About the Client:
A leading healthcare analytics organization partnered with RAS to solve a major industry challenge: understanding and controlling rapidly rising Medical and Pharmacy (RX) costs for employer groups.
Their environment managed vast and complex datasets—Medical and RX claims, census information, MEDC/EDC codes, service classifications, and Johns Hopkins risk indices. While strong in retrospective analysis, the client lacked one critical capability: an AI-driven forecasting system capable of predicting future utilization and cost trends at scale.
They needed accurate monthly, quarterly, and annual projections at both the individual and group level—and they needed it automated.
Challenge:
RAS quickly identified that the client’s roadblocks went beyond building a single predictive model. They faced three core challenges:
Highly Complex and Fragmented Claims Data:
Medical claims carried thousands of unique MEDC/EDC and service codes, while RX claims involved wide-ranging drug codes, risk patterns, and utilization behaviors. Each domain had independent cost drivers and non-linear relationships that made a unified model impossible.
No Scalable or Automated Data Architecture
The client lacked:
- Automated ingestion pipelines
- Standardized data transformations
- Model versioning
- A production-grade forecasting workflow
There was no existing model baseline—RAS had to build a complete AI ecosystem from the ground up.
Need for Future-Proof Infrastructure:
Any solution needed to accommodate:
- New employer groups
- Evolving RX code categories
- Changing MEDC/EDC classifications
- Expanding forecasting windows RAS had to engineer an architecture that could scale without repeated rework.
Solution: A Fully Integrated AI Platform Built for Accuracy, Scale & Automation
RAS delivered an end-to-end predictive platform powered by Azure and Databricks—built to transform the client’s operations from manual analysis to continuous AI-driven forecasting.
Enterprise-Grade Data Foundation
RAS implemented a fully automated Medallion Architecture (Bronze → Silver → Gold), using Azure Data Factory and Databricks to orchestrate ingestion of:
- Medical & RX claims
- Census data
- MEDC/EDC codes
- Service codes
- RX code data
- Johns Hopkins risk metrics
This standardized and cleansed the data to create reliable features for machine learning.
Purpose-Built Predictive Models for Medical and RX Costs
Based on extensive exploratory analysis, RAS built two specialized XGBoost models—one for Medical, one for RX—because each domain had its own distinct cost drivers.
Medical model features included:
- High-impact EDC codes
- Service codes
- Demographic attributes
- Bill types
- Geographic factors
RX model features included:
- RX codes
- Pharmacy risk scores
- Drug categories
- Utilization patterns
- Census attributes
XGBoost was selected for its strength with mixed distributions, missing values, and outliers—ideal for claims data complexity.
Automated ML Operations & Forecasting Engine
Using MLflow and Databricks Workflows, RAS created a production-ready engine that automatically:
- Retrains models
- Generates monthly, quarterly, and yearly forecasts
- Stores predictions in SQL Server and ADLS
- Supports multiple model versions
- Allows parameterized runs (group ID, forecast period, model version)
The result: a fully governed, scalable platform capable of supporting new employers and new code categories with minimal effort.
Impact & Results:
RAS delivered measurable business value and a transformative operational shift.
Exceptional Model Accuracy at Group Level
- 98% accuracy for Medical cost predictions (R² = 54%)
- 94% accuracy for RX cost predictions (R² = 72%)
These results validated that the models captured true cost drivers across EDC, MEDC, Service, and RX code ecosystems.
Operational Transformation
The client now benefits from:
- Continuous, automated forecasting
- Centralized, high-quality data pipelines
- Governed ML model lifecycle
- Scalable architecture supporting new employer groups and code hierarchies
Strategic Advantage
With RAS’s solution, the client now operates on par with top healthcare intelligence platforms—offering proactive insights, early identification of high-cost populations, and stronger renewal strategies.