Open Collections

UBC Faculty Research and Publications

Large-scale external validation and comparison of prognostic models: an application to chronic obstructive… Guerra, Beniamino; Haile, Sarah R; Lamprecht, Bernd; Ramírez, Ana S; Martinez-Camblor, Pablo; Kaiser, Bernhard; Alfageme, Inmaculada; Almagro, Pere; Casanova, Ciro; Esteban-González, Cristóbal; Soler-Cataluña, Juan J; de-Torres, Juan P; Miravitlles, Marc; Celli, Bartolome R; Marin, Jose M; ter Riet, Gerben; Sobradillo, Patricia; Lange, Peter; Garcia-Aymerich, Judith; Antó, Josep M; Turner, Alice M; Han, Meilan K; Langhammer, Arnulf; Leivseth, Linda; Bakke, Per; Johannessen, Ane; Oga, Toru; Cosio, Borja; Ancochea-Bermúdez, Julio; Echazarreta, Andres; Roche, Nicolas; Burgel, Pierre-Régis; Sin, Don D; Soriano, Joan B; Puhan, Milo A Mar 2, 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12916_2018_Article_1013.pdf [ 936.09kB ]
JSON: 52383-1.0364137.json
JSON-LD: 52383-1.0364137-ld.json
RDF/XML (Pretty): 52383-1.0364137-rdf.xml
RDF/JSON: 52383-1.0364137-rdf.json
Turtle: 52383-1.0364137-turtle.txt
N-Triples: 52383-1.0364137-rdf-ntriples.txt
Original Record: 52383-1.0364137-source.json
Full Text

Full Text

RESEARCH ARTICLE Open AccessLarge-scale external validation andcomparison of prognostic models: anapplication to chronic obstructivepulmonary diseaseBeniamino Guerra1, Sarah R. Haile1, Bernd Lamprecht2,3, Ana S. Ramírez4, Pablo Martinez-Camblor5,Bernhard Kaiser6, Inmaculada Alfageme7, Pere Almagro8, Ciro Casanova9, Cristóbal Esteban-González10,Juan J. Soler-Cataluña11, Juan P. de-Torres12, Marc Miravitlles13, Bartolome R. Celli14, Jose M. Marin15,Gerben ter Riet16, Patricia Sobradillo17, Peter Lange18, Judith Garcia-Aymerich19, Josep M. Antó20, Alice M. Turner21,Meilan K. Han22, Arnulf Langhammer23, Linda Leivseth24, Per Bakke25, Ane Johannessen26, Toru Oga27,Borja Cosio28, Julio Ancochea-Bermúdez29, Andres Echazarreta30, Nicolas Roche31, Pierre-Régis Burgel32,Don D. Sin33, Joan B. Soriano34,35, Milo A. Puhan36,37* and for the 3CIA collaborationAbstractBackground: External validations and comparisons of prognostic models or scores are a prerequisite for their usein routine clinical care but are lacking in most medical fields including chronic obstructive pulmonary disease(COPD). Our aim was to externally validate and concurrently compare prognostic scores for 3-year all-causemortality in mostly multimorbid patients with COPD.Methods: We relied on 24 cohort studies of the COPD Cohorts Collaborative International Assessment consortium,corresponding to primary, secondary, and tertiary care in Europe, the Americas, and Japan. These studies includeglobally 15,762 patients with COPD (1871 deaths and 42,203 person years of follow-up). We used networkmeta-analysis adapted to multiple score comparison (MSC), following a frequentist two-stage approach; thus,we were able to compare all scores in a single analytical framework accounting for correlations among scoreswithin cohorts. We assessed transitivity, heterogeneity, and inconsistency and provided a performance ranking of theprognostic scores.Results: Depending on data availability, between two and nine prognostic scores could be calculated for each cohort.The BODE score (body mass index, airflow obstruction, dyspnea, and exercise capacity) had a median area under thecurve (AUC) of 0.679 [1st quartile–3rd quartile = 0.655–0.733] across cohorts. The ADO score (age, dyspnea, and airflowobstruction) showed the best performance for predicting mortality (difference AUCADO – AUCBODE = 0.015[95% confidence interval (CI) = −0.002 to 0.032]; p = 0.08) followed by the updated BODE (AUCBODE updated – AUCBODE= 0.008 [95% CI = −0.005 to +0.022]; p = 0.23). The assumption of transitivity was not violated. Heterogeneity acrossdirect comparisons was small, and we did not identify any local or global inconsistency.(Continued on next page)* Correspondence: miloalan.puhan@uzh.ch36Epidemiology, Biostatistics and Prevention Institute, University of Zurich,Hirschengraben 84, Room HRS G29, CH -8001 Zurich, Switzerland37Epidemiology & Department of Epidemiology, Johns Hopkins BloombergSchool of Public Health, Baltimore, MD, USAFull list of author information is available at the end of the article© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (, which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver( applies to the data made available in this article, unless otherwise stated.Guerra et al. BMC Medicine  (2018) 16:33 from previous page)Conclusions: Our analyses showed best discriminatory performance for the ADO and updated BODE scores in patientswith COPD. A limitation to be addressed in future studies is the extension of MSC network meta-analysis to measuresof calibration. MSC network meta-analysis can be applied to prognostic scores in any medical field to identify the bestscores, possibly paving the way for stratified medicine, public health, and research.Keywords: COPD, Prognostic scores, Large-scale external validation, Performance comparison, Network meta-analysisBackgroundPrognostic scores, commonly based on coefficients fromregression models, provide a probability of a certain ad-verse outcome for an individual over a specified timehorizon. Prognostic scores have become increasinglypopular over the last two decades [1–5]. They servemultiple purposes such as informing individuals andhealth care providers about disease and outcome risks,supporting risk-stratified and personalized prevention ortreatment decisions, identifying participants for research,or adjusting for confounding [6–9].Numerous prognostic models have been developedin various fields of medicine [10–13]. Just for predict-ing the risk of cardiovascular disease in the generalpopulation, a recent review identified 363 prognosticmodels or scores [14]. For patients with chronicobstructive pulmonary disease (COPD), prognosticscores have been developed mostly to predict the riskof death [15–30], but scores also exist to predictexacerbations [31] or deteriorating of health-relatedquality of life [27, 32].Major obstacles for using prognostic scores in practiceand research are, however, the frequent lack of external val-idations, comparisons of their predictive performance, andassessments of their applicability in practice [2, 33–38].Practitioners and researchers are left with uncertaintyabout which prognostic score to use and may be reluctantto use them at all [39]. Ideally, prognostic scores would beexternally validated in several different populations andtheir performance summarized [40, 41]. However, such ex-ternal validations and concurrent comparisons are rarelyperformed [42]. In addition, for even more comprehensivecomparison, the performance of prognostic scores may becompared