In clinical research, there is a fundamental mismatch between a recorded medical event and a biological reality. When a researcher looks at a database, they see a "prescription filled" or a "medication dispensed." However, the conceptual event they actually care about is "medication ingested." This discrepancy—often called the proxy gap or verification gap—creates a significant layer of uncertainty in real-world evidence (RWE).
The Proxy Problem: Possession is Not Ingestion
Most large-scale studies rely on administrative data, such as insurance claims or electronic health records (EHRs), to determine if a patient is taking a drug. As Fu‐An Chen et al. note, relying on these records introduces uncertainty because there is no direct verification of drug consumption.
The most common metric for adherence in these datasets is the Proportion of Days Covered (PDC), which calculates how many days a patient possessed a medication based on their refill dates. However, Andrew J. Zimolzak et al. point out that PDC measures medication possession rather than actual ingestion. A patient may appear perfectly adherent in the records while the pills remain in the bottle. This creates a "reality check" problem: Qiao Gao et al. found that the real-world effectiveness of diabetes medications is often lower than in controlled trials, likely because the trials ensure consumption while real-world data only tracks the intent to treat.
The Verification Gap: Resistance vs. Non-compliance
This uncertainty makes it difficult for clinicians to interpret treatment failure. If a patient’s condition does not improve, is the drug ineffective (biological resistance), or did the patient simply not take it (behavioral non-compliance)? Atul Butte et al. describe this as a "verification gap" that introduces noise into effectiveness studies.
Min Xu demonstrates this challenge in treating H. pylori: without strict adherence monitoring, it is nearly impossible to tell if a surviving infection is due to antibiotic resistance or a failure to complete the regimen. Similarly, GuiSen Wu highlights that relying on qualitative reports from family members lacks the rigor of objective assays, leading to potential overestimations of how well patients are following their prescriptions.
Confounding in Large-Scale Analytics
The prescription-consumption gap does not just affect individual care; it skews population-level insights. In pharmacogenomics, researchers try to link genetic markers to drug responses. However, Iris Kalka et al. warn that if a family shares a behavioral habit of skipping doses, a researcher might incorrectly attribute their poor clinical outcomes to shared genetics rather than shared non-adherence.
Machine learning models also struggle with this "black box" of outpatient behavior. Juan Lü et al. and Isaac Ronald Ward et al. explain that because administrative records track dispensing rather than consumption, models cannot fully account for a patient's total drug exposure. This is exacerbated by the "OTC blind spot": Justin Reese et al. and Jacob E. Kurlander et al. note that EHRs rarely capture over-the-counter purchases, meaning a patient in a "control group" might actually be self-medicating with the very drug being studied.
The Limits of Current Solutions
To bridge this gap, researchers have turned to digital and behavioral proxies, but these too have limitations.
- Electronic Monitoring: Devices that track when a pill bottle is opened provide better data than claims, but Xiaoqiu Liu et al. note that opening a box is still just a proxy for swallowing a pill.
- Digital Reporting: Jackson A. Killian shows that digital adherence tools can produce false-positive risk signals when patients take their medication but simply forget to log the dose.
- Collection Behavior: Yi Ling Eileen Koh et al. argue that the timing and completeness of prescription collection is a scalable, objective proxy for adherence, yet they acknowledge it still fails to verify correct administration technique at home.
Ultimately, the authors across these studies suggest that until we can bridge the gap between the pharmacy counter and the patient's home, our understanding of drug effectiveness will remain shadowed by the uncertainty of unobserved behavior.