Insights_subhead

Insights with Impact: more data isn’t better data

Why Prime’s adherence predictors top bulk consumer data


Prime researched adherence predictors with and without consumer purchasing data. The study took platinum at the March 2017 AMCP. More importantly, the results show where your money is better spent. 

A new trend within the pharmacy benefit management industry is emerging — purchasing consumer data to enhance health care predictive modeling. This data is not free (or even cheap). Yet despite concerns over privacy, it is often suggested by consultants. 

So Prime did what Prime does best. We tested the effectiveness of buying this bulk data. 

The purchased consumer data containing information on 250 million Americans’ interests, including: 

  • Lifestyle
  • Ethnicity
  • Purchasing behaviors
  • Wealth
  • Home ownership
  • Vehicle ownership
  • Family composition
  • Self reported diseases and 
  • Occupation

These data were added to Prime’s existing data and our effectiveness in predicting adherence was assessed.  

The consumer data augmented Prime’s Medicare data with over 300 potentially new predictors. Combined, we studied nearly 400 potential predictors in statistical models. Could this new data augment our existing predictors? 

The challenge

We wanted to know what could reliably predict adherence for the three medicine adherence Medicare Star rating quality performance measures: 

  • One for high blood pressure medicines (renin angiotensin system [RAS] antagonists), 
  • One for cholesterol lowering medicines (statins), and 
  • One for diabetes medicines (excluding insulin)

The stakes are high — Medicare Star ratings drive our clients’ financial reimbursement model. And Prime often includes quality ratings in its own performance goals with clients. Star ratings and quality performance affect our financial results.

Prime’s existing predictors 

Prime’s adherence predictors are robust and effective. We use more than 80 potential predictors from Prime’s Medicare prescription claims data:

  • Census data
  • Pharmacy risk group score (PRG – severity of illness)
  • Medication use
  • Prior adherence
  • Medication synchronization
  • Benefit design
  • Enrollment data 

According to our previously published research, in the three drug categories measured for Medicare Star ratings, these are the three best adherence predictors: 

  • Member behavior: Past 100 percent adherence (4 to 8 times as likely to be adherent in the future)
  • Benefit design feature: Having a 90-day supply claim in the previous year (28 percent to 31 percent higher odds of being adherent)
  • Benefit design feature: Having a $4 or less generic drug cost share per 30-day supply (20 percent to 33 percent higher odds of being adherent)

Did adding more data enhance our data? No. 

Adding 300 more consumer predictors from the purchased database did not improve Prime’s adherence predictors. We did find a few characteristics that predict adherence at about the same weight as our claims census data. 

But the consumer data could provide no further insight into what might influence adherence, as Prime’s claims data could. Plus, the cost of purchasing, storing and linking the data far outweigh the value obtained. 

This study was not done to disprove an industry trend

We identify and perform research to advance Prime’s business, and our clients’ performance. But we also understand that our findings contribute to the knowledge base that helps shape the future of managed care. 

Prime’s current adherence predictors are robust and effective. This does not mean it is perfect. Testing the influence of consumer data could have led to a better way to help our clients and members. This time it did not. 

So while consumer data may be valuable for informing other aspects of health care, for predicting adherence, it does not appear to be more valuable than health data. 

Follow our Insights with Impact series in Prime’s newsroom, or contact us for more information.