Inequities in Telehealth Use Associated with Payer Type During the COVID-19 Pandemic

Introduction

Arkansas was the only southern state to expand Medicaid under the Patient Protection and Affordable Care Act at its first opportunity on January 1, 2014. The state took a unique premium assistance approach to Medicaid expansion that was facilitated by the approval of a Section 1115 demonstration waiver by the Centers for Medicare and Medicaid Services (CMS). Medicaid expansion coverage in Arkansas for most enrollees is provided by premium assistance to individual qualified health plans (QHPs) offered on the Health Insurance Marketplace. Arkansas adults meeting Medicaid eligibility requirements before coverage expansion continue to be enrolled in the Primary Care Case Management (PCCM) program managed by the state.

There are notable differences between QHP and PCCM policies, and one is regarding telehealth benefits. QHPs historically have reimbursed for telehealth services when the originating site, that is, where the patient is located, is the patient’s home. By contrast, the PCCM has required the originating site to be a clinic where the patient is receiving services. This policy was amended, however, on March 18, 2020, when the governor of Arkansas declared a public health crisis owing to the COVID-19 pandemic, thereby permitting telehealth services to be delivered to PCCM patients and providers to be reimbursed when the originating site was the patient’s home.^1,2

Since the beginning of the pandemic, telehealth has become a critical pathway to deliver care to patients while adhering to social distancing practices and reducing the risk of COVID-19 transmission.³ Although the variation in telehealth use by enrollees and providers in private plans and networks during the pandemic has been characterized,^4,5 there is limited description of telehealth access barriers among Medicaid beneficiaries over the same period.⁶

This study assessed the association between payer type (PCCM vs. QHP) and telehealth use and tested whether the aforementioned change in Medicaid telehealth reimbursement policy led to equitable access across payer types.

First, telehealth utilization in the first 3 months after the public health crisis declaration was examined by payer type. Both populations under study were matched on sociodemographic, geographic, and clinical profiles. The existence of established patient–provider relationships, which has been reported to be a key factor in the level of telehealth use during the pandemic, was included as an additional control variable.⁷ Two specific domains of delivery that were most frequently used, synchronous and mobile health (m-health) visits, were subsequently examined.

A synchronous visit is a live, two-way (patient and provider) interactive delivery of health care. Traditionally, synchronous visit reimbursement was restricted to video visits; however, this restriction was relaxed during the COVID-19 pandemic and some synchronous services were allowed to be delivered by telephone appointments as well.⁸ m-Health refers to health care and public health practices supported by mobile communication devices, such as cell phones and tablet computers.⁹ Examples of m-health include messages that promote healthy behavior through text messages and application software (apps). Mental or behavioral health versus physical health care were also studied separately to account for the higher mental or behavioral health needs of PCCM enrollees compared with QHP enrollees.¹⁰

Second, a difference-in-differences framework was used to test for a differential increase in telehealth utilization from the pre-pandemic baseline among PCCM enrollees compared with QHP enrollees.^11–13

Previous work has documented that, even before the pandemic, Medicaid enrollees incurred disproportionately fewer, and delayed access to, office visits and telehealth services compared with enrollees in private plans.^10,14 At the beginning of the pandemic, health care provider hours and staffing were reduced at hospitals and clinics.^2,15 This study tested a hypothesis that, even with a change in Medicaid reimbursement policy during the public health emergency, there was an access difference between PCCM and QHP enrollees for telehealth services during the pandemic, which consequently created an unmet health care demand and resulted in missed care opportunities among PCCM enrollees during this vulnerable period.

Methods

In this retrospective cohort study, enrollees from each program (PCCM and QHP) were selected based on having established tenure in the program with sufficient records in the Arkansas All-Payer Claims Database (APCD) during the study period. Among these two populations, matched pairs were selected based on the probability of being a PCCM enrollee, which was derived from a logistic regression model on sociodemographic and geographic variables, medical fragility, and prior mental or behavioral health care utilization. Generalized linear mixed models were fitted to estimate payer-type effects. This study was approved and determined as nonhuman subject research by University of Arkansas for Medical Sciences Institutional Review Board (FWA00001119).

DATA SOURCES

The APCD is a component of the Arkansas Healthcare Transparency Initiative administered by the Arkansas Center for Health Improvement and governed by the Arkansas Insurance Department. The APCD includes enrollment records and hospital, emergency department, outpatient, office-based, dental, and pharmacy claims for Arkansans with health coverage through nonself-insured private payers, Medicaid, and Medicare. Data for this study were obtained from June 2019 through June 2020. The median household income and proportion of residents with broadband internet access were obtained from the 2019 American Community Survey 5-Year Estimates^16,17 at the 5-digit ZIP codes of patient residence. These ZIP code-level variables were included to serve as proxy indicators for estimates of access to synchronous services.

SAMPLE SELECTION AND STUDY POPULATION

There were 64,477 low-income adult parents or caregivers (Medicaid aid categories 20 or 25) aged 18–64 years who had at least one instance of enrollment in the PCCM at the beginning of the pandemic (March–June 2020). Those who also had a record in a QHP or Medicaid Aid Category 61 or 65, which are for pregnant women, within the June 2019 to June 2020 observation period were removed (n = 16,860). Other exclusion criteria were a dual eligibility with Medicare and Medicaid coverage (n = 942), not having at least 180 days of continuous enrollment between June 2019 and March 2020 period and at least 60 days of continuous enrollment between March 2020 and June 2020 (n = 11,384), and missing values for race, county, income, broadband accessibility, or insurance market region (n = 4,284).

The result was an analytic population of 31,007 individuals exclusively enrolled in the PCCM who could be followed from June 1, 2019, through June 30, 2020. We then generated a comparison sample using the same exclusion criteria but composed of low-income adults enrolled in QHPs through premium assistance over the same time period. This resulted in a sample of 128,003 QHP enrollees.

