Patient Demographics and Clinic Type Are Associated With Patient Engagement Within a Remote Monitoring Program

Abstract

Background: Remote physiological monitoring (RPM) is accessible, convenient, relatively inexpensive, and can improve clinical outcomes. Yet, it is unclear in which clinical setting or target population RPM is maximally effective.

Objective: To determine whether patients' demographic characteristics or clinical settings are associated with data transmission and engagement.

Methods: This is a prospective cohort study of adults enrolled in a diabetes RPM program for a minimum of 12 months as of April 2020. We developed a multivariable logistic regression model for engagement with age, gender, race, income, and primary care clinic type as variables and a second model to include first-order interactions for all demographic variables by time. The participants included 549 adults (mean age 53 years, 63% female, 54% Black, and 75% very low income) with baseline hemoglobin A1c ≥8.0% and enrolled in a statewide diabetes RPM program. The main measure was the transmission engagement over time, where engagement is defined as a minimum of three distinct days per week in which remote data are transmitted.

Results: Significant predictors of transmission engagement included increasing age, academic clinic type, higher annual household income, and shorter time-in-program (p < 0.001 for each). Self-identified race and gender were not significantly associated with transmission engagement (p = 0.729 and 0.237, respectively).

Conclusions: RPM appears to be an accessible tool for minority racial groups and for the aging population, yet engagement is impacted by primary care location setting and socioeconomic status. These results should inform implementation of future RPM studies, guide advocacy efforts, and highlight the need to focus efforts on maintaining engagement over time.

Introduction

Remote physiological monitoring (RPM) has been shown to improve clinical outcomes^1,2 and is a promising tool for health care delivery given its accessibility, convenience, and favorable cost profile. However, adoption of RPM has been hindered by lack of uniform reimbursement, variable patient engagement, and unequal connectivity and patient access to internet services. In addition, RPM research varies in intervention complexity, program design, and clinical outcomes. Together, these heterogeneities in the reported literature impede widespread acceptance and implementation of telemedicine programs and leave the question unanswered: in which clinical setting or target population RPM is maximally effective?³ Despite these challenges, adoption of RPM has rapidly accelerated during the COVID-19 pandemic⁴ and is expected to persist postpandemic.

Although the COVID-19 pandemic has forced the adoption of telemedicine by the mainstream, a divide in access to care persists. A recent study found that increasing age, male gender, non-White race, lower education and income level, and rurality continue to be predictors of unreadiness to engage in telemedicine.⁵ This is consistent with survey results published in 2018, which found that increasing age, Medicare insurance, lower education level, and low annual income were associated with decreased likelihood to have a smartphone and to use a health application.⁶ In contrast, a 2013 analysis of Medicare claims for rural beneficiaries showed that those who received a telemedicine visit were more likely than those who did not live in poorer communities.⁷ Ideally, RPM and other telehealth modalities can be used to improve health equity by offering an accessible means of health care for those facing disparate health outcomes.

Our institution has launched a statewide diabetes and hypertension RPM program targeting low-income and rural adult patients with type 2 diabetes mellitus. To ensure the success of our program, we have taken steps to optimize its feasibility and accessibility to the target population. The program uses an automated 2-in-1 glucose- and blood pressure-monitoring device that is embedded with a SIM card. Patients are asked to check their blood pressure and blood sugar, as they would do with any standard blood pressure cuff or glucose meter. When the device comes into range of any cellular tower, it transmits data to a secure online web server for provider review. For patients not living within reach of a cellular signal, the compact device can be transported. All data points are stored on the device and will be transmitted when within cellular reach. The embedded SIM card obviates the need for patients to own a smartphone or other cellular device or to have broadband access. By removing unnecessary technological barriers, this RPM device could improve clinical care for high-risk patients with limited access to traditional internet or phone.