indirectly using common comparator scoressimilar to network meta-analysis (NMA) [43–48] ofrandomized trials.Our aim was to use multiple score comparison(MSC) in order to externally validate and concurrentlycompare prognostic scores for 3-year mortality in patientswith COPD.MethodsWe followed a prespecified study protocol and describedthe detailed statistical methods elsewhere [43].Study design and participantsThis study was based on 26 cohort studies of the COPDCohorts Collaborative International Assessment (3CIA)consortium. Details have been reported elsewhere (andsummarized in Table 2) [49]. All cohorts were approvedby ethics committees, and participants gave writteninformed consent [49]. We also included the Phenotypeand Course (PAC)-COPD and Copenhagen cohorts inthe final database, even if they were used in the large-scale update of the ADO (age, dyspnea, and airflowobstruction) index [15]. We considered this approachreasonable, since they form only a small part of the finaldatabase, but we verified in a sensitivity analysis if theyaffected the results.Prognostic scoresStarting from the literature review of two studies [32, 42]and searching among their references, PubMed-related ar-ticles, and through our research network, we identified 19prognostic scores, of which we included 10 in our analysis.The scores (see Table 1 for details) were the BODE (bodymass index, airflow obstruction, dyspnea, and severe exac-erbations) [17], updated BODE [16], ADO ( we includedin the analysis only the updated ADO index and not theoriginal ADO index [16] because the updated ADO wasgenerated from large-scale external validation; however,we will name it simply ADO) [15], eBODE (severe acuteexacerbation of COPD plus BODE) [18], BODEx (bodymass index, airflow obstruction, dyspnea, severe acute ex-acerbation of COPD) [18], DOSE (dyspnea, obstruction,smoking and exacerbation frequency) [27], SAFE (SaintGeorge’s Respiratory Questionnaire (SGRQ) score, air-flow limitation and exercise capacity) [28], and B-AE-D(body mass index, acute exacerbations, dyspnea; we usedthe optimized version and not the original B-AE-D score)[23]. The Global Initiative for Chronic Obstructive LungDisease (GOLD) classification [50, 51] and the 2011–2016GOLD classification (often referred to as new GOLD inthe recent COPD literature) [51] were also used in theanalysis, even if they were not designed for prognosticpurposes. Apart from original ADO and original B-AE-Dscore the other seven identified scores from the literaturewere excluded from the analysis, since our database didnot include at least one of their predictors or did not in-clude them simultaneously in at least one cohort.Guerra et al. BMC Medicine  (2018) 16:33 Page 2 of 13Table1ScoringrulesofprognosticscorestopredictmortalityinpatientswithCOPDScorePredictorGOLD[50,51]GOLD(2011–2016)[51]BODE[17]BODEupd.[16]ADO[15]e-BODE[18]BODEx[18]DOSE[27]SAFE[28]B-AE-D[23]BMI0(>21)0(>21)0(>21)0(>21)0(>=21)1(<=21)1(<=21)1(<=21)1(<=21)6(18.5–21)9(<18.5)FEV1%pred.0(>=80%)0(ifFEV1pp>=50and<=1exacerbationsperyear)0(>=65%)0(>=65%)0(>=81%)0(>=65%)0(>=65%)0(>=50%)0(>=80%)1(50–79%)2(otherwise)1(50–64%)1(36–64%)1(65–60%)1(50–64%)1(50–64%)1(31–49%)1(50–79%)2(30–49%)2(36–49%)2(<=35)2(51–64%)2(36–49%)2(36–49%)2(<=30)2(36–49%)3(<30%)3(<=35)3(35–50%)3(<=35)3(<=35)3(<=35)4(<=35%)mMRC0(ifmMRC>=2andCAT>=10)0(0–1)0(0–1)0(0)0(0–1)0(0–1)0(0–1)0(0–2)1otherwise1(2)1(2)1(1–2)1(2)1(2)1(2)6(3)2(3)2(3)2(3)2(3)2(3)2(3)10(4)3(4)3(4)3(4)3(4)3(4)3(4)6MWT(m)0(>=350)0(>=350)0(>=350)0(>=400)1(250–349)4(250–349)1(250–349)1(300–399)2(150–249)7(150–249)2(150– 249)2(200–299)3(<=149)9(<=149)3(<=149)3(<=199)Age(years)0(40–49)2(50–59)4(60–69)5(70–79)7(>=80)Prev.exacerbation(SeeFEV1pp)0(0)0(0)0(0–1)0(0)1(1–2)1(1–2)1(2–3)3(1)2(>2)2(>2)2(>3)7(>=2)CAT(SeemMRC)Smoking0(non-smoker)1(currentsmoker)Guerra et al. BMC Medicine  (2018) 16:33 Page 3 of 13Table1ScoringrulesofprognosticscorestopredictmortalityinpatientswithCOPD(Continued)ScorePredictorGOLD[50,51]GOLD(2011–2016)[51]BODE[17]BODEupd.[16]ADO[15]e-BODE[18]BODEx[18]DOSE[27]SAFE[28]B-AE-D[23]Qualityoflife(SGRQ)0(<=30)1(31–49)2(50–64)3(>=65)Totalscore0–30–30–100–150–140–120–90–80–90–26Abbreviations:BMIbodymassindex,FEV1%pred.forcedexpiratoryvolumein1spercentagepredicted,mMRCmodifiedMedicalResearchCouncildyspneascale,6MWT6-minwalktest,CATCOPDAssessmentTest,SGRQSaintGeorge’sRespiratoryQuestionnaire;previousexacerbationsarereferredtothepreviousyear,GOLDGlobalInitiativeforChronicObstructiveLungDisease,BODEbodymassindex,airflowobstruction,dyspneaandsevereexacerbations,BODEupd.BODEupdated,ADOage,dyspnea,airflowobstruction(weuseinouranalysistheupdatedversionoftheADOscore),e-BODEsevereacuteexacerbationofCOPDplusBODE,BODExbodymassindex,airflowobstruction,dyspnea,severeacuteexacerbationofCOPD,DOSEdyspnea,obstruction,smoking,andexacerbationfrequency,SAFESaintGeorge’sRespiratoryQuestionnaire(SGRQ)score,air--flowlimitationandexercisecapacity,B-AE-Dbodymassindex,acuteexacerbations,dyspnea(weusetheoptimizedversionofthescore,introducedinthesamepaper).MissingcellscorrespondtovariablesthatdonotconstitutethescoreofthecorrespondentcolumnGuerra et al. BMC Medicine  (2018) 16:33 Page 4 of 13Outcome and performance measure for externalvalidation and comparison of prognostic scoresWe evaluated a number of performance measurescommonly used to assess the prognostic properties ofprediction models and scores [43]. We deemed the areaunder the curve (AUC) to be the most appropriate per-formance measure for our purposes, mainly because itsrange is independent of the data, it is easy to interpret,and an analytic formula for its variance is available [52].Statistical analysisWe followed a prespecified study protocol. We first per-formed direct head-to-head comparisons using randomeffects meta-analysis and then examined the networkevidence merging all available direct and indirect evi-dence [53]. We used a novel methodology, i.e., MSCmeta-analysis, adapted from multiple treatment com-parison network meta-analysis [54, 55]. Methodologicaldetails are reported in the section “Detailed Methods” inAdditional file 1 and in a recent paper [43]. R codes areavailable (provided in the section “R Code for MSCmeta-analysis” in Additional file 1).Direct comparisons (random effects pairwise meta-analysis)We directly compared prognostic scores by pairwise ran-dom effects meta-analysis [56, 57]. We used forest plotsto visually investigate statistical heterogeneity as well asthe I2 statistic. Such standard meta-analysis has limita-tions, since it does not take into account the correlationsamong multiple scores evaluated on the same set of pa-tients [58], and it does not give a clear indication ofwhich prognostic score performs best. Thus, we adoptednetwork meta-analysis, an approach that