Data come from Health Information National Trends Survey (HINTS) 4 (Cycle 1), a national probability survey of adults aged 18 or older in the civilian non-institutionalized population of the United States that assesses usage and trends in health information access and understanding (Figure 1). This study uses data from the fourth data collection wave, carried out between October 2011 and February 2012 by the National Cancer Institute. A full description of HINTS methodology is available elsewhere [24,25]. Briefly, the latest iteration used a two-stage stratified sample of addresses present on the Marketing Systems Group database to which questionnaires were mailed for self-administration (in both English and Spanish). This was followed by a reminder card and an additional three mailings of the questionnaire depending on non-response. The sampling frame of addresses was divided into high and low ethnic/racial minority strata, with high minority areas oversampled to provide more exact estimates for minority populations. For each selected household, participants were identified by randomly allocating household to one of two selection methods. In the next-birthday method, the adult in the household whose birthday is soonest completes the survey, and in the all-adult method, any adult in the household can complete the survey. Response rates for the two methods were 37.9% and 35.3% respectively, yielding an overall response rate of 36.7%.

Out of a total of 3959 participants, only those who provided information on Internet use and at least one health-risk behavior (3911/3959, 98.79%) were included in the analytic sample. In univariable analysis, differences in categorical and continuous variables between those who did and did not access the Internet, or between those who did and did not use online resources for health-related information and support, were compared with chi-square and t test, respectively. In multivariable analysis that controlled for sociodemographic characteristics, the association between health behaviors and Internet access and use was determined with logistic and linear regressions. Sampling weights based on the Horvitz-Thompson estimator to account for sample design and non-response (jackknife with 50 replicates) were applied to all analyses to calculate accurate standard errors of estimates [32]. This was implemented using the svyset command in STATA version 12 to declare the appropriate survey design.

As shown in Table 1, poor sun-protective behavior was by far the most prevalent of the health-risk behaviors; nearly nine out of ten participants did not follow sun-safe guidelines (87.7%, 95% CI 86.4-88.9). This was followed by low fruit/vegetable intake (56.0%, 95% CI 53.6-58.3), obesity (28.9%, 95% CI 26.9-31.0), current smoking (17.8%, 95% CI 15.8-20.0), excessive alcohol consumption (15.1%, 95% CI 12.8-17.7), and having a sedentary life-style (14.5%, 95% CI 12.5-16.9).

General Internet use was common as nearly four out of five participants indicated that they had ever used it (78.2%, 95% CI 76.1-80.1). Univariable analysis showed that participants who engaged in any health-risk behaviors (with the exception of excessive alcohol consumption and low fruit/vegetable intake) were significantly less likely to have ever used the Internet (see Table 1).

However, after controlling for sociodemographic and other characteristics in multivariable analysis, only participants with unprotected sun exposure remained less likely to have ever used the Internet (Table 2). Younger age, being female, married, of white ethnicity, and having a college education were all independently associated with ever using the Internet (Table 2).

Among those who had ever accessed the Internet, over three-quarters of participants (78.2%, 95% CI 75.4-80.7) had used it to obtain health-related information during the last year. Participants with low fruit/vegetable consumption were less likely to have sought health-related information online in the last year both in univariable analysis (Table 1), and after controlling for sociodemographic and other confounders in multivariable analysis (Table 2). Being female, in poor health, and having a college education were independently associated with use of the Internet to access health-related information in the last year (Table 2).

Nearly half of Internet users reported that they would look online first whenever they urgently required health-related information (47.8%, 95% CI 44.8-50.7). A higher proportion of those with, rather than without, low fruit and vegetable intake said they would use the Internet as a first source for information on health and medical topics. This was the case both in univariable analysis (Table 1) and when adjusting for potential confounders in multivariable analysis (Table 2).

Over half of all those who had ever been online also reported using the Internet to access some sort of health-related behavioral support in the last year (56.9%, 95% CI 53.7-60.0). Both univariable (Table 1) and multivariable (Table 2) analysis showed that smokers were nearly twice as likely as non-smokers to have used the Internet to obtain behavioral support during last year. Presumably this was primarily due to getting support for stopping smoking as this difference disappeared when information seeking for quitting smoking was excluded from the definition of behavioral support (OR 1.10, 95% CI 0.71-1.69). Participants who were younger, female, college educated, and with higher BMI were also more likely to have accessed behavioral support online during the last year (Table 2).

