Prescriptive analytics for reducing 30-day hospital readmissions after general surgery.

Area of Science:

• Surgical outcomes research
• Health informatics
• Data science in healthcare

Background:

• Hospitals face financial pressure to reduce patient readmission rates.
• The American College of Surgeons National Surgical Quality Improvement Program (NSQIP) provides valuable surgical data.
• Data science methods can be used to predict and prevent readmissions.

Purpose of the Study:

• To develop predictive and prescriptive models for 30-day surgical readmissions using NSQIP data.
• To offer real-time, personalized treatment recommendations to reduce readmission risk.
• To evaluate the effectiveness of interventions like pre-operative blood transfusions.

Main Methods:

• Utilized NSQIP data from 722,101 surgeries.
• Applied various classification methods to predict 30-day readmissions.
• Developed prescriptive methods to recommend pre-operative blood transfusions to increase hematocrit.

Main Results:

• Predictive models achieved 87% out-of-sample accuracy (AUC) using all NSQIP variables.
• Pre-operative variables alone yielded 74% AUC accuracy.
• Personalized interventions reduced readmissions by 12% for eligible patients (hematocrit <30).

Conclusions:

• Pioneered a data-driven methodology for personalized treatment recommendations to reduce readmissions.
• The approach demonstrates potential for significant annual cost savings in the U.S. healthcare system.
• Predicted reduction in readmissions could save over $20 million annually.