Digital inequalities in health information seeking behaviors and experiences in the age of web 2.0: A population-based study in Hong Kong

Background Inequalities in health information seeking behaviors (HISBs) using mass media and internet websites (web 1.0) are well documented. Little is known about web 2.0 such as social networking sites (SNS) and instant messaging (IM) and experiences of HISBs. Methods We surveyed representative Hong Kong Chinese adults (N = 10143, 54.9% female; 72.3% aged 25–64 years) on frequency of HISBs using traditional sources, internet websites, SNS (e.g., Facebook, Twitter), and IM (e.g., WhatsApp, WeChat) and experiences measured using Information Seeking Experience Scale. Adjusted prevalence ratios (aPRs) for HISBs and experiences by sociodemographic and health-related characteristics were yielded using multivariable Poisson regression with robust variance estimators. aPRs for experiences by HISBs using internet websites, SNS, and IM adjusting for sociodemographic and health-related characteristics were also yielded. Results Being female, higher educational attainment, not smoking, and being physically active were associated with HISBs using any source (all P<0.05). Older age had decreased aPRs for HISBs using traditional sources (P for trend = 0.03), internet websites (P for trend<0.001), and SNS (P for trend<0.001) but not for IM (aged 45–64 years: aPR = 1.48, 95% CI 1.07, 2.03). Lower educational attainment and income were associated with negative experiences including feelings of effort and difficulties in understanding the information (all P for trend<0.05). Older age had increased aPRs for difficulties in understanding the information (P for trend = 0.003). Compared with internet websites, HISBs using IM was associated with feelings of frustration (aPR = 1.39, 95% CI 1.08, 1.79), difficulties in understanding the information (aPR = 1.36, 95% CI 1.12, 1.65), and quality concern (aPR = 1.20, 95% CI 1.08, 1.32). Conclusions We identified correlates of web-based health information seeking and experiences in Hong Kong Chinese adults. Providing greater access to and improved information environment of web 2.0 to the target groups may help address digital inequalities.


Design and participants
The Hong Kong Family and Health Information Trends Survey (FHInTS) is a periodic territory-wide telephone survey on the general public's behaviors and views regarding information use, individual and family well-being, and health communication, under the project named "FAMILY: A Jockey Club Initiative for a Harmonious Society". The target population was Cantonese-speaking Hong Kong residents aged 18 years or above. We have conducted five waves of FHInTS since 2009, and the details have been reported elsewhere [5].
The present study was part of the fifth wave of FHInTS that included two phases of the fieldwork. We conducted the phase 1 survey from April to July 2016 and the phase 2 survey from February to May 2017. As we used the same battery of instruments in phase 1 and 2 surveys, datasets were combined to improve the sample size. Each phase used the dual-frame probability-based telephone survey method. Landline and mobile telephone numbers were randomly generated using known prefixes assigned to telecommunication service providers under the Numbering Plan provided by the Government Office of the Communications Authority. Invalid numbers were removed according to the computer and manual dialing records. Telephone numbers of respondents from previous waves were also filtered. For the landline survey, once a household was successfully reached, an eligible family member whose next birthday was the closest to the interview day was invited for the survey. No second-level sampling was used in the mobile survey. All telephone interviews were conducted by trained interviewers of the Public Opinion Program (POP) at the University of Hong Kong. All data were collected by interviewers using a Web-based Computer Assisted Telephone Interview (Web-CATI) system invented in-house by the research team, which allowed real-time data capture and consolidation.

