Identifying Risky Drug-Seeking Behavior at the Point of Care
10:30am - 11:30amTuesday, February 12
Orlando - Orange County Convention Center
Brigham and Women's Hospital employed visualization techniques as well as descriptive and predictive analytics on a large longitudinal prescription dataset (PDMP - prescription drug monitoring program). A web-based tool, MeDSS, was then developed which dynamically generates charts on the patient's trajectory and does complex computations on risk predictors within seconds. A crossover study was conducted with participating physicians to determine how the inclusion of risk predictors from machine-learning models, incorporated into a tool with an improved UI design, increases comprehension of PDMP data, efficiency and recognition of high-risk factors--and thus assists with prescriber decision-making when a controlled substance is prescribed.
Develop predictive features of drug overdose risk from raw attributes in any prescription drug dataset
Construct patient's spatiotemporal timeline graph from longitudinal prescription data to quickly present clinicians with indicators of risk across modalities (payment, dosage, distance, etc.)
Integrate data-driven decision-making tools into existing workflows, specifically EHR systems and use in daily clinical practice, before prescribing controlled substances
Create dashboard that joins disparate datasets to get better and complete understanding of patient profile. This includes heatmaps monitoring of emerging polypharmacy cocktails as new drugs appear on the market