This systematic review has been performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [29]. For a detailed description, see the PRISMA checklist in Multimedia Appendix 1.

A comprehensive analysis of the databases of PsycINFO and PubMed was performed. JMIR was also systematically searched. Starting from the word “surrogate seeking,” relevant core terms were identified and used for database analysis by applying the pearl finding and growing strategy [30]. These results were combined by using Boolean operators with family-related terms (mother, father, family, caregiver, parent, child). To consider the linguistic variations, these terms were truncated accordingly: (Mother* OR Father* OR Famil* OR Caregiver* OR Parent* OR Child*) AND (Internet OR Web OR Online OR Cyber* OR eHealth OR e-Health OR Health Information OR Information Seeking).

Studies from January 1994 to June 2018 were considered. The year 1994 was chosen because in this year, the first International World Wide Web Conference took place [31]. The existing web did not have essential health services at that time, and internet use was not common.

Records were summarized in a text-based database. After elimination of the duplicates, titles and relevant abstracts were reviewed. The full texts of the remaining records were reviewed to determine whether they met all inclusion criteria. A protocol of the process for selecting studies is available in Multimedia Appendix 2.

To study OHIS by parents for their children, we included papers that met the following criteria: (1) the participants were parents; (2) the focus of the investigated behavior was OHIS on publicly available websites; (3) the online health-seeking behavior was for their own child; and (4) the study was written in English, presented quantitative data, and was published in a journal between 1994 and 2018.

First, the participants were parents. We defined parents as the primary caregivers who substantially support the child over a stable period in daily routines like feeding, hygiene, play, sleep, or health. Studies including other caregivers (eg, grandparents, other family members) besides parents were excluded if the percentage of other caregivers was greater than 5% of the total sample.

Second, the focus of the investigated behavior was OHIS on publicly accessible websites. Excluded papers included those about special online behaviors (evaluation of one specific website) or areas that are only accessible with registration (support groups, discussion boards, chats), papers that focused only on offline information-seeking behavior (books, television, physicians), and papers with a focus on non–health relevant search behavior.

Third, the online health-seeking behavior was for their own child. Studies with self-seeking behavior only (searching for own medical issues) and studies in which a self-seeking and proxy-seeking distinction was not made or was not possible (eg, pregnancy) were excluded.

Fourth, only papers written in English, presenting quantitative data, and published in a journal between 1994 and 2018 were included.

Based on studies that met the inclusion criteria, we manually reviewed their references to identify further studies that may not have been found through the literature review. Further, we used Google Scholar in June 2018 to identify cited papers that met the inclusion criteria.

One author (CK) manually merged the studies from the different databases, removed duplicates, screened titles and abstracts for relevance, and hand-searched additional citations. The remaining records after screening by title and abstract were independently checked for eligibility by an author (CK) and a psychology master’s student (PS) (Cohen κ=0.84). In cases of nonagreement (7 out of 136), studies were discussed and a consensus for inclusion or exclusion was reached.

The formal study characteristics were extracted from all 33 papers by 1 main author (CK) and can be found in their entirety in Multimedia Appendix 3 with a description of the studies (author, year of publication, location, survey period), study design (survey methodology, prospective vs nonprospective, cross-sectional vs longitudinal, hypothesis generating vs hypothesis testing, sampling technique), and sample characteristics (sample size, amount of parents in the sample, clinical vs community sample, parental gender, parental age, race, education, income, occupation, health insurance, child’s age). Subsequently, the content focus of each paper was coded in 3 category clusters: (1) studies with OHIS related to a child’s specific disease, (2) studies with OHIS related to a treatment or circumstance, and (3) studies that investigated parent-child OHIS in general.

We extracted the quantitative surveyed prevalences on parental OHIS as well as the related item that was used to assess prevalence because the study-specific prevalences are based on varying defined timespans. Significant and nonsignificant associated factors related to these items were extracted as well. To develop a research agenda, further information on theoretical frameworks, study limitations, and mentioned research gaps were extracted from the reviewed studies.

The heterogeneity of the sample composition of studies and the lack of a sufficient sample size of studies with similar outcome variables made the use of meta-analytical methods inappropriate for this review. Therefore, data were summarized by conducting a descriptive analysis and narrative synthesis. Frequency counts of key variables were coded and summarized.

