A systematic review of factors associated with side‐effect expectations from medical interventions

Abstract Background Fear of side‐effects can result in non‐adherence to medical interventions, such as medication and chemotherapy. Side‐effect expectations have been identified as strong predictors of later perception of side‐effects. However, research investigating predictors of side‐effect expectations is disparate. Objective To identify factors associated with side‐effect expectations. Search strategy We systematically searched Embase, Ovid MEDLINE, Global Health, PsycARTICLES, PsycINFO, Web of Science and Scopus. Inclusion criteria Studies were included if they investigated associations between any predictive factor and expectations of side‐effects from any medical intervention. Data extraction and synthesis We extracted information about participant characteristics, medication, rates of side‐effects expected and predictors of side‐effect expectations. Data were narratively synthesized. Main results We identified sixty‐four citations, reporting on seventy‐two studies. Predictors fell into five categories: personal characteristics, clinical characteristics, psychological traits and state, presentation format of information, and information sources used. Using verbal risk descriptors (eg ‘common’) compared to numerical descriptors (eg percentages), having lower quality of life or well‐being, and currently experiencing symptoms were associated with increased side‐effect expectations. Discussion and conclusions Decreasing unrealistic side‐effect expectations may lead to decreased experience of side‐effects and increased adherence to medical interventions. Widespread communications about medical interventions should describe the incidence of side‐effects numerically. Evidence suggests that clinicians should take particular care with patients with lower quality of life, who are currently experiencing symptoms and who have previously experienced symptoms from treatment. Further research should investigate different clinical populations and aim to quantify the impact of the media and social media on side‐effect expectations.


| INTRODUC TI ON
Patients often fail to take medication as prescribed. Non-adherence to prescribed treatments is thought to cost up to $52 000 (US$ 2015) per person annually worldwide. 1 One of the main reasons why people do not take their medication is for fear of side-effects. [2][3][4] However, the cause of side-effects attributed to medication is often unclear. While some may be directly caused by the medication, others may arise from the nocebo effect. This is a phenomenon whereby symptoms are attributed to an exposure, but they are not directly caused by the physical properties of the exposure. There is good evidence that expectation of symptoms from inert 'placebo' exposures such as sham pills, inhalers and odours can cause symptoms in those expecting them. 5 Heightened side-effect expectations are associated with later perception of side-effects. Meta-analytic results indicate that patient expectations for post-chemotherapy side-effects are associated with development of side-effects from chemotherapy. [6][7][8] Similarly, a prospective cohort study of parents vaccinating their child for influenza found that parents' side-effect expectations were the strongest predictor of parental report of side-effects. 9 Symptoms reported in the placebo arm of randomized placebo-controlled trials may also arise from patient and investigator expectation. 10,11 Multiple systematic reviews and meta-analyses have found similar rates and profiles of symptoms reported in the placebo and active drug arms of randomized placebo-controlled trials across a range of medications. [10][11][12][13][14][15][16][17] There is little research investigating how side-effect expectations develop. Beliefs about high dosage of the medication and explicit suggestions that the medication causes side-effects may contribute to side-effect expectations. 5 How information about medical interventions, such as pharmacotherapy, chemotherapy and surgery, is framed by a health-care professional or patient information leaflet may also affect side-effect expectations.
Previous attempts to decrease side-effect expectations and subsequent side-effect experience include reducing information given to patients about potential side-effects. 18 This is problematic as it runs contrary to notions of informed consent and patient autonomy, and may breach laws ruling that information given to patients should not be 'cherry picked'. 19 Therefore, it is important to identify other factors that influence side-effect expectations to provide alternative avenues for interventions which do not face this ethical issue.
The aim of this study was to provide an overview of the current literature on side-effect expectations by conducting an exploratory systematic review to identify factors associated with expectations of more frequent side-effects from medical interventions. We investigated psychological factors, identifying factors to target in interventions to reduce the nocebo effect, and personal and clinical factors, identifying populations who are particularly at risk of inaccurate expectations. Thus, results will provide us with two useful implications: how to minimize side-effect expectations, and populations which may be particularly susceptible to heightened side-effect expectations.

