Patients who monitor online media for health information may experience frequent exposure to exaggeration and misrepresentation of medical science [1-5]. Two typical examples of such infelicitous reporting are the depiction of observed correlations as causal connections—for example, between lifestyle behaviors and disease outcomes—and the inflation of preclinical animal testing results, often followed by the inference of these to humans [6,7]. This is misleading when one considers that about 88% of the pharmaceutical developments that reach the first human trials will never reach the phase of market approval [8]. Such misrepresentations are present in numerous easily accessible health news sites and are spread freely on social media such as Facebook.

Earlier research by our group, on the media coverage of innovative diabetes therapies, found that 83% of Dutch newspaper reports about innovative diabetes treatments lack any reference to clinical trials in humans [9]. Similarly, in the United States, a study on health news appraisals found that most authors do not satisfactorily discuss the quality of the evidence [6].

Although, to our knowledge, there is no literature on the effects of news reporting on the patients’ attitudes, it is highly plausible that messages about promising future treatments could affect the readers’ sentiments such as enthusiasm and curiosity. Moreover, a patient’s level of hope may increase, which is positive as having hope is associated with more favorable diabetes outcomes [10,11]. The effects may also turn out to be negative when, for example, feelings of impatience or disbelief are more prominent.

Overall, 3 aspects of Web-based reporting may influence attitudes. First, the tone of the reports, using intensified language (eg, revolutionary and breakthrough), a common phenomenon in health news coverage [12], may amplify these sentiments.

Second, attitudes may be affected by the references to an innovation’s developmental phase. Important innovations are covered for many years and during different research stages. For the readers, it may remain unclear as to how long it would take for the innovation to be available in clinical practice. An example is the artificial pancreas, a concept for the treatment of diabetes that has been reported since 1972 [13]. Third, reports on future cure-focused innovations, such as pancreatic cell transplantation for diabetes, may potentially have a stronger impact on the readers’ sentiments than news about noncure-focused treatments.

To increase the understanding of the associations between the characteristics of news about future treatments for chronic illnesses and the readers’ sentiments, we assessed the posts about diabetes research on Facebook pages, together with the corresponding user comments.

A retrospective observational study was performed on a corpus of diabetes news messages posted on publicly accessible Facebook pages between January 1, 2014, and January 1, 2018, and the associated reader comments. Facebook pages enable public figures and businesses to create a public presence on Facebook. Every person on Facebook can connect with these pages by liking them, after which they receive updates in their news feed and can interact with them [14]. The 2 most-followed publicly accessible diabetes pages in the Netherlands were selected: (1) Juvenile Diabetes Research Foundation (JDRF) Nederland [15], the Dutch division of an international type 1 diabetes research foundation, with over 27,000 Facebook followers and (2) Diabetes Fonds [16], a Dutch charity funding of research on all types of diabetes, with over 36,000 Facebook followers.

Data extraction and preparation comprised multiple steps (Figure 1). First, all news posts and associated readers’ comments were extracted for the 4 years from January 1, 2014, to January 1, 2018. The Facepager tool, version 3.8.2., developed by Jünger and Keyling [17], was used to scrape the publicly available data from the Facebook pages, including all reader comments. Replies were excluded (ie, comments on comments) as their content and sentiments are influenced by the initial comments on the news posts. Second, all nonscience news–related posts were identified and removed from the corpus. Furthermore, 3 criteria for diabetes research posts were applied: (1) it must contain information about the development of an innovative therapy, technique, product, instrument, or insights into preventive behaviors; (2) it must contain a reference to a traceable scientist or scientific institution (including medical companies); and (3) it must refer to an innovation which is not (or not yet) applied in the Dutch standard diabetes care. Therefore, nonmedical and nonscientific topics (eg, personal experiences, practical tips, travel stories, and fundraising) were excluded. The consensus between 2 raters in a subsample of 14.99% (271/1808) of the comments on initial Facebook posts was used to resolve disagreements. The third step was to extract the source message (eg, Web-based newspaper item) whenever a hyperlink was available and to merge it with the post content.

All user comments were evaluated by 2 raters as to whether they contained sentiments (positive, negative, or both). Focus was put on both textual expressions (ie, words and sentences) and the use of emoticons [18,19]. The initial interrater reliability in a subsample of 15% of the sentiments was high (kappa=0.80). For the remainder of the sample, any disagreements were resolved by consensus and, where uncertainties remained, by a third reviewer.