The conceptual model was developed with an inductive approach by the 2 authors (CK, HMF) based on the investigated variables in the studies. The CMIS by Johnson and Meischke [25] provides a theoretical framework and served as a basic structure to categorize the extracted variables. The underlying assumption of the CMIS is that characteristics of the individual and characteristics of the medium jointly influence health information–seeking behavior. Specifically, the model considers antecedents in the seeker (demographics, personal experience, beliefs, and salience), the characteristics and perceived utility of the information carrier (eg, health information on a website), and the final health information–seeking action (eg, decision to see a doctor). Based on the Johnson and Meischke [25] classification, we renamed the category names to make it more suitable for an internet search and distributed the constructs according to this distribution. Demographics, personal experience, beliefs, and salience of the CMIS are subsumed under “personal factors” and “environmental factors.” Characteristics and utilities of the CMIS are classified under “online search factors.” Information-seeking actions of the CMIS correspond to the outcome category labeled “health decision making and behaviors.” Finally, we have added the relational categories “relational factors” and “search subject” to our model. These are unique to health information seeking by proxy.

One author (CK) scanned the papers for quantified variables and created a binary coding system (1=variable is investigated; 0=variable is not investigated) with definitions for 49 variable categories to examine the frequencies of considered variables in the whole study sample (Multimedia Appendix 4). A psychology bachelor’s student (AS) and 1 author (CK) coded the variables in the studies independently (Cohen κ=0.69).

A total of 33 studies met the inclusion criteria (Figure 1 [32]). All studies were cross-sectional. Studies were conducted using in-person questionnaires (n=23), online surveys (n=5), interviews (n=2), telephone surveys (n=2), and mailed questionnaires (n=1).

The papers were sorted into 3 groups based on the focus of the paper (Table 1). A total of 13 of the studies focused on OHIS related to a specific disease or disease cluster, including asthma [34], attention-deficit/hyperactivity disorder [35], brachial plexus birth palsies [36], congenital heart disease [33,37,38], diabetes [39], hearing loss [40], hydrocephalus [41], scoliosis [42], skin disorders [43], and rare diseases [44,45]. An additional 13 studies addressed specific circumstances, mainly prior to a surgical procedure [46-51], after childbirth [52], 24 hours before an emergency department visit [53], during stay in a neonatal intensive care unit [54], in a pediatric outpatient clinic [55,56], in palliative care [57], and regarding attitudes towards human papillomavirus vaccination [58]. In addition, 7 studies dealt with general OHIS without specified diseases or particular circumstances [14,59-64].

The samples from all studies included a total of 8665 participants and varied from a sample size of 70 [35] to 848 [58] participants, with a median of 209 participants (Table 2). A total of 26 out of 33 studies specified the proportion of mothers and fathers in the samples; with 4758 mothers and 1353 fathers, 77.86% were mothers (4758/6111). Of the 33 studies, 29 (88%) were conducted in the Western world, of which one-third of all studies (n=11) were conducted in the United States. Samples from other parts of the world included Nigeria [60], Singapore [51], Saudi Arabia [34], and Brazil [48].

Reported parental mean ages were all between 30 and 42 years, but only half of studies (n=16) reported ages of parent and child. The child’s age varied from neonates [37,54] to adults [40,42,44,45,55], but studies with reported mean ages or distributions consisted mainly of toddlers, preschoolers, and school-aged children aged 1 to 12 years. Adolescents were targeted in only one study explicitly [58]. Only 2 studies [34,62] differentiated between mothers’ and fathers’ demographic data and listed their information separately.

The samples consisted of highly educated parents, with more than 50% [33,34,36,41,44,48,54] and up to more than 75% of parents holding academic degrees [52,58,63], but 14 studies did not report any educational levels. The proportion of persons with only primary education varied between 0% [52] and 21.9% [57] among the studies that reported on education levels.