Measures
Health information seeking behaviors (HISBs). Frequency of HISBs was asked as "How often have you searched for health information in the past 12 months from sources including traditional sources (i.e., television/radio/newspaper/magazine), internet websites, SNS (e.g., Facebook, Twitter), and IM (e.g., WhatsApp, WeChat)?" Responses included at least once a week, 1-3 times in a month, once in several months, seldom, or never. The frequencies were dichotomized into at least once a week/1-3 times in a month/once in several months and seldom/never (reference) due to the non-normal distributions.
Web-based HISBs experiences. Experiences of web-based HISBs were measured using the Information Seeking Experience (ISEE) Scale [23]. Skill barrier was measured using the widely used three items, as "It took a lot of effort to get the information you needed;" "You felt frustrated during your search for the information;" and "The information you found was too hard to understand." The mental barrier was measured using the single item, as "You were concerned about the quality of the information." Responses scored on a Likert scale from 1 = very much agree to 4 = very much disagree. Agreement with ISEE items was dichotomized into very much agree/somewhat agree and somewhat disagree/very much disagree (reference) [23][24][25].
Sociodemographic characteristics. Sociodemographic characteristics included sex, age, marital status, employment status, educational attainment, and monthly household income.
Health-related characteristics. Lifestyle characteristics included smoking status (never, ex-smoker, or current smoker), alcohol drinking (never, ex-drinker, occasional drinker, less than once a month, 1-3 days/month, or 1 day/week or more), and frequency of moderate physical activity (none, 1-3 days/week, or 4 days/week or more). History of doctor-diagnosed chronic diseases (e.g., cardiovascular diseases, respiratory diseases, liver diseases, allergies, and others) was dichotomized into none and any. Depression symptoms were measured using the two-item Patient Health Questionnaire (PHQ-2) that has two DSM-IV diagnostic core criteria for major depression disorder [26]. Each item scores on a Likert scale from 0 = not at all to 3 = nearly every day, with a total score of � 3 indicating possible presence of a depression disorder [26]. The Chinese version of PHQ-2 has been validated in Hong Kong [27]. Cronbach's alpha was 0.72 in the present sample.

Statistical analyses
All data were weighted according to sex, age, and educational attainment distributions of the Hong Kong general population. Survey phases and frames were accounted for survey design effects. Missing data were handled by available case analyses as there were minimal missing values for all variables (< 0.25%). Adjusted prevalence ratios (aPRs) for HISBs and experiences by sociodemographic and health-related characteristics were yielded using multivariable Poisson regression with robust variance estimators. aPRs for experiences by different web-based sources adjusting for sociodemographic and health-related characteristics were yielded in respondents who exclusively used internet websites, SNS, or IM (at least once a week/1-3 times in a month/once in several months), whereas those seldom/never used the three sources or used multiple sources were excluded. The modified Poisson regression estimation of relative risk was used to avoid potential exaggeration because of high prevalence (> 10%) of frequent HISBs and negative experiences [28]. Note that log-binomial regression also estimates relative risk but is subject to narrower confidence intervals than they should be and convergence problems [28]. Stata's "estat gof" command was used to yield goodness-of-fit statistics and the "nbreg" command was used to check the equi-dispersion assumption of Poisson regression. All Poisson regression models were supported as all goodness-of-fit chi-squared tests and tests of dispersion were found not statistically significant (all P = 1.00).
As secondary analysis, multinomial logistic regression was used to examine sociodemographic and health-related correlates of preferred web-based sources: SNS, IM, and internet websites (reference outcome) (S3 File). To test the robustness of results of Poisson regression, ordered logistic regression was used by treating agreement with web-based health information seeking experiences as an ordinal variable (1 = very much disagree, 2 = somewhat disagree, 3 = somewhat agree, 4 = very much agree) (S4 File). All analyses were conducted using Stata 15.1 (StataCorp LP, College Station, TX, USA). A two-sided P-value of < 0.05 was considered statistically significant.

