P224: Context and Comparative Effectiveness of Covid-19 Treatments: Predicting mortality, Rehospitalization, and Activities of Daily Living (ADLs) Across Covid-19 Variants and Treatments
Poster Presenter
Brittin Wagner
Senior Research Scientist
PointClickCare Canada
Objectives
We (1) describe epidemiology of pharmacotherapy across a representative sample of US residents across demographic and clinical characteristics including comorbidities for each Covid-19 variant; and (2) quantify impacts of pharmacotherapy for COVID-19 on death, hospitalization, and ADLs.
Method
EHR data for 628,160 instances of Covid-19 in the US includes patients’ demographic and clinical attributes, vaccine status modeled with logit, GLM, and ML models to predict outcomes death, hospitalization, and ADLs across facility-level attributes and Covid-19 variants.
Results
Time-dependent predictors like variant explain treatment receipt and treatment impact on outcomes like mortality; we also find statistically significant differences for how clinical and demographic characteristics (such as having a condition like Dementia) covary with treatment receipt and outcomes such as death, and that these attenuate over time and variant (from pre-Delta until Omicron III).
Conclusion
Time-dependent predictors like variant explain treatment receipt and treatment impact on outcomes like mortality; we also find statistically significant differences for how clinical and demographic characteristics (such as having a condition like Dementia) covary with treatment receipt and outcomes such as death, and that these attenuate over time and variant (from pre-Delta until Omicron III).
The impacts of risk factor count and specific risk factors like dementia attenuate over the course of the Covid-19 pandemic.
Paxlovid and Molnupiravir have greater impacts over time; pooled models ignores covariation of treatments with time-based measures including variant.
Remdesivir has significant positive association with death, but only during Delta.
Vaccination was most impactful during Omicron 1.
Identifying time- and variant- dependent effectiveness of pharmacotherapies for viral infections like Covid-19 requires including patient-level and facility-level factors, as well as factors related to time and characteristics of the disease, such as variant. In models not accounting for time and variant (pooled “ALL VARIANT” models), effectiveness of treatments and the impacts of demographic and clinical characteristics are distorted; this is the case in stats models (logit and GLM) as well as machine learning models depicting feature importances. Including patient-centered outcomes, such as ADLs, sheds light on how trajectories of treatment disparities (eg for people with dementia) attenuate as health system familiarity and capacity, as well as vaccine availability, increase.