In the setting of this easy-to-use device and RPM program, we sought to identify demographic factors or clinical settings that predict data transmission frequency over time, as a surrogate marker for device use and program engagement. Patients of our ongoing implementation program achieve a mean hemoglobin A1c (hbA1c) reduction of 1.3% over 12 months⁸; however, not all patients equally benefit. Some patients fail to transmit data altogether. If we assume that improvements in hbA1c are related to providers' ability to act on data, success depends on data transmission. Data transmission, in turn, is dependent upon patient engagement as manifested by device use. Prior work confirms that the degree of patient engagement and activation correlates with the degree of clinical benefit from RPM.⁹ Through a series of regression models, we aimed to determine whether patients' demographic characteristics or clinical settings are associated with data transmission and engagement, defined as at least 3 days per week of data transmission over time.

Methods

Sample/Cohort

All patients enrolled in a statewide diabetes RPM program for a minimum of 12 months as of April 2020 were considered for analysis in this prospective cohort study. The patient population of the RPM program includes adults with type 2 diabetes mellitus and a baseline hbA1c of 8.0% or higher.

Patients who withdrew from the program before 12 months were excluded. We also excluded those patients who, according to their primary care site, could not be contacted (i.e., were lost to follow-up), were deceased, or had transferred their care to a nonparticipating site. In addition, patients were excluded if age data were missing, as this is a required variable for model inclusion.

Independent Variables

We obtained demographic data, including age, gender, race, ethnicity, health insurance, and annual household income from all sources, by self-report and/or electronic medical record extraction at the time of enrollment. The enrollment questionnaire is available in Supplementary Figure S1. Trained clinical staff collected and entered these values into Research Electronic Data Capture's (REDCap) secure, web-based software platform hosted by the Medical University of South Carolina (MUSC).¹⁰ A variable for week was also included so that the effect of “time-in-program” could be assessed.

Age was collapsed into a categorical variable with three options: 18–44 years old, 45–64 years old, and 65 years or more. Annual household income was reported by patients as a categorical variable in $10,000 increments, starting at $0. For purposes of analysis, race, insurance, and household income were collapsed to binary variables, as given in Table 1. Owing to small sample size, non-White and non-Black groups (e.g., Asian) were included in the non-Black cohort. Patients missing race data or self-classified as “other” were also included in the non-Black group. Clinic type was assigned based on the location of the patient's primary care office and analyzed as a categorical variable with three possible responses: free/volunteer, federally qualified health center (FQHC), or academic clinic.

Table 1.

Clinical and Sociodemographic Characteristics: Overall Sample, and by Absence or Presence of Transmission Data

VARIABLE	OVERALL (%) (N = 549)	WITHOUT TRANSMISSION DATA (%) (N = 63)	WITH TRANSMISSION DATA (%) (N = 486)	p
Age (years, mean ± SD)	53.1 ± 11.8	48.0 ± 12.7	53.7 ± 11.5	<0.001a
Age group (years)
18–44	127 (23.1)	29 (46.0)	98 (20.1)	<0.001b
45–64	328 (59.7)	27 (42.9)	301 (61.9)
65 and older	94 (17.1)	7 (11.1)	87 (17.9)
Gender
Male	201 (36.6)	27 (42.9)	174 (35.8)	0.27b
Race
Black	298 (54.3)	22 (34.9)	276 (56.8)	<0.001c
Non-Black
Whited	215 (39.2)	26 (41.3)	189 (38.9)
American Indian or Alaskan native	16 (2.9)	7 (11.1)	9 (1.9)
Asian	4 (0.7)	2 (3.2)	2 (0.4)
Native Hawaiian or Pacific Islander	7 (1.3)	3 (4.8)	4 (0.8)
Missing race data	9 (1.6)	3 (4.8)	6 (1.2)
Ethnicity
Hispanic	103 (18.8)	29 (46.0)	74 (15.2)	<0.001b
Non-Hispanic	445 (81.2)	34 (54.0)	411 (84.6)
Missing	1 (0.2)	0 (0.0)	1 (0.2)
Annual household income
$0–19,999	411 (74.9)	50 (79.4)	361 (74.3)	0.38b
$20,000 or more	138 (25.1)	13 (20.6)	125 (25.7)
Health insurance
Uninsured	258 (47.0)	42 (66.7)	216 (44.4)	<0.001b
Insurede	291 (53.0)	21 (33.3)	270 (55.6)
Primary care clinic type
FQHC	345 (62.8)	42 (66.7)	303 (62.4)	<0.001b
Academic	132 (24.0)	2 (3.2)	130 (26.8)
Free	72 (13.1)	19 (30.2)	53 (10.9)

