The relationships between health anxiety, online health information seeking, and cyberchondria: Systematic review and meta-analysis

Background: Cyberchondria refers to an abnormal behavioral pattern in which excessive or repeated online searches for health-related information are distressing or anxiety-provoking. Health anxiety has been found to be associated with both online health information seeking and cyberchondria. The aims of the present systematic review and meta-analysis were to examine the magnitude of these associations and identify any moderator variables. Methods: A systematic literature search was performed across several databases (PsycINFO, PubMed, Embase) and reference lists of included studies. Results: Twenty studies were included across two independent meta-analyses, with 7373 participants. Random effects meta-analyses showed that there was a positive correlation between health anxiety and online health information seeking [r = 0.34, 95% CI (0.20, 0.48), p < .0001], and between health anxiety and cyberchondria [r = 0.62, 95% CI (0.52, 0.71), p < .0001]. A meta-regression indicated that the age of study participants [Q(1) = 4.58, p = .03] was partly responsible for the heterogeneity found for the relationship between health anxiety and cyberchondria. Limitations: The generalizability and validity of our findings are restricted by the methodological limitations of the primary studies, namely, an over-reliance on a single measure of cyberchondria, the Cyberchondria Severity Scale. Conclusions: Our review found a positive correlation between health anxiety and online health information seeking, and between health anxiety and cyberchondria. Further research should aim to explore the contexts for these associations as well as address the identified limitations of the extant literature.

The growth of the internet has led to health information being more accessible than ever before. In the United States, more than 100 million internet users search for health information online (Fox & Duggan, 2013). Health-related information is free to access anonymously on the internet and it is available any time on a range of devices (e.g., desktop or laptop computers, tablets, smartphones). In 2016, 51% of adults in Great Britain used the internet to search for health information, compared to 18% in 2007 (Prescott, 2016). The large and ever-increasing numbers of people obtaining health information online suggest that it might have become the most popular method by which to attain such information (Dobransky & Hargittai, 2012).
Access to health information online has potential benefits insofar as educating people about the nature, causes, prevention, and treatment of specific health conditions. However, for some people who are distressed or anxious about their health, the internet may be accessed for the purpose of self-diagnosing or obtaining reassurance (White & Horvitz, 2009). Indeed, people who are more anxious about their health appear to search the internet for health information more frequently (Baumgartner & Hartmann, 2011;Eastin & Guinsler, 2006;Muse et al., 2012) and for greater amounts of time (Singh & Brown, 2014). Consensus has not been reached as to the directionality of the relationship between online health information searching and health anxiety (Starcevic & Berle, 2015). That is, distress and anxiety about health could be a primary motivator for searching for health information online. Alternatively, searching online for health information in the absence of any significant anxiety could be a precursor to increased health anxiety. Moreover, health anxiety resulting from online searches may in turn precipitate further or more detailed searches. The direction of the relationship between searching online for health information and health anxiety may also vary from one person to another.
Heightened health anxiety or distress associated with excessive or repeated searches online for health-related information is referred to as cyberchondria (Starcevic & Berle, 2013). The distinction between online health information seeking and cyberchondria relates to the reasons for the behavior and its consequences (Starcevic & Berle, 2013). That is, cyberchondria not only refers to online health information seeking, but involves excessive searches that are driven by and/or lead to distress and anxiety. Thus, rather than simply lying on a continuum of behavior, online health information seeking and cyberchondria have different aims (learning about a condition versus relieving anxiety about a condition). While online health information seeking is not in and of itself "maladaptive", cyberchondria involves spending an excessive amount of time online (to the expense of more productive activities) and experiencing an increase in anxiety after searching.
Cyberchondria is considered an abnormal behavioral pattern, rather than a condition or diagnostic entity (Starcevic, 2017) and is thought to be especially common among people with high levels of health anxiety (Starcevic & Berle, 2013). Studies have shown that searching for health information may indeed increase levels of distress and uncertainty about one's feared condition (Baumgartner & Hartmann, 2011;Doherty-Torstrick, Walton, & Fallon, 2016;Singh & Brown, 2016;White & Horvitz, 2009), and potentially lead to greater functional impairment (Doherty-Torstrick, Walton, & Fallon, 2016) providing preliminary indirect support for the construct.
A key contributing factor related to cyberchondria is the ambiguity of online health information, such that it is often inaccurate, misleading, or incomplete (Eysenbach, Powell, Kuss, & Sa, 2002). Individuals seeking reassurance about their health may spend much of their time attempting to determine the validity of health-related information. This process contributes to the cycle in which repeated online searches increase distress and anxiety (Starcevic & Berle, 2013).
There appears to be variation in the literature regarding the magnitude of the association between health anxiety and online health information seeking, and between health anxiety and cyberchondria. For instance, some studies have found small to moderate (e.g., r = 0.21) relationships between internet use and health anxiety (e.g., Fergus & Dolan, 2014), whereas others have found a strong relationship (e.g., r = 0.5; Baumgartner & Hartmann, 2011). Similarly, there is variation in the strength of the relationship between health anxiety and cyberchondria, such that one study reported a small relationship (r = 0.23; Selvi, Turan, Sayin, Boysan, & Kandeger, 2018), whereas other studies have found a strong relationship (e.g., r = 0.62; Fergus, 2015). It is critical that we have a good understanding of the magnitude of the association between both health anxiety and online health information seeking, and health anxiety and cyberchondria, so that possible mechanisms and maintaining factors can then be explored in future research. It would also be important to conduct an integrative review to quantify the magnitude of these relationships and identify their moderators and covariates.
Further to these considerations, we conducted the present systematic review and meta-analysis with the following aims: (a) to examine the relationship between health anxiety and online health information seeking, (b) to examine the relationship between health anxiety and cyberchondria, and (c) to identify potential moderator variables that may influence these relationships. Improving our understanding of these relationships is important for determining in what ways health anxiety might be associated with problems related to Internet use (excessive time spent online or life interference arising from searching online), as well as for gaining insight into the relationship between health anxiety and the counterproductive behaviors which are thought to characterize cyberchondria.

