Machine Learning

, Volume 84, Issue 1–2, pp 109–136 | Cite as

Informing sequential clinical decision-making through reinforcement learning: an empirical study

  • Susan M. ShortreedEmail author
  • Eric Laber
  • Daniel J. Lizotte
  • T. Scott Stroup
  • Joelle Pineau
  • Susan A. Murphy


This paper highlights the role that reinforcement learning can play in the optimization of treatment policies for chronic illnesses. Before applying any off-the-shelf reinforcement learning methods in this setting, we must first tackle a number of challenges. We outline some of these challenges and present methods for overcoming them. First, we describe a multiple imputation approach to overcome the problem of missing data. Second, we discuss the use of function approximation in the context of a highly variable observation set. Finally, we discuss approaches to summarizing the evidence in the data for recommending a particular action and quantifying the uncertainty around the Q-function of the recommended policy. We present the results of applying these methods to real clinical trial data of patients with schizophrenia.


Optimal treatment policies Fitted Q-iteration Policy uncertainty 


  1. Adams, C. E. (2002). Schizophrenia trials: past, present and future. Epidemiologia E Psichiatria Sociale, 11(13), 144–151. Google Scholar
  2. Andrews, D. W. K. (2000). Inconsistency of the bootstrap when a parameter is on the boundary of the parameter space. Econometrica, 68(2), 399–405. MathSciNetzbMATHCrossRefGoogle Scholar
  3. Bagnell, A., Ng, A., & Schneider, J. (2001). Solving uncertain Markov decision problems (Tech. Rep. CMU-RI-TR-01-25). Robotics Institute, Carnegie Mellon University. Google Scholar
  4. Berry, D. A. (2006). A guide to drug discovery: Bayesian clinical trials. Nature Reviews. Drug Discovery, 5, 27–36. CrossRefGoogle Scholar
  5. Biswas, S., Liu, D. D., Lee, J. J., & Berry, D. A. (2009). Bayesian clinical trials at the University of Texas M. D. Anderson cancer center. Clinical Trials, 6, 205–216. CrossRefGoogle Scholar
  6. Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123–140. MathSciNetzbMATHGoogle Scholar
  7. Brunskill, E., Leffler, B. R., Li, L., Littman, M., & Roy, N. (2008). A continuous-state offset-dynamics reinforcement learner. In D. A. McAllester & P. Myllymäki (Eds.), Proceedings of 24th conference on uncertainty in artificial intelligence (UAI 2008) (pp. 53–61). Google Scholar
  8. Carpenter, J. R. , Kenward, M. G., & White, I. R. (2007). Sensitivity analysis after multiple imputation under missing at random: a weighting approach. Statistical Methods in Medical Research, 16(3), 259–275. MathSciNetzbMATHCrossRefGoogle Scholar
  9. Dawson, R., & Lavori, P. W. (2004). Placebo-free designs for evaluating new mental health treatments: the use of adaptive strategies. Statistics in Medicine, 23, 3249–3262. CrossRefGoogle Scholar
  10. Dearden, R., Friedman, N., & Andre, D. (1999). Model based Bayesian exploration. In B. Kathryn, & H. P. Laskey (Eds.), Proceedings of 5th conference on uncertainty in artificial intelligence (UAI 1999) (pp. 150–159). San Mateo: Morgan Kaufmann. Google Scholar
  11. Diggle, P., Heagerty, P., Liang, K. Y., & Zeger, S. (2002). Analysis of longitudinal data. Oxford: Oxford University Press. Google Scholar
  12. Doshi, F., Pineau, J., & Roy, N. (2008). Reinforcement learning with limited reinforcement: using Bayes risk for active learning in POMDPs. In A. McCallum & S. Roweis (Eds.), Proceedings of the 25th annual international conference on machine learning (ICML 2008) (pp. 256–263). New York: Omnipress. CrossRefGoogle Scholar
  13. Efron, B. (1979). Bootstrap methods: another look at the jackknife. The Annals of Statistics, 7(1), 1–26. MathSciNetzbMATHCrossRefGoogle Scholar
  14. Efron, B., & Tibshirani, R. J. (1993). An introduction to the bootstrap. New York: Chapman & Hall. zbMATHGoogle Scholar
