Preventing adverse drug reactions (ADRs) is a major challenge for health care systems worldwide, both to improve patient outcomes and to reduce economic burdens [1,2]. Monitoring drug exposure at both the patient and institutional levels is a key strategy to address this issue. Furthermore, information on how drug exposure influences patient response is fundamental to evaluating a medication’s safety and effectiveness. Therefore, various tools such as Computerized Provider Order Entry, clinical decision support systems (CDSS), and deprescribing programs have been implemented to limit unnecessary or inappropriate prescriptions [3,4]. Nonetheless, significant challenges remain: generating only relevant alerts to avoid clinician alert fatigue and providing meaningful indicators to assess the effectiveness of deprescribing programs. All of these initiatives depend on accurate assessment of patients’ drug exposure across all stages of care [5].

Learning health system (LHS) aims to address this challenge by capturing data from routine practice, transforming them into knowledge, and reintegrating that knowledge to improve care in a continuous cycle. A typical LHS includes three iterative steps:

Hence, LHSs could contribute to improving current practices in drug exposure measurement by enabling the development of new methods for knowledge extraction from real-world data.

Accurately quantifying drug exposure is fundamental for assessing medication safety and effectiveness. A robust theoretical framework integrates pharmacokinetics, pharmacoepidemiology, and clinical pharmacology, with 4 dimensions: exposure definition, data sources, metrics, and biases [6]. Pharmacokinetic exposure can be described by parameters such as the area under the curve, maximum and minimum concentrations (Cmax, Cmin), or time in therapeutic range; despite being closely related to drug effects, these concentration-based measures are rarely used, even in clinical trials where the administered dose remains the standard [7]. Pharmacoepidemiology defines exposure by drug receipt and refines it with duration, dose, and adherence. Aggregate indicators such as medication possession ratio, proportion of days covered, and the World Health Organization–defined daily dose are useful in chronic care but fail to capture rapid regimen changes or cumulative burden in hospital settings.

Because inpatient treatments vary from day to day, choosing an appropriate observation window is crucial. Assessments limited to admission or discharge underestimate drug burden; a comprehensive evaluation should encompass the entire hospital stay. In addition to quantifying overall exposure, specific prescribing determinants provide complementary insight into the quantity and quality of therapy. Polypharmacy (PP) and hyperpolypharmacy (HPP) reflect the number of concomitant drugs, while potentially inappropriate medications (PIMs) and drug-drug interactions (DDIs) capture the appropriateness of the regimen. These indicators are particularly relevant for older or multimorbid patients.

Whether evaluating overall exposure or these determinants, the same methodological challenges arise. Exposure estimated from prescriptions may overstate administration, while unrecorded medications obtained elsewhere cause underestimation; differences in length of stay introduce further bias, as longer stays inevitably yield more exposure and more determinants. These issues, combined with the dynamic nature of inpatient therapy, expose the limitations of simple point-in-time indicators that record the presence of drugs or determinants at admission, during hospitalization, or at discharge.

To more faithfully characterize drug exposure, especially in contexts of transient high drug burden such as intensive care units or transplantation services, composite measures that integrate both intensity and accumulation are needed. Dose intensity (total dose or number of drugs per unit time), dose density (dosing frequency), and cumulative exposure (sum of doses over the stay) capture both the quantity and the temporal dynamics of medication use. By providing a fuller picture than conventional point-in-time metrics, these cumulative indicators can support more nuanced analyses of prescribing practices and promote safer, more rational medication use.

The reuse of health data has become an essential component of medical research, offering insights into patient care, treatment effectiveness, and disease patterns. Through the integration of EHRs, administrative databases, and insurance claims, researchers can analyze large-scale populations and track health care trends over time [8].

Among the most powerful tools facilitating the practice-to-data step of LHS are CDWs, which aggregate structured and unstructured health information from multiple sources within hospital systems [9]. By consolidating laboratory results, drug administration records, and patient demographics, CDWs provide invaluable real-world evidence for pharmacovigilance, pharmacoepidemiology, and drug safety research [10]. Moreover, unlike traditional clinical trial data, which often lack generalizability to broader populations, they provide real-world insights into prescribing practices, ADRs, and treatment effectiveness in heterogeneous patient groups [11].

