Digital Divide: Marked Disparities in Computer and Broadband Internet Use and Associated Health Inequalities in the United States

2.1. Data Sources

Data for the present analysis came from the 2017 American Community Service (ACS) Micro-data Sample, the 2013-2017 ACS Summary File, the National Mortality Database, and the County Health Rankings and Roadmaps.^5-11 The ACS is the primary census database for producing socioeconomic, demographic, housing, and labor force characteristics of various population groups at the national, state, county, and local levels.^7-10,12 The unique advantage of the ACS is that it is conducted annually with a sample size of over 3 million records.^7,9,12 The ACS uses a complex, multistage probability design and is representative of the civilian non-institutionalized population, covering all communities in the US.^7-10,12 The household response rate for the ACS exceeds 98%.^7,12 All data are based on self-reports obtained via mail-back or online questionnaires, telephone, interviews or in-home personal interviews.^7,9,12 Substantive and methodological details of the ACS are available in census and previous publications.^7-10,12 The National Mortality Database, maintained by the National Center for Health Statistics, contains annual as well as temporal mortality data by age, sex, race/ethnicity, cause of death, place of death, and other demographic characteristics.^5,6 The mortality database was used to calculate age-adjusted all-cause and cause- specific mortality rates at the county level for the period 2013-2017. The County Health Rankings and Roadmaps database, compiled by the Robert Wood Johnson Foundation, provided county-level data on life expectancy, health, morbidity, and health-risk behaviors from 2014 to 2017.¹¹

2.2. Dependent Variables

To analyze sociodemographic disparities in internet and computer use at the individual level, we used broadband internet access and computer access as dependent variables. In the 2017 ACS, broadband internet access was defined as whether the respondent or any member of a household had

broadband (high-speed) internet service such as cable, fiber optic, or DSL (digital subscriber line) service installed in that household.^1,7-10 Computer access was defined by whether the respondent or any member of the household owned or used any of the following types of computer: desktop, laptop, smartphone, tablet or other portable wireless computer, or some other type of computer.^17-10 In the 2013-2017 ACS Summary File, the area- level indicator of internet access was defined as the percentage of households with an internet subscription that included broadband service, cellular data plan, satellite, fixed wireless, or dialup service.^1,7,9,10

To analyze health characteristics associated with internet and computer use at the individual level, we considered disability status and health insurance status as outcomes variables. Disability status and health insurance status are the only health-related variables available in the ACS. Analyses of disability status and health insurance status were carried out for 2,385,595 individuals aged ≥18 years. The ACS defined individuals as having a disability if they reported serious vision, hearing, cognitive, ambulatory, self-care, or independent living difficulties.^7-10,12 The ACS concept of disability captures these six aspects of disability to define an overall measure or specific disability types.^7-10,12 To derive vision-related disability, the ACS respondents are asked if they are “blind or ...have serious difficulty seeing even when wearing glasses.” Hearing difficulty is derived from a question that asks respondents if they are “deaf or ...have serious difficulty hearing.” Cognitive difficulty involves serious difficulty concentrating, remembering, or making decisions due to a physical, mental, or emotional condition. Ambulatory difficulty is based on a question that asks respondents if they have “serious difficulty walking or climbing stairs.” Selfcare difficulty is based on the question whether or not the respondent has difficulty dressing or bathing. Independent living difficulty is determined if the respondent reports having difficulty doing errands alone such as visiting a doctor's office or shopping due to a physical, mental, or emotional condition.^7-10,12

In addition to the individual-level disability and health insurance variables, we considered a number of health, morbidity, mortality outcomes at the area (county and Zip code) level, including life expectancy, all-cause and cause-specific mortality rates, disability rates, health insurance rates, hospital admissions rates, and prevalence of mental distress, poor health, smoking, physical inactivity, and obesity.^5,6,10,11,13

2.3. Independent Variables

Race/ethnicity was classified into 6 categories as shown in Table 1 and included the major racial/ethnic groups such as non-Hispanic Whites, non-Hispanic Blacks/African-Americans, American Indians/Alaska Natives (AIANs), Asian/Pacific Islanders (APIs), Hispanics, and a residual category of other races that included multiple race groups. Nativity/immigrant status was defined on the basis of an individual's place of birth. US-born were those born in one of the 50 states, Washington, DC, or one of the US territories. Immigrant or foreign-born refers to those born outside these areas and who were not a US citizen at birth.^7-10,12

Using the social determinants of health framework and past research as a guide, we considered, in addition to race/ethnicity and immigrant status, the following sociodemographic covariates that are known to be associated with disability and health insurance: age, gender, marital status, educational attainment, poverty status, employment status, housing tenure, and region of residence.^12,13 These covariates were measured as shown in Table 1.