  15. Engel, Y., Mannor, S., & Meir, R. (2005). Reinforcement learning with Gaussian processes. In L. D. Raedt & S. Wrobel (Eds.), Proceedings of the 22nd international conference on machine learning (ICML 2005) (pp. 201–208). New York: ACM.  10.1145/1102351.1102377. CrossRefGoogle Scholar
  16. Ernst, D., Geurts, P., & Wehenkel, L. (2005). Tree-based batch mode reinforcement learning. Journal of Machine Learning Research, 6, 503–556. MathSciNetGoogle Scholar
  17. Ernst, D., Stan, G. B., Goncalves, J., & Wehenkel, L. (2006). Clinical data based optimal STI strategies for HIV: a reinforcement learning approach. In Proceedings of the machine learning conference of Belgium and The Netherlands (Benelearn) (pp. 65–72). Google Scholar
  18. Fard, M. M., Pineau, J. (2009). MDPs with non-deterministic policies. In D. Koller, D. Schuurmans, Y. Bengio, & L. Bottou (Eds.), Advances in neural information processing systems (pp. 1065–1072). Cambridge: MIT Press. Google Scholar
  19. Gelman, A., Carlin, J. B., Stern, H., & Rubin, D. B. (1995). Bayesian Data Analysis. New York: Chapman & Hall. Google Scholar
  20. Gelman, A., Mechelen, I. V., Verbeke, G., Heitjan, D. F., & Meulders, M. (2005). Multiple imputation for model checking: completed-data plots with missing and latent data. Biometrics, 61, 74–85. MathSciNetzbMATHCrossRefGoogle Scholar
  21. Guez, A., Vincent, R., Avoli, M., & Pineau, J. (2008). Adaptive treatment of epilepsy via batch-mode reinforcement learning. In Proceedings of the innovative applications of artificial intelligence (IAAI). Google Scholar
  22. Hastie, T., Tibshirani, R., & Friedman, J. (2001). Elements of statistical learning. Berlin: Springer. zbMATHGoogle Scholar
  23. Irodova, M., & Sloan, R. H. (2005). Reinforcement learning and function approximation. In Proceeding of the twentieth national conference on artificial intelligence (AAAI) (p. 2005). American Association for Artificial Intelligence, Menlo Park. Google Scholar
  24. Kaelbling, L. P., Littman, M. L., & Moore, A. (1996). Reinforcement learning: a survey. The Journal of Artificial Intelligence Research, 4, 237–385. Google Scholar
  25. Kaelbling, L. P., Littman, M. L., & Cassandra, A. R. (1998). Planning and acting in partially observable stochastic domains. Artificial Intelligence, 101, 99–134. MathSciNetzbMATHCrossRefGoogle Scholar
  26. Kakade, S., Kearns, J., & Langford, J. (2003). Exploration in metric state spaces. In Proceedings of the 20th Annual International Conference on Machine Learning (ICML 2003). Google Scholar
  27. Kay, S. R., Flazbein, A., & Opler, L. A. (1987). The positive and negative syndrome scale (PANSS) for schizophrenia. Schizophrenia Bulletin, 13(2), 261–276. Google Scholar
  28. Laber, E. B., Qian, M., & Murphy, S. A. (2010). Statistical inference in dynamic treatment regimes (Tech. Rep. 506). Dept. of Statistics, University of Michigan Google Scholar
  29. Lagoudakis, M. G., & Parr, R. (2003). Least-squares policy iteration. Journal of Machine Learning Research, 4, 1107–1149. MathSciNetCrossRefGoogle Scholar
  30. Little, R. J. A., & Rubin, D. B. (1987). Statistical analysis with missing data. New York: Wiley. zbMATHGoogle Scholar
  31. Lizotte, D. J., Laber, E., & Murphy, S. A. (2009) Assessing confidence in policies learned from sequential randomized trials (Tech. Rep. 481). Department of Statistics, University of Michigan. Google Scholar
  32. Lizotte, D., Bowling, M., & Murphy, S. (2010). Efficient reinforcement learning with multiple reward functions for randomized controlled trial analysis. In Proceedings of the twenty-seventh international conference on machine learning (ICML 2010). (pp. 695–702). New York: Omnipress. Google Scholar
  33. Mannor, S., Simester, D., Sun, P., & Tsitsiklis, J. (2007) Biases and variance in value function estimates. Management Science 53(1). Google Scholar
