CLINICAL METHODOLOGY

How we measure outcomes

Transparent methodology built on NCQA standards, published trial data, and validated ML models.

Medication Adherence (PDC)

We measure adherence using Proportion of Days Covered (PDC), the NCQA-endorsed standard for medication adherence measurement.

PDC = (Days with medication on hand) / (Total days in measurement period) × 100

A patient is considered adherent when PDC ≥ 80%, consistent with CMS quality measures. For GLP-1 specifically, we account for:

Reference: Nau DP. Proportion of Days Covered (PDC) as a Preferred Method of Measuring Medication Adherence. Pharmacy Quality Alliance, 2012.

Our dropout risk model uses a gradient-boosted classifier (XGBoost) trained on patient features to predict 90-day discontinuation probability.

Validated via 5-fold cross-validation on historical data:

Category	Score Range	Recommended Action
Low	0-30	Standard monitoring
Medium	31-60	Automated refill reminder, flag for review
High	61-100	Outreach call within 48 hours, care coordinator referral

Reference: Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System. KDD 2016.

Per-patient weight forecasting uses exponential smoothing with trend damping, calibrated against clinical trial trajectories.

Population-level forecasting aggregates individual trajectories with confidence intervals. Key adjustments:

Reference: Wilding JPH et al. Once-Weekly Semaglutide in Adults with Overweight or Obesity. NEJM 2021;384:989-1002.

Reference: Jastreboff AM et al. Tirzepatide Once Weekly for the Treatment of Obesity. NEJM 2022;387:205-216.

Financial impact estimates are derived from published health economics literature:

Component	Per-Patient Estimate	Source
Healthcare cost reduction	$280 per 1% body weight loss	Cawley et al., J Health Econ 2015
Absenteeism reduction	$1,200/year per 10% loss	CDC Worksite Health ScoreCard
Disability claims	$2,400/year per 10% loss	Finkelstein et al., JOEM 2010
Drug cost (GLP-1)	$12,000/year avg	GoodRx market data 2025

Reference: Cawley J et al. Savings in Medical Expenditures Associated with Reductions in Body Mass Index. J Health Economics 2015.

K-Means clustering identifies distinct patient response patterns across the population:

Segment	Characteristics
Strong Responder	>15% weight loss, high adherence, consistent engagement
Moderate Responder	5-15% weight loss, regular refills
Plateau	Initial weight loss followed by stagnation
At Risk	Declining engagement, widening refill gaps
Disengaged	Minimal observations, low adherence

Segmentation enables targeted intervention strategies rather than one-size-fits-all approaches.

All clinical data undergoes validation before entering the analytics pipeline:

Reference: Regenstrief Institute. LOINC — Logical Observation Identifiers Names and Codes. https://loinc.org

Program retention is modeled using Kaplan-Meier survival analysis with optional Cox Proportional Hazards for covariate adjustment.

This produces the retention curve shown on the dashboard, answering: "What percentage of patients remain active at day X?"

Reference: Kaplan EL, Meier P. Nonparametric Estimation from Incomplete Observations. JASA 1958;53(282):457-481.

PEER REVIEW

This methodology is reviewed by our clinical advisory board and updated quarterly. For questions or feedback, contact clinical@pathriva.com