Remote patient monitoring sustains reductions of hemoglobin A_1c in underserved patients to 12 months

Abstract

Aims

We sought to determine whether underserved patients enrolled in a statewide remote patient monitoring (RPM) program for diabetes achieve sustained improvements in hemoglobin A_1c at 6 and 12 months and whether those improvements are affected by demographic and clinical variables.

Methods

Demographic and clinical variables were obtained at baseline, 6 months and 12 months. Baseline HbA1c values were compared with those obtained at 6 and 12 months via paired t-tests. A multivariable regression model was developed to identify patient-level variables associated with HbA_1c change at 12 months.

Results

HbA_1c values were obtained for 302 participants at 6 months and 125 participants at 12 months. Compared to baseline, HbA_1c values were 1.8% (19mmol/mol) lower at 6 months (p<.01) and 1.3% (14mmol/mol) lower at 12 months (p<.01). Reductions at 12 months were consistent across clinical settings. A regression model for change in HbA_1c showed no statistically significant difference for patient age, sex, race, household income, insurance, or clinic type.

Conclusions

Patients enrolled in RPM had improved diabetes control at 6 and 12 months. Neither clinic type nor sociodemographic variables significantly altered the likelihood that patients would benefit from this type of technology. These results suggest the promise of RPM for delivering care to underserved populations.

1. INTRODUCTION

Diabetes mellitus affects approximately 30 million people in the U.S. (1), where it is estimated to contribute to over 7 million hospitalizations per year and is associated with $245 billion in annual direct and indirect costs (2). Despite this enormous spending, the U.S. has a higher prevalence of diabetes and poorer health outcomes than other nations worldwide (3).

Social determinants and barriers to care disproportionately threaten the health of underserved populations, especially those living with diabetes (4,5,6). Rural regions in the U.S. face poor transportation and technological infrastructure. Health literacy, financial constraints, and educational level also impede the ability to prevent and attain control of chronic diseases (7). Compounding these barriers is the lack of primary care workforce, particularly in rural areas (8). Some counties in South Carolina average one primary care provider to every 10,000 residents (9).

Telemedicine represents a flexible method of data communication that can facilitate treatment oversight and management of chronic disease states, regardless of access to local care (10,11,12). Remote patient monitoring (RPM), a form of telemedicine that allows transmission of data from the patient to the healthcare provider, could reduce barriers to care by enabling providers to monitor patients and to adjust treatment plans outside of traditional office visits. RPM is particularly well suited for diabetes care delivery, as the American Diabetes Association recommends self-monitoring of blood glucose with provider data review and interpretation (13).

Current literature provides optimism for the role of telemedicine in chronic diabetes management, but the feasibility of widespread implementation of RPM for low-income, rural populations is still in question. Early studies, including one from our institution, demonstrated effectiveness of RPM in HbA_1c reduction over six months (14). Subsequent studies confirmed these findings within specific populations and over longer time periods (15,16), but studied interventions include resource-intensive coaching or education that are not feasible for low-resourced rural clinics. Indeed, scalability has been identified as a major barrier to current remote health interventions directed at diabetes care (17). Evidence is inconclusive as to whether RPM, in the absence of more intense coaching or counseling, can lead to large-scale, lasting reductions in disparities in diabetes care for low-income and rural populations.

Evidence is also inconclusive and inconsistent about whether telemedicine-based interventions universally enhance diabetes care or whether efficacy differs by population subgroup. In a survey of a vulnerable urban population, most adults reported smartphone ownership and willingness to use a mobile health application; however, increasing age, Medicare insurance, lower education level, and low annual income were associated with decreased likelihood to have a smartphone and lower willingness to use a health application (18). Additionally, a systematic review found technological illiteracy to be one of the most commonly cited barriers to telemedicine use in diabetes, particularly among low-income populations (17). An early study further suggested that gender differences may impact patients’ likelihood to complete a three-month RPM program and to receive clinical benefit from it (19). More work is needed to determine patient-level factors that influence success rates with RPM for diabetes management, as it is not clear whether all population groups are equally likely to benefit from this intervention (20).

Using data from participants of a statewide RPM program launched by our institution to improve management of type 2 diabetes in low-income and rural populations, we seek to understand the degree and duration of HbA_1c reduction among RPM participants and to identify patient-level variables that influence the likelihood to achieve HbA_1c reduction at 6 and 12 months.

