy of one systematic review [8] in order Dretzke  et al. Systematic Reviews  2014,  3 :140 Page 2 of 10  to assess the effect on sensitivity and precision of thesearch. Results Eight relevant systematic reviews were identified, withpublication dates from 2007 to 2014 (see Additional file2 for review selection process and characteristics of sys-tematic reviews): seven [10-16] through the searches and the HTA report/new systematic review [8] itself. Alleight systematic reviews had the same aim, namely toexplore the association between  ‘ aspirin resistance ’  andthe risk of future cardiovascular events in patients on as-pirin therapy for secondary prevention. Volume of included studies and method of determinationof prognostic factor (platelet function test) The number of included primary studies and theconsistency between the eight reviews were explored. Pri-mary studies with a publication date up to 2006 were con-sidered, in order to make the search periods comparableacross reviews. Thirty-eight unique studies were includedacross these reviews for this period. The numbers of stud-ies included in individual reviews varied between 2 and 25for the time period up to 2006 (see Additional file 3 fornumber and overlap between reviews). No study was in-cluded in all eight systematic reviews. Only four studieswere consistently represented in 7/8 (88%) of the system-atic reviews, six studies in  ≥ 75% of the reviews, nine stud-ies in  ≥ 63%, 14 studies in  ≥ 50%, with the remainder of primary studies ( n =24) represented in only 1 – 3 of theeight reviews. The prognostic factor can be measured by anumber of different methods (PFTs), and no review re-stricted inclusion to a specific PFT. As a consequence of including different primary studies, varying proportions of studies using a particular method (PFT) were representedin the eight reviews (Figure in Additional file 4). Differentinclusion criteria unrelated to the type of PFTare thoughtto have contributed to these discrepancies to some extentand are further discussed below. The systematic review by Li et al. [15], for example, included only studies wherecompliance had been verified. However, it does not appearthat differing selection criteria explain all of the consider-able variation. Reliably identifying prognostic studies con-tinues to be a problem in systematic reviews of prognosticfactors. Search strategies Search strategies, in particular with regard to filters forstudy identification, are less well developed in the progno-sis field and potentially less able to identify relevant stud-ies [17]. This appears to be reflected within this examplein the number and type of search terms, as well as combi-nations of terms, which varied substantially between thereviews. Table 1 shows the types of search terms includedin the respective search strategies. None of the reviews re-ported use of a filter for prognostic studies, though theHTA report [8] incorporated search terms relating to pre-diction and prognosis. There were similarities in terms of sources searched, with all searching at least two majorelectronic databases (such as MEDLINE, Embase, theCochrane library) and using citation checking. Wherestudies specified the initial number of citations identifiedwith their search strategy, the numbers varied greatly:36,573 [11], 16,583 [8], 3,847 [15], 3,882 [16] and 1,978 [13], a reflection of the sensitivity (breadth) of the initialsearch. Two reviews [10,11] attempted to reduce the num- ber of hits by subsequently including additional limits.The initial search yield did not necessarily appear to cor-relate with a greater number of included studies; however,a formal assessment of this across reviews was not pos-sible due to a lack of clarity in the reporting of yields atdifferent stages.In order to test whether the introduction of a prognos-tic filter could have limited the sensitivity (and reducedthe number of hits), whilst retaining the precision (i.e.identify relevant studies), the literature searches for theHTA report [8] were rerun, for MEDLINE only, withadditional validated Haynes filters relating to prognosisand clinical prediction [17]. It was found that the  ‘ progno-sis ’  Haynes filter picked up 77% of studies identified by the srcinal broader search strategy whilst reducing the volume of overall hits to 82%; the  ‘ clinical predictionguide ’  Haynes filter picked up 62% of studies, whilst redu-cing the volume to 72%. These results show that studiesmeeting the eligibility criteria would have been missedusing either of the prognosis/prediction filters, thoughthey both significantly reduced the larger initial quantity of evidence. In the current context, it is unlikely that cit-ation checking (forward and backward) would have led tothe identification of the missed studies, but this remainsto be empirically tested. Inclusion criteria There were