  34. Monahan, G. (1982). A survey of partially observable Markov decision processes. Management Science, 28, 1–16. MathSciNetzbMATHCrossRefGoogle Scholar
  35. Murphy, S. M. (2003). Optimal dynamic treatment regimes. Journal of the Royal Statistical Society, Series B, 65(2), 331–366. MathSciNetzbMATHCrossRefGoogle Scholar
  36. Murphy, S. A. (2005). An experimental design for the development of adaptive treatment strategies. Statistics in Medicine, 24, 1455–1481. MathSciNetCrossRefGoogle Scholar
  37. Murphy, S. A., Oslin, D., & Rush, A. J. (2007). Methodological challenges in constructing effective treatment sequences for chronic disorders. Neuropsychopharmacology, 32(2), 257–262. CrossRefGoogle Scholar
  38. NAP (2010). The prevention and treatment of missing data in clinical trials. The National Academies Press, Panel on Handling Missing Data in Clinical Trials. Committee on National Statistics, Division of Behavioral, Social Sciences and Education. Google Scholar
  39. Neter, J., Kutner, M. H., Nachtsheim, C. J., & Wasserman, W. (1996). Applied linear statistical models. New York: McGraw-Hill. Google Scholar
  40. Parr, R., Li, L., Taylor, G., Painter-Wakefield, C., & Littman, M. (2008). An analysis of linear models, linear value-function approximation, and feature selection for reinforcement learning. In A. McCallum, & S. Roweis (Eds.), Proceedings of the 25th annual international conference on machine learning (pp. 752–759). New York: Omnipress. CrossRefGoogle Scholar
  41. Pineau, J., Bellemare, M. G., Rush, A. J., Ghizaru, A., & Murphy, S. A. (2007). Constructing evidence-based treatment strategies using methods from computer science. Drug and Alcohol Dependence S52–S60. Google Scholar
  42. Robins, J. M., Rotnitzky, A., & Scharfstein, D. (1999). Sensitivity analysis for selection bias and unmeasured confounding in missing data and causal inference models. In M. E. Halloran & D. Berry (Eds.), Statistical models in epidemiology: the environment and clinical trials (pp. 1–92). Berlin: Springer. Google Scholar
  43. Rubin, D. B. (1996). Multiple imputation after 18+ years (with discussion). Journal of the American Statistical Association, 91, 473–489. zbMATHCrossRefGoogle Scholar
  44. Rush, A. J., Fava, M., Wisniewski, S. R., & Lavori, P. W. (2004). Sequenced treatment alternatives to relieve depression (STAR*D): rational and design. Controlled Clinical Trials, 25(1), 119–142. CrossRefGoogle Scholar
  45. Schafer, J. L. (1997). Imputation of missing covariates under a multivariate linear mixed model (Tech. rep.). Dept. of Statistics, The Pennsylvania State University. Google Scholar
  46. Schafer, J. L. (1999). Multiple imputation: a primer. Statistical Methods in Medical Research, 8(1), 3–15. CrossRefGoogle Scholar
  47. Schafer, J. L., & Yucel, R. M. (2002). Computational strategies for multivariate linear mixed models with missing values. Journal of Computational and Graphical Statistics, 11, 421–442. MathSciNetCrossRefGoogle Scholar
  48. Scharfstein, D. O., Rotnitzky, A., & Robins, J. M. (1999). Adjusting for nonignorable drop-out using semiparametric nonresponse models. Journal of the American Statistical Association, 94(448), 1096–1120. MathSciNetzbMATHCrossRefGoogle Scholar
  49. Shao, J. (1994). Bootstrap sample size in nonregular cases. Proceedings of the American Mathematical Society, 122(4), 1251–1262. MathSciNetzbMATHCrossRefGoogle Scholar
  50. Shelton, C. R. (2001). Balancing multiple sources of reward in reinforcement learning. In Advances in neural information processing systems (NIPS 2000) (pp. 1082–1088). Google Scholar
  51. Smallwood, R. D., & Sondik, E. J. (1973). The optimal control of partially observable Markov processes over a finite horizon. Operations Research, 21, 1070–1088. CrossRefGoogle Scholar
  52. Strehl, A. L., & Littman, M. L. (2004). An empirical evaluation of interval Estimation for Markov decision processes. In ICTAI (pp. 128–135). Los Alamitos: IEEE Computer Society. Google Scholar
  53. Strehl, A. L., & Littman, M. L. (2005). A theoretical analysis of model-based interval Estimation. In L. D. Raedt & S. Wrobel (Eds.), Proceedings of the 22nd international conference on Machine learning (ICML 2005 ) (pp. 856–863). New York: ACM.  10.1145/1102351.1102459. CrossRefGoogle Scholar
  54. Strehl, A., Li, L., Wiewiora, E., Langford, J., & Littman, M. (2006). PAC model-free reinforcement learning. In W. W. Cohen & A. Moore (Eds.), Proceedings of the 23rd annual international conference on machine learning (ICML 2006) (pp. 881–888). CrossRefGoogle Scholar
  55. Strens, M. J. A. (2000). A Bayesian framework for reinforcement learning. In P. Langley (Ed.), Proceedings of the seventeenth international conference on machine learning (ICML 2000) (p. 943–950). San Francisco: Morgan Kaufmann. Google Scholar
  56. Stroup, T. S., McEvoy, J. P., Swartz, M. S., Byerly, M. J., Glick, I. D., Canive, J. M., McGee, M., Simpson, G. M., Stevens, M. D., & Lieberman, J. A. (2003). The National Institute of Mental Health clinical antipschotic trials of intervention effectiveness (CATIE) project: schizophrenia trial design and protocol development. Schizophrenia Bulletin, 29(1), 15–31. Google Scholar
  57. Sutton, R. S., & Barto, A. G. (1998). Off-policy bootstrapping. In Reinforcement learning: an introduction Cambridge: MIT Press. Google Scholar
  58. Swartz, M. S., Perkins, D. O., Stroup, T. S., McEvoy, J. P., Nieri, J. M., & Haal, D. D. (2003). Assessing clinical and functional outcomes in the clinical antipsychotic of intervention effectiveness (CATIE) schizophrenia trial. Schizophrenia Bulletin, 29(1), 33–43. Google Scholar
  59. Tetreault, J., Bohus, D., & Litman, D. (2007). Estimating the reliability of MDP policies: a confidence interval approach. In Proceedings of the human language technology conference (pp. 276–283). Google Scholar
  60. Thall, P., & Wathen, J. (2007). Practical Bayesian adaptive randomisation in clinical trials. European Journal of Cancer, 43(5), 859–866. CrossRefGoogle Scholar
  61. Thall, P. F., & Wathan, J. K. (2000). Covariate-adjusted adaptive randomization in a sarcoma trial with multistate treatments. Statistics in Medicine, 19, 1011–1028. CrossRefGoogle Scholar
  62. van Buuren, S. (2007). Multiple imputation of discrete and continuous data by fully conditional specification. Statistical Methods in Medical Research, 16(3), 219–242. MathSciNetzbMATHCrossRefGoogle Scholar
  63. van Buuren, S., Brand, J. P. L., Groothuis-Oudshoorn, C. G. M., & Rubin, D. B. (2006). Fully conditional specification in multivariate imputation. Journal of Statistical Computation and Simulation, 76(12), 1049–1064. MathSciNetzbMATHCrossRefGoogle Scholar
  64. Wang, T., Lizotte, D., Bowling, M., & Schuurmans, D. (2005). Bayesian sparse sampling for on-line reward optimization. In L. D. Raedt & S. Wrobel (Eds.), Proceedings of the 22nd international conference on machine learning (ICML 2005) (pp. 956–963). New York: ACM.  10.1145/1102351.1102472. CrossRefGoogle Scholar
  65. Zhao, Y., Kosorok, M. R., & Zeng, D. (2009). Reinforcement learning design for cancer clinical trials. Statistics in Medicine, 28, 3294–3315. CrossRefGoogle Scholar

Copyright information

© The Author(s) 2010

Authors and Affiliations

  • Susan M. Shortreed
    • 1
    Email author
  • Eric Laber
    • 2
  • Daniel J. Lizotte
    • 2
  • T. Scott Stroup
    • 3
  • Joelle Pineau
    • 1
  • Susan A. Murphy
    • 2
  1. 1.School of Computer ScienceMcGill UniversityMontrealCanada
  2. 2.Department of StatisticsUniversity of MichiganAnn ArborUSA
  3. 3.NYS Psychiatric InstituteNew YorkUSA

Personalised recommendations