Table 3 presents the prevalences of OHIS by proxy and associated factors, separated into general OHIS and OHIS for specific conditions in the child. In studies that reported prevalence by parental OHIS in general (n=9), prevalence ranged from 52% to 98%. Only 3 studies explicitly distinguished between general OHIS and specific OHIS [36,53,62]. Recent studies from 2017 or later showed the highest prevalences, with around three-fourths [51,59,60] to roughly 9 out of 10 parents who searched for health information related to their child [14,33,35,44]. Likewise, the health-related internet use among parents of children with rare diseases seems to be relatively high [44,45], even for older studies that deal with rare conditions [36,37]. Most of the data are related to OHIS before or because of surgical intervention [46-51]. In those cases, the prevalence varied between 38% and 90%.

In Table 3, all variables investigated in relation to parental OHIS are reported. Only a small proportion of studies provided bivariate or multivariate analysis of associated factors with OHIS, often presenting only selective data with significant outcomes. Education was shown to be the most common associated factor with parental OHIS [34,36,40,41,47,53,54,56,57], although some studies found no significant association with education [35,46,62].

The gender of the parent was not related to whether a parent searched the internet for their child in most studies [35,42,53,54,57,62], but it was related in one study [33]. The influence of the age of the parents on OHIS was inconsistent; some studies found younger age to be associated [56,62], one found older age to be associated [53], and other studies found no association with age and search behavior [14,35,40,42,47,57].

Google was reported to be the most common starting point for gathering health information [33,42,44,48,50,51,53,62,63]. The most recent studies found that 9 out of 10 parents use Google [33,44,51] and many of the daily internet users go online via their mobile phones [45]. The rising trend in mobile phone use over desktop computer use was already evident in the studies since 2013 [41,44,45,47,52,53,56]. There is some evidence that smartphone owners are more likely to look for health-related information relating to their child [47,56] than people without a smartphone. Yardi et al [14] reported first that smartphones have overtaken the desktop computer as the most used device for proxy health information seeking.

Parents described information from the web mostly as helpful and useful, with a fundamentally positive attitude towards OHIS [33,34,37,38,41-52,56,57,63,64]. The most frequent underlying search motive was the need for a better understanding of the child's condition, which gave parents the opportunity to play a more active role in the management of their child's health [14,33,34,42,43,45,49,51,52,55].

Parents used the internet to decide if their child needed a doctor [14,60,63] and in some cases also in emergency situations [53]. Likewise, they searched the web before a doctor’s visit to prepare for the appointment and after the doctor’s visit to address unanswered questions [14,33,42]. The web was also reported to be used as a second medical opinion, but the amount varied from 1% to 57% across studies covering different medical circumstances [44,45,51,52].

Unfortunately, only a few studies examined search content in detail. Information about characteristics of specific diseases, current treatments, and diagnoses were the most common search topics [36,44,45,50,51,55,59], while looking for alternative treatments [40,44,50,51,55] was comparatively less common. However, the choice of treatment could be influenced by the information from the internet [46-48,56]. Studies that did not restrict their items on search content to a specific disease showed a greater variety of search content, including searches for health purposes like children’s nutrition [44,45,52,61] or development [44,52,61].

A consistent finding across studies over time was the search for or use of support groups [14,33,39-41,44,45,49-51,55,57]. In particular, parents of children with chronic, acute, or rare diseases showed a high need for support groups [40,41,44,45,57].

Parents perceived the information on the web as easy to understand [14,43,46-48], but studies found that parents sometimes had problems distinguishing between trusted and untrusted websites [14,33] or finding reliable information [54,60]. Some studies showed only a small proportion of parents who considered the reliability and trustworthiness of the information [14,49,51], while other studies showed greater skepticism of the participants towards the internet as a reliable source [41,54]. Further, parents did not necessarily navigate to the pages that they trusted or that provided trustworthy information [61]. The web as a trusted source was ranked lowest, but it is used frequently as a source of health information [46,54], and even unreliable information was reported as helpful [52,54].

Although parents wished for more guidance regarding good websites from their physicians [14,41,48,52,53], parents rarely or never discussed their findings with them [14,34,40,46-48,50,51]. Reasons for not discussing findings included a lack of time and a fear of doctors’ disapproval [33,51,52,59]. Other problems mentioned included nonnative language information [34,38], technical language [34,51], and information overload [33,38,51].

