Do First Generics Reach Medicaid Pharmacy Markets Equally After Approval? An ML.NET Framework for Measuring Diffusion and Access Gaps
Year: 2026
Type: article
Abstract: Generic drug approvals are expected to improve medication access, increase competition, and reduce treatment cost. However, regulatory approval does not necessarily mean that a generic medicine becomes visible across all pharmacy markets immediately after launch. Some generic products may first appear in high-volume markets, while others may spread broadly across many markets, remain delayed, or remain non-observed in public utilization data for several quarters. This study proposes a reproducible data engineering and machine learning framework to measure and forecast generic drug diffusion after first generic approval using Numeric Distribution (ND), Weighted Distribution (WD), and ML.NET. The study uses FDA first-generic approval records as launch-start proxies and Medicaid State Drug Utilization Data as a public proxy for state-level pharmacy access markets. A SQL Server pipeline was developed to clean first-generic launch records, import Medicaid utilization records, match brand and generic product names, calculate quarterly state-level diffusion metrics, and generate ML.NET training data. Numeric Distribution measures the percentage of active state-level molecule markets where generic utilization is observed. Weighted Distribution measures the share of total molecule prescription volume represented by the states where generic utilization is observed. The difference between WD and ND, called Access Gap, is used to identify whether generics reached high-volume markets before broad geographic diffusion. The final dataset produced 1,188 quarterly ND/WD observations and 444 first-four-quarter launch observations across 111 complete generic launch cases. The ML.NET training dataset contained 333 rows, with 213 Fast adoption rows, 12 Medium adoption rows, and 108 Slow adoption rows. The ML.NET multiclass classification model achieved 95.18% micro accuracy, 66.67% macro accuracy, and 0.2232 log loss. The lower macro accuracy reflects the limited number of Medium examples and is an important indicator of class imbalance. The results reveal several meaningful launch diffusion patterns. Droxidopa showed the largest high-volume-first pattern, with Q1 ND of 16.67% but WD of 84.30%, producing a 67.63-point Access Gap. Tasimelteon showed a similar pattern, reaching only 14.29% ND in Q2 but 76.29% WD, meaning one observed state represented most molecule prescription volume. Dabigatran etexilate showed persistent high-volume-first diffusion, remaining below 50% ND while maintaining WD above 84% during the first four quarters. Liraglutide, acetaminophen/ibuprofen, amantadine, amlodipine, budesonide/formoterol, and several other launches showed immediate broad and high-volume diffusion. Medium adoption cases included darunavir, deflazacort, pazopanib, and saxagliptin. Several slow cases showed no observed generic diffusion in the first four quarters under the current public-data matching method. This study demonstrates that ND and WD provide richer insight than prescription volume alone. A generic launch can appear limited by numeric state coverage but still cover most prescription volume if it reaches high-volume states first. The paper contributes a .NET-based health informatics pipeline, an ND/WD measurement framework for public pharmaceutical data, and an ML.NET model for classifying generic adoption patterns. The study should be interpreted as an applied computer science and data engineering contribution rather than a clinical or pharmacological evaluation.
Author: Khan, Tanvir Mahmud
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7165934041
FWCI:
Cites:
Cited by:
Related to: