Demographic, geographic, medical, and use data came from the corporate data warehouse (CDW). Race and ethnicity are determined based on self report. During this study’s period, when a patient was enrolled in RTM, their Department of Veterans Affairs (VA) provider placed an order for the device through the Prosthetics Department. As the company making the devices only had 1 product, we were able to determine enrollment in RTM based on the Data Universal Numbering System number (DUN & Bradstreet). The company provided information on the average number of times per week that patients used the mat each calendar month from January 2019 until October 2021.

We included individuals who were enrolled in RTM in VA between January 2019 and June 2021 (as identified in the CDW) and for whom we were also able to obtain SmartMat use data. Of those enrolled in January 2019 or later (n=1675), we were able to link 1641 individuals to other data in the CDW. We excluded 504 people with less than 12 months of follow-up, leaving 1137. Due to how the variables were constructed, individuals for whom we could not determine a home facility (n=123) were included in most analyses, but were missing for area deprivation index (ADI), VA district or region, and facility complexity.

The device under investigation is a daily-use telemedicine foot temperature monitoring SmartMat made by Podimetrics. Patients with a high risk of ulceration (mainly due to a history of foot ulceration; lower limb, foot, or toe amputation; or Charcot foot) are eligible for RTM (Kyle Nordrum, DPT, personal communication). The device is ready to use without any configuration or set-up. A temperature scan takes 20 seconds, and the temperature data are transmitted to the cloud using an embedded cellular component. All scans are timestamped, allowing for objective measures of use. The software detects “hot spots,” defined as asymmetries of ≥2.2˚C between the same region on the left and right foot or different regions on the same foot. Temperature asymmetries that persist for at least 2 days are predictive of ulceration [21]. When hot spots are detected, either clinical support staff from the company or the VA provider follows-up with the patient to evaluate risk factors and make recommendations about actions to take. Patients are also called if they fail to use the mat for several days in a row to assess reasons for nonuse and encourage re-engagement.

While daily use is recommended, using the device at least 2 times per week is thought to be sufficient for the detection of hot spots (Jon Bloom, MD, personal communication). As we had a relatively long follow-up (12 mo), we defined noncompliance as using the mat less than 2 times per week for at least a quarter of the months under observation. Specifically, patients were considered noncompliant if they used the mat less than 2 days per week for 4 or more months out of 12.

We evaluated demographic, geographic, clinical, and facility factors, as well as health care use as potential correlates of nonadherence. Details about the data sources, definitions, and categories are included in Table 1.

We calculated the percentage of patients who were noncompliant, with 95% CIs. To evaluate associations between characteristics and noncompliance, we estimated OR and corresponding 95% CIs using a logistic regression model that included all covariates. The Akaike information criterion (AIC) was used to assess contributions of each covariate and groups of covariates to model fit along with a likelihood ratio test for a model that excluded the covariate or group of covariates [25]. AIC helps to quantify how well a model fits the data it was generated from relative to other models fit on the same data. AIC penalizes models that use more parameters to reduce the potential for overfitting. Lower AIC scores are considered evidence of better model fit.

Missing data were recovered using multiple imputation by chained equations using all covariates and the outcome and results displayed are from pooling the 20 imputed datasets [26-28]. Generalized variance inflation factors [29] for each of the covariates were calculated to assess correlation between covariates, and the impact it may have had on regression results. A variance inflation factor of 4 or more was used as evidence of substantial collinearity [30]; there was no evidence of substantial collinearity.

We conducted several sensitivity analyses. First, because there is no empirical basis for our definition of noncompliance [31], and some prior studies have used a higher cutoff, we conducted analyses using different cut points of minimum days per week on average (2 and 3) and months in the past 12 (11, 9, and 6) for defining noncompliance. Second, we graphically explored the association between hospitalization and separately, amputation, on SmartMat use by examining use in the 6 months prior to, and the 6 months after a hospitalization or amputation (separately). In these analyses, we categorized days per week into 4 categories: no use, <2 days per week, 2 to <5 days per week, and 5‐7 days per week. We also graphed use over time in those with a hospitalization or amputation (separately) relative to those who had neither a hospitalization nor an amputation.

This program evaluation qualified as a nonresearch quality improvement activity conducted under the authority of VHA operations. It complies with the VHA definition of “non-research operations activities” outlined in section 5a of the 2019 VHA Program Guide 1200.21: VHA Operations Activities That May Constitute Research, meeting both specified conditions: (1) the evaluation was designed and implemented for internal VHA purposes and (2) not designed to produce information to expand the knowledge base of a scientific discipline.

Even though the requirements of monitoring are minimal (20 s per day to obtain a scan), nearly 4 in 10 patients did not use the SmartMat at least 2 days per week for at least 75% of the months under observation. Based on AIC differences, the strongest factors associated with noncompliance were behavioral factors (poor glucose control [as measured by hemoglobin A_1c] and smoking), and to a lesser extent, factors such as a recent history of osteomyelitis and an elevated Gagne comorbidity index, indicating a high comorbidity burden. Patients who are unable to manage their blood glucose levels or quit smoking, as well as those with osteomyelitis and numerous chronic comorbidities may also have challenges regularly using a home temperature monitoring device. Noncompliance was also higher among those who had a hospitalization or amputation. Although some results were different in descriptive analyses (eg, BMI and ADI) or statistically significant (eg, ADI) in our multivariable model, the absence of large AIC differences indicated that they did not contribute importantly to model fit and therefore may not be important factors for compliance. Additional research that replicates these findings and that can help us understand reasons for noncompliance in patients would be helpful to inform future interventions.