allowed us toweight and then pool the results coming from differentcohorts.MSC meta-analysisMethodological details are reported in detail in [43]. Inbrief, we used an example of implementation of networkmeta-analysis for treatment effectiveness comparison[54], adapting it to our purposes, namely to concurrentlyexternally validate and compare prognostic scores fromindividual patient data across different cohorts [43]. Wehave explicitly included correlations [58] between thescores on a cohort level. We use a frequentist two-stagemeta-regression model, as proposed in [54]:1. Ordinary meta-analysis (stage I) to obtain the directestimates for pooled differences in AUC (using theinverse-variance weighted means of the correspondingcohorts). The meta-analyses were done within eachgroup of cohorts where data for the same prognosticscores were available.2. In stage II, we merged the estimates for thedifferences in AUC from the groups of cohorts,looking for the weighted least squares solution tothe regression problem equation. Based on the directestimates and their variances from the first stage, weestimated the pooled differences in AUC that obeyedfundamental consistency equations. Thus in stage II,the stage I estimates for the differences in AUC werecombined across groups of cohorts to give overallperformance estimates for the entire network.In order to provide a ranking of the scores, we used a fre-quentist version of the surface under the cumulative rankingcurve (SUCRA) [59, 60] score showing the likelihood of thescore to be better than any other score and summarizingrelative performances and confidence intervals.The last steps were to ensure that the heterogeneity,transitivity, and consistency assumptions were met [46].Heterogeneity in the MSC analysis was evaluated by thepooled heterogeneity variance among groups (τ2pooled). Weassessed “transitivity” through analysis of variance(ANOVA) tests. Thus, we assessed the comparability ofthe cohorts across whom the predictive performance of ascore may vary because of a “spectrum effect” [61] or“case mix” [37, 62, 63]. We also assessed consistency [46]between direct evidence and MSC meta-analysis estimatesusing the Q likelihood-ratio test statistic to evaluate theglobal consistency and analysis of residuals and leveragesto evaluate the local consistency [54]. For more details,see “Detailed Methods” in Additional file 1 and [43].Handling of missing dataIf a variable was missing for > 30% of the patients, wediscarded the specific variable for that particular specificcohort, since the effects of such predictors could be gen-erally distrusted [1]. Otherwise we performed multipleimputation with chained equations (the analysis of thepatterns of missingness allowed us to consider the miss-ing data missing completely at random apart from thedependence on the cohort) [4]. We combined the esti-mates of the 30 different analyses (one for each imputeddataset, for each of which we followed all the previouslyhighlighted frequentist two-stage meta-regression modelapproaches) using Rubin’s rules.ResultsCohort and participant characteristicsThe cohorts varied greatly in terms of geographic loca-tion, sample size, and number of events and included abroad spectrum of patients with COPD from primary,secondary, and tertiary care settings (Table 2). Meanforced expiratory volume in 1 s percentage (FEV1) rangedfrom 30 to 70% of the predicted values, mean modifiedMedical Research Council (mMRC) dyspnea scores fromGuerra et al. BMC Medicine  (2018) 16:33 Page 5 of 13Table2StudycharacteristicsCohortNumberofEventsNumberofpatientsPersonyearsMeanage,yearsMen,%MeanFEV1%pred.MeanmMRCPastexacerbators,%Meanno.prev.exacerbationsCurrentsmoker,%MeanBMIMean6MWT,mMeanSGRQMeanCATCOPDgene337448410,60363(9)5657.4(22.8)1.50.164327.9(6.1)376.1(124.1)36.9(22.9)Sevillaa205596156266(10)9543.5(13.3)10.251.162429.2(5.7)Copenhagenb1862287661861(9)5470.5(23.7)1.37125(4.2)Genkols126954270865(10)6146.9(17),c5218639671(9)9944.5(16.5)10.160.621728.1(5.2)380.1(111.9)ICECOLDERIC47400107167(10)5755.3(16.5),mMRCmodifiedMedicalResearchCouncil(MMRC)dyspneascale;pastexacerbatorsaredefinedaspatientswithmorethanoneexacerbationinthepreviousyear;meannumberpreviousexacerbationsarereferredtothepreviousyear,BMIbodymassindex,6MWT6-minwalktest,SGRQSaintGeorge’sRespiratoryQuestionnaire,CATCOPDAssessmentTestThecohortsarepresentedindecreasingorderofnumberofevents.Mostofthevariablesavailableprovidedbythe3CIAcollaborationforthedifferentcohortsareshown.Inparticular,weshowallthevariablesconstitutingthescoresanalyzedinourstudy.Wepresentthestandarddeviationforallindividualvariableswhosedistributionisapproximatelynormal;thisisnotthecaseforcount(withsmallnumbers)orcategoricalvariables,likenumberofpreviousexacerbationsormMRC)a CohortsbelongingtotheCollaborativeCohortstoAssessMulticomponentIndicesofCOPDinSpain(COCOMICS)collaborationbCohortsbelongingtotheADOcollaboration.Forinformationconcerningthecohorts,see[49]c SincenoneofthescorescouldbeevaluatedinthecohortRequenaI(mainlybecausethevariabledyspnoeawasmissingfor95%ofthepatients,i.e.,for165outof174patients),thiscohortwasexcludedfromtheanalysisdSincetherewasnoeventinafollow-upof3years,thecohortA1ATDwasexcludedfromtheanalysisMissingcellscorrespondtovariablesthatarecompletelymissinginthecohortofthecorrespondentrowGuerra et al. BMC Medicine  (2018) 16:33 Page 6 of 131.0 to 2.8 (the scale goes from 0 to 4, with 4 being theworst), mean number of exacerbations in the previousyear (where available) from 0.2 to 1.7, and mean 6-minwalk distance (where available) from 218 to 487 m.Direct comparisons of prognostic scores for mortality inpatients with COPDThe direct comparisons are shown in the upper-righttriangle of Table 3, i.e., a league table (that also includesthe MSC meta-analysis in the lower-left triangle). Forty-one direct comparisons of the AUC of prognostic scoreswere possible; indeed, no direct evidence was availablefor the comparison between SAFE and the eBODE,BODEx, DOSE, and B-AE-D scores (cells D6, E6, F2,G6, I10 in the league Table 3).The updated BODE score performed statistically signifi-cantly better than GOLD, new GOLD, and the B-AE-Dscores, whereas the AUC of the updated BODE score washigher than for the other scores but not statistically signifi-cantly so. We deemed overall statistical heterogeneity ofdirect comparisons moderate. However, in our MSC meta-analysis the direct comparisons should be interpreted withcaution, since they do not take into account that multiplescores were evaluated on the same set of patients and arethus likely to bias the interpretation of which prognosticscore performs best [58].Groups of cohorts evaluating the same prognostic scoresGrouping of cohorts where the same prognostic scorescould be calculated was the first step to consider corre-lations introduced by predictions performed on thesame sample of patients. Figure 1 shows the grouping ofcohorts. In group 1 (constituting four cohorts:Copenhagen, HUNT, Japan, SEPOC, as shown in Fig. 1)information on FEV1, age, and dyspnea was available tocalculate the GOLD and ADO scores for each partici-pant. In contrast, group 6 consisted of four cohorts (LaPrincesa Madrid, Requena II, Tenerife, Terrassa II)where nine prognostic scores (all except for the SAFEscore) could be calculated for each participant. Figure 1provides a visual representation of these groups togetherwith the number of events (i.e., deaths). For example,the dark green line represents group 1 where the GOLDand ADO scores could be compared against each other.The closed polygons show the comparisons that are pos-sible for each group of cohorts. Group 6 is represented bythe dark yellow polygon that includes nine scores. Thus,unlike multiple treatment network meta-analyses, whereusually two or at most three treatments are compared ineach trial, Fig. 1 shows that in each of the cohorts of ourdatabase we can compare between two and nine prognos-tic scores.MSC meta-analysis of prognostic scores to predict 3-yearmortality in patients with COPDThe lower-left part of Table 3 shows all comparisons be-tween the AUCs of the 10 prognostic scores taking intoaccount the correlation among multiple comparisons forthe same patients as well as direct and indirect evidenceof the entire network (Fig. 1). The median AUC of theGOLD classification of airflow obstruction severity was0.613 (interquartile range 0.587 to 0.637) and is shown inboldface in the upper-left cell as an anchor to interpretthe differences in AUC between the prognostic scores.Compared to GOLD, all prognostic scores showed statisti-cally significantly higher AUCs except for the B-AE-D andGOLD 2011–2016 (cells B1-L1 in Table 3). Compared tothe BODE score (the most commonly used prognosticscore in COPD, median AUC 0.679 [interquartilerange 0.655 to 0.733]), the ADO, updated BODE, andeBODE showed higher AUCs, whereas all other scoresperformed worse.Figure 2 shows the comparisons of all scores against theBODE score and that the ADO score and the updatedBODE performed better than the other scores (i.e.,AUCADO – AUCBODE = +0.015 [95% CI –0.002 to 0.032],p = 0.08; AUCBODE updated – AUCBODE = 0.008 [95% CI =−0.005 to +0.022]; p = 0.23). The sensitivity analysis under-taken excluding from the database the two cohorts used inthe large-scale update of the ADO index [15] shows nosignificant differences.Heterogeneity, transitivity, and inconsistencyGlobal heterogeneity was relatively small (τ2pooled =0.00011) (we did not use a τ2 for each group (τ2g) since thisis not recommended when there are groups with a singlecohort [54]). The groups of the MSC meta-analysis werebalanced with regard to characteristics of the different co-horts that may modify the predictive performance of thescores (all a priori defined characteristics that were gener-ating case mix were not statistically significantly differentacross groups), and we could thus assume transitivity.The consistency analyses did not suggest local or globalinconsistency. Visual analysis of the Q-Q plot and studen-tized residuals indicated robust local consistency. Thelikelihood-ratio test statistic showed overall consistency(Q likelihood-ratio test = 25.29 ≅ χ2(0.95, 16) = 26.30, pvalue = 0.06).DiscussionOur study has two main findings. Firstly, our resultsindicate that the ADO index has the best ability to predict3-year mortality in patients with COPD, followed by theupdated BODE and eBODE indices. Given its simplicity,the ADO index may be the most attractive optionacross care settings to inform patients and healthcare professionals about prognosis and to informGuerra et al. BMC Medicine  (2018) 16:33 Page 7 of 13Table3Leaguetablepresentingthemultiplescorecomparison(MSC)meta-analysis(lower-lefthalfofthetable)andthedirectrandomeffectsmeta-analysis(upper-righthalfofthetable)Directmeta-analysisMSCmeta-analysisGOLDB-AE-DGOLD(2011–2016)DOSEBODExSAFEeBODEBODEBODEupdatedADO12345678910GOLDAAUC=0.613(1stQu.0.587,3rdQu.0.637)ΔAUC=0.030(−0.005,0.065)0.017(−0.005,0.038)0.036(0.008,0.064)0.054(0.029,0.079)0.047(0.027,0.068)0.064(0.004,0.123)0.071(0.040,0.102)0.080(0.041,0.119)0.090(0.072,0.109)B-AE-DBΔAUC=0.010(−0.010,0.031)−0.004(−0.025,0.017)0.004(−0.012,+0.020)0.025(0.011,0.039)NA0.069(−0.016,−0.121)0.079(0.004,0.154)0.082(−0.012,−0.152)0.076(0.051,0.101)GOLD(2011–2016)C0.012(−0.001,0.024)0.001(−0.019,0.021)0.009(−0.002,0.021)0.028(0.017,0.039)0.055(0.038,0.072)0.047(0.016,0.079)0.059(0.046,0.073)0.051(0.027,0.076)0.067(0.053,0.080)DOSED0.022(0.006,0.037)0.011(−0.007,−0.029)0.010(−0.005,0.025)0.018(0.008,−0.029)NA0.039(0.007,0.070)0.033(−0.000,0.065)0.043(−0.002,0.088)0.061(−0.044,−0.079)BODExE0.041(0.027,0.055)0.030(0.014,0.047)0.029(0.015,0.043)0.019(0.005,0.033)NA0.030(−0.001,0.061)0.031(−0.017,0.079)0.039(−0.028,0.105)0.050(0.034,0.066)SAFEF0.061(0.034,0.087)0.050(0.018,0.082)0.049(0.022,0.076)0.039(0.010,0.068)0.020(−0.009,0.048)NA0.011(−0.000,0.023)0.005(−0.009,0.018)−0.007(−0.029,0.015)eBODEG0.065(0.046,0.085)0.055(0.032,0.078)0.054(0.034,0.074)0.044(0.023,0.064)0.024(0.007,0.042)0.005(−0.025,0.035)−0.001(−0.020,0.017)0.002(−0.031,0.034)0.024(−0.018,0.066)BODEH0.068(0.052,0.084)0.057(0.034,0.080)0.056(0.039,0.074)0.046(−0.027,−0.065)0.027(0.009,0.045)0.007(−0.019,0.034)0.003(−0.016,0.021)0.005(−0.006,0.017)−0.004(−0.023,0.016)BODEupd.I0.076(0.058,0.095)0.066(0.041,0.091)0.065(0.045,0.085)0.055(−0.033,−0.076)0.036(0.015,0.056)0.016(−0.012,0.043)0.011(−0.009,0.031)0.008(−0.005,0.022)−0.005(−0.032,0.022)ADOL0.083(0.070,0.096)0.072(0.052,0.093)0.071(0.056,0.087)0.070(−0.052,−0.089)0.042(0.026,0.058)0.022(−0.005,0.050)0.018(−0.003,0.038)0.015(−0.002,0.032)0.007(−0.012,0.026)Abbreviations:AUCareaunderthecurve;thelower-lefthalfofthetablereferstotheMSCmeta-analysis.Theupper-righthalfofthetablereferstodirectcomparisonsusingconventionalrandomeffectsmeta-analysis.Thefirstcell(firstrow,firstcolumn)givesareferencevalue(inboldface),namelythemedianand1stand3rdquartilesoftheAUCoftheGOLDclassificationacrosscohortsasananchortointerpretthedifferencesinAUCbetweentheprognosticscores.Ineveryothercell,eachpairofscoresiscomparedusingthedifferenceinAUC.Lower-lefthalfofthetablewereportinthecorrespondentcellthedifferencebetweentheAUCsofthescoreintherowandthescoreinthecolumn;instead,fortheupper-righthalfofthetablewereportthedifferencebetweentheAUCsofthescoreinthecolumnandthescoreintheroworthe.WedecidedforthisrepresentationtomakeavisualcomparisonbetweendirectandMSCcomparisoneasier;inthisway,itisenoughtolookatcorrespondingvaluesmirroredatthemaindiagonal.The95%confidenceintervalisindi-catedinparentheses.Forbetterreadabilityofthetablethesign“+”isomitted,whilethesign“–”isindicatedGuerra