Demographic and socioeconomic characteristics of the overall sample population and of subsets that never transmitted data and those that transmitted at least once during the program.

^aTwo sample t-test.

^bChi square test.

^cFisher's exact test.

^dWhite or Caucasian, depending on term used by patient or electronic health record.

^eAny insurance reported by patient.

FQHC, federally qualified health center.

Dependent Variable Data Collection

A data manager exported all transmitted data directly from the device vendor's server (ForaCare^®, Moorpark, CA) onto a secure research server for all patients included in the study cohort from the time of enrollment to March 31, 2020. We specifically analyzed transmission frequency over the first 52 weeks of program participation. We determined the presence of engagement per week per patient, defined as remote data transmission a minimum of 3 distinct days per week. Engagement was reported in binary form: engaged or not. Continuity of transmission engagement by individual patients was not assessed.

Analyses

First, we compared demographic variables between patients with and without transmission data, using two sample t-tests for continuous variables and chi square and Fisher's exact tests for categorical variables. Analysis for collinearity of independent variables revealed high correlation between ethnicity and race as well as insurance and clinic type. As a result, we opted not to include ethnicity and insurance in subsequent modeling. The dependent variable was calculated as binary presence of engagement within prespecified demographic subgroups who transmitted at least 3 days per week, for each of 52 weeks for each patient. This is represented as transmission engagement over time (Fig. 1).

Fig. 1.

Transmission engagement over time. (a) Percentage of cohort, by race, who achieve data transmission engagement by week after remote monitoring initiation, where engagement is defined as a minimum of one transmission on 3 or more separate days per week. Week 1 is the first week after monitoring initiation; week 52 is the final week (after 1 year of monitoring). Non-Black includes participants identifying as White, Caucasian, Native American, Pacific Islander, Native Hawaiian, Alaskan Native, missing race data, or other. Non-Black represented by black line. Black represented by gray line. (b) Percentage of cohort, by clinic site, who achieve data transmission engagement by week after remote monitoring initiation. Academic clinic represented by black line, FQHC by dark gray line, and free clinics by light gray line. (c) Percentage of cohort, by gender, who achieve data transmission engagement by week after remote monitoring initiation. Female represented by black line. Male represented by gray line. (d) Percentage of cohort, by age group, who achieve data transmission engagement by week after remote monitoring initiation. Age 18 to 44 years represented by black line, ages 45 to 64 years by dark gray line, and age 65 years and older by light gray line. (e) Percentage of cohort, by self-reported income, who achieve data transmission engagement by week after remote monitoring initiation. Higher income (black line) is defined as an annual household income of $20,000 or more, and low income (gray line) is defined as <$20,000 per year. FQHC, federally qualified health center.

Second, we developed a multivariable logistic regression model for engagement with age, gender, race, income, and primary care clinic type as variables (Table 2, Model 1). A second model was created to include first-order interactions for all demographic variables by time (Table 2, Model 2). Statistical analyses were performed using SAS version 9.4, and significance was determined at the 5% level. The institutional review board at the MUSC approved this study protocol.

Table 2.

