Background In clinical trials, when comparing treatments in a subgroup of patients defined by an event that occurred after randomization is required, the standard estimator that adjusts for the post-treatment variable does not have a causal interpretation.
Purpose To address this problem, we formulate clinically relevant causal estimands using the principal stratification framework developed by Frangakis and Rubin [3], and propose a new estimation method for the principal causal effect.
Methods We consider the comparison of the duration of response among patients who responded to chemotherapy in a cancer clinical trial. Our goal is to estimate the local average treatment effect, that is, the treatment difference among patients who would have responded to either treatment. In order to identify this estimand, we make the assumption that the value of the counterfactual indicator of response is independent of both the actual response status and the outcome variable of interest conditional on the covariates. The proposed estimator is a weighted average of the standard estimators for responders where weights are the probability that the response would have occurred had the patient received the other treatment.
Results The proposed method is applied to data from a randomized phase III clinical trial in patients with advanced non-small-cell lung cancer. The average difference for the duration of response among responders estimated by the proposed method and the standard one was 16.1 (days) and 9.5 (days), respectively. We also evaluate the performance of the proposed method through simulation studies, which showed that the proposed estimator was unbiased, while the standard one was largely biased.
Conclusions We have developed an estimation method for the local average treatment effect. For any type of outcome variables, our estimator can be easily constructed and can be interpreted as the treatment effect among patients who would have had the event in either treatment group.
Get full access to this article
View all access options for this article.
References
1.
1. Rosenbaum PR. The consequences of adjustment for a concomitant variable that has been affected by the treatment. J R Stat Soc Series A-General1984; 147: 656–666.
2.
2. Robins JM, Greenland S. Identifiability and exchangeability of direct and indirect effects. Epidemiology1992; 3: 143–155.
3.
3. Frangakis CE, Rubin DB. Principal stratification in causal inference. Biometrics2002; 58: 21–29.
4.
4. Rubin DB. More powerful randomization-based p-values in double-blind trials with non-compliance. Stat Med1998; 17: 371–385.
5.
5. Rubin DB. Comment on ‘Causal inference without counterfactuals’, by A. P. Dawid. J Am Stat Assoc2000; 95: 435–437.
6.
6. Gilbert PB, Bosch RJ, Hudgens MG. Sensitivity analysis for the assessment of causal vaccine effects on viral load in HIV vaccine trials. Biometrics2003; 59: 531–541.
7.
7. Zhang JL, Rubin DB. Estimation of causal effects via principal stratification when some outcomes are truncated by ‘death’. J Educ Behave Stat2003; 28: 353–368.
8.
8. Angrist JD, Imbens GW, Rubin DB. Identification of causal effects using instrumental variables (with discussion). J Am Stat Assoc1996; 91: 444–472.
9.
9. Rubin DB. Bayesian inference for causal effects: the role of randomization. Annals Stat1978; 6: 34–58.
10.
10. Rubin DB. Estimating causal effects of treatments in randomized and nonrandomized studies. J Educ Psychol1974; 66: 688–701.
11.
11. Holland PW. Statistics and causal inference (with discussion). J Am Stat Assoc1986; 81: 945–970.
12.
12. Huber PJ. The behavior of maximum likelihood estimates under nonstandard conditions. In Proceedings of the Fifth Berkley Symposium in Mathematical Statistics and Probability. University of California Press, 1976: 221–233.
13.
13. Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika1986; 73: 13-22.
14.
14. Robins JM. Marginal structural models versus structural nested models as tools for causal inference. In Halloran ME, Berry D eds. Statistical models in epidemiology: the environment and clinical trials. Springer-Verlag, 1999: 95–134.
15.
15. Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology2000; 11: 550–560.
16.
16. Robins JM, Rotnitzky A, Zhao LP. Estimation of regression coefficients when some regressors are not always observed. J Am Stat Assoc1994; 89: 846–866.
17.
17. Rubin DB, Thomas N. Affinely invariant matching methods with ellipsoidal distributions. Annals Stat1992; 20: 1079–1193.
18.
18. Rubin DB, Thomas N. Characterizing the effect of matching using linear propensity score methods with normal distributions. Biometrika1992; 79: 797–809.
19.
19. Rubin DB, Thomas N. Matching using estimated propensity scores: Relating theory to practice. Biometrics1996; 52: 249–264.
20.
20. Negoro S, Masuda N, Takada Y et al. Randomised phase III trial of irinotecan combined with cisplatin for advanced non-small cell lung cancer. Br J Cancer2003; 88: 335–341.
21.
21. Robins JM, Rotnitzky A, Zhao LP. Analysis of semiparametric regression models for repeated outcomes in the presence of missing data. J Am Stat Assoc1995; 90: 106–121.
22.
22. Robins JM, Rotnitzky A, Scharfstein DO.Sensitivity analysis for selection bias and unmeasured confounding in missing data and causal inference models. In: Halloran ME, Berry D eds. Statistical models in epidemiology, the environment, and clinical trials. Springer, 1999: 1–94.
23.
23. Balke A, Pearl J. Bounds on treatment effects from studies with imperfect compliance. J Am Stat Assoc1997; 92: 1171–1178.
