The Annals of Statistics

Information-regret compromise in covariate-adaptive treatment allocation

Asya Metelkina and Luc Pronzato

Full-text: Access denied (no subscription detected)

We're sorry, but we are unable to provide you with the full text of this article because we are not able to identify you as a subscriber. If you have a personal subscription to this journal, then please login. If you are already logged in, then you may need to update your profile to register your subscription. Read more about accessing full-text


Covariate-adaptive treatment allocation is considered in the situation when a compromise must be made between information (about the dependency of the probability of success of each treatment upon influential covariates) and cost (in terms of number of subjects receiving the poorest treatment). Information is measured through a design criterion for parameter estimation, the cost is additive and is related to the success probabilities. Within the framework of approximate design theory, the determination of optimal allocations forms a compound design problem. We show that when the covariates are i.i.d. with a probability measure $\mu$, its solution possesses some similarities with the construction of optimal design measures bounded by $\mu$. We characterize optimal designs through an equivalence theorem and construct a covariate-adaptive sequential allocation strategy that converges to the optimum. Our new optimal designs can be used as benchmarks for other, more usual, allocation methods. A response-adaptive implementation is possible for practical applications with unknown model parameters. Several illustrative examples are provided.

Article information

Ann. Statist. Volume 45, Number 5 (2017), 2046-2073.

Received: February 2016
Revised: September 2016
First available in Project Euclid: 31 October 2017

Permanent link to this document

Digital Object Identifier

Primary: 62K05: Optimal designs
Secondary: 62P10: Applications to biology and medical sciences

Optimal design treatment allocation bounded design measure equivalence theorem multi-armed bandit problem


Metelkina, Asya; Pronzato, Luc. Information-regret compromise in covariate-adaptive treatment allocation. Ann. Statist. 45 (2017), no. 5, 2046--2073. doi:10.1214/16-AOS1518.

Export citation


  • [1] Atkinson, A. C. (1982). Optimum biased coin designs for sequential clinical trials with prognostic factors.Biometrika6961–67.
  • [2] Atkinson, A. C. (1999). Optimum biased-coin designs for sequential treatment allocation with covariate information.Stat. Med.181741–1752.
  • [3] Atkinson, A. C. (2002). The comparison of designs for sequential clinical trials with covariate information.J. Roy. Statist. Soc. Ser. A165349–373.
  • [4] Atkinson, A. C. (2014). Selecting a biased-coin design.Statist. Sci.29144–163.
  • [5] Atkinson, A. C. and Biswas, A. (2005). Bayesian adaptive biased-coin designs for clinical trials with normal responses.Biometrics61118–125.
  • [6] Baldi Antognini, A. and Giovagnoli, A. (2010). Compound optimal allocation for individual and collective ethics in binary clinical trials.Biometrika97935–946.
  • [7] Baldi Antognini, A. and Giovagnoli, A. (2015).Adaptive Designs for Sequential Treatment Allocation. CRC Press, Boca Raton.
  • [8] Baldi Antognini, A. and Zagoraiou, M. (2015). On the almost sure convergence of adaptive allocation procedures.Bernoulli21881–908.
  • [9] Ball, F. G., Smith, A. F. M. and Verdinelli, I. (1993). Biased coin designs with a Bayesian bias.J. Statist. Plann. Inference34403–421.
  • [10] Biswas, A., Park, E. and Bhattacharya, R. (2012). Covariate-adjusted response-adaptive designs for longitudinal treatment responses: PEMF trial revisited.Stat. Methods Med. Res.21379–392.
  • [11] Cappé, O., Garivier, A., Maillard, O.-A., Munos, R. and Stoltz, G. (2013). Kullback–Leibler upper confidence bounds for optimal sequential allocation.Ann. Statist.411516–1541.
  • [12] Cook, R. D. and Wong, W. K. (1994). On the equivalence of constrained and compound optimal designs.J. Amer. Statist. Assoc.89687–692.
  • [13] Dragalin, V. and Fedorov, V. (2006). Adaptive designs for dose-finding based on efficacy-toxicity response.J. Statist. Plann. Inference1361800–1823.
  • [14] Dragalin, V., Fedorov, V. and Wu, Y. (2008). Adaptive designs for selecting drug combinations based on efficacy-toxicity response.J. Statist. Plann. Inference138352–373.
  • [15] Dragalin, V., Fedorov, V. and Wu, Y. (2008). Two-stage design for dose-finding that accounts for both efficacy and safety.Stat. Med.275156–5176.
  • [16] Efron, B. (1971). Forcing a sequential experiment to be balanced.Biometrika58403–417.
  • [17] Fedorov, V. V. (1989). Optimal design with bounded density: Optimization algorithms of the exchange type.J. Statist. Plann. Inference221–13.
  • [18] Fedorov, V. V. and Leonov, S. L. (2014).Optimal Design for Nonlinear Response Models. CRC Press, Boca Raton, FL.
  • [19] Goldenshluger, A. and Zeevi, A. (2011). A note on performance limitations in bandit problems with side information.IEEE Trans. Inform. Theory571707–1713.
  • [20] Goldenshluger, A. and Zeevi, A. (2013). A linear response bandit problem.Stoch. Syst.3230–261.
  • [21] Hu, F. and Rosenberger, W. F. (2006).The Theory of Response-Adaptive Randomization in Clinical Trials. Wiley, Hoboken, NJ.
  • [22] Hu, F. and Zhang, L.-X. (2004). Asymptotic properties of doubly adaptive biased coin designs for multitreatment clinical trials.Ann. Statist.32268–301.
  • [23] Hu, J., Zhu, H. and Hu, F. (2015). A unified family of covariate-adjusted response-adaptive designs based on efficiency and ethics.J. Amer. Statist. Assoc.110357–367.
  • [24] Hu, Y. and Hu, F. (2012). Asymptotic properties of covariate-adaptive randomization.Ann. Statist.401794–1815.
  • [25] Lai, T. L. and Robbins, H. (1985). Asymptotically efficient adaptive allocation rules.Adv. in Appl. Math.64–22.
  • [26] Metelkina, A. and Pronzato, L. (2017). Supplement to “Information-regret compromise in covariate-adaptive treatment allocation.”DOI:10.1214/16-AOS1518SUPP.
  • [27] Pocock, S. J. and Simon, R. (1975). Sequential treatment assignment with balancing for prognostic factors in the controlled clinical trial.Biometrics31103–115.
  • [28] Pronzato, L. (2006). On the sequential construction of optimum bounded designs.J. Statist. Plann. Inference1362783–2804.
  • [29] Pronzato, L. (2010). Penalized optimal designs for dose-finding.J. Statist. Plann. Inference140283–296.
  • [30] Pukelsheim, F. (1993).Optimal Experimental Design. Wiley, New York.
  • [31] Rabie, H. and Flournoy, N. (2013). Optimal designs for contingent response models with application to toxicity-efficacy studies.J. Statist. Plann. Inference1431371–1379.
  • [32] Rosenberger, W. F. and Lachin, J. M. (2002).Randomization in Clinical Trials. Theory and Practice. Wiley, New York.
  • [33] Rosenberger, W. F., Sverdlov, O. and Hu, F. (2012). Adaptive randomization for clinical trials.J. Biopharm. Statist.22719–736.
  • [34] Sahm, M. and Schwabe, R. (2001). A note on optimal bounded designs. InOptimum Design2000 (Cardiff).Nonconvex Optim. Appl.51131–140. Kluwer Academic, Dordrecht.
  • [35] Wiens, D. P. (2009). Robust discrimination designs.J. R. Stat. Soc.,B71805–829.
  • [36] Wu, C. F. J. and Wynn, H. P. (1978). The convergence of general step–length algorithms for regular optimum design criteria.Ann. Statist.61273–1285.
  • [37] Wynn, H. P. (1982). Optimum submeasures with applications to finite population sampling. InStatistical Decision Theory and Related Topics III(S. S. Gupta and J. O. Berger, eds.).Proc.3rd Purdue Symp.2485–495. Academic Press, New York.
  • [38] Zhang, L.-X. and Hu, F. (2009). A new family of covariate-adjusted response-adaptive designs and their properties.Appl. Math. J. Chinese Univ.241–13.
  • [39] Zhang, L.-X., Hu, F., Cheung, S. H. and Chan, W. S. (2007). Asymptotic properties of covariate-adjusted response-adaptive designs.Ann. Statist.351166–1182.

Supplemental materials

  • Supplement to “Information-regret compromise in covariate-adaptive treatment allocation”. In this Supplement, we give the proofs of Theorems 2.1, 2.2 and 4.1.