Time Windows Used When Identifying Current Drug Use and Polypharmacy Authors: Sabrina Giometto, Mette Reilev, Martin Thomsen Ernst, Henrik Støvring, Anton Pottegård Published: 2026 ABSTRACT ABSTRACTPurposeThe length of the time window used to assess “current drug use” or “number of medications used” will influence the estimates hereof; however, no consensus exists on the optimal width of such time windows. We aimed to explore how the estimated prevalence of drug use in general, and of polypharmacy in particular, is affected by definitions used.MethodsWe conducted a drug‐utilization study divided into two parts. In the first part, we focused on current drug use. Using population‐based registries from Denmark, we identified adults (i.e., individuals aged ≥ 18) during 2020–2022, and among them, current use of different drugs, including those with typically chronic or episodic patterns of use. The second part of the study focused on polypharmacy. We estimated its prevalence, based on different definitions, using population‐based registries from Denmark in a cohort of older adults (i.e., individuals aged ≥ 65) in 2022. We also evaluated the accuracy of different criteria for predicting polypharmacy using simulations.ResultsEvaluating current drug use, the proportion of individuals classified as exposed increased with the length of the time window for all drugs, reaching a plateau considering a 120–150‐day window for statins, glucose‐lowering drugs, and selective serotonin reuptake inhibitors, and a 180–300‐day window for opioids, whereas no plateau was reached for non‐steroidal anti‐inflammatory drugs within 360 days. The prevalence of polypharmacy ranged from 21% (10 different 4th level Anatomical Therapeutic Chemical (ATC) groups in 1 year) to 92% (two different 4th level ATC groups in 1 year) depending on the applied definition. In the simulation, the best criterion for identifying polypharmacy required at least two dispensations during the one‐year study period for each of at least five drugs, with sensitivity ranging between 0.93 and 1.0, and specificity between 0.72 and 1.0.ConclusionsTime windows up to 120 days are too short to identify baseline drug use in the Danish setting. How polypharmacy is defined significantly influences its estimate, suggesting a need to use multiple definitions in each study.Keywords: current drug use, polypharmacy, prevalence, registry data, time window length Purpose Pharmacoepidemiologic studies mainly rely on secondary data, meaning that the variables of interest, including outcome and exposure, must be operationalized, that is, constructed from raw data. To avoid measurement error, it is crucial to define and measure exposure to drugs accurately, most commonly using predefined time windows of 60, 90, 120, 180, or even 360 days to assess drug use [1, 2, 3]. However, the estimated prevalence of use is known to be sensitive to the length of the time window [4]. Estimating the prevalence of polypharmacy presents additional challenges, as the impact of the length of the time window increases when multiple drugs are involved and as there is no definitive cut‐off for the number of concomitant medications required to define polypharmacy. The aim of the study was to explore how time windows considered for the assessment of drug use affect estimates of current (prevalent) drug use, and to investigate polypharmacy prevalence by explicitly considering the time window applied, the number of concomitant medications required, and the assessment method used. Methods The study was divided into two parts: the first part focused on current drug use, corresponding to “baseline drug use” in a cohort study, estimating its prevalence using population‐based registry data. The second part focused on polypharmacy, estimating its prevalence using registry data, and evaluating the accuracy of different criteria for predicting polypharmacy using simulations. Data Sources—Registry Data We obtained information on birth, death, and migration dates from the Danish Person Registry [5] which was linked with information from The Danish National Prescription Registry [6] on pharmacy dispensations (date, ATC‐code, and amount) and hospitalizations from Danish National Patient Registry [7] (date of admission and discharge, ICD‐10 diagnosis). The Danish dispensing system is based on original pack dispensing, that is, typically packages of 100 tablets for most drugs, and the days supply is not recorded explicitly. Linkage was done based on an encrypted version of the unique person identifier assigned to all individuals living in Denmark [5]. Data Sources—Data Generating Mechanisms for Simulations We simulated multiple study settings, each defined by a mean number of drugs per individual (five or eight) and a median refill gap (i.e., time from one dispensation to the subsequent) of 60, 75, 90, or 105 days. For each setting, we generated 10 000 individuals, assigning each a Poisson‐distributed number of drugs. Refill gaps between dispensations followed a log‐normal distribution within a one‐year period, with a standard deviation on the log‐scale of 0.5, consistent with empirical estimates [8]. A more explicit description of the data generating mechanism for the simulation and the simulation itself are reported in Table S1. Study Population We identified two different main cohorts using registry data, one was applied to investigate the prevalence of current drug use, and the other to investigate the prevalence of polypharmacy. The first main cohort included individuals aged 18 years or older who experienced a myocardial infarction (ICD‐10‐CM code I21*) during 2020–2022, for whom the myocardial infarction date was considered as index date. We considered three additional cohorts using as index date: (i) the date of death; (ii) a random date during 2022; (iii) the date of osteoporotic fracture (ICD‐10‐CM code M80*). The second main cohort, used specifically to investigate the prevalence of polypharmacy, included individuals older than 65 in 2022. Study