Logistic Regression Models of Engagement by Demographic Variables and Clinic Type

	MODEL 1		MODEL 2
	OR (95% CI)	p	OR (95% CI)	p
Age group, years (Ref.: 18–44)
45–64	2.108 (1.979–2.246)	<0.001	2.202 (1.947–2.491)	<0.001
65+	2.744 (2.523–2.985)	<0.001	2.333 (1.969–2.764)	<0.001
Male	1.032 (0.980–1.087)	0.237	0.969 (0.872–1.077)	0.559
Black	0.990 (0.938–1.046)	0.729	1.149 (1.029–1.282)	0.014
Clinic type (Ref.: academic)
FQHC	0.486 (0.455–0.519)	<0.001	0.547 (0.477–0.627)	<0.001
Free	0.455 (0.414–0.500)	<0.001	0.383 (0.316–0.463)	<0.001
Annual household income <$20,000	0.853 (0.805–0.904)	<0.001	0.787 (0.698–0.888)	<0.001
Week	0.973 (0.971–0.974)	<0.001	0.975 (0.968–0.982)	<0.001
Interactions
Week by age, years (Ref. 18–44)
Week by age (45–64)			0.998 (0.994–1.003)	0.441
Week by age (65+)			1.006 (1.000–1.012)	0.041
Week by male			1.002 (0.999–1.006	0.182
Week by Black			0.994 (0.991–0.998)	0.002
Week by annual household income <$20,000			1.003 (0.999–1.007)	0.134
Week by clinic (Ref.: academic)
Week by clinic (FQHC)			0.995 (0.991–1.000)	0.040
Week by clinic (Free)			1.007 (1.001–1.013)	0.032

Multivariable logistic regression models for engagement with age, gender, race, income, and primary care clinic type as variables (Model 1) as well as a second model (Model 2) to incorporate first-order interactions of those variables with time-in-program.

CI, confidence interval; OR, odds ratio.

Results

Sample Characteristics

A total of 674 patients met criteria for inclusion. Of those, 121 patients were excluded due to withdrawal from the program before 12 months of participation, and 4 patients were excluded due to missing age variable, leaving a total of 549 patients included in the analysis. A total of 63 patients had 0 transmissions. These patients were included in the analysis, similar to an intention-to-treat framework; however, characteristics of this population are also reported separately in Table 1. Most patients in the sample cohort were younger than 65 years (82.9%), were female (63.4%), had an annual household income <$20,000 (74.9%), and received care at an FQHC (62.8%) (Table 1). A slight majority identified as Black (54.3%) (Table 1). Compared with the overall cohort, patients who failed to transmit any data over the course of the program were younger (mean age 48.0 years), more often identified as Hispanic (46.0%), and were more likely to attend a free clinic (30.2%) and to lack insurance (66.7%).

Transmission Engagement Over Time

Greater proportions of Black patients and patients receiving care at an academic medical center transmitted regularly throughout 52 weeks of program enrollment (Fig. 1a, b). Similar transmission engagement rates were seen among male and female patients (Fig. 1c). Adults 65 years or older were most likely to be engaged with transmission, whereas patients in the youngest subgroup were least likely to transmit 3 days or more per week (Fig. 1d). Engagement with data transmission was greater in those with a higher ($20,000 or more) than lower (<$20,000) annual income (Fig. 1e).

Logistic Regression

In the logistic regression model, significant predictors of transmission engagement included age, academic clinic type, higher annual household income, and shorter time-in-program (p < 0.001 for each; Table 2, Model 1). Specifically, patients aged 45 years or more were two to three times as likely to be engaged with data transmission as those aged 44 years or less (odds ratio [OR] 2.108 for adults 45–64 years confidence interval [95% CI: 1.979–2.246]; OR 2.744 for adults aged 18–44 years [95% CI: 2.523–2.985]). Patients from FQHCs and free clinics were about half as likely to be engaged as participants from academic clinics (OR 0.486 for FQHC [95% CI: 0.455–0.519]; OR 0.455 for free clinics [95% CI: 0.414–0.500]). Patients with income <$20,000 were ∼15% less likely to be engaged than the higher income group (OR 0.853 [95% CI: 0.805–0.904]). For each advancing week in the program, transmission engagement rates decreased by ∼3% (OR 0.973). Self-identified race and gender were not significantly associated with transmission engagement (p = 0.729 and 0.237, respectively).

