Disparities in Offering Enrollment in Remote Patient Monitoring for COVID-19

Abstract

Introduction:

Remote patient monitoring (RPM) programs are increasingly common. There is a risk that inequitable use of RPM will perpetuate existing health care disparities. We conducted a study to determine if enrollment in a COVID-19 RPM program was offered differentially across demographic groups.

Methods:

From March through September 2020, patients with COVID-19 were evaluated within a large academic health system with a standardized care pathway that directed providers to refer the patients for RPM. We conducted a retrospective cohort study to evaluate the effects of social vulnerability and urbanicity of residence on the odds of referral. We estimated vulnerability using the CDC social vulnerability index (SVI) and used logistic regression to determine odds ratios (ORs) for referral based on SVI and urbanicity.

Results:

Of 16,739 patients who had a qualifying health care encounter, 2,946 (17.6%) were referred for RPM. Patients in census tracts with higher social vulnerability were less likely to be referred than those in tracts with lower vulnerability (OR 0.73, 95% confidence interval 0.63–0.84). Patients living in Micropolitan/Large Rural Cities or Small Towns/Small Rural Towns were more likely to be referred than those in Metropolitan/Urban areas. In the full regression model, including both SVI and urbanicity, urbanicity was the strongest predictor of referral, and patients living in Metropolitan/Urban areas were the most likely to be referred.

Conclusions:

We found disparities in who is offered access to remote monitoring despite the use of standardized care pathways. Health systems need to evaluate how they implement RPM programs and care pathways to ensure equitable care delivery.

Introduction

The COVID-19 pandemic forced health systems across the United States to implement novel telehealth programs so that patients could continue to receive care while protecting health care workers and the public.¹ Remote patient monitoring (RPM) systems, developed primarily for chronic disease management,² were adapted during the pandemic to reduce in-person care and facilitate safe early discharge from hospitals for patients with COVID-19 infection.^3–5 The Centers for Medicare and Medicaid Services and the Department of Health and Human Services worked quickly to expand coverage for telehealth services, including RPM, and removed barriers to access, such as requirements that patients have an established, in-person relationship with a provider before engaging in telehealth care.⁶

Reliance on telehealth services can improve patient access to care but may increase health disparities through differential abilities to connect and use telehealth services or through discrepancies in who is offered these services. Marginalized communities already suffer inferior care for routine in-person visits.^7,8 Concern has been raised that limited English skills, digital health literacy, and internet access could amplify these disparities further when care is delivered by a telehealth system, and early evidence has borne out these concerns.^9,10

Whether health care systems differentially offer access to telehealth technologies is less known. Previous research has shown that different patient groups are offered different in-person medical treatment based on gender, race/ethnicity, and geography.^7,8,11 Demographic factors also influence which patients are offered access to online patient portals, and how patients access and utilize telehealth care.^10,12,13 We were unable to find published work exploring disparities in offering other telehealth services, including RPM.

A possible solution to minimizing the potential impact of telehealth on disparities is to embed it in care pathways. Care pathways decrease care delivery variation through electronic standard checklists, note templates, and decision support tools.^14,15 Racial health disparities were decreased by implementing a checklist or eliminated by using a care pathway in rheumatology and oncology, respectively.^16,17 To our knowledge, the use of care pathways in RPM implementation has not been studied.

Early in the COVID-19 pandemic, M Health Fairview (MHFV), a large midwestern health system, implemented an RPM platform (GetWell Network, Inc., Bethesda, MD) for patients presumed to have COVID-19. The RPM was implemented along with a system-wide “care pathway.”⁴ The purpose of this study was to evaluate whether disparities existed in offering enrollment in the RPM platform across patient socioeconomic and geographic characteristics despite care pathway implementation.

Methods

SAMPLE POPULATION

MHFV is a large midwestern health system with 74 clinics and 14 hospitals throughout the state of Minnesota. Coinciding with the start of the COVID-19 pandemic in March 2020, patients in the MHFV health system with COVID-19 symptoms were offered medical evaluation through virtual urgent care platforms, emergency departments, outpatient clinics, and by phone with triage registered nurses or primary care physicians. Those suspected of having COVID-19 based on clinical symptoms, exposure, and/or a positive test for SARS-CoV-2 were offered referral to a COVID-19 RPM program (GetWell Loop) with the purpose of providing ongoing access to medical care while limiting in-person care.