2. SUBJECTS, MATERIALS AND METHODS

2.1. Statewide RPM program model

In 2015, our institution launched a statewide quality improvement and implementation program focusing on RPM as a tool for diabetes and hypertension management, particularly for low-income and rural populations. This program and its structure are based on a previous, successful, randomized trial protocol out of the same institution (14). The program is coordinated by central staff at an academic medical center, with clinic champions at each participating clinical site, as discussed further in Section 2.1.2. Patient data, including blood glucose and blood pressure (BP) measurements, are transmitted in real-time through the RPM device. Outcome data, as outlined in Section 2.2.1 below, are collected and reported by the local clinical staff at 6-month intervals through a secure online platform. At the time of this analysis, more than 800 patients across the state of South Carolina had enrolled. The program continues to enroll new patients.

2.1.1. Eligibility and recruitment criteria

Patients are eligible to participate in the RPM program if they have an HbA_1c of 8.0% (64mmol/mol) or higher and receive their diabetes care from a participating site. Participating sites include two free or volunteer clinics, two federally qualified predominantly rural health centers (FQHC), and one urban academic clinic. The program has been initiated in a non-randomized stepwise fashion among the various clinics, with rolling enrollment, in accordance with aims of this implementation program.

Prior to enrollment, all patients agree to participate for 12 months, although they are allowed to withdraw at any time. If diabetes care transitions to a provider outside of the participating site during the program period, the participant is withdrawn.

2.1.2. Remote data monitoring

Participants are provided a 2-in-1 blood pressure and glucose-monitoring device (FORA D40g, ForaCare®, Moorpark, CA) and glucose monitoring supplies free of charge. The device is equipped with an embedded SIM card to allow direct data transmission to an online, secure website. The device uses 3G cellular connectivity to transmit data, obviating the need for patients to have their own data plan, cellular phone, or internet connection. Program staff at our institution review the transmitted data and alert partnering sites of abnormal values to allow for treatment adjustment in the inter-visit period. Partnering sites and providers are also encouraged to review the data directly.

The RPM program has a central coordinating hub, housed within our academic center. Central staff perform administrative functions, coordinate program operations, train local clinical staff and route abnormal clinical data. Local clinical staff enroll patients and oversee day-to-day medication titration. Through a series of orientation sessions, local clinical staff are trained by the central staff on the program structure, device operation, enrollment and data submission processes, and how to review transmitted patient data. Central staff send biweekly reports of patients who fail to transmit data or whose transmitted readings exceed set thresholds to local clinical staff. Central staff encourage local sites and providers to adjust treatment based on incoming glucose and BP data, but there is no centrally mandated treatment algorithm. Clinics are empowered to delegate treatment adjustments to the appropriate staff and to modify the workflow to integrate with current clinical workflows.

2.2. Current study

2.2.1. Collection of baseline, 6-month, and 12-month data

This study included all patients who enrolled in the aforementioned statewide RPM program between September 2017 and September 2019 and who submitted either 6- or 12-month follow-up data as of March 2020. Local sites were asked to obtain clinical and demographic variables for patients enrolled during the study period and submit them to the central site through REDCap’s (Research Electronic Data Capture) secure, web-based software platform (21). Demographic characteristics included age, sex, race, ethnicity, annual household income from all sources, and health insurance status at the enrollment visit. All of these demographic variables were self-reported by the patient at enrollment and entered into REDCap by local clinical staff. Annual household income was reported as a categorical variable in $10,000 increments, starting at $0. Health insurance was reported according to primary insurer. Clinical variables, including HbA_1c, BP, and weight, were measured and submitted to REDCap by local clinical staff at the enrollment visit (baseline), as well as 6 and 12 months after enrollment. These values were stored on REDCap for analysis by central program staff.

2.2.2. Paired t-test for HbA_1c comparison

In our primary analysis, we compared HbA_1c values obtained at baseline and at 6 and 12 months through paired t-tests. All patients with both a baseline and 6-month value were included in the 6-month cohort. Similarly, all patients with both baseline and 12-month values were included in the 12-month cohort. Patients with both 6- and 12-month data points were represented in both cohorts and analyses. A 12-week window around the 6- and 12-month HbA_1c measurement was allowed to accommodate common real-life barriers (e.g., transportation issues, insurance requirements of 90-day wait from prior HbA_1c measurement, unforeseen appointment cancellations). Data falling more than 6 weeks before or after the 6- or 12-month mark (i.e., outside of the 12-week window) were treated as missing data. Missing data were excluded from analysis given the limitations of imputation methods. Based on known differences in staffing models among participating clinics, we performed subgroup analyses based on clinic type: academic, FQHC, and free (clinics staffed by volunteer providers).