MEASURES

Outcome measures include the number of enrollees who completed any type of telehealth claim submitted for reimbursement during the period of March 18, 2020 to June 30, 2020. Telehealth utilization was captured through insurance claims records with an algorithm combining Current Procedural Terminology (CPT), procedure modifiers, place of service, and revenue codes. This algorithm (detailed in Appendix A1) was developed through a review of reimbursement documents from Arkansas Medicaid, Arkansas Blue Cross and Blue Shield (the largest QHP in the state), and CMS.^8,18–20 Two specific domains of delivery, synchronous and m-health visits, were subsequently examined. The definitions of synchronous telehealth and m-health can vary and are not always mutually exclusive.^21,22

This study followed the framework presented by the Center for Connected Health Policy and defined synchronous as live, two-way interaction between a person and a provider using audiovisual telecommunication technology and m-health as the use of mobile communication devices to provide health care and education and promote better public health practices.²¹ As the telehealth rules were relaxed both by commercial and public insurance carriers, some synchronous services were allowed to be audio only, which introduces variation in the modality in studying this outcome.¹⁸

Synchronous telehealth utilization was identified through procedure codes and modifiers, revenue codes, and place of service codes (Appendix Tables A4 and A5), whereas m-health utilization was identified through procedure codes (see Appendix A1 for the full algorithm). Finally, telehealth services of any domain were separated by mental or behavioral health versus physical health to account for differential needs by health conditions.⁷

Health care utilization for mental or behavioral disorders was identified through Clinical Classifications Software (CCS). CCS is a tool developed by the Agency for Healthcare Research and Quality to aggregate ICD-9/10-CM codes into detailed, clinically meaningful categories. The list of CCS conditions used are listed in Appendix Table A3.²³

Covariates were sociodemographic factors (age, sex, race/ethnicity, median household income, and broadband availability), rural and urban classification area, insurance regions, underlying medical fragility using the Charlson Comorbidity Index (CCI),²⁴ preexisting mental or behavioral health conditions (yes/no), and preexisting primary care physician–patient relationships (yes/no). These covariates were all identified or constructed from insurance enrollments and claims records from June 1, 2019 to March 17, 2020, or linked at the 5-digit ZIP code level from the 2019 American Community Survey 5-year summary data.

STATISTICAL ANALYSIS

Randomly assigning adults to PCCM or QHP coverage was not feasible, so this study used a propensity score match study design to assess the association between telehealth utilization and the payer type. In a well-balanced match, where PCCM and QHP enrollees are systematically very similar (across sociodemographic and geographic characteristics and underlying medical needs), observed statistical outcome differences between two payer types highlight systematic telehealth access differences.

At the first analytic stage, a one-to-one matching strategy was implemented using a combination of exact category and greedy matching propensity scores with caliper = 0.2.²⁵ To be matched, enrollees from both groups had to be identical in the following categories: gender, race/ethnicity, income group, CCI group, age category, and broadband internet access. In addition, using SAS (SAS Institute, Cary, NC) proc psmatch, both groups were matched by propensity score. In total, 30,491 PCCM enrollees (98.3%) were matched one-to-one with QHP enrollees.

At the second analytic stage, generalized linear integrated mixed models were fitted. An indicator of each PCCM and QHP-matched dyad was included as a random variable in the mixed models. A binomial distribution with a logit-link function was used for the binary outcome of having at least one telehealth visit, or not. Differences in PCCM and QHP telehealth utilization were determined using adjusted odds ratios (aORs), 95% confidence interval (CI), and p-values (α = 0.05). Similarly, outcomes of having at least one synchronous visit, m-health visit, mental or behavioral health visit, and physical health visit were fitted.

Finally, to account for baseline pre-pandemic telehealth utilization differences in both groups, we estimated a difference-in-differences model to determine whether having a private insurance coverage was associated with a higher rate of increases in telehealth utilization postpublic emergency declaration. To assess the baseline telehealth use, we reviewed APCD claims records from September 1, 2019, through March 17, 2020, for the PCCM and QHP-matched populations. We then compared changes in those trends after the onset of the public health crisis, March 18, 2020 through June 30, 2020, for both groups.

Results

Our sample included 30,491 PCCM and matched QHP enrollee dyads. Table 1 presents demographic characteristics for our sample of continuously enrolled PCCM and QHP enrollees postmatch (Table 1). Before matching, study groups were significantly different in terms of baseline sociodemographic, geographic, and clinical characteristics (Appendix A1). In particular, QHP enrollees were older and more likely to identify as non-Hispanic white than PCCM enrollees. Although 61% of QHP enrollees were women, an even higher percentage of PCCM enrollees were women (85%). For the propensity score match comparison, standardized differences are presented. Variance ratio statistics are all between the accepted threshold of 0.5 to 2.0 and included in Appendix Table A2. Figure 1 presents the prevalence of telehealth use by domain and type of care (Fig. 1).

