Persistent drug shortages in the U.S. pharmaceutical market threaten patient access and may raise acquisition costs for pharmacies and health systems. Prior work frequently treats shortages as binary events and applies static study designs, limiting insight into how shortage characteristics—especially duration and causal mechanisms—drive price responses. The authors ask when shortages lead to higher pharmacy acquisition costs and whether the pass-through from shortage signals to prices is heterogeneous across drugs and shortage contexts.
The analysis links U.S. Food and Drug Administration (FDA) drug shortage records with weekly National Average Drug Acquisition Cost (NADAC) prices. The focus is on generic, non-injectable drugs. The preprint reports use of these two data sources to construct weekly observations of shortage status and acquisition price for individual drug products. The authors note that all data produced in the study are available from the authors upon reasonable request.
The study applies three complementary empirical strategies. First, a baseline two-way fixed-effects model is used to estimate the average association between an active shortage designation and weekly NADAC, controlling for time-invariant drug characteristics and common time effects. Second, duration-specific models examine how associations depend on shortage persistence by estimating effects for shortages of different consecutive-week lengths and by relating cumulative shortage weeks to price changes. Third, a double machine-learning framework is employed to estimate heterogeneous conditional average treatment effects (CATEs) across drugs and weeks, allowing exploration of how estimated shortage-price pass-through varies with observable covariates such as market structure and stated shortage reasons.
The preprint reports clustered standard errors in the baseline specification. Specific covariates, model equations, sample sizes, and tuning or validation details for the double machine-learning approach are reported in the full manuscript; the abstract summarizes the principal modeling choices and outcomes.
In the baseline two-way fixed-effects model, an active FDA shortage designation by itself was not associated with a statistically significant increase in weekly NADAC when standard errors were clustered. However, effects differ when shortage duration is considered. Shortages that persisted for more than four consecutive weeks were associated with approximately 7% higher NADAC relative to non-shortage periods. When the authors modeled cumulative exposure, each additional cumulative shortage week was associated with approximately 0.37% higher NADAC. These duration-specific findings indicate that shortage persistence, rather than brief or transient designations, is a key driver of acquisition-cost increases.
Using a double machine-learning framework, the authors estimate Conditional Average Treatment Effects (CATEs) and report wide variation across drugs and weeks. The distribution of estimated CATEs indicates that the price pass-through of shortages is not uniform: some drugs experience much larger acquisition-cost increases than others when subject to shortage designations.
Shortages associated with allocation restrictions, disruptions to raw-material supply or distribution, and cases where a lack of manufacturers is documented were characterized by higher estimated CATEs. In other words, supply-chain constraints and limited supplier competition appear to amplify the degree to which shortages translate into higher NADAC prices.
The results suggest a nuanced view: a shortage label alone does not always raise acquisition costs, but persistent shortages and those tied to concrete supply-chain failures or limited manufacturer presence do. This implies that monitoring efforts and policy responses should focus not only on the incidence of shortages but also on their persistence and proximate causes. Targeted interventions addressing allocation rules, raw-material and distribution vulnerabilities, and incentives to diversify manufacturing could mitigate adverse price effects and protect access.
This work is a medRxiv preprint and has not been peer reviewed. The abstract summarizes modeling approaches and headline results, but details such as full model specifications, sample sizes, validation diagnostics for the machine-learning component, and robustness checks are presented in the full manuscript; readers should consult the full text for those specifics. As stated by the authors, the preprint should not be used to guide clinical practice.
The authors declare no competing interests. They confirm adherence to relevant ethical guidelines and participant-consent procedures. The preprint states that all data produced in the study are available upon reasonable request to the authors.