To be included in the study cohort, individuals had to be 18 years of age or older; have a positive SARS CoV-2 polymerase chain reaction (PCR) test at an MHFV location between March 1, 2020, and March 31, 2021; have a current address within the state of Minnesota; have an active chart within the MHFV electronic health records system; and have consented to their electronic health data being used for research purposes.

RPM PLATFORM AND CARE PATHWAY

MHFV has for several years utilized Care Maps, more commonly known as care pathways, to promote delivery of standard, evidence-based care for chronic disease and preventive care. During the pandemic, the system created a COVID-19 Care Map. To maximize the likelihood that patients with diagnosed or suspected COVID-19 were enrolled in the RPM, the referral was embedded in the Care Map standard electronic orders and workflows for outpatient, emergency, and inpatient care. Patients hospitalized for COVID-19 were offered enrollment at the time of discharge.

The RPM was launched and available for referrals on March 23, 2020. A description of the RPM platform has been previously published.⁴ Potential RPM enrollees had to have a cell phone or computer with internet access and comprehend written English. Patients engaged with the RPM via web application (app). Educational content about COVID-19 and symptom questionnaires were pushed to participants with email notifications. Patients could use the app to ask questions of the care team or alert the care team to new concerns. Care team members responded using the RPM interface or by phone depending on the question or concern. Interpreter services were not available during the initial role out of the RPM.

MEASURES

Patient cohort data were extracted from the University of Minnesota's Academic Health Center Information Exchange Clinical Data Repository, which contains medical records for more than 2 million MHFV patients. Data were extracted for the period of March 1, 2020, through March 31, 2021. The categorical variables extracted from the Clinical Data Repository were sex, race, ethnicity, and COVID-19 status, in addition to the continuous variable of age in years. Geographic data were also extracted for each individual: address, census tract-level Federal Information Processing Series (FIPS) code, county, and state.

Due to concerns about the reliability and completeness of patient-specific socioeconomic data in the Clinical Data Repository, including data for racial and ethnic identification, the 2018 CDC social vulnerability index (SVI) for Minnesota was used to estimate individuals' social vulnerability.^18,19 The Minnesota SVI dataset uses 16 U.S. census variables to estimate the relative social vulnerability of each census tract in the state. The SVI is grouped into four domains: (1) Socioeconomic Status; (2) Household Composition and Disability; (3) Minority Status and Language; and (4) Housing and Transportation. Every census tract in Minnesota has a percentile ranking (0–1) for each SVI domain and overall SVI. A ranking of 1 indicates the highest vulnerability in the state of Minnesota.

SVI does not include urbanicity of residence, and previous work has shown that health outcomes can differ based on both vulnerability and urbanicity.²⁰ Due to unexpected findings from preliminary analyses using SVI, location of residence was explored as a confounder on RPM referral rates. Minnesota census tract-level urbanicity data were obtained using the 2010 (2019 revision) US Department of Agriculture Economic Research Service Rural-Urban Commuting Area Codes.²¹ Informed by the Rural Health Research Center at the University of Washington,²² the 21 code scheme was collapsed into 4 categories. Minnesota census tracts were categorized, from most urban to least urban as: Metropolitan/Urban, Micropolitan/Large Rural City, Small Town/Small Rural Town, or Rural/Isolated Small Rural Town. MHFV had 68 clinics and 14 hospitals in Metropolitan/Urban areas, 3 clinics in Micropolitan/Large Rural City areas, 2 clinics in Small Town/Small Rural Towns areas, and 1 clinic in a Rural/Isolated Small Rural Town.

STATISTICAL ANALYSES

The primary outcome of interest, being referred to the RPM, was dichotomous and coded as yes or no. Demographic differences between those referred and not referred to RPM were compared using chi-square tests for categorical variables and Student's t test for the continuous variable.

The association between referral to the RPM and social vulnerability was assessed using a series of logistic regressions adjusted for age and sex. First, the relationship between referral to the RPM and overall SVI was assessed. Subsequent logistic regressions were performed for each of the four SVI domains to assess their individual relationship with RPM referral. SVI and its domains were treated as continuous variables for all analyses, and all analyses were adjusted for age and sex.