Table 1. Demographic Characteristics Following Propensity Score Matching

VARIABLES	CATEGORY	TREATMENT (N = 30,491)	CONTROL (N = 30,491)	SD
		n (%)	n (%)
Gender	Male	4,514 (14.8)	4,514 (14.8)	0
	Female	25,977 (85.2)	25,977 (85.2)
Race	White	19,225 (63.1)	19,225 (63.1)	0
	Black	8,633 (28.3)	8,633 (28.3)
	Hispanic	1,140 (3.7)	1,140 (3.7)
	Other	1,493 (4.9)	1,493 (4.9)
RUCA	Urban	18,133 (59.5)	17,843 (58.5)	0.04
	Large	5,916 (19.4)	6,102 (20.0)
	Small	4,678 (15.3)	4,792 (15.7)
	Isolation	1,764 (5.8)	1,754 (5.8)
Income	<100% FPL	1,573 (5.2)	1,573 (5.2)	0
	100–150% FPL	17,003 (55.8)	17,003 (55.8)
	150–200% FPL	9,139 (30.0)	9,139 (30.0)
	≥200% FPL	2,776 (9.1)	2,776 (9.1)
Insurance region	Central	9,527 (31.2)	8,950 (29.4)	0.06
	Northern	6,398 (21.0)	6,429 (21.1)
	Northwest	4,020 (13.2)	4,015 (13.2)
	South	1,947 (6.4)	2,013 (6.6)
	Southeast	2,972 (9.7)	3,045 (10.0)
	Southwest	2,633 (8.6)	2,879 (9.4)
	West Central	2,994 (9.8)	3,160 (10.4)
Charlson Comorbidity Index category	0	24,792 (81.3)	24,792 (81.3)	0
	1–2	5,120 (16.8)	5,120 (16.8)
	3–4	434 (1.4)	434 (1.4)
	>4	145 (0.5)	145 (0.5)
Age category, years	<30	13,331 (43.7)	13,331 (43.7)	0
	30–40	12,048 (39.5)	12,048 (39.5)
	40–50	4,225 (13.9)	4,225 (13.9)
	50–60	837 (2.7)	837 (2.7)
	≥60	50 (0.2)	50 (0.2)
Broadband internet access percentage	<50%	268 (0.9)	239 (0.8)	0.03
	50–70%	8,133 (26.7)	7,884 (25.9)
	70–80%	9,446 (31.0)	9,502 (31.2)
	80–90%	10,990 (36.0)	11,196 (36.7)
	≥90%	1,654 (5.4)	1,670 (5.5)
Mental health disorders	Yes	15,124 (49.6)	15,124 (49.6)	0
	No	15,367 (50.4)	15,367 (50.4)

Findings presented in Table 2 indicate that an estimated 35% more QHP enrollees had at least one telehealth use compared with PCCM enrollees (aOR 1.35; 95% CI: 1.29–1.42). In addition, compared with those younger than 30, those aged 30–40 years, 40–50 years, and 50–60 years had 1.17 (95% CI: 1.12–1.24), 1.26 (95% CI: 1.17–1.35), and 1.16 (95% CI: 1.01–1.34) higher adjusted odds, respectively, of having any telehealth use. Female enrollees (aOR = 1.41; 95% CI: 1.31–1.52), attribution to primary care provider (aOR = 2.24; 95% CI: 2.14–2.36), residence in isolated area (aOR = 1.14; 95% CI: 1.03–1.27) were also associated with increased telehealth utilization compared with reference categories (e.g., male enrollees).

Table 2. Multivariable Relationship Between Insurance Type and Telehealth Use During First 3 Months of the COVID-19 Pandemic, by Domain

	ADJUSTED ODDS RATIO
	ANY TELEHEALTH	DOMAIN
		SYNCHRONOUS	m-HEALTH
Insurance (ref: PCCM)
QHP	1.35 (1.29–1.42)***	1.31 (1.25–1.37)***	5.91 (4.25–8.21)***
Attributed to PCP (ref: no)
Yes	2.24 (2.14–2.36)***	2.26 (2.15–2.37)***	1.62 (1.25–2.10)***
Gender (ref: male)
Female	1.41 (1.31–1.52)***	1.41 (1.31–1.52)***	0.98 (0.70–1.37)
Race (ref: White)
Black	0.94 (0.88–0.99)**	0.94 (0.88–0.99)**	0.80 (0.59–1.10)
Hispanic	0.90 (0.79–1.04)	0.91 (0.79–1.04)	0.91 (0.45–1.84)
Other	0.92 (0.83–1.03)	0.91 (0.82–1.02)	1.49 (0.95–2.34)*
RUCA (ref: urban)
Isolated	1.14 (1.03–1.27)**	1.16 (1.04–1.29)**	0.92 (0.53–1.59)
Large rural	0.87 (0.82–0.93)***	0.87 (0.82–0.94)***	1.10 (0.78–1.53)
Small rural	0.90 (0.83–0.97)**	0.90 (0.83–0.97)**	0.98 (0.68–1.41)
Income percent of FPL (ref: ≥200%)
<100%	1.00 (0.86–1.15)	1.01 (0.87–1.16)	1.23 (0.59–2.55)
100–150%	1.03 (0.94–1.13)	1.03 (0.93–1.13)	1.15 (0.68–1.95)
150–200%	0.95 (0.87–1.04)	0.95 (0.87–1.04)	0.97 (0.58–1.61)
Insurance region (ref: central)
Northeast	0.90 (0.84–0.97)**	0.88 (0.82–0.94)***	2.18 (1.50–3.15)***
Northwest	0.92 (0.85–0.99)**	0.92 (0.85–0.99)**	1.20 (0.77–1.87)
SC	0.55 (0.50–0.61)***	0.56 (0.50–0.62)***	0.76 (0.39–1.45)
Southeast	0.75 (0.67–0.82)***	0.73 (0.66–0.81)***	1.40 (0.83–2.37)
Southwest	0.82 (0.75–0.90)***	0.81 (0.74–0.89)***	1.37 (0.82–2.30)
WC	0.71 (0.65–0.78)***	0.68 (0.62–0.74)***	3.23 (2.19–4.76)***
CCI group (ref: 0)
1–2	1.64 (1.56–1.73)***	1.63 (1.54–1.72)***	1.83 (1.42–2.36)***
3–4	2.19 (1.89–2.53)***	2.16 (1.86–2.50)***	3.12 (1.84–5.29)***
>4	2.56 (2.00–3.28)***	2.47 (1.93–3.15)***	2.18 (0.77–6.15)
Age category, years (ref: <30)
30–40	1.17 (1.12–1.24)***	1.17 (1.11–1.23)***	1.15 (0.89–1.48)
40–50	1.26 (1.17–1.35)***	1.25 (1.17–1.35)***	0.95 (0.66–1.36)
50–60	1.16 (1.01–1.34)**	1.13 (0.97–1.30)	1.49 (0.83–2.67)
≥60	1.23 (0.68–2.21)	0.95 (0.50–1.78)	4.62 (1.10–19.3)**
Broadband internet access (ref: ≥90%)
<50%a	0.79 (0.59–1.05)*	0.79 (0.60–1.06)	0.43 (0.08–2.43)
50–70%	0.89 (0.80–1.00)**	0.89 (0.79–1.00)**	0.95 (0.53–1.70)
70–80%	0.87 (0.78–0.97)**	0.88 (0.78–0.98)**	0.69 (0.39–1.22)
80–90%	0.92 (0.83–1.02)	0.92 (0.83–1.02)**	0.96 (0.57–1.62)
Mental health (ref: no)
Yes	4.46 (4.22–4.72)***	4.46 (4.22–4.72)***	3.06 (2.26–4.15)***

