Comparing medication adherence using a smartphone application and electronic monitoring among patients with acute coronary syndrome

Abstract

Aim:

The purpose of this study was to determine the extent of agreement between adherence measures obtained using two technological interventions, electronic monitoring (EM) and a smartphone application (App).

Background:

Clinicians, patients, and researchers depend on valid measurements of medication adherence to inform the delivery of preemptive care when needed. Technology is routinely used for monitoring medication adherence in both clinical practice and research, yet there is a dearth of research comparing novel App based approaches to traditional approaches used for assessing medication adherence.

Methods:

Adherence rates were captured on both the EM and the App for 3697 daily observations from 44 participants with acute coronary syndrome over 90 days immediately following discharge from acute care. For EM, adherence was measured using EM equipped pill bottles. For the App, adherence was measured by having participants upload daily photos to the App prior to taking their daily aspirin. Agreement was assessed using a Bland-Altman analysis.

Results:

The mean adherence rate was higher on the App, 92%, than the EM, 78% (p < 0.001). The mean difference in adherence rates between these methods was 14% (95% Confidence Interval: − 23%, − 5%).

Conclusions:

These findings illustrate a lack of agreement between technological interventions used for measuring adherence in cardiovascular patient populations, with higher adherence rates observed with the App compared to EM. These findings are salient given the increased reliance on telehealth due to the ongoing COVID-19 pandemic.

1. Introduction

Medication adherence is important to both patients living with chronic conditions and the clinicians who prescribe medications because nonadherence is tied to poor health outcomes and increased healthcare costs (Du et al., 2017; Marcum et al., 2017; Packard & Hilleman, 2016; Wenger et al., 2017). Medication adherence for cardiovascular disease is associated with 39% lower odds of hospitalization or myocardial infarction (Du et al., 2017). Yet, a large meta-analysis found that approximately one-third of patients were not adherent to their prescribed drug regimen after a cardiac event, regardless of the type of cardiovascular drug prescribed (Naderi et al., 2012). Preventable healthcare costs due to nonadherence surpass $100 billion in America, annually (Iuga & McGuire, 2014). Unlike many factors associated with increased costs, medication nonadherence is among the few modifiable factors that can be addressed by clinicians and their patients interested in implementing low-cost telehealth interventions (Treskes et al., 2018). Therefore, ongoing monitoring of adherence using technology can be vital to informing the delivery of preemptive care interventions capable of mitigating negative health outcomes (Salisbury et al., 2016). The increased reliance on telehealth during the Coronavirus Disease 2019 (COVID-19) pandemic highlights the urgent need for effective monitoring of medication adherence (Dosaj et al., 2021). In this study we address a specific telehealth approach, Mobile Health (mHealth) (e.g., smartphone and tablet apps) interventions. mHealth allows for bidirectional communication in the encounter and often requires the use of an App (Iyengar, 2020).

Investigators have shown the utility of reminder systems, delivered using technology, on improving medication adherence (McFarland et al., 2021; Thakkar et al., 2016). Less attention, however, has focused on which approaches are most appropriate for measuring adherence in community dwelling patients with complex chronic conditions. Electronic monitoring (EM) devices (e.g., electronic pill bottle caps) are often used by telehealth clinicians for assessing medication adherence in cardiovascular patients (Gandapur et al., 2016). EM devices detect and store the dates and times of the opening and closure of a drug container. They are considered a valid and robust technological approach to measuring when the prescribed dose of a drug was taken (Gandapur et al., 2016).

Authors of scoping reviews have reported that EM adherence data is correlated with other methods, but wide variability between methods has been noted (El Alili et al., 2016; Lam & Fresco, 2015; Williams et al., 2013). Compared to EM, medication adherence rates have been found to differ by 17% using self-report, 8% using pill count, and 6% using ratings by caregivers or clinicians (El Alili et al., 2016). Such discrepancies suggest a need for innovative approaches to measure medication adherence. Current advances in mobile technology suggest that smartphone applications (Apps) might be useful for this purpose. Apps are helpful in that they do not require the use of costly, specialized EM devices nor the use of a pill bottle from which to dispense medications. Patients who use alternative dispensing methods such as weekly pill containers risk having their adherence underestimated because they do not register pill bottle openings when taking their medications (El Alili et al., 2016).