Hospital-based CDWs offer a unique advantage by capturing detailed, real-time data on medication administration, making them particularly valuable for drug exposure. Unlike outpatient prescription databases, which, in most cases, indicate only intended treatments rather than actual drug intake, hospital CDWs provide precise records of administered medications, dosage adjustments, and the timing of drug exposure. Moreover, they facilitate precise evaluation of actual drug administration, offering dynamic analyses over the entire course of hospitalization [12,13].

However, despite their potential, the reuse of health data and CDW-based analyses faces several challenges. Data quality issues, such as missing or inconsistent medication records, variations in documentation practices across hospitals, and incomplete capture of drug administration details, can impact the reliability of findings [8]. Additionally, differences in exposure definitions such as whether drug exposure is measured based on prescriptions, administrations, or estimated treatment durations can lead to variability in results. These discrepancies highlight the need for harmonized methodologies to ensure consistency in developing new drug exposure, particularly in hospitalized patients where medication regimens are frequently modified. Standardized approaches to data integration, exposure quantification, and analytical techniques are crucial to improving the robustness and comparability of real-world evidence on medication use.

In this context, we introduce two novel cumulative metrics designed to provide a more comprehensive evaluation of drug exposure and prescribing determinants during hospitalization:

These metrics aim to offer a dynamic, patient-centered evaluation of drug exposure, addressing limitations of conventional static indicators and could serve as quality indicators in initiatives promoting safer and more rational use of medications.

The objective of this exploratory study was to develop and validate, as a methodological proof of concept, 2 CDE metrics (CDE and CDED) derived from real-time hospital CDW data. This work should be considered a preliminary methodological step rather than a definitive evaluation. The proposed new indicators were applied to quantify 4 key prescribing determinants (PP, HPP, DDIs, and PIMs) and were compared with traditional point-in-time measures to assess their added value within an LHS for characterizing drug exposure in hospitalized patients undergoing complex and dynamic treatment regimens.

The data for this study were obtained from three data sources:

PP and HPP were defined based on the daily number of distinct drugs administered during hospitalization. Therefore, the daily number of distinct ATC classes administered during each hospitalization day was computed using structured medication administration data from eHOP. A patient was classified as having PP if the number of ATC classes per day was ≥5, and HPP if the number of ATC classes was ≥10 [18]. Chemotherapy regimens used for transplant conditioning (ATC class L01) administered before the transplantation day were excluded from this calculation to avoid an artificial overestimation of polymedication due to high-dose chemotherapy protocols. This classification was applied daily throughout hospitalization, allowing for the assessment of PP and HPP at admission, discharge, and during the hospital stay.

A DDI refers to a clinically significant unintended modification in drug exposure or response caused by the coadministration of another drug [19]. The detection of DDIs was performed using the Theriaque database, which provides predefined lists of interacting drug pairs (term 1 and term 2) based on ATC codes. A DDI was recorded when both interacting drugs were administered on the same day. DDIs were further categorized into major and moderate severity levels. The daily number of DDIs per patient was computed, and aggregated values were calculated at admission, discharge, and across hospitalization to assess exposure variations over time.

PIMs are drugs for which the risks outweigh the expected benefits in older adult patients, due to an increased risk of ADRs, toxicity, or limited clinical effectiveness [20]. For each patient aged 65 years or older, all administered medications were cross-referenced daily with the GO-PIM ATC-based classification, and any drug listed in GO-PIM was classified as a PIM. The daily number of PIMs per patient was determined, and aggregate values were calculated at admission, discharge, and throughout hospitalization to evaluate temporal patterns of PIMs exposure.

To compare cumulative exposure metrics with traditional methods, conventional metrics were assessed at 2 hospitalization time points: admission and discharge. For each prescribing determinant (PP, HPP, DDIs, and PIMs), presence at admission or discharge was evaluated as a binary variable indicating whether the patient met the definition at these time points.

The number of prescribing determinants at each time point was also quantified as follows:

To quantify the overall exposure to the prescribing determinants throughout hospitalization, 2 cumulative exposure metrics were computed: CDE and cumulative exposure density. CDE was defined as the total number of days a patient was exposed to each prescribing determinant during hospitalization.

For PP and HPP, exposure was recorded as a binary status per day (exposed and nonexposed), and the sum of exposure days was computed across the full hospital stay. However, for DDIs and PIMs, CDE accounted for the number of occurrences per day: if multiple DDIs or PIMs were recorded on a given day, each event contributed separately to the total exposure count. For instance, if a patient experienced 4 distinct DDIs on a single day, this was counted as 4 exposure days in the cumulative measure. To account for variations in hospitalization duration, CDED was calculated by normalizing CDE to the total length of stay, providing an intensity-adjusted measure (Figure 1).

