Medication adherence is often discussed as a patient behaviour issue, but in longitudinal claims and electronic health record (EHR) studies it is also a measurement issue. The way adherence and persistence are defined, calculated, and reported can change the interpretation of drug exposure over time.
For clinical development and real-world evidence teams, this matters because claims and EHR databases are used to understand how medicines are prescribed, dispensed, continued, switched, or stopped outside tightly controlled trial settings. This article explains how drug adherence and persistence are commonly approached in longitudinal databases, where the main interpretation risks sit, and why transparent definitions are essential.
Medication adherence describes the extent to which a patient takes medicine in line with the agreed regimen. In database research, drug adherence is usually inferred from records such as prescriptions, dispensing events, or claims. This makes it measurable at scale, but it also means the analysis is usually based on proxies for medicine-taking rather than direct observation of use.
Adherence can be separated into different phases. These include:
Primary and secondary adherence refers to whether a patient obtains the medicine after it is first prescribed. Secondary adherence refers to refill behaviour after treatment has started. This distinction can matter in longitudinal database studies because patients who never fill an initial prescription may be missed by measures that depend on later refill records.
Persistence is related, but distinct. It usually refers to how long a patient continues treatment before discontinuing. A patient may be persistent but poorly adherent if they stay on a medicine while missing doses or refilling late. Another patient may be adherent while taking treatment but stop earlier than expected. Treating adherence and persistence as the same concept can hide clinically meaningful patterns.
This is especially important in longitudinal claims and EHR databases because these data sources follow treatment use over time. They can show whether medicines appear to be started, continued, interrupted, restarted, switched, or stopped, but the interpretation depends on how the study defines each event.
Longitudinal claims and EHR databases can provide repeated observations across patient follow-up. This makes them useful for studying drug adherence and persistence in routine care, where treatment use is affected by prescribing decisions, refill behaviour, treatment changes, multimorbidity, and healthcare utilisation.
Claims databases are often valuable because they capture reimbursed medicines at scale. They can support population-level analysis of dispensed medicines, refill intervals, treatment gaps, and switching patterns.
EHR databases can add different forms of clinical context. They may include prescriptions, diagnoses, encounters, recorded outcomes, patient characteristics, and other information relevant to interpreting adherence patterns. In some settings, EHRs may be linked to dispensing data or other administrative datasets, which can strengthen adherence estimation.
EHR prescribing records can also allow follow-up to begin from the point a therapy was ordered, rather than only from the point a medicine was dispensed. That can be useful when the study question includes primary adherence or early treatment discontinuation.
The value of these databases comes from their scale and longitudinal structure, but neither source should be treated as a direct window into actual medicine-taking.
Claims databases can show when a medicine was dispensed or reimbursed, how much was supplied, and how refill behaviour changed over time. These data can support common adherence analyses, including:
These patterns can be useful in pharmacoepidemiology, health services research, and real-world evidence generation.
The strength of claims data is that they often capture routine care at scale. This can make them less selective than clinical trial data and more reflective of real-world health service use. They can also support analyses across long observation windows, where adherence and persistence may change.
The limitation is that dispensing or reimbursement is not the same as ingestion, correct administration, or correct timing. A filled prescription suggests that medicine was available to the patient, but it does not prove that the medicine was taken as prescribed.
EHR databases can support medication adherence research by capturing prescribing records and clinical information over time. In some studies, adherence can be estimated by comparing medication supply against the prescribed regimen, particularly when prescription and dispensing information can be analysed together.
Clinical Context
EHR data may also help explain adherence patterns. Diagnoses, disease control, comorbidities, clinical encounters, and treatment changes can provide context that claims data alone may not contain. For example, poor disease control may prompt questions about whether a patient had adequate medication supply, whether treatment was changed, or whether the recorded prescription reflected actual use.
Data Interpretation
However, EHR-based adherence estimation is not straightforward. Prescribing records may show what was intended, not what was dispensed or taken. Free-text directions, dose changes, overlapping prescriptions, and discontinuation records can all affect how medication supply is estimated. Different data-processing approaches can produce different adherence estimates from the same underlying records.
When switches or discontinuations are documented in the EHR, they may also inform censoring rules. This can help distinguish a genuine gap in medicine availability from a planned treatment change, although the approach still needs to be stated clearly.
For this reason, EHRs are useful for drug adherence research, but the method used to convert recorded prescriptions into adherence estimates needs to be explicit.
Adherence in longitudinal databases is commonly estimated by calculating whether a patient had medicine available across a defined observation period. Two widely used approaches are medication possession ratio and proportion of days covered.
Common Measures
Medication possession ratio usually compares the amount of medicine supplied with the length of the observation period. Proportion of days covered estimates the proportion of days in a period where the patient is considered to have medicine available.
A practical difference is that medication possession ratio is a simpler calculation based on total days’ supply over a defined period and may exceed 100% if oversupply is included, although it can also be capped at 100% by definition. Proportion of days covered is typically more complex, as it accounts for overlapping prescriptions and requires additional assumptions about how to handle those overlaps.
