Assessing the Association Between Network-based Provider Communities and Patient Mortality in the Medicare Population with Multiple Chronic Conditions.

Understanding the complexity of care delivery and care coordination for patients with multiple chronic conditions is challenging. Network analysis can model the relationship between providers and patients to find factors associated with patient mortality. We constructed a network by connecting the providers through shared patients, which was then partitioned into tightly connected communities using a community detection algorithm. After adjusting for patient characteristics, the odds ratio of death for one standard deviation increase in degree centrality ratio between primary care providers (PCPs) and non-PCPs was 0.95 (0.92-0.98). Our result suggest that the centrality of PCPs may be a modifiable factor for improving care delivery. We demonstrated that network analysis can be used to find higher order features associated with health outcomes in addition to patient-level features.

Comparison of Methods to Reduce Bias From Clinical Prediction Models of Postpartum Depression.

Importance: The lack of standards in methods to reduce bias for clinical algorithms presents various challenges in providing reliable predictions and in addressing health disparities. Objective: To evaluate approaches for reducing bias in machine learning models using a real-world clinical scenario. Design, Setting, and Participants: Health data for this cohort study were obtained from the IBM MarketScan Medicaid Database. Eligibility criteria were as follows: (1) Female individuals aged 12 to 55 years with a live birth record identified by delivery-related codes from January 1, 2014, through December 31, 2018; (2) greater than 80% enrollment through pregnancy to 60 days post partum; and (3) evidence of coverage for depression screening and mental health services. Statistical analysis was performed in 2020. Exposures: Binarized race (Black individuals and White individuals). Main Outcomes and Measures: Machine learning models (logistic regression [LR], random forest, and extreme gradient boosting) were trained for 2 binary outcomes: postpartum depression (PPD) and postpartum mental health service utilization. Risk-adjusted generalized linear models were used for each outcome to assess potential disparity in the cohort associated with binarized race (Black or White). Methods for reducing bias, including reweighing, Prejudice Remover, and removing race from the models, were examined by analyzing changes in fairness metrics compared with the base models. Baseline characteristics of female individuals at the top-predicted risk decile were compared for systematic differences. Fairness metrics of disparate impact (DI, 1 indicates fairness) and equal opportunity difference (EOD, 0 indicates fairness). Results: Among 573 634 female individuals initially examined for this study, 314 903 were White (54.9%), 217 899 were Black (38.0%), and the mean (SD) age was 26.1 (5.5) years. The risk-adjusted odds ratio comparing White participants with Black participants was 2.06 (95% CI, 2.02-2.10) for clinically recognized PPD and 1.37 (95% CI, 1.33-1.40) for postpartum mental health service utilization. Taking the LR model for PPD prediction as an example, reweighing reduced bias as measured by improved DI and EOD metrics from 0.31 and -0.19 to 0.79 and 0.02, respectively. Removing race from the models had inferior performance for reducing bias compared with the other methods (PPD: DI = 0.61; EOD = -0.05; mental health service utilization: DI = 0.63; EOD = -0.04). Conclusions and Relevance: Clinical prediction models trained on potentially biased data may produce unfair outcomes on the basis of the chosen metrics. This study's results suggest that the performance varied depending on the model, outcome label, and method for reducing bias. This approach toward evaluating algorithmic bias can be used as an example for the growing number of researchers who wish to examine and address bias in their data and models.

Hierarchical patient-centric caregiver network method for clinical outcomes study.

In clinical outcome studies, analysis has traditionally been performed using patient-level factors, with minor attention given to provider-level features. However, the nature of care coordination and collaboration between caregivers (providers) may also be important in determining patient outcomes. Using data from patients admitted to intensive care units at a large tertiary care hospital, we modeled the caregivers that provided medical service to a specific patient as patient-centric subnetwork embedded within larger caregiver networks of the institute. The caregiver networks were composed of caregivers who treated either a cohort of patients with particular disease or any patient regardless of disease. Our model can generate patient-specific caregiver network features at multiple levels, and we demonstrate that these multilevel network features, in addition to patient-level features, are significant predictors of length of hospital stay and in-hospital mortality.

Predicting Adverse Drug Reactions on Distributed Health Data using Federated Learning.

Using electronic health data to predict adverse drug reaction (ADR) incurs practical challenges, such as lack of adequate data from any single site for rare ADR detection, resource constraints on integrating data from multiple sources, and privacy concerns with creating a centralized database from person-specific, sensitive data. We introduce a federated learning framework that can learn a global ADR prediction model from distributed health data held locally at different sites. We propose two novel methods of local model aggregation to improve the predictive capability of the global model. Through comprehensive experimental evaluation using real-world health data from 1 million patients, we demonstrate the effectiveness of our proposed approach in achieving comparable performance to centralized learning and outperforming localized learning models for two types of ADRs. We also demonstrate that, for varying data distributions, our aggregation methods outperform state-of-the-art techniques, in terms of precision, recall, and accuracy.