2.2.3. Multivariable linear regression model

We additionally developed a multivariable linear regression model for the change in HbA_1c over 12 months, with age, baseline HbA_1c, sex, race, income, insurance status, and primary care clinic type as variables. For purposes of analysis, race, ethnicity, household income, and insurance status were collapsed to binary outcomes, as represented in Table 1. Statistical analyses were performed using SAS version 9.4, and significance was determined at the 5% level. The Internal Review Board deemed this work to be quality improvement and thus advised that approval for human research was not required.

Table 1.

Sample characteristics

Variable (%)*	6-month cohort	12-month cohort
Sample size (n)	302	125
Baseline HbA1c, mean ±SD	10.7 ± 2.0 (93 ± 17 mmol/mol)	10.4 ± 2.0 (90 ±17 mmol/mol)
Follow-up HbA1c, mean ±SD	8.9 ± 2.2 (74 ± 18 mmol/mol)	9.1 ± 2.2 (76 ± 18 mmol/mol)
Age (years), mean ±SD	54.5 ± 11.5	57.5 ± 10.6
Sex
Male	37.1	28.0
Female	63.5	72.0
Race
Non-black†	37.8	32.0
Black	62.2	68.0
Ethnicity
Hispanic	12.9	8.0
Non-Hispanic	87.1	92.0
Annual household income
$0–19,999	72.8	30.4
$20,000 or more	27.2	69.6
Health insurance
None‡	46.7	29.6
Any insurance §	53.3	70.4
Primary care clinic type
FQHC	55.6	35.2
Academic	26.8	48.8
Free	17.6	16.0

Clinical and sociodemographic characteristics of participants with reported HbA_1c values at 6 months ± 6 weeks or 12 months ± 6 weeks compared to baseline

^*Variables are expressed in percentage, except where otherwise noted.

^†White or other

^‡No insurance

^§Commercial or private insurance, Medicare, Medicaid, military, other

3. RESULTS

3.1. Description of cohorts

HbA_1c values were obtained for 302 participants at 6 months and 125 participants at 12 months. The 6- and 12-month cohorts were similar in baseline HbA_1c, average age, sex, race, and ethnicity, with most participants being female, black, and in their mid-50s (Table 1). Follow-up HbA_1c values at 6 or 12 months were also similar for both cohorts (8.9% or 74mmol/mol vs 9.1% or 76mmol/mol, respectively).

In contrast, income level, insurance status, and primary care clinic type differed between the two cohorts. More participants in the 6-month than the 12-month cohort reported an annual income of less than $20,000 (72.8% vs. 30.4%) and lacked insurance (46.7% vs. 29.6%). The type of primary clinic also varied between the groups. The overall percentage of patients who received care at an FQHC decreased from 55.6% for the 6-month cohort to 35.2% for the 12-month cohort, while the percentage of those treated at an academic clinic increased from 26.8% to 48.8%.

3.2. Paired t-test comparison

Compared to baseline, RPM was associated with a 1.8% or 19mmol/mol absolute reduction in HbA_1c at 6 months (n= 302; p<0.01) and a 1.3% or 14mmol/mol reduction at 12 months (n= 125; p<0.01) (Table 2). A significant reduction in HbA_1c was achieved at both 6 and 12 months, regardless of clinic type. In the academic clinic, HbA1c decreased by 1.7% at 6 months (10.1 to 8.4%; n=81; p<0.01) and 1.3% at 12 months (10.1 to 8.8%; n=61; p<0.01). Among FQHCs, HbA1c decreased by 1.9% at 6 months (11.1 to 9.2%; n=168; p<0.01) and 1.3% at 12 months (10.5 to 9.2%; n=44; p<0.01). In free/volunteer clinics, HbA1c decreased by 1.3% at 6 months (10.4 to 9.1%; n=53; p<0.01) and 1.5% at 12 months (10.8 to 9.4%; n=20; p<0.01).

Table 2.

