Application of machine learning to develop a cost-effectiveness analysis of pressure injury prevention for improving hospital safety

Abstract: Hospital-acquired pressure injuries are harmful, if not lethal to patients and costly to hospitals. These conditions cost the U.S. $26.8 billion, resulting in 2-3 million cases and 60,000 deaths. The UK faces a similar crisis of 700,000 cases resulting in 26,000 deaths at a cost of 2.1 billion GBP. Proper prevention methods using international guidelines could help reduce pressure injury incidence, but with so many hospitalized patients, it is difficult to justify the allocation of costly nursing time to all patients. We conducted a cost-effectiveness analysis coupled with Machine Learning applied to an electronic health record system to harness real-world data on pressure injury outcomes. A machine learning Markov model was conducted to extract transition probabilities from staged pressure injuries and adapt transition probabilities to real-time updates in incidence rates. We then fit these probabilities to an economic model to calculate the incremental cost-effectiveness ratio (ICER in $/QALY) of what proportion of the hospitalized patient population to risk-stratify for pressure injury risk. We analyzed a sample of over 30,000 patients between 2012-2014 to determine that risk-stratifying all patients, not just high-risk cohorts, achieved cost-effective care at an ICER of <$50,000/QALY. Hospitals should invest thoroughly in pressure injury prevention to save money in the short run and reduce harm, which may lead to financial surplus for quality improvement in other areas in the long run.

Papers that will be discussed during the seminar:
Addressing the multisectoral impact of pressure injuries in the USA (PDF , 178kb)
Value of hospital resources for effective pressure injury prevention (PDF , 533kb)
Using clinical data to predict high-cost performance coding issues associated wi (PDF , 513kb)