The study was a cross-sectional and descriptive correlational design using a secondary data analysis. The primary data source was from the National Alliance for Caregiving (NAC) and the American Association of Retired Persons (AARP). This dataset was selected for this study because of (1) its up-to-date information on Internet use by dementia caregivers and (2) continuous refinement of sampling and data collection over the past decade [24,25].

The NAC and AARP survey collected sociodemographic and caregiving-related data about persons with dementia and caregivers as well as their Internet and technology use. Interviews using a standardized questionnaire were programmed into a computerized telephone system and were conducted from March to June 2009. Interviewees were 6806 adults living in communities in California, Delaware, Illinois, Kansas, Ohio, Virginia, and the state of Washington in the United States. Random digit dialing based on surnames produced a set of telephone numbers stratified by geographic population density. Oversampling was done in racial and ethnic minorities (African American, Asian, and Hispanic groups) and older adults (age 50 years or older). One respondent was randomly selected from each household. If there were multiple care recipients (2 or more) for 1 caregiver, interviewers focused on the information for 1 primary care recipient who was receiving the most assistance from the caregiver [24,25].

This study used a subset of the data belonging to persons with dementia and their caregivers. Dementia caregivers were defined as persons who provided unpaid care or assistance to a family, relative, friend, or anyone living with Alzheimer dementia, any other type of dementia, or dementia-related conditions (confusion or forgetfulness). Among the 1768 informal caregivers in the dataset, 450 eligible caregivers were included in this study after excluding those caring for persons younger than 18 years (n=173), anyone with incomplete data regarding their care recipients’ dementia condition (n=7), and nondementia caregivers (n=1138) (See Figure 1).

The data acquisition and use was approved by the NAC and AARP. All data provided were deidentified to follow Health Insurance Portability and Accountability Act privacy rules. The University of Virginia’s Institutional Review Board for Social and Behavioral Sciences reviewed this project and confirmed the exempt status. Data analyses included both preliminary analyses (reliability tests, intercorrelation analyses, an exploratory factor analyses, and relevant statistical assumptions checks) and main analyses (descriptive analyses and a hierarchical binary logistic regression analysis). A total of 13 cases were dropped from the main analysis based on listwise deletion across all independent and dependent variables. This represented only 2.9% of the sample; thus, no data imputation was conducted [28].

Before the descriptive and regression analysis, all statistical assumptions were checked including univariate/multivariate normality, linearity, and multicollinearity. To correct for univariate normality, the variable of age of persons with dementia, number of hours for caregiving, and financial hardship was transformed using a log₁₀ function. For multivariate normality, 2 outliers were identified based on the Mahalanobis distance function. When comparing results from a model with 2 outliers to those from the model without the outliers, there were no differences in R ², coefficients, F statistics, or P values, although the values of the Mahalanobis distance were corrected. To glean the maximum information from the available samples, the final results were reported from the model that did not exclude those 2 outliers. The final sample size (N=437) in the main analysis was sufficient for conducting multiple regression with 16 independent variables because the suggested sample size was 160-320 [28-30].

To answer the first and second research questions, percentage responding weighted frequency and means (SD) were reported as well as results of univariate descriptive statistics (independent t tests, Mann-Whitney U tests, or chi-square tests) to compare health-related Internet users to non-health-related Internet users in the dementia caregivers. To answer the third research question, a hierarchical binary logistic regression analysis was completed to identify factors of the health-related Internet user group (0=non-health-related Internet users; 1=health-related Internet users). Based on the stress process model [13], 16 independent variables were entered into the regression model. Block 1 included age and gender of persons with dementia, whereas Block 2 included sociodemographics of caregivers, including age, gender, education level, income, race and ethnicity, resident care, and relationship to persons with dementia. Separately, functional dependency in terms of ADLs and IADLs was included in Block 3 and caregiving history was included in Block 4. Subjective responses of caregiving stress were included in Block 5. SPSS 20.0 (IBM Corp, Armonk, NY, USA) was used for data analyses. The significance-level criterion for all statistical tests was alpha=.05, 2-tailed. To infer generalizable findings, the study applied a composite score for the population weight, which was calculated based on age, gender, and race/ethnicity, and the results compared to the 2008 population estimates released by the Population Division of the US Census Bureau on May 14, 2009 [24].