| ME THODS
We conducted a systematic review in accordance with PRISMA criteria 20 to identify factors associated with expectations of sideeffects from medical interventions. We searched Embase, Ovid MEDLINE, Global Health, PsycARTICLES and PsycINFO through OvidSP, as well as searching Web of Science and Scopus. Our final search term was (symptom* OR side effect OR adverse effect OR adverse event OR adverse reaction) ADJ3 expect* (see Supporting Information S1). Databases were searched from inception to 6 March 2019. References and forward citations of included articles were also searched.

| Inclusion criteria
Studies were included if they met the following criteria: Participants: any age, or health status.
Predictors/exposures: investigated the association between psychological, social, contextual, or demographic factors and expectation that a medical intervention causes side-effects (in an actual or hypothetical situation).
Outcome: expectation that any active medical intervention (eg pill, vaccine, asthma inhaler, chemotherapy, surgery) caused side-effects. Studies investigating combined expectations about side-effect frequency and severity were included; those which investigated only expectations about side-effect severity were excluded. Studies investigating whether side-effect expectations predicted later perception of side-effects were excluded.
Study reporting: published in English. Studies were not excluded based on publication type.

| Data extraction
We extracted information about study design, inclusion criteria, participant characteristics, medical intervention, rates of side-effects expected and predictors of side-effect expectations.

| Risk of bias
Risk of bias was measured using an amended version of the Downs & Black checklist, 21 a validated checklist, 22 which is suitable for use in systematic reviews with appropriate modifications 23 and which can be applied to reliably and validly evaluate randomized and nonrandomized studies, including observational studies using cross-sectional and cohort methods. 24 The modified version of this checklist has been used previously by our group. 4,25 The checklist evaluates studies on five dimensions: reporting (out of 10); external validity (out of two); internal validity-bias (out of three); confounding-selection bias (out of three); power (out of one). Scores were summed to give a total out of nineteen. Studies were rated as good quality if | 733 SMITH eT al.
they scored a total of 16 or over; moderate quality if they scored [11][12][13][14][15]; and poor quality if they scored 10 or under. Studies were rated as poor quality for individual constructs if they scored: six or under for reporting; one or under for internal validity (bias), confounding (selection bias) and external validity; and if they did not include a justification for the sample size used.
LS and RW completed risk of bias ratings separately for 10 studies. Any discrepancies in scoring were discussed. LS and RW then completed ratings for 35 and 27 studies, respectively, which were cross-checked by the other author. Any discrepancies were solved through discussion. Final scores were approved by both authors.

| Procedure
LS came up with the search terms, carried out the search, screened papers, extracted data and completed risk of bias assessment. RW screened a random sample of 100 citations to full-text screening F I G U R E 1 Flowchart depicting the selection of studies for the systematic review, with reasons for exclusion stage, screened ten additional full texts and completed risk of bias assessment. Guidance was provided by GJR.
Data were narratively synthesized, taking study design and predictive validity into account when considering the strength of evidence for predictive factors. For psychological factors, experimental studies were considered to provide the strongest evidence, followed by longitudinal studies, then cross-sectional studies. We counted cross-sectional studies with factorial designs as experimental studies. For demographic characteristics which did not change, all study designs were considered equal.

| Study characteristics
A total of 14 297 citations were found by the original search. After removing duplicates, 7441 citations remained. After title, abstract and full-text screening, nineteen citations remained. Forty-five citations were identified by reference searching and forward citation tracking; none of these were found by the original search. Thus, 64 citations, reporting on 72 studies, met inclusion criteria (see Figure 1).
Inter-rater agreement for title, abstract and full-text screening for the random sample of 100 citations was 100%; agreement for fulltext screening of ten additional full texts was also 100%.
Studies investigated side-effect expectations for a range of medical interventions, including: chemotherapy; surgery; various medications including statins; and blood transfusion (for full list see Table 1).
Most studies investigated hypothetical scenarios in which participants imagined they needed a specified medical intervention and made judgments about the possibility of side-effects based on information given to them (n = 41). Twenty-five studies investigated real situations, in which participants were going to receive the medical intervention. Six investigated hypothetical situations, but a proportion of participants were taking or about to start taking that medication.
We identified four basic methods in the literature to measure side-effect expectations (see Supporting Information S2): likelihood of side-effects using a Likert-type scale (n = 26); probability of side-effects as a percentage (n = 9); frequency of side-effects as a number (eg out of 100 taking the medication, n = 6); or visual analogue scale (n = 4). The remainder of studies used a combination of these methods.
Forty-six studies were cross-sectional, with 36 using a factorial design. Sixteen studies used prospective cohort designs; nine were randomized controlled trials.

