Evaluating the Impact of Telemedicine Services on Community Health: A County-Level Analysis

Telemedicine and the Service Literature on Innovation

Increasingly, ICT-enabled telemedicine services are being used to deliver complex, off-site healthcare services that were once the sole domain of in-person interactions with onsite tools (Osmundsen et al. 2015), and support seamless coordination and resource integration (Go Jefferies, Bishop, and Hibbert 2021). Telemedicine services allow patients to remotely access medical specialists by visiting local hospitals and clinics rather than traveling to distant ones to access this care—a critical factor given that a five-minute increase in travel time can reduce hospital demand by up to 41%, a phenomenon known as “home bias” (Raval and Rosenbaum 2021). More generally, telemedicine services exemplify a fast-growing category of ICT-enabled innovations that have emerged to be among the key service research priorities with the potential to shed light on how services should be managed and delivered, including in “turbulent times” (Barrett et al. 2015; Ostrom et al. 2021; Srivastava and Shainesh 2015). In particular, during the COVID-19 pandemic, remote patient diagnosis and treatment using telemedicine services experienced substantial, unanticipated growth (Berry, Yadav, and Hole 2024; Bestsennyy et al. 2021; Dudley and Sung 2020; Hollander and Carr 2020; Rotenstein and Friedman 2020), with peak use accounting for 69% of doctor-patient visits in the United States in April 2020; similar use patterns were also observed in Europe and Asia (Pearl and Wayling 2022).

We characterize telemedicine as high human social presence and low automated social presence (van Doorn et al. 2017). In essence, telemedicine services reflect ICT-mediated interactions that remain substantively human-driven by both patients and providers. As such, telemedicine services represent a promising frontier for ICT-enabled innovations in the health sector, offering new opportunities for service innovation and enhancing the efficiency and effectiveness of healthcare delivery (Barrett et al. 2015). Furthermore, telemedicine is not a standalone ICT-enabled innovation but is deeply interconnected with other ICT advancements. According to Huang and Rust (2018), the integration of rapid technology evolution and applications into service processes is fundamentally altering service delivery and consumption by unbundling and enabling remote delivery of services, leading to the emergence of new types of technology-enabled services (Makridis and Mishra 2022).

Telemedicine and the Literature on Community Health

By defining community geographically at the county level in our study, telemedicine services are intended to complement conventional health services and benefit community health (Berry et al. 2022; Danaher and Gallan 2016; Ostrom et al. 2021). Thus, it is critical to examine how telemedicine services which (i) involve novel and innovative avenues of delivering care (Berry, Yadav, and Hole 2024; Danaher and Gallan 2016; Fisk et al. 2023; Parasuraman and Colby 2015), (ii) address health disparities (Sunar and Staats 2022), and (iii) integrate healthcare services for seamless coordination (Bavafa, Hitt, and Terwiesch 2018; Go Jefferies, Bishop, and Hibbert 2021; Rajan, Tezcan, and Seidmann 2019; Saghafian et al. 2018) impact community health. Investigating community health which involves multiple stakeholders of a service system—such as the clinic-based service system in this study—has emerged to be among the key priority areas in service research (Anderson et al. 2013; Blocker and Barrios 2015; Danaher and Gallan 2016; Danaher et al. 2024; Huang et al. 2021; Keiningham, Aksoy, and Malthouse 2024; Ostrom et al. 2021).

This line of inquiry aligns with the CBA to health and safety programs (e.g., Aljafari et al. 2024; Bell, Lee, and Gruca 2024; Farmer et al. 2001; Nilsen 2006) and stakeholder theory (Freeman 1984). CBA provides a useful lens for examining the role of telemedicine services in local communities because “the belief that the community-based approach is beneficial appears to have become a deeply held conviction in public health. . .. ‘It is almost an article of faith that locating programs in the community and involving community members in planning, implementation, and evaluation can be an effective strategy for improving population health’” (Nilsen 2006, p. 140). Based on critical analyses of the important principles, Nilsen (2006) suggested that substantial resource requirements, a long-term program view, and multifaceted interventions are the most important dimensions of successful application of CBA. This framing contrasts with most extant studies that take the adoption and implementation of telemedicine service programs (or online/digital platforms) as given or “free.” Therefore, our analysis of heterogeneity in the impact of telemedicine services is further informed by stakeholder theory (Freeman 1984), emphasizing relationships among and needs of all community stakeholders who are integral to healthcare services. The CBA to stakeholders has been frequently adopted to examine operations of firms and their impact on the sustainability and corporate social responsibility around various communities. In this vein, our research is akin to recently published studies examining how recent technology developments, such as chatbots, can support population-level health (Ollier et al. 2023), and how AI-related jobs contribute to community-level economic activity, well-being, and social welfare (Makridis and Mishra 2022).

Unintended Consequences of Telemedicine and the Gap in Literature

Telemedicine services, while offering numerous potential benefits, may have limitations and unintended consequences. Mindful of the wisdom in Brock and von Wangenheim’s (2019, p. 129) observation, that “AI certainly holds a lot of promise but it is not a panacea,” it is conceivable that telemedicine-enabled service transformation can lead to unintended failures and unforeseen harms (cf. Berry et al. 2022; Polonsky et al. 2024). Hence, it is imperative that we take a balanced approach to telemedicine services, acknowledging the barriers and challenges.

