Date Published: April 30, 2018
Publisher: Oxford University Press
Author(s): Miguel Angel Luque-Fernandez, Aurélien Belot, Linda Valeri, Giovanni Cerulli, Camille Maringe, Bernard Rachet.
In this paper, we propose a structural framework for population-based cancer epidemiology and evaluate the performance of double-robust estimators for a binary exposure in cancer mortality. We conduct numerical analyses to study the bias and efficiency of these estimators. Furthermore, we compare 2 different model selection strategies based on 1) Akaike’s Information Criterion and the Bayesian Information Criterion and 2) machine learning algorithms, and we illustrate double-robust estimators’ performance in a real-world setting. In simulations with correctly specified models and near-positivity violations, all but the naive estimators had relatively good performance. However, the augmented inverse-probability-of-treatment weighting estimator showed the largest relative bias. Under dual model misspecification and near-positivity violations, all double-robust estimators were biased. Nevertheless, the targeted maximum likelihood estimator showed the best bias-variance trade-off, more precise estimates, and appropriate 95% confidence interval coverage, supporting the use of the data-adaptive model selection strategies based on machine learning algorithms. We applied these methods to estimate adjusted 1-year mortality risk differences in 183,426 lung cancer patients diagnosed after admittance to an emergency department versus persons with a nonemergency cancer diagnosis in England (2006–2013). The adjusted mortality risk (for patients diagnosed with lung cancer after admittance to an emergency department) was 16% higher in men and 18% higher in women, suggesting the importance of interventions targeting early detection of lung cancer signs and symptoms.
Under the structural framework (see DAG in Figure 1) described above for population-based cancer epidemiology, we estimated 1-year adjusted mortality risk differences for cancer diagnosed after admittance to a hospital emergency department versus receiving a nonemergency cancer diagnosis. The high proportion of lung cancer diagnosed after admittance to an emergency department in England (emergency presentation) has been hypothesized to be mainly due to multiple steps that patients undergo between identification of the first symptoms and final diagnosis by the health-care system.
Given the increasing availability of a different range and variety of data in population-based cancer epidemiology, the proposed structural framework (Figure 1) constitutes a basis for further development of comparative effectiveness research in population-based cancer epidemiology. Developed for a binary treatment and outcome, the framework can be easily extended to handle time-to-event outcomes and might be adapted to specific comparative effectiveness scenarios. For instance, we considered cancer patients’ comorbidity and stage as confounders, but this might not be the case with other comparative effectiveness research questions. We recently published an article in which we argued that multivariate adjustment for cancer-related comorbid conditions (those with an onset date close before or after the date of cancer diagnosis) to evaluate the effectiveness of cancer treatment might be inappropriate, as it could induce collider stratification bias (38).