Celebrities’ impact on health-related knowledge, attitudes, behaviors, and status outcomes: protocol for a systematic review, meta-analysis, and meta-regression analysis

We will conduct this systematic review, meta-analysis, and meta-regression analysis adhering to the following protocol and will report any changes to the protocol that arise as we proceed. The methods and design of this systematic review protocol are in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA-P), available as an additional file to this protocol (see Additional file 1). This protocol is registered with PROSPERO (CRD42015019268).

Both quantitative and qualitative studies will be included in this systematic review to identify all empirical research evidence pertaining to the impact of celebrities on health-related outcomes.

All quantitative impact evaluations, including experiments (e.g., randomized controlled trials), quasi-experiments (e.g., interrupted time-series analyses), and observational designs (e.g., pooled time-series and cross-sectional analyses) will be included. Based on the results of our pilot search, we anticipate that most studies encountered will likely be observational in nature (see Additional file 2 for pilot search strategy). Within observational evaluations, both ecological studies (i.e., comparisons of groups rather than individuals) and individual-level studies (e.g., surveys of opinions) will be assessed. It is of value to include both ecological and individual-level studies since ecological studies alone face limitations with respect to identifying causal effects, whereas individual-level studies have an increased risk of reporting bias [25].

Qualitative studies will complement the knowledge gained from quantitative analyses, especially in providing rich insights into the contexts in which celebrities affect health-related outcomes. Qualitative research evidence will specifically allow us to better understand the meaning of the quantitative data, explore how individuals or groups may perceive certain celebrity interventions, and better understand the biological, psychological, and social mechanisms underlying the factors that mediate any influence.

We will only include studies that evaluate the impact of celebrity health activities on individuals or groups of individuals. Studies that evaluate the impact of celebrities on corporations, governments, organizations, or similar entities will not be considered.

Studies from all settings and countries will be considered, which is appropriate given that the influence of celebrities manifests itself in different places and across geography and culture.

Interventions include any health-related campaigns, news events, programs, or statements that primarily revolve around a celebrity or several celebrities—whether or not the celebrities intended to cause a health effect. Existing literature defines celebrity status in a number of ways, with a recent emphasis on celebrity as a type of capital that results from accumulated media visibility and recognizability [26, 27]. We aim to be liberal in our understanding of who are “celebrities,” with the recognition that various cultures may endow celebrity status in different ways and this may or may not be linked to media presence. Therefore, we chose to define celebrities as well-known and highly visible individuals in society, including but not limited to athletes, entertainment stars, media personalities, politicians, religious leaders, and socialites. Interventions will be included regardless of whether or not a celebrity played an active first-hand role in influencing a health-related outcome. In some cases, a celebrity may personally advocate for health-related behaviors; in other cases, a celebrity may be indirectly involved, such as a newspaper reporting on a particular health condition with which a celebrity was recently diagnosed. The goal is to broadly evaluate the impact of celebrities on health-related outcomes rather than only purposefully designed celebrity interventions that specifically aim to affect health-related outcomes.

Randomized controlled interventions are not necessary for a study to be included in this review. Examples of possible control interventions include, but are not limited to, comparator groups who were not exposed to the celebrity health activity. Studies without a control group will still be included in the review for the purposes of qualitative synthesis; however, only studies that assess the intervention in comparison to a control group will be included in the meta-analysis.