Post hoc analyses were conducted using the four levels of urbanicity previously described. A logistic regression was used to assess the relationship between urbanicity and referral to the RPM. Finally, a hierarchical logistic regression was performed that first entered urbanicity and in the second block entered the four SVI domains. Again, these analyses adjusted for age and sex.

Reported for regressions are the regression coefficient (B), odds ratio (OR; which is calculated as eβ, where β is the standardized regression coefficient), and the 95% confidence interval (CI) for the OR.

This study was reviewed and approved under the Expedited category by the University of Minnesota's Institutional Review Board as it involved no greater than minimal risk.

All data management and analyses were conducted using SPSS v27.0 (IBM Corp., Armonk, NY).

Results

The analysis include 16,739 patients; 2,946 were referred to the RPM platform and 13,793 were not. These groups were significantly different based on ethnicity, age, and urbanicity, but did not significantly differ on the basis of sex or race (Table 1).

Table 1.

Study Sample Demographics

	RPM REFERRED, n (%)	RPM NOT REFERRED, n (%)	TOTAL, n (%)
Total	2,946	13,793	16,739
Sex
Female	1,622 (55.1)	7,474 (54.2)	9,096 (54.3)
Male	1,324 (44.9)	6,318 (45.8)	7,642 (45.7)
Race
White	2,242 (76.1)	10,419 (75.5)	12,661 (75.6)
Black or African American	348 (11.8)	1,589 (11.5)	1,937 (11.6)
Unknown	195 (6.6)	1,015 (7.4)	1,210 (7.2)
Asian	116 (3.9)	577 (4.2)	693 (4.1)
American Indian or Alaska Native	29 (1)	111 (0.8)	140 (0.8)
Mixed/Multiple Races	14 (0.5)	51 (0.5)	65 (0.6)
Native Hawaiian or Other Pacific Islander	2 (0.1)	29 (0.2)	31 (0.2)
Other	0 (0)	2 (<0.1)	2 (<0.1)
Ethnicitya
Hispanic/Latino	141 (4.8)	598 (4.3)	739 (4.4)
Not Hispanic/Latino	2,721 (92.4)	12,665 (91.8)	15,386 (91.9)
Unknown	84 (2.9)	530 (3.8)	614 (3.7)
Age: mean (SD)b	51.7 (18.3)	49.1 (19.3)	49.5 (19.2)
Urbanicityb
Metropolitan/Urban	2,594 (88.1)	10,724 (77.7)	13,318 (79.6)
Micropolitan/Large Rural City	139 (4.7)	824 (6.0)	963 (5.8)
Small Town/Small Rural Town	177 (6.0)	1,931 (14.0)	2,108 (12.6)
Rural/Isolated Small Rural Town	36 (1.2)	314 (2.3)	350 (2.1)

Differences in demographic factors, between those who were referred to the RPM and those who did not, were compared using chi-square tests for categorical data and Student's t test for continuous data.

^a Significantly different at p < 0.05.

^b Significantly different at p < 0.001.

RPM, remote patient monitoring; SD, standard deviation.

There was a significant association between sociocultural vulnerability as measured by the SVI and referral for remote monitoring (OR = 0.73, 95% CI = 0.63–0.84; χ² = 20.2, df = 1, p < 0.001) (Table 2). The likelihood of being referred to RPM decreased as social vulnerability increased. Table 2 shows the results of logistic regressions predicting RPM referral with the four domains of social vulnerability as predictors.

Table 2.

Association of Remote Patient Monitoring Referral with Social Vulnerability Index and Its Four Domains

	B	ODDS RATIO (95% CI)	p
SVI	−0.319	0.73 (0.63–0.84)	<0.001
Socioeconomic Status	−0.186	0.83 (0.67–1.04)	0.099
Household Composition and Disability	−0.031	0.97 (0.81–1.16)	0.738
Minority Status and Language	0.146	1.16 (1.01–1.33)	0.041
Housing Type and Transportation	−0.200	0.82 (0.68–0.98)	0.034

For each SVI, odds ratios represent change in odds of referral to RPM as social vulnerability increases. Analyses adjusted for age and sex.

CI, confidence interval; SVI, social vulnerability index.