Our findings provide up-to-date information on Internet access in the United States and demonstrate its widespread use to obtain health-related and medical information and support. In agreement with other national data [3,33], we find that over three-quarters of adults in the United States have ever gone online. Of these, the same proportion has used the Internet to look for health or medical information in the last year and nearly half to obtain behavioral support. This study is the first to show that there are few differences in Internet access and use for health-related support and information between people who do or do not engage in specific health-risk behaviors. This provides empirical evidence that Internet-based interventions to change health-risk behaviors generally reach those who are the intended target of health promotion and lends further credence to the potential of the Internet as a platform for improving public health [21].

Nonetheless, the findings also suggest that the Internet may not be equally effective for addressing all types of health-risk behaviors. In particular, the Internet may be less effective for promoting sun-protective behaviors and related awareness campaigns as Internet access is lower in the at-risk population, even after taking sociodemographic confounders into account. The reasons for this are unclear. It may in part reflect lower Internet penetration of rural areas where poor sun-protective behavior can be more prevalent [34,35], though the extent to which an area was urban or rural was not directly assessed in the current study. Among those with access to the Internet, participants with a diet low in fruit and vegetables were more likely to report using the Internet as the first source for health-related information but were less likely to have used the Internet to obtain health-related information in the last year. This finding is in agreement with work from the United Kingdom, which suggests that daily recommended intake of fruit and vegetables is associated with consistent Internet use after controlling for known confounders [36]. Our results also indicate that the Internet may be particularly effective for providing behavioral support for smoking cessation as current smokers were nearly twice as likely to seek support online, primarily for help with stopping smoking. However, even then, few use intensive online support to aid quit attempts [19].

There were also sociodemographic correlates of Internet use that were mostly independent of health-related behaviors. Access to the Internet and gaining health-related information and support online was associated with being younger, female, having at least college level education and less so with white ethnicity and being married. Importantly, the observed associations of health-risk behaviors with reduced access to the Internet were attenuated but not eradicated when controlling for sociodemographic determinants. Although this suggests that the Internet may be a good medium to deliver health promotion messages and interventions to those with health-risk behaviors, it also indicates a need to be aware that older, male, non-white, and less educated people could be less likely to benefit from the availability of online health-related support. Indeed, many of the characteristics that were associated with limited access or use of the Internet to obtain health-related and medical information in this study such as unemployment, worse education, and being single are also linked with detrimental health behaviors (eg, [37,38]).

Our results have a number of implications. The Internet appears to have sufficient reach to engage people who display various risky health behaviors and, given its other advantages, is therefore a good medium to deliver online interventions to address excessive alcohol use, overeating, and physical inactivity. Based on our findings, smoking cessation interventions in particular may benefit from being delivered online. However, as access to the Internet and its use for obtaining health-related information is more limited among people with inadequate sun protection and with low fruit and vegetable intake, Internet-based interventions to change these behaviors may be less effective and require additional promotion. For instance, it may be important to supplement such interventions with print material and tailored advertising in health care outlets to reach the target population. Moreover, even though access to the Internet has grown exponentially over the last 15 years, this access is not equal across all population characteristics [39], and our results highlight the need to be aware of sociodemographic determinants of Internet use for health information. In order to avoid increasing health inequalities and decreasing effectiveness, online material will need to be tailored to characteristics such as male gender, older age, and lower educational attainment to engage these users. This can be done successfully, for instance, in the area of online support for smoking cessation [40,41].

The study has a number of limitations inherent to most surveys. Findings rely on self-reported data, and this may have introduced biases due to systematic misreporting or forgetting. For instance, participants may underreport their alcohol [42] and tobacco consumption [43] due to social desirability concerns. Moreover, given the cross-sectional nature of the study, no causal interpretations can be made, and we cannot exclude the possibility that unmeasured confounding factors explain some of the observed associations. For example, particular trait characteristics that influence health-risk behaviors (such as greater impulsivity, which is associated with alcohol use and smoking [44]) may also determine use of the Internet to access health-related information. Notwithstanding these limitations, given the anonymous nature of data collection, misreporting is unlikely to have made a substantial contribution to results, and common confounders in the analysis were controlled. Strengths of the study include its representativeness of the US population and its large sample size. However, findings will need to be replicated in longitudinal analyses and other countries to confirm and clarify the reported associations.

Overall, our results suggests that the Internet has a wide reach and should be an effective tool to provide support and information for improving most health-risk behaviors but that sociodemographic characteristics of users need to be taken into consideration when developing online health promotion material.