The IBM SPSS Statistics program, version 25, was used to evaluate the differences in the probabilities that sentiments (positive, negative, and hopefulness) were reflected in the user comments, depending on the developmental phase, intended therapeutic effect, and the presence of language intensifiers in the text. Crude and mutually adjusted binary logistic regression models were used to calculate odds ratios (ORs) and 95% CIs. Furthermore, it was assessed whether Facebook pages ID, commenter ID, and gender data contributed to the logistic regression models.

Table 2 shows that between January 1, 2014, and January 1, 2018, a total of 173 diabetes news messages about innovative methods to treat diabetes were posted on the 2 largest publicly accessible Facebook pages in the Netherlands. These posts evoked 3710 reader comments, containing a total of 2727 positive, 880 negative, and 363 neutral sentiments and 513 verbal expressions of having hope.

First, it was tested whether the news messages about the innovative ways to cure diabetes were associated with different sentiments than the news messages related to other therapeutic effects. Table 3 shows that diabetes prevention (OR 0.55, 95% CI 0.37-0.84), improved blood glucose regulation (OR 0.68, 95% CI 0.56-0.84), and symptom relief (OR 0.31, 95% CI 0.21-0.44) were associated with less positive sentiments as compared with potential diabetes cures. Moreover, Table 3 shows that the analyses of negative sentiments show a similar pattern, although this was only significant in blood glucose regulation (OR 1.38, 95% CI 1.12-1.70; for being associated with more negative sentiments). Table 4 shows that, in line with the readers’ positive sentiments, hopefulness was most frequently expressed when Facebook news reported on cure-focused therapies.

Furthermore, it was examined whether the commenters’ sentiments were related to the covered innovations’ developmental phases. Tables 3 and 4 show that, compared with the success in the larger patient trials, evidence from the preclinical phases led to more positive sentiments (OR 1.46, 95% CI 1.07-1.99). Earlier observational (OR 0.31, 95% CI 0.15-0.66) and fundamental findings (OR 0.71, 95% CI 0.53-0.95), however, led to less positive sentiments. Tables 3 and 4 also show that this sentiment pattern was, for the most part, confirmed in the analysis of negative sentiments and the expressions of hopefulness.

It was examined whether the intensification of language was associated with sentiments and hopefulness. However, Tables 3 and 4 show that there were no significant relationships between the language intensification of Facebook posts about diabetes research and the sentiments (positive, negative, and hopefulness) of those who reacted to it on Facebook.

To verify the robustness of our findings, additional variables and levels were tested for any effect on our regression analysis. It was found that the Facebook pages ID (JDRF Nederland vs Diabetes Fonds) did not contribute to the regression models. Furthermore, data on the commenters’ gender were available for 71% (2634/3710) of the comments with identifiable sentiment (the first batch of 2 data extractions). Analysis of this sample showed that controlling for gender did not greatly alter the patterns and magnitudes of our results. In addition, the necessity to include commenter ID as a level in our regression model was rejected owing to the flat distribution of comments by the commenters in the same subsample: 80% (1495/1870) of the commenters commented once only.

Our final model only contained the 3 main independent variables: therapeutic effect, developmental phase, and language intensification.

In this analysis of 4 years of Facebook posts and comments on diabetes news and user sentiments, posts about potentially curative innovations were associated with more positive general sentiments than the posts not about potential cures, as expected.

However, unexpectedly, innovations supported by evidence from phases just before human clinical trials showed the strongest positive association with improved general sentiments. The observational research results were associated with the most negative general sentiments in the user comments. In addition, and contrary to our expectations, this study found no evidence for the associations between language intensity and the readers’ sentiments.

A strong positive association was found between cure-focused innovations and positive sentiments. The explanation for this finding is likely to be the absence of cure-focused therapies, to date, for the debilitating disease that diabetes still is. However, negative sentiments may also be provoked by cure-focused innovation, for example, when frustrations about perceived false promises have the upper hand. The strong negative association between the written expression of hopefulness and the news about noncure-focused innovations suggests that the concept of hopefulness only plays a role regarding the desire for a cure.

When looking at the developmental phases, general sentiments were most positively associated with positive results in the preclinical phases closely before human trials. This finding conflicts with the scientific standard that the proof of concept is demonstrated by doing randomized clinical trials. Although it was not assessed by us, an explanation for the preclinical positivity may be the overly optimistic way in which news outlets frequently cover animal studies [30].