Significant associations were found between the SVI domains Minority Status and Language and Housing and Transportation with referral to the RPM platform. Increasing SVI Minority Status and Language was associated with a 16% increase in referral odds (OR = 1.16, 95% CI = 1.01–1.33; χ² = 4.1, df = 1, p = 0.041), while increasing Housing and Transportation was associated with an 18% decrease in referral odds (OR = 0.82, 95% CI = 0.68–0.98, χ² = 4.5, df = 1, p = 0.034).

There were significant differences in RPM referral depending on the urbanicity of communities, in which individuals lived (χ² = 175.06, df = 3. p < 0.001). Table 3 shows the results of this logistic regression. Compared to individuals living in Metropolitan/Urban communities, those living in Micropolitan/Large Rural Cities were about two times more likely to be referred for RPM and those in Small Town/Small Rural Towns were 1.5 times more likely to be referred. There were no significant differences in the likelihood to be referred to RPM between those living in Rural/Isolated Small Rural Town communities and those in Metropolitan/Urban communities.

Table 3.

Association of Remote Patient Monitoring Referral with Urbanicity

URBANICITYa	B	ODDS RATIO (95% CI)	p
Micropolitan/Large rural city	0.75	2.12 (1.50–3.00)	<0.001
Small town/Small rural town	0.40	1.50 (1.01–2.21)	0.043
Rural/Isolated small rural town	−0.22	0.81 (0.55–1.18	0.261

^a Metropolitan/Urban is the reference group. Analyses adjusted for Age and Sex.

Table 4 shows the results of the fully adjusted logistic regression model including age, sex, urbanicity, and social vulnerability. Once adjusted for SVI, the effect of urbanicity on odds of RPM referral reversed from the unadjusted model. In this model, patients living in Metropolitan/Urban communities were the most likely to be referred, and SVI was no longer a significant predictor of referral. Patients living in Small Town/Small Rural Town or Rural/Isolated Rural communities were less than half as likely as those living in Metropolitan/Urban communities to be referred for RPM. When the four SVI domains were included as independent variables in place of SVI, the effect of urbanicity was unchanged and the only SVI domain significantly associated with referral was Minority Status and Language (B = −0.478, OR = 0.62, 95% CI = 0.53–0.73).

Table 4.

Full Model Predicting Remote Patient Monitoring Referral

VARIABLE	B	ODDS RATIO (95% CI)	p
Age	0.007	1.01 (1.00–1.01)	<0.001
Sex	0.036	1.04 (0.96–1.12)	0.380
Urbanicitya
Micropolitan/Large rural city	−0.332	0.72 (0.60–0.87)	<0.001
Small town/Small rural town	−0.953	0.39 (0.33–0.45)	<0.001
Rural/Isolated small rural town	−0.735	0.48 (0.34–0.68)	<0.001
SVI	−0.073	0.93 (0.81–1.07)	0.321

^a Metropolitan/Urban is the reference category.

The interaction between urbanicity and each SVI theme was explored using logistic regression controlling for age and sex. The results are presented in Table 5. The equation was significant (χ² = 260.25, df = 5. p < 0.001). However, the only theme that significantly contributed to predicting referral was the interaction between urbanicity and Minority Status and Language (B = −0.47, Wald = 45.06, df = 1, p < 0.001). The interaction was explored by computing the effect of Minority Status and Language at each level of urbanicity. The results indicate significant effects of Minority Status and Language for participants living in Metropolitan/Urban areas (OR = 0.67, 95% CI = 0.58–0.78, p < 0.001). Minority Status and language were not significantly related to odds of being referred to RPM within any other levels of urbanicity.

Table 5.

Association of Remote Patient Monitoring Referral with Interaction Between Urbanicity and Each Social Vulnerability Index Theme

SVI THEME INTERACTION TERMa	B	ODDS RATIO (95% CI)	p
Socioeconomic status	0.09	1.10 (0.93–1.30)	0.287
Household composition and disability	−0.003	1.00 (0.86–1.15)	0.972
Minority status and language	−0.47	0.62 (0.54–0.72)	<0.001
Housing type and transportation	−0.04	0.96 (0.84–1.11)	0.592

^a Analyses adjusted for Age and Sex. Each row represents the interaction between urbanicity and that theme.

