The COVID-19 Research Database Consortium provided data for the study. The consortium, facilitated by Datavant, is a private and public partnership across industries in the United States to facilitate data sharing and promote public health research. The Consortium provided access to Office Ally and Analytics IQ PeopleCore Consumer linked databases. The Office Ally database provided deidentified US claims data from 100 million unique patients and 3.4 billion medical claims. The Analytics IQ PeopleCore consumer database is a nationally representative database of 242.5 million US adults aged 19 years and older. Analytics IQ PeopleCore Consumer data provided deidentified patient-level data including health characteristics, medical care, and social determinants of health to help decision makers better understand their patients. With the linked identifiers (common tokens) provided by COVID-19 Research Database, we combined the Office Ally claims data with Analytics IQ PeopleCore Consumer data, which enabled us to retrieve patient-related information and examine telehealth usage across demographic and socioeconomic indicators. Telehealth in this study was defined as a range of web-based communications including remote monitoring, telephone calls, and videoconferencing.

The COVID-19 Research Database was established in compliance with regulatory standards to protect patient privacy. The COVID-19 Research Database received a waiver of patient consent by the Western Institutional Review Board for the use of Health Insurance Portability and Accountability Act (HIPAA)–certified deidentified data on April 20, 2020. Exemption status was granted by the Western Institutional Review Board for HIPAA-limited data sets and non–HIPAA-covered data on May 14, 2020. This exemption covers all research performed in the COVID-19 Research Database. In addition, researchers with approved study proposals are granted access only to specific data sets that are necessary to answer their research questions. Only deidentified and limited data sets are made available through the database and certified before access was granted. Individual project institutional board approval was not needed.

The study period was from March 2020 to April 2021. To investigate telehealth usage in low-income populations, data were retrieved from claim records of 2,850,831 patients whose household incomes were at or below 200% of the federal poverty level. Telehealth claims were identified by screening for current procedural terminology modifier codes 95, GT, and GQ. The current procedural terminology is a medical code set that uses a uniform language for coding and reporting health services and medical procedures. The modifiers 95, GT, and GQ supplement claim forms by adding extra information about the services provided. In this case, these codes informed us that the services were delivered via telehealth.

The data were aggregated at the patient level to investigate telehealth usage; that is, whether a patient used telehealth during the study period. A patient with ≥1 telehealth claims during the study period was assigned a value of 1 to the dependent variable, otherwise 0. R software (The R Foundation) was used for the analysis. A multiple logistic regression analysis was used to determine the odds of using telehealth among patients in different subgroups during the study period. Categorical variables were created to divide patients into groups by demographic and socioeconomic characteristics. The total number of claims of each patient during the study period was included in the logistic regression analysis to control its potential impact on the dependent variable—telehealth usage. A P value of <.01 was considered statistically significant.

We analyzed 2,850,831 unique patients and their claim records. The results indicate that among patients in low-income positions, 9.84% of them had ≥1 telehealth claim during the study period. In comparison, among patients whose incomes are above the low-income levels (200% of the federal poverty level), 12.86% of them had ≥1 telehealth claim during the study period. The total number of claims of each patient during the study period ranged from 1 to 16 (mean 3.75, SD 3.59). Nearly 60% of participants were female, 75% of them had a high school education or less, 49% of them were unemployed, and 62% of them identified as non-Hispanic White. Patient characteristics are summarized in Table 1.

As shown in Table 2, the results of our logistic regression analysis suggest that gender, age, education, race, ethnicity, and employment influence telehealth usage among people whose household incomes were at or below 200% of the federal poverty level. The P values suggest that Asian and Hispanic patients are more likely to use telehealth than non-Hispanic White and -Black patients; telehealth usage was not significantly different between non-Hispanic White and -Black patients. Patients who identified as male were 12% less likely to use telehealth than those who identified as female. Additionally, patients with high school or less education were 5% less likely to use telehealth than those with a bachelor’s degree or higher. Our results also suggest that patients in the age group of 18-44 years are 32% more likely to use telehealth than those in the ≥65-year age group. Meanwhile, patients with full-time employment were 15% more likely to use telehealth than those who were unemployed. The results support the hypothesis that telehealth usage is positively associated with education and employment status but negatively associated with age. The results also support the hypothesis that women use telehealth more than men.

