Comparing adverse events data from systematic reviews and consumer social media posts




Poster session 1


Sunday 16 September 2018 - 12:30 to 14:00

All authors in correct order:

Golder S1, Smith K2, O'Connor K3, Abeed S3, Gonzalez-Hernandez G3
1 Cochrane Adverse Effects Methods Group, UK
2 Regis University, USA
3 University of Pennsylvania, USA
Presenting author and contact person

Presenting author:

Su Golder

Contact person:

Abstract text
Technological advances have enabled the searching of social media for healthcare consumers' posts on adverse events. However, validation of social media as a data source is required against other sources such as systematic reviews. Systematic reviews are a key source of adverse event data as they pool data, improving generalisability and decreasing uncertainty.

We sought to compare information on adverse events from systematic reviews and from healthcare consumer posts on social media.

We identified 38 reviews of adalimumab from Epistemonikos and DARE and 10,188 social media posts mentioning adalimumab. We compared the adverse event categories mentioned in systematic reviews and social media posts and the rank order of these adverse event categories. We then carried out a more detailed analysis with the included randomised controlled trials (RCTs) from the reviews that reported rates in the treatment and placebo arms.

Only 20 of the 38 reviews reported on named adverse effects with adalimumab monotherapy. The variety of statistical analysis (events per patient years, risk ratios, rates) and the lack of numerical data reported in the systematic reviews made comparisons difficult. A comparison of adverse events mentioned in these 20 systematic reviews with the top 16 adverse events categories identified from Twitter revealed that 'neurologic: anxiety/depression/insomnia/panic/mood' was the only category not covered in the systematic reviews despite ranking 5th on Twitter. We were only able to rank order adverse events within 10 systematic reviews. The rank order in these reviews differed from the social media posts (Table 1).

We carried out an analysis of the 11 RCTs included in the systematic reviews. Only 'infection' had complete data in all the RCTs and 'injection site' was reported in nine. One major issue, therefore, was that investigators may be primarily interested in determining efficacy outcomes hence adverse outcomes that rank lower than the most common ones (i.e. lower than infections and local reactions) may not be measured.

Reporting bias was prevalent, with reviews focusing on selected adverse events. These adverse events may not be the most salient according to healthcare consumer posts on social media.

Patient or healthcare consumer involvement:
For this stage of our project, we did not involve patients or consumers.


Relevance to patients and consumers: 

Adverse events are an important patient outcome. Information on adverse events helps improve informed balanced decision making. Identifying information on adverse events, however, can be challenging. Social media contains many mentions of adverse events posted directly by patients or consumers. This research finds differences in the adverse events reported in reviews and those posted by consumers. This enforces the need to prevent reporting bias in systematic reviews of adverse events and the potential value of social media to ascertain the most salient adverse events as recognised by healthcare consumers.