Interaction Effects

First-order interactions were significant for time-in-program by race, age 65 years and older, free clinic, and FQHC (p < 0.05 for each; Table 2, Model 2). Gender, household income, and age 45 to 64 years did not interact with time-in-program to impact the likelihood to transmit regularly over time. Although Black patients started at a higher rate of engagement, those rates declined proportionally by ∼0.6% per week compared with non-Black patients (OR 0.994 [95% CI: 0.991–0.998]). By week 52 of program participation, the proportion of transmission engagement for Black patients approached that of non-Black patients. In patients aged 65 years and older, transmission engagement declined slightly less over time proportionally compared with younger patients (1.006 [95% CI: 1.000–1.012]), widening the difference in transmission engagement among the groups. There appears to be a small yet significant interaction between patients of free or FQHC clinic and time-in-program. Transmission engagement declined slightly less over time, proportionally, in patients of free clinics (OR 1.007 [95% CI: 1.001–1.013]) and slightly more in patients of FQHCs (OR 0.995 [95% CI 0.991–1.000]) compared with academic clinic patients.

Discussion

In this analysis of transmission frequency among patients enrolled in a diabetes and hypertension RPM program, increasing age, higher household income, and receipt of care at an academic medical center were predictive of engagement in home monitoring. Not surprisingly, engagement in thrice-weekly data transmission declined for all patients as time progressed. Male and female patients were equally engaged over time. Although overall there was not a significant difference in transmission engagement by race (Table 2, Model 1), there was a significant difference in engagement through time between Black and non-Black patients (Table 2, Model 2). Specifically, Black patients started at a higher rate of engagement but those rates dropped more quickly than seen among non-Black peers. Furthermore, a smaller proportion of Black patients were represented in the “never transmitter” group (Table 1). These observations on race suggest that RPM could be an effective avenue to provide care for patients of varying racial backgrounds, but that maintaining engagement over time will be important.

Our results suggest that telemedicine is accessible and attainable for older patient populations. Patients aged 45 years and older were more likely than their younger peers to transmit regularly. In fact, the group aged 65 years and older showed the highest rates of transmission engagement over time with less drop-off in transmission and a lower percentage of patients who failed to transmit any data. These results refute potential concerns for usability of telemedicine among older groups.

Patients reporting an annual household income <$20,000 were less likely to be engaged with data transmission than their higher income peers. This observation raises an important concern regarding the role of telemedicine in addressing health equity. Disparities in social determinants and health outcomes have long been known but have become even more glaring during the COVID-19 pandemic.^11,12 The surge of telemedicine during the pandemic raises the question of whether virtual care and remote monitoring will enable vulnerable populations to access care or whether it will further widen the digital divide by privileging technologically enabled communities. Our finding of decreased transmission engagement among lower income patients suggests that RPM may not equally serve all socioeconomic subgroups and warrants further investigation.

Finally, patients who received their primary care from academic clinics had greater rates of data transmission than did patients who received care at free clinics or FQHCs. This effect endured over time. Increased staffing and support at academic health centers could account for that difference, as free clinics largely rely on volunteers for staffing and may not have access to trained social workers or health educators. An analysis is currently under way to assess for clinic structure and staffing characteristics that impede RPM engagement in the various clinical settings.