Hematopoietic stem cell transplantation (HSCT) patients represent a particularly high-risk group regarding ADRs, as they are exposed to an exceptionally high medication burden throughout hospitalization. Several studies have explored associations between prescribing determinants such as PP, HPP, PIMs, and DDIs, and adverse outcomes such as mortality or rehospitalization, using conventional metrics [21,22]. While these studies have highlighted potential risks, their findings remain inconsistent, likely due to methodological differences and reliance on conventional, point-in-time metrics. In most cases, drug exposure was assessed only at the time of transplantation, failing to account for the cumulative burden experienced throughout hospitalization. This limitation may lead to an incomplete assessment of medication-related risks, reinforcing the need for more dynamic and temporally sensitive approaches to exposure assessment.

HSCT treatment regimens involve prolonged use of antineoplastics, immunosuppressants, antimicrobials, and supportive therapies, making these patients especially vulnerable to PP, HPP, DDIs, and PIMs [21,23]. Given the duration and complexity of treatment, accurately quantifying CDE of prescribing determinants is essential to identify medication-related risks such as increased toxicity or adverse drug events. It could also help optimize therapeutic strategies by adjusting drug regimens and limiting unnecessary medication burden while refining prescribing guidelines through real-world evidence on exposure patterns and associated risks.

We focused on all adult patients hospitalized at Rennes University Hospital between November 1, 2020, and November 30, 2021, for autologous HSCT. Patients were identified through the diagnosis-related group of their hospital stay (ie, 27Z03) available in the Rennes University Hospital CDW, corresponding to autologous HSCT patients. Comorbidities were assessed using the Charlson Comorbidity Index, calculated from ICD-10 (International Statistical Classification of Diseases, Tenth Revision) diagnostic codes available in the CDW. For patients with multiple admissions during the study period, only the first hospitalization during which the transplant procedure was performed was retained for analysis, serving as the index stay for the computation of cumulative exposure metrics.

The study was conducted in accordance with the ethical principles of the 1964 Declaration of Helsinki and its amendments, or equivalent standards. Given its retrospective and noninterventional design based on existing database data, a waiver of informed consent was granted. Confidentiality and anonymity were strictly maintained throughout: the extracted data contained no nominative information and were fully deidentified prior to analysis, in compliance with the institutional data protection procedures. The protocol was approved by the local ethics committee (approval 24.12, May 2024) and complies with French clinical research regulations. The study used routinely collected hospital data. No compensation was provided to participants and no direct contact with patients occurred.

This study included 69 adult patients hospitalized at Rennes University Hospital between November 1, 2020, and November 30, 2021, for autologous HSCT. Patients were identified using the diagnosis-related group code (eg, 27Z03). For each patient, only the hospitalization during which the transplant procedure was performed was retained as the index stay, so that each patient contributed a single stay to the analysis. The cohort’s median age was 60 years (range 21‐70 years), with 22 patients out of 69 (31.9%) aged 65 years or older. The median length of hospital stay was 19 days (range 14‐52 days).

The primary indications for HSCT were multiple myeloma (44/69, 64%), non-Hodgkin lymphoma (19/69, 28%), and Hodgkin lymphoma (6/69, 9%). Patients exhibited a median Charlson Comorbidity Index score of 4, reflecting a significant burden of comorbid conditions.

This exploratory study aimed to develop and evaluate 2 CDE metrics: CDE and CDED, and to compare them to conventional, point-in-time indicators in a cohort of autologous HSCT recipients. These metrics were designed to capture the number of days during which patients were exposed to PP, HPP, DDIs, and PIMs, with CDED normalizing this duration by length of stay. Compared with binary measures at admission or discharge, CDE and CDED offered a more nuanced picture of drug burden by incorporating treatment duration and intensity.

Across all prescribing determinants, the number of drugs and DDIs at admission showed no significant correlation with CDE or CDED, and the presence of PP, HPP, or DDIs at admission was not associated with higher CDE or CDED values. These findings suggest that admission-based conventional metrics do not reliably reflect the overall drug burden experienced by patients throughout hospitalization. Conversely, while the number of medications at discharge was not correlated with cumulative exposure to PP, a moderate correlation was observed with cumulative exposure to HPP. Similarly, the number of DDIs at discharge showed a significant positive correlation with cumulative exposure to DDIs.

