Evaluation of the Effect of a Continuous Treatment: A Machine Learning Approach with an Application to Treatment for Traumatic Brain Injury

Abstract

SummaryFor a continuous treatment, the generalised propensity score (GPS) is defined as the conditional density of the treatment, given covariates. GPS adjustment may be implemented by including it as a covariate in an outcome regression. Here, the unbiased estimation of the dose–response function assumes correct specification of both the GPS and the outcome‐treatment relationship. This paper introduces a machine learning method, the ‘Super Learner’, to address model selection in this context. In the two‐stage estimation approach proposed, the Super Learner selects a GPS and then a dose–response function conditional on the GPS, as the convex combination of candidate prediction algorithms. We compare this approach with parametric implementations of the GPS and to regression methods. We contrast the methods in the Risk Adjustment in Neurocritical care cohort study, in which we estimate the marginal effects of increasing transfer time from emergency departments to specialised neuroscience centres, for patients with acute traumatic brain injury. With parametric models for the outcome, we find that dose–response curves differ according to choice of specification. With the Super Learner approach to both regression and the GPS, we find that transfer time does not have a statistically significant marginal effect on the outcomes. © 2015 The Authors. Health Economics Published by John Wiley & Sons Ltd.

Predicting Outcome of Endovascular Treatment for Acute Ischemic Stroke: Potential Value of Machine Learning Algorithms

“…Super Learner's predictive performance has been found to surpass individual machine learning models in various clinical studies (17, 19, 20). …”

Should a propensity score model be super? The utility of ensemble procedures for causal adjustment

“…Super Learner has also been considered in the context of longitudinal data, where it has been found useful in the presence of model misspecification . Overall, Super Learner is now being widely adopted in the causal inference literature and in applications …”

The Association between Dietary Vitamin A and Carotenes and the Risk of Primary Liver Cancer: A Case–Control Study

“…These studies were performed on mice that had already induced cancer and were based on a relatively high intake of antioxidant supplementation. As mentioned above, trials in humans demonstrated that supplementation with beta-carotene and vitamin A could increase the risk of lung cancer [16,17], but increased vitamin A intake from food could have inverse effects on some cancer incidence [13,25]. These findings may suggest that moderate antioxidant intake is preferable, and that high intakes of antioxidants are unsafe.…”

“…The generalized propensity score (GPS) approach is an alternative to regression for estimation of the dose–response function of continuous values or multivalued treatments, for evaluation of the treatment level received, and for observation of the covariates [25]. The key feature of the GPS is that its balancing property is similar to the propensity score of binary treatments and to answering complex questions in non-experimental settings [26].…”

“…First, we conducted stratified analyses to determine whether the above-mentioned associations were modified by lifestyle behaviors (smoking, alcohol use, and tea drinking). Secondly, we tested the relationship between intake of VA and PLC by dose–response functions using the GPS method [25], which is defined as the conditional probability of a response variable (i.e., PLC) at each particular level of the exposure (or treatment) (i.e., vitamin A) given the pre-“exposed” variables (covariates), such as sex, age, BMI, education and income level, smoking, alcohol drinking, tea drinking, physical activity, and energy intake. GPS was analyzed using Stata 12.0, and other data were analyzed using R 3.2.3.…”