Drug trials can be conceptualized as a dynamic system, where the investigational product (i.e., “the drug”) serves as the input, and the resulting estimates of drug efficacy and safety represent the output (see Fig. 1). Within this system, the individuals selected to participate in the study are carefully chosen to represent specific characteristics of the target population.

Fig. 1

Visual representation of the systemic perspective on drug trials

Curiously, despite substantial efforts and investments directed towards ensuring the selection of the population and sound output assessment, relatively little attention is devoted to guaranteeing the quality and precision of the input data. Similar to other logical processes, the adage “garbage in, garbage out” holds true, emphasizing the importance of reliable input data about the drug to ensure meaningful output. In drug development, typically, the study protocol outlines a dosing regimen that is assumed to be followed religiously by the patients involved.

However, it is crucial to acknowledge that in ambulatory care settings, adherence to the dosing regimen specified in the study protocol often deviates from expectations. In 2012, Blaschke et al. [1] conducted an analysis of a database containing electronically compiled dosing history data from 95 clinical studies. Their findings revealed that half of the 16,907 study participants exhibited substantial deviations from the dosing regimen outlined in the study protocol.

For instance, when it comes to once-daily medications, only a small fraction of study participants adhere strictly to a 24-h dosing schedule. Instead, most patients exhibit variability in the timing of their medication intake, typically within a few hours interval. This common variability in medication adherence is depicted in Fig. 2 for 4 patients with typical patterns of dosing schedules while having taken 100% of their prescribed doses.

Fig. 2

Dosing chronology plots of four patients. Index date of follow-up in the study is shown on the horizontal axis, and 24-h clock time is shown on the vertical axis. Blue dots indicate the electronically recorded time and date of dosing. All four patients adhered 100% to the prescribed one dose per day, but there were variations in the timing of intake

Furthermore, besides variability in the time of drug intake, occasional occurrences of missed doses or extra doses are not uncommon. For instance, in a study analyzing drug trials for once-daily prescribed anti-hypertensive medications, Vrijens et al. [2] emphasized that approximately half of the patients experienced a monthly rate of missing a single day’s dose. Figure 3 depicts patterns of single missed doses and errors in medication intake that are common in ambulatory care.

Fig. 3

Dosing chronology plots of four patients. Index date of follow-up in the study is shown on the horizontal axis, and 24-h clock time is shown on the vertical axis. Blue dots indicate the electronically recorded time and date of dosing. The vertical tan lines depict missed doses

Interestingly, this suboptimal adherence behavior is often considered acceptable and even desired in drug trials, as it reflects how the medication is likely to be taken in real-life situations. Acknowledging the inherent variability in patients’ adherence patterns ensures that the findings derived from the trial are more representative of actual clinical practice. It accounts for the diverse factors that can influence adherence, such as individual routines, lifestyle demands, and occasional forgetfulness or mistakes [3].

By allowing for this accepted variability in medication exposure, clinical trials aim to generate findings that can be extrapolated to real-world scenarios. This approach increases the generalizability and applicability of the study results once the drug is available on the market. It acknowledges that patients’ adherence behaviors in routine clinical practice may not follow a strictly predefined dosing regimen.

Extensive research on medication adherence in drug trials reveals a concerning prevalence of more significant adherence errors. In addition to the previously mentioned variations in timing and occasional missed or extra doses, more severe issues such as interruptions in dosing, or even complete discontinuation of medication are frequently observed within these trials [1]. The occurrence of drug holidays, characterized by two or more consecutive days without medication dosing, poses a particular concern. In such instances, participants intentionally or unintentionally cease taking the medication for a brief period or longer (see 4 examples in Fig. 4). Conversely, overdosing is also prevalent in most trials, characterized by the intake of additional doses or the consumption of doses too closely together. This issue is a cause for concern as it increases the likelihood of encountering side effects. These findings shed light on the complex nature of medication adherence within drug trials, indicating that it extends beyond minor deviations in dosing timing. The identification of more significant adherence errors emphasizes the importance of understanding the full spectrum of adherence behavior and its potential impact on treatment outcomes.

