A review of studies examining the relationship between progression ...

More documents

Recommendations

Info

4.2. METHODS IN STUDIES USING INDIVIDUAL PATIENT DATAFour papers used IPD to estimate the correlation between PFS and OS. 14-16,20 In Buyse et al 14 therank correlation coefficient ρ between PFS and OS was estimated “through a Hougaard bivariatecopula distribution of these end points over the entire time range, or using the Kaplan-Meierestimates of PFS at 6 months and OS at 12 months.” In Burzykowski et al 15 the association betweenPFS or TTP and survival was quantified through Spearman’s rank correlation coefficient. Twopapers assessed the “nonparametric Kendall tau rank correlation for bivariate censored data” 16,20 andone assessed the correlation between PFS and OS using the Fleischer model. 20Landmark analysis was used in the five papers to assess the prognostic impact of being alive andprogression-free at various time points on future survival. 16-20 In the landmark analysis, multivariateCox proportional hazards models were constructed for OS and these were stratified by progressionfreestatus at various time points. Hazard ratios are reported for survival in patients who were aliveand progression free at these time points compared to those who were not. Three of the models werestratified by trial protocol 16,17,19 while one reported separate analyses for each trial and a combinedanalysis adjusted for study protocol. 18All of the models were adjusted for multiple known prognostic factors, such as age, gender andperformance status. Halabi et al 16 also conducted a subgroup analysis for trial participants whoreceived docetaxel on the basis that this was the only known agent to prolong survival in castrateresistant prostate cancer. Halabi et al divided their sample into a training and testing set and appliedthe estimates of PFS to the testing data set and estimated misclassification error rates. Two papersevaluated model discrimination using the concordance index (c-index). 17,19 The c-index computes theprobability that, for a pair of randomly chosen comparable patients, the patient with the higher riskprediction (e.g progression free at 3 months) will experience an event (e.g death) before the lowerrisk patient (e.g progressed before 3 months). A completely random prediction would have a c-indexof 0.5, and a perfect rule will have a c-index of 1.0. 194.3. RESULTS4.3.1. Results from studies using aggregate data from multiple trialsIn studies using aggregate data from multiple trials, the size of the correlations between the surrogateoutcomes of PFS/TTP and the final outcome of OS were similarly wide ranging (0.24 to 0.89).Whilst the range of correlation coefficients reported for TTP (0.24 to 0.75) and PFS (0.481 to 0.89)22
were overlapping, the values were consistently lower for TTP in those studies reporting bothoutcomes. 13,27Louvet et al reported a regression slope of 0.68, suggesting that each month of PFS is associatedwith 21 days of additional OS. 22 Hotta et al did not report the slope but their regression equationsuggests that 26% of the variation in median OS is accounted for by the variation in PFS. 25 Subgroupanalysis in one study found that treatment with platinum-based therapies was found to be asignificant factor in the association between median OS and median PFS. None of the other factorsconsidered in the subgroup analysis were found to have significant interaction terms. 25 Hotta et alalso reported the association between survival post-progression and median OS which were stronglycorrelated (r 2 =0.89), although this is to be expected given that they defined survival post-progressionas the difference in median OS and median PFS. 254.3.2 Results from studies examining individual level correlationsIn studies using IPD data to examine individual level correlations, the size of the correlationcoefficient for PFS and OS varied from 0.30 16 to 0.82, 14 although the variation is likely to be stronglyrelated to the method used, as studies that reported more than one method also showed significantvariation between estimates. 14,20 The one study reporting the correlation between TTP and PFS at theindividual level gave an estimate of 0.682 which was similar to their estimate for PFS.All of the landmark analyses conducted showed significant hazard ratios suggesting that people whoare alive and progression free at the time points considered have a lower risk of subsequentmortality. Model discrimination as determined by the c-index was good (0.60 to 0.68) in all caseswhere it was estimated.5. STUDIES EXAMINING THE RELATIONSHIP BETWEEN THETREATMENT EFFECT FOR PFS/TTP AND THE TREATMENT EFECTFOR OS ACROSS MULTIPLE RCTSThirteen studies examined the relationship between the treatment effect on PFS or TTP and thetreatment effect on OS. 9-11,13-15,17,21,23,24,26-28 Three studies used IPD to estimate measures of triallevelsurrogacy 14,15,17with the remaining papers using aggregate date from multiple trials. Theanalysis by Foster et al 17 used IPD data from three RCTs but as this was a small number they treatedeach trial centre as a separate unit giving 32 points for analysis.23
Page 1 and 2: A REVIEW OF STUDIES EXAMINING THE R
Page 3 and 4: EXECUTIVE SUMMARYIntroductionProgre
Page 5 and 6: CONTENTS1. INTRODUCTION ...........
Page 7 and 8: 1. INTRODUCTION1.1. BACKGROUNDProgr
Page 9 and 10: equivalent incremental benefit in O
Page 11: 2. REVIEW METHODS2.1. SEARCH STRATE
Page 14 and 15: Table 1: Characteristics of include
Page 16 and 17: Table 2: Methods used to analyse th
Page 18 and 19: Reference Data type Relationships a
Page 20 and 21: Reference Data type Relationships a
Page 24 and 25: 5.1. DEFINITION OF TREATMENT EFFECT
Page 26 and 27: 5.3. OTHER MEASURES OF CORRELATION
Page 28 and 29: The slope parameters in the linear
Page 30 and 31: Table 4: Summary of results by tumo
Page 33 and 34: many of the papers included in this
Page 35 and 36: 8. REFERENCES1. Therasse, P., Arbuc
Page 37 and 38: 25. Hotta, K., Kiura, K., Fujiwara,
Page 39 and 40: APPENDIX 2: SEARCH STRATEGIESThe in
Page 41: 38 26 or 27 or 28 or 29 (1206968)39

A review of studies examining the relationship between progression ...

Create successful ePaper yourself

Delete template?

Save as template?