Discussion

Patients of a large academic health system with COVID-19 were least likely to be referred to an RPM if they lived in the rural areas of the state. Location of residence was a more powerful predictor of referral compared to more traditional mediators of social vulnerability.

We were surprised that the SVI domain Minority Status and Language was positively associated with referral despite patients having to be able to read and converse in English to be referred to the RPM program. Since sociocultural vulnerability was based on the characteristics of the neighborhood in which an individual lives and not characteristics of the individual, we explored whether geographic area of residence might be the underlying cause of this unexpected association. Controlling for urbanicity did reverse the direction of the effect of minority status and language on referral, likely due to greater concentrations of minority groups and languages in more urban areas.

We hypothesize that the health care system's systematic implementation of a standardized, decision-support COVID-19 care pathway minimized the effect of provider bias on offering access to the RPM. A possible explanation for the geographic disparity in referral rates is that patients living in nonmetropolitan areas were more likely to be seen at nonmetropolitan care facilities. Some of these clinics and hospitals joined the health system in 2017 and might be less familiar with the system's care pathways and potentially less adherent. It is also possible that some providers at these more remote sites were unaware of changes to RPM coverage policies during the public health emergency that allowed for referral even without patients having a preexisting relationship with the health system. Due to limitations in our dataset, we were unable to explore such an association or how closely providers adhered to the care pathway in the present study.

LIMITATIONS

We used census-tract level data as an estimate of individuals' exposure to social vulnerability.²³ Different results may have been obtained using patient-level data, but community-level data allow health systems a practical method of testing for disparities in care delivery.

Due to the software only being available in English, COVID-19 patients who could not speak and read English were systematically excluded from referral. We were therefore unable to evaluate RPM referral for these populations. It is possible that a study of an RPM with more language options would uncover different biases in RPM referral. In addition, because Minority Status and Language is one of the four domains of the SVI, excluding non-English speaking patients from our RPM program may have diminished the measured effect of SVI on referral.

Our analyses of the effects of urbanicity on RPM referral were not prespecified at the start of our study. In light of the surprising initial results, we believe that the post hoc analyses we conducted were necessary to avoid drawing erroneous conclusions from our data, but it does increase the risk of our results being due to chance and multiple comparisons (type 1 error).

A significant limitation of the present study is the overall low adherence to the health system's care pathway that directed providers to refer eligible patients with known or suspected COVID-19 to the RPM. Only 17.6% of patients were referred, which is substantially lower than outcomes for other order set-based care pathways in the system. Typical utilization rates in our health system vary from 43% on the low end for controlled substance agreements for patients on chronic opioids to 92% on the high end for use of well-child visit order sets. The rapid rollout and frequent updating of the COVID-19 care pathway and the relative novelty of RPM within our system likely played roles with the low adherence. It is possible that our findings would be different if we evaluated referrals to a more established RPM program and are most generalizable to similar novel programs.

Conclusion

Although standardized care pathways have been suggested as a means of reducing disparities in health care delivery, we found that their implementation may be insufficient to eliminate these disparities in RPM programs. RPM is a growing part of health care. Ensuring those services are accessible and equitably offered to patients across all domains of social vulnerability, including geography, is essential. Health systems must intentionally consider disparities in the design, implementation, and evaluation of care pathways and RPM programs or risk reproducing existing in-person care inequities.

Authors' Contributions

J.A.T.: conceptualization, methodology, writing—original draft, and writing—review and editing; D.H.: conceptualization, methodology, resources, formal analysis, data curation, writing—original draft, and writing—review and editing; R.N.K.: conceptualization, writing—original draft, and writing—review and editing; M.H.M.: methodology, software, formal analysis, and writing—original draft; P.A.: conceptualization, methodology, resources, writing—original draft, writing—review and editing, supervision, and funding acquisition.

Disclosure Statement

The authors have no conflicts of interest to declare.

Funding Information

This research was partially funded by the UMN Campus Public Health Officer's Collaborative Outcomes:Visionary Innovation & Discovery (CO:VID) grants program (principal investigator [PI]: P.A.) and a UMN Medical School COVID Rapid Response Grant (PI: P.A.). This research was supported by resources provided by the National Institutes of Health's National Center for Advancing Translational Sciences, grant UL1TR002494. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health's National Center for Advancing Translational Sciences.