Approximately 15.5 million people in the United States have a diagnosis of acute coronary syndromes (ACS) (Mozaffarian et al., 2016). In ACS patients, poor medication adherence has been found to be associated with higher morbidity and mortality (Du et al., 2017). These negative health outcomes underscore the need to accurately measure daily medication adherence in order to identify those at risk of being nonadherent so that clinicians can intervene. This has greater importance today as in-person appointments to ensure medication adherence may be limited by the COVID-19 pandemic. The purpose of this study was to determine the extent of agreement between adherence measures obtained using two mHealth approaches, EM and an App.

2. Methods

This secondary analysis examined data from participants in the intervention arm of a pilot randomized control trial (RCT) (Riegel et al., 2020). The aim of the original RCT was to compare the efficacy of a mobile App rooted in behavioral economics and created by Wellth Inc., a health-based mobile application development company, on medication adherence. The participants in this analysis were drawn exclusively from the intervention arm of the original study. Medication adherence was captured using both EM and the App that had been downloaded by each participant to their personal device.

2.1. Sample and setting

Following approval from the Institutional Review Board, the parent study enrolled 130 participants from two acute care urban academic medical centers in the United States with a total of 62 participants randomized into the intervention arm between February 1, 2017 to August 16, 2018 (Riegel et al., 2020). Eligibility was determined by age (21 years and older), admission diagnosis (myocardial infarction or unstable angina), having once-per-day aspirin prescribed at discharge, and confirmation of independence in administration of their medications. Participants were required to have a smartphone with a sufficient data plan or home Wi-Fi to use the App and avoid any excess charges, sufficient vision, and basic English speaking and comprehension skills. Potential participants were excluded if they had a cardiac event occurred during hospitalization, were discharged to a location other than home, were unable to use their smartphone or complete a successful medication check-in with the App, or had a life expectancy of less than 6 months.

Participants in the intervention arm of the original study downloaded the App to their smartphones at time of discharge with the assistance of research staff. The App sent daily notifications to participants’ smartphones to remind them to take their aspirin at a predetermined time they selected. Participants had a 12-h window in which to check-in (e.g. take their medication) each day. If participants did not take their pill at the predetermined time, the App would send additional reminders during the 12-h period. A check-in was considered completed if the participant took a photo of the pill in the hand via the App’s camera feature prior to ingesting it. Therefore, App adherence refers to participants taking a picture of the pill and uploading it. Participants received financial incentives both for being in the study and for uploading daily pictures to the App. Following loss aversion principles, participants utilizing the App received $50 at the beginning of every month and $2 was deducted for every day a photo was not taken. In this manner, intervention participants could earn up to an additional $150 over the course of 90 days (Kahneman et al., 1991). This design was used to penalize for not taking medications instead of rewarding for taking medications, which could promote over-medication.

All participants in this analysis were provided with an EM device, either CleverCaps® (Compliance Meds Technologies, Miami, FL) or Medication Event Monitoring System (MEMS®) (Aardex Ltd., Zug, Switzerland), that was pre-filled with a 90-day supply of aspirin at time of discharge. Participants were instructed to take their aspirin from only the EM device during the study period. As such, EM adherence refers to participants opening their medication bottle to withdraw medication each day. Of the 62 participants randomized to the intervention, 18 were excluded from this analysis due to technical failure of the CleverCaps® device, failure to return the cap for data download, or study withdrawal. This resulted in a sample of 44 participants and 3697 daily medication adherence data points captured using EM and the App. Unlike the parent study, which compared the effects of the App compared to usual care (e. g. EM), this analysis uniquely examined data from participants in the intervention arm who used both the App and EM.

2.2. Measures

Medication adherence was measured daily over a 90-day period and reported as a percentage (%) of drug taken for both the EM device and the App. If a rehospitalization occurred, those days when participants were not expected to take their aspirin independently were removed prior to analysis. The main outcome of interest, mean difference in medication adherence, was computed based on the difference in the proportion of medication taken as measured using EM and the App. Additionally, the proportion of days a participant used either the EM device or the App exclusively was computed. A central assumption of research using EM devices is that each time a cap is opened, a dose is ingested. This assumption can be violated if participants remove more than one dose at a time when the cap is opened or do not store their medication in the EM device and, therefore, do not open it daily. Demographic surveys and the validated Self Care of Coronary Heart Disease Inventory (SC-CHDI) were completed by all participants at the time of enrollment (Vaughan Dickson et al., 2017). Specifically, one of the SC-CHDI items asks participants how routinely they use a system to help remember medicines, such as using a pill box or reminders; this item was used in analysis. Responses are recorded on a scale from 1, “Never or rarely,” to 5, “Always or daily.” To ascertain if responses to this SC-CHDI item were associated with adherence rates, numeric responses were dichotomized as “Never or rarely” or “Sometimes to always or daily.” Participants’ electronic health records were reviewed to collect clinical information (e.g., prior MI, prior coronary artery bypass grafting or percutaneous intervention, additional risk factors) and rehospitalization events. All participants were dispensed aspirin in an EM bottle at the time of discharge from the hospital’s pharmacy.