Material and Methods
This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) standard (Moher, Liberati, Tetzlaff, & Altman, 2009). The protocol for our systematic review and meta-analysis was pre-registered at PROSPERO (CRD42017069599).

Literature search
One author (RDM) searched PsycINFO, PubMed, and Embase from database inception to March 21, 2018 for relevant literature. The following Boolean expressions were used in PsycINFO and Embase: "(cyberchondria OR online OR internet OR web) AND (search* OR seek* OR brows* OR reassur*) AND (health anxiety OR illness anxiety OR hypochondri*)". The following Boolean expressions were used in PubMed: All articles retrieved were uploaded to Covidence (Covidence, 2018), which is an online screening and data extraction tool. After the removal of duplicates, two authors (RDM and SA) screened titles and abstracts. Studies that were not relevant were excluded. The two authors then assessed the full text of articles to judge their eligibility in accordance with the inclusion criteria. The same two authors (RDM and SA) also inspected the reference lists of selected studies for remaining relevant studies. Disagreement between the two authors was resolved through discussion.

Inclusion criteria
Studies were included in the current systematic review if they fulfilled the following criteria: (a) the study (observational or experimental) investigated the relationship between health anxiety and online health information seeking or cyberchondria, (b) the study included both a measure of health anxiety and online health information seeking or cyberchondria, (c) the study was published in a peer-reviewed journal (d) the study was published in English.

Quality assessment
One author (RDM) assessed the quality of the included studies by using an adapted version of the Quality Assessment Tool for Quantitative Studies developed by the Effective Public Health Practice Project (Ávila, Lucchetti, & Lucchetti, 2017). This tool consists of 19 items that assess 8 criteria: (a) study design, (b) blinding, (c) representativeness -selection bias, (d) representativeness -withdrawals and dropouts, (e) confounders, (f) data collection methods, (g) data analysis, and (h) reporting. The rating for each criterion ranges from 1 (low risk of bias; strong) to 3 (high risk of bias; weak). Studies can have between 4 and 8 component ratings based on the 8 criteria. A global rating is assessed according to the component ratings. For example, a study with 6 ratings could be rated as "strong" if there are no WEAK ratings and at least 3 STRONG ratings, "moderate" if there is one WEAK rating or less than 3 STRONG ratings, or "weak" if there are two or more WEAK ratings.

Data extraction
One author (RDM) developed a data extraction form that was used to extract relevant information from included studies. This information included: first author, journal, publication year, country, study design, sample size, cyberchondria or online health information search measure, health anxiety measure, whether confounding variables were controlled, mean age of participants, Pearson's r value, and quality assessment rating.
Authors of eligible studies were contacted when studies did not provide effect sizes or essential statistics for effect size calculation.