2.4. Statistical Methods

Multivariate logistic regression was used to model the association between race/ethnicity and socioeconomic factors and the binary outcomes of broadband internet access, computer use, disability, and health insurance.¹⁴ The two- sample t test was used to test the difference in prevalence between any two groups. To examine area-level associations, Zip-code and county-level correlations between computer and internet use and health, mortality, and socioeconomic characteristics were computed and tested for statistical significance. Analyses were carried out using SAS Version 9.4.¹⁴

3.1. Racial/Ethnic and Socioeconomic Disparities in Broadband Internet and Computer Use

For the total US population in 2017, 94.2% had a computer, which includes smartphones, and 82.9% used a broadband internet subscription. Despite high levels of overall access, there were significant disparities in computer and internet use by sociodemographic characteristics. In 2017, persons aged 65 and older were less likely to have access to computer and broadband internet than those younger than age 65 (Table 1). Approximately 82.7% of persons aged 18-24 had access to broadband internet, compared with 80.5% among those aged ≥65. Approximately 97.8% of persons aged 18-24 reported computer use, compared with 82.8% of those aged ≥65. After controlling for sociodemographic characteristics, persons aged 18-24 had 1.7 times higher odds of broadband internet use and 19.8 times higher odds of computer use, compared with those aged ≥65.

In 2017, 89.7% of APIs had broadband internet service, compared with 66.0% of AIANs, 77.2% of Blacks, 78.8% of Hispanics, and 83.5% of non-Hispanic Whites. APIs (97.4%) were more likely and AIANs (80.3%) less likely than other racial/ethnic groups to own or use a computer (including smartphones). After controlling for sociodemographic characteristics in the multivariate logistic regression models, APIs had 38% higher odds of broadband internet use and 51% higher odds of computer use, compared with non-Hispanic Whites. AIANs, Blacks, and Hispanics had significantly lower adjusted odds of broadband internet and computer use than their non-Hispanic White counterparts (Table 1).

Education and income/poverty level were strongly and consistently linked to both broadband internet and computer use. In 2017, persons with less than a high school education were significantly less likely to have a broadband service than those with a college degree (72.3% vs 89.7%). Persons with less than a high school education were significantly less likely to own or use a computer than those with a college degree (80.5% vs 98.1%). Persons below the poverty level reported significantly lower broadband internet use (72.9% vs. 88.6%) and computer use (83.7% vs. 98.2%), compared with those with incomes at or above 500% of the poverty threshold. After controlling for sociodemographic characteristics, persons with less than a high school education had, respectively, 59% and 82% lower odds of internet and computer use than those with a college degree. Persons below the poverty level had, respectively, 52% and 78% lower adjusted odds of internet and computer use than those with incomes at or above 500% of the poverty threshold.

3.2. Disparities in Disability and Health Insurance by Broadband Internet and Computer Use

Table 2 shows disparities in the prevalence of various types of disability and health insurance coverage by internet and computer use. In 2017, persons without broadband internet access were 1.34 times more likely to have a disability than those with access (17.4% vs. 13.0%). Persons who did not own or use a computer were 2.8 times more likely to have a disability than those using a computer (39.7% vs. 14.2%). Controlling for sociodemographic characteristics accounted for most of the excess risk of overall and different types of disability among those without broadband access. However, after controlling for sociodemographic characteristics, compared to those with a computer, persons who did not own or use a computer had 42% higher odds of overall disability, 43% higher odds of mental disability, 32% higher odds of ambulatory disability, 51% higher odds of self-care disability, 42% higher odds of independent living disability, 32% higher odds of hearing disability, and 36% higher odds of vision disability.

In 2017, persons without broadband access were 77% more likely to be without health insurance than those with access (11.4% vs. 6.5%). Persons who did not own or use a computer were 30% more likely to lack health insurance than those using a computer (10.0% vs. 7.8%). After controlling for sociodemographic characteristics, those lacking access to broadband internet and computers had, respectively, 29% and 11% higher odds of uninsurance than their counterparts with broadband and computer access.

