This study is a cross-sectional study using a systematic approach to search and review social media posts about physical therapy interventions. Therefore, the study followed the Cochrane Collaboration recommendations14 to systematically select information, reported the finding following recommendations for systematic review report,15 and was prospectively registered with the PROSPERO register database (CRD42021276941).

The searches were performed on the social media platforms Instagram and Twitter during the month of October 2021 (14th and 15th October for searches in Portuguese, and 18th and 19th October for searches in English) by two independent researchers (Fig. 1). A new user profile was created just for the searches in this study to minimize the influence of the social media algorithm for targeting content. The searches were performed using the hashtag “#reabilitação” for posts in Portuguese, and “#rehabilitation” for posts in English. No limit for publication period was applied.

Fig. 1.

Flow diagram of post selection process.

(0.26MB).

As we intended to include at least 100 posts per platform in each language, we reran the search on Twitter for posts in Portuguese on 26th October 2021, considering we included just 86 posts from the first search (Fig. 1).

To be included, the posts had to (i) be written in Portuguese or English; (ii) include, in any location of the post (caption, text, hashtags, or images), the terms “fisioterapia”, “fisioterapeuta” and/or “fisio” for posts in Portuguese, and “physical therapy”, “physical therapist”, “physiotherapy”, “physiotherapist” and/or “physio” for posts in English; (iii) suggest, explain, or present at least one physical therapy intervention; and (iv) report some objective, purpose, or focus of the discussed intervention targeting a condition or an outcome. Posts were excluded if they (i) were expressly intended to endorse, sell, or offer services, products, or lectures/courses; (ii) presented interventions outside the physical therapy scope, e.g. prescription of medication or surgery techniques; (iii) were from scientific journals profiles; (iv) and/or only shared links from other social media. In the case of reposts, the original post was searched and analyzed for eligibility. English language was selected considering that the majority of scientific evidence is published in this language, and a secondary language (Portuguese) was included because language is an important barrier for the understanding of scientific information12 and the language translation is likely a factor influencing the consistency of information.16

• -

  1st step: the posts reached using #reabilitação/#rehabilitation were sorted by publication date (most recent to oldest), and each independent researcher pre-selected the most recent 150 posts for both social media platforms, and both languages, applying the eligibility criteria (Fig. 1). Twitter was the exception, each reviewer pre-selected 100 posts in Portuguese (Fig. 1). This platform presented less eligible content in Portuguese, so we reached outdated posts (early 2000s). Therefore, we included fewer posts to try and maintain a chronological uniformity.

• -

  2nd step: another two independent researchers confirmed the eligibility of pre-selected posts and excluded duplicates. In cases of inconsistency in the selection, a fifth researcher was consulted to resolve the differences. During this step we also verified the number of posts per profile. We set a limit of a maximum of four most recent posts per profile to reach a greater diversity of content authors.

Data extraction was performed by two independent researchers and coordinated to be realized in the same day. Conflicts were resolved by consensus. Data extracted from the post were: publication date; number of likes, comments, and shares (when available); description of the interventions; objective of the intervention; references cited supporting the information; presence of endorsement or advertisement of products or services. Data extracted from the profile which posted the content were: number of followers; professional information disclosed in the profile description; and presence of endorsement or advertising in other posts in the last two years from the date of the post selected.

The analyses and classifications were performed by two independent researchers, and when consensus was not reached, a third researcher was consulted.

We analyzed whether the posts cited scientific references as basis for their information.

We considered a potential COI if the analyzed post included professional contacts for selling products or services; or if in the past two years, the profile had published posts endorsing, advertising, or encouraging the consumption of products or services offered by them or in partnerships with organizations or other professionals. We also considered the presence of a potential COI in posts presenting only the positive outcomes from the cited reference, omitting negative and/or non- significant results (selection bias).

The feature of the publication regarding information dissemination was classified as:

• -

  Facilitated information: when the profile was not considered to have COI, and the post content was developed to facilitate knowledge acquisition of the referenced scientific information. For this, the posts could use different tools, such as infographics, illustrations, or videos.

• -

  Facilitated information with potential COI: the post content was developed to facilitate knowledge acquisition of the referenced scientific information; however, the post or profile presented indications of having potential COI.