Effective implementation of telemedicine services requires seamless coordination among the multiple stakeholders of clinic-based services. System-level complexity and dynamism—characterized by multiple stakeholders, relationships, interactions, and constant changes in activities and progress—make it susceptible to failures and unforeseen harms (cf. Polonsky et al. 2024). A significant limitation is the inability of telemedicine services to fully replace in-person visits, especially when physical examination is crucial for accurate diagnosis, such as for chest pain or weight-based medication adjustments (Saljoughian 2021). The absence of hands-on assessment can compromise diagnostic accuracy and the quality of treatment and care, as a telemedicine service setting is not amenable to direct touch, physical presence, and emotional connection. Some patients may have limited capability to effectively communicate with doctors in a telemedicine service setting (Hwei and Octavius 2021). Online interactions can be impersonal and even alienating (Haimi 2023).

Inequity is another substantial barrier to telemedicine services, particularly for older adults, those with disabilities, or those lacking the necessary technology literacy, infrastructure, or trust (Berry, Yadav, and Hole 2024; Haimi 2023; Hwei and Octavius 2021). Even preference for technology-enabled healthcare options can vary (Ostrom et al. 2015; Saljoughian 2021). Telemedicine services may not meet all the needs of all community members (Berry, Yadav, and Hole 2024; Haimi 2023). There are concerns that the benefits of telemedicine services may disproportionately favor a small sub-population of tech-savvy individuals. Privacy and security are also major concerns, with complex terms and regulatory challenges that can discourage engagement in telemedicine services (Hwei and Octavius 2021). These issues can paradoxically exacerbate healthcare disparities rather than reduce them.

Overall, we find that most studies, to date, focus on addressing drivers of telemedicine service adoption and utilization (e.g., broadband subscription, cost, insurance, patient characteristics, infectious and contagious diseases and other care types; see Supplemental Appendix B.2 for a detailed summary of the literature). Similarly, some studies examine one or a few performance metrics at the patient, specialty, and/or hospital levels (e.g., patient travel distance, patient visits, chronic care management, and physician productivity). Yet, questions regarding the effectiveness of telemedicine services in improving health outcomes at scale in communities, as we do in this study, still remain largely unaddressed (Danaher and Gallan 2016). As noted earlier, we operationalize community health in terms of county-level health outcomes that include overall health outcomes, premature mortality, preventable hospital stay, chronic illness care, and others (Anderson et al. 2013; Berry et al. 2022; Ostrom et al. 2021), using counties as the geographic units commonly employed in research on the delivery of healthcare services (e.g., Aljafari et al. 2024; Bell, Lee, and Gruca 2024; Farmer et al. 2001; Nilsen 2006).

Hypothesis Development

Proposed Conceptual Framework

We integrate the relationships posited in the above hypotheses into a conceptual framework depicted in Figure 1. The framework provides a nuanced understanding of how availability of telemedicine services can improve community health while acknowledging the moderating effects of the salient contextual factors relevant to a community—that is, socioeconomic status, digital infrastructure, innovation capacity, and demographics.

OPEN IN VIEWER

Figure 1.

Conceptual framework for the impact of availability of telemedicine services on community health.

Independent Variables and the Sample

We develop a county-level measure of telemedicine services by aggregating clinic-level data on the availability of telemedicine services. We first identify the yearly availability of telemedicine services for each clinic based on whether the following telemedicine activities are conducted at a clinic: (i) host site where service is provided via telecommunication systems; (ii) host site for referral provider, with providers at another site who provide services; or (iii) host site for general practitioner, with specialists at another site who assist in rendering a diagnosis. Clinics respond to the three items: Yes or No. For each study year (2010–2022), we construct an indicator of availability of telemedicine services at the clinic level: (1) if a clinic indicated any of the three measured items, and (0) otherwise (see Supplemental Appendix A.2 for details).

To investigate the effect of availability of telemedicine services on county-level health outcomes, we aggregate the clinic-level telemedicine measure to the county level. Thus, in the second step, we determine how many clinics have (1) for availability of telemedicine services for each county and year and then divide by the number of clinics in the county that year to calculate a proportion measure for each county-year. Thus, using this proportion measure, we can examine the effect of availability of telemedicine services at the county level.

With the two data sources, county-level data on health outcomes and clinic-level data on availability of telemedicine services, we match these data using county as the matching criterion. Our matched sample includes 87 counties over 11 years, resulting in 957 county-year observations before accounting for missing values.

In Figure 2, we plot trends in the proportion of availability of telemedicine services for clinics and counties over the study time interval. Both plots indicate an increasing proportion of clinics providing telemedicine services, reflecting the level of availability of telemedicine services is increasing in Minnesota clinics.

OPEN IN VIEWER

Figure 2.

Trends in the availability of telemedicine services in the state of Minnesota.

Moderating Variables

We acquire data from CHR (socioeconomic status); Federal Communications Commission (FCC) (internet speed); Indiana Business Research Center (IBRC) (innovation index); and US Census (race, gender, and age). Specifically, CHR provides the measure for socioeconomic status, which is a composite measure reflecting the level of income, education, employment, safety, and social support in a county. This measure captures the percentage of population with high school degree and some college education, unemployment rate, children in poverty and income inequality, children in single-parent households and access to social opportunities, as well as violent crime and deaths. CHR provides reverse coded z-scores for this measure among all counties in a state, so higher values represent undesirable outcomes.

Digital infrastructure is measured by internet speed of a county reported by FCC. Specifically, internet speed is based on the value reported for residential fixed broadband connections with a downstream speed of at least 10 Mbps (number of housing units with connections per 1,000 housing units).

To study the moderating effect of innovation capacity, we acquire county-level innovation index data from IBRC at the Indiana University Kelley School of Business (https://www.statsamerica.org/innovation/). The data include the 2023 Headline Innovation Index which aggregates underlying core indexes for innovation inputs and outputs. The index reflects a region’s capacity for innovation, considering human capital, knowledge creation, business dynamics, employment, productivity, and overall economic well-being.