et al. BMC Medicine  (2018) 16:33 Page 8 of 13treatment decisions whose effectiveness may dependon life expectancy. Secondly, we presented a com-prehensive approach for external validation and con-current comparison of prognostic scores and its firstapplication. MSC meta-analysis is a method adaptedfrom network meta-analysis that meets the call fornew approaches for external validation andconcurrent comparison of risk prediction models andscores that should take advantage of data sharing,individual patient data (IPD), and advanced analyticaltechniques [36, 37, 45, 64, 65].In practice, the GOLD score using just lung function isstill used most commonly to grade disease severity, whichis traditionally related to prognosis as in other fields (e.g.,Fig. 1 Network plot. Network representing which prognostic scores belong to the different groups. Each node represents a score and eachclosed polygon represents a group of cohorts where the same prognostic scores are available. The thickness of the lines represents the totalnumber of deaths in the specific groupFig. 2 Comparison of AUC of prognostic scores. Difference in AUC (shown with confidence interval with 95% confidence level) among thedifferent scores and the BODE index (chosen here as the reference score) in the MSC meta-analysis. As a reference we use the median of theAUC of the BODE score 0.679 (1st Qu. 0.655, 3rd Qu. 0.733)Guerra et al. BMC Medicine  (2018) 16:33 Page 9 of 13cancer). FEV1% pred. (thus, GOLD classification) is animportant parameter at the population level in the predic-tion of important clinical outcomes such as mortality andhospitalization. The revised combined COPD assessmentand their further developments integrate the severity ofairflow limitation assessment, also providing informationregarding symptom burden and risk of exacerbation [51].However, the results of our analysis show that, when theaim is to predict mortality in individuals, other scoressuch as ADO, updated BODE, and eBODE are substan-tially better than the GOLD classifications (in our analysis,GOLD and GOLD 2011–2016). We note that the AUCfor the best score (ADO) is 0.69, a moderately good dis-criminative performance; however, we can often not ex-pect a much higher discriminative performance in clinicalsettings (for instance, see [31]).The predictive performance of a prognostic score is im-portant, but it is not the only criterion for choosing aprognostic score for practice. Indeed, with an eye towardsapplicability, the time, cost, and burden for patients andpractitioners to measure the predictors of a prognosticscore should be taken into consideration [66]. We deem aprognostic score such as ADO to be easily available if itonly includes simple questions, easily available informa-tion from medical charts, and spirometry (performed forthe diagnosis of COPD) [50, 51].Scores to predict mortality are also useful beyond esti-mating prognosis. Nowadays, no treatments to lower therisk of mortality are currently available for patients withCOPD; thus, for this outcome, prediction scores cannotprovide risk-stratified treatment guidance. However, prog-nostic scores may help to make randomized trials with all-cause mortality as primary outcome more efficient thanprevious trials by only including patients at higher risk[67]. Also, prognostic scores for all-cause mortality are par-ticularly attractive for multimorbid patients such as COPDpatients, where cardiovascular disease, diabetes, renal dis-ease, and lung cancer, among other conditions, also contrib-ute to mortality [68, 69]. Patients with COPD often receiveless than optimal prevention and treatment of cardiovascu-lar disease, which may partly reflect a therapeutic nihilism.Of course, there are patients who are unlikely to benefitfrom long-term cardiovascular prevention because of shortlife expectancy. However, a prognostic score provides abetter basis for decisions on cardiovascular prevention,lung cancer screening, or other treatments and may limitunder- and over-treatment in COPD [1, 70, 71].Many prognostic models and scores (as in the models’simplified forms) are never validated in practice, andmany investigators develop a second model instead ofrelying on existing scores at least as a starting point.Such practice has led to numerous prognostic scores forthe same conditions that are left without external valid-ation. Thus, we introduced MSC meta-analysis, whichaddresses the lack of external validation and compari-sons of prognostic scores by comparing their predictiveperformance in external validation cohorts and simul-taneously considering the entire network of direct andindirect comparisons. Thereby, it allows for a compari-son of predictive performance that is not limited bynon-comparable spectrum of populations, as is com-monly the case when evaluating the results of independ-ent validation studies. MSC meta-analysis can be appliedto any medical field, with the availability of individualpatient data being the only major limiting factor.Strengths of our study include the careful analyticalapproach to MSC meta-analysis and the availability ofthe R code, which allows for widespread use and poten-tial further development of the method. For the particu-lar application of MSC meta-analysis here, a majorstrength is the large high-quality database of the 3CIAcollaboration with the broadest possible COPD patientspectrum. The diverse case mix and broad patientspectrum greatly increase the probability that our resultsare generalizable to all COPD patients. A limitation ofthe study is that, ideally, a network meta-analysis is con-ducted prospectively and jointly planned for all of thecohorts involved to ensure equality of the clinical set-tings and homogeneity of study design, conduct, andvariable definitions, though this will rarely be the case inreality. Another limitation of our analysis is that we onlyused AUC as a performance measure, which we did fortheoretical and practical reasons [43]. In general, im-provements in AUC have to be interpreted with caution[72]. Furthermore, we cannot exclude the possibility ofcase-mix effects due to variables that were not availablein the database or unknown.Further research needs include the extension of MSC toinclude measures of calibration, which is arguably as im-portant as discrimination. For the area of COPD, it wouldbe attractive to apply MSC to risk scores for exacerbations[51, 73]. However, there are likely too few thoroughly de-veloped and externally validated scores to predict exacer-bations in patients with COPD [31]. Finally, given thelarge number of risk scores in the medical field and thelack of external