variations between reviews both in the levelof reporting of inclusion and exclusion criteria and theactual criteria applied. Only three reviews [8,11,16] gave details on whether patients who were receiving aspirinas monotherapy and/or dual therapy (aspirin +other an-tiplatelet agent) were eligible; this distinction is import-ant as the presence of a second antiplatelet agent may interact with the prognostic factor assessment (plateletfunction) and also be associated with the outcome of interest (cardiovascular events).Further, only 5/8 [8,11,14-16] reviews specified whether prospective and/or retrospective study designs were eli-gible. Both concomitant therapy and study design arefactors that may have an effect on results, and if notspecified, may introduce bias into overall findings. Dretzke  et al. Systematic Reviews  2014,  3 :140 Page 3 of 10  Further, the terms prospective or retrospective may beused to describe the study design but may not apply inthe same way to data collection/analysis. For example,sample collection may have occurred before the out-come of interest, but analysis of the sample (i.e. meas-urement of the prognostic factor by PFT) may have beenundertaken after the outcomes occurred. One such ex-ample is the study by Eikelboom et al. [18], which is de-scribed as a nested case – control study (within an RCT)but could also meet the criteria for a prospective (prog-nostic) study as sample collection preceded clinical events.Other eligibility criteria variously specified in individ-ual reviews only were the reporting of specific outcomestatistics [14], blinding of investigators [11] measure- ment of patient compliance [15] or English language[16]. Again, these criteria may have an impact on com-pleteness and robustness of the evidence identified andsynthesised by each review.Regardless of how eligibility criteria have been defined,screening for eligibility may be hampered by poorreporting. The HTA report [8] found that the levels of reporting within primary studies varied dramatically interms of whether (i) the results of the PFT (for assess-ment of prognostic factor) were reported, (ii) resultswere dichotomized ( ‘ resistant ’  and  ‘ sensitive ’ ), (iii) car-diovascular outcomes were reported, and (iv) outcomeswere linked to  ‘ resistant ’  and  ‘ sensitive ’ groups. Not all of this information was necessarily discernable from theabstract, thus necessitating a detailed reading of wholearticles at an early screening stage.Differences in selection criteria can lead to differentincluded studies, and as a result, different conclusions asto the prognostic utility of a factor of interest. Evenwhere included, studies are consistent across reviews, thequality of reporting within included studies may also influ-ence any conclusions drawn by review authors. Quality assessment of included studies Quality assessment of primary prognostic studies is a de- veloping field, and the different approaches used acrossthe eight reviews may in part be a reflection of theirpublication date. Both quality assessment (where under-taken) and use of quality findings in interpreting resultswere highly variable within this example (see Table 2).Three reviews used an item related to study quality as aneligibility criterion, e.g.  ‘ investigators blinded to patients ’ aspirin status ’  [11] which is in contrast to effectivenessquestions where quality assessment can, but does not usu-ally, form part of the study selection process. This may beproblematic where such quality items have not been re-ported in the article.Prognostic studies may be of varying study design, e.g.cohort or case – control design, or one arm of an RCT,and may include a diagnostic test (such as a PFT) foridentifying prognostic factors, and this is reflected in thechoice of different quality assessment tools. Two of thereviews (Wisman et al. [16] and the HTA report [8]) modified one or more checklists to include items rele- vant to the particular topic area. The HTA report basedquality assessment on QUADAS (quality assessment of diagnostic accuracy studies) [19] for assessing quality of test accuracy studies and guidelines for appraising prog-nostic studies [20], the latter of which has since beenfurther developed into the Quality in Prognosis Study (QUIPS) tool [22]. It is important that quality assess-ment focuses on those aspects relevant to the prognosisquestion rather than be (solely) led by the study design,particularly where the prognostic aspect may not havebeen the main or only focus of the study. Presentation of results Primary studies mainly reported results as dichotomousfrequency data, i.e. the number of clinical events in  ‘ as-pirin resistant ’  (prognostic factor positive) and  ‘ aspirinsensitive ’  (prognostic factor negative) arms, where athreshold value was used to define the two groups.Fewer studies reported adjusted/unadjusted odds ratios(OR) and/or hazard ratios (HR). The extent to whichdata was transformed for use in meta-analysis (or pres-entation in forest plots) was variable across reviews. All Table 1 Search terms used in the systematic reviews Search termsrelating toSystematic reviews*HTA report [8] Li et al. [15] Pusch et al. [12] Sofi et al. [14] Krasopoulos et al. [11]Wisman et al. [16] + Canivano andGarcia [10] Aspirin  ✓ ✓ ✓ ✓ ✓ ✓ ✓ Resistance  ✓ ✓ ✓ ✓ ✓ ✓ ✓ Platelet (function)  ✓ ✓ ✓ ✓ ✓ ✓ Outcomes/condition  ✓ ✓ ✓ ✓ Names of PFTs  ✓ Prognosis/prediction  ✓ Filter for study design *Search terms not listed in the systematic review by Snoep et al. [13].  + Not clear if all search terms listed. Dretzke  et al. Systematic Reviews  2014,  3 :140 Page 4 of 10  reviews calculated RRs or ORs from frequency data.The test results for the prognostic factor were in theform of continuous data (i.e. level of platelet aggrega-tion), but these were frequently dichotomised into posi-tive/negative using a threshold, or presented as tertilesor quartiles. There was often no explanation regardingthe choice of threshold.The HTA report [8] presented results for multiple(PFT) thresholds for designating  ‘ aspirin resistance ’  (pres-ence of prognostic factor) where reported; calculatedthresholds where PFT test results were reported as ter-tiles or quartiles by collapsing these into two groups:presented HRs where available and presented adjustedresults where available. Adjusted results can revealwhether a test has prognostic utility over and aboveother prognostic factors. None of the included studiesreported prognostic models for individual risk predic-tion (i.e. a model containing PFT results and otherprognostic factors for absolute risk prediction), so find-ings were limited to an average association between theprognostic factor and outcome. Few systematic reviewsconsidered adjusted results or the potential importanceof models.Time-to-event analyses may be more appropriate whenaccounting for different lengths of follow-up; HRs werepresented in the HTA report [8], which also calculatedHRs from other available data where possible (usingmethods described by Parmar et al. [23] and Perneger[24]). In contrast, two of the other reviews (Li et al.[15], Sofi et al. [14]) converted HRs to relative risks (RR) for inclusion in synthesis using meta-analysis. Thereis a balance between maximising data for analyses (im-proving effect estimates) and the assumptions that have tobe made in order to do this (limiting conclusions that canbe drawn). Presenting the same data using as many out-come statistics as reported or calculable does allow explor-ation of whether results are sensitive to the choice of outcome statistic.Five reviews undertook meta-analyses on variously de-fined composite cardiovascular outcomes (major adversecardiac events (MACE)) [15], composite cardiovascularendpoints [16], clinical ischaemic events [13] or any car- diovascular events [11,14]) while only one [13] also pre- sented individual outcomes (re-occlusion and myonecrosisafter PCI) (fixed or random effects assumptions were usedand a pooled RR or OR presented (see Table 3). All main Table 2 Quality assessment approaches Review Quality assessment undertakenand methodFindingspresented/use of summary scoreFindings used in contextof results/sensitivityanalysisComment Canivano and Gracia [10] None N/A N/AHTA report [8] Quality assessment tool derivedfrom QUADAS [19] and the Haydenchecklist relating to prognosticstudies [20]Results of the qualityassessment werepresentedImpact commented onbut sensitivity analyses notdeemed possible.Krasopoulos et al. [11] Study eligibility criterion:investigators to be blinded topatients ’  aspirin statusQuality rating for risk of bias (A to D) butnot explicit on howthis was derivedNo Terminology used wasconfusing (e.g.  ‘ allocationof blindness ’  and  ‘ compliancewith blindness ’ ). The term ‘ allocation concealment ’  usedin the context of observationalstudies is not appropriateLi et al. [15] Study eligibility criterion: onlythose studies with verifiedcompliance. Newcastle-Ottawachecklist [21] for cohort studiesFindings presented NoPusch et al. [12] None N/A N/ASofi et al. [14] Study eligibility criterion:prospective study designN/A N/ASnoep et al. [13] Quality criteria relating to: controlfor confounders, measurement of exposure, completeness of follow-upand blinding, and, for case – controlstudies, matching and case definitionNo NoWisman et al. [16] Modified QUADAS tool [19] (for quality assessment of diagnosticaccuracy studies). 11 items assessedFindings presented Sensitivity analysis.Studies scoring  ‘ low risk of bias ’  on eight or moreof the quality items wereconsidered to begood quality Dretzke  et al. Systematic Reviews  2014,  3 :140 Page 5 of 10
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!