Data on county demographics is collected from US Census annual reports. The unit of observation is county-year. Race reflects the proportion of selected racial sub-populations (classifications that include Black community members). Here, we focus on the proportion of: (i) Black or African American alone or in combination; (ii) Black or African American alone; (iii) Hispanic, Black, or African American alone or in combination; and (iv) Hispanic, Black, or African American alone. These race category designations are reported in the US Census data. Gender reflects the proportion of females in the total population of a county. Age reflects the proportion of a county’s population who are 65 years and older.

For ease of interpretation for the moderation effect analysis, the measures of internet speed, innovation index, race, gender, and age are all calculated by removing the mean values over all counties in a year to obtain a relative measure (similar to the treatment of socioeconomic status in CHR). In line with the study setting, we include a suitable set of both time-variant and time-invariant control variables in our analysis. Details regarding the control variables are provided in Supplemental Appendix Section A.3, while descriptive statistics and correlations are presented in Supplemental Appendix Table A2.

Testing Hypothesis 1

Panel Data Analysis

We estimate the effect of availability of telemedicine services on county-level health outcomes using the following panel data analysis with random effect model specification (Equation 1):

Y i t = β 0 + β 1 W i t + β 2 X i t + β 3 Z i + α i + α i ( t ) + ε i t ,

(1)

where W_it is the proportion of availability of telemedicine services for county i in year t, and the estimation of β 1 captures the effect of this county-level proportion of availability of telemedicine services. X i t is a vector of the time-variant control variables: socioeconomic status, comparison of telemedicine, and total reimbursements by year. Z i is a vector of time-invariant variables that we obtained from various sources to reflect the county heterogeneity: initial telemedicine availability, January precipitation, median household income, population density, population changes, ICU beds, and innovation index. Meanwhile, we control for county differences, α i , by allowing variation in county effects, and year differences, α i ( t ) , by specifying variations of years within counties. To ensure robustness, we also estimate a fixed effect model with time-variant variables only and find results largely consistent with the random effect model.

The random effect model estimation results are presented in Table 1. The results indicate that increased availability of telemedicine services is negatively associated with overall health outcomes (M1: b = −.122; p < .05). Since this outcome variable is reverse coded, an increase in availability of telemedicine services is associated with improved overall health outcomes. Based on the marginal effect, one standard deviation increase in availability of telemedicine services is associated with 2.0% improvement in overall health outcomes. Increased availability of telemedicine services is also associated with reduced premature death rate (M2: b = −.048; p < .01) and preventable hospital stays (M3: b = −.073; p < .01). In addition, the results show that increased availability of telemedicine services is associated with reduced low birthweight rate (M4: b = −.042; p < .001), smoking rate (M5: b = −.068; p < .01), and need for diabetes monitoring (M6: b = −.105; p < .001). We repeat the panel data analysis with a fixed effect model specification and find the estimation results are largely consistent (see Table 2). In sum, we find empirical support for Hypothesis 1.

OPEN IN VIEWER

Table 1.

Random Effect Model Estimation Results For The Effect Of Availability Of Telemedicine Services On County-Level Health Outcomes.

	Effect of Availability of Telemedicine Services
Variables	M1: Overallhealth outcomes (z-score)	M2: Premature death rate (years of potential life lost (YPLL) before age 75 per 100,000 population) a	M3: Preventable hospital stay rate (discharges per 1,000 Medicare enrollees for outpatient-treatable conditions) a	M4: Low birthweight rate (% of all births) a	M5: Smoking rate (% of adults) a	M6: Diabetes monitoring (number of diabetic Medicare enrollees) a
Intercept	21.275*** (4.976)	14.573*** (1.424)	4.208* (2.088)	3.371** (1.121)	4.915** (1.889)	16.469** (5.935)
January precipitation	−.265 + (.143)	−.120** (.040)	−.203*** (.058)	.032 (.032)	.036 (.053)	.268 (.179)
Median household income	−1.938*** (.457)	−.537*** (.131)	.009 (.192)	−.137 (.103)	−.180 (.174)	−1.191* (.547)
Population density	.011 (.076)	−.009 (.022)	−.023 (.032)	.016 (.017)	−.015 (.029)	.385*** (.086)
Population changes	1.834 (1.141)	.054 (.317)	−.163 (.453)	.391 (.256)	−.243 (.422)	.878 (1.478)
Year 2015 to 2018 dummy	.041 (.031)	.025* (.011)	−.223*** (.018)	−.003 (.007)	−.053*** (.015)	−.130*** (.011)
Year 2020 dummy	.040 (.052)	.089*** (.019)	4.709*** (.029)	.029* (.012)	−.094*** (.024)	— —
Year 2022 dummy	.042 (.050)	.126*** (.018)	4.005*** (.028)	.041*** (.012)	.119*** (.023)	— —
ICU beds	.032 (.035)	.009 (.010)	.010 (.014)	−.002 (.008)	.026* (.013)	.232*** (.046)
Total reimbursements	.010 (.019)	.014* (.007)	.021* (.011)	−.008 + (.005)	−.026** (.009)	−.005 (.009)
Initial telemedicine availability	.266* (.108)	.114*** (.030)	−.081 + (.043)	.023 (.024)	−.010 (.040)	.237 + (.140)
Comparison of telemedicine	.020 (.028)	.014 (.010)	.000 (.016)	.006 (.007)	.002 (.013)	.022 + (.011)
Socioeconomic status	.411** (.149)	.222*** (.049)	.042 (.075)	−.009 (.035)	.256*** (.066)	.013 (.083)
Internet speed	.023 (.028)	.005 (.010)	−.027 + (.016)	−.007 (.007)	−.009 (.013)	−.021 (.013)
Innovation index	−.002 (.008)	.001 (.002)	−.006 + (.003)	.002 (.002)	.003 (.003)	.015 (.010)
Race	−1.546 (1.894)	−.933 (.589)	−.021 (.872)	.619 (.434)	−2.320** (.763)	1.960 (1.506)
Gender	−5.884 (4.157)	.017 (1.281)	.193 (1.889)	−1.252 (.960)	−4.715** (1.736)	−3.340 (2.893)
Age	.853 (1.246)	−.000 (.382)	−.621 (.559)	.550 + (.288)	−1.861*** (.512)	−.367 (1.022)
Availability of telemedicine services	−.122 * (.047)	−.048 ** (.017)	−.073 ** (.027)	−.042 *** (.011)	−.068 ** (.023)	−.105 *** (.019)
Log Likelihood	−415.028	478.468	105.603	837.247	268.603	258.483
Num. obs.	890	878	813	877	780	574