Ethics
The Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster approved this study (UW 09-324). Verbal informed consent of all respondents was documented using the Web-CATI system under close supervision. Telephone interviews were tape-recorded for quality checking with respondents' consent. Records were then erased six months after completing the survey.
Of all respondents (N = 10143), over one third (36.9%) sought health information (at least once a week/1-3 times in a month/once in several months) using internet websites, followed by traditional sources (i.e., television/radio/newspaper/magazine; 35.4%), SNS (17.9%), and IM (12.9%) ( Table 2). Prevalence of HISBs using all four sources increased from phase 1 to phase 2 (all P < 0.001). Nearly three quarters (74.4%) agreed that they were concerned about the quality of the information. Nearly half agreed that it took a lot of effort to get the information needed (46.4%) and the information found was too hard to understand (45.0%). Less than a third (29.8%) agreed that they felt frustrated during the search for the information. Prevalence of agreeing that they were concerned about the quality (72.3% to 76.3%, P = 0.03) and that the information found was too hard to understand (40.6% to 49.0%, P < 0.001) increased from phase 1 to phase 2.
Being female was associated with HISBs using any source (all P < 0.001) ( Table 3). Younger age was associated with HISBs using traditional sources (P for trend = 0.03), internet websites (P for trend < 0.001), and SNS (P for trend < 0.001), whereas the age group of 45-64 years was associated with HISBs using IM (adjusted prevalence ratio [aPR] = 1.48, 95% CI 1.07, 2.03). The older group was more likely to seek health information using IM compared with internet websites (S3 File). Being cohabitated or married was associated with HISBs using IM (aPR = 1.46, 95% CI 1.23, 1.73). Higher educational attainment was associated with HISBs using any source (all P for trend < 0.001), and stronger associations were observed for internet websites (Secondary education: aPR = 4.19, 95% CI 3.39, 5.18; Tertiary education: aPR = 6.38, 95% CI 5.15, 7.91). Higher monthly household income was associated with HISBs using traditional sources (P for trend = 0.003), internet websites (P for trend < 0.001), and IM (P for trend = 0.02), but no association was observed for SNS (P for trend = 0.08). Not smoking and being physically active (i.e., moderate physical activity > 1 day/week) were associated with HISBs using any source.
Lower educational attainment was associated with skill barriers, including feelings of effort, frustration, and difficulties in understanding the information (all P for trend < 0.001) ( Table 4). Higher household income had decreased aPRs for feelings of effort (P for trend = 0.001; � HK $40000: aPR = 0.84, 95% CI 0.72, 0.98) and difficulties in understanding the information (P for trend = 0.02). Older respondents reported that the information found was too hard to understand (P for trend = 0.003) but were less concerned about the quality (P for trend = 0.02). The robustness of results was supported using ordered logistic regression (S4 File).
Quality concern was the most common negative web-based health information seeking experiences across different sources (74.5%-82.1%) ( Table 5). Compared with internet websites, HISBs using IM was associated with feelings of frustration (aPR = 1.39, 95% CI 1.08, 1.79), difficulties in understanding the information (aPR = 1.36, 95% CI 1.12, 1.65), and being concerned about the qualities (aPR = 1.20, 95% CI 1.08, 1.32) . Similar findings of web-based sources as the most prevalent were observed in a national-wide survey in the United States [30,31]. Women tended to be more health-conscious and are often caregivers, and hence being more motivated to seek health information [32]. This was supported by our findings that being female was associated with HISBs using any source. Compared with other SES indicators such as employment status and income, educational attainment was strongly associated with HISBs using any source. Education may provide people with higher health literacy defined as knowledge, skills, and confidence to access, process, and use health information [4,14]. Health literacy has shown associations with HISBs using multiple sources from health professionals, Frequency of health information seeking behavior was treated as a dummy variable (1 = "at least once a week/1-3 times in a month/once in several months" vs 0 = "seldom/never"). c Agreement with web-based health information seeking experiences was treated as a dummy variable (1 = "very much agree/somewhat agree" vs 0 = "somewhat disagree/very much disagree"). https://doi.org/10.1371/journal.pone.0249400.t002 family and friends, and mass media to web-based sources [33]. A deepening divide for those with lower educational attainment in HISBs using internet websites (web 1.0) was observed in our study. Web-based health information may require additional ICTs training, social support, time, and ICTs literacy that the disadvantaged are lacking [34]. Not smoking and being physically active were associated with HISBs using any source. The finding supported HISBs as a proactive approach for health promotion as posited in the health and wellness model [35]. Similar findings were shown in our previous study indicating that more health application possession in people who were physically active to log health records and track health measures (e.g., blood pressure and heart rate) [36]. Reverse causation is possible, as frequent health information seeking can provide behavioral change strategies, b Frequency of health information seeking behavior was treated as a dummy variable (1 = "at least once a week/1-3 times in a month/once in several months" vs 0 = "seldom/never"). c US $1 = HK $7.8.
https://doi.org/10.1371/journal.pone.0249400.t003 Table 4. Adjusted a associations of sociodemographic and health-related characteristics with web-based health information seeking experiences b .