• -

  Simple dissemination: the post or profile did not present indications of having potential COI, yet it did not facilitate knowledge acquisition of the referenced scientific information; or it facilitated the understanding, but the content was not based on scientific references.

• -

  Simple dissemination with potential COI: the post or profile presented indications of having potential COI and did not facilitate knowledge acquisition of the referenced scientific information; or it facilitated the understanding, but the content was not based on scientific references.

• -

  Self-promotion: posts which presented scientific information (or not), with or without facilitation of the understanding, and directly associated professional contacts with the content disclosed, such as clinic/clinician name and/or brand, telephone, e-mail, etc. Posts presenting pre-post treatment performed by the profile's owner were also considered self-promotion.

To verify the reach of the posts and engagement of followers, the absolute number of followers, likes, comments and shares were analyzed.

We analyzed the following aspects:

• -

  Level of evidence regarding intervention: the Oxford centre for Evidence-based Medicine scale was applied considering the methodological design used in the references.17

• -

  Methodological quality: the scales were applied according to the methodological design. Randomized controlled trials were assessed using PEDro scale (good-quality ≥6 points, low-quality ≤5 points)18-20; Systematic reviews were assessed using the tool AMSTAR II (good-quality = none, one or more non-critical weakness, low-quality = one or more critical flaws)21; Observational studies were assessed via Epidemiological Appraisal Instrument (EIA) (good-quality >1 point; low-quality ≤1 point)22,23; and Clinical guidelines were assessed via the tool AGREE II (good-quality ≥50% at domain 3, low-quality <50% at domain 3).24,25 Other designs were not assessed considering the low relevance for intervention evidence.

• -

  Impact of the references: was assessed using the greater citation numbers on Scopus or Web of Science database, and the Altmetric score.

• -

  Consistency of the information: verified whether the post information was consistent with the cited reference. Posts were classified as ‘accurate’ when the post information matched the reference; ‘inaccurate’ when the post information was not supported by the reference; and ‘partial’ when the information agreed with the reference, however, the post (i) did not present all the primary results of the study; (ii) presented only image from manuscript omitting important parts; or (iii) did not present all the results that fulfill the aims mentioned in the original reference, when primary outcomes were not determined.

Data were analyzed descriptively using measures of central tendency and variability, or frequencies, percentages, and absolute values. After checking the primary data, in a post hoc analysis, we aimed to verify whether (i) posts using scientific references reach more people or engage more followers compared to posts without references; (ii) posts of profiles with potential COI reach more people or engage more followers compared to posts of profiles without COI; (iii) the non-use of scientific reference is related to profiles with COI; and (iv) facilitation of knowledge acquisition is related to scientific references of good quality. Therefore, the Mann-Whitney test was used to verify whether there were differences regarding the number of followers and likes between posts that cited or not scientific reference and between profiles with or without COI. Data were analyzed using R statistical software 4.1.1 (packages stats and pairwiseCI). Additionally, a Pearson's chi-square test was used to verify the association between the use of scientific reference and presence of COI, and between the type of dissemination and the quality of reference used. Data were analyzed using SPSS version 26.0 (IBM SPSS Inc., Chicago, IL, USA). The confidence level was set at 5% for all the analyses.

From 1,145 pre-selected posts, 632 posts were included in the analysis, of which 339 posts were in English and 293 in Portuguese (Fig. 1). The included posts were published between June 2011 and September 2021. The most cited intervention was exercise (Fig. 2A), and the main target of the interventions was pain management (Fig. 2B). The dataset used in the analyses is available in the supplementary material 1.

Fig. 2.

A – Interventions included in social media posts; B – Targets of the interventions included in social media posts.

(0.41MB).

From 632 posts, 14% (n = 89) cited references in which their information was based, and 86% (n = 543) did not cite any source of information (Fig. 3A).

Fig. 3.

Data on use of references, potential conflict of interest and communication aspects (n = 632).

(0.21MB).

We identified that 57% (n = 362) of the posts/profiles presented materials which suggest potential COI (Fig. 3B and supplementary material 2 [Table 2.A]). For selection bias, 6% (n = 6) of the posts presented only the positive results from the cited references (supplementary material 2 [Table 2.C]).