Note. ***p < .001, **p < .01, *p < .05, + < .10. Standard errors reported in brackets.

a

Dependent variable is calculated as ln (1 + y). The first and sixth measures are reverse coded by County Health Ranking & Roadmaps (CHR) so that, like the other measures, higher values indicate undesirable outcomes. County-level data are from CHR, and limited data for certain years are disclosed for diabetes monitoring. We apply random effect models for analysis. We control for the county random effect, year by county random effect, time-invariant county characteristics, and time-variant variables.

OPEN IN VIEWER

Table 2.

Fixed Effect Model Estimation Results For The Effect of Availability of Telemedicine Services on County-Level Health Outcomes.

	Effect of Availability of Telemedicine Services
Variables	M1: Overall health outcomes (z-score)	M2: Premature death rate (years of potential life lost (YPLL) before age 75 per 100,000 population) a	M3: Preventable hospital stays rate (discharges per 1,000 Medicare enrollees for outpatient-treatable conditions) a	M4: Low birthweight rate (% of all births) a	M5: Smoking rate (% of adults) a	M6: Diabetes monitoring (number of diabetic Medicare enrollees) a
Total reimbursements	−.000 (.023)	.009 (.009)	.010 (.015)	−.004 (.006)	−.034** (.011)	−.006 (.011)
Comparison of telemedicine	.012 (.027)	.015 (.010)	.001 (.017)	.002 (.007)	.003 (.012)	.020* (.010)
Socioeconomic status	.130 (.204)	.093 (.082)	.017 (.123)	−.018 (.049)	.142 (.104)	.024 (.099)
Internet speed	.026 (.030)	.005 (.011)	−.032 (.019)	−.003 (.008)	−.013 (.013)	−.022 (.015)
Race	−3.255 (2.157)	−1.470* (0.739)	0.306 (1.553)	−0.238 (0.509)	−0.584 (1.171)	2.044 (1.435)
Gender	−7.725 (6.791)	−1.410 (2.988)	3.639 (4.822)	−0.598 (1.977)	−17.143*** (5.110)	2.157 (3.116)
Age	2.729 (2.292)	0.308 (1.106)	0.946 (1.642)	0.834 (0.699)	−2.562* (1.045)	−0.733 (1.363)
Availability of telemedicine services	−.099* (.050)	−.048* (.020)	−.071* (.028)	−.037** (.013)	−.075*** (.022)	−.104*** (.020)
Log Likelihood	−262.296	648.218	259.849	1014.347	450.434	531.338
Num. obs.	921	904	842	904	794	595

Note. ***p < .001, **p < .01, *p < .05. Heteroskedasticity-robust standard errors are reported in brackets. County fixed effects and year groups are included in the model.

a

Dependent variable is calculated as ln (1 + y). The first and sixth measures are reverse coded by County Health Ranking & Roadmaps (CHR) so that, like the other measures, higher values indicate undesirable outcomes. We control for the county and year group fixed effects and time-variant county characteristics (time-invariant variables are absorbed by the county fixed effects).

Robustness Checks

Causal estimation using GRF analysis

We conduct GRF analysis to obtain causal evidence in observational studies (Athey and Imbens 2019; Imbens and Rubin 2015). We implement this supervised matching learning method based on causal forest inference to examine the impact of availability of telemedicine services across counties, following Wager and Athey (2018). Extending the idea of a standard random forest (Breiman 2001), GRF estimates the effect based on an average of multiple trees/partitions using a random forest approach. This causal forest employs an “honest” estimation whereby the sample is randomly split into two subsamples, with one used to develop the trees/partitions and the other to make inference regarding the effect. GRF excels in providing estimates that are robust, consistent, and asymptotically normal. Recently, this method has been used to study the effect of information disclosure on payments to physicians in a healthcare setting (Guo, Sriram, and Manchanda 2021). We apply the GRF method to estimate the average partial effect because we have a continuous treatment variable. The effect estimation is derived from the full sample (see Supplemental Appendix B.1.1). As illustrated in Figure 3 and the average treatment effects in Supplemental Appendix Table B1, the GRF analysis supports the panel data analysis findings, showing that increased availability of telemedicine services in a county is positively associated with improved county-level health outcomes, including overall health outcomes, premature death rate, preventable hospital stays, low birthweight rate, smoking rate, and need for diabetes monitoring.

OPEN IN VIEWER

Figure 3.

Estimated heterogeneous effects of availability of telemedicine services per county, τ ^ .

Lagged independent variable

The impact of telemedicine services may take time to be realized in community outcomes, so we test the sensitivity of our findings to time effects by replacing telemedicine with its 1-year lagged value. Table 3 presents the results, which indicate that lagged telemedicine remains significantly associated with the outcome variables, supporting the robustness of our findings.

OPEN IN VIEWER

Table 3.

Robustness Check: Lagged Independent Variable Using Random Effect Model Estimation Results For The Effect of Availability of Telemedicine Services on County-Level Health Outcomes.

	Effect of Availability of Telemedicine Services
Variables	M1: Overall health outcomes (z-score)	M2: Premature death rate (years of potential life lost (YPLL) before age 75 per 100,000 population) a	M3: Preventable hospital stays rate (discharges per 1,000 Medicare enrollees for outpatient-treatable conditions) a	M4: Low birthweight rate (% of all births) a	M5: Smoking rate (% of adults) a	M6: Diabetes monitoring (number of diabetic Medicare enrollees) a
Lagged availability of telemedicine services	−.107* (.048)	−.055** (.017)	−.074** (.028)	−.046*** (.011)	−.067** (.023)	−.096*** (.020)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Log Likelihood	−417.344	480.718	95.672	837.031	270.178	255.586
Num. obs.	894	882	817	881	784	574

Note. ***p < .001, **p < .01, *p < .05. Standard errors reported in brackets.

a

Dependent variable is calculated as ln (1 + y). The first and sixth measures are reverse coded by County Health Ranking & Roadmaps (CHR) so that, like the other measures, higher values indicate undesirable outcomes. County-level data are from CHR, and limited data for certain years are disclosed for diabetes monitoring. We apply random effect models for analysis. We control for the county random effect, year by county random effect, time-invariant county characteristics, and time-variant variables.

Clinical care quality

Our estimates could be influenced by existing clinical care quality, as counties with better clinical care infrastructure may experience different telemedicine uptake patterns, or availability of telemedicine could reflect broader variations in local healthcare quality. Overlooking this could bias our results, as we might mistakenly attribute to telemedicine effects that are actually driven by underlying differences in clinical care quality. To address this, we incorporate a lagged county-level measure of clinical care quality obtained from CHR (https://www.countyhealthrankings.org/health-data/community-conditions/health-infrastructure/clinical-care). CHR defines clinical care as “anything relating to the direct medical treatment or testing of patients,” and emphasizes that care should be “accessible, timely, safe, effective and affordable, providing the right care for the right person at the right time.” This measure reflects access to physicians, mental health providers, dentists, insurance, and hospital stays for ambulatory-care sensitive conditions. It is operationalized as a reverse-coded z-score, with higher values indicating poorer quality. As shown in Table 4, controlling for clinical care quality does not materially alter the association between telemedicine services and health outcomes, alleviating concerns regarding the influence of clinical care quality on our results.

OPEN IN VIEWER

Table 4.

Robustness Check: Controlling for Clinical Care Quality Using Random Effect Model Estimation Results For the Effect of Availability of Telemedicine Services on County-Level Health Outcomes.

	Effect of Availability of Telemedicine Services
Variables	M1: Overall health outcomes (z-score)	M2: Premature death rate (years of potential life lost (YPLL) before age 75 per 100,000 population) a	M3: Preventable hospital stays rate (discharges per 1000 Medicare enrollees for outpatient-treatable conditions) a	M4: Low birthweight rate (% of all births) a	M5: Smoking rate (% of adults) a	M6: Diabetes monitoring (number of diabetic Medicare enrollees) a
Clinical care quality	.199 (.227)	.008 (.080)	.467 *** (.123)	.035 (.053)	−.187 + (.106)	.093 (.104)
Availability of telemedicine services	−.121* (.047)	−.048** (.017)	−.071** (.027)	−.042*** (.011)	−.070** (.023)	−.105*** (.019)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Log Likelihood	−415.208	476.862	111.327	835.453	268.841	257.539
Num. obs.	890	878	813	877	780	574

Note. ***p < .001, **p < .01, *p < .05, + < .10. Standard errors reported in brackets.

a

Dependent variable is calculated as ln (1 + y). The first and sixth measures are reverse coded by County Health Ranking & Roadmaps (CHR) so that, like the other measures, higher values indicate undesirable outcomes. County-level data are from CHR, and limited data for certain years are disclosed for diabetes monitoring. We apply random effect models for analysis. We control for the county random effect, year by county random effect, time-invariant county characteristics, and time-variant variables.

Additional analyses to assess robustness and generalizability

Specifically, we evaluate the results using (i) an alternative measure of availability of telemedicine services; (ii) alternative samples to account for incomplete responses from clinics and missing values; (iii) DID analysis incorporating variations in state telemedicine parity laws; (iv) a control for nearby counties; (v) instrumental variable and control function analysis; (vi) within sample DID analysis; and (vii) matching approaches (see Supplemental Appendix B.1.2). These analyses confirm the robustness of our findings across alternative measures and sample criteria, and support the generalizability and causal interpretation of the positive impact of availability of telemedicine services both within our sample and nationwide.

Testing Hypotheses 2, 3, 4, and 5

While increased availability of telemedicine services improves county-level health outcomes, as shown in our main analysis, we now examine heterogeneity in the impact of telemedicine services on health outcomes as it relates to county-level socioeconomic status, digital infrastructure (i.e., internet speed), innovation capacity, and demographics (i.e., race, gender, and age) as proposed in Hypotheses 2 to 5. The results of the analysis are reported in Table 5.

OPEN IN VIEWER

Table 5.

Contingency Effects: Heterogeneous Effects of Availability of Telemedicine Services on County-Level Health Outcomes Depending on County Socioeconomic Status, Digital Infrastructure, Innovation Capacity, and Demographics.

Panel A. County Socioeconomic Status, Digital Infrastructure, and Innovation Capacity.