2.3. Statistical analysis

Sample characteristics were examined using descriptive statistics (e.g. mean, standard deviation [±], median, interquartile range). Daily adherence observations on the EM device and App were summarized into a total percentage (%) representing the proportion of days when medication was taken (e.g. adherence rate). Nonparametric approaches (e.g. Mann Whitney) compared differences in the distribution of adherence rates captured on the EM compared to the App. Differences in adherence rates between the two methods, EM and App, were plotted against the mean adherence of the two methods, per subject, in a Bland- Altman Plot (Altman & Bland, 1983). Confidence intervals and levels of agreement were tabulated to assess the degree of difference in adherence rates captured on the two methods. Nonparametric approaches examined the association between SC-CHDI responses and differences in adherence rates (e.g. Mann Whitney). All analyses were conducted in R (Vienna, Austria).

3. Results

Of the 44 participants, 57% were male, 39% identified as Black, 52% identified as White, and the average age was 59 (±10.5) years. Half of the participants were married (50%), most lived with others (80%), and the majority of participants, 68%, held bachelor’s degrees or higher degrees. At the time of enrollment, 52% of participants stated that they “Never or rarely” used a system to help them remember to take their medications (Supplementary Table 1).

Participants had an average of 82 (±21) daily observations over the 90-day study period. Despite medication adherence being relatively high among the study participants, there was a statistically significant difference between EM and App measured adherence. Average adherence rate was 78% (±27%) measured using EM and 92% (±17%) using the App (p = 0.001). The mean difference in adherence rates (EM–App) between methods was −14% (95% CI: −23%, −5%) (figure below). The negative difference indicates that participants’ adherence rates measured on the EM device were lower compared to the rates measured on the App. Jointly, the higher average adherence observed on the App and the difference observed on the Bland-Altman Plot indicates more frequent use of the App than the EM.

figure.

Participants who reported using a system to help remember medicines “Sometimes to Always or daily” had an average adherence rate of 83% compared to an average of 55% among those participants who “Never or rarely” used a system to help remember their medicines. However, this 28% difference in adherence rates was not statistically significant. Further, nonparametric approaches indicated no significant differences between participants based on the response to the SC-CHDI item. Additional sensitivity analyses were conducted to see if there was an association between time since the study began and the sample’s average adherence using a generalized linear model, however no statistically significant association was observed.

4. Discussion

The purpose of this study was to determine the extent of agreement between adherence measures obtained by EM device and an App. We compared adherence in a sample of individuals in whom both methods were prescribed as part of the monitoring for an RCT. The two methods did not agree, particularly among participants with intermediate levels of measured adherence. Although most participants opened the EM device and then used the App, as expected, the significantly higher mean adherence observed on the App highlights how participants more frequently and regularly favored the picture taking App compared to the EM. It is also possible that participants chose to take aspirin obtained over the counter that was not kept in pharmacy dispensed EM equipped bottles. Future work should examine this hypothesis. Our finding suggests that these two measures are not interchangeable, and that the App may be a more sensitive measure of medication adherence.

Though telehealth and mHealth continue to gain traction amidst the COVID-19 pandemic, these two interventions are fundamentally different. Few prior studies of medication adherence in cardiovascular patients have evaluated mobile phone-based approaches; we identified only two other studies using photo-taking approaches to measure medications adherence (Galloway et al., 2011; Holstad et al., 2019). Galloway and colleagues compared taking photographs with a cellular phone to a capsule count and an EM event monitoring system and found that taking photographs underestimated adherence, while the EM event monitoring system overestimated adherence (Galloway et al., 2011). Holstad and colleagues reported characteristics of a self-performed picture pill count for a clinical trial of people with HIV living in rural Georgia and found that this approach was a reliable and valid method of measuring adherence (Holstad et al., 2019). Compared to these studies, we found that using a photo-taking approach captured higher rates of medication adherence relative to either an EM device or a self-report questionnaire (Galloway et al., 2011; Holstad et al., 2019). We found a difference in adherence rates between the EM device and the App to be similar to, if not higher than, the differences reported in meta-analyses comparing self-report measures to EM, which range from 9.5% to 17% (El Alili et al., 2016; Shi et al., 2010). Based on the findings of these meta-analyses, the 14% difference we observed suggests that the App is an accurate method for measuring medication adherence and potentially better than EM.