Most included posts were characterized as self-promotion post (43%; n = 270), with only 9% (n = 51) using approaches to facilitate knowledge acquisition of the scientific information disclosed (Fig. 3C and supplementary material 2 [Table 2.A]).

The social media profiles posting information on physical therapy interventions had a mean ± SD of 5,162 ± 37,240 followers (Table 1). In total, the posts received a mean of 88 ± 593 likes, 2 ± 9 comments, and 1 ± 7 share (Table 1). The results for each language and social media are presented in supplementary material 2 (Table 2.B).

One hundred and one references were cited as source of information. We could not access the full text of six references (three original manuscripts, two books, and one undergraduate thesis). The characteristics of references used in the included posts are presented in Fig. 4 and supplementary material 2 (Table 2.C).

Fig. 4.

Data on references analysis.

(0.25MB).

The most frequent methodological design was systematic reviews (33%) (Fig. 4A). In addition, the most frequent level of evidence based on the Oxford scale was 1a (20%) which indicates results from systematic reviews (Fig. 4B).17 Regarding the methodological quality, most references were of poor quality (39%) (Fig. 4C). The average number of citations on Scopus or Web of Science was 25 (SD=58, median=2, min-max values=0–417) and the average Altmetric score was 152 (SD=558, median=31, min-max values=0–5,188). When comparing the information presented in the posts with the information from references, half of the posts presented accurate information (51%) (Fig. 4D).

Details of the post hoc analysis are presented in the supplementary material 2 (Tables 2.D–G). The post hoc analysis indicated that posts using scientific references as source of information did not reach more people via the number of followers compared to posts without references (median difference= –73 followers, 95% CI: –297, 302). In contrast, posts without references caught more attention of the audience receiving more likes than posts which used references as source of information (median difference= –6 likes, 95% CI: –8, –2).

Profiles presenting potential COI did not reach higher numbers of followers (median difference= 201 followers, 95% CI: –2,356) or receive more likes (median difference= 0 likes, 95% CI: –4, 3) compared to profiles without evident COI.

Posts with no reference as source of information were associated to the presence of a potential COI (Chi-Square, p<0.001; Phi=0.248), and posts which used some approach to facilitate knowledge acquisition were associated to information based in studies with good methodological quality (Chi-Square, p = 0.02, Phi=0.288).

We cannot affirm that the search strategy was specific enough to capture all posts regarding physical therapy interventions on the platforms. Our decision regarding the search terms was based on a pilot study: i) we searched different hashtags, and #rehabilitation/#reabiltitação presented the greatest number of linked posts; ii) posts linked to #rehabilitation/#reabilitação were most likely to fulfill the eligibility criteria; and iii) most posts linked to abbreviated hashtags, such as #physio, presented content related to activities of daily living without technical information. Therefore, we chose the search strategy to facilitate a systematic search and improve our chances to reach posts that discussed physical therapy interventions. The used metrics to check social media engagement is another potential limitation, and we acknowledge that they do not necessarily represent real engagement with the presented information. We included posts from two social media platforms, and the non-inclusion of others, such as Facebook, is likely another study limitation. Nevertheless, besides the limitations, our results properly reflect the content which is available for general population of physical therapists.

Most of the analyzed posts were missing vital information needed to be trustworthy and to report useful information for supporting clinical decision. The lack of important information to properly understand interventions is not a flaw unique to social media content producers. Recent studies reported that infographics published by scientific journals do not report sufficient information to allow professionals to make clinical decisions,30 and official government websites present inaccurate content about the management of low back pain.31 Therefore, during the process of knowledge acquisition and clinical making decisions, it is crucial to search for trustful sources with complete information. This does not mean that social media should be disregarded. However, consuming information on social media requires caution. Readers should check the presented information references and be aware of the methodological quality, COI, studied population, intervention description, comparators, outcomes, clinical meaningfulness, harms, and others.30,32,33 Hence, it is advisable to look for high-quality studies and read the full-text version. We acknowledge that the access and interpretation of scientific papers are important barriers mentioned by physical therapists.34 Nevertheless, there are valuable sources, such as PEDro database (https://pedro.org.au/), that provide the tools to facilitate the understanding of research and trials in physical therapy.