| Predictors of side-effect expectation
Results from adjusted and unadjusted analyses are reported together in the text. Where studies reported both adjusted and unadjusted analyses, only adjusted results are reported narratively. Only results from good and moderate quality studies are reported narratively; poor quality studies are reported in summary tables. We evaluated strength of evidence on a case-by-case basis to take into account study design. Where study design was the same, we used the following quantifications for the strength of evidence. 'Good evidence' was used when 80% or more of the studies investigating a factor found an association. 'Some evidence' was used when 60% to 80% of studies investigating the factor found an association. Where all studies found an association, but only few studies investigated an association, we also used the term 'some evidence.' 'No evidence' was used when less than 60% of studies found an association to account for the effect of publication bias.

| Personal characteristics
There was no evidence that gender was associated with sideeffect expectations (see Table 2). Of seven studies, two found an association between female gender and greater side-effect expectations. 38,51 One study found an association for two of five outcomes, 84,85 whereas another found an association between female gender and increased estimates of the probability of side-effects, but not increased likelihood of side-effects. 35 Three studies found no evidence for an association. 71,72,87 There was no evidence for an effect of age on side-effect expectations, with studies reporting mixed findings. Of nine studies, one found an association between older age and increased side-effect expectations 51 ; another found an association between older age and expectations of serious, but not mild side-effects. 71 One study found mixed evidence for an association between younger age and side-effect expectations for nausea, but not vomiting, 72 while another found an association between younger age and expectations for pain, but not fatigue. 73 Five studies found no evidence for an association between age and side-effect expectations. 35,38,72,84,85,87 There was no evidence for the effect of education on side-effect expectations, with studies reporting mixed findings. Of nine studies, one found an association between higher education and These results are from the same group of participants.

TA B L E 1 (Continued)
increased side-effect expectations, 51 while another found an association between higher education and increased expectations about the likelihood of fatigue, but not pain. 73 Three studies found mixed evidence for an association between lower education and increased side-effect expectations, with one study finding an association for one of five outcomes 84,85 ; another finding an association with expected nausea, but not vomiting 72 ; and the last finding an association with minor, but not major, complications. 87 Four studies found no evidence for an association. 35,38,71,72 There was no evidence for the effect of ethnicity on side-effect expectations, with studies reporting mixed findings. One study found evidence that people of white ethnicity gave increased estimates of the likelihood of side-effects compared to non-white ethnicities. 38 Conversely, one study found evidence that ethnic minorities gave increased estimates of the likelihood of side-effects compared to white ethnicities for four of five outcomes. 84,85 Three studies found no evidence for an association between ethnicity and side-effect expectations. 73,81,87 One Australian study found no association with being born overseas and side-effect expectations. 71 Studies investigating associations between side-effect expectations and employment and job role were heterogenous, providing no evidence for an association. One study found that students estimated that a higher percentage of people would experience side-effects from over-the-counter medications than doctors. 34 Another study found mixed evidence that patients estimated a higher frequency of side-effects than doctors. 69 One study found that people who considered their job to be health care-related estimated a higher frequency of side-effects than those who did not. 87 Another study found no association between side-effect expectations and employment. 84,85 Both studies investigating the association between poorer numeracy and increased side-effect expectations found an association, 77 with a third study finding mixed evidence for an association between poorer numeracy and increased probability of certain side-effects. 44 One study also found mixed evidence for an association between poorer health literacy and increased side-effect expectations. 84,85

Side-effect characteristics
There was some evidence that side-effects perceived as being less severe were expected to occur more often (see Table 3). Of five studies, three found an association 34,43,87 ; two studies found no evidence for an association. 28,29,84,85 All studies used experimental designs.
There was no evidence that increased objective likelihood or frequency of side-effects was associated with increased side-effect expectations. Of five studies, two experimental studies found an association with increased perceived likelihood of side-effects. 43,77 Three studies (one experimental, two longitudinal) found no evidence for an association. 38,72 Previous experience with illness or treatment There was no evidence that previous experience of a treatment or illness was associated with increased side-effect expectations. One cross-sectional study found that previous experience of surgery for breast cancer was associated with decreased expectations for postsurgical fatigue, but found no evidence for an association with postsurgical pain. 73 Two experimental studies found no evidence that previous experience of endocrine treatment, or history of illness, were associated with increased side-effect expectations. 48 There was some evidence that previous experience of symptoms from a medical intervention was associated with increased side-effect expectations. Of three studies, one longitudinal study found an association between having previously experienced side-effects from the treatment and increased side-effect expectations. 64 Two studies (one cross-sectional, one experimental) found an association with previous side-effects for mild, but not severe, side-effect expectations. 71,87 Another cross-sectional study found no evidence that knowing more side-effects from endocrine therapy (free recall, before being given study treatment information) was associated with increased side-effect expectations. 49

