Advanced Analytics for National Institutes of Health (NIH)
Advanced Analytics for National Institutes of Health
The NIH Pharmacy Department provides pharmaceutical care to patients participating in NIH research. NIH pharmacists manage commercially available and investigational drugs involved in thousands of research protocols and blind studies.
Since NIH does not receive reimbursement from health insurance companies, managing its budget and forecasting costs for is a mission-critical task. The department needed a simple way for staff to track upcoming costs for planning and procurement purposes.
Using advanced data science techniques, GCOM developed a powerful forecast model that encompasses all drugs administered on site or sent home with a patient for the next 90 days.
The forecast model collects and cleans historical before feeding it into a tensor flow package – a type of deep learning framework. Calculated cost predictions are then surfaced into visualizations in an intuitive dashboard for NIH staff.
GCOM data scientists selected a tensor flow because simpler strategies—like traditional time series forecasting—do not account for the complexities of the NIH’s clinical activities. Since activities at the NIH’s 18 institutes affect one another in many ways, focusing on a single institute’s cost over time would be insufficient. A more complex multivariate analysis is required.
Compared to the total drug cost moving average, the neural net model built by GCOM is 93% more accurate – almost twice as accurate than a moving average forecast.