The outcomes of interest in this review can be divided into short-, medium-, and long-term effects of celebrity intervention on health-related outcomes. In the short-term, celebrities may alter the public’s health-related knowledge, such as by changing the public’s understanding of a certain disease’s etiology, risk, diagnosis, and treatment. In the medium-term, celebrities may change the health-related attitudes and/or behaviors of the public. The Health Belief Model proposed that individuals’ perceptions of susceptibility, severity, benefits, and barriers related to health-related issues are precursors for their “readiness to act” and perform various health-related behaviors [28]. As such, we defined “health-related attitudes” to encompass this range of perceptions that help to shape individuals’ intentions to perform health-related behaviors. For instance, fear of needles may be viewed as a health-related attitude that will influence whether an individual seeks vaccination. A health behavior was originally defined by Kebl and Cobb as any activity undertaken by a person who believes himself to be healthy for the purpose of preventing disease or detecting disease in an asymptomatic stage [29]. For the purposes of this review, we are interested in examining both positive and negative celebrity influences as it applies to all individuals, both healthy and unhealthy, and thus defined a “health-related behavior” as any action that may promote or diminish one’s health status, be that physical, mental, or emotional well-being. For example, the decision of a high-risk patient to be screened for cancer would be a positive health behavior that can potentially benefit long-term health status, whereas declining treatment contrary to medical advice would be a negative health behavior that can lead to diminished health status. Finally, in the long-term, health status outcomes such as the incidence and prognosis of a preventable disease will be considered. These long-term health status outcomes are this review’s primary outcomes of interest; however, the secondary short- and medium-term outcomes serve as helpful secondary surrogate measures in the meta-analysis and potential mediating factors for study in the meta-regression analysis. Collecting evidence of celebrity influence on a wide range of primary and secondary outcomes from short- to long-term will allow for a comprehensive assessment of celebrity influence on health. The logic model illustrates this rationale and the interaction between these outcome measures in determining short-, medium-, and long-term health-related outcomes (see Fig. 1).

We constructed a generic search strategy and adapted it for the MEDLINE database in consultation with health sciences and social sciences information specialists at McMaster University (see Table 2). With the goal of conducting the most comprehensive review possible, we aimed to maximize the sensitivity of our search by encompassing a broad range of potential celebrity interventions and health-related outcome measures in the search terms. We maintained specificity by taking a conservative approach to defining the inclusion and exclusion criteria (see “Selection of studies” section below).

We adapted the generic search strategy for the following ten electronic databases (see Additional file 3 for all adapted search strategies): CINAHL (1982 to July 2014); Communication Complete (1915 to July 2014); EMBASE (1947 to July 2014); Journals @ Scholars Portal (no date restrictions); OVID MEDLINE In-Process & Other Non-Indexed Citations (1946 to July 2014); Proquest Dissertations & Theses A&I electronic database (all dates); PsycINFO (1806 to July 2014); PubMed (1966 to July 2014); Social Sciences Citation Index (1976 to July 2014); and Sociological Abstracts (1952 to July 2014). We searched a combination of health and social sciences databases and additionally looked for gray literature. All searches were run on July 31, 2014. There were no language restrictions and the databases were searched from their earliest date of inception.

Two members of the research team (YM/NN) independently screened the title and abstract of each study for inclusion based on a pre-established eligibility screening form (Additional file 4). This eligibility criteria form was used to screen for both quantitative and qualitative studies and, moreover, helped to categorize which studies belonged to which study design. Articles were selected during the title/abstract screening if reviewers answered “yes” to all three questions:

Disagreements were resolved through discussion between the two reviewers. If there had been a case in which a consensus could not be reached, the principal investigator (SJH) would have been consulted, although that was never necessary. Studies that were eligible for inclusion based on title and abstract screening were then reviewed in a full-text screening exercise. The same inclusion criteria were used during full-text screening as the title/abstract screening but with two adjustments: first, the full-text screening required a precise examination of whether celebrity interventions were being evaluated and if a health-related knowledge, attitude, behavior, or status outcome was being measured; and second, we narrowed our third eligibility criteria to only include empirical studies that had been peer-reviewed. Doing so enhanced the specificity of our search, allowing us to exclude studies that were likely to be less robust and credible. Again, the two independent reviewers used an eligibility screening form to determine which studies met the full-text inclusion criteria (see Additional file 5 for the full-text screening form). Disagreements were resolved through discussion; if consensus could not be reached, the principal investigator would have been consulted, although that was never necessary. Figure 2 visualizes the first part of a PRISMA flowchart showing the number of articles included and excluded at each stage of screening.