Fig. 4

Dosing chronology plots of four patients. Index date of follow-up in the study is shown on the horizontal axis, and 24-h clock time is shown on the vertical axis. Blue dots indicate the electronically recorded time and date of dosing. The vertical tan lines depict missed doses. Extended periods without dosing (drug holidays) are shown by vertical tan bars, the width of which reflects the number of days without dosing

A motivating example highlighting the issue discussed in this paper can be seen in the disappointing outcomes of the MOUNTAIN Phase III study, which investigated zuranolone in major depressive disorder (MDD) [4]. As it is typical for the interpretation of Phase III studies according to the intention-to-treat (ITT) principle, the authors report incredibly high medication adherence rates, claiming ‘overall adherence to study drug was 98.3%’. However, this figure is highly improbable in a population known to struggle with adherence issues [1]. Noncompliance with antidepressant therapy remains a significant concern for MDD patients. Strikingly, despite this purportedly exceptional adherence rate, an exploratory post hoc analysis revealed that 9% of the samples showed no detectable plasma zuranolone concentration. When these patients with no measurable plasma zuranolone concentration were excluded, a notable difference in clinical outcomes between intervention and control groups was observed, which was not the case in the ITT analysis.

To avoid these types of discrepancies, it is essential to employ precise and reliable measurement methods of medication adherence that can differentiate between minor and more significant deviations.

1.1 Methods of Measuring Medication Adherence in Drug Trials

Medication adherence plays a vital role in determining the efficacy and safety of medicines. Therefore, it is crucial to assess adherence with precision and accuracy during drug trials. A review by Mantila et al. [5] has reported the frequency of methods used to measure medication adherence in registration trials, which resulted in the approval of new medicines in Europe. In the following sections, we will explore the theoretical framework of these measurement methods and examine their actual performance, shedding light on the challenges encountered during trial execution. The following items will provide a description of the authors’ experience in utilizing those measures to evaluate medication adherence. We will present them in the order corresponding to their reported usage.

1.2 Pill/Dose Count: Used in 90.2% of Trials

In theory, the method consists in counting manually the number of pills or tablets remaining in a medication container at specific time points, typically at study visits. The assumption behind pill count is that the number of pills dispensed minus the number of pills remaining reflects the number of doses that have been taken by the study participant. By comparing the expected number of pills to be taken with the actual count, researchers can estimate the average level of adherence to the prescribed medication regimen. While manual pill count is simple, and well-accepted, it is a sparse method providing only an average measure of drug consumption and it is well known to have several potential sources of measurement error, including human error in counting pills, patients not consuming the pills removed from the packaging, and instances where patients drop pills prior to their visit.

In drug trials for registration, the situation may be worse as both sponsors and investigators have strong incentives to adhere to the study protocol, which leads to a high emphasis on recruiting adherent patients. However, pill counting suffers from a significant drawback: it is susceptible to the creation of fraudulent reports that indicate good adherence [6]. Patients are encouraged to bring back empty study drug packages to avoid retraining by investigators on study requirements and eventually being excluded from the trial. This results in a recognized desirability bias towards favorable outcomes in pill count measurements, which is further amplified when adherence data from excluded non-adherent patients are censored for further evaluation and thus not reported.

To inflate adherence estimates, an arbitrary threshold, often around 80%, is frequently set to dichotomize between an adherent and non-adherent patient, allowing for some tolerance in adherence levels. However, this low threshold is seldom justified using pharmacometrics analysis and may permit gaps in dosing that compromise the effectiveness of the medication, for instance, it allows a treatment interruption of 2 full weeks out of 10.