Change in HbA_1c compared to baseline, by clinic type

	N	Baseline hgbA1c % (mmol/mol)	Follow-up hgbA1c* % (mmol/mol)	Change in HbA1c % (mmol/mol)	P value
6-month cohort
Overall	302	10.7 ± 2.0 (93 ± 17)	8.9 ± 2.2 (74 ± 18)	−1.8 ± 2.4 (−19 ± 25)	<.0001
Academic clinic	81	10.1 ± 1.9 (87 ± 16)	8.4 ± 2.1 (68 ± 17)	−1.7 ± 2.2 (19 ± 25)	<.0001
FQHC†	168	11.1 ± 2.1 (98 ± 19)	9.2 ± 2.2 (77 ± 18)	−1.9 ± 2.5 (−21 ± 28)	<.0001
Free clinic	53	10.4 ± 1.8 (90 ± 16)	9.1 ± 2.0 (76 ± 17)	−1.3 ± 2.3 (−14 ± 25)	.0002
12-month cohort
Overall	125	10.4 ± 2.0 (90 ± 17)	9.1 ± 2.2 (76 ± 18)	−1.3 ± 2.3 (−14 ± 25)	<.0001
Academic clinic	61	10.1 ± 2.0 (87 ± 17)	8.8 ± 2.1 (73 ± 17)	−1.3 ± 2.4 (−14 ± 26)	.0001
FQHC†	44	10.5 ± 2.0 (91 ± 17)	9.2 ± 2.3 (77 ± 19)	−1.3 ± 2.3 (−14 ± 25)	.0007
Free clinic	20	10.8 ± 1.8 (95 ± 16)	9.4 ± 2.3 (79 ± 19)	−1.5 ± 2.3 (−16 ± 25)	.0100

Absolute change in HbA_1c from baseline to 6 months or 12 months, with subgroup analysis based on clinic type. Data are presented as mean ± SD unless otherwise indicated. P values represent results of t-tests.

^*At 6 or 12 months ± 6-week window.

^†Federally Qualified Health Center

3.3. Multivariable regression model

After adjustment for sociodemographic and other covariates, the multivariable regression model showed no statistically significant difference in degree of HbA_1c change based on patient age, sex, race, household income, insurance, or clinic type (Table 3). A higher baseline HbA_1c was associated with a larger reduction in HbA_1c at 12 months (β= −0.64; p <0.01).

Table 3.

Multivariable linear regression model for change in HbA_1c over 12 months^*

Covariate	Estimate	Standard Error	t Value	P value
Age	−0.034	0.019	−1.8	0.074
Baseline HbA1c	−0.638	0.099	−6.48	<.0001
Male	0.072	0.435	0.17	0.8686
Black	0.547	0.459	1.19	0.2363
Low annual household income	0.075	0.431	0.17	0.8615
Any insurance	−0.454	0.598	−0.76	0.4495
Primary care clinic type (reference: Academic)
FQHC†	0.247	0.467	0.53	0.5976
Free	−0.310	0.756	−0.41	0.6825

Modeling the primary outcome of change in HbA_1c over 12 months (n=125).

^*± 6-week window

^†Federally Qualified Health Center

4. DISCUSSION

4.1 Our work demonstrates clinically significant reductions in HbA_1c in association with a relatively easy-to-implement diabetes RPM. The finding of a sustained reduction in HbA_1c at 12 months aligns with other studies suggesting that RPM helped patients attain and maintain improved glycemic control (15,16). However, a similarly sized randomized trial conducted in Malaysia found no change in HbA_1c at 52 weeks (22). Those contrasting results may be attributable to inherent differences in study population, setting, and intervention. Importantly, our findings confirm those reported by Davis et al (16) in similarly underserved population and suggest that RPM can be leveraged to improve the health of patients facing barriers to care.

Others have reported achieving significant HbA_1c reductions for a predominantly rural and low-income population (16,23). Our work adds to the current literature, however, by showing that patients of academic, free, and community health clinics all derive similar benefit. Neither the paired t-test nor the regression model revealed a significant difference in degree of HbA_1c reduction based on clinic type, suggesting that the RPM intervention offered similar clinical benefit despite differences in resource levels and primary care settings. The ability of a low-resource RPM intervention to improve diabetes care is particularly relevant for rural health systems that are threatened by a lack of resources, geographic barriers, poor transportation infrastructure, and a dwindling primary care workforce.

To our knowledge, this study is also the first to find that a relatively low-resource RPM program can deliver similar improvements in diabetes care for a large, heterogeneous population. Differences in sociodemographic variables did not affect the change in HbA_1c, suggesting that RPM may be similarly beneficial for patients regardless of sex, race, income level, or insurance status.