These hypothesis generating results offer valuable insight for prescribing clinicians. Notably, these findings highlight the need to work with patients to consider various approaches by which to measure medication adherence, specifically in cardiovascular populations. For example, recent systematic reviews have found that behavioral interventions may improve the proportions of patients who satisfactorily adhere to their prescriptions compared to either usual care or educational only interventions (Cross et al., 2016). In other patient populations, investigators have reported wide variation in adherence rates between EM and traditional pill count approaches, with up to 25% of participants being falsely deemed nonadherent when adherence was measured using EM devices alone (Van Onzenoort et al., 2010). In this study, much of the disagreement in adherence rates between the EM device and the App seems to have been due to patient preference, favoring the App over the EM device. This suggests patient preference is an important element to consider when choosing a method to measure adherence. Therefore, EM may not be an adequate approach for monitoring adherence for all patients. App-based adherence monitoring offers one novel approach by which to assess the long-term impact of interventions aimed at improving medication adherence.

The study design has inherent limitations. The sample size available for this secondary analysis limited our ability to identify characteristics associated with disagreement between measures. Future research with larger samples is needed to examine trends in utilization between EM devices and apps based on patient characteristics. Given the difference observed in EM recordings and pictures taken of medication using the App it appears that the EM equipped bottles were not consistently utilized to store medication before photos were taken. While EM remains the gold standard for assessing medication adherence, the assumption in studies utilizing EM is that a single dose of medication is dispensed each time the bottle is open, which cannot always be guaranteed (Lieb et al., 2020). Not only could participants have taken out multiple doses at once but also they may have stored medication in various other locations or containers aside from the EM equipped bottle. We did not directly ask about alternative dispensing methods or participant-reported reasons for possibly using one method over another. Therefore, it is not possible to know if the EM equipped bottle was exclusively used to dispense the medication photographed.

Despite EM equipped pill bottles being the standard in medication adherence research, results from the present study support the need to examine the utility of other EM equipped containers and to compare adherence rates with the App in future studies. For example, recent research indicates the substantial number of community dwelling patients who are interested in technology enabled pillboxes that can be leveraged monitor medication adherence (Choi, 2019). Another limitation is that participants were not blinded to the intervention. It is possible that they changed their behavior because they were being monitored closely. However, the longitudinal design of this study may have accurately captured routine behavior given that the Hawthorne Effect decreases with time (McCarney et al., 2007). The financial incentive also may have led some users to favor using the App more than the EM device. Yet, adherence rates captured on both the EM and the App are similar to, if not superior to, rates observed in other clinical trials with patients diagnosed with acute coronary syndrome and other chronic conditions (Conn & Ruppar, 2017; Ho et al., 2014). Future research, without financial incentives, is needed to further ascertain the difference in adherence rates between EM and Apps.

Another limitation was that there were issues with EM device failures that contributed to missing data and reduced our sample size. The nature of this investigation required that participants’ have adherence data on both the EM and App and resulted in a smaller sample qualifying for inclusion in this analysis. Although we did not detect a difference in agreement between measures among those reporting that they used a system to remind them to take their medications, we did not directly assess the use of alternative methods of pill delivery such as a pill box or aspirin kept in medication bottles that were not equipped with EM. Thus, we cannot be sure about how many patients were using such methods. Given that patient preferences vary for storing and managing medications, future studies comparing telehealth approaches to measure medication adherence will need to use a patient-centered approach and incorporate multiple methods by which to capture adherence (e.g. EM equipped pill boxes). Despite these constraints, this analysis identified that participants frequently used both the App and EM device and that the difference in adherence rates was due to a lack of EM use in a subset of the sample.

5. Conclusions

Using a smartphone App to measure adherence was associated with differences in adherence estimates when compared with the use of an EM device. In particular, a subset of individuals did not use the EM device as frequently as the App. Future work is needed to identify patient characteristics associated with differences in adherence measures across methods. If our speculation that patients might prefer App-based measurement approaches is true, the ubiquitous presence of mobile phones could make adherence measurement more efficient and cost- effective. Medication adherence monitoring will remain a relevant concern for clinicians and patients as the COVID-19 pandemic continues to impede routine clinical care across health care settings. Data collected using multiple methods of medication adherence monitoring will provide applicable and valuable clinical insights that can optimize care planning for clinicians offering telehealth care services. Optimizing telehealth is a growing priority as more patients seek care using telehealth platforms during the ongoing COVID-19 pandemic. It is essential that telehealth clinicians be able to meet the needs of ACS patients in an ever-changing society.