3.3. Area-Level Associations between internet and Computer Use and Health, Mortality and Socioeconomic Indicators

During 2013-2017, there were marked geographic disparities in computer and internet use, with many Zip codes in the Southeast, Southwest, Appalachia, Upper Midwest, and the rural US having lower access to computer and broadband internet (Figure 1). Similar geographic patterns were observed for county-level data (data not shown). Geographic disparities in rates of overall disability and cognitive/ mental disability were also marked, with many ZIP codes in the Southeastern and Appalachian regions showing high rates (Figure 2).

Close

Figure 1:

Percentage of Households with Computer and Internet Access, United States, 2013-2017 (32,989 Zip Codes) Source: Data derived from the 2013-2017 American Community Survey.

Export to PPT

Close

Figure 2:

Overall Disability and Cognitive Disability Rates, United States, 2013-2017 (32,989 Zip Codes) Source: Data derived from the 2013-2017 American Community Survey.

Export to PPT

Zip-code-level correlations indicate statistically significant associations between internet and computer use and various health and mortality indicators (Table 3). Zip-code areas with low internet and computer use had substantially higher rates of disability, uninsurance, ethnic minority population, poverty and unemployment, and lower education. Similar, albeit stronger correlations, were observed at the county level. Internet access and computer use was positively associated with life expectancy (γ=>0.66) and inversely associated with disability (γ=>0.70).

During 2013-2017, communities with low internet use (<62.3%) had 6.6 years shorter life expectancy than communities with high internet use (>85.2%) [75.2 years vs. 81.8%]. The corresponding difference in life expectancy associated with low vs. high computer use was 6.7 years. Communities with lower internet and computer use also had substantially higher rates of all-cause mortality, CVD mortality, cancer mortality, diabetes mortality, homicide, suicide, self-assessed fair/poor health, mental distress, disability, preventable hospitalization, uninsurance, smoking, obesity, and physical inactivity (Table 4).

3.4. Rural-Urban Disparities in Internet and Computer Use

Figure 3 shows rural-urban disparities in internet and computer use during 2013-2017. Compared to metropolitan areas, nonmetropolitan areas had lower internet access (80.3% vs. 69.7%) and computer use (88.4% vs. 80.5%). Consistent rural-urban gradients were observed, with people in rural areas and small urban towns having the lowest level of internet use (65.3% and 70.2% respectively) and computer use (77.0 and 80.9% respectively) and those in large metropolitan areas and medium-size metropolitan areas reporting the highest level of internet use (81.9% and 78.0% respectively) and computer use (89.4 and 86.9% respectively).

Close

Figure 3:

Internet and Computer Use Among US Adults Aged ≥18 years by Levels of Urbanization, (2013 Rural-Urban Continuum Code), United States, 2013-2017

Source: Data derived from the 2013-2017 American Community Survey.

Gradients in internet and computer use by level of urbanization were statistically significant at p<0.00l.

Metropolitan (urban) counties include: (1) large metro = counties in metropolitan areas of ≥1 million population, (2) medium metro = counties in metropolitan areas of 250,000-999,999 population, (3) small metro = counties in metropolitan areas of <250,000 population.

Nonmetropolitan (rural) counties include: (4) urban nonmetropolitan counties or small urban towns = population 2,500 to 49,999, (5) nonmetropolitan rural counties or small rural towns with a population of <2,500.

Export to PPT

4.1. Limitations

This study has limitations. Our study provides only limited analysis of health and health care disparities at the individual level for those lacking access to broadband internet and computers. In the ACS microdata sample, presence of a disability and whether or not individuals had access to health insurance were the only two health-related variables available. No information was available regarding chronic conditions, poor physical health, mental distress, hospitalization, affordability of health care costs, patient-provider communication, difficulty in obtaining health care due to transportation, satisfaction with health care, smoking, obesity, and physical inactivity among individuals without broadband and computer access. Although we analyzed many of these health outcomes at the community level, the patterns of disparities in these outcomes associated with lack of internet and computer access may differ from those at the individual level.

Another potential limitation of the study is that it cannot account for all racial/ethnic and socioeconomic gaps related to the uptake of health care technology. In addition, the study does not address the potential negative health outcomes due to utilizing telehealth. For example, a patient may require in person care and risks being misdiagnosed or the severity of their health issue negated.

Compliance with Ethical Standards

Financial Disclosure:

None to report.

Fundingl Support:

None.

Conflicts of Interest:

None.

Disclaimer:

The views expressed are the authors’ and not necessarily those of the US Department of Health and Human Services or the Health Resources and Services Administration.

Ethical approval:

This study used existing de-identified publicly available data and was deemed exempt.