In medical research, we often treat a "prescription" as a synonym for "treatment." If an electronic health record (EHR) shows a patient was prescribed a statin, or if insurance claims show they picked up a bottle of pills, researchers typically label that patient as "exposed" to the drug.
However, a growing body of research suggests this is a massive leap of faith. There is a persistent "verification gap" between what is recorded in administrative data and what actually happens at the patient’s bedside or kitchen table. This gap introduces significant noise into real-world evidence, making it difficult to distinguish between a drug that doesn't work and a patient who simply isn't taking it.
The Proxy Problem: Possession vs. Ingestion
The fundamental challenge in using Real-World Data (RWD) is that most datasets track medication possession, not medication ingestion.
As Andrew J. Zimolzak and Isaac S. Kohane point out in Early Detection of Poor Adherers to Statins, metrics like the "Proportion of Days Covered" (PDC) are derived from pharmacy fill records. While these are excellent for predicting "Pay for Performance" bonuses for insurance plans, they are imperfect proxies for clinical adherence. A patient may have a 100-day supply of medication in their cabinet, but that provides no guarantee that a single pill was consumed.
This discrepancy creates what Ronald Chow and colleagues call "exposure misclassification" in The association between statin and COVID-19 adverse outcomes. If a study concludes that a drug had no effect, but half the "exposed" group never took the medication, the drug’s true biological potential remains hidden behind a veil of behavioral non-compliance.
Distinguishing Biological Failure from Behavioral Non-compliance
Why does this distinction matter? For clinicians, the "why" behind a treatment failure dictates the next step. If a patient’s HIV viral load remains high despite a prescription for antiretrovirals, the doctor needs to know: is the virus resistant to the drug (biological failure), or is the patient missing doses (behavioral failure)?
Giota Touloumi and Nikos Pantazis emphasize in Virologic and Immunologic Response to cART by HIV-1 Subtype that the inability to monitor adherence introduces a significant confounding variable in longitudinal studies. Without precise data, researchers cannot accurately assess the effectiveness of different drug regimens across diverse populations.
This uncertainty extends to nearly every corner of medicine:
- Cardiology: Elsie Ross and Nicholas J. Leeper found that achieved LDL levels—not prescribed statin intensity—were the true predictors of heart events, likely because achieved levels implicitly account for whether the patient actually took the drug Statin Intensity or Achieved LDL?.
- Pediatrics: Atul Butte and Kendra K. Radtke argue that the "data desert" in pediatric medicine is exacerbated by the lack of adherence verification, which introduces noise into studies of off-label drug use Advancing evidence-based treatment of infectious diseases in children with real-world data.
The Limits of Hardware and Qualitative Solutions
If administrative data is insufficient, can we solve the problem with better hardware or direct reporting? The evidence is mixed.
In Effectiveness of Electronic Reminders to Improve Medication Adherence in Tuberculosis Patients, Xiaoqiu Liu and James Lewis demonstrate that electronic monitor boxes can reduce "poor adherence" by 50%. Yet, even these devices rely on a proxy: the opening of a box. They do not provide definitive proof of ingestion.
Furthermore, qualitative reports—asking the patient or their family—often lack rigor. As noted in research on antipsychotic prescription and conversion to psychosis, subjective reports can lead to an overestimation of compliance, masking the true dose-response relationship and hindering precision medicine.
Toward Robust Causal Inference
To move forward, researchers are developing sophisticated statistical frameworks to handle this "messy" reality.
In Estimating the effect of PEG in ALS patients, Pallavi S. Mishra‐Kalyani and Qi Long utilized a Bayesian causal inference framework to account for missing outcomes and inconsistent treatment use. By acknowledging that an "initiation" of treatment (like a feeding tube) doesn't tell you the "frequency" of its use, they were able to uncover positive effects that simpler analyses missed.
The goal of this research direction is not just to "fix" the data, but to change how we interpret it. By treating medication adherence as a dynamic, behavioral variable rather than a static binary (prescribed vs. not), researchers like Yuan Lu and Mui Van Zandt are building more resilient models of how treatments perform in the "Wild West" of real-world clinical practice Analysis of Dual Combination Therapies Used in Treatment of Hypertension.