The mean age of persons with dementia was 78.37 years (SD 14.13) and the majority were women (69.1%, 309/447). They had moderate levels of functional impairments of ADLs and IADLs (mean 2.21, SD 2.01 and mean 4.21, SD 1.71, respectively). The mean age of their caregivers was 50.30 years (SD 14.98). The majority of them were women (62.0%, 277/447), non-Hispanic whites (71.6%, 320/450), and children or grandchildren of persons with dementia (74.9%, 335/447). Caregivers were educated at the level of high school or less (53.5%, 240/449) and had overall household incomes greater than US $30,000 per year (70.3%, 314/447). The residence areas of the caregivers were evenly distributed (urban: 29.3%, 131/447; suburban: 37.8%, 169/447; rural: 30.7%, 137/447). Caregivers spent a mean 29.96 (SD 46.93) hours per week for a mean of 5.22 (SD 7.96) years performing the role of caregiver for the person with dementia.

Approximately 59% (265/450, 58.9%) of dementia caregivers were identified as health-related Internet users. Several of the caregiver’s characteristics were statistically different between health-related Internet users and non–health-related Internet users. Health-related Internet users were younger (P=.01), were more educated (P<.001), had a higher level of household income (P<.001), and spent fewer hours per week caregiving (P=.004). Health-related Internet users were more likely to be a child or grandchild of persons with dementia (78.4%, 207/265) rather than their spouse (3.4%, 9/265) compared to non–health-related Internet users (child or grandchild: 70.0%, 128/185; spouses: 10.9%, 20/185; P=.02). However, the sociodemographic characteristics of persons with dementia were not statistically different whether their caregivers were health-related Internet users or not (Table 1).

The results of the hierarchical binary logistic regression analysis are shown in Table 2. The overall model explained 23.9% of the variance to predict who health-related Internet users were. The group classifications predicted 80.3% of the health-related Internet user group and 55.4% of the non–health-related Internet user group (Tables 2 and 3).

In Block 1, the age and gender of persons with dementia explained 1.3% of the variance (Nagelkerke R ²=.013; χ² ₂=3.7; P=.15). Both the age and gender of persons with dementia were not significant factors. After adding caregiver’s sociodemographic factors into Block 2, the overall model became significant because Block 2 added 16.9% of the explained variance (Nagelkerke R ²=.182; χ² ₁₅=54.4; P<.001). Caregiver’s age (Wald χ² ₁=4.5; P=.03; OR 0.980, 95% CI 0.963-0.998) and education levels (Wald χ² ₁=25.8; P<.001; OR 3.523, 95% CI 2.168-5.726) were significant factors. Block 3 including ADLs and IADLs did not significantly increase explained variance (Nagelkerke R ²=.003; χ² ₂=0.8; P=.67). However, overall model and predictive values of caregiver’s age and education levels still remained significant (P<.05 for all). Block 4 included the number of hours and duration of caregiving years which did not significantly increase explained variance (Nagelkerke R ²=.016; χ² ₂=5.1; P=.08). However, the overall model including Blocks 1 to 4 still remained significant (Nagelkerke R ²=.201; χ² ₁₉=60.3; P<.001). There were noticeable changes in individual factors. Caregiver’s age became an insignificant factor (Wald χ² ₁=2.5; P=.11). After controlling for sociodemographics of persons with dementia and their caregivers, IADLs became a significant factor (Wald χ² ₁=4.0; P=.045; OR 1.201, 95% CI 1.004-1.436). After controlling for functional dependency of persons with dementia and all sociodemographics, the number of hours for caregiving was a significant factor (Wald χ² ₁=4.8; P=.03; OR 0.519, 95% CI 0.288-0.933). However, caregiver’s education levels remained a factor with similar predictive strength (Wald χ² ₁=25.1; P<.001; OR 3.536, 95% CI 2.158-5.794).