	Effect of Availability of Telemedicine Services on Overall Health Outcomes
Variables	Socioeconomic status	Digital infrastructure	Innovation capacity
Contingency factor:	M1: Socioeconomic status	M2: Internet speed (Fixed Broadband ≥ 10Mbps)	M3: Innovation Index
Availability of telemedicine services	−.061*** (.018)	−.058** (.019)	−.060** (.019)
Contingency factor	.118*** (.035)	.023 (.020)	−.019 (.079)
Contingency factor × availability of telemedicine services	.098*** (.013)	−.043** (.014)	−.051*** (.015)
Log Likelihood	−393.231	−414.996	−410.972
Num. obs.	890	890	890

Note. ***p < .001, **p < .01, *p < .05. Model M1 is based socioeconomic status, which is a z-score from County Health Outcome Maps data. Model M2 is based on the internet variable from FCC, and we use the data on Residential Fixed Broadband Connections with a Downstream Speed of at least 10 Mbps. Model M3 incorporates the county innovation index, which has been scaled for inclusion as an interaction term with the availability of telemedicine services in the regression analysis.

OPEN IN VIEWER

Panel B. County Demographic Characteristics.

	Effect of Availability of Telemedicine Services on Overall Health Outcomes
Variables	Selected racial groups				Gender	Age
Contingency factor:	M4: Black or African American alone or in combination	M4a: Black or African American alone	M4b: Hispanic, Black or African American alone or in combination	M4c: Hispanic, Black or African American alone	M5: Gender	M6: Older Adults (≥65)
Availability of telemedicine services	−.050** (.018)	−.049** (.018)	−.051** (.018)	−.051** (.018)	−.047** (.018)	−.061** (.020)
Contingency factor	.000 (.042)	−.007 (.040)	−.016 (.031)	.012 (.028)	−.032 (.034)	.177 (.116)
Contingency factor × availability of telemedicine services	−.024* (.009)	−.023* (.009)	−.036*** (.010)	−.041*** (.011)	−.029* (.012)	.053*** (.013)
Log Likelihood	−362.630	−362.701	−359.680	−359.556	−363.614	−354.028
Num. obs.	890	890	890	890	890	890

Note. ***p < .001, **p < .01, *p < .05. The county-level data are retrieved from county population estimates reported in the “County Population by Characteristics: 2010 to 2019,” https://www.census.gov/data/tables/time-series/demo/popest/2010s-counties-detail.html, and “County Population by Characteristics: 2020 to 2023,” https://www.census.gov/data/tables/time-series/demo/popest/2020s-counties-detail.html. The contingency factors include selected racial groups, gender, and age for a county in a year. The measures are then demeaned by the average across counties in the state in a year, which reflect the relative levels compared to the state average.

In Panel A of Table 5, the positive coefficient for the interaction term: socioeconomic status × availability of telemedicine services (M1: b = .098, p < .001) shows that lower socioeconomic status weakens the positive impact of increased availability of telemedicine services, lending support to Hypothesis 2. In essence, telemedicine services may not reach their full potential toward improving health outcomes for counties with low socioeconomic status.

In Panel A, negative coefficients for the interaction terms: internet speed × availability of telemedicine services (M2: b = −.043, p < .01), and innovation index × availability of telemedicine services (M3: b = −.051, p < .001) suggest that higher internet speed and greater innovation capacity in a county strengthen the positive impact of increased availability of telemedicine services, lending support to Hypotheses 3 and 4. Thus, superior digital infrastructure, evidenced by high internet speed, and greater innovation capacity enhance the enabling role of telemedicine services in improving county-level health outcomes.

In Panel B of Table 5, we observe that benefits from availability of telemedicine are greater when the proportion of the Black or African American sub-population is higher (M4: b = −.024, p < .05), and these results are robust across various US Census group designations for Black community members (M4a: b = −.023, p < .05; M4b: b = −.036, p < .001; M4c: b = −.041, p < .001). These results support Hypothesis 5a. Furthermore, we observe that counties with a higher proportion of females (gender) have better overall health outcomes as availability of telemedicine services increases (M5: b = −.029, p < .05), lending support to Hypothesis 5b. Finally, we find that counties with a higher proportion of older adults (age) have lower overall health outcomes as availability of telemedicine services increases (M6: b = .053, p < .001). This result lends support to Hypothesis 5c and suggests that telemedicine services are not likely to reach full potential in improving health outcomes in counties with high proportions of older adults. Possible explanations of this result include older adults’ limited adaptability to telemedicine services or the lack of telemedicine services tailored to their needs.

In summary, we find empirical support for heterogeneity in the positive impact of availability of telemedicine services on health outcomes across counties. Furthermore, we find that heterogeneity is explained by the moderating effects of socioeconomic status, digital infrastructure, innovation capacity, and demographics. We do not observe any negative impacts of telemedicine services on the set of studied health outcomes, alleviating concerns of potential unintended harms (Berry et al. 2022; Polonsky et al. 2024). The relationship between availability of telemedicine services and community health, and the moderating effects on this relationship, are captured in an integrated conceptual framework (i.e., Figure 1, shown earlier).

Post-hoc Analyses of Pandemic Timeframe

Pre- and mid-pandemic subgroup analysis

For this post-hoc analysis, we investigate whether the impact of availability of telemedicine services on overall health outcomes across counties changed during the COVID-19 pandemic. We assess the effect of availability of telemedicine services by splitting our sample into two subgroups: (i) pre-pandemic (2010–2018), and (ii) mid-pandemic (2020–2022). As shown in Supplemental Appendix B.1.3, Table B8, the main effects of availability of telemedicine services remain consistent across both subgroups, suggesting consistent impacts on overall health outcomes before and during the pandemic.

