Azeez Akinbode Unveils New Predictive Framework for Healthcare Systems with Incomplete Records

By Tolu Oke

Azeez Kunle Akinbode, a researcher in Applied Statistics (Business Analytics) at Bowling Green State University, has developed a groundbreaking predictive framework that enhances the accuracy of healthcare models even when patient data is incomplete.

His study, titled “Impact of Incomplete Records on Predictive Models That Use Electronic Health Record Data,” investigates how predictive modeling systems function when faced with missing patient information a frequent challenge in hospitals, emergency centers, and low-access communities. The publication has been added to the OhioLINK Electronic Theses and Dissertations (ETD) Center and accepted for presentation at the 2025 INFORMS Annual Meeting in Atlanta, Georgia.

Akinbode’s research examined 126 predictive model combinations built from the seven American Diabetes Association (ADA) screening variables, testing them across four patient subpopulations: sporadic, occasional, frequent, and non-frequent users.

“Predictive models are only as strong as the data they rely on,” Akinbode explained. “Our findings show that even when information is incomplete, a stable and reliable prediction process can still be achieved with a few key patient characteristics.”

The study revealed that predictive accuracy measured by the Area Under the Receiver Operating Characteristic (AUC) reached a plateau after just three variables: age, race, and family history. Models using these variables achieved an AUC of 0.95 or higher across all subgroups, showing that practical screening decisions can be made using minimal but critical patient information.

Beyond its academic relevance, the findings hold significant promise for healthcare systems in developing countries such as Nigeria, where electronic health records are still evolving and data completeness remains a persistent challenge.

“Our findings suggest that healthcare systems can still derive actionable insights from limited data,” he said. “This can help strengthen early disease detection and screening efforts even in resource-constrained settings.”

The research offers valuable guidance for improving Nigeria’s digital health infrastructure and advancing public health analytics. It highlights how data-driven insights can be applied effectively in primary care and emergency settings, where patient information is often incomplete.

Akinbode is set to present his findings at the upcoming INFORMS Annual Meeting in Atlanta, where he will further explore how data completeness and model stability interact contributing to broader global discussions on fairness, bias mitigation, and operational efficiency in predictive healthcare analytics.