The final model including Blocks 1 to 5 significantly explained 23.9% of the variance (Nagelkerke R ²=.239; χ² ₂₂=73.0; P<.001) with Block 5′s significant increase of 2.0% of explained variance (χ² ₃=12.7; P=.01). Caregiver’s age still remained an insignificant factor (Wald χ² ₁=3.4; P=.07). However, caregiver’s education levels remained a factor with similar predictive strength (Wald χ² ₁=21.9; P<.001; OR 3.348, 95% CI 2.0-5.552). After controlling for sociodemographics of persons with dementia and their caregivers, IADLs became an insignificant factor (Wald χ² ₁=3.2; P=.07; OR 1.183, 95% CI 0.985-1.420). After controlling for functional dependency of persons with dementia and all sociodemographic factors, the number of hours for caregiving remained a significant factor, but decreasing in strength of prediction (Wald χ² ₁=6.3; P=.01; OR 0.452, 95% CI 0.243-0.840). Newly emerging significant factors were identified. Age of persons with dementia was shown as a significant factor (Wald χ² ₁=4.5; P=.03; OR 0.278, 95% CI 0.085-0.906). Caregiver’s emotional stress (Wald χ² ₁=4.0; P=.046; OR 1.249, 95% CI 1.004-1.555) and caregiver’s financial hardship (Wald χ² ₁=6.4; P=.01; OR 4.606, 95% CI 1.416-14.978) were significant factors after controlling for sociodemographics, caregiving history, and functional dependency of persons with dementia.

This study examined the sociodemographic and caregiving characteristics of health-related Internet users among dementia caregivers. In this study, using 2009 NAC and AARP survey data, 59.1% of dementia caregivers were identified as health-related Internet users. Caregiver’s age, levels of education and income, hours spent caregiving each week, and relationship to persons with dementia were univariate factors discriminating the health-related Internet use group from non-health-related users. After controlling for confounding effects, age and dependency of IADLs of persons with dementia, caregiver’s emotional stress, and caregiver’s financial hardship were newly emerging factors of health-related Internet use. Caregiver’s sociodemographics and their subjective responses of caregiving stress were the most significant factors to identify health-related Internet users followed by workload assisting in IADLs of persons with dementia [31].

In all, 59% of the prevalence of health-related Internet use is lower than that of the general public (80% in the 2010 Pew Internet & American Life Project) [21], but similar to those found in other types of caregivers (eg, 42%-60% for cancer caregivers) [17]. Compared to findings reported in the recent Pew Internet Health Tracking Survey (2012), our dementia caregiving participants reported a much higher rate (59%) than 39% reported by general caregivers of an adult or child with significant health issues [32]. This finding suggests that using the Internet has become more prevalent and significant behavior seeking for health-related resources among caregivers in the United States. This intensity seems to result from (1) the huge growth in availability and the widespread adoption of the Internet and relevant technologies [33], (2) health care consumers’ strong motivation, (3) their positive perceptions regarding Internet-based resources [34], and (4) the promising benefits of Internet-based approaches (ie, convenience and confidentiality) [34]. Thus, the Internet has been acknowledged as a promising modality for implementing interventions or distributing caregiving resources.

Caregiver’s sociodemographic characteristics are strong factors in determining their behavior regarding health-related Internet use, including age, education levels, income, and their relationship to persons with dementia. Similar to previous study findings, the predictive values of age and education levels were confirmed in this study. Younger and more highly educated caregivers reported they used Internet-based resources for better health and caregiving purposes similar to the same findings in the general population or of cancer caregivers [17,33-36]. Interestingly, those who had a minimum of college-level education were 3.35 times more likely to be health-related Internet users than those who were educated at the level of high school or less. Age and educational attainment were the most significant factors for eHealth literacy [37,38]. In addition, younger generations were generally considered to be more technology-friendly and more prepared to use Internet-based resources compared to their older counterparts. Additionally, a higher education level may be associated with either a higher level of knowledge of health-related resources or better computer skills [15,38].

Higher levels of both household income and self-reported financial hardship were associated with a greater likelihood of being health-related Internet users. Initially, those 2 findings may appear to conflict with one other because a basic assumption is that a person with higher income would experience a lower level of self-reported financial hardship. However, this inconsistency has been reported in previous research; as with the study findings, individuals with higher levels of income use the Internet more [33,35]. Individuals with higher household incomes are more likely to own computers and handheld mobile devices and spend more time using the Internet in their daily life than their lower-income counterparts [33,37]. However, another study has described the opposite association between income and health-related Internet use. A study of national surveys reported that those with lower incomes were also more likely to participate in online support groups than those with higher incomes [36]. Interestingly, this study found the unique impact of the subjective response of financial hardship after controlling for income levels. One possible explanation is that individuals with higher incomes may have diverse means to reduce their stress levels. However, those who are experiencing financial hardship have limited use of traditional face-to-face resources for stress reduction and they try to find alternatives through the Internet [33].

