Primary Submission Category: Heterogeneous Treatment Effects
Resource-constrained optimal rules for HIV care retention in rural Kenya
Authors: Lina Montoya, Harriet Adhiambo, Thomas Odeny, Elvin Geng, Maya Petersen,
Presenting Author: Lina Montoya*
Missed clinic visits can compromise HIV treatment success. A recent Sequential Multiple Assignment Randomized Trial (ADAPT-R) of 1,816 HIV-positive patients in Kenya showed that, on average, conditional cash transfers (CCTs) for on-time clinic visits increased viral suppression (VS), an indicator of treatment success, compared to standard of care. We applied SuperLearning to data from ADAPT-R to estimate an optimal dynamic treatment rule for CCT use. Preliminary results suggest that this rule would assign CCTs to all persons (i.e., CCT was not harmful in the short term for any participants). In practice, however, resources may constrain the proportion of persons who can receive a CCT. One response is to selectively administer CCTs to only those persons most likely to benefit. For example, standard univariate effect modification analyses suggest that CCTs were more effective for persons living further from the clinic or self-employed. Thus, we use the approach of Luedtke and van der Laan (2016) to estimate optimal stochastic allocation rules for administering CCTs under a range of constraints on the maximum proportion of patients who can receive a CCT. We further evaluate the expected counterfactual probability of VS under each resource-constrained optimal rule and contrast it with the expected counterfactual outcome under the static rule in which everyone receives a CCT. Our work provides an applied illustration of resource-constrained optimal dynamic treatment rules.