Over the last decade, population health predictive analytics has gained traction due to the availability of robust clinical, financial and operational data and increased pressure to lower costs and improve care outcomes. The traditional approach of reactive patient care has been replaced by a predictive, data-driven approach. Predictive analytics plays a major role in patient medication adherence, the extent to which a patient takes medication based on recommendations from their healthcare provider. Predictive analytics can help identify patients who are non-adherent and design an effective intervention strategy.
Patient medication adherence is a worldwide and persistent issue. As far back as 2003, the World Health Organization said that increased adherence had a far greater impact on population health than any improvements in medical treatments, highlighting the pressing need to focus on the issue. High rates of non-adherence are especially concerning in cases of chronic diseases. Studies show a 15% primary non-adherence rate for four chronic diseases, with the highest in lipid-lowering medications (statins) at 21%. Studies also estimate that non-adherence racks up $300 billion annually in avoidable U.S. healthcare costs, $44 billion of which is linked to cardiovascular disease and statin medication non-adherence. These figures, along with encouragement from government programs such as CMS Stars, are driving the healthcare sector to study adherence patterns using predictive analytics to adjust care managers’ intervention strategies
We conducted a study to explore how a next-gen approach to population health predictive analytics could improve medication adherence. Our study incorporated diverse datasets such as patient drug fill dates, days of supply and therapy drug NDC codes to analyze therapy progress and gaps on a year-to-date basis, in addition to patient-level data from third-party data vendor Symphony Health Solutions including integrated claims, health records, provider affiliation and limited social determinants of health data for over 27 million deidentified patients in Pennsylvania, New Jersey and Delaware.
Our study suggested that leveraging predictive analytical insights can help improve medication adherence in terms of:
- Patient demographics, by helping providers and health plans segment the population based on demographics and tailor intervention strategies accordingly
- Patient copays, by helping identify the right cost-sharing strategies for health plans to ensure medication adherence.
- Patient prescription supply, increasing medication adherence by making purchases more convenient for patients.
Medication adherence for statins is often negatively impacted by coronary artery disease (CAD) events, so prioritizing interventions and messaging for those patients is important.
Our study identified these key features for health plans to design an effective medication adherence strategy:
- Diversify your datasets: Like the common expression “garbage-in garbage-out,” all data analyses are dependent on data quality and diversity. Predictive model analyses are no different, and in fact may need large investments in master data management and big-data capabilities to ensure quality and diversity of datasets. The growing trend is to include non-traditional datasets such as Census, Bureau of Labor Statistics, Centers for Medicare & Medicaid Services and genomic data as well. Working with third-party data vendors makes it easier to access this required data. It also provides data-compliant work environments to “deidentify” members and provide member-specific strategies. Mitigation of personal data bias, access to broader data attributes and data integration support are some of the other benefits of working with third-party vendors.
- Automate your predictive analytics: The traditional approaches to predictive analytics are outdated since they are time-consuming and expertise-dependent. Employing machine learning and automation capabilities in the modeling process is recommended.
- Translate outputs to strategy: Without industry expertise, access to comprehensive datasets and next-gen modeling techniques will be moot. We recommend translating complex modeling results and statistical inferences to an intervention strategy that can be easily understood by stakeholders who don’t have an extensive data science background. This approach develops deeper stakeholder trust and mindshare to help you further improve.