Both methods rely on assumptions about days’ supply, refill timing, overlapping prescriptions, and what should happen when treatment changes.
Methodological Choices
Several practical choices can alter the estimate. These include:
A study that allows a longer treatment gap before classifying non-adherence may produce different results from one using a shorter gap.
This is why adherence estimates should not be read as purely objective properties of the dataset. They are partly produced by the study design, the available fields, and the rules applied during data processing.
Persistence is usually measured as the duration of continuous treatment before discontinuation. In longitudinal claims or EHR data, this often means defining a treatment start date, following the patient forward, and identifying when treatment is considered to have stopped.
Discontinuation rules
The main methodological decision is the discontinuation rule. Researchers need to define how long a gap can occur before a patient is classified as non-persistent. They also need to decide how to handle restarts, switches, dose changes, and patients who leave the database or reach the end of follow-up.
These choices should be reported as part of the persistence definition, not treated as background processing. A different gap rule, grace period, or censoring rule can change who is counted as persistent.
What persistence does and does not show
Persistence can be particularly useful where the clinical question is about continuing therapy over a defined period. For example, a one-year persistence measure may help describe how many patients remain on a treatment after initiation. But this does not necessarily show whether those patients took each dose correctly during that year.
Adherence and persistence therefore answer related but different questions. Adherence asks how closely medicine use aligns with the regimen during a defined period. Persistence asks how long treatment continues before stopping under the study definition.
Medication adherence studies often use thresholds to classify patients as adherent or non-adherent. A common example is an 80% threshold, although the clinical meaning of any cut-off can vary by medicine, condition, endpoint, and study context.
Thresholds are convenient because they turn a continuous measure into a category. They can also make results easier to report. The risk is that they may imply a level of certainty that the underlying data do not fully support. A patient with 79% coverage and a patient with 81% coverage may be categorised differently, even if their medicine-taking behaviour is similar.
Definitions also affect comparability. If two studies use different permissible gaps, different observation windows, different handling of early refills, or different assumptions about dose changes, their adherence and persistence estimates may not be directly comparable.
For clinical research teams, the key point is not that one definition is universally correct. The definition should match the research question and be reported clearly enough for readers to judge the finding.
Many adherence methods work most cleanly when the focus is one medicine for one condition over a clear follow-up period. Real-world patients often do not fit that pattern. They may have several chronic conditions, multiple prescribers, changing regimens, and medicines with different dosing schedules.
Single-drug measures
In this setting, single-drug adherence may be a limited measure. A patient may be adherent to one medicine and not another. They may stop a drug because of a planned treatment change, an adverse event, a change in diagnosis, or clinical review. Without enough clinical context, a database may record the change but not fully explain it.
Patient-level complexity
Multimorbidity also makes adherence interpretation more patient-specific. Treatment burden, regimen complexity, age, comorbid disease, and healthcare-system factors can all shape whether a patient continues or implements therapy as prescribed. Longitudinal databases can help describe these patterns, but they may not capture the full reasons behind them.
For studies using claims and EHR data, this argues for careful cohort definition, clear treatment episodes, and cautious interpretation when patients are exposed to multiple therapies over time.
The recurring challenge in drug adherence and persistence research is comparability. When studies use different terminology, different measurement rules, and different reporting practices, it becomes difficult to compare findings or apply them across settings.
Consistent terminology
Standard terminology helps separate adherence from persistence and distinguishes initiation, implementation, and discontinuation. Transparent reporting helps readers understand how the analysis handled observation windows, permissible gaps, refills, discontinuation, switching, and data limitations.
Clear methods
This is particularly important for longitudinal claims and EHR databases because each database has its own structure, coding practices, and missingness patterns. A method that works well in one dataset may require adaptation in another. Reporting those choices clearly is part of making the evidence interpretable.
For clinical development and real-world evidence teams, the practical lesson is straightforward: adherence and persistence outputs should be accompanied by enough methodological detail to understand what was measured, what was inferred, and what remains uncertain.
Longitudinal claims and EHR databases can provide valuable evidence on drug adherence and persistence in routine care. They allow researchers to examine treatment use over time, including refills, gaps, switching, discontinuation, and continuation across large patient populations.
Their value depends on careful interpretation. Claims data may show that medicine was dispensed or reimbursed. EHR data may show that medicine was prescribed or clinically reviewed. Neither automatically proves that the medicine was taken correctly, consistently, or for the intended reason.
The strongest adherence and persistence studies are therefore explicit about definitions, data sources, follow-up rules, and measurement assumptions. Clear reporting makes the findings easier to interpret.
Quanticate’s statistical consultancy team supports sponsors with the design, analysis, and interpretation of studies using longitudinal claims and EHR databases, including drug adherence and persistence definitions, treatment gap rules, PDC and MPR approaches, censoring decisions, and reporting frameworks. If you are using real-world data to assess medication use over time, or need support to define and justify adherence measures for your study, request a consultation today.