DOI QR코드

DOI QR Code

Doubly-robust Q-estimation in observational studies with high-dimensional covariates

고차원 관측자료에서의 Q-학습 모형에 대한 이중강건성 연구

  • Lee, Hyobeen (Department of Statistics, Korea University) ;
  • Kim, Yeji (Department of Statistics, Korea University) ;
  • Cho, Hyungjun (Department of Statistics, Korea University) ;
  • Choi, Sangbum (Department of Statistics, Korea University)
  • Received : 2021.01.15
  • Accepted : 2021.02.18
  • Published : 2021.06.30

Abstract

Dynamic treatment regimes (DTRs) are decision-making rules designed to provide personalized treatment to individuals in multi-stage randomized trials. Unlike classical methods, in which all individuals are prescribed the same type of treatment, DTRs prescribe patient-tailored treatments which take into account individual characteristics that may change over time. The Q-learning method, one of regression-based algorithms to figure out optimal treatment rules, becomes more popular as it can be easily implemented. However, the performance of the Q-learning algorithm heavily relies on the correct specification of the Q-function for response, especially in observational studies. In this article, we examine a number of double-robust weighted least-squares estimating methods for Q-learning in high-dimensional settings, where treatment models for propensity score and penalization for sparse estimation are also investigated. We further consider flexible ensemble machine learning methods for the treatment model to achieve double-robustness, so that optimal decision rule can be correctly estimated as long as at least one of the outcome model or treatment model is correct. Extensive simulation studies show that the proposed methods work well with practical sample sizes. The practical utility of the proposed methods is proven with real data example.

동적 치료 요법(dynamic treatment regimes; DTRs)은 다단계 무작위 시험에서 개인에 맞는 치료를 제공하도록 설계된 의사결정 규칙이다. 모든 개인이 동일한 유형의 치료를 처방받는 고전적인 방법과 달리 DTR은 시간이 지남에 따라 변할 수 있는 개별 특성을 고려한 환자 맞춤형 치료를 제공한다. 최적의 치료 규칙을 파악하기 위한 회귀 기반 알고리즘 중 하나인 Q-학습 방법은 쉽게 구현될 수 있기 때문에 더욱 인기를 끌고 있다. 그러나 Q-학습 알고리즘의 성능은 Q-함수를 제대로 설정했는지의 여부에 크게 의존한다. 본 논문에서는 고차원 데이터가 수집되는 DTRs 문제에 대한 다양한 이중강건 Q-학습 알고리즘을 연구하고 가중 최소제곱 추정 방법을 제안한다. 이중강건성(double-robustness)은 반응변수에 대한 모형 혹은 처리변수에 대한 모형 둘 중 하나만 제대로 설정되어도 불편추정량을 얻을 수 있음을 의미한다. 다양한 모의실험 연구를 통해 제안된 방법이 여러 시나리오 하에서도 잘 작동함을 확인하였으며 실제 데이터 예제를 통해 방법론에 대한 예시를 제시하였다.

Keywords

Acknowledgement

이 연구는 2020학년도 고려대학교 교내학술연구비 지원(K2008341)과 정부(교육부)의 재원으로 한국연구재단의 지원(2019R1F1A1052239, 2019R1A4A1028134)을 받아 수행된 기초연구사업임.

