Effective, Safe, and Measurable Implementation of Predictive Healthcare Solutions
Event Type
Oral Presentations
TimeThursday, April 1512:50pm - 1:10pm EDT
LocationPatient Safety Research and Initiatives
DescriptionPredictive healthcare uses machine learning to improve outcomes in clinical and/or operational settings. “Big data” has become a hot buzzword as a potential panacea for problems in healthcare. But the fact of the matter is that while data science and machine learning have produced many algorithms, with increasing levels of accuracy and precision, few of these algorithms have actually been implemented and evaluated in care settings in a way that has shown demonstrable improvement in healthcare outcomes.

Using experience in executing over 20 applied predictive healthcare projects in an academic medical system over the past six years, we developed a Predictive Healthcare Process including phases for Feasibility Assessment, Impact Assessment, and Scaling to Operations. We will describe this process and associated deliverables using example projects such as Palliative Connect (1,2), which uses a prediction model based on information in the EHR to identify patients most likely to benefit from a palliative care consultation.

We will present in detail a user-centered design methodology developed as part of the process: the Data Science “MadLibs” tool. Data Science MadLibs provide a simple but very powerful framework for specifying a problem space that is viable for a predictive solution. The tool can also be used for other improvement projects to ensure that multidisciplinary teams are tackling a project in a way that has the right stakeholders at the table, a practical intervention and evaluation plan, and measurable outcomes.

1. Courtright K, Chivers C, Becker M, Regli S, Pepper L, Draugelis M and O’Connor N. (2019). Electronic Health Record Mortality Prediction Model for Targeted Palliative Care Among Hospitalized Medical Patients: a Pilot Quasi-experimental Study. Journal of General Internal Medicine; 34 10.1007/s11606-019-05169-2.
2. Courtright K, Chivers C, Becker M, Regli S, Draugelis M, and O'Connor N. (2019). Palliative Connect: Triggered Palliative Care Consultation Using an EHR Prediction Model (FR420A). Journal of Pain and Symptom Management; 57: 408-409. 10.1016/j.jpainsymman.2018.12.120.