Internet services have given users convenient access to various types of online information resources. A worldwide study in 2018 reported that more than 4 billion users searched for information via the internet [
1], most of whom searched for health information. More than 70% of U.S. adults were reported to perform such searches [
2], as did European adults [
3] and Chinese citizens [
4]. They use the information to facilitate diagnosis, manage conditions, consult with their doctors and make informed medical decisions for themselves and others [
5‐
9]. However, its overall quality remains low [
10,
11]. This situation is exacerbated by the COVID-19 pandemic and the unprecedented "infodemic", which is characterized by a flood of rumours and conspiracy theories through various online platforms [
12]. During the pandemic, low-quality information resulted in negative consequences for patients due to anxiety about deteriorating conditions, delayed treatments [
13] and even death in extreme cases [
14]. The quality of information has been defined in different ways. It has been defined as "fitness for use" [
15], which is difficult to apply to the judgement of health information due to its vagueness. It has also been defined as meeting or exceeding consumer expectations [
16] with a focus on the value of particular information to consumers and the satisfaction of users’ needs. Personal judgement of its quality is a process that involves the subjective evaluation of information based on the user's own information needs and characteristics [
17]. Hence, users may make judgements for different reasons or based on certain criteria (e.g., whether it is objective, comprehensive, and accurate). Therefore, it is important to understand how users use these criteria to make judgements and what factors affect the way they use these criteria.
The objective of this study is to determine how patients apply these different criteria in their judgement of the quality of online health information during the pandemic in terms of how frequently they apply these criteria and how important they consider them. In particular, we investigate whether there is consistency between the likelihood of using a particular criterion and its perceived importance among groups of users with different demographics (age, gender, and educational level) and levels of health literacy.
Different conceptualizations of information quality have been proposed in previous literature. The process-oriented model views information quality as a product of measurement where accuracy is guaranteed during the measurement process [
18], while the system-oriented model defines information quality as various steps of information collection, retrieval, and display [
19]. In comparison, the user-oriented model conceptualizes information quality as users’ subjective perception of whether information satisfies their information needs [
20], and information quality judgement is the process by which users apply quality criteria to determine the overall quality of certain information [
20]. This study adopts the user-oriented approach to investigate how users make subjective judgements on the quality of health information.
Several user-oriented models were found in previous literature. Bovee, Srivastava, and Mak (2003) proposed an intuitive model of information quality, proposing a three-layer model with integrity, accessibility, interpretability, and relevance in the first layer. Integrity included four criteria: accuracy, completeness, consistency, and existence. Datedness, usefulness, and other criteria were included as the second layer of the relevance criterion. Age and volatility were included within datedness as the third layer of the criteria [
21]. Kahn, Strong, and Wang (2002) developed a two-by-two conceptual model that defined two dimensions of information quality: one dimension for product quality and service quality and the other for conforming to specifications and meeting or exceeding consumer expectations. They assigned multiple criteria to these quadrants, such as freedom from errors, conciseness, completeness, and consistency within the quadrant "product quality * conforms to specifications" and timeless and security within the quadrant "service quality * conforms to specifications" [
17]. Stvilia et al. (2007) developed a three-category framework, including intrinsic, relational, and reputational categories. Within each category, several criteria were included (e.g., currency and complexity within the intrinsic category and accuracy and naturalness in the relational category) [
22].
Sellitto and Burgess (2005) developed a weighted framework consisting of criteria with different weights [
5]. Criteria "reputable" was assigned the highest weight, followed by "no advertising" and "creation”. Stvilia, Mon, and Yi (2009) extended Stivilia's model (2007) and suggested a model of consumer health information quality with five constructs: accuracy, completeness, authority, usefulness, and accessibility, each of which contained several subcriteria (e.g., credibility and reliability in the construct of accuracy) [
23]. Subsequently, Kim and Oh (2009) developed a fine-grained list of criteria for information quality judgement, including clarity, rationality, novelty, and quickness [
24], which were not included in the list of Stvilia et al. Sun et al. (2015) further extended the list by identifying additional criteria: solution feasibility, certification, politeness and humour [
25]. Other lists of judgement criteria have also been found in the works of other scholars (e.g., see Choi & Shah, 2016 [
26]; Emamjome et al., 2013 [
27]). They share most of their criteria with the aforementioned lists.
The issue with these studies is that terms were used inconsistently within different lists of criteria. For instance, the consistency criterion from Stvilia's study was replaced by the term readability in Zhu et al.'s study with the same definition [
28], which was further changed to representation interpretability in Ge et al.'s study [
29]. Besides, currency was interchanged with timeliness and quickness in different studies [
30]. A summary of the aforementioned criteria is presented in Table
1. Recently, Sun et al. (2019) conducted a systematic review of quality judgement of online health information and summarize all the criteria in previous literature into a list of twenty-five criteria [
11]. This comprehensive list of criteria was used in this study.