validations and comparisons of risk scores,there is a great need for comparative studies that may useMSC in order to inform clinical practice and researchabout the most predictive scores [31].ConclusionsBorrowing from network meta-analysis, we presented acomprehensive approach for external validation andconcurrent comparison of multiple prognostic scores.While our analyses showed best performance for theADO and updated BODE scores to predict mortality forpatients with COPD, MSC meta-analysis can be appliedGuerra et al. BMC Medicine  (2018) 16:33 Page 10 of 13to prognostic scores in any medical field to identify thebest scores, possibly paving the way for stratifiedmedicine, public health, and research.Additional filesAdditional file 1: The Appendix. (DOCX 90 kb)AbbreviationsADO: Age, dyspnea, airflow obstruction; AUC: Area under the curve; B-AE-D: Bodymass index, acute exacerbations, dyspnea; BODE: Body mass index, airflowobstruction, dyspnea and severe exacerbations; BODEx: Body mass index, airflowobstruction, dyspnea, severe acute exacerbation of COPD; COPD: Chronicobstructive pulmonary disease; DOSE: Dyspnea, obstruction, smoking andexacerbation frequency; e-BODE: Severe acute exacerbation of COPD plus BODE;GOLD: Global Initiative for Chronic Obstructive Lung Disease; MSC: Multiple scorecomparison; NMA: Network meta-analysis; SAFE: Saint George’s RespiratoryQuestionnaire (SGRQ) score, air-flow limitation and exercise capacity;SUCRA: Surface under the cumulative ranking curveAcknowledgementsWe would like to thank Sarah Crook, Violeta Gaveikaite, Laura Werlen, andAlex Marzel (all from the University of Zurich, Zurich, Switzerland) for theircomments. We also thank the reviewers for their valuable comments.FundingMAP obtained funding for the current study. All authors organized fundingfor their respective cohorts, which has been described in detail elsewhere[49].Availability of data and materialsThe datasets supporting the conclusions of this article are reported within aprevious article [49]. The programming language used for the main analysiswas R version 3.0.2.Authors’ contributionsAll the authors had full access to all of the data in the study and takeresponsibility for the integrity of the data. BG, SRH, and MAP designed thestudy. All the authors contributed to the acquisition, analysis, orinterpretation of data. BG, SRH, and MAP drafted the manuscript. All authorsprovided a critical revision of the manuscript for important intellectualcontent. BG, SRH, and MAP undertook the statistical analysis. All authorsprovided necessary support to contribute their data to the 3CIAcollaboration. JBS and MAP supervised the study. All authors read andapproved the final manuscript.Ethics approval and consent to participateAll cohorts were approved by ethics committees, and the participants gavewritten informed consent [49].Consent for publicationNot applicable.Competing interestsThe authors declare that they have no competing interests.Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.Author details1Epidemiology, Biostatistics and Prevention Institute, University of Zurich,Zurich, Switzerland. 2Department of Pulmonary Medicine, KeplerUniversitatsklinikum GmbH, Linz, Austria. 3Faculty of Medicine, JohannesKepler Universitat Linz, Linz, Austria. 4Facultad de Medicina UASLP,Universidad Autonoma de San Luis Potosi, San Luis Potosi, Mexico.5Dartmouth College Geisel School of Medicine, Dartmouth, NH, USA.6Department of Pulmonary Medicine, Paracelsus MedizinischePrivatuniversitat, Salzburg, Austria. 7Hospital Universitario de Valme, Sevilla,Spain. 8Internal Medicine, Hospital Universitario Mutua de Terrassa, Terrassa,Spain. 9Pulmonary Department and Research Unit, Hospital Universitario NSLa Candelaria, Tenerife, Spain. 10Network and Health Services ResearchChronic Diseases (REDISSEC), Hospital Galdakao, Bizkaia, Spain. 11Servicio deNeumología, Hospital Universitari Arnau de Vilanova, Lleida, Spain.12Pulmonary Department, Clinica Universidad de Navarra, Pamplona, Spain.13European Respiratory Society (ERS) Guidelines Director, Barcelona, Spain.14Pulmonary and Critical Care Medicine, Brigham and Women’s Hospital,Boston, MA, USA. 15IISAragón and CIBERES, Hospital Universitario MiguelServet, Zaragoza, Spain. 16Department of General Practice, Academic MedicalCenter, University of Amsterdam, Amsterdam, The Netherlands. 17HospitalUnivarsitario de Cruces, Barakaldo, Vizcaya, Spain. 18Department of PublicHealth, Section of Social Medicine, University of Copenhagen, Copenhagen,Denmark. 19ISGlobal, CIBER Epidemiología y Salud Pública (CIBERESP),Universitat Pompeu Fabra (UPF), Barcelona, Spain. 20ISGlobal, Centre forResearch in Environmental Epidemiology (CREAL), IMIM (Hospital del MarMedical Research Institute, Universitat Pompeu Fabra (UPF), CIBEREpidemiología y Salud Pública (CIBERESP), Barcelona, Spain. 21Institute ofApplied Health Research, University of Birmingham, Birmingham, UK.22Division of Pulmonary and Critical Care, University of Michigan, Ann Arbor,MI, USA. 23Department of Public Health and Nursing, Norvegian University ofScience and Technology, Trondheim, Norway. 24Centre for ClinicalDocumentation and Evaluation, Northern Norway Regional Health Authority,Bodø, Norway. 25University of Bergen, Haukeland University Hospital, Bergen,Norway. 26Department of Global Public Health and Primary Care, Universityof Bergen, Bergen, Norway. 27Department of Respiratory Care and SleepControl Medicine, Graduate School of Medicine, Kyoto University, Kyoto,Japan. 28Department of Respiratory Medicine, Hospital SonEspases-IdISBa-CIBERES, Palma de Mallorca, Spain. 29Instituto de InvestigaciónSanitaria Princesa (IISP)-Servicio de Neumología- Hospital Universitario de laPrincesa, Universidad Autónoma de Madrid, Madrid, Spain. 30UniversidadNacional de la Plata, Hospital San Juan de Dios de La Plata, Buenos Aires,Argentina. 31Hopitaux Universitaires Paris Centre, Service de PneumologieAP-HP, Paris, France. 32Hopital Cochin; Universite Paris Descartes, Paris,France. 33University of British Columbia, James Hogg Research Centre,Vancouver, Canada. 34Instituto de Investigación del Hospital Universitario dela Princesa (IISP), Universidad Autónoma de Madrid, Servicio de Neumología,Madrid, Spain. 35Scientific and Methodological Consultant of, Barcelona, Spain. 36Epidemiology, Biostatistics and PreventionInstitute, University of Zurich, Hirschengraben 84, Room HRS G29, CH -8001Zurich, Switzerland. 