Drugs To estimate the prevalence of baseline drug use we identified all dispensations in 2021–2022 of drugs used chronically—that is, those with a regular pattern of use over time—and episodically—that is, those typically used a shorter period of time or on an as‐needed basis. Drugs expected to be used chronically included: statins (ATC: C10AA*), glucose‐lowering drugs (ATC: A10*), and selective serotonin reuptake inhibitors (SSRIs) (ATC: N06AB*). Drugs expected to be used predominantly on an episodic basis included: opioids (ATC: N02A*), and non‐steroidal anti‐inflammatory drugs (NSAIDs) (ATC: M01A*, excluding M01AX*). To estimate the prevalence of polypharmacy we identified all dispensations in 2021–2022, excluding anti‐infectives for systemic use (Anatomical Therapeutic Chemical (ATC) code: J* and P01AB01)—the largest group of clearly episodic drugs—, to focus on chronically used concomitant drugs. Exposure Definitions To estimate the prevalence of baseline drug use we considered an individual exposed if they had at least one dispensation within a predefined period from the index date, considering different lengths for these periods ranging from 10 to 360 days, with 10‐day increments. To examine the prevalence of polypharmacy we used five different definitions, four based solely on whether a dispensation was filled, and one that also considered the duration of the single dispensation: (i) five dispensations with different ATC codes (fifth level) in 2022, that is, within 365 days; (ii) two to 10 dispensations with different ATC‐groups (forth level) in 2022, that is, within 365 days; (iii) five dispensations with different ATC‐groups in the second half of 2022; (iv) five dispensations with different ATC‐groups in the last quarter of 2022; (v) five overlapping dispensations at any point during 2022, when the duration of each dispensation was defined as: 90 days, 120 days, equal to the number of tablets dispensed (90 days, if the table count was not available), and equal to number of defined daily doses (DDDs) dispensed. Analysis To estimate the prevalences of baseline drugs, we calculated the number of individuals exposed and divided it by the size of the population at the index date. To estimate the prevalence of polypharmacy, we calculated the number of individuals with polypharmacy, according to the five different definitions, and divided it by the size of the population at the index date. To evaluate the accuracy of different criteria for predicting polypharmacy, we performed a simulation study [9]. We defined the true polypharmacy status by considering whether each individual had at least 1 day of overlapping treatment with five or more drugs assuming different dispensation lengths (60, 90 and 120 days), that is, the number of days an individual is considered treated with a specific drug after it has been dispensed. Finally, we applied three criteria for predicting polypharmacy: having at least two, three, or four dispensations during the one‐year study period for each of at least five drugs. Sensitivity and specificity were computed as the proportions of correctly identified individuals with and without polypharmacy, respectively. The analyses were conducted both in R, version 4.2.1 and Stata, version 17.0. Results In the evaluation of prevalence of baseline drug use we observed that the proportion of individuals classified as exposed increased from 0 to 360 days before the index date for all drugs irrespective of index date definition (Figure 1, Figure S1). A plateau, defined as a variation of ≤ 5% over three 10‐day periods, was reached earlier for statins, GLD, and SSRI (120–150 days) than for opioids (180–300 days) and NSAIDs, where it was not reached at all within the 360‐day period when considering date of myocardial infarction as index date. Even though the estimated prevalence of baseline drug use was dependent on the definition of the index date, the variations in estimates depending on the length of the time window were highly consistent across all four index dates. When extending the time window from 1 year to 5 years for the main cohort of individuals who experienced a myocardial infarction, the estimated prevalence of baseline drug use increased by 175% for NSAIDs, 113% for opioids, 67% for SSRIs, 14% for statins, and 5% for glucose‐lowering drugs (Figure S2). Proportion of individuals classified as treated by drug, considering the date of myocardial infarction as index date, based on Danish registry data, 2020–2022. NSAIDs: Non‐steroidal anti‐inflammatory drugs; SSRIs: Selective serotonin reuptake inhibitors. The prevalence of polypharmacy remained relatively stable across definitions (i), (iii) and (v), ranging between 55% (five different ATC groups in the second half of 2022, definition (iii), Figure 2) and 63% (five unique ATC codes). It was sensitive to the included number of different ATC groups, ranging from 21% (10 different ATC groups) to 92% (two different ATC groups), and it dropped to 37% when restricting the time period to a three‐month window (Figure 2). Prevalences of polypharmacy according to five different definitions, based on Danish registry data, 2022. ATC: Anatomical therapeutic chemical; DDD: Defined daily dose. Definitions of estimated prevalence of polypharmacy: (i) five dispensations with different ATC codes (fifth level) in 2022, that is, within 365 days; (ii) two to 10 dispensations with different ATC‐groups (forth level) in 2022, that is, within 365 days; (iii) five dispensations with different ATC‐groups in the second half of 2022; (iv) five dispensations with different ATC‐groups in the last quarter of 2022; (v) five overlapping dispensations at any point during 2022, when the duration of each dispensation was defined as: 90 days, 120 days, equal to the number of tablets dispensed (90 days, if the table count was not available), and equal to number of defined daily doses (DDDs) dispensed. According to the simulation, the best criterion for predicting polypharmacy was to require at