Table 1
Models of information quality judgment and their criteria
| 2003 | Integrity, accessibility, Interpretability, and Relevance |
| 2002 | Accessibility, appropriate amount of information, believability, completeness, concise representation, consistent representation, ease of manipulation, free of error, interpretability, objectivity, relevancy, reputation, security, timeless, understandability, and value-added |
| 2005 | Auhority and currency, accuracy, objectivity, and privacy |
| 2007, 2008 | Accuracy/validity, cohesiveness, complexity, semantic consistency, structural consistency, currency, informativeness, naturalness, precision, relevance, accessibility, security, and verifiablity |
| 2009 | Authoritative, complementarity, privacy, attribution, justifiability, transparency, financial disclosure, and advertising policy |
| 2013 | Completeness, originality, objectivity, novelty, accuracy, content quality, verifiability, reliability, amount of data, relevancy, credibility, user feedback, timeless, understandability, value-added, conciseness, consistency, and accessibility |
Weiskopf & Wen | 2013 | Completeness, correctness, concordance, plausibility, and currency |
| 2015 | Accuracy, specificity, objectivity, completeness, relevance, language expression, valuable words, novelty, understandability, profession, originality, external links, external certification, quickness, interactivity, effectiveness, and solution feasibility |
| 2016 | Responsiveness, alternativeness, completeness, emotional support, verifiability, and trustworthy |
Factors influencing the judgement of health information quality
Several factors have been found to influence the judgement of health information quality. Benotsch, Kalichman, and Weinhardt (2004) conducted a survey of HIV patients to determine how patients evaluated the quality of online health information for certain health websites [
18]. Educational and income levels were found to be significantly related to the overall assessment of online health information quality in addition to health literacy (including knowledge) and other psychological factors. Another national survey of Americans’ health information seeking indicated that educational level affected users' ability to navigate within the online environment to seek health information [
31], which affected their judgement of online health information.
Another set of studies focused on the factors that are related to particular criteria for the quality of health information, i.e., trustworthiness and credibility. Age, sex, and education were found to be significantly related to these criteria [
32]. Three reviews summarized a list of possible influencing factors found in empirical studies: gender, education, health status, income, age, health literacy, race, and health beliefs. However, health status has been found to be significantly related to trustworthiness and credibility in certain studies but not in others [
33‐
37]. In addition, controversial results on health beliefs have been found (e.g., Atkinson et al. found that people with poor health status trusted online health information more, while Cotten et al. observed the opposite [
33‐
35,
38,
39]). Furthermore, health beliefs have been found to affect the intention to use health information rather than the direct judgement of quality [
40]. Therefore, these two factors were left for further investigation, while gender, age, education level, and health literacy were retained in this study. In addition, the health literacy scale was replaced with the eHealth literary scale, which is more appropriate for the online environment. Race was not used in this study since the participants were all Chinese.
Although quite a few studies have focused on the overall judgement of health information quality or examined some of its dimensions/criteria while ignoring other criteria, no previous studies have investigated both how likely users are to use certain quality judgement criteria and how important they consider these criteria. In other words, it remains unclear whether the likelihood of using certain criteria is consistent with the perceived importance of these criteria among users with different demographic characteristics. This study investigated how these demographic variables and eHealth literacy affect these patterns and consistency.
Measure and data analysis
The demographics of the participants were collected and categorized as follows. In terms of age, the participants were categorized into five groups from 18 years old to 60 years and above (18–29; 30–39; 40–49; 50–59; 60 and above). Educational levels were also categorized into five groups from junior high school or below to Ph.D. degree (junior high school or below; high school; undergraduate; master’s; Ph.D.). When gender was considered, two groups (female; male) were used to categorize the participants (no participant indicated gender as unclear or transgender).
In addition to the demographic variables, two sets of questions were used in this study. The first set was items to assess the intention of using a particular criterion in the judgement of online health information and the perceived importance of that criterion, adopted from Sun et al.'s review article [
11]. For each criterion, the participant was asked to indicate whether he or she used it in evaluating the quality of online health information. If yes, the participants proceeded to indicate how important the criterion was using a five-point Likert scale. The second set was the eHealth Literacy scale (eHEALS) to assess consumers’ ability to engage in eHealth [
43]. It was used to replace the traditional health literacy scales since the context of this study was online health information seeking and evaluation. This scale has been validated by various empirical studies and shows good psychometric properties. The distribution of the eHealth literacy scores among the participants was skewed (S = -0.446), with a median total score of 30 for the eHEALS scale (range: 8–40). Since there was no consensus on the cut-off score of the eHEALS scale to categorize people into high and low eHealth literacy, we used the median score to divide the participants into these two user groups.
All the measures used in the study were forwards and backwards translated following the COSMIN guidelines. Each type of translation involved two bilingual translators whose mother tongue was the original language: one was the domain expert, and the other was naive about the domain. The backwards translation was compared with the original questions, and discrepancies were resolved by discussion among the authors and translators.
Two kinds of statistical analyses were adopted in this study. To determine whether a particular criterion was used by a patient, chi-square analysis was conducted to compare the group differences in the frequency of criterion use since it was a binary variable. For the perceived importance of each criterion, nonparametric analysis (Kruskal–Wallis test and 1-way ANOVA for post hoc test) was used to compare the group differences (the normal distribution assumption was not guaranteed). It should be noted that in the analysis of importance, participants who indicated that they did not use a particular criterion were excluded since the measurement of the five-point Likert scale assumed that the participants used the criterion regardless of how important they considered it (even if it was considered unimportant, the value was 1 in the scale). Participants who did not use the criterion would introduce bias into the analysis interpretation of the results. The scores of the eHealth literacy scale items were summed to indicate participants’ level of literacy when they conducted online health information seeking and evaluation [
44]. The data analysis was conducted using SPSS 26 software.