Paper detail

Causal inference with multiple versions of treatment and application to personalized medicine

The development of high-throughput sequencing and targeted therapies has led to the emergence of personalized medicine: a patient's molecular profile or the presence of a specific biomarker of drug response will correspond to a treatment recommendation made either by a physician or by a treatment assignment algorithm. The growing number of such algorithms raises the question of how to quantify their clinical impact knowing that a personalized medicine strategy will inherently include different versions of treatment. We thus specify an appropriate causal framework with multiple versions of treatment to define the causal effects of interest for precision medicine strategies and estimate them emulating clinical trials with observational data. Therefore, we determine whether the treatment assignment algorithm is more efficient than different control arms: gold standard treatment, observed treatments or random assignment of targeted treatments. Causal estimates of the precision medicine effects are first evaluated on simulated data and they demonstrate a lower biases and variances compared with naive estimation of the difference in expected outcome between treatment arms. The various simulations scenarios also point out the different bias sources depending on the clinical situation (heterogeneity of response, assignment of observed treatments etc.). A RShiny interactive application is also provided to further explore other user-defined scenarios. The method is then applied to data from patient-derived xenografts (PDX): each patient tumour is implanted in several immunodeficient cloned mice later treated with different drugs, thus providing access to all corresponding drug sensitivities for all patients. Access to these unique pre-clinical data emulating counterfactual outcomes allows to validate the reliability of causal estimates obtained with the proposed method.