Impact of medication nonadherence and drug-drug interaction testing on the management of primary care patients with polypharmacy: a randomized controlled trial

Abstract

Background

Clinical management of patients with chronic cardiometabolic disease is complicated by polypharmacy. Consequently, when patients clinically deteriorate, physicians are challenged to distinguish both medication nonadherence and drug-drug interactions (DDI) from chronic disease progression.

Methods

In this randomized controlled trial, we enrolled U.S. board-certified Primary Care Physicians (PCPs) serving patients with cardiometabolic disease. PCPs were randomized and managed their patients with (intervention), or without (control), a novel chronic disease management test (CDMT) that can detect medication nonadherence and DDIs. Patients’ medical records were abstracted at baseline and 3-month follow-up. Primary outcomes were the CDMT’s impact on both the PCPs’ detection of medication nonadherence and DDI, and the frequency of performing medication nonadherence- and DDI-related clinical actions. Secondary outcomes examined the types of clinical actions performed. Primary and secondary outcomes were analyzed by logistic regression using single variable and clustered multivariable modeling to adjust for similarities in patient characteristics, by PCP practice.

Results

Sixteen intervention and 20 control PCPs shared de-identified records for 126 and 207 patients, respectively. There were no significant demographic differences between the two study arms, among PCPs or patients. At baseline, there was no significant difference between the intervention and control PCPs in the percentage of clinical actions performed for medication nonadherence (P = 0.98) and DDI (P = 0.41). At 3-month follow-up (after CDMT), 69.1% of intervention compared to 20.3% of control patients with medication nonadherence had a related clinical action performed (P < 0.001). Regarding DDI, 37.3% of intervention compared to 0.5% of control patients had a relevant clinical action performed in follow-up (P < 0.001). Across the range of medication nonadherence- and DDI-related actions, the intervention compared to the control PCPs were more likely to adjust the medication regimen (24.1% vs. 9.5%) and document medication nonadherence in the patient chart (31.0% vs. 14.3%) at follow-up (P = 0.04).

Conclusions

Although intervention and control PCPs similarly detected and acted upon medication nonadherence and DDI at baseline, intervention PCPs’ detection increased significantly after using the CDMT. Also, the clinical actions performed with CDMT support were more clinically rigorous. These outcomes support the clinical utility of CDMT in the management of symptomatic patients with cardiometabolic disease and polypharmacy.

Trial registration

https://clinicaltrials.gov/study/NCT05910684.

Background

Clinical management of patients with chronic cardiometabolic disease is complicated by polypharmacy, defined as the concomitant administration of five or more medications [1, 2]. The prevalence of polypharmacy among adults in the United States (U.S.) is highest among those with heart disease and has risen dramatically from 40.6% (95% confidence interval [CI], 34.5–46.7) between 1999 and 2000 to 61.7% (95% CI, 55.2–68.2) between 2017 and 2018 [3]. Consequently, when clinicians are confronted with diagnosing the etiology of a patient’s clinical deterioration, they are challenged to distinguish polypharmacy-related factors, such as medication nonadherence and drug-drug interactions (DDI), from the natural progression of the underlying chronic diseases.

Medication nonadherence rates in a recent meta-analysis ranged from 7.0 to 83.5% [4], with the highest rates among adults taking medications for cardiovascular disease and diabetes [5, 6]. This systematic review of 178 studies [4] revealed that medication nonadherence was detected better, albeit incompletely, by self-report (76.5%) than by reliance on pharmacy data (69.4%) or electronic monitoring (44.1%), further evidence that the detection of medication nonadherence remains a diagnostic challenge. For example, measurements of drugs or drug metabolites in body fluid (most commonly blood or urine) and the direct observation of drug ingestion accurately detect systemic drug presence [7]. However, these methods may be invasive or impractical for patients with multiple comorbidities and/or polypharmacy [7]. Indirect detection methods, such as monitoring indicators in pharmacy records and other electronic devices that make it more difficult for patients to falsify adherence, still present limitations as patients may fill prescriptions or engage with the devices and still not ingest the medications. Consequently, enhanced efforts to obtain accurate data are needed because medication nonadherence has been shown to not only impact quality and length of life, but also accounts for approximately 50% of clinical treatment failures, 125,000 deaths, and as many as 25% of annual U.S. hospitalizations [8].