These findings can be explained by the specific therapeutic trajectory of autologous HSCT. Drug exposure evolves dynamically throughout the hospital stay; it is not determined solely by the patient’s status at admission or discharge but by successive clinical phases such as conditioning, aplasia, and management of complications or auto-graft versus host disease [25-27]. Numerous medications (antimicrobials, antivirals, antifungals, immunosuppressants, and supportive care drugs) are initiated to prevent or manage infections, mucositis, nausea, and hematologic toxicities, and many are maintained until discharge and sometimes beyond [23,28]. This progressive accumulation of treatments contributes to cumulative exposure and helps explain why discharge metrics better correlate with cumulative measures.

The heterogeneity of associations reported in the literature between conventional metrics and clinical outcomes in malignant hematological diseases further underscores the limitations of static indicators. Studies have linked PP, PIMs, and DDIs to increased toxicity, reduced survival, or poorer transplantation outcomes, but the strength and consistency of these associations vary widely depending on definitions, study populations, (autologous vs allogeneic HSCT), and the timing of measurement [21,22,29,30]. Such divergences highlight the need for dynamic and integrative exposure measures that better reflect the complexity of pharmacological trajectories.

To further explore whether cumulative metrics capture distinct aspects of drug exposure, we conducted an FAMD. When CDE variables were included, the principal components were largely driven by conventional metrics such as the number of medications and the presence of PP or DDIs at admission and discharge; CDE variables were less strongly represented, suggesting that they capture specific dimensions not fully reflected by static measures. In contrast, when CDED was included, this exposure density measure contributed more substantially to the first dimensions, alongside some conventional metrics, indicating that normalizing cumulative exposure by length of stay captures relevant contrasts between patients.

PIMs were not incorporated into the FAMD because they apply only to patients aged  65 years and older, and including a determinant restricted to a subpopulation would have introduced structural heterogeneity. The dynamics of PIM exposure, often linked to chronic or preexisting treatments, differ from those of other determinants: conventional PIM metrics tend to align closely with cumulative measures because these medications are commonly part of long-term regimens started before hospitalization and continued during and after the inpatient period [31,32].

This study has several limitations. First, the relatively small sample size (n=69) and the single-center design inevitably limit the statistical power of the analyses and the generalizability of the findings. Accordingly, this work should be interpreted primarily as an exploratory methodological development rather than as a definitive evaluation of drug exposure in HSCT patients. The study population consisted exclusively of autologous HSCT recipients, a highly specific group treated under standardized therapeutic protocols. This deliberate choice was made because these patients are exposed to dense and complex drug burden, providing a relevant framework to test and refine the proposed metrics. However, the conceptual approach underlying CDE and CDED is not restricted to autologous HSCT recipients: these cumulative measures are designed to be clinically agnostic and can be applied to other hospitalized populations, regardless of their pharmacological profile. While their relevance may be particularly evident in contexts of high therapeutic intensity such as oncology, intensive care, or surgery, they are generalizable to any setting where drug exposure evolves dynamically over time. Future work should therefore focus on validating these metrics in larger, multicenter, and more diverse clinical populations, ideally including outcome analyses to establish their clinical significance.

Second, clinical outcomes such as toxicity, adverse drug events, or hospital readmission were not assessed in this study. While this limits the immediate clinical applicability and interpretation of cumulative exposure, the primary purpose of this work was methodological, aiming to develop and illustrate novel exposure metrics. However, ongoing work is being conducted to evaluate the relationship between these metrics and patient outcomes, in order to support their prognostic and practical value.

Third, although the use of a CDW represents a key strength by offering precise, structured information on actual drug administrations, it is not without limitations. In particular, CDWs do not systematically capture clinical indications, prescriber rationale, or therapeutic intent in a structured format. Retrieving such information would require manual review of unstructured clinical documents such as discharge summaries, without guarantee of consistency or completeness. This limitation can complicate the interpretation of the clinical appropriateness of certain prescribing patterns such as PP, HPP, or DDIs, which may, in some cases, be justified depending on the clinical context, severity of illness, or treatment objectives. Furthermore, certain contextual elements such as patient-brought medications or off-protocol adaptations may not be consistently documented in structured data fields.