It took a lot of effort to get the information you needed (n = 3530)
You felt frustrated during your search for the information (n = 3506)

The information you found was too hard to understand (n = 3560)
You were concerned about the quality of the information (n = 3546) reinforcement of a psychological commitment, and social support to engaging in healthier behaviors such as quitting smoking and frequent vigorous physical activity [37]. Compared with internet websites (web 1.0), IM appeared to reduce digital inequalities in HISBs in older people in our study. Similar result that age was not a significant predictor of HISBs using web 2.0 was found in the United States [6]. Mobile phone for web 2.0 has higher penetration rate than personal computers for web 1.0 among the elderly in Hong Kong [22], possibly due to low-cost internet access and wide coverage of public free WiFi services (5 1943 hotspots in 2017) [38]. Text messaging-based IM could be more popular due to the low requirements for technology skills. IM has been found as a feasible and effective intervention modality in promoting healthy behaviors in older people [39]. Given the continuous and expanding penetration, IM could be promising health communication channels to reach populations across sociodemographic characteristics [6]. Another explanation is that IM with a more interactive and user-centered environment increases the participation of the disadvantaged groups and hence facilitating HISBs [10]. Middle-aged quitters of smoking perceived benefits from emotional and informational support through participating in IM peer discussion groups in our relapse prevention trial [40]. Our findings go beyond the physical barrier by showing that people with lower educational attainment and income had more skill barriers, including a lot of effort and frustration during the search and difficulties in understanding the web-based health information. This confirmed the SES disparities in experiences of web-based HISBs identified in our qualitative interview [15] and studies including patient populations only [41,42]. The findings supported the "Inverse Care Law" [43], which suggests that the disadvantaged groups are most in need of healthcare but may benefit less from health-related ICTs. Notably, the decline of health and ICTs literacy with age might explain the greater difficulties in understanding the information in the older respondents [14]. Quality concern was the most common (74.4%) negative experience in our study. Such mental barrier may be due to the spread of health misinformation on web-based sources that allow the anonymity of content generator and disseminator and low rigor in monitoring and fact-checking [15]. However, older respondents were found to have less quality concern about the information. One possible explanation is that the elderly may have lower ICTs literacy associated with less exposure to and knowledge of ICTs, which may lead to credulity in web-based sources compared with those with better literacy [6]. Community-based interventions, such as collaborative learning and increased social support, may improve people's confidence in dealing with the web-based information [44]. Healthcare professionals could leverage online platforms to disseminate evidence-based content, correct misinformation, and build trust with the communities. Technology companies can implement mechanisms for vetting and validating the credibility of information. For example, Twitter has now used labels and warning messages to add context and instructions on some Tweets containing disputed or misleading information [45].
IM was associated with more negative experiences among the three web-based sources. Frustration and difficulties in understanding may be attributable to the lower readability of health information on IM as IM applications are designed with shorter text, smaller font size, and more crowded visual presentation than internet websites. Health information on IM may Table 5. Weighted a n (%) and adjusted b prevalence ratios (aPRs) for web-based health information seeking experiences c by different sources. c Agreement with web-based health information seeking experiences was treated as a dummy variable (Agree: 1 = "very much agree/somewhat agree" vs Disagree: 0 = "somewhat disagree/very much disagree"). Respondents reporting seldom/never used the three sources or used multiple sources were excluded.

It took a lot of effort to get the information you needed
https://doi.org/10.1371/journal.pone.0249400.t005 be from a small and closer social network not involving healthcare professionals. Nearly 70% of respondents were concerned about health information on WeChat in a national-wide survey in China [9]. Healthcare professionals can use WhatsApp Business or WeChat Official Account for delivering quality health information to the public. The study had some limitations. The cross-sectional data restricted the inference of temporal sequence between health-related characteristics and HISBs and experiences. Prospective and intervention studies are warranted to investigate the causal relations. All data were selfreported, which were subject to recall bias and social desirability bias. Ecological momentary assessments of smoking and alcohol drinking behaviors and objective measurements of physical activity can be used in future studies. We examined general health information seeking and experiences. Future studies are needed to differentiate the purpose, such as health promotion, disease prevention, treatment, or management. The study sample was from the general Chinese population in Hong Kong, one of the most urbanized and developed cities in China. The generalizability to rural and underdeveloped Chinese communities outside Hong Kong is unclear. However, our findings might foresee the digital inequalities in web-based HISBs and experiences in places with improving cyber-infrastructure and increasing penetration of web 2.0.

Conclusions
We identified correlates of web-based health information seeking and experiences in Hong Kong Chinese adults. Providing greater access to and improved information environment of web 2.0 to the target groups may help address digital inequalities.