All Things Predictive

Recently, I was talking with a colleague about a particularly difficult problem she was having in trying to ferret out why their in-hospital sepsis patients did not do as well as the patients that presented to the hospital in septic shock. Despite trying several tried and true performance improvement (PI) methods, the answers simply were not coming. As I discussed this further it became clear that because these patients were all very different (by service, provider, age, etc.) the interaction of these variables was essentially playing chaos-maker with their PI program, because no traditional data analysis method can account for the myriad of variables staring my colleague in the face.

I was struck that this would be a problem perfectly suited to predictive analytics (PA). A predictive model has the ability to see patterns in data that conventional data analysis (t-test, chi-squared, or ANOVA) simply can’t. In addition it will assign a probability to factors that contribute to the current state, in this case in-house sepsis. Predixion’s PA platform allows for easy and user-friendly analysis of these types of complex, multi-variant clinical conundrums. Using your institution’s data, Predixion will scan the data looking for patterns that contribute to your problem and assign a probability score that allows one to determine the magnitude of the contribution by the identified variable. For example, Predixion may show you that patient age, increasing WBC, systolic blood pressure trend, use of beta-blocker and admission to general surgery are the variables that most closely associate with patients who develop in-house sepsis. This gives actionable and real-time data to use for targeted screening and interventions to quickly lower the sepsis rate.