Given the likely variability among studies and the challenges we anticipate in developing a standardized form that can be applied to all included studies, three stages of data extraction will be conducted. In the first stage, general study information will be gathered including details about participant demographics, interventions, and methods employed (see Additional file 6 for a draft form for stage one of data extraction). This will help make data extraction more efficient by grouping together studies with similar study designs. The next phase of data extraction will focus specifically on defining outcome variables. This information will be used to further group studies by outcomes, thus facilitating the development of a third data extraction form specifically tailored to each subgroup of studies. The third data extraction form will focus on collecting precise quantitative data and key findings specific to the outcome measures reported in the second data extraction form. As there will be a large number of studies in this review, data extraction will be completed by two pairs of extractors (four reviewers total), with each pair randomly assigned half of the studies.

The three data extraction forms will be developed through a series of calibration exercises involving the investigators, data extractors, and a statistician. Specifically, each extractor will independently extract data from three purposively selected studies with diverse methodologies and celebrity activities. After data extraction is performed, discrepancies will be discussed until full agreement is reached among the extractors. We will also ask each extractor for feedback to modify the data extraction forms such that they are user-friendly and applicable to the wide range of studies that are included. These calibration exercises will be repeated until all four reviewers agree on a reliable set of data extraction forms and achieve near-perfect consistency in the extraction process. After finalizing the standardized data extraction forms, they will be implemented in Distiller SR online software to maximize efficiency of the data extraction process. Similar to the calibration exercises, disagreements will be resolved by discussion between reviewers to achieve consensus and consultation with the principal investigator if necessary.

Two extractors will independently describe the main findings from the included qualitative studies using a standardized form implemented in Distiller SR. Extractors will record specific text fragments from the studies wherever possible and appropriate. In line with our third research question, we will also extract information on any factors identified as mediators of celebrities’ impact on health-related outcomes and make note of any themes or trends that are uncovered. Overall themes across all included qualitative studies will be allowed to emerge naturally from the data in keeping with thematic synthesis methodology; they will not be pre-identified by a priori hypotheses [30, 31]. Any discrepancies about the details and conclusions from qualitative studies will be addressed through discussion between reviewers, consulting the principal investigator if consensus is not possible.

Studies that met the inclusion criteria will be assessed for bias using either the Cochrane Risk of Bias Tool for Randomized Trials or the Cochrane Risk of Bias in Non-Randomized Studies for Interventions (ROBINS-I) tool, depending on study design [32, 33]. Although the kinds of biases are broadly equivalent for randomized and non-randomized studies—namely selection, confounding, group equivalence, spill-overs, and reporting biases—there are important differences in their operationalization. Raters will resolve disagreements by discussion and through consultation with the principal investigator if necessary. Qualitative studies will be assessed for quality using the Critical Appraisal Skills Programme (CASP) approach, given its widely cited use in qualitative reviews and recommendation from the Cochrane Qualitative Research Methods Group. We will explore reporting biases such as publication bias by using funnel plots and assessing the asymmetry both visually and using the Egger test [34]. In the presence of visual asymmetry, exploratory analyses will be performed.

Studies will first be categorized by outcome domain (short-term, medium-term, and long-term health-related outcomes). Within each outcome domain, three types of analysis will be conducted, resulting in at least three sets of results: 1) summaries of studies demonstrating that celebrities have positive, negative, mixed, or no impacts on the health-related outcome; 2) synthesized quantitative measurements of the direction and magnitude of any identified impact that celebrities have on the health-related outcome; and 3) evidence identifying the comparative influence of different factors, circumstances, or conditions that mediate whether celebrities have positive or negative effects on the health-related outcome. A meta-analysis will facilitate all three lines of inquiry, while the subgroup analyses, meta-regression analysis, and the qualitative review will contribute to the third line of inquiry.