Skipr
Read the article in Skipr
Predicting the unpredictable
An electronic patient record contains a wealth of structured and unstructured information about patients, their condition and their medical history. This data can help to improve patient care. For example, big data techniques make it possible to better predict the necessary care for a patient. In recent years, much has been published about these and other promises related to using big data in healthcare, but realizing this potential still remains to be seen. Making good on the promises requires usable and convincing examples. The aim of our study is to show that practical applications are closer than expected, and they are even relatively easy to integrate with existing systems.
In this report, we show that big data analysis contributes to predicting one of the most impactful forms of care: admission to an intensive care unit. The study demonstrates that we can make predictions using only administrative data: information that’s already available in a structured format in every hospital in the electronic patient record. To this end, we analyzed an anonymous dataset of half a million patient journeys, which allowed us to make predictions about the admission of individual patients to intensive care. The ratings of the predictive model varied from ‘excellent’ (AUC 0.90) for elective patients to ‘reasonable’ (AUC 0.74) for emergency patients.
This proof-of-concept study shows that with a limited, yet widely available dataset, it’s possible to make useful predictions. By using the wealth of information already available, we’re contributing to improving healthcare.
Download the full report 'Predicting the unpredictable' here »