Cross-sectional analysis for COVID-19-related deaths

Furthermore, we examine how the availability of telemedicine services may have impacted community health during the COVID-19 pandemic by facilitating better coordination between in-person and remote virtual care. To that end, we construct a cross-sectional sample that incorporates pre-pandemic availability of telemedicine services data and pandemic mortality (i.e., COVID-19-related deaths). We collected data on COVID-19-related deaths compiled by The New York Times (2023) from state and local sources. We use reported daily deaths, which include both confirmed and probable deaths from COVID-19, to calculate the average number of daily deaths in each county 2020 to 2022. For pre-pandemic availability of telemedicine services and other control variables, we compute the average value across the years 2010 to 2018 for each county. For robustness checks, we develop two alternative measures for COVID-19-related deaths: (i) rolling averaged daily deaths, representing county-level daily deaths averaged over a 30-day period, and (ii) rolling averaged daily deaths per 100,000 county residents. We also conduct robustness checks for the time effects of the measure of daily deaths, exploring different time windows, using the year 2020 only, and the years 2020 to 2021.

Table 6 presents the results of post-hoc regression analysis examining the impact of pre-pandemic availability of telemedicine services on COVID-19-related deaths and its implications for contingency factors. In Model 1, the significant and negative coefficient estimation (M1: b = −.052, p < .01) suggests that higher pre-pandemic availability of telemedicine services is associated with a reduction in new daily COVID-19-related deaths. Calculating the corresponding marginal effect reveals that predicted new daily deaths decrease from 0.155 to 0.101 when pre-pandemic availability of telemedicine services increases by one standard deviation (mean + SD = 0.470 + 0.192 = 0.662). This predicted marginal effect indicates a 36.9% reduction in deaths. This finding is robust when considering alternative measures of COVID-19-related deaths (see Models M2 and M3) and different time windows (see Models M4 and M5).

OPEN IN VIEWER

Table 6.

Pre-pandemic Telemedicine Services and COVID-19-Related Deaths: Cross Sectional Estimation Results for the Effect of Availability of Telemedicine Services on County-Level COVID-19-Related Deaths.

	Effect of Availability of Telemedicine Services on COVID-19-related Deaths
Types:	Alternative Measures			Time Window		Contingency Implications
Sample:	Years: 2020–2022			Year: 2020	Years: 2020–2021	Years: 2020–2022	Year: 2020	Years: 2020–2021
Dependent variable:New daily COVID-19 deaths	M1	M2: Alternative Measure 1	M3: Alternative Measure 2	M4	M5	M6	M7	M8
Availability of telemedicine services	−.052** (.020)	−.051** (.019)	−.030* (.014)	−.082* (.034)	−.063** (.024)	−.051*** (.014)	−.082*** (.025)	−.062*** (.017)
Socioeconomic status × Availability of telemedicine services						−.002(.023)	−.007(.038)	−.002(.028)
ICU beds × Availability of telemedicine services						−.064* (.026)	−.121** (.046)	−.080* (.031)
Internet speed × Availability of telemedicine services						−.017(.017)	−.008(.029)	−.018(.021)
Control variables included	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
R 2	.826	.828	.469	.790	.820	.857	.826	.853
Num. obs.	82	82	82	82	82	82	82	82

Note. ***p < .001, **p < .01, *p < .05. Heteroskedasticity robust standard errors reported in brackets. To run this analysis, we apply ordinary least squares (OLS) regression models. Model 2 includes the alternative measure 1 for outcome variable: rolling averaged daily deaths. Model 3 includes the alternative measure 2 for outcome variable: rolling averaged daily deaths per 100,000 people. For the terms in the interaction, we center variables to reduce multicollinearity issues and increase the interpretability of the results.

We also assess implications of three contingency factors: (i) socioeconomic status, with higher values indicating worse socioeconomic status; (ii) number of ICU beds, with higher values indicating greater need for hospitalized patient care (Janke et al. 2021); and (iii) internet speed, with higher values indicating superior digital infrastructure. In Model M6, we report the results for these interactions with new daily COVID-19-related deaths from 2020 to 2022. The results show that availability of telemedicine services plays a more substantial role in reducing deaths in counties with more ICU beds (M6: b = −.064, p < .05), while its impact is not significant in counties with lower socioeconomic status or superior digital infrastructure. These results are consistent across different time windows in Models M7 and M8.

Contributions

This study is among the very first to expand on the scope of earlier studies showing individual, clinic, or hospital benefits from telemedicine services by investigating their broad-scale impact on health outcomes at the county level. The empirical findings here, taken together, along with the proposed framework (Figure 1), contribute toward advancing the service literature by underscoring the potential for leveraging ICTs—specifically, clinic-based telemedicine services—for improving community health (Anderson and Ostrom 2015; Berry et al. 2022; Berry, Yadav, and Hole 2024; Ostrom et al. 2021). Our study responds to prior calls to revisit and investigate health and safety programs through a CBA (e.g., Aljafari et al. 2024; Bell, Lee, and Gruca 2024; Farmer et al. 2001; Nilsen 2006), and extends stakeholder theory by documenting broad community-wide health benefits from the availability of telemedicine, an ICT-based service innovation (Freeman 1984).

Our study sheds light on an augmented seamless-coordination framework of service innovation by showing how availability of telemedicine services, both directly and indirectly, drive community-wide value configurations and resource integration (Go Jefferies, Bishop, and Hibbert 2021). Telemedicine services amplify the value co-creation through the enhanced connectivity of clinics within healthcare, thus deepening the understanding of frontline service contexts (Makridis and Mishra 2022; Vargo and Lusch 2008). Increased availability of telemedicine services at a clinic facilitates the integration of diverse expertise across providers and incorporates patient-generated inputs through digital platforms, enhancing the application of knowledge in regions. This convergence pushes healthcare closer to a pure service economy, seamlessly merging remote expertise with local delivery to create system-wide synergies, reflecting the idea that “the whole is greater than the sum of its parts” (Antons and Breidbach 2018, p. 17).

