By Abhinav Shashank, CEO and co-founder, Innovaccer.
Once while I was scrolling through the news feed on my phone, there was one specific line that really made me wonder: “There’s a 40 percent chance of gusty and blustery winds today.” Statements such as this one strongly influence people’s behavior, as they are based on evidence or data findings from years of surveying, studying, and analyzing past trends and occurrences. However, my question is “Why are we not able to make such claims in healthcare- even today?”
Can we predict the vulnerabilities a patient might face in the future or the current health risks a population segment faces?
Is risk scoring the answer we have been looking for?
Almost all kinds of care organizations have some risk scoring methodology to target care interventions. With quality, costs, and patient experience taking the center stage in healthcare, care organizations need to stratify patients based on their need for immediate intervention.
The need of the hour is to address high-risk issues that impact large groups of patients and ensure that these needs are met in a timely fashion. Often, frequent fliers among high-risk patients come into the emergency department as if it’s their second home.
What if we take the method of risk scoring to a whole new level?
Traditionally, providers and health systems have relied on claims-based risk models, such as CMS-HCC, ACG and DxCG, which were built to forecast the risk of populations/sub-populations but not for individual patients. Hence, these models give an accurate prediction of the average risk of the population but exhibit very poor accuracy if used to predict risk for individual patients.
Although risk scoring has turned out to be a key factor in addressing the needs of the patient population, this method cannot provide all the important insights that are needed to drive necessary interventions. Since healthcare already has the right data from sources such as EHRs, claims, labs, pharmacy, social determinants of health (SDoH) and others, can we predict the future cost of care instead of just stating the risk score of the patient?
The right machine learning-driven approach to predict the future cost of care for patients
It all starts with the right data. The first step is to integrate the data from multiple sources- whether it is clinical or non-clinical data, such as SDoH. The data from these sources can allow us to use the comprehensive patient’s data for multiple predictive models to predict future health cost with greater accuracy.