Spousal caregivers were significantly less likely to use available services, which is consistent with previous reports [14]. This reluctance by spousal caregivers seems to originate from emotional barriers: spouses tend to perform the caregiving role without external help because using resources may be considered as betraying the spousal relationship [14]. However, the significant impact of their relationship seems to result from the confounding effect of their age. After controlling for their age, the predictive value is no longer significant in this study sample. Additionally, children of persons with dementia would naturally be considerably younger than their parent and, thus, likely to be more technology-prepared [39].

One of the salient findings of this study was that the higher the emotional stress experienced by dementia caregivers, the more health-related Internet use they reported. When they reported very much emotional stress (score 5) on a 5-point Likert-type scale, they were 3.05 times more likely to be health-related Internet users than those who reported no emotional stress (score 1). This tendency was found in a previous study that applied qualitative methods to the online postings of dementia caregivers. The emotional concern and seeking psychosocial support represents the second-commonest theme for dementia caregivers after behavior of seeking information [20]. Caregiver stress has been shown as a need variable that facilitates their use of resources [40], especially in dementia caregivers [6,15,40]. Dementia caregivers with higher levels of emotional and psychological stress are more likely to use the traditional face-to-face resources of health and human services [15]. The stress-appraisal theory [41] and the stress process model [13] explain that there is a positive relationship between recognized stress levels and efforts to alleviate stress. In addition, a systemic review of networked technologies in dementia caregiver stated a potential impact of ITC intervention on caregiver stress [42]. Thus, health-related Internet use may be considered a coping strategy for caregivers to relieve their subjective stress or burden [41] and a mediator to modulate the impact of caregiving stress on negative outcomes [13].

The predictive value of functional dependency of IADLs appears after controlling for sociodemographic impact. When caregivers are taking care of totally dependent persons with dementia (IADLs score=6), they were 2.99 times less likely to be health-related Internet users than those taking care of totally independent persons with dementia (IALDs score=0). However, when adding the number of hours for caregiving, the predictive value of IADLs dependency became insignificant. The fewer hours caregivers spent caregiving, the more likely they were to be health-related Internet users. For those caring for cancer survivors, having fewer direct caring responsibilities increases the likelihood of health-related Internet use [17]. When caregivers spent longer times providing direct caregiving tasks, they did not have the time or energy to search for information and seek support via the Internet. Respite care may be especially beneficial to caregivers because it may free up the time they need to access the Internet for health-related purposes. Thus, this finding may suggest more appropriate ways to apply interventions designed for providing physical assistance and, thereby, reducing the workload of dementia caregivers.

This study had several limitations: (1) limited inference of causality due to the cross-sectional design, (2) limited generalizability due to the use of a convenience sampling method, (3) possible responder bias in the self-reported data, and (4) difficulties in controlling data quality as a secondary data analysis.

Conceptually, dichotomous grouping (health-related Internet users vs non–health-related Internet users) has limitations in explaining their complex behavior of seeking resources for caregiving. Using current data, the multiple grouping depending on intensity of use (eg, high vs moderate vs nonuser groups) and multinomial logistic regression may show a more detailed description in this sample. Moreover, data collected in 2009 limitedly reflect current trends in the scientific community or daily practice because Internet research is 1 of the most rapidly changing in the field.

Future up-to-date studies would greatly benefit from the use of a longitudinal design that utilizes other multiple measurements of health-related Internet use or caregiver stress. Because the variables tested in this study are predisposing and need variables to facilitate health care service use [14], the inclusion of enabling variables related to health service utilization should provide a contextually better in-depth understanding to shed new light on the complex picture of health-related Internet use in this population. To define the medical condition of dementia using International Classification of Diseases codes will be more accurate than caregiver-reported condition.

The Internet has become a significant resource for dementia caregivers for health-related purposes. This research adds to our knowledge of the prevalence and factors of health-related Internet use by dementia caregivers. Subjective responses of caregiving stress are a need factor leading to increased Internet use for health and caregiving purposes. Significant demographic factors provide helpful information to identify those who are less likely to use Internet-based resources. The lowest utilization is detected in those who were older, a spouse, less educated, with lower incomes, and devoting longer times to caregiving. Thus, this study helps us identify underserved groups regarding virtual health care resources. Clinical researchers should consider our findings to develop tailored interventions and effective care delivery approaches targeting the virtually underserved caregiving population. Additionally, the study findings may assist policy makers seeking to distribute information, resources, and services via the Internet to help dementia caregivers and their care recipients with dementia.