The risk of DDIs also increases with polypharmacy [2]. Adverse drug reactions (ADRs) have been estimated to cost the U.S. healthcare system $30.1 billion annually [9], with recent reports estimating that approximately 18% of ADRs are due to DDIs [10, 11]. Current approaches to detecting DDIs include electronic databases [12] and expert consensus evidence-based decision support [13]. Although software programs are highly accurate in the evaluation of potentially harmful DDIs [14], their utility has been limited in clinical practice for various reasons, including an inability to determine the clinical appropriateness of the potentially harmful drug combinations [15]. To this end, an expert consensus panel developed a transparent and systematic evidence-based process for evaluating DDIs to support clinical decision-making [13]. Despite being successful in several aspects, they determined that more work is needed to more objectively identify clinical relevance of the DDI, such as severity grading; factors that predict patient harm; and a quality of evidence rating system [13]. Consequently, additional research is warranted to develop a provider- and patient-friendly approach to accurately and distinguish medication nonadherence and DDI that addresses these limitations.

In 2022, we conducted a randomized controlled study utilizing QURE Healthcare’s scientifically validated virtual patients [16] to evaluate the ability of primary care physicians (PCPs) to identify medication nonadherence, DDI, and disease progression. We assessed the impact of results from a novel, saliva-based chronic disease management test (CDMT) on the PCPs’ clinical decision-making, reflecting clinical utility [16]. In brief, we found that CDMT results improved the detection of medication nonadherence from 1.3 to 39% (P < 0.001) and the detection of DDIs from 6.5 to 57.1% (P < 0.001) [16]. In order to corroborate our virtual patient evidence, we conducted this randomized controlled study in actual patients with chronic cardiometabolic disease to quantify the ability of PCPs to detect and adjust clinical management due to medication nonadherence and DDI and to examine their clinical management actions.

Methods

Between July 2022 and December 2023, we enrolled 279 board-certified PCPs in a randomized controlled study to examine the clinical utility of the CDMT on PCPs’ ability to detect and treat medication nonadherence and DDI. PCPs randomized to the intervention arm (Fig. 1) were provided the opportunity to use the CDMT on patients in their practice who satisfied eligibility criteria (detailed below in “Patient Inclusion Criteria”) then, as appropriate, utilize the results to guide clinical decisions and actions. Both intervention and control PCPs provided patient data, at baseline and at least 3 months after the baseline visit, through de-identified medical record abstraction. PCPs randomized to the control arm used the same patient inclusion criteria, but they were not provided with the CDMT.

Fig. 1

Physician enrollment

Ethics

This study was conducted in accordance with ethical standards, approved by the Advarra Institutional Review Board, Columbia, MD, USA, and listed on clinicaltrals.gov (NCT05910684, 20/06/2023). Voluntary, informed consent was obtained from all participants.

Chronic disease management test

Aegis Sciences Corporation [17] developed a chronic disease management test (CDMT, MedProtect CDM™) that detects medication nonadherence and DDI through saliva sample analysis [16]. Of particular importance is the breadth of cardiometabolic conditions covered when using CDMT, including coronary artery disease, heart failure, hypertension, and diabetes. The CDMT detects > 150 drug metabolites in the saliva specimen, reconciles them with the current medications list provided by the PCP via the requisition form, and generates respective medication nonadherence or DDI results. Drug classes include angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, beta blockers, biguanides, diuretics, statins, sulfonylureas, and antithrombotic agents. Moreover, the CDMT reports DDIs based on objective detection of ingested substances with known ability to contribute to interactions and classifies the severity of any DDIs detected.

PCP recruitment and enrollment

We recruited PCPs from our virtual patient study [16], as well as additional PCPs who met the study inclusion criteria from a national sample of PCPs in the USA (Fig. 1). PCPs from the virtual patient study maintained their original randomization assignment (intervention or control) and newly recruited PCPs were randomized by a balanced randomized block size allocation.(Fig. 1) [18]. All PCPs received a fair market value honorarium for their participation.

