Can Positive Framing Reduce Nocebo Side Effects? Current Evidence and Recommendation for Future Research

Although critical for informed consent, side effect warnings can contribute directly to poorer patient outcomes because they often induce negative expectations that trigger nocebo side effects. Communication strategies that reduce the development of nocebo side effects whilst maintaining informed consent are therefore of considerable interest. We reviewed theoretical and empirical evidence for the use of framing strategies to achieve this. Framing refers to the way in which information about the likelihood or significance of side effects is presented (e.g., negative frame: 30% will experience headache vs. positive frame: 70% will not experience headache), with the rationale that positively framing such information could diminish nocebo side effects. Relatively few empirical studies (k = 6) have tested whether framing strategies can reduce nocebo side effects. Of these, four used attribute framing and two message framing. All but one of the studies found a significant framing effect on at least one aspect of side effects (e.g., experience, attribution, threat), suggesting that framing is a promising strategy for reducing nocebo effects. However, our review also revealed some important open questions regarding these types of framing effects, including, the best method of communicating side effects (written, oral, pictorial), optimal statistical presentation (e.g., frequencies vs. percentages), whether framing affects perceived absolute risk of side effects, and what psychological mechanisms underlie framing effects. Future research that addresses these open questions will be vital for understanding the circumstances in which framing are most likely to be effective.

To be included, studies needed to have administered an ostensible treatment or medication (either active or placebo) and implemented a positive side effect framing manipulation and compared it with at least one other frame. Positive side effect framing manipulations were broadly defined as any in which information provided to participants was altered to accentuate the likelihood or significance of side effects in a positive light. In terms of dependent variables, studies must have measured the frequency and/or intensity of side effects and have reported appropriate descriptive statistics from which to calculate an effect size. To broaden the scope of the review, measurement could include the general experience of symptoms and/or those specifically attributed to treatment. Data collected from previously published, adapted, and study-specific questionnaires were considered. However, studies must have reported the frequency or intensity of those side effects framed during the consent process.

Study Coding
Citations (n = 255) were downloaded to Endnote (Endnote X8 Thomson Reuters, USA) where duplicates (n = 108) were removed. References were then exported to Rayyan (Ouzzani et al., 2016) where titles and abstracts were independently categorised according to inclusion and exclusion criteria by the lead author (K.B.) and an additional researcher not otherwise involved in the review (J.D., acknowledgements). At this stage, seven potentially relevant studies were identified. In a subsequent step, additional citations were sought by searching both the reference lists and published citations (via Google scholar) of the papers that appeared to fit the search criteria. This identified one further citation. The full text of the eight identified studies was subsequently assessed and two citations removed due to not adhering to search criteria. Both reviewers were in full agreement regarding included and excluded studies.
We note here that, while a criterion for inclusion was reporting of framed side effects, reporting was potentially ambiguous in one study (Fernandez et al., 2019). In this instance, eight side effects were mentioned during consent, but framing appeared to apply to any side effect experienced (wording: if you do experience a side effect, you might take this as a signal or reminder that the medication is absorbed and active in your body). We interpret this liberally as framing any side effect that might be experienced, thus legitimizing data extraction from the full GASE measure (Rief et al., 2011), which was not separated into those side effects listed and those not.

Data Extraction
For each included study, data was extracted pertaining to the following categories: sample characteristics, type of treatment administered, type of framing manipulation employed, communication method used when delivering framed information, statistical presentation of framed side effects, number of framed and non-framed side effects, outcome measure employed to assess side effects, whether side effects were measured generally or attributed to the treatment (for the largest effect only), measures used to assess expectancy for side effects, measures used to assess anxiety.
Statistical data was extracted for the primary outcome (frequency / intensity of framed side effects), and three secondary outcomes where available (frequency / intensity of non-framed side effects, effect of frame on side effect expectancy and effect of anxiety). In keeping with previous reviews concerning the effect of framing on health-related outcomes (O' Keefe and Jensen, 2007;2009;Gallagher and Updegraff, 2012), Pearson's r was selected as the effect size. Where several results were reported for the primary outcome (e.g. frequency and intensity of side effect), we report the largest effect. We decided not to compute an averaged effect size because we did not perform quantitative synthesis and, given the review was the first of its type, we wanted to discuss the maximum potential impact of framing. We note, however, that overall combined effects are likely to be smaller. In some articles, Pearson's r was directly available. In other cases, means and standard deviations, Chi-square statistics or odds ratios were reported (for further information see Table 1). In these instances, the effect size was estimated from formulas suggested in the literature.

Quality Assessment
The intention of the review was not specifically to identify randomised controlled trials (RCTs), but any experimental evidence assessing the role of framing. In fact, only one study (Webster et al., 2018) reported being designed as an RCT. Placing weight on quality assessment designed for RCTs may therefore be considered stringent. For full transparency, we present data below based on the Cochrane Collaboration's Risk of Bias tool (Higgins et al., 2011), but allow the reader to form their own opinion regarding publication quality based on experiment type. Categories were assessed by the lead author (K.B.) and an additional researcher not otherwise involved in the review (J.D.). Discrepancies were resolved by a third researcher (B.C.).  Table (where the plus symbol represents 'low risk of bias', the minus symbol 'high risk of bias', and the question mark 'unclear risk of bias') a Two methods of participant blinding were typical of the papers reviewed. In the first, participants were told that k groups existed, that they differed, and that they would be randomly assigned to one (O'Connor et al., 1996;Webster et al., 2018). In the second, participants were randomly assigned to an experimental condition, but were unaware others existed (Caplandies et al., 2017;Faasse et al., 2018;Wilhelm et al., 2018;Fernandez et al., 2019). b Six indices were calculated from the GASE (Rief et al., 2011). When investigating analgesia and adverse events two are omitted. It is unclear whether these omissions were pre-planned. c Only statistically significant Chi-square values are provided for side effect frequency. Some descriptive statistics are presented in graphical format and can only be estimated approximately. d While no power analysis is reported, this study contains the largest sample of those reviewed. e Exploratory analysis is reported for one subscale of the BMQ (Weinman and Hankins, 1999) only. It is unclear whether additional tests were performed and omitted.