Intervention characteristics
There was some evidence that decreased medication effectiveness (perceived and stated) was associated with increased side-effect expectations. Of four studies, one cross-sectional study found an association between decreased medication effectiveness (perceived) and increased side-effect expectations, 49 two experimental studies found mixed evidence (stated effectiveness), 77 and one experimental study found no evidence for an association (perceived effectiveness). 38 Another experimental study found an association between including extra information about the effectiveness of the treatment and decreased side-effect expectations. 76

Current symptoms and quality of life
There was some evidence that current experience of symptoms was associated with increased side-effect expectations. Of six studies, two (one experimental, one longitudinal) found an association between existing physical symptoms and increased sideeffect expectations. 48,51 One longitudinal study found evidence for an association at one of four timepoints investigated, 64 while a cross-sectional study found that pre-surgical fatigue was associated with increased expectations of post-surgical fatigue; there were no associations with pre-surgical pain. 73 Two studies (one experimental and one cross-sectional) found no evidence for an association. 48,49 Two studies investigated the severity of existing symptoms with relation to side-effect expectations, with one longitudinal study finding an association between increasing severity of existing symptoms and increased side-effect expectations 51 and one cross-sectional study finding no evidence for an association. 49 There was some evidence that lower pre-treatment quality of life was associated with increased side-effect expectations, with two cross-sectional studies finding an association. 39,49 Another longitudinal study found an association between worse general well-being and increased side-effect expectations. 51 An experimental study found evidence for an association between chronic illness and increased side-effect expectations for two of five outcomes. 84,85 One experimental study found no association between health status and side-effect expectations. 38

Anxiety and other traits
There was some evidence that heightened health anxiety was associated with increased side-effect expectations (see Table 4), with one experimental study finding an association. 84,85 Another longitudinal study found evidence that an anxious preoccupation cancer coping style was associated with increased likelihood and severity of expectations for multiple side-effects. 86 There was no evidence that higher trait and state anxiety were associated with increased sideeffect expectations. Two studies (one experimental and one crosssectional) found an association between increased trait anxiety and side-effect expectations, 48,73 while two studies (one experimental, one longitudinal) found no evidence for an association. 48,64 One longitudinal study found no evidence for an association between state anxiety and side-effect expectations. 86 There was no evidence that other psychological traits were associated with side-effect expectations. One cross-sectional study found no evidence for an association between combined depression and anxiety score and side-effect expectations. 49 Another longitudinal study found no evidence for an association between emotional distress and side-effect expectations. 64 Two studies (one experimental and one cross-sectional) investigated the association between optimism and side-effect expectations, neither finding evidence for an association. 73,84,85 There was some evidence that pre-intervention distress was associated with side-effect expectations. One cross-sectional study found that pre-surgical distress and fear were associated with increased expectations of side-effects from surgery. 73 Decisional conflicts about treatment were associated with increased likelihood and severity of side-effect expectations in one longitudinal study (64 study 1), but not another (64 study 2).

Beliefs about medicines
Few studies investigated the association between beliefs about medications and side-effect expectations, with mixed results. There was some evidence that negative beliefs about the overuse of medications were associated with increased side-effect expectations, with one cross-sectional study finding an association 49 and one experimental study finding an association for four of five outcomes. 84,85 There was no evidence for an association between negative beliefs about harm that medications could cause and side-effect expectations, with an experimental study finding an association for four of five outcomes 84,85 ; another cross-sectional study found no evidence for an association. 49 There was no evidence that more concerns about medications compared to beliefs about their necessity were associated with sideeffect expectations, with three studies (two experimental and one cross-sectional) finding no evidence for an association. 48,49 There was some evidence that increased perceived sensitivity to medicines was associated with increased side-effect expectations, with one experimental study finding an association. 84,85 Using a monitoring coping style to deal with illness was associated with increased likelihood and severity of side-effects in one longitudinal study (64 study 2), but not another (64 study 1). There was no evidence for an association between side-effect expectations and somatosensory amplification (cross-sectional), 49 or social desirability (longitudinal). 64