Higher needs for medical care were also associated with a higher rate of telehealth use—specifically, the adjusted odds of using telehealth at least once for those with CCI scores between 1 and 2, 3 and 4, and greater than 3 were 1.64 times (95% CI: 1.56–1.73), 2.19 times (95% CI: 1.89–2.53), and 2.56 times (95% CI: 1.89–2.53) higher, respectively, than those with CCI = 0. Existing mental health conditions were independently associated with higher rates of completed telehealth visits (aOR = 4.46, 95% CI: 4.22–4.72).

Conversely, Black race (aOR = 0.94; 95% CI: 0.88–0.99), rural areas (aOR = 0.87; 95% CI: 0.82–0.93 for large rural areas and aOR = 0.90; 95% CI: 0.83–0.97 for small rural areas), and insurance regions other than Central were associated with lower rates of completed telehealth visits compared with reference categories. Compared with individuals who live in ZIP codes with more than 90% having broadband internet access, those in ZIP codes with 50–70% and 70–80% access rates had 0.87 (95% CI: 0.78–0.97) and 0.89 (95% CI: 0.80–1.00) adjusted odds, respectively, of having at least one telehealth use. Less than 1% of individuals lived in ZIP code geographic regions where <50% of the residents in that area had broadband internet access.

Separating utilization by domain (synchronous and m-health) and by type of care (mental/behavioral health and physical health) did not change the primary finding. A higher proportion of QHP enrollees had at least one synchronous (aOR = 1.31; 95% CI: 1.25–1.37) and m-health (aOR = 5.91; 95% CI: 4.25–8.2) use than PCCM enrollees (Table 2), as well as mental/behavioral health telehealth session (aOR = 1.13; 95% CI: 1.07–1.19; Appendix Table A1). In the multivariable logistic regression model for mental or behavioral telehealth utilization, prior mental or behavioral utilization was removed from covariates because all patients who received at least one mental or behavioral telehealth service had had a prior mental or behavioral health care visit.

Finally, the pre-post analysis showed that in 6 months leading up to the public health crisis declaration, 356 PCCM (1.2%) and 325 QHP (1.1%) enrollees received at least one telehealth service. Compared with the baseline, the proportion of enrollees who used telehealth services increased significantly more with the relative risk ratio of 1.76 (p < 0.001) for the QHP group compared with the PCCM group in the first 3 months following the public health crisis declaration. Figure 2 provides a graphic presentation of changes in telehealth utilization at the beginning of the pandemic for the QHP and PCCM population (Fig. 2). The full model specification of the difference-in-differences analysis is presented in Appendix A1.

Discussion

The findings of this study demonstrate that care for Medicaid beneficiaries administered through private QHPs included significantly higher telehealth utilization during the first 3 months of the COVID-19 pandemic, as well as a higher rate of increase compared with the pre-pandemic level, than beneficiaries enrolled in the state-administered PCCM program. In the pre-pandemic period, the exceedingly low telehealth utilization among both PCCM and QHP enrollees suggests that the originating site restriction by the PCCM program had a limited effect, if any, on telehealth utilization outside the context of the pandemic and public health emergency.

However, the rate of telehealth utilization among PCCM enrollees during the first 3 months of the pandemic would have likely been considerably smaller than that observed without the suspension of this originating site restriction. As the decision to offer telehealth services during the COVID-19 outbreak was firmly tied to adequate reimbursement, equipment investment cost, and patient demand, clinics that served the most vulnerable patients covered under the PCCM (low-income parents and caretakers) may have been less likely to have had the incentive or capability to establish a telehealth service line.

Our findings are consistent with prior studies showing age, race, and income disparities in usage of telehealth services during the pandemic.^26–29 The finding that Black enrollees and those who live in areas with low internet access were less likely to participate in telehealth during the pandemic under either the PCCM or QHPs is a cause for concern. The pandemic differentially affected low-income communities and communities of color as represented by greater reduction in life-expectancy, the ultimate health outcome.^30,31 These findings likely suggest not only individual barriers to receiving appropriate care but also systemic issues related to deployment, availability, and access for these groups and health care marketing that target specific populations.²⁹

Of interest, and also consistent with findings by Patel et al.,⁷ enrollees in both groups who had been seen by primary care providers in the 6 months leading up to the public health emergency declaration were significantly more likely to have a telehealth visit. In addition, another previous work noted that new availability of telemedicine visits through patients’ primary care providers may have changed the patient satisfaction of telehealth,³² which may in turn, have increased the telehealth uptake. The mediation effect of primary care provider attribution is outside the scope of this study. In our future study, we will focus on those who had existing patient–provider relationships at the beginning of the pandemic and further examine driving factors in the differential uptake of telehealth services.

There are several potential limitations when interpreting our findings. First, telehealth utilization rates were captured through administrative data using an algorithm developed in a combination of CPT, place of service, and ICD 9-/10-CM and is subject to clerical errors. Second, audio-only (telephone) visits were permitted for some synchronous telehealth services. Although the choice of a telephone and video visit is important in studying the digital divide,³³ we were unable to ascertain the modality from the claims record.^30,34,35 Finally, to fully understand the effect of easing of originating site restrictions, future investigation of claims for site of service to identify the originating site for the telehealth service is warranted.

