Predicting COVID-19 related healthcare resource utilization across a statewide patient population.

Kasthurirathne, Suranga; Park, Jeremy; Wild, David; Khan, Babar; Haggstrom, David; Grannis, Shaun

BACKGROUND The COVID-19 pandemic has highlighted the inability of health systems to leverage existing system infrastructure to rapidly develop and apply broad analytical tools that could inform state and national-level policymaking as well as patient care delivery at hospital settings. COVID-19 has also led to highlighted systemic disparities in health outcomes and access to care based on race/ethnicity, gender, income-level and urban-rural divide. While the US seems to be recovering from the COVID-19 pandemic due to widespread vaccination efforts and increased public awareness, there is an urgent need to address the aforementioned challenges. OBJECTIVE Inform the feasibility of leveraging broad, statewide datasets for population-health driven decision making by developing robust analytical models that predict COVID-19 related healthcare resource utilization across patients served by Indiana's statewide Health Information Exchange (HIE). METHODS We leveraged comprehensive datasets obtained from the Indiana Network for Patient Care (INPC) to train decision forest-based models that predicted patient-level need of healthcare resource utilization. To assess models for potential biases, we tested model performance against sub-populations stratified by age, race/ethnicity, gender, and residence (urban vs. rural). RESULTS We identified a cohort of 96,026 patients from across 957 zip codes for model development. We trained decision models that predicted healthcare resource utilization using the most impactful features (~100) out of a total of 1172 features created. Each model and stratified sub-population under test reported precision scores > 70%, accuracy and AUC ROC scores > 80%, and sensitivity scores ~>90%. We noted statistically significant variations in model performance across stratified sub-populations identified by age, race/ethnicity, gender, and residence (urban vs. rural). CONCLUSIONS This study presents the possibility of developing decision models capable of predicting patient-level healthcare resource utilization across a broad statewide region with considerable predictive performance. However, our models present statistically significant variations in performance across stratified sub-populations of interest. Further efforts are necessary to identify root causes of these biases and to rectify them. CLINICALTRIAL