The cumulative nature of drug exposure, although conceptually aligned with long-standing practices in public health, remains underexplored in pharmacoepidemiology. Established exposure metrics such as pack-years in tobacco research or cumulative indices in occupational health have long recognized the importance of duration and timing. Conceptually, this approach aligns with the broader exposome framework, which aims to characterize the totality of environmental, biological, and behavioral exposures an individual encounters across the life course [33].

Within this framework, pharmaceuticals represent a major, yet insufficiently characterized component of the chemical exposome, particularly regarding their longitudinal accumulation. Although medications are intentionally administered, their cumulative impact, especially in contexts of PP and prolonged or intensive treatment, can significantly modulate health outcomes beyond the intended therapeutic effect [34].

Recent methodological developments have sought to better account for the temporal dimension of drug exposure. Among them, the weighted cumulative exposure model has extended time-sensitive modeling to pharmacological exposures by estimating the effect of prior doses using flexible, spline-based weight functions over specified exposure windows [35]. However, these approaches rely predominantly on prescription or dispensing data and are implemented within regression-based analytic frameworks, requiring prior assumptions about exposure timing and pharmacological latency. While such models offer valuable insights for studying long-term drug effects in ambulatory care, they are less suited to settings where drug exposure is highly dynamic and context dependent. In this context, CDE and CDED leverage real-time administration data recorded in hospital CDWs, enabling standardized and reproducible quantification of actual in-hospital drug burden.

At the population level, CDE and CDED offer valuable tools to inform hospital-wide strategies for medication management. By providing standardized, temporally resolved indicators of drug exposure, they could be integrated into institutional dashboards to monitor prescribing practices across wards or clinical services. For example, identifying units with persistently high cumulative exposure to PP may help guide resource allocation, staff training, or targeted medication review initiatives. Beyond real-time monitoring, these metrics also allow for longitudinal tracking of drug exposure trends, making them particularly relevant for evaluating the impact of deprescribing policies or institutional changes in prescribing practices. Such applications directly respond to limitations identified in previous work, which highlighted that commonly used point-in-time indicators fail to adequately capture the dynamic nature of drug exposure during hospitalization and emphasized the need for cumulative, temporally informed metrics better suited to quality monitoring and prescribing optimization [36-38].

At the individual level, these new metrics could enhance both the specificity and the clinical relevance of decision support systems. Most CDSS currently rely on static medication lists and rule-based logic applied at fixed points in time, often ignoring how long or how intensively a patient has been exposed to potentially risky prescriptions. This contributes to high rates of nonspecific alerts and clinician alert fatigue [39,40].

By incorporating cumulative exposure into alerting mechanisms, CDSS could move beyond simple presence or absence logic to integrate the duration and intensity of drug exposure, thereby identifying patients with a sustained drug burden more accurately. For example, alerts could be designed to trigger only when exposure to high-risk interactions or inappropriate medications exceeds a predefined cumulative threshold, supporting more clinically meaningful interventions and reducing unnecessary interruptions. Embedding these indicators into patient-level dashboards within EHRs would also provide clinicians with a longitudinal view of drug burden at the bedside, directly informing prescribing and deprescribing decisions.

However, the definition of clinically validated cumulative exposure thresholds remains to be established. Demonstrating associations between these thresholds and meaningful patient outcomes is a critical next step, beyond the methodological proof-of-concept presented here, and would provide the necessary foundation for their integration into CDSS. This represents a relevant direction for future research, particularly for refining risk stratification and alert prioritization within digital clinical tools.

In summary, CDE and CDED address key limitations of existing exposure indicators by leveraging real-world, time-resolved data. Their integration into both institutional governance frameworks and clinical decision tools opens new perspectives for optimizing pharmacotherapy and promoting safer, more appropriate medication use in hospital settings.

This study proposes 2 cumulative exposure metrics, CDE and CDED, derived from daily drug administration data recorded in hospital CDWs. By incorporating the temporal dimension of treatment, these metrics offer a more comprehensive assessment of in-hospital drug exposure than conventional point-in-time metrics. Their reproducibility and adaptability support applications ranging from research to quality monitoring and medication safety. Further analyses are underway to examine how these cumulative profiles relate to patient trajectories and clinical events and to explore their potential contribution to hospital surveillance and prescribing evaluation efforts.