It is worth noting that the 6- and 12-month cohorts differed with regards to distribution of income level, insurance status, and clinic type. This is likely related to several factors, many of which can be linked to the natural history of the program. This statewide program was introduced to clinics at different time points, owing to standard contracting and onboarding procedures as well as the practical design of this implementation program. At the time of analysis, each of the analyzed clinic sites was still enrolling, and thus many patients eligible for a 6-month follow-up visit had not reached the 12-month point yet. As a result, those patients were not eligible for the 12-month cohort. If a greater proportion of FQHC patients fit this scenario, the observed differences in insurance and income status between the two cohorts could be affected.

Additionally, clinic staffing and patient characteristics affect both rate of enrollment and follow-up completion rates at each site. Intrinsic characteristics at both the patient and clinic levels could affect the likelihood that: a) patients remain engaged, b) patients present to clinic for follow-up testing, and c) sites report follow-up data to the central site for analysis. For example, staff turnover at one clinic could reduce enrollment rates and data reporting. A mixed methods analysis of site-level factors is currently underway to further investigate these trends.

4.2. Limitations

Our study has several limitations. First, it lacks a control group, an inherent limitation in a program designed as a quality improvement initiative. Consequently, we cannot quantify the change in HbA_1c of eligible but not enrolled patients, and thus we cannot estimate the impact of regression to the mean. This phenomenon could partly explain our finding that a higher baseline HbA_1c portends a greater reduction over time, even after adjustment for other demographic and clinical variables (Table 3). As mentioned above, the structure and design of this program were based on a pilot study that showed significantly greater HbA_1c reductions at 6 months for the intervention group than the control group (14). The similarity of program design and findings between the earlier pilot and our current program increase our confidence that we observed a true change in HbA_1c. That being said, it is imperative that future work include an appropriate control group to enable comparison of HbA_1c change in populations with and without access to RPM services.

The sample size in this work, particularly of the 12-month cohort, is small owing to the timeframe during which data was collected. Outcomes data, HbA_1c and BP, were collected for all patients enrolled for a minimum of 6 months as of March 2020, when the COVID-19 pandemic began majorly affecting clinical operations across our state. The program is ongoing and we plan to repeat the analysis with a larger cohort in the future to confirm our findings.

Missing outcome data is another limitation to this work. Of the participants in the 6-month cohort, approximately 60% were missing follow-up HbA_1c data. The most common reasons for missing data were 1) failure to present to clinic for HbA_1c collection and/or BP measurement within the allotted timeframe and 2) withdrawal from the program. It is important to note that we examined clinic-collected outcomes, rather than data that could be patient-transmitted or acquired directly through the RPM device. We predict that patient-level factors, such as income level and employment status, impact the likelihood of data missingness. Lack of transportation and competing priorities are examples of mediators of this effect. Additionally, previous work has shown that females are more likely than males to complete a 3-month diabetes monitoring program, suggesting that demographic factors may play a role in engagement in similar programs (24). Given that we did not include patients in analysis if they were missing 6- or 12-month HbA_1c data, this could skew our results.

To minimize the chance of missingness, central program staff communicated regularly with trained local clinical staff to identify patients and time windows for follow-up HbA_1c data collection and to encourage sites to report back on any available data collected within their health system. We also performed sensitivity analyses to compare demographic characteristics of patients included in our analyses and those excluded due to missing data (Supplemental Table S1 and S2). Data were more likely to be missing at 6 months among patients with higher baseline HbA_1c, younger age, non-black race, and FQHC as primary clinic site. At 12 months, missing data were more common among younger, male, non-black, uninsured, and FQHC patients. Additional research is needed to characterize reasons for missing data and to inform the development of interventions to optimize data collection and reporting.

4.3. Conclusion

While results should be confirmed through comparison of the intervention with a control group, the results here suggest that RPM is an effective, scalable tool for delivering inter-visit care to patients of diverse geographic and socioeconomic backgrounds, affording sustained HbA_1c reductions regardless of clinic type.

Highlights.

• Remote patient monitoring (RPM) is an effective, accessible tool for diabetes care

• Rural and underserved populations achieved significant improvements in HbA1c

• HbA1c reductions were sustained at 6 and 12 months of RPM program participation

• Patients of varying demographics and clinic types achieved similar clinical benefit