Recurrent Event Estimands: With or Without Competing Terminal Event

Presentation on theme: "Recurrent Event Estimands: With or Without Competing Terminal Event"— Presentation transcript:

1 Recurrent Event Estimands: With or Without Competing Terminal Event
Jiawei Wei JSM, July 30, 2018 Vancouver Convention Center

2 Recurrent event data Recurrent events involve repeated occurences of the same type of event over time Settings of interest: comparing test treatment and control in chronic disease Without terminal event (e.g. relapsing-remitting multiple sclerosis) With terminal event (e.g. chronic heart failure) How to measure a treatment effect under the repeated occurrence of an event? | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

3 Estimand description according to draft ICH E9 (R1)
Estimand: defines what is to be estimated to address a specific scientific question of interest specify the treatment effect more precisely A. Population Subjects targeted by the scientific question B. Variable Quantities required to address the scientific question C. Intercurent event Events occurring after treatment initiation that either preclude observation of the variable or affect its interpretation D. Summary measure On which the treatment comparison will be based | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

4 Intercurrent Events ... Randomization End of Study Patient 1 1st event
2nd event 3rd event Patient 2 1st event 2nd event discontinuation due to AE ... Patient N 1st event discontinuation due to death Randomization End of Study | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

5 Without Terminal Event
Intercurrent event: Treatment discontinuation

6 Estimand 1 (Treatment-policy)
Population: defined through inclusion/exclusion criteria to reflect the targeted patient population Variable: number of relapses up to two years Intercurrent event: effect of test treatment versus placebo regardless of treatment discontinuation Summary measure: | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

7 Account for the relapses that occurred after treatment discontinuation
3 events by 2 years Patient 2 1st relapse 2nd relapse discontinuation due to AE Randomization 2 years Statistical methods (estimators): Negative Binomial (NB) model, Lin-Wei-Yang-Ying (LWYY) model (censoring at end of study) Missing data (due to drop out) imputation should be in line with the treatment policy estimand. | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

8 Estimand 2 (Hypothetical)
Population: defined through inclusion/exclusion criteria to reflect the targeted patient population Variable: number of relapses up to two years Intercurrent event: interested in the effect of test treatment vs placebo in the hypothetical scenario where treatment discontinuation was abolished i.e. patients would continue their treatment for the intended duration Summary measure: | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

9 Hypothetical scenario where treatment discontinuation was abolished
2 events by 2 years Patient 2 1st relapse 2nd relapse discontinuation due to AE Randomization 2 years Statistical methods (estimators): NB model, LWYY (censoring at the time of treatment discontinuation) Sensitivity analysis needed as assumptions for the predictions cannot be verified from the observed data | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

10 Which statistical approach is targeting which estimand?
`Estimate' values are calculated based on simulation with “true” treatment effect = 0.65 The numerical value for the treatment policy estimand is closer to 1 than for the hypothetical estimand because the effect is diluted under treatment policy NB and LWYY give consistent mean effects for both estimands Cox, WLW and PWP models are not appropriate since their target values are different from the estimand values *WLW: Wei-Lin-Weissfeld model; PWP: Prentice-Williams-Peterson model | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

11 With Terminal Event Intercurrent event: Treatment discontinuation
Cardiovascular death (CVD) Non-cardiovascular death

12 Estimand 1 (HHF) Population: defined through inclusion/exclusion criteria to reflect the targeted patient population Variable: number of HHF while the patient is alive Intercurrent event: effect of test treatment versus placebo 1. regardless of treatment discontinuation and 2. while being alive Summary measure: HHF: hospitalization of heart failure Expected number of events per unit time alive = Average number of events while alive Average exposure time | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

13 Focus on the period of time where patients are alive
2 events by time of death Patient 1 1st HHF 2nd HHF discontinuation due to non-CVD 2 events by time of death Patient 2 1st HHF 2nd HHF discontinuation due to CVD Randomization End of Study In certain scenarios, favors a test treatment with a worse effect on CVD since the worst patient outcome of CVD precludes all future HHF for that patient. Statistical methods (estimators): NB (censoring at the time of CVD) LWYY (censoring at the time of CVD) Joint frailty model (JFM): account for correlation between HHF and CVD | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

14 Estimand 2 (HHF+CVD) Population: defined through inclusion/exclusion criteria to reflect the targeted patient population Variable: number of unfavorable events, i.e. number of HHF or CVD, up to and including the time of death Intercurrent event: effect of test treatment versus placebo 1. regardless of treatment discontinuation and 2. while being alive Summary measure: | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

15 Focus on the period of time where patients are alive
2 events by time of death Patient 1 1st HHF 2nd HHF discontinuation due to non-CVD 3 events by time of death Patient 2 1st HHF 2nd HHF discontinuation due to CVD Randomization End of Study Weights all “bad events” equally, a natural extension of time-to-first-composite-event analyses to the recurrent HHF setting Statistical methods (estimators): NB (censoring at the time of CVD) LWYY (censoring at the time of CVD) JFM: account for correlation between HHF and CVD | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

16 Which statistical approach is targeting which estimand?
`Estimate' values are calculated based on simulation with “true” treatment effect on HHF = 0.7 For Estimand 1, the treatment effect is stronger as 𝐻𝑅 𝐶𝑉 increases, since the worst patient outcome of CVD precludes all future HHF for that patient For Estimand 2, the treatment effect is fairly constant when 𝐻𝑅 𝐶𝑉 varies LWYY targets both estimands of interest | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