Conclusions

This study demonstrates that among equally financially vulnerable patient populations, premium assistance-eligible individuals enrolled in commercial plans with premium subsidies utilized telehealth at a higher rate than a matched group of enrollees in a traditional PCCM Medicaid plan. The acceleration of telehealth utilization during the pandemic offers promise for future strategies to address rural health needs, specialty access, and chronic disease management. However, systematic assessment of opportunities for enhanced telehealth support and identification of barriers to success are necessary to avoid accentuating the digital divide for low-income communities, communities of color, and those enrolled in public insurance.

Authors’ Contributions

K.N.L. conceptualized and designed the study, drafted the initial version of the article, and reviewed and revised the article. A.G. conceptualized and designed the methodology, and reviewed and revised the article. J.C.W conceptualized and administered the project, and reviewed and revised the article. E.T. and J.L. conducted data curation and investigation, designed analytic methodology and performed the final analyses, and reviewed and revised the article. J.T. conceptualized the design, provided senior supervision and resources, and reviewed and revised the article. All authors approved the final version of the article as submitted and agree to be accountable for all aspects of the work.

Disclosure Statement

No competing financial interests exist.

Funding Information

Supported by The Commonwealth Fund, a national, private foundation based in New York City that supports independent research on healthcare issues and makes grants to improve healthcare practice and policy. The views presented here are those of the author and not necessarily those of The Commonwealth Fund, its directors, officers, or staff.

REFERENCES

• 1. Arkansas Department of Human Services. Public Health Emergency Rule Suspensions. 2020. Available at https://humanservices.arkansas.gov/covid-19/dhs-response-to-covid-19/updates-for-providers/public-health-emergency-rule-suspensions/ (last accessed June 21, 2021). Google Scholar
• 2. Lewis C, Seervai S, Shah T, Abrams M, Zephyrin L. Primary care and the COVID-19 pandemic. Commonwealth Fund. 2020. DOI: 10.26099/73k0-a831. Crossref, Google Scholar
• 3. Koonin LM, Hoots B, Tsang CA, et al. Trends in the use of telehealth during the emergence of the COVID-19 pandemic—United States, January–March 2020. MMWR Morb Mortal Wkly Rep 2020;69:1595–1599. Crossref, Medline, Google Scholar
• 4. Patel SY, Mehrotra A, Huskamp HA, et al. Trends in outpatient care delivery and telemedicine during the COVID-19 pandemic in the US. JAMA Intern Med 2020;181:388–391. Crossref, Google Scholar
• 5. Cantor JH, McBain RK, Pera MF, Bravata DM, Whaley CM. Who is (and is not) receiving telemedicine care during the COVID-19 pandemic. Am J Prev Med 2021;61:434–438. Crossref, Medline, Google Scholar
• 6. Nouri S, Khoong EC, Lyles CR, Karliner L. Addressing equity in telemedicine for chronic disease management during the Covid-19 pandemic. NEJM Catal Innov Care Deliv. 2020. DOI: 10.1056/CAT.20.0123. Crossref, Google Scholar
• 7. Fischer SH, Uscher-Pines L, Roth E, Breslau J. The transition to telehealth during the first months of the COVID-19 pandemic: Evidence from a national sample of patients. J Gen Intern Med 2021;36:849–851. Crossref, Medline, Google Scholar
• 8. Center for Medicare and Medicaid Services. List of Telehealth Services for Calendar Year 2021, as of March 30, 2021. Available at https://www.cms.gov/Medicare/Medicare-General-Information/Telehealth/Telehealth-Codes (last accessed June 30, 2021). Google Scholar
• 9. The Center for Connected Health Policy. What is telehealth? Available at https://www.cchpca.org/what-is-telehealth/ (last accessed December 28, 2021). Google Scholar
• 10. Goudie A, Martin B, Li C, et al. Higher rates of preventive health care with commercial insurance compared with medicaid: Findings from the arkansas health care independence “private option” program. Med Care 2020;58:120–127. Crossref, Medline, Google Scholar
• 11. Angrist J, Pischke J. Mostly harmless econometrics. Princeton, NJ: Princeton University Press, 2008. Crossref, Google Scholar
• 12. Warton M, Parker M, Karter A. How D-I-D you do that? Basic difference-in-differences models in SAS^®. In: Western Users of SAS® Software. San Francisco, CA: SAS Institute, 2016. Google Scholar
• 13. Basu P, Small D. Constructing a more closely matched control group in a difference-in-differences analysis: Its effect on history. Observ Stud 2020;6:103–130. Crossref, Google Scholar
• 14. Park J, Erikson C, Han X, Iyer P. Are state telehealth policies associated with the use of telehealth services among underserved populations? Health Aff (Millwood) 2018;37:2060–2068. Crossref, Medline, Google Scholar
• 15. Vogt TM, Zhang F, Banks M, et al. Provision of pediatric immunization services during the COVID-19 pandemic: An assessment of capacity among pediatric immunization provider participating in the vaccines for children program—United States, May 2020. MMWR Morb Mortal Wkly Rep 2020;69:859–863. Crossref, Medline, Google Scholar
• 16. American Community Survey. Types of internet subscriptions by selected characteristics. Secondary types of internet subscriptions by selected characteristics. 2019. Available at https://data.census.gov/ (last accessed January 8, 2021). Google Scholar
• 17. American Community Survey. Income in the past 12 months (in 2019 inflation-adjusted dollars). Available at https://data.census.gov/ (last accessed December 31, 2020). Google Scholar
• 18. Arkansas Department of Human Services. Arkansas Medicaid Provider Manual Little Rock, AR. 2020. Available at https://medicaid.mmis.arkansas.gov/Download/provider/provdocs/Manuals/SectionI/Section_I.doc (last accessed June 30, 2021). Google Scholar
• 19. Arkansas Blue Cross and Blue Shield. Provider manual. 2021. Available at https://www.arkansasbluecross.com/docs/librariesprovider9/providers/arkbluecross/provider-manual.pdf?sfvrsn=ba860fd_16 (last accessed July 30, 2021). Google Scholar
• 20. Center for Medicare and Medicaid Services. Telehealth Services. 2021. Available at https://www.cms.gov/Outreach-and-Education/Medicare-Learning-Network-MLN/MLNProducts/MLN-Publications-Items/CMS1243327 (last accessed June 30, 2021). Google Scholar
• 21. The Center for Connected Health Policy. What is telehealth? Available at https://www.cchpca.org/what-is-telehealth/ (last accessed December 28, 2021). Google Scholar
• 22. American Academy of Pediatrics. Telehealth support: Provider to patient visits. 2021. Available at https://www.aap.org/en-us/professional-resources/practice-transformation/telehealth/Pages/Provider-to-Patient-Visits.aspx (last accessed June 25, 2021). Google Scholar
• 23. Owens PL, Heslin KC, Fingar KR, AJ Weiss. Statistical Brief #240. Healthcare Costs and Utilization Project (HCUP). Rockville, MD, Agency for Healthcare Research and Quality, 2018. Available at https://www.hcup-us.ahrq.gov/reports/statbriefs/sb240-Co-occurring-Physical-Mental-Substance-Conditions-Hospital-Stays.jsp (last accessed February 20, 2021). Google Scholar
• 24. Sundararajan V, Henderson T, Perry C, Muggivan A, Quan H, Ghali WA. New ICD-10 version of the Charlson comorbidity index predicted in-hospital mortality. J Clin Epid 2004;57:1288–1294. Crossref, Medline, Google Scholar
• 25. Rosenbaum PR. Observational studies, 2nd ed. New York: Springer, 2002. Crossref, Google Scholar
• 26. Mehrota A, Ray K, Brockmeyer DM, Barnett ML, Bender JA. Rapidly converting to “virtual practices”: Outpatient care in the era of COVID-19. NEJM Catal Innov Care Deliv 2020. DOI: 10.1056/CAT.20.0091 Crossref, Google Scholar
• 27. Ye S, Kronish I, Fleck E, et al. Telemedicine expansion during the COVID-19 pandemic and the potential for technology-driven disparities. J Gen Intern Med 2021;36:256–258. Crossref, Medline, Google Scholar
• 28. Patel SY, Mehrotra A, Huskamp HA, Uscher-Pines L, Ganguli I, Barnett ML. Variation in telemedicine use and outpatient care during The COVID-19 pandemic in the United States. Health Aff (Millwood) 2021;40:349–358. Crossref, Medline, Google Scholar
• 29. Valesquez D, Mehrota A. Ensuring the growth of telehealth during COVID-19 does not exacerbate disparities in care. Health Affairs Blog 2020. DOI: 10.1377/hblog20200505.591306. Crossref, Google Scholar
• 30. Uscher-Pines L, Sousa J, Jones M. Telehealth use among safety-net organizations in California during the COVID-19 pandemic. JAMA 2021;325:1106–1107. Crossref, Medline, Google Scholar
• 31. Woolf SH, Masters RK, Aron LY. Effect of the covid-19 pandemic in 2020 on life expectancy across populations in the USA and other high income countries: Simulations of provisional mortality data. BMJ 2021;373:1343. Crossref, Google Scholar
• 32. Holtz BE. Patients perceptions of telemedicine visits before and after the coronavirus disease 2019 pandemic. Telemed J E Health 2021;27:107–112. Link, Google Scholar
• 33. Clare CA. Telehealth and the digital divide as a social determinant of health during the COVID-19 pandemic. Netw Model Anal Health Inform Bioinform 2021;10:26. Crossref, Medline, Google Scholar
• 34. Rodriguez JA, Betancourt JR, Sequist TD, Ganguli I. Differences in the use of telephone and video telemedicine visits during the COVID-19 pandemic. Am J Managed Care 2021;27:21–26. Crossref, Medline, Google Scholar
• 35. Payán DD, Rodriguez HP. Telehealth disparities. Health Aff (Millwood) 2021;40:1340. Crossref, Medline, Google Scholar