We are only aware of a single study [21] that evaluated compliance with a SmartMat for foot temperature monitoring, and that study found lower noncompliance using a more stringent definition. Unlike our study, the Frykberg et al [21] study excluded days during which a participant had a contraindication to using the mat (eg, for an open plantar wound) and considered those who did not use the mat for >28 consecutive days as lost to follow-up (18.6% of the sample) [21]. As our sensitivity analyses indicated a reduction in use following a hospitalization or amputation, differences in use between the prior study and ours may be partly because we did not exclude any days. In any case, noncompliance in this study is similar to prior studies of foot temperature monitoring involving handheld thermometers. For example, in a trial of 151 people randomized to daily foot temperature monitoring using a handheld thermometer, 62% of participants measured foot temperatures at least 70% of days (equivalent to 38% noncompliance) [14]. Likewise, a study of daily temperature monitoring in Peru observed 60% compliance when treating those who did not return logbooks as nonadherent (vs adherence of 87% when leaving them out of the analysis, equivalent to 40% and 13% noncompliance) [32]. Estimates for foot temperature monitoring are also similar to general adherence to a variety of self-management activities in people with diabetes observed in meta-analyses [33]. In summary, even though RTM is a relatively low burden intervention, noncompliance rates are strikingly similar despite different definitions of adherence and different activity or intervention burdens, suggesting that factors other than time burden likely impact compliance.

The Unified Theory of Acceptance and Use of Technology 2 (UTAUT2) [34] is a helpful framework for understanding why individuals accept and use technology and may help to understand factors impacting compliance with RTM. UTAUT2 outlines how 7 constructs influence consumers’ intention to use a technology: (1) performance expectancy (does it help?), which is analogous to relative advantage in the diffusion of innovation literature [35] and perceived usefulness in the technology acceptance model [36], (2) effort expectancy (how easy is it to use?), (3) facilitating conditions (are there resources available to support use?), (4) social influence (do those close to the individual support their use of the technology?), (5) hedonic motivation (is it fun?), (6) price value (is it worth it financially?), and (7) habit (does it become a habit?).

Because we lack data from patients’ perspectives, it is unclear whether patients perceived a benefit, especially those who did not have any hot spots, or had many hot spots. Unlike a scale or a blood pressure monitoring device, which provides direct feedback with each use, and a sense of accomplishment for those who are losing weight or lowering their blood pressure, patients were not routinely provided direct feedback. They were called if they did not use the mat, or if temperature asymmetries were detected. If perceived benefit is low, use may decrease over time. Even though the apparent time burden is low, there may be steps (such as removing socks and shoes) that may be challenging and prevent people from using the mat more regularly. Given that an annual SmartMat subscription includes access to company personnel to answer questions, the third construct from the UTAUT2, facilitating conditions, may be high. As social influence is known to be important in diabetes management [37], social influence may be an important factor for this technology. Unfortunately, we had no direct measures of social influence, and marital status (a poor proxy) was not associated with compliance, which is understandable because marital status does not provide a direct measure of whether someone has a positive social influence. RTM is intended for disease management and was not designed to be fun, so hedonic motivation may be low. Gamification (eg, points and badges for streaks or other goals) could make it more fun and might improve compliance [38]. Price value is likely low since VA provides the SmartMat free of charge to patients. Future studies that collect self-report information from patients, including patient interviews, could help elucidate the extent to which factors in the model facilitate use.

A major strength of our study was its large size and ability to examine associations between various characteristics and SmartMat noncompliance. A potential limitation is that we did not have data on each day’s use, but instead an average number of days per week per month of follow-up. Though less precise, this level of detail is informative to understand trends. Third, there are no standard definitions of adherence or noncompliance, and our definition was different from the 1 used in a prior study [21], though it was based on expert opinion. Fourth, we lacked information on factors not readily available in the medical record such as patient preferences, perceived benefits, beliefs, attitudes, social support, and environmental factors that may have impacted use of the mats. This information would be useful to collect in future studies. Finally, because our sample included VA patients, individuals were primarily older White males, so results may not generalize to more diverse samples. Future research, particularly randomized trials testing different approaches to increase compliance (eg, gamification, incentives, patient and caregiver education, motivational interviewing, and reminders or alerts) in different patient populations, will be valuable to informing how to make technologies such as these more impactful.

Our study found that a large fraction of patients did not use the SmartMat as directed, and thus they would be unlikely to benefit from it. Testing approaches to proactively provide additional support for self-monitoring to patients with poor glucose control, current smoking, or high comorbidity burden (factors associated with high noncompliance) is an important area of future research. Future research should also seek to understand patients’ perspectives on their experience with SmartMats and why they may have routinely used, rarely used, or stopped using the mat. Reducing the risk of ulcer recurrence and amputation could have enormous benefits for individual patients and lower health care costs. Thus, ensuring that patients effectively employ tools to reduce the risk of ulcer recurrence is paramount.