least two dispensations for each of at least five overlapping drugs within a 1‐year period, with sensitivity (range: 0.93–1.0) and specificity (range: 0.72–1.0) being stable across settings (Table S2, Table S3). Overall, the sensitivity was drastically affected by the setting, whereas specificity was less affected by it. Discussion This study explored how length of time windows influences estimates of current drug use and how such time windows, the number of drugs considered, and the assessment method used—that is, incorporating or not the days supply—impacts polypharmacy prevalence estimates. We observed that the proportion of exposed individuals increases when using time windows up to 120–150 days, reaching stability thereafter for chronically used drugs, whereas stability was reached later or not at all for those mainly used episodically. The prevalence of polypharmacy ranged from 21% to 92%, according to different definitions, highlighting how the estimate is impacted by the specific definition of polypharmacy used. Regarding the assessment method used, we observed that the estimate of 62.6% based on the definition of 5 filled different ATC groups aligns with those ranging from 57.2% to 62.3% obtained from definitions that take dispensation duration into account. Considering the time window, the 62.6% estimate over 1 year is comparable to the 54.9% observed when considering a 6‐month period, while restricting the window to 4 months reduces the estimate to 37.2%. This highlights how the time window considered affects the estimated prevalence of polypharmacy, in addition to the prevalence of baseline drug use. This study utilizes high‐quality nationwide dispensing data; however, it is uncertain whether a dispensation accurately reflects actual drug consumption. The results are specific to Denmark—a setting with specific prescribing traditions and “original pack dispensing” of full packages of medicines. Thus, study findings are mainly applicable in similar settings like other Scandinavian countries. Nevertheless, considerations regarding appropriate time windows for assessing drug use are universal, and it remains important not to define such windows too narrowly. A previous study compared the estimation of point versus period prevalence, considering different ways of data handling, showing how the former should be preferred when the interest is the average number of individuals on a certain drug [4]. Similar to our findings, prevalence estimates varied considerably depending on the decision rule applied. Another study examined how the run‐in period length affects the estimated incidence of statins new users and again found considerable variation in estimates. Compared to a 10‐year gold standard, shorter periods (eight to 12 months) showed lower positive predictive values (around 48% to 61%), with values increasing as the run‐in period lengthened (up to 89% for 5 years) [10]. In a Danish setting, time windows up to 120 days are too short to identify baseline and total drug use. For drugs used chronically, such narrow windows would lead to underestimation. However, caution is also needed in the opposite direction: for drugs used episodically, overly wide time windows may lead to overestimation, resulting in classifying as a current user someone who is not. Moreover, caution is warranted when analyses require longer observation periods in data sources with limited follow‐up, as requiring a long look‐back period may lead to the exclusion of a substantial proportion of individuals. Nevertheless, based on our analyses, the proportion of individuals classified as treated reaches a plateau after approximately 180 days in most cases, suggesting that an extensive observation window is generally not required for an accurate identification of current drug use. Polypharmacy prevalence is very sensitive to definitions, suggesting a need to use multiple definitions in each study. Conclusion This study empirically shows that the estimated prevalence of drug use is influenced by the time window considered. In particular, polypharmacy prevalence estimates vary depending on the definition used, with shorter time windows yielding markedly lower prevalence. These findings highlight the importance of reporting multiple definitions of polypharmacy and avoiding overly narrow time windows when defining drug use, including polypharmacy. Funding The authors have nothing to report. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Detailed description of the data generating mechanism for the simulation and the analyses conducted within it. Figure S1: Proportion of individuals classified as treated by drug, considering three different index dates. Figure S2: Proportion of individuals classified as treated by drug, considering a 5‐year time window and MI date as index date. Table S2: Sensitivity and specificity of three different criteria for defining polypharmacy within a 1‐year period from simulations, in a setting where individuals had a mean number of drugs of five. Table S3: Sensitivity and specificity of three different criteria for defining polypharmacy within a 1‐year period from simulations, in a setting where individuals had a mean number of drugs of eight. References Associated Data Supplementary Materials Table S1: Detailed description of the data generating mechanism for the simulation and the analyses conducted within it. Figure S1: Proportion of individuals classified as treated by drug, considering three different index dates. Figure S2: Proportion of individuals classified as treated by drug, considering a 5‐year time window and MI date as index date. Table S2: Sensitivity and specificity of three different criteria for defining polypharmacy within a 1‐year period from simulations, in a setting where individuals had a mean number of drugs of five. Table S3: Sensitivity and specificity of three different criteria for defining polypharmacy within a 1‐year period from simulations, in a setting where individuals had a mean number of drugs of eight.