There were 6 studies that reported anxiety, distress, or worries caused by information from the internet [33,42,44,45,50,51]. The proportion of affected parents was between 14% [51] and 52% [45]. Nicholl et al [44] reported that online searching lowered anxiety in 16% of their participants, but the number of people with increased fear was twice as large. Likewise, attention should also be paid to parents who spend a lot of time searching the web for health purposes or who visit many different sites, like in Shah et al [36] or Porter and Edirippulige [40], which indicates that some parents may not be able to find the health information they need.

The conceptual model (Figure 2) consists of 6 categories: personal factors within the seeker, environmental factors, relational factors between seeker and search subject, factors within the search subject (ie, the child in regard to personal and health variables), online search factors (search channels, content, behaviors, and appraisals), and outcomes. Factors of the search subject and the relational factors are unique for OHIS by proxy.

The most commonly studied variables were the age of the parent (21/33), their gender (23/33), their education (21/33), and their access to the internet (25/33), as well as the age of the child (21/33). Attitude on usefulness of online health information was the only variable from the other categories that was frequently included in studies (22/33). The most neglected study variables were relational variables. However, variables from the category of online search factors were also rarely assessed, especially trustworthiness of information (2/33), search satisfaction during the search (2/33), and the perceived search success after the search (1/33). In addition, influencing factors such as trait health anxiety (0/33), health knowledge (2/33), and eHealth literacy (2/33) were also considered in less than 10% of the studies.

The surveyed variables of the studies that met the inclusion criteria for narrative synthesis (N=33) were counted. On average, studies considered 12.12 variables (SD 5.36) out of 49 coded variables (Multimedia Appendix 4). The study by Whyte and Hunter [64] presented the fewest number of variables (n=2) and Kasparian et al [33] considered the most variables (n=28).

Parents are heavy users of health-related information on the internet for their children across highly diverse circumstances. Across studies, results showed that the majority of parents have searched the web at least once for general health information for their child. This indicates that the percentage of parents who search for their child is well above the national averages for self-seekers [1]. Education was the most consistent predictor for proxy OHIS across all studies. Well-educated parents used the internet for their children more than parents with little education. The most recent studies show that Google search engine was used by almost all parents as a starting point for OHIS.

Further, information retrieved from the web was reported to be used by caregivers for decision making about children’s health. Physicians should be aware that parents reported using information found on the web under certain circumstances for treatment choice or to make health care utilization decisions. Although parents rarely or never discussed information from the web with their doctor, studies showed that they would like more support from doctors on how to find reliable sources. However, there is a tremendous lack of understanding about which criteria parents use to make decisions and about individual and external factors that contribute to parental empowerment. More research is needed on offering parents tangible knowledge and appropriate guidance, using the web for health purposes in an effective way, and problem solving approaches to the multifaceted problems that come up during or after OHIS for their child (eg, unmet information needs, obstacles in parent-doctor communication, false proxy lay diagnoses by parents with wrong conclusions for child’s needed treatment, unnecessary or missed doctor visits, and parental health anxiety due to online health information). All 6 studies that surveyed anxiety and distress by proxy showed a significant proportion of affected parents.

Overall, this review identified the need for more developed research in the area of OHIS. As can be seen from the results of the review, most studies provided descriptive information, but process- and theory-driven advances in this research area are still in infancy. To facilitate more systematic research in the area of OHIS, we present a summary of research gaps in the context of the conceptual framework provided. A total of 17 studies included in this review named research gaps explicitly or gave suggestions for further research. We synthesized these into the results of the developed model on proxy OHIS and the current research on proxy health searches. This is a framework that can be used for future studies.

This systematic review has several limitations. We included studies from a period of 17 years, while the manner of OHIS has undoubtedly changed much faster. The circumstances in which the studies were conducted may be difficult to compare due to differences at the point of data collection, geographical location, characteristics of the parents, and the underlying diseases of the children. Further, almost all studies were conducted in clinical settings, and the findings in this review may not generalize to other populations. There is a strong need for research on representative samples of parents. Estimates of the prevalence of proxy OHIS should be treated with caution, as it was often not consistently defined in the previous studies, with different time periods being queried and the health status of the child varying.

Our systematic review has important implications for future research. The results suggest that more studies on parental OHIS are needed to understand parental online search behaviors and support parents in their medical decision making by proxy. There is evidence that parental proxy OHIS is a very common but understudied behavior. Our presented agenda has highlighted research gaps that will hopefully lead to more systematic, theoretically informed research in this field.