Debiased Machine Learning and Network Cohesion for Doubly-Robust Differential Reward Models in Contextual Bandits

A common approach to learning mobile health (mHealth) intervention policies is linear Thompson sampling. Two desirable mHealth policy features are (1) pooling information across individuals and time and (2) incorporating a time-varying baseline reward. Previous approaches pooled information across individuals but not time, failing to capture trends in treatment effects over time. In addition, these approaches did not explicitly model the baseline reward, which limited the ability to precisely estimate the parameters in the differential reward model. In this paper, we propose a novel Thompson sampling algorithm, termed ''DML-TS-NNR'' that leverages (1) nearest-neighbors to efficiently pool information on the differential reward function across users and time and (2) the Double Machine Learning (DML) framework to explicitly model baseline rewards and stay agnostic to the supervised learning algorithms used. By explicitly modeling baseline rewards, we obtain smaller confidence sets for the differential reward parameters. We offer theoretical guarantees on the pseudo-regret, which are supported by empirical results. Importantly, the DML-TS-NNR algorithm demonstrates robustness to potential misspecifications in the baseline reward model.

A Meta-Learning Method for Estimation of Causal Excursion Effects to Assess Time-Varying Moderation

Twin revolutions in wearable technologies and smartphone-delivered digital health interventions have significantly expanded the accessibility and uptake of mobile health (mHealth) interventions across various health science domains. Sequentially randomized experiments called micro-randomized trials (MRTs) have grown in popularity to empirically evaluate the effectiveness of these mHealth intervention components. MRTs have given rise to a new class of causal estimands known as "causal excursion effects", which enable health scientists to assess how intervention effectiveness changes over time or is moderated by individual characteristics, context, or responses in the past. However, current data analysis methods for estimating causal excursion effects require pre-specified features of the observed high-dimensional history to construct a working model of an important nuisance parameter. While machine learning algorithms are ideal for automatic feature construction, their naive application to causal excursion estimation can lead to bias under model misspecification, potentially yielding incorrect conclusions about intervention effectiveness. To address this issue, this paper revisits the estimation of causal excursion effects from a meta-learner perspective, where the analyst remains agnostic to the choices of supervised learning algorithms used to estimate nuisance parameters. The paper presents asymptotic properties of the novel estimators and compares them theoretically and through extensive simulation experiments, demonstrating relative efficiency gains and supporting the recommendation for a doubly robust alternative to existing methods. Finally, the practical utility of the proposed methods is demonstrated by analyzing data from a multi-institution cohort of first-year medical residents in the United States (NeCamp et al., 2020).