Negative associations of the fundamental research with positive sentiments can be explained using construal level theory [31]. First, as the success of therapies in the earliest research stages is far away in time, the patient’s thinking about these innovations becomes more abstract and the consequent anticipation may decrease. Frequently, the medical applicability of very early stage therapies is indeed abstract. A second explanation may be that bad personal experiences, with waiting for other cure-focused innovations, affect the so-called experiential distance (ie, perception of the chance that treatment may become a reality, based on earlier experiences).

Specific research topics may explain the strong association between observational research and the less positive general sentiments. Both disbelief and powerlessness in readers may have arisen from Facebook posts that describe how patients should have behaved in the past to prevent their disease. Furthermore, the association between the less positive general sentiments and the news about near-market innovations may be related to fears and frustrations regarding low availability and the doubt on medical insurance coverage.

A possible explanation for the absence of associations between language intensifiers in the news content and the sentiments may have its origin in the characteristics of our target population. Patients and others interested in diabetes seemed to be able to distinguish between the objective content and the subjective use of language. People commenting on the investigated diabetes pages are involved in the burdens of the chronic disease—enduring much and awaiting therapies for many years—and their emotions may not (or may no longer) be as affected by the subjective language.

Previous studies suggest that exaggeration of medical research is a problem [1-7]. Our study shows no evidence that language intensifiers are associated with the sentiments of the online diabetes populations. However, improved positive sentiments were found regarding the preclinical trials—just before evidence from humans—that give reason to suspect an undesired effect of health news exaggeration. The Facebook comments were enthusiastic and full of hope, despite the fact that about 88% of the pharmaceutical developments that reach the first human trials will never reach the phase of market approval [8].

These findings suggest that exaggeration is not limited to language intensification and other verbal inflation of research findings. It is the sheer coverage frequency of specific health research that may lead to positivity and hope, which is not always justified. The mental shortcut availability heuristic relies on the immediate examples that come to a person’s mind, possibly putting too much weight on medical information when they read about it frequently [32]. At the same time, the importance of having hope when suffering from a chronic disease must not be underestimated. Hope mediates the relationship between psychological distress and health status and is an essential factor to cope with a disease [24,25].

High-quality health information is increasingly important. Responsible news authors must give context, interpret scientific findings, filter what is important to their target group, and act as an honest and valid conduit, especially as the role of social media is increasing every year [33]. Furthermore, the specific importance of social media to patients must be emphasized. Platforms such as Facebook or Twitter provide tailored information, increase the accessibility of news, and function as social and emotional peer supporters [34-36]. Being well-informed about scientific developments fulfills an essential need for the health information–monitoring patient [37], and it is known that the patients’ subjective well-being also clearly and positively affects health and all-cause mortality [38].

To our knowledge, our study is the first to quantitatively investigate the associations between the health news characteristics and the sentiments of readers dealing with chronic illnesses.

Moreover, an extra focus was put on the written expressions of being hopeful, enabling the confirmation of general sentiment associations in 1 specific disease-related sentiment. By assessing common language intensifiers, it was possible to differentiate between the objective characteristics of the news posts and a subjective language component. Our large sample size enabled us to mutually adjust the 3 news-related variables. Moreover, the validity of our sentiment outcome increased as offhand comments were observed, written down in an unforced situation. By using a Python regular expression search, the reliability of finding all language intensifiers was high. In addition, the kappa values for rating sentiments and coding message characteristics were high, and coding consensus was achieved regarding occasional discrepancies.

This study does, however, have limitations. First, it is not known what our population’s exact proportion sizes are regarding the patients, their social context, and others who were perhaps only momentarily interested in diabetes and left a comment. Moreover, although the language intensifiers were included, other journalistic language elements, such as emotionalization of news, were not included as a potentially predicting or modifying factor. One final limitation of the study may be the bias that theoretically occurs when readers with either very strong (rejecting) or neutral sentiments refrain from commenting owing to the sentiment itself.

By observing the news posts and comments on diabetes research on 2 large Dutch Facebook pages, we found that the readers’ sentiments are associated with both the innovations’ developmental phase and the intended therapeutic effect. However, no evidence was found on the association between sentiments and the presence of commonly used language intensifiers in the health news texts.

Our finding that comments on diabetes news on Facebook have the most positive sentiment, and most frequently express hopefulness when clinical efficacy is not yet proven, suggests that the news authors and editors must exercise caution when acting as a conduit for medical research news. More experimental research is necessary, in various populations, to determine a healthy balance between being optimally informed and avoiding having false hope.