Patient inclusion criteria

All PCPs were asked to identify 6–14 patients in their practice who satisfied the following inclusion criteria: (1) receiving pharmacological treatment for at least two cardiometabolic conditions: type 2 diabetes mellitus, heart failure, hypertension, and/or coronary artery disease; (2) being prescribed five or more drugs; (3) presenting with new and/or undiagnosed clinical symptoms; (4) being at least 21 years old, and (5) evoking PCP concern regarding medication nonadherence, DDI, or the use of non-prescribed substances, e.g., over-the-counter supplements or medications. The patients who were determined to be eligible based on these criteria were detailed separately for the intervention PCPs (Fig. 2) and control PCPs (Fig. 3).

Fig. 2

Intervention patient data

Fig. 3

Control patient data

Data collection and intervention

The primary data for this study were obtained from de-identified, abstracted patient medical records from both intervention and control PCPs. Medical record data were captured at baseline and at a follow-up occurring at least 3 months after the baseline data collection for each patient. The medical records were de-identified by the PCPs in compliance with Sect. 164.514 of the U.S. Department of Health and Human Services’ HIPAA Privacy Rule and were linked to CDMT results through a unique identification number. The study team developed an abstraction sheet with category determinations informed by national guidelines and evidence-based recommendations on medication nonadherence and DDIs. A third-party abstracter used this sheet to extract data related to clinical actions performed for analysis. Discussions were held with primary care physician experts on the study team in case of medical record disagreement.

Intervention PCPs obtained patient consent from eligible patients for the CDMT according to their standard clinical practice. These PCPs obtained an in-office saliva sample and submitted the samples directly to the Aegis Sciences Corporation laboratory for complimentary analysis. CDMT results were returned to the PCPs.

Primary and secondary outcomes

The primary outcomes were to determine if the CDMT results [1] improved PCPs’ detection of medication nonadherence and/or a DDI and [2] led to subsequent medication nonadherence- or DDI-related clinical actions. The secondary outcome was to examine the ways in which intervention PCPs modified their clinical practice in the setting of CDMT results in order to address medication nonadherence or DDI.

Statistical analyses

Single binary independent variables and categorical variables in intervention and control arms were analyzed using the chi-square test. Primary and secondary outcomes between study arms were analyzed by logistic regression using both a single variable and a clustered multivariable model, in order to adjust for potential patient characteristic similarities within PCP practice. The sample size calculation was based on a chi-square test of the ability of PCPs to detect medication nonadherence and/or DDI among intervention patients compared to control patients. In order to achieve an alpha of 0.05 and a power of 80%, we assumed a baseline incidence of 20% for both primary outcomes and an absolute effect size of 17.5%. Therefore, using standard power calculations, we would be required to enroll 104 patients in each study arm. However, in order to account for potential clustering effects within PCP practices, we increased the sample size by 20% to require a minimum sample size of 125 patients in each arm. All analyses were performed in Stata 18.0 [19].

Results

Physician characteristics

After applying inclusion/exclusion criteria (Fig. 1), a total of 279 PCPs were enrolled and randomized. However, 106 PCPs withdrew from the study for reasons including time constraints (n = 34); institutional or organizational restrictions (n = 27); personal circumstances or changes (n = 12); lack of interest or willingness (n = 18); misunderstanding of study expectations (n = 7); miscellaneous or administrative issues (n = 5); and technical or logistical issues (n = 3). Additionally, 137 PCPs were lost to follow-up (87 intervention and 50 control), resulting in 36 PCPs completing the study (16 intervention and 20 control). Comparing PCP characteristics between the two study arms, we found no significant differences by age, gender, years of experience, or practice demographics (P > 0.05 for all; Table 1).

Table 1 Baseline characteristics of physicians

Patient characteristics

A total of 126 intervention (Fig. 2) and 207 control patients (Fig. 3) were included in the analysis. When comparing patient characteristics between intervention and control study arms, respectively, we found no significant differences by age (mean ± SD, 67.6 ± 12.3 vs. 68.3 ± 12.9 years); sex (male, 57.3 vs. 59.3%); or number of medications at baseline (11.5 ± 6.9 vs. 10.7 ± 6.2; all P > 0.05; Table 2). When examining co-morbidities, we found no significant differences between the intervention and control patients by age, sex, number of medications taken at baseline, or prevalence of diseases. We note that although intervention patients were slightly more likely to have been diagnosed with non-cardiometabolic disease than control patients (71.6% vs. 63.8%) this difference was not significant (P = 0.16).