Verbal and numerical presentation
There was good evidence that describing the incidence of side-effects verbally, using words such as 'often' or 'rarely', was associated  with greater side-effect expectations than when describing incidence numerically, using percentages or natural frequencies (see Table 5). Of eight studies, five found an association. 28,29,35,38,57,77 Two studies found mixed evidence for an association; 45,77 one study found no evidence for an association. 26 Two studies found that using only verbal descriptors led to greater expectations of likelihood and severity of side-effects than using combined verbal and numerical descriptors. 77 Two studies investigated the use of combined numerical and verbal information, compared to just numerical information.
One study found mixed evidence for an association between combined numerical and verbal information and increased side-effect expectations, 60 while the other found no evidence for an association. 66 Another study found that the order of the verbal descriptors of incidence (eg presenting side-effects which 'often' or 'rarely' occurred first) did not affect side-effect expectations. 28,29 All studies used experimental designs.
There was no firm evidence for the type of numerical predictor most associated with increased side-effect expectations.
One study found evidence that incidences presented as natural frequencies (eg 'affects 1 in 50 people') led to greater estimates of the likelihood of side-effects than percentages and combined natural frequencies and percentages; 88 another study found very little (one of seven outcomes) evidence for this association. 59 One study found that there was a wider spread in the verbal labels assigned by participants to incidences described as percentages than natural frequencies. 79 One study found that estimated percentages of incidence of side-effects were greater when communicated as an increase in the number needed to harm (eg 'for every 500 women…one additional woman will have') and relative risk (eg 'the risk…is doubled') than when communicated as an increase in absolute risk (eg 'the risk…is 0.02% higher') in situations with no information about the baseline rate of people affected by that side-effect. 36 Two studies found no evidence that the response format (percentage or natural frequency) for estimates of side-effect expectations affected outcomes. 28,29,38 All studies used experimental designs.

Framing information
There was no evidence that personalizing information (eg 'you should take two tablets' compared to 'two tablets should be taken') was associated with side-effect expectations, with studies reporting mixed findings. Of five studies, two found that non-personalized information was associated with increased expectations of the likelihood of side-effects. 31 One study found that personalized information was associated with increased estimates of likelihood and severity of side-effects. 48  an association between personalized information and side-effect expectations. 48 study 2,43 All studies used experimental designs.
There was some evidence that negatively framed information was associated with increased side-effect expectations. Of three studies, one experimental study found an association, 48  Four studies investigated the effect of individual statements on side-effect expectations. Participants in one study gave higher estimates for the incidence of side-effects when the baseline rate of side-effects was not communicated (compared to communicated). 36 Two studies found that using a verbal descriptor ('more people had bone loss') increased side-effect expectations compared to giving no information about medication effectiveness or side-effect incidence. 77 One study found no evidence that using a verbal qualifier (eg 'will affect' compared to 'may affect') was associated with side-effect expectations. 60 All studies used experimental designs.

| Information sources
There was no evidence that the number of sources used to gain information about a medical intervention was associated with increased side-effect expectations, with studies reporting mixed results. One poor quality cross-sectional study found an association; 52 as this was a conference abstract, the quality rating score was artificially low.
This study also found that using the internet, the National Cancer Institute and American Cancer Society as sources of information about cancer were associated with increased side-effect expectations, whereas consulting newspapers and primary care physicians were associated with decreased side-effect expectations. 52 Another cross-sectional study found no evidence for an association between number of sources used to gain information about the intervention and side-effect expectations. 49 How often participants read patient information leaflets when taking a new medication was also not associated with side-effect expectations (experimental study). 84,85 One longitudinal study found that using more media sources to gain information about an illness and its treatment was associated with stating that treatment side-effects were more likely. 51

| D ISCUSS I ON
Fear of side-effects is one of the most commonly cited reasons for not adhering to medical interventions. 9 Side-effect expectations have also been associated with decreased intention to adhere to medications. 48 Side-effects from medical interventions may not be directly attributable to the treatment itself, but may instead arise through a psychological phenomenon known as the nocebