Predictive Analytics and Intelligent Data Integration for Improving Decision-Making in Healthcare Management
Abstract
Healthcare organizations increasingly generate heterogeneous data through electronic health records (EHRs), laboratory systems, pharmacy transactions, claims, scheduling platforms, medical devices, and patient-facing applications. Yet data fragmentation, inconsistent terminology, missingness, temporal misalignment, and delayed access frequently prevent managers from converting information into timely decisions. This paper examines an integrated framework that combines standardsbased data integration with predictive analytics to support capacity planning, readmission-risk management, workforce allocation, and proactive care coordination. A simulation-based experimental design was developed using 60,000 synthetic patient encounters and two analytical pathways: an EHR-only baseline and an integrated longitudinal dataset incorporating utilization, pharmacy, claims, and appointment features. Three models were compared: regularized logistic regression using the baseline feature set, regularized logistic regression using integrated features, and histogram-based gradient boosting using the integrated dataset. The simulation demonstrated that integration improved AUROC from 0.659 to 0.686 for logistic regression and improved F1 from 0.310 to 0.346 at an operational threshold of 0.20. A complementary latency simulation indicated a reduction in median decision-data assembly time from approximately 183 seconds to 133 seconds when standardized and cached integration services were used. The findings suggest that predictive performance and managerial responsiveness depend not only on model choice but also on the quality, interoperability, provenance, and timeliness of the data layer. The paper proposes a closed-loop architecture in which data quality monitoring, prediction, managerial action, and outcome feedback are treated as one governance process. The results are simulation findings rather than evidence from a real hospital and are therefore intended to demonstrate the engineering and analytical value of the proposed architecture rather than claim clinical effectiveness.
KEYWORDS: Predictive analytics, healthcare management, electronic health records, data integration, FHIR, OMOP, machine learning, clinical decision support, interoperability, data quality
Full Text:
PDF 95-107Refbacks
- There are currently no refbacks.