Table 2 Baseline characteristics of patients

Chronic Disease Management Test (CDMT) results

Of the 126 intervention patients with CDMT results, 42.1% (53/126) received results indicating medication nonadherence alone, 9.5% (12/126) received results indicating DDI alone, and 42.9% (54/126) received results indicating the presence of both. Consequently, only 5.6% (7/126) of the CDMT results were negative for both medication nonadherence and DDI. When examining medication nonadherence, the CDMT detected 3.1 ± 1.8 (mean ± SD) instances of drug metabolites that were not in alignment with the patient’s treatment (e.g., nonadherence to a prescribed medication, detection of a medication not indicated as prescribed), with an average of 2.0 ± 1.5 prescribed but not detected and 1.1 ± 1.4 detected but not prescribed. Regarding DDI, the CDMT detected 2.9 ± 2.6 moderate or severe DDIs per test.

Medication nonadherence and DDI detection

At baseline, the percentages of intervention and control PCPs who identified medication nonadherence or DDI in their patients were inferred from the PCP clinical actions performed (e.g., noted in chart, counseled patient, augmented or discontinued pharmacotherapy or other non-prescription substance use). At 3-month follow-up, after the CDMT results were shared, 107 patients were recognized within medical records as nonadherent to medications (P < 0.001). Regarding DDI, after the CDMT results were shared, 66 patients were identified (P < 0.001).

Of note, nine medication nonadherence-related actions were performed among the 19 intervention patients without CDMT-detected medication nonadherence. Also, two DDI-related actions were performed among the 60 intervention patients without CDMT confirmation of DDI.

Medication nonadherence- and DDI-related action frequency

For medication nonadherence, 18.3% of intervention patients compared to 18.4% of control patients (P = 0.98) had a medication nonadherence-related action performed at baseline, such as counseling, noting nonadherence on chart, adjusting/switching medications causing nonadherence, or offering tools to mitigate nonadherence (Table 3a). At 3-month follow-up, intervention patients received a medication nonadherence-related action 69.1% of the time compared to 20.3% of the time for control patients (P < 0.001), representing a difference-in-difference of 48.9%. Consequently, at 3 months, the intervention arm was more likely to be associated with a change in clinical practice than the control arm (odds ratio, 8.8; 95% confidence interval [CI], 4.1 to 19.0; P < 0.001). A full logistic regression model accounting for provider, patient, and practice characteristics, as well as potential PCP practice clustering effects, revealed an odds ratio of 11.9; 95% CI, 2.8 to 51.4; P < 0.001) (see Additional File 1: Table S1(a) for details.)

Table 3 Clinical actions taken in response to (a) medication nonadherence or (b) drug-drug interactions

Regarding DDI, only 0.8% of intervention patients had a DDI-related action performed at baseline compared to 1.9% of control patients (p = 0.41; Table 3b). These DDI-relevant clinical management actions included counseling, noting new DDI in the patient’s chart, adjusting/switching medications causing DDI, noting new medication allergy in chart, or offering a monitoring plan for potential increased symptoms of DDI. At 3-month follow-up, intervention patients received a DDI-related action 37.3% of the time compared to 0.5% of the time for control patients (p < 0.001), demonstrating a difference-in-difference of 38.0%. Simple logistic regression showed an odds ratio of 301.9 (95% CI, 15.4 to 5904.7). A full logistic regression model using the same variables as those used for the medication nonadherence-related actions exhibited an odds ratio of 372.2 (95% CI, 16.6 to 8351.9). (See Additional File 1: Table S2(b) for details.)

Types of medication nonadherence-related actions

At baseline, there was no significant difference between intervention and control patients in the types of medication nonadherence-related actions that were performed (P = 0.20, Table 4). Nonetheless, the intervention compared to the control PCPs more frequently counseled their patients (91.3% vs. 73.3%), but less frequently adjusted their patients’ medications (0.0% vs. 10.5%) or offered a tool to improve adherence, such as smart reminders, alarm setting, or digital pills (0.0% vs. 5.3%).