Furthermore, it is important to note that while returned tablet counts are typically recorded in the randomization system (IRT/RTSM) for drug accountability, they often do not make their way to the study data repository (EDC). As a result, pill count data are often unavailable to study statisticians and are not utilized for risk-based quality management (RBQM). For instance, when Rudd et al. reported in 1988 [7] that a significant number of patients (35%) participating in drug trials had pill counts well above 100%, it should have served as a warning sign. However, regrettably, even to this day, such cautionary measures are often overlooked.

In summary, manual pill counting is a convenient system that masks the uncomfortable truth of medication nonadherence in clinical trials. It is often employed by investigators as a checkbox method to document presumed good adherence, regardless of the actual level of drug exposure. This information is typically utilized to assert a high level of adherence to the prescribed dosing regimen outlined in the protocol, which is essential for interpreting data analysis based on the well-established ITT principle.

1.3 Patient Self-Report: Used in 27.0% of Trials

In theory, patient self-reporting of medication adherence encompasses a wide range of methods, including retrospective questionnaires, prospective diaries, or electronic diaries (e-diaries). It is often used in complement to pill count.

In practice, similar to manual pill counting, retrospective questionnaires are susceptible to desirability bias, making it easy to generate a favorable adherence record by simply answering positively to the questions. The wide choice of questionnaire (e.g., 121 as reported by Kwan et al. [8]), method of administration (e.g., at the study site vs at home, in person vs by phone), the person administering the questionnaire (e.g., independent nurse or study investigator) and their attitude (empathic vs authoritative) can significantly influence the results, leading to considerable variability and lack of reliability. The retrospective time period covered by the questionnaire (e.g., 1 day, few days, 1 month, or unspecified) is also a major factor contributing to the variability. A recall period longer than 4 days backwards about medication adherence is nearly impossible for anyone, in particular if the medication-taking process is habitual, which is desirable [9].

Prospective diaries, when recorded on paper (i.e. paper diaries), encounter similar challenges [10]. Non-adherent patients often fill in the diaries hastily in the parking lot or waiting room before study visits. It is common to observe seemingly perfect adherence reported in paper diaries, almost as if individuals were robots. If scrutinized appropriately, such perfect data would be flagged as fraudulent in an RBQM. However, paper diaries are typically used to document adherence checkboxes at the site but are not electronically entered into the study database.

Electronic diaries aim to overcome these issues by timestamping the records. In the case of electronic diaries, an upward bias is typically introduced through reminder functions in the system. Patients are reminded at specific times to take their medication, and later the system prompts them to indicate whether or not they have taken it. If they state that they did not, they are asked to provide a reason. This creates an effective nudging system to generate a perfect adherence record, even though the action of recording the event is often disconnected from the actual medication intake, leading to exaggerated levels of adherence. To overcome this problem, it is possible to require the patient to capture a video of each intake to prove ingestion. This approach has, however, shown low accuracy and poor acceptability by patients because of the additional burden [11].

In summary, patient self-reporting of medication adherence is prone to biases and variability due to factors such as desirability bias, questionnaire choice and administration method, recall period covered, hurriedly filled diaries, and the introduction of reminder functions in electronic diaries that can easily conceal the uncomfortable truth about medication non-adherence.

1.4 Bioanalytical Methods: Used in 4.1% of Trials

In theory, using drug concentration in body fluids (such as blood, plasma, or urine) as a direct measure of adherence seems promising. This method relies on the assumption that the presence of the drug indicates its ingestion and, therefore, adherence to the prescribed medication regimen. In this sense, it is considered a reliable approach to assess adherence.

The results are not influenced by patient recall or reporting biases, as they directly reflect the presence or absence of the drug in the body.

Although drug concentration in body fluids is considered a reliable measure of adherence, its precision is compromised by inherent limitations as it does not account for various factors such as individual variations in drug metabolism, absorption, distribution, elimination, and interactions with other medications or substances that can influence drug concentrations. To facilitate the interpretation of concentration and reduce its variability, sampling is typically carried out at trough, i.e. just before the next scheduled dose.