Statistical analysis
The statistical analysis was conducted using R statistical software version 3.4.3 (R Core Team, 2016). In order to normalize the distribution of the raw data (Pearson's r values), these values were transformed to Fisher's z scale and its variance (Borenstein, Higgins, & Rothstein, 2009). Synthesis of individual effect sizes to summary effect sizes was completed by conducting random effects meta-analyses using a restricted maximum likelihood method.
Results were converted back from Fisher's z to Pearson's r for interpretation. Heterogeneity and variance among effect sizes of studies were examined by calculating the Q statistic, which is the standardized sum of the squared deviations of all effects about the mean (Borenstein, Higgins, Hedges, & Rothstein, 2015) and the I 2 statistic, which reflects the proportion of true to total variance (Borenstein et al., 2015). A Bajaut plot was visually inspected to identify sources of heterogeneity. According to Bajaut, Mah, Pignon, and Hill (2002), studies that fall in the top right quadrant of the plot have greater influence on the overall result and contribute most to study heterogeneity. An examination of the characteristics of these studies can allow for the identification of potential moderator variables that contribute to heterogeneity. A moderator analysis using a meta-regression model was conducted in order to identify sources of heterogeneity. Potential moderator variables included age, quality of studies, and control of confounding variables. An outlier and influence diagnostic procedure was used to determine the presence of potential outliers and influential cases (Vierchtbauer & Cheung, 2010). This procedure extends diagnostic procedures from standard linear regression analyses to the context of meta-analysis.
Publication bias was examined by visually inspecting both a funnel plot and a contour enhanced funnel plot. Begg's adjusted rank correlation test and Egger's regression test were used to assess publication bias (Begg & Mazumdar, 1994;Egger et al., 1997).
We conducted two separate random effects meta-analyses. The first meta-analysis examined the relationship between health anxiety and internet use (i.e., online health information seeking) and the second meta-analysis examined the relationship between health anxiety and cyberchondria.
The overall quality of the included studies ranged from 'moderate' to 'strong'. whereas the other 10 studies used a variety of assessment tools for internet searching such as self-report measures of frequency and duration of online searches and tracking of webpages viewed across five months or during a 15-minute task. The most commonly used measure of health anxiety was the Short Health Anxiety Inventory (SHAI; Salkovskis, Rimes, Warwick, & Clark, 2002), followed by the Whitely Index (WI; Pilowsky, 1967), and the Health Anxiety Inventory (HAI; Salkovskis et al., 2002). All other measures of health anxiety were used only once. Inspection of the funnel plot did not reveal asymmetry ( Figure 4). However, examination of the contour enhanced funnel plot indicated an over-representation of study effect sizes outside the significance contours, which may suggest publication bias ( Figure 5).

Second meta-analysis -the relationship between health anxiety and cyberchondria
There was a positive correlation between health anxiety and cyberchondria [r = 0.62, 95% CI (0.52, 0.71), p < .0001; Figure 6] and a high level of heterogeneity [Q = 76.49, p < .0001, I 2 = 90.13%, 95% CI (79.04%, 97.17%)]. Examination of the Bajaut plot revealed that one study was located in the upper right quadrant (Figure 7). A closer examination of this study did not lead to the identification of potential moderator variables. An outlier and influence diagnostic procedure revealed that one study (i.e., Selvi et al., 2018) was identified as a potential outlier. A sensitivity analysis was conducted, which involved re-running the analysis without the identified outlier. The results indicated a similar summary effect size as the original analysis [r = 0.66, 95% CI (0.61, 0.71), p < .0001]. Consequently, this study was retained in subsequent analyses. A moderator analysis was performed to identify sources of heterogeneity. A meta-regression indicated that age [Q(1) = 4.58, p = .03] was partly responsible for the heterogeneity. However, quality of studies [Q(1) = 1.72, p = .19] and control for confounding variables [Q(1) = 1.72, p = .19] did not contribute to heterogeneity among effect sizes. Neither Egger's test (p = .71) nor the rank correlation test were significant (p = .38). Inspection of the funnel plot did not reveal asymmetry ( Figure 8).
However, examination of the contour enhanced funnel plot indicated an over-representation of study effect sizes outside the significance contours, which may suggest publication bias ( Figure 9).