We carried out a simple linear regression analysis to investigate the impact of household income on telehealth use within each household size. The dependent variable is the percentage of patients who used telehealth at each income level, and the independent variable is income level. Our linear regression analysis of the patients by household size suggests that within the low-income population, income is a contributor to telehealth usage in households of >2 people. Among 1-2–person households, the association of household income and telehealth usage is insignificant (P=.86 and P=.23, respectively). As shown in Figure 1, for patients in 3-10–person households, income is significantly (P<.001) associated with the percentage of patients using telehealth at each income level. For example, the telehealth usage of patients in 3-person households ranged from 5% (for those with a household income of US $3000) to 10% (for those with a household income of US $42,000). We validate the impact of income by performing multiple logistic regression analyses at the patient level within each household size. Our results suggest that within each household size, income is significantly (P<.001) and positively associated with the odds that a patient uses telehealth, accounting for the effects of age, gender, education, employment, and race. The results support the hypothesis that there are systemic differences in telehealth usage among people who live at or below 200% of the federal poverty level. Factors that impact telehealth usage include age, gender, race, ethnicity, education, employment status, household size, and income. However, among household size and income variables, our hypothesis is supported for at least 3-person households and their income. Thus, there is a positive relationship between household income and telehealth use among people who live at or below 200% of the federal poverty level—this is attributed to living in households with at least 3 people.

Enhancement of digital inclusion supports a reduction in inequities by addressing issues that are specific to subgroups of people [32,33]. Education (health and digital literacy) is intricately linked with employment, income, and technology use [34,35]. Thus, education can be a tool to promote equity in telehealth, employment, and income status. In many instances, local health care and public health systems can conceptualize telehealth inequities to their specific communities and tailor support services that consider characteristics beyond income [36]. Telehealth should be appropriately positioned with support strategies to foster engagement in ways that overcome modifiable barriers and consider nonmodifiable factors such as race, ethnicity, age, and gender [37]. For example, adding broadband internet access to the public health infrastructure, providing computers or laptops, and supporting education that advocates for the use of technology for health-related reasons are just the beginning. This may be a reasonable starting point for Hispanic communities, but a different strategy may be needed for non-Hispanic Black and -White communities, men, and older adults, which consider gender, generational, and cultural perspectives to promote telehealth use.

While this study is not an exhaustive examination of factors that influenced telehealth use, we did consider key factors that may contribute to inequities in usage. The study used the COVID-19 Research Database and is subject to the limitations of administrative databases. In the Office Ally database, the validity of the data is reliant upon the facilities to report accurate data and code visits correctly. The Analytic IQ PeopleCore Consumer database relies on the accuracy of consumer reporting. This study did not consider contextual factors such as the availability of providers who used telehealth, residential segregation, and the lack of a racial and ethnic minority workforce. Future studies should consider these factors and the variety of cultural perspectives in communities. Focus groups of patients and providers in these and other communities may help explore additional information not captured in surveys and claims data that explicate attitudes and challenges with telehealth access and use. Future work could parse out the influence of sociodemographic characteristics on the type of visits used by this population. Such information could be used to develop community-specific programs that facilitate telehealth access either through education or access to technology equipment.

Our study concludes that among people whose incomes are below the federal poverty threshold, Hispanic and Asian patients were more likely to use telehealth than non-Hispanic White and -Black patients. Patients who are employed full-time, female, aged between 18 and 44 years, and had completed a bachelor’s degree were more likely to use telehealth. Income is positively associated with telehealth usage in 3- to 10-person households. As we seek to promote telehealth usage, it is imperative that we consider the socioeconomic and demographic factors among subgroups of people who experience poverty. Due to the long-standing challenges in the US health care system, inequities have the potential to become entrenched in our society unless we take decisive action to address these challenges. By focusing on communities in low-income positions, we provide professionals and decision makers with additional insight to promote public health in an increasingly digital society. The tragic events of COVID-19 the pandemic show us that we need to bolster the public health infrastructure and take a more meaningful and targeted approach to health equity concerns.