Appendix

Appendix A1. Technical Notes

COVID Period Analysis

Telehealth visits were measured as a binary outcome variable. Dichotomous outcomes include having had at least one (1) telehealth use (yes/no), (2) synchronous telehealth use (yes/no), (3) m-health use (yes/no), (4) telehealth use for mental/behavioral care, and (5) telehealth use for nonmental/behavioral care. Individual generalized linear models were fit regressing a PCCM/QHP dichotomous primary independent indicator on the outcomes of interest. Matched pairs were accounted for in individual generalized linear models by using a random effect indicator for the dyads. In reality, the number of matched pairs was too high for fitting a random effects model and we therefore collapsed matched propensity scores into 100 groups (percentiles) and used this variable as a random effect.

A generalized logistic regression was utilized and aORs were reported for the differences between matched PCCM and QHP. Maximum likelihood method was utilized for the estimation of each model.

Pre-Post Analysis

We used a difference-in-differences (DIDs) framework to estimate the differential increase in telehealth service use associated with payer types during the first few months of the COVID-19 pandemic. DIDs require a parallel trend assumption. As seen in Figure 2, the satisfaction of parallel trend assumption was visually verified. Our DID model, which captures a one-time post-public health emergency change in the probability of receiving a telehealth service, is outlined in the equation hereunder:

where is the odds of the telehealth use for individual i and time t. s (i) is the group to which i belongs (QHP or PCCM). QHP(s) is the indicator function with domain {QHP,PCCM} that equals to 1 when s = QHP, and 0 otherwise. COVID(t) is the indicator function with domain {Before COVID, During COVID} that equals to 1 when During COVID (March 18, 2020–June 30, 2020) and 0 when Before COVID (January 1, 2020–March 1, 2020). Finally, is the residual term for the telehealth use of individual i at time t. A relative risk ratio was also modeled using binomial distribution with log link. REPEATED statement was used to account for correlation within the experiences of the same individual.