17 Summary In settings without terminal event, the treatment policy and hypothetical estimands seem to be the most commonly used estimands in current practice In settings with terminal event Estimand 1 (HHF) appears more suitable in settings where test and control treatment are very unlikely to differ with respect to their effect on CVD Estimand 2 (HHF+CVD) provides an overall treatment effect, and could be used in settings where treatment effects on both HHF and CVD are plausible Various recurrent event data approaches exist but it is not always clear which estimand they are targeting Need to step back and precisely specify estimands which are clinically meaningful, interpretable, and estimable Statistical approaches need to be aligned to the estimands of choice and robustness of conclusions ought to be assessed through a sensitivity analysis. | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

18

19 References Andersen, P. K. & Gill, R. D. (1982): Cox’s regression model for counting processes: a large sample study. Ann. Stat., 10, 1100–1120. Cannon CP (1997): Clinical perspectives on the use of composite endpoints. Controlled clinical trials 18: Chi GYH (2005): Some issues with composite endpoints in clinical trials. Fundamental & Clinical Pharmacology 19: Cook R.J. (1995): The design and analysis of randomized trials with recurrent events. Statistics in Medicine, Vol. 14, Cook RJ, Lawless JF (1997): Marginal analysis of recurrent events and a terminating event. Statistics in Medicine, Vol 16, Cook RJ, Lawless JF, Lakhal-Chaieb L, Lee K-A (2009): Robust estimation of mean functions and treatment effects for recurrent events under event-dependent censoring and termination: application to skeletal complications in cancer metastatic to bone. JASA, 104:485, Cowling BJ, Hutton JL, Shaw JEH (2006): Joint modeling of event counts and survival times. Appl. Statist. 55, Part 1: pp Freemantle N et al. (2003): Composite outcomes in randomized trials: greater precision but with greater uncertainty? JAMA Vol 289. No Freemantle N, Calvert M. (2007): Weighing the pros and cons for composite outcomes in clinical trials. Journal of Clinical Epidemiology 60: | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

20 References (continued)
Ghosh D & Lin DY (2000): Nonparametric analysis of recurrent events and death. Biometrics, 56, 554–562. Ghosh D & Lin D Y (2002): Marginal regression models for recurrent and terminal events. Statistica Sinica, 12, 663–688. Li QH, Lagakos SW (1997): Use of Wei-Lin-Weissfeld Method for the Analysis of a Recurring and Terminating Event. Statistics in Medicine 16: Lin, D. Y., Wei, L. J., Yang, I. & Ying, Z. (2000): Semiparametric regression for the mean and rate functions of recurrent events. J. R. Stat. Soc., B, 62, 711–730. Liu L, Wolfe RA, Huang X (2004): Shared frailty models for recurrent events and terminal event. Biometrics 60: McCullagh P, Nelder JA (1989): Generalized Linear Models, 2nd edn. London: Chapman and Hall Metcalfe C et al (2003): The use of hospital admission data as measure of outcome in clinical studies of heart failure. European Heart Journal 24: Metcalfe C, Thompson SG (2006): The importance of varying the event generation process in simulation studies of statistical methods for recurrent events. Statistics in Medicine 26: Metcalfe C, Thompson SG (2007): Wei, Lin and Weissfeld‘s marginal analysis of multivariate failure time data: should it be applied to a recurrent events outcome) Statistical Methods in Medical Research 16: Montori VM, et al., (2005): Validity of composite end points in clinical trials. BMJ Vol | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

21 Back-up

22 Case study: ValHeFT Study: Placebo-contolled study
Placebo arm: 2499 patients Valsartan arm: 2511 patients, i.e. total N=5010 Mean duration of follow-up: 23 months (range: 0 – 38 months) N Engl J Med, Vol. 345, No. 23, December 6, 2001, pp | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

23 Case study: summary of number of HHF and CVD
Number of HF hosp. events Placebo N= N (%) Test treatment NT = N (%) Total NTOT = N (%) 1878 (75.15) 1974 (78.61) 3852 (76.89) 1 344 (13.77) 317 (12.62) 661 (13.19) 2 146 (5.84) 130 (5.18) 276 (5.51) 3 56 (2.24) 51 (2.03) 107 (2.14) 4 36 (1.44) 19 (0.76) 55 (1.10) 5 21 (0.84) 13 (0.52) 34 (0.68) 6 5 (0.20) 3 (0.12) 8 (0.16) 7 6 (0.24) 1(0.04) 7 (0.14) 8 2 (0.08) 5 (0.10) 9 0 (0.00) 2 (0.04) 10 1 (0.04) 12 1 (0.02) Number of HHF 1189 922 2111 Number of CVD 419 (16.77) 427 (17.01) 846 (16.89) Number of HHF and CVD 1608 1349 2957 | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

24 Case study: results for recurrent event endpoints
Table: Summary of analysis methods (RR: rate ratio; LCIL: lower 95% condence interval limit; UCIL: upper 95% condence interval limit) for Estimand 1 (HHF) and Estimand 2 (HHF+CVD). | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018

25 Acceptance of recurrent event endpoint by regulators
Commonly used in areas where mortality is relatively low (e.g., Multiple Sclerosis) EMA (1999, 2015 draft) guidance for chronic HF acknowledges recurrent HFH as potentially acceptable primary endpoint in some circumstances highlighting the importance of terminal events for analysis and interpretation ESC CV Round Table: “... particularly suitable for diseases where reductions in repeat hospitalizations are of interest (e.g. HF with preserved ejection fraction or acute decompensated HF).” FDA precedence: In the HF area recurrent HFH has been used as primary endpoint for pivotal/late stage trials of devices (CHAMPION), gene therapies (CUPID-2) and more recently drugs (PARAGON) | Recurrent Event Estimands | JSM | Jiawei Wei | July 30th 2018