A second possible explanation for increased transmission among academic medical center patients is greater access to cellular towers. Patients of an academic medical center are more likely to live in an urban setting while FQHCs or free clinics are more likely to serve rural populations. Rural patients may fail to transmit due to lack of cellular signal despite regular home testing. More than 95% of patients who failed to transmit a single value received care at either a free clinic or an FQHC, which may signal inability of the device to connect to a cellular tower. Advocacy for connectivity and equity in communications access will continue to be important as telemedicine expands.

Varying transmission engagement among clinic types could also be reflective of underlying differences in patient populations. For example, patients of free clinics were much more likely to report a lack of health insurance and to identify as Hispanic. Lack of health insurance is known to correlate with lower education level,¹³ which may in turn be associated with lower health literacy and likelihood to maintain engagement in RPM. Transient living environments and lack of access to stable health care understandably hinder regular engagement in care over 12 months and may explain some of the observed differences in transmission engagement among patients of free clinics. We are currently conducting interviews of RPM participants to better understand the observed trends and factors that may facilitate or impede data transmission and program engagement.

It is important to note that the cohort in this program is vulnerable to selection bias. Patients with type 2 diabetes, defined as an hbA1c of 8.0% or higher, who received primary care at a participating clinic, were offered enrollment in an opt-in manner. A number of patients withdrew from the program before 12 months and thus were excluded from analysis. Reasons for withdrawal included transfer of care to a nonparticipating clinic, voluntary withdrawal, and death. A disproportionate number of these received care at free clinics. Excluded patients comprised a total of 24 patients from academic medical centers (15.4% of 156 total academic center patients), 65 from FQHCs (15.9% of 410 patients), and 36 from free clinics (33.3% of 108 total patients).

There are many strengths of this analysis. First, the year-long duration of follow-up allows assessment of trends over time, including expected attrition due to waning attention and competing priorities. Second, the sample size of this cohort is larger than that of many other RPM studies and includes patients of diverse backgrounds. Generalizability is further strengthened by the involvement of multiple enrolling sites operating under diverse business and clinical care models. We believe that the heterogeneity of sites substantiates the scalability and external validity of implementing this work into a real-life clinical setting, although the heterogeneity does not allow for standardization of treatment protocol as would be desired in a controlled trial. This lack of standardization was intentional in an effort to work within the current clinical workflow of participating sites, each of which operates under its own staffing model. Along these lines, a third strength of this work is the ability to implement this program in a variety clinical practice models, and a fourth strength is the ease of the use of the device by patients. Patients of rural areas and low education level were able to effectively transmit data with this device, suggesting that this technology is accessible to at-risk populations.

This prospective cohort study builds on an existing body of literature supporting the use of RPM for management of chronic diseases such as diabetes and hypertension, while further broadening our understanding of the role of RPM within certain population subgroups. RPM appears to be an accessible tool for diverse racial groups and for the aging population, but its effectiveness is impacted by health care access, primary care location setting, and, to a lesser degree, socioeconomic status. Mixed methods research including in-depth interviews is underway and may reveal barriers to transmission engagement specifically faced by underserved populations and clinics, so that RPM can be optimized for these groups. As the United States continues to expand telemedicine use, we must consider how to adapt technology to best serve vulnerable populations, lest we risk widening health inequities.

Disclaimer

The content is solely the responsibility of the authors and does not necessarily represent the official views nor an endorsement by NIH, HRSA, HHS, or the U.S. Government.

Disclosure Statement

E.K., W.P.M., P.M., A.S., S.O.S, M.H., J.M., and J.Z. report grants from National Center for Advancing Translational Sciences of the National Institutes of Health and grants from Health Resources and Services Administration, during the conduct of the study.

Funding Information

This publication was supported, in part, by the National Center for Advancing Translational Sciences of the National Institutes of Health (NIH) under Grant Number UL1 TR001450 and by the Health Resources and Services Administration (HRSA) of the U.S. Department of Health and Human Services (HHS) as part of the National Telehealth Center of Excellence Award (U66 RH31458).

Supplementary Material

Supplementary Figure S1