Did Coefficient Estimates from Proc Genmod Model with Binomial Distribution and Log Link, with Repeated Subject
PARAMETER	ESTIMATE	STANDARD ERROR	95% CONFIDENCE LIMITS		Z	PR > |Z|
Intercept	−5.0269	0.0705	−5.165	−4.8887	−71.32	<0.0001
Period	3.1019	0.0699	2.9648	3.2389	44.36	<0.0001
Insurance	−0.2421	0.1063	−0.4505	−0.0337	−2.28	0.0228
Period × insurance	0.5643	0.1056	0.3573	0.7713	5.34	<0.0001

As a sensitivity analysis, we also modeled the log odds ratio using a logistic regression, which also demonstrated a significantly higher rate of change in the telehealth use among QHP population.

Did Coefficient Estimates from Logistic Regression with Repeated Subject
PARAMETER	ESTIMATE	STANDARD ERROR	95% CONFIDENCE LIMITS		Z	PR > |Z|
Intercept	−5.0203	0.0709	−5.1593	−4.8812	−70.76	<0.0001
Period	3.253	0.0706	3.1147	3.3913	46.1	<0.0001
Insurance	−0.2435	0.1069	−0.4531	−0.0339	−2.28	0.0228
Period × insurance	0.6328	0.1065	0.4242	0.8415	5.94	<0.0001

Appendix Table A1. Multivariable Relationship Between Insurance Type and Telehealth Use During First 3 Months of the COVID-19 Pandemic, by Type of Care

	ADJUSTED ODDS RATIO
	TYPE OF CARE
	MENTAL/BEHAVIORAL HEALTH	NOT MENTAL/BEHAVIORAL HEALTH
Insurance (ref: PCCM)
QHP	1.13 (1.07–1.19)***	1.09 (1.03–1.15)**
Attributed to PCP (ref: no)
Yes	1.72 (1.62–1.83)***	2.63 (2.48–2.79)***
Gender (ref: male)
Female	1.39 (1.27–1.52)***	1.43 (1.31–1.57)***
Race (ref: White)
Black	0.67 (0.62–0.73)***	1.22 (1.14–1.31)***
Hispanic	0.97 (0.89–1.05)	1.14 (1.05–1.23)***
Other	0.89 (0.76–1.03)	0.90 (0.77–1.04)
RUCA (ref: urban)
Isolated	1.07 (0.93–1.22)	1.20 (1.05–1.36)**
Large rural	0.86 (0.79–0.94)***	0.87 (0.80–0.94)***
Small rural	0.99 (0.90–1.08)	0.84 (0.77–0.93)***
Income percent of FPL (ref: ≥200%)
<100%	0.98 (0.82–1.18)	0.92 (0.78–1.08)
100–150%	0.96 (0.86–1.07)	0.96 (0.86–1.07)
150–200%	0.92 (0.83–1.02)*	0.93 (0.84–1.03)
Insurance region (ref: Central)
Northeast	0.98 (0.90–1.07)	0.97 (0.90–1.06)
Northwest	1.00 (0.92–1.09)	0.92 (0.84–1.01)*
South Central	0.69 (0.61–0.78)***	0.50 (0.43–0.58)***
Southeast	0.83 (0.73–0.95)**	0.68 (0.61–0.77)***
Southwest	0.80 (0.71–0.90)***	1.21 (1.09–1.34)***
West Central	0.91 (0.82–1.01)*	0.76 (0.68–0.85)***
CCI group (ref: 0)
1–2	1.19 (1.11–1.26)***	2.04 (1.92–2.17)***
3–4	1.31 (1.10–1.55)**	3.14 (2.70–3.65)***
>4	1.50 (1.15–1.96)**	4.47 (3.55–5.62)***
Age category, years (ref: <30)
30–40	1.14 (1.07–1.21)***	1.28 (1.21–1.36)***
40–50	1.16 (1.06–1.26)***	1.49 (1.38–1.62)***
50–60	0.92 (0.76–1.10)	1.40 (1.20–1.64)***
≥60	0.30 (0.07–1.26)	1.23 (0.63–2.43)
Broadband internet access (ref: ≥90%)
<50%a	0.59 (0.39–0.90)**	0.94 (0.66–1.34)
50–70%	0.86 (0.75–0.99)**	1.13 (0.99–1.29)*
70–80%	0.97 (0.85–1.10)	1.02 (0.89–1.16)
80–90%	1.00 (0.88–1.13)	1.01 (0.89–1.14)
Mental health (ref: no)
Yes	NA	1.63 (1.53–1.73)***

Appendix Table A2. Primary Care Case Management and Qualified Health Plan Enrollees Background Characteristics, Pre- and Postpropensity Score Matching