Due to the limited frequency of drug concentration measurements, the validity of this measure is hindered by a phenomenon known as “white-coat adherence.” This refers to a temporary increase in adherence observed a few days before a scheduled visit [12]. In drug studies, this effect is further reinforced by a common practice of making a phone call prior to the visit, reminding patients to strictly adhere to the medication schedule.

More advanced biosensing technologies [13] that allow for continuous drug monitoring offer a transformative approach to drug exposure assessment, enabling real-time tracking of therapeutic drug concentrations. Wearable and in vivo sensors hold promise for automated, non-invasive monitoring, revolutionizing treatment optimization. However, challenges in sensor integration, clinical validation, and data privacy remain critical considerations for widespread adoption.

When comparing groups in randomized controlled trials, bioanalytical methods have however a limitation tied to typical different pharmacometric characteristics of the drugs, potentially introducing a systematic bias in adherence assessment as a behavior. This issue becomes more critical when a placebo group is involved, as no drug exposure is available to assess adherence behavior in this context. This limitation precludes appropriate causal inference analysis that require an unbiased measure of adherence behavior in all randomized groups [14].

To summarize, bioanalytical methods, although direct and reliable are currently sparse, limiting the reliability of this adherence measure to a few days preceding the sampling. Furthermore, they are subject to large variability, affected by white-coat adherence, and thus limited in their ability to differentiate between minor and more significant deviations in medication adherence.

1.5 Electronic Monitoring: Used in 2.7% of Trials

The concept, pioneered by the MEMS® Cap, involves integrating a microchip into pharmaceutical packages commonly used in clinical trials (such as pill bottles or blisters) and has been expanded to include injectables, inhalers, cream tubes, and eye drop containers. The chip automatically timestamps each action required to access or administer the medication, providing real-time tracking of dosing histories. It should be noted that removing the drug from its package does not necessarily indicate ingestion. Studies have demonstrated that electronically recorded dosing histories align with bioanalytical measures in 97% of cases, while providing rich and continuous assessment of drug exposure in-between visits [15].

For solid oral forms, some companies have moved the sensor from the package to the pill itself (“smart pill”) with the intention of proving ingestion. A microcircuit integrated into an ingestible drug is activated by gastric acid to generate a weak radio signal that contains information on the drug’s identity and the timing of ingestion. This signal is detected, amplified, and retransmitted to a more distant source via a signal-detection skin patch or necklace worn by the patient. While this technique of proving ingestion has strong value in some settings (e.g., Phase I studies), it is tough to deploy at scale due to drug stability, patient intrusiveness, and logistics [16].

Incorporating electronic monitoring into pharmaceutical packages for drug trials involves several important considerations that require sound preparation. First and most important the logistic preparation for electronic monitoring entails selecting and deploying the appropriate devices for data collection. The utilization of electronic monitoring packages, such as smart pill bottles or blisters, generates substantial and intricate data related to medication adherence. This involves not only data storage to handle the extensive volume of data generated but also an appropriate processing to ensure its relevance and usefulness for various stakeholders involved in the trials. Electronically compiled adherence data can then be used to support several objectives:

For patients, the processed data can be utilized to provide feedback on their adherence behavior, fostering engagement and motivation in the study.

Investigators and study monitors can leverage the processed data to monitor patient adherence patterns, identify potential issues, and make informed decisions regarding patient care.

Sponsors, who bear the responsibility for ensuring the success of the trial, can utilize the processed data to evaluate the impact of adherence on study outcomes and adjust strategies if needed.

In summary, the incorporation of electronic monitoring in pharmaceutical packages is a passive method that compiles the adherence data automatically without burdening patients. Subsequent processing and storage of the data are vital steps to enhance the accuracy, reliability, and relevance of data, contributing to the success and integrity of the drug trials.