Discussion
This review examined 20 studies that explored the relationship between either health anxiety and online health information seeking, or health anxiety and cyberchondria. Improving our understanding of the function and correlates of online health information seeking is important. This understanding is particularly relevant for people who find such searches to be counterproductive and may experience increases rather than decreases in distress and anxiety about their health. To the best of our knowledge, our work is the first integrative review of this area.
Our review found a medium sized positive association between health anxiety and online health information seeking. This finding confirms assertions that one of the predictors of online health information seeking may be the extent to which an individual is experiencing health anxiety (Baumgartner & Hartmann, 2011;Eastin & Guinsler, 2006;Muse et al., 2012;Singh & Brown, 2014). Additionally, the medium-size strength of the association suggests that other factors might contribute to online health information seeking. That is, online health information seeking may not only be driven by health anxiety. Consequently, there remains scope for more nuanced assessment of the role of health anxiety in online health information seeking. For instance, future research could investigate whether reassurance seeking mediates the relationship between health anxiety and online health information searches.
Despite finding an association between health anxiety and online health information seeking, there appears to be significant variability in the strength of this association across studies. For instance, associations ranged from small (e.g., r = 0.07; Lee & Hawkins, 2016) to large (e.g., r = 0.55; Doherty-Torstrick, Walton, & Fallon, 2016). A meta-regression indicated that age, quality of studies, and control for confounding variables did not explain such heterogeneity. However, there may be other study-related factors, such as study setting, sample characteristics, language, study design, and outcome measures, which explain the variation in the strength of the relationship between health anxiety and online health information searches. Further research should aim to better characterize the sources of such variation and identify individual differences which predict online health information seeking among people with high levels of health anxiety.
A second focus of our review was on the relationship between health anxiety and the notion of cyberchondria, which implies distressing and counterproductive outcomes from online health information seeking. We found a strong relationship between health anxiety and measures of cyberchondria. This relationship appears stronger than that between health anxiety and online health information seeking, which suggests that while most people search for health information online, people with health anxiety might be especially prone to experiencing counterproductive outcomes from such searches. This also suggests that attempts to seek reassurance about one's health by repeatedly searching online for health-related information may maintain health anxiety. However, the imperfect associations between symptoms of health anxiety on one hand and online health information seeking and cyberchondria on the other leaves open the possibility that the latter behaviors are somewhat independent from pathological health anxiety. Indeed, as far as health anxiety and cyberchondria are concerned, it has been demonstrated that they are both related and distinct (Fergus & Russell, 2016;Mathes et al., 2018).
With regards to potential moderator variables that may influence the relationship between health anxiety and cyberchondria, there was also heterogeneity in the strength of associations across the relevant studies. A meta-regression indicated that age explained some of this variation. That is, studies with older participants found a stronger association between health anxiety and cyberchondria, which may suggest that younger people with health anxiety find their searches to be relatively more reassuring, or at least, not as escalating of their anxiety. Future research could further investigate the role of age, general health status, and other potential moderator variables that may influence the relationship between health anxiety and cyberchondria.
The limitations of the primary studies in our review serve to limit the strength of our conclusions. For instance, most of the studies relied upon a single measure of cyberchondria (i.e., the CSS). Consequently, the strength of associations reported in the present review rely on the validity and reliability of this measure, which has been criticized for taking a broad approach to cyberchondria by including items that may be irrelevant and non-specific (Starcevic & Berle, 2015). Additionally, the original five-factor structure for this instrument has been called into question because one subscale (i.e., mistrust of medical professionals) has been found to have poor model fit (e.g., Barke, Bleichhardt, Rief, & Doering, 2016;Fergus, 2014). Second, most of the studies relied upon self-report as a measure of online health information searching. Future studies could employ a controlled experimental design or use real-time measures to avoid the problem of recall bias. A third limitation pertains to the dearth of studies that reported health anxiety in the context of clinically significant conditions, such as illness anxiety disorder, depression, and anxiety. Fourth, perhaps our review was not exhaustive, such that potentially relevant unpublished studies were not identified or included. Also, two studies were excluded because necessary statistics required to calculate an effect size were not provided by the authors. Finally, while cyberchondria is defined as involving heightened health anxiety or distress, it may be important to investigate the relationship between health anxiety and cyberchondria without the distress items of the CSS due to possible item overlap (e.g., Fergus, 2014). However, anecdotally, we note that studies which have reported on the relationship between health anxiety and the separate domains of cyberchondria have also found significant positive correlations with the nondistress domains of cyberchondria .
In conclusion, the findings of our systematic review and meta-analysis indicated a medium sized positive association between health anxiety and online health information seeking, and a strong association between health anxiety and cyberchondria. The findings revealed a high level of heterogeneity in both meta-analyses. A meta-regression indicated that age was a significant moderator of the strength of the association between health anxiety and cyberchondria. Limitations of the literature identified by our review suggest a need to recruit diverse samples, including those with "clinical" levels of health anxiety and illness anxiety disorder. Furthermore, the associations between health anxiety and cyberchondria should be examined using a diverse range of cyberchondria-related measures (e.g., potentially structured interviews as well as self-report measures). Future research may then inform attempts to develop relevant prevention and treatment avenues for people who suffer from distress and health anxiety in the context of online health information seeking.