VARIABLES	CATEGORY	UNMATCHED (N = 128,013)			PS MATCHED (N = 60,982)
		PCCM (N = 31,007)	QHP (N = 97,006)	P	PCCM (N = 30,491)	QHP (N = 30,491)	SD	VR	P
		n (%)	n (%)		n (%)	n (%)
Gender	Male	4,522 (14.6)	38,042 (39.2)	<0.001	4,514 (14.8)	4,514 (14.8)	0	1	1.000
	Female	26,485 (85.4)	58,954 (60.8)		25,977 (85.2)	25,977 (85.2)
Race	White	19,253 (62.1)	66,181 (68.2)	0.001	19,225 (63.1)	19,225 (63.1)	0		1.000
	Black	9,071 (29.3)	21,596 (22.3)		8,633 (28.3)	8,633 (28.3)
	Hispanic	1,164 (3.8)	4,043 (4.2)		1,140 (3.7)	1,140 (3.7)
	Other	1,519 (4.9)	5,176 (5.3)		1,493 (4.9)	1,493 (4.9)
RUCA	Urban	18,283 (59.0)	52,270 (53.9)	<0.001	18,133 (59.5)	17,843 (58.5)	0.04		0.085
	Large rural	6,082 (19.6)	19,830 (20.4)		5,916 (19.4)	6,102 (20.0)
	Small rural	4,858 (15.7)	17,180 (17.7)		4,678 (15.3)	4,792 (15.7)
	Isolation	1,784 (5.8)	7,716 (8.0)		1,764 (5.8)	1,754 (5.8)
Income percent of FPL	<100%	1,708 (5.5)	4,678 (4.8)	<0.001	1,573 (5.2)	1,573 (5.2)	0		1.000
	100–150%	17,290 (55.8)	55,595 (57.3)		17,003 (55.8)	17,003 (55.8)
	150–200%	9,218 (29.7)	28,638 (29.5)		9,139 (30.0)	9,139 (30.0)
	≥200%	2,791 (9.0)	8,085 (8.3)		2,776 (9.1)	2,776 (9.1)
Insurance region	Central	9,546 (30.8)	25,214 (26.0)	<0.001	9,527 (31.2)	8,950 (29.4)	0.06		<0.001
	Northern	6,473 (20.9)	19,591 (20.2)		6,398 (21.0)	6,429 (21.1)
	Northwest	4,035 (13.0)	13,756 (14.2)		4,020 (13.2)	4,015 (13.2)
	South	1,962 (6.3)	6,484 (6.7)		1,947 (6.4)	2,013 (6.6)
	Southeast	3,138 (10.1)	9,011 (9.3)		2,972 (9.7)	3,045 (10.0)
	Southwest	2,825 (9.1)	10,830 (11.2)		2,633 (8.6)	2,879 (9.4)
	West Central	3,028 (9.8)	12,110 (12.5)		2,994 (9.8)	3,160 (10.4)
Charlson Comorbidity Index category	0	25,135 (81.1)	71,012 (73.2)	<0.001	24,792 (81.3)	24,792 (81.3)	0		1.000
	1–2	5,213 (16.8)	21,334 (22.0)		5,120 (16.8)	5,120 (16.8)
	3–4	480 (1.5)	3,003 (3.1)		434 (1.4)	434 (1.4)
	>4	179 (0.6)	1,647 (1.7)		145 (0.5)	145 (0.5)
Age category, years	<30	13,748 (44.3)	30,078 (31.0)	<0.001	13,331 (43.7)	13,331 (43.7)	0		1.000
	30–40	12,132 (39.1)	23,593 (24.3)		12,048 (39.5)	12,048 (39.5)
	40–50	4,237 (13.7)	17,803 (18.4)		4,225 (13.9)	4,225 (13.9)
	50–60	839 (2.7)	19,545 (20.2)		837 (2.7)	837 (2.7)
	≥60	51 (0.2)	5,977 (6.2)		50 (0.2)	50 (0.2)
Broadband internet percent	<50%	269 (0.9)	992 (1.0)	<0.001	268 (0.9)	239 (0.8)	0.03		0.104
	50–70%	8,353 (26.9)	26,298 (27.1)		8,133 (26.7)	7,884 (25.9)
	70–80%	9,573 (30.9)	30,775 (31.7)		9,446 (31.0)	9,502 (31.2)
	80–90%	11,141 (35.9)	33,817 (34.9)		10,990 (36.0)	11,196 (36.7)
	≥90%	1,671 (5.4)	5,114 (5.3)		1,654 (5.4)	1,670 (5.5)
Mental health disorders	Yes	15,606 (50.3)	44,739 (46.1)	<0.001	15,124 (49.6)	15,124 (49.6)	0	1	1.000
	No	15,401 (49.7)	52,257 (53.9)		15,367 (50.4)	15,367 (50.4)

Appendix Table A3. Clinical Classifications Software Numbers Used in Identification of Preexisting Mental or Substance Use Disorder

CCS NO.	CCS DESCRIPTION
650	Adjustment disorders
651	Anxiety disorders
652	Attention-deficit, conduct, and disruptive behavior disorders
655	Disorders usually diagnosed in infancy, childhood, or adolescence
656	Impulse control disorders, not elsewhere classifiable
657	Mood disorders
658	Personality disorders
659	Schizophrenia and other psychotic disorders
660	Alcohol-related disorders
661	Substance-related disorders
662	Suicide and intentional self-inflicted injury
663	Screening and history of mental health and substance abuse codes
670	Miscellaneous mental health disorders

Post-COVID-19 Telehealth Identification Algorithm

1.	From medical claims, extract claim lines with dates of services between and including March 1, 2020 and June 30, 2020 which meet any of the following conditions: a. With procedure modifier in (“95,” “GT,” “GQ,” “G0”) b. Or with revenue code = “0780” c. Or place of service code in (2,02) d. Procedure code in list of codes in Exhibit A3 below.
2.	For each applicable claim, modify the type of telehealth using the conditions below: a. If Procedure modifier = “GQ” then type of telehealth is Asynchronous b. If procedure modifier in (“95,” “GT,” “G0”) or place of service code in (2,02) then type of telehealth is Synchronous

Appendix Table A4. List of HCPCS Level I and II Procedure Codes Used

TELEHEALTH CATEGORY	PROCEDURE CODE
Synchronous	99421–99423, 98970–98972, G2061–G2063, Q3014, G0425–G0427, G0406–G0408, G0459, G0508–G0509
Asynchronous	G2010, G2250
Remote monitoring	99453–99454, 99457, 99091
Mobile health	99441–99443, 98966–98968
Mobile health	G2012, G2251–G2252
E-consults	99446–99449, 99451–99452

Appendix Table A5. List of Procedure Modifiers, Revenue Codes and Place of Service Codes Used

CODE	TYPE OF CODE	DESCRIPTION	TELEHEALTH CATEGORY
95	Procedure modifier	Synchronous telemedicine service rendered via a real-time interactive audio and video telecommunications system	Synchronous
GT	Procedure modifier	Telehealth service rendered via interactive audio and video telecommunication systems	Synchronous
GQ	Procedure modifier	Telehealth service rendered via asynchronous telecommunications system	Asynchronous
G0	Procedure modifier	Telehealth services for diagnosis, evaluation or treatment of symptoms of an acute stroke	Synchronous
0780	Revenue code	Telemedicine, general classification	Synchronous
02	Place of service code	Telehealth place of service code	Synchronous