37Epidemiology & Department of Epidemiology, JohnsHopkins Bloomberg School of Public Health, Baltimore, MD, USA.Received: 14 August 2017 Accepted: 26 January 2018References1. Steyerberg EW. Clinical prediction models. In: Gail M, Krickeberg K, Sarnet J,Tsiatis A, Wong W, editors. Statistics for Biology and Health. Berlin: Springer;2010. ISBN: 978–1–4419-2648-7.2. Moons KGM, Royston P, Vergouwe Y, Grobbee DE, Altman DG. Prognosisand prognostic research 1: what, why, and how ? BMJ. 2009;338:1317–20.3. Steyerberg EW, Moons KGM, Van Der Windt DA, Hayden JA, Perel P,Schroter S, et al. Prognosis Research Strategy (PROGRESS) 3: prognosticmodel research. PLoS Med. 2013;10:9.4. Harrell FE. Regression modelling strategies. In: Bickel P, Diggle P, FeinbergSE, Gather U, Olkin I, Zeger S, editors. Statistics for Biology and Health.Berlin: Springer; 2015. ISBN: 978–3–319-19424-0.5. Laupacis A, Wells G, Richardson WS, Tugwell P. Users’ guides to themedical literature: V. How to use an article about prognosis. JAMA.1994;272:234–7.6. Puhan MA, Yu T, Stegeman I, Varadhan R, Singh S, Boyd CM. Benefit-harmanalysis and charts for individualized and preference-sensitive prevention:example of low dose aspirin for primary prevention of cardiovasculardisease and cancer. BMC Med. 2015;13:11.7. Hayes DF, Markus HS, Leslie RD, Topol EJ. Personalized medicine: riskprediction, targeted therapies and mobile health technology. BMC Med.2014;12:8.8. Yu T, Vollenweider D, Varadhan R, Li T, Boyd C, Puhan MA. Support ofpersonalized medicine through risk-stratified treatmentGuerra et al. BMC Medicine  (2018) 16:33 Page 11 of 13recommendations — an environmental scan of clinical practiceguidelines. BMC Med. 2013;11:7.9. Nickel CH, Bingisser R, Morgenthaler NG. The role of copeptin as adiagnostic and prognostic biomarker for risk stratification in the emergencydepartment. BMC Med. 2012;10:7.10. Vuong K, McGeechan K, Armstrong BK, Cust AE. Risk prediction models forincident primary cutaneous melanoma: a systematic review. JAMADermatology. 2014;150:434.11. Kagen D, Theobald C, Freeman M. Risk prediction models for hospitalreadmission: a systematic review. JAMA. 2011;306:1688–98.12. Yourman LC, Lee SJ, Schonberg MA, Widera EW, Smith AK. Prognosticindices for older adults: a systematic review. JAMA. 2012;307:182–92.13. Collins GS, Mallett S, Omar O, Yu L-M. Developing risk prediction models fortype 2 diabetes: a systematic review of methodology and reporting. BMCMed. 2011;9:14.14. Damen JAAG, Hooft L, Schuit E, Debray TPA, Collins GS, Tzoulaki I, et al.Prediction models for cardiovascular disease risk in the general population:systematic review. BMJ. 2016;353:11.15. Puhan MA, Hansel NN, Sobradillo P, Enright P, Lange P, Hickson D, etal. Large-scale international validation of the ADO index in subjectswith COPD: an individual subject data analysis of 10 cohorts. BMJOpen. 2012;2:1–10.16. Puhan MA, Garcia-Aymerich J, Frey M, ter Riet G, Antó JM, Agusti A, et al.Expansion of the prognostic assessment of patients with chronicobstructive pulmonary disease: the updated BODE index and the ADOindex. Lancet. 2009;374:704–11.17. Celli BR, Cote CG, Marin JM, Casanova C, Montes de Oca M, Mendez RA, etal. The body-mass index, airflow obstruction, dyspnea, and exercise capacityindex in chronic obstructive pulmonary disease. N Engl J Med. 2004;350:1005–12.18. Soler-Cataluña JJ, Martinez-Garcia MA, Sanchez LS, Tordera MP, Sanchez PR.Severe exacerbations and BODE index: two independent risk factors fordeath in male COPD patients. Respir Med. 2009;103:692–9.19. Esteban C, Quintana JM, Aburto M, Moraza J, Arostegui I, Espana PP, et al.The health, activity, dyspnea, obstruction, age, and hospitalization:prognostic score for stable COPD patients. Respir Med. 2011;105:1662–70.20. Briggs A, Spencer M, Wang H, Mannino D, Sin DD. Development andvalidation of a prognostic index for health outcomes in chronic obstructivepulmonary disease. Arch Intern Med. 2008;168:71–9.21. Schembri S, Anderson W, Morant S, Winter J, Thompson P, Pettitt D, et al. Apredictive model of hospitalisation and death from chronic obstructivepulmonary disease. Respir Med. 2009;103:1461–7.22. Esteban C, Quintana JM, Aburto M, Moraza J, Capelastegui A. A simple score forassessing stable chronic obstructive pulmonary disease. QJM. 2006;99:751–9.23. Boeck L, Soriano JB, Brusse-Keizer M, Blasi F, Kostikas K, Boersma W, et al.Prognostic assessment in COPD without lung function: the B-AE-D indices.Eur Respir J. 2016;47:1635–44.24. Eisner MD, Trupin L, Katz PP, Yelin EH, Earnest G, Balmes J, et al.Development and validation of a survey-based COPD severity score. Chest.2005;127:1890–7.25. Cardoso F, Tufanin AT, Colucci M, Nascimento O, Jardim JR.Replacement of the 6-min walk test with maximal oxygen consumptionin the BODE index applied to patients with COPD: an equivalencystudy. Chest. 2007;132:477–82.26. Williams JEA, Green RH, Warrington V, Steiner MC, Morgan MDL, Singh SJ.Development of the i-BODE: validation of the incremental shuttle walkingtest within the BODE index. Respir Med. 2012;106:390–6.27. Jones RC, Donaldson GC, Chavannes NH, Kida K, Dickson-Spillmann M,Harding S, et al. Derivation and validation of a composite index of severityin chronic obstructive pulmonary disease: the DOSE Index. Am J Respir CritCare Med. 2009;180:1189–95.28. Azarisman MS, Fauzi MA, Faizal MP, Azami Z, Roslina AM, Roslan H. TheSAFE (SGRQ score, air-flow limitation and exercise tolerance) Index: a newcomposite score for the stratification of severity in chronic obstructivepulmonary disease. Postgrad Med J. 2007;83:492–7.29. Esteban C, Quintana JM, Moraza J, Aburto M, Aguirre U, AguirregomoscortaJI, et al. BODE-Index vs HADO-score in chronic obstructive pulmonarydisease: which one to use in general practice? BMC Med. 2010;8:28.30. Quintana JM, Esteban C, Unzurrunzaga A, Garcia-Gutierrez S, Gonzalez N, BarrioI, et al. Predictive score for mortality in patients with COPD exacerbationsattending hospital emergency departments. BMC Med. 2014;12:66.31. Guerra B, Gaveikaite V, Bianchi C, Puhan MA. Prediction models forexacerbations in patients with COPD. Eur Respir Rev. 2017;26:1–13.32. Siebeling L, Musoro JZ, Geskus RB, Zoller M, Muggensturm P, Frei A, et al.Prediction of COPD-specific health-related quality of life in primary care COPDpatients: a prospective cohort study. NPJ Prim Care Respir Med. 2014;24:7.33. Royston P, Moons KGM, Altman DG, Vergouwe Y. Prognosis and prognosticresearch 2: developing a prognostic model. BMJ. 2009;338:1373–7.34. Altman DG, Vergouwe Y, Royston P, Moons KGM. Prognosis and prognosticresearch 3: validating a prognostic model. BMJ. 2009;338:1432–5.35. Moons KGM, Altman DG, Vergouwe Y, Royston P. Prognosis and prognosticresearch 4: application and impact of prognostic models in clinical practice.BMJ. 2009;338:1487–90.36. Collins GS, Moons KGM. Comparing risk prediction models: should be routinewhen deriving a new model for the same purpose. BMJ. 2012;3186:1–2.37. Riley RD, Ensor J, Snell KIE, Debray TPA, Altman DG, Moons KGM, et