Reducing Prescription Drug Non-Adherence: A Data-Driven Approach for Health Insurance Companies
Prescription drug non-adherence is a major cost driver for health insurance companies, leading to worsening patient outcomes, increased hospitalizations, and higher long-term claims costs. Business analytics professionals can leverage advanced data modeling to identify trends, diagnose root causes, predict high-risk patients, and implement interventions to improve medication adherence.
The Problem: High Rates of Prescription Drug Non-Adherence
A health insurer identifies that a significant percentage of patients with chronic conditions (e.g., diabetes, hypertension) are not refilling their medications as prescribed. Leadership asks:
🔹 What are the overall trends in prescription adherence rates?
🔹 What factors contribute to medication non-adherence?
🔹 Can we predict which policyholders are most likely to be non-adherent?
🔹 What interventions should be implemented to improve adherence? (Prescriptive)
1. Descriptive Analytics: Measuring Prescription Adherence Trends
Question: What are the overall trends in prescription adherence rates?
The analytics team collects pharmacy claims data to measure adherence rates by different categories:
📊 Proportion of Days Covered (PDC) per patient (a standard measure of adherence)
📊 Adherence rates by drug class (e.g., statins, insulin, antihypertensives)
📊 Adherence trends over time (monthly, quarterly, yearly)
📊 Adherence by patient demographics, region, and prescriber specialty
Solution:
- Calculate adherence percentages per drug type and patient group.
- Use trend graphs to identify seasonal patterns in adherence.
- Compare adherence rates by age, gender, income level, and chronic condition type.
Key Insight: The analysis reveals that adherence rates are lowest among younger patients with hypertension and among those in lower-income ZIP codes.
2. Diagnostic Analytics: Identifying Root Causes of Non-Adherence
Question: What factors contribute to medication non-adherence?
After measuring adherence trends, the next step is to determine why certain policyholders fail to take their medications as prescribed.
Solution: Ordered Probit Model
An Ordered Probit Model is used because medication adherence is often measured on an ordinal scale, such as: 1 = Low Adherence (PDC < 50%) 2 = Moderate Adherence (PDC 50-79%) 3 = High Adherence (PDC ≥ 80%)
🔹 Dependent Variable: Medication Adherence Level (1 = Low, 2 = Moderate, 3 = High)
🔹 Independent Variables:
- Out-of-Pocket Cost (copay amount for the drug)
- Pharmacy Access (distance to nearest in-network pharmacy)
- Patient Socioeconomic Status (income level, insurance tier)
- Number of Prescriptions (polypharmacy burden)
- Chronic Condition Type (diabetes, hypertension, etc.)
- Provider Engagement (how often the patient visits their doctor)
Implementation Steps:
- Gather historical pharmacy claims data with adherence levels.
- Fit an Ordered Probit Model to determine how each variable impacts adherence.
- Analyze marginal effects to quantify how changes in independent variables influence adherence probabilities.
Key Insight: The model reveals that higher out-of-pocket costs and limited pharmacy access are the strongest predictors of non-adherence, particularly among lower-income patients.
3. Predictive Analytics: Forecasting Future Non-Adherence
Question: Can we predict which policyholders are most likely to be non-adherent?
With a clear understanding of why patients fail to adhere, the next step is to predict future non-adherence before it happens.
Solution: Propensity Score Matching (PSM)
A Propensity Score Matching Model is used to predict which patients are at high risk of non-adherence by comparing them to similar patients who adhere to medications.
🔹 Dependent Variable: Non-Adherence (Binary: 1 = Non-Adherent, 0 = Adherent)
🔹 Independent Variables:
- Chronic Disease Type
- Number of Prescriptions Filled in the Last 6 Months
- Pharmacy Distance
- Cost-Sharing (Copay Percentage)
- Insurance Plan Type (HMO vs. PPO vs. Medicare Advantage)
- Prior Hospitalizations
Implementation Steps:
- Identify patients who are currently adherent and match them with similar non-adherent patients based on demographic and clinical characteristics.
- Train a Propensity Score Matching Model to identify new policyholders at high risk of non-adherence.
- Assign risk scores to patients based on their likelihood of missing medications.
Key Insight: The model identifies high-risk patients 60 days before they become non-adherent, allowing for early intervention strategies.
4. Prescriptive Analytics: Implementing Solutions to Improve Adherence
Question: What interventions should be implemented to improve adherence?
With predictive insights, the final step is to develop targeted interventions to improve medication adherence and reduce long-term healthcare costs.
Solution: Panel Data Model to Evaluate Policy Interventions
A Panel Data Model is used to track adherence rates over time and measure the impact of different interventions on improving adherence.
🔹 Dependent Variable: PDC Score (Adherence Rate per Patient over Time)
🔹 Independent Variables:
- Introduction of Copay Assistance Programs (Binary: 1 = Yes, 0 = No)
- Pharmacy Home Delivery Enrollment (Binary: 1 = Enrolled, 0 = Not Enrolled)
- Provider Adherence Counseling (Binary: 1 = Received, 0 = Did Not Receive)
- Text Message Reminders Sent (Binary: 1 = Yes, 0 = No)
Implementation Steps:
- Track adherence rates before and after the rollout of copay assistance programs and pharmacy home delivery services.
- Use a Panel Data Model to assess which interventions significantly improve adherence over time.
- Adjust future policy decisions based on the most effective interventions.
Key Outcome: The insurer finds that patients enrolled in home delivery programs and receiving text message reminders have a 25% higher adherence rate, leading to lower hospitalization rates and reduced claims costs.
Conclusion: A Data-Driven Strategy to Improve Medication Adherence
By applying Descriptive, Diagnostic, Predictive, and Prescriptive Analytics, the insurer can:
✅ Measure adherence trends using standard pharmacy claims analysis.
✅ Identify the root causes of non-adherence using an Ordered Probit Model.
✅ Predict high-risk patients using Propensity Score Matching.
✅ Evaluate intervention effectiveness using a Panel Data Model.
As a result, the company reduces medication non-adherence rates, lowers hospitalization claims, and improves policyholder health outcomes.