Developing Statistical Methods on Event History Data Subject to Data Complexities for HIV Disease Progression and Policy Evaluation

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT In 2015, the World Health Organization (WHO) introduced Treat-All guidelines for people living with HIV, which recommend immediate initiation of antiretroviral therapy (ART) treatment upon diagnosis regardless of disease severity. Since then, most countries worldwide have adopted the policy. However, the understanding of the impact of such policy is quite limited, especially regarding HIV disease progression. Focused on event history outcome (represented by WHO clinical stages and death), we recently conducted a preliminary analysis. We used data from the Central Africa region of the International epidemiology Database to Evaluate AIDS (CA-IeDEA) for a multistate model based on a target trial design (where two cohorts were constructed, one before and one after the policy adoption). This work illuminated several limitations. For example, the assumption of non-informative censoring was unlikely to hold for all censored individuals due to loss of follow-up or transfer out. Also, the relatively small sample size of the CA-IeDEA hindered our capacities to 1) explore more clinically relevant and biologically plausible models for HIV disease progression and 2) explore population heterogeneities regarding the impact of the Treat-All on the outcome. In the proposed study, we plan to address these limitations by developing new statistical methods and leveraging the multi-regional, i.e., the global-IeDEA data, which will provide a substantially larger sample. We will develop procedures to address informative (dependent) censoring for the multistate models under the target trial design to allow for sensitivity analysis. For example, we propose parametric, nonparametric, and semi-parametric approaches to handle censoring at random. In addition, we offer a controlled multiple imputation method to handle censoring not at random. We will compare and validate those methods using both internal and external data. Finally, we will comprehensively analyze the global-IeDEA data, where the sensitivity analysis will ensure the robustness of our findings. The proposed work will advance research in HIV care by providing more detailed information on possible evolutionary courses of HIV disease progression and factors that modify the effectiveness of Treat-All. Our analysis is a first step towards developing more precise patient treatment options and resource allocation, thereby improving patient outcomes. The proposed statistical methods may also have applications to model other diseases that evolve through predefined clinical states with intermittent data collection schema subject to similar data complexities.