Table 4 Types of medication nonadherence and DDI actions taken at baseline and follow-up

After introduction of the CDMT, the intervention compared to the control PCPs had a different distribution of medication nonadherence-related action (Table 4; P = 0.04). Specifically, intervention PCPs were less likely to counsel their patients (77.0 vs. 90.5%) but were more likely to adjust their patients’ medication regimen (24.1% vs. 9.5%) and note medication nonadherence on the patient chart (31.0% vs. 14.3%). None of the PCPs at follow-up offered an auxiliary tool (e.g., smart reminders, alarm setting, digital pills) that might improve medication adherence.

Types of DDI-related actions

At baseline, there were no significant differences between intervention and control patients in the types of DDI-related actions performed (P = 0.12, Table 4). At 3-month follow-up, 63.8% of the intervention PCPs offered counseling; 25.5% adjusted DDI-related medications; 34.0% noted a new DDI on the patient’s chart; 8.5% implemented a monitoring plan; and 6.4% noted a new drug allergy (P = 0.003, Table 4). Only one control patient received a DDI-related action at 3-month follow-up, the notation of a new drug allergy.

MNA- and DDI-related actions in non-positive CDMT results

Among the 19 intervention patients without CDMT-detected medication nonadherence, counseling was performed seven times, and a chart notation was performed two times. Among the 60 control patients without CDMT-detected DDI, a drug allergy was noted on the chart two times.

Discussion

We conducted this randomized controlled study in order to quantify the ability of PCPs to detect medication nonadherence and DDI, and to adjust clinical management accordingly. In addition, we analyzed the types of responses PCPs offered in these two adverse drug event (ADE) scenarios. We developed these study aims because we acknowledged that the highest rates of medication nonadherence have been reported in adults taking medications for cardiovascular disease [5] and diabetes [6]. Current approaches to detect medication nonadherence are limited [7, 20], and there is an identified need for research to develop a provider- and patient-friendly approach to accurately distinguish medication nonadherence and DDI [13]. Additionally, because we demonstrated the clinical utility of CDMT in scientifically validated virtual patients with chronic cardiometabolic disease [16], we set out to corroborate our virtual patient evidence among actual patients with similar clinical characteristics.

We observed no significant difference at baseline between the intervention and control PCPs in their ability to detect and adjust clinical management due to medication nonadherence or DDI. Although the medication nonadherence and DDI baseline data were not validated by CDMT, these data were inferred from the PCPs’ clinical actions data, reflecting their customary approach to ADR management related to medication nonadherence and DDIs. Additionally, we believe that these inferred data are valuable since the literature indicates that the highest rates of medication nonadherence are detected by patient self-report [4]. Nonetheless, these self-reported rates of medication nonadherence are still relatively low and support the need for enhanced provider and patient responsibility in fostering more effective discussions about medication adherence [21]. Our data further confirm the presence of this clinical practice gap in that the control PCPs, who cared for their patients without the use of CDMT, performed fewer clinical actions addressing these two ADRs than did the intervention PCPs.

Our study demonstrated that the CDMT was associated with [1] increasing the rate of detecting medication nonadherence and DDIs, and [2] increasing the number of clinically relevant actions taken to address DDIs and medication nonadherence among intervention versus control PCPs. At baseline, the clinical actions were primarily counseling (91%) and chart documentation (22%); there were no medication adjustments (change dosage or frequency, stop, or switch to an alternative medication) made. At 3-month follow-up, after the CDMT, the clinical actions were distributed among counseling (77%), medication adjustment (24%), and chart documentation (31%). The shift towards more clinically rigorous and impactful actions may reflect a greater degree of confidence regarding the certainty of medication nonadherence status.

When considering the impact of the CDMT on clinical actions among the Intervention PCPs, the clinical actions performed for medication nonadherence increased significantly, from 18.3 to.

69.1% (P < 0.001; Table 3a). Similarly, the clinical actions to address DDI increased significantly, from 0.8 to 37.3% (P < 0.001; Table 3b). These data suggest that the detection of medication nonadherence and DDIs is not being sufficiently recognized or treated in standard primary care. A few providers in this study exhibited a blanket patient education and counseling approach, regardless of patient presentation or results. However, the CDMT’s influence on personalized healthcare decision-making related to polypharmacy remained significant. Notably, the size and directional increase of medication non-adherence/DDI action even after accounting for other significant variables and confounders would suggest that the CDMT can be used across a variety of provider and practice settings.