al. Externalvalidation of clinical prediction models using big datasets from e-healthrecords or IPD meta-analysis: opportunities and challenges. BMJ. 2016;353:11.38. Mallett S, Royston P, Waters R, Dutton S, Altman DG. Reporting performanceof prognostic models in cancer: a review. BMC Med. 2010;8:11.39. Hemingway H. Ten steps towards improving prognosis research: problemswith prognosis research. BMJ. 2014;4184:1–10.40. Fraccaro P, van der Veer S, Brown B, Prosperi M, O’Donoghue D, Collins GS,et al. An external validation of models to predict the onset of chronickidney disease using population-based electronic health records fromSalford. UK BMC Med. 2016;14:15.41. Thangaratinam S, Allotey J, Marlin N, Dodds J, Cheong-See F, von DadelszenP, et al. Prediction of complications in early-onset pre-eclampsia (PREP):development and external multinational validation of prognostic models.BMC Med. 2017;15:11.42. Marin JM, Alfageme I, Almagro P, Casanova C, Esteban C, Soler-Cataluña JJ,et al. Multicomponent indices to predict survival in COPD: the COCOMICSstudy. Eur Respir J. 2013;42:323–32.43. Haile SR, Guerra B, Soriano JB, Puhan MA. Multiple Score Comparison: anetwork meta-analysis approach to comparison and external validation ofprognostic scores. BMC Med Res Methodol. 2017;17:1–12.44. Kessels AG, Riet G, Puhan MA, Kleijnen J, Bachmann LM, Minder C. A simpleregression model for network meta-analysis. OA Epidemiol. 2013;1:1–8.45. Li T, Puhan MA, Vedula SS, Singh S, Dickersin K. Network meta-analysis-highlyattractive but more methodological research is needed. BMC Med. 2011;9:5.46. Salanti G. Indirect and mixed-treatment comparison, network, or multiple-treatments meta-analysis: many names, many benefits, many concerns for thenext generation evidence synthesis tool. Res. Synth. Methods. 2012;3:80–97.47. Salanti G, Marinho V, Higgins J. A case study of multiple-treatmentsmeta-analysis demonstrates that covariates should be considered. J ClinEpidemiol. 2009;62:857–64.48. Sauter R, Held L. Network meta-analysis with integrated nested Laplaceapproximations. Biom J. 2015;57:1038–50.49. Soriano JB, Lamprecht B, Ramírez AS, Martinez-Camblor P, Kaiser B,Alfageme I, et al. Mortality prediction in chronic obstructive pulmonarydisease comparing the GOLD 2007 and 2011 staging systems: a pooledanalysis of individual patient data. Lancet Respir Med. 2015;3:443–50.50. Vestbo J, Hurd SS, Agustí AG, Jones PW, Vogelmeier C, Anzueto A, et al.Global strategy for the diagnosis, management, and prevention of chronicobstructive pulmonary disease GOLD executive summary. Am J Respir CritCare Med. 2013;187:347–65.51. Decramer M, Vogelmeier C, Agustí AG, Bourbeau J, Celli BR, Chen R, et al.Global strategy for the diagnosis, management, and prevention of chronicobstructive pulmonary disease. 2015. Hanley JA, McNeil BJ. The meaning and use of the area under a receiveroperating characteristic (ROC) curve. Radiology. 1982;143:29–36.53. Caldwell DM, Ades AE, Higgins J. Simultaneous comparison ofmultiple treatments: combining direct and indirect evidence. BMJ.2005;331:897–900.54. Lu G, Welton NJ, Higgins J, White IR, Ades AE. Linear inference for mixedtreatment comparison meta-analysis: A two-stage approach. Res SynthMethods. 2011;2:43–60.55. Mills EJ, Ioannidis JPA, Thorlund K, Schünemann HJ, Puhan MA, Guyatt GH.How to use an article reporting a multiple treatment comparisonmeta-analysis. JAMA. 2012;308:1246–53.56. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials.1986;7:177–88.Guerra et al. BMC Medicine  (2018) 16:33 Page 12 of 1357. Debray TPA, Damen JAAG, Snell KIE, Ensor J, Hooft L, Reitsma JB, et al. Aguide to systematic review and meta-analysis of prediction modelperformance. BMJ. 2017;356:11.58. Franchini AJ, Dias S, Ades AE, Jansen JP, Welton NJ. Accounting forcorrelation in network meta-analysis with multi-arm trials. Res SynthMethods. 2012;3:142–60.59. Salanti G, Ades AE, Ioannidis JPA. Graphical methods and numericalsummaries for presenting results from multiple-treatment meta-analysis: anoverview and tutorial. J Clin Epidemiol. 2011;64:163–71.60. Rücker G, Schwarzer G. Ranking treatments in frequentist networkmeta-analysis works without resampling methods. BMC Med Res Methodol.2015;15:9.61. Ransohoff DF, Feinstein A. Problems of spectrum and bias in evaluating theefficacy of diagnostic tests. N Engl J Med. 1978;299:926–30.62. Vergouwe Y, Moons KGM, Steyerberg EW. External validity of risk models:use of benchmark values to disentangle a case-mix effect from incorrectcoefficients. Am J Epidemiol. 2010;172:971–80.63. Debray TPA, Vergouwe Y, Koffijberg H, Nieboer D, Steyerberg EW, MoonsKGM. A new framework to enhance the interpretation of external validationstudies of clinical prediction models. J Clin Epidemiol. 2015;68:279–89.64. Debray TPA, Riley RD, Rovers MM, Reitsma JB, Moons KGM. Individualparticipant data (IPD) meta-analyses of diagnostic and prognostic modelingstudies: guidance on their use. PLoS Med. 2015;12:1–12.65. Pennells L, Kaptoge S, White IR, Thompson SG, Wood AM. Assessing riskprediction models using individual participant data from multiple studies.Am J Epidemiol. 2014;179:621–32.66. Ioannidis JP, Tzoulaki I. What makes a good predictor? The evidence appliedto coronary artery calcium score. JAMA. 2010;303:1646–7.67. Vestbo J, Anderson JA, Brook RD, Calverley PMA, Celli BR, Crim C, et al.Fluticasone furoate and vilanterol and survival in chronic obstructivepulmonary disease with heightened cardiovascular risk (SUMMIT): a double-blind randomised controlled trial. Lancet. 2016;387:1817–26.68. Fabbri LM, Luppi F, Beghé B, Rabe KF. Complex chronic comorbidities ofCOPD. Eur Respir J. 2008;31:204–12.69. Divo M, Cote C, De Torres JP, Casanova C, Marin JM, Pinto-Plata V, et al.Comorbidities and risk of mortality in patients with chronic obstructivepulmonary disease. Am J Respir Crit Care Med. 2012;186:155–61.70. Reilly BM, Evans AT, Schaider JJ, Das K, Calvin JE, Moran LA, et al.Impact of a clinical decision rule in the emergency department. JAMA.2002;288:342–50.71. McGinn TG, Guyatt GH, Wyer PC, David Naylor C, Stiell IG, Richardson WS.Users’ guides to the medical literature. JAMA. 2000;284:79–84.72. Tzoulaki I, Liberopoulos G, Ioannidis JPA. Assessment of claims of improvedprediction beyond the Framingham risk score. JAMA. 2009;302:2345–52.73. Wedzicha JA, Brill SE, Allinson JP, Donaldson GC. Mechanisms and impact ofthe frequent exacerbator phenotype in chronic obstructive pulmonarydisease. BMC Med. 2013;11:10.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript your next manuscript to BioMed Central and we will help you at every step:Guerra et al. BMC Medicine  (2018) 16:33 Page 13 of 13


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items