By engaging patients as both customers and co-creators, telemedicine services nurture relational, customer-centric exchanges that underpin sustainable community health (Berry, Yadav, and Hole 2024). Patients actively shape care processes in setting goals, managing treatments, and personalizing care routines, especially in chronic care contexts. Their participation also extends to scheduling, self-monitoring, digital communication, and assuming greater autonomy and control over their care. These practices exemplify patient-driven value co-creation and adaptive service innovation, aligning with Go Jefferies, Bishop, and Hibbert (2021).

Moreover, telemedicine acts as a network-of-networks resource integrator (Vargo and Lusch 2008), linking clinics, patients, insurers, and community organizations for value co-creation at individual, organizational, and population levels. As “healthcare is an elementary service that applies to every individual” in a community (Aljafari et al. 2024, p. 1346), our county-level findings demonstrate how seamless ICT-enabled coordination via telemedicine generates, delivers, and disseminates collective health benefits at scale to entire communities.

At a more nuanced level, our findings reveal the heterogeneous impact of telemedicine services on county-level health outcomes, contingent on socioeconomic status, digital infrastructure, innovation capacity, and demographics of counties. Telemedicine alone cannot directly bridge structural disparities; however, we find that it yields greater benefits in counties with robust digital infrastructure, strong innovation capacity, and higher proportions of Black and female residents, whereas lower-resource areas experience diminished or mixed effects. These heterogeneous effects highlight the importance of context in ICT-enabled service innovations (Aljafari et al. 2024; Castrogiovanni 1991), especially in domains like health and safety programs typically assessed at a community level (Nilsen 2006). Strategically adapting telemedicine delivery to local conditions can help address multi-stakeholder needs and ensure sustainable, high-quality care for all community stakeholders (Freeman 1984; Huang et al. 2021; Keiningham, Aksoy, and Malthouse 2024).

Practical Implications

To put the practical implications of this study into perspective, the following background information is relevant. Data on the use of telemedicine services in the United States can be traced back to 2005, starting with 0.02 visits per 1000 persons and rising to 6.57 visits per 1000 persons in 2017 (Barnett et al. 2018). During the COVID-19 pandemic, the use of telemedicine services experienced steep and unanticipated growth, with 40% of surveyed US consumers planning to use telehealth and telemedicine services in the future (Bestsennyy et al. 2021; Dudley and Sung 2020; Hollander and Carr 2020; Rotenstein and Friedman 2020). Specifically, the COVID-19 period demonstrated the potential of telemedicine services to: (i) deliver care at scale— that is, at a community level (Bestsennyy et al. 2021, p. 13)—and (ii) improve equity and access to care (Smith 2022) leading US state governments to seek verification if such potential can indeed be realized during the post-COVID-19 period (cf. MDH 2024). The practical implications of our study include verifying such potential by empirically demonstrating that availability of telemedicine services improve community health, both in aggregate as well as along specific dimensions. By empirically evaluating the moderating effects of socioeconomic status, digital infrastructure (internet speed), innovation capacity, and demographics (race, gender and age), we highlight equity and access to care issues and show that telemedicine’s impact varies across contexts, supporting its potential to reduce health disparities when adapted to local conditions.

These practical implications of our study are both consequential and of contemporary interest. Specifically, health centers in the South region and in rural geographies of the United States have disproportionately fewer telehealth visits (as a percentage of all visits) than their counterparts in the Midwest, Northeast, and West regions and urban geographies, respectively (Demeke et al. 2021). Also, reviewing state-level telemedicine laws and reimbursement policies (pre-COVID-19, to coincide with our data collection period), we find that: (i) all 50 states and Washington DC had some form of Medicaid reimbursement for telemedicine services; and (ii) Minnesota and 39 other states had regulation governing telemedicine service reimbursement by private payers (CCHP 2019, p. 2,19). Thus, our study findings have practical implications by way of being relevant and generalizable to other states and regions because the regulatory context of telemedicine services in Minnesota is comparable to many other US states.

Limitations and Directions of Future Research

Our study has several limitations that provide avenues for future research. First, using more granular data, such as medical specialties, patient conditions, and cost and quality of care, could clarify contextual factors, reduce confounding, and enable our understanding of telemedicine’s dynamic, long-term effects. Second, future research should examine how telemedicine services evolve with ICTs and artificial intelligence, incorporating the role of personal technologies like smartphones and wearables, as well as the dynamics of human versus automated social presence for healthcare communication. Third, additional perspectives on health outcomes, such as patient satisfaction and affordability, are necessary to fully understand the impact of the enabling role of ICTs in improving the delivery of telemedicine services to communities. While our study focused on community health, examining the behavior of individual patients in using telemedicine services can provide a complementary perspective. Fourth, as we primarily rely on geographic units based on counties to define community, future research could analyze non‑geographic communities, such as professional networks or groups formed around shared interests or activities, for additional insights. Deeper theoretical work is needed to bridge gaps between access and quality in healthcare, and between health outcomes and overall well-being, while extending community and stakeholder perspectives. Fifth, with healthcare being a bundle of services, goods, and experiences involving many stakeholders, more research is warranted to further uncover the interactions among the development, adoption, and implementation of ICTs, along with care provider knowledge and equity in care delivery. Finally, given that when it comes to “leveraging technology to advance services” (Ostrom et al. 2015, pp. 143-44), similarities have been acknowledged between telemedicine services for non-urgent care and online banking services (Carenet Health 2020); hence, the generalizability of the conceptual framework proposed in this study should be examined in the context of other service sectors such as to evaluate the community implications of online banking services.