We also observed limited DDI detection and clinical management actions performed at baseline among the intervention PCPs, providing further evidence regarding the diagnostic challenge posed by DDI assessment in clinical practice. While practicing PCPs may have access to electronic DDI databases, [12, 14, 15], challenges in technical or time-dependent barriers likely still prevent more systematic use. Prescribers are also increasingly reporting “alert-fatigue” due to the increased frequency, limited relevance, and perceived discrepancies of computerized or integrated drug-drug interaction alert systems [22,23,24,25].

Notably, after the CDMT results were provided, intervention PCPs engaged in a number of DDI- and medication nonadherence-related actions, primarily counseling (64%); chart documentation (40%); medication adjustments (26%); and implementation of a DDI monitoring plan (8.5%; P = 0.003, Table 4). These data suggest that the provision of an uncomplicated patient- and provider-friendly tool may support improved clinical management.

Interestingly, although a small number of control PCPs offered patients tools to improve medication adherence (e.g., smart reminders, alarm setting, digital pills) at baseline, no control PCPs elected to utilize these tools at 3-month follow-up. This preference may reflect a prior unsatisfactory experience with these devices, a lack of experience with them, or perhaps they simply forgot. While further investigation may be conducted to understand this observation, the unfortunate current reality is that there are limited resources utilized by PCPs to support medication nonadherence detection.

Limitations of our study include the size of the patient population. Although the patient demographics were similar between intervention and control, we could not ascertain that the patients in this study fully represented all patients with cardiometabolic multimorbidity and polypharmacy due to a lack of available published data. While it might be possible that subgroups not represented in this study could exist for whom the results of this study would not be valid, we find such a case to be unlikely given the general nature of the test and the eligibility requirements of the study and is, in any case, outside the scope of this study.

Further, it is possible that the population of study PCPs may not be generalizable to the general PCP population. For example, our intervention PCP population was comprised of 94% males but 2021 data showed that 57.7% of Family Medicine and 60.8% of Internal Medicine physicians practicing in the U.S. identified as male [26]. Additionally, we did not include patient race/ethnicity in our analyses and there are data to suggest that these demographic factors play a role in medication adherence [27] as well as DDI [28].

Of note, there was also an appreciable number of PCPs lost to follow-up. These PCPs may have been unable to participate for reasons similar to those expressed by PCPs who withdrew, or perhaps as a reflection of the known challenges associated with ADR monitoring and reporting [29]. More research is needed to determine the impact of attrition bias on these findings. PCPs as participants, unlike patients, are more accessible and are known to be lost to follow-up in research for reasons that are less likely to systematically affect the study’s conclusions, such as time or resource constraints [30]. Additionally, the primary outcomes of this study are clearly defined, objectively measured behaviors that are less susceptible to recall bias or subjective interpretation, further mitigating the potential impact of missing data.

While the adverse clinical impacts of medication nonadherence and DDI are well documented [31, 32], we were unable to determine the long-term clinical efficacy of the CDMT in this study. Analyzing whether or not the clinical actions performed by the PCPs translated to improved long-term cardiometabolic clinical outcomes for patients, such as reductions in biomarkers or disease progression, would require an observation period beyond the scope of this study. Additionally, because the CDMT was provided to PCPs free of charge in this study, we did not perform a cost–benefit analysis or incremental cost-effectiveness ratio (calculated as [cost in intervention arm − cost in control arm]/[effect in intervention arm − effect in control arm]) [33]. Finally, although the CDMT provides data on severity of DDI supported by their use of First Databank MedKnowledge Drug-Drug Interaction Module™ [34], we did not analyze these data in this study population.

Conclusions

The optimal clinical management of patients with chronic cardiometabolic disease and polypharmacy requires awareness and active clinical management of medication nonadherence and DDIs. Because of the limited and variable efficacy of patient self-reporting, monitoring systems, and currently available management tools, the use of a provider- and patient-friendly diagnostic tool is necessary. Our study provides evidence for the clinical utility of a noninvasive test that can enable physicians to better detect medication nonadherence and DDI, and to subsequently perform clinical actions that will promote improved clinical outcomes. Further investigation is warranted to determine its long-term clinical efficacy and incremental cost-effectiveness ratio.