References

  1. Arjas E and Saarela O (2010). Optimal dynamic regimes: presenting a case for predictive inference, The International Journal of Biostatistics, 6.
  2. Bellman R (1957). Dynamic Programming, Princeton University Press, Princeton.
  3. Bertsekas D and Tsitsiklis J (1996). Neuro-Dynamic Programming, Athena Scientific, Massachusetts.
  4. Breiman L, Friedman JH, Olshen RA, and Stone CJ (1984). Classification and Regression Trees, Wadsworth, California.
  5. Chakraborty B, Murphy SA, and Strecher V (2010). Inference for non-regular parameters in optimal dynamic treatment regimes, Statistical Methods in Medical Research, 19, 317-343. https://doi.org/10.1177/0962280209105013
  6. Chakraborty B (2011). Dynamic treatment regimes for managing chronic health conditions: a statistical perspective, American Journal of Public Health, 101, 40-45. https://doi.org/10.2105/AJPH.2010.198937
  7. Geng Y, Zhang HH, and Lu W (2015). On optimal treatment regimes selection for mean survival time, Statistics in Medicine, 34, 1169-1184. https://doi.org/10.1002/sim.6397
  8. Gunter L, Zhu J, and Murphy SA (2011). Variable selection for qualitative interactions, Statistical Methodology, 8, 42-55. https://doi.org/10.1016/j.stamet.2009.05.003
  9. Henderson R, Ansell P, and Alshibani D (2010). Regret-regression for optimal dynamic treatment regimes, Biometrics, 66, 1192-1201. https://doi.org/10.1111/j.1541-0420.2009.01368.x
  10. Hernan MA, Lanoy E, Costagliola D, and Robins JM (2006). Comparison of dynamic treatment regimes via inverse probability weighting, Basic & Clinical Pharmacology & Toxicology, 98, 237-242. https://doi.org/10.1111/j.1742-7843.2006.pto_329.x
  11. Kereiakes DJ, Obenchain RL, Barber BL, et al. (2000). Abciximab provides cost-effective survival advantage in high-volume interventional practice, American Heart Journal, 140, 603-610. https://doi.org/10.1067/mhj.2000.109647
  12. Kravitz RL, Duan N, and Braslow J (2004). Evidence-based medicine, heterogeneity of treatment effects and the trouble with averages, Milbank Quarterly, 82, 661-687. https://doi.org/10.1111/j.0887-378X.2004.00327.x
  13. Lee BK, Lessler J, and Stuart EA (2010). Improving propensity score weighting using machine learning, Statistics in Medicine, 29, 337-346. https://doi.org/10.1002/sim.3782
  14. Longford NT and Nelder JA (1999). Statistics versus statistical science in the regulatory process, Statistics in Medicine, 18, 2311-2320. https://doi.org/10.1002/(SICI)1097-0258(19990915/30)18:17/18<2311::AID-SIM257>3.0.CO;2-T
  15. Lu W, Zhang HH, and Zeng D (2011). Variable selection for optimal treatment decision, Statistical Methods in Medical Research, 22, 493-504. https://doi.org/10.1177/0962280211428383
  16. Murphy SA (2003). Optimal dynamic treatment regimes (with discussions), Journal of the Royal Statistical Society: Series B, 65, 331-366. https://doi.org/10.1111/1467-9868.00389
  17. Murphy SA (2005). A generalization error for q-learning, Journal of Machine Learning Research, 6, 1073-1097.
  18. Pineau J, Bellemare MG, Rush AJ, Ghizaru A, and Murphy SA (2007). Constructing evidence-based treatment strategies using methods from computer science, Drug and Alcohol Dependence, 88, 52-60.
  19. Polley E, LeDell E, Kennedy C, Lendle S, and van der Laan M (2016). Package 'superlearner', Retrieved from: https://cran.r-project.org/web/packages/SuperLearner/SuperLearner.pdf
  20. Ridgeway G, McCaffrey D, Morral A, Griffin BA, and Burgette L (2016). Twang: toolkit for weighting and analysis of nonequivalent groups.
  21. Robins JM, Hernan MA, and Brumback B (2000). Marginal structural models and causal inference in epidemiology, Epidemiology, 11, 550-560. https://doi.org/10.1097/00001648-200009000-00011
  22. Robins JM, Orellana L, and Rotnitzky A (2008). Estimation and extrapolation of optimal treatment and testing strategies, Statistics in Medicine, 27, 4678-4721. https://doi.org/10.1002/sim.3301
  23. Robins JM (2004). Optimal structural nested models for optimal sequential decisions. In Proceedings of the Second Seattle Symposium on Biostatistics, Springer, New York.
  24. Sorensen TI (1996). Which patients may be harmed by good treatments?, Lancet, 348, 351-352. https://doi.org/10.1016/S0140-6736(05)64988-4
  25. Thall PF, Millikan RE, and Sung HG (2000). Evaluating multiple treatment courses in clinical trials, Statistics in Medicine, 19, 1011-1028. https://doi.org/10.1002/(SICI)1097-0258(20000430)19:8<1011::AID-SIM414>3.0.CO;2-M
  26. Therneau T, Atkinson B, and Ripley B (2014). Package 'Rpart', Retrieved from: http://cran.r-project.org/web/packages/rpart/rpart.pdf
  27. Tian L, Alizaden AA, Gentles AJ, and Tibshirani R (2014). A simple method for estimating interactions between a treatment and a large number of covariates, Journal of the American Statistical Association, 109, 1517-1532. https://doi.org/10.1080/01621459.2014.951443
  28. Tibshirani R (1996). Regression shrinkage and selection via the lasso, Journal of the Royal Statistical Society: Series B, 58, 267-288.
  29. Van der Laan MJ and Petersen ML (2007). Causal effect models for realistic individualized treatment and intention to treat rules, The International Journal of Biostatistics, 3.
  30. Van der Laan MJ, Polley EC, and Hubbard AE (2007). Super learner, Statistical Applications in Genetics and Molecular Biology, 6.
  31. Wallace MP and Moodie EEM (2015). Doubly-robust dynamic treatment regimen estimation via weighted least squares, Biometrics, 71, 636-644. https://doi.org/10.1111/biom.12306
  32. Wallace MP, Moodie EEM, and Stephens DA (2017). Model validation and selection for personalized medicine using dynamic-weighted ordinary least squares, Statistical Methods in Medical Research, 26, 1641-1653. https://doi.org/10.1177/0962280217708665
  33. Westreich D, Lessler J, and Funk MJ (2010). Propensity score estimation: neural networks, support vector machines, decision trees (CART) and meta-classifiers as alternatives to logistic regression, Journal of Clinical Epidemiology, 63, 826-833. https://doi.org/10.1016/j.jclinepi.2009.11.020
  34. Zhao P and Yu B (2006). On model selection consistency of lasso, Journal of Machine Learning Research, 7, 2541-2563.
  35. Zou H (2006). The adaptive lasso and its oracle properties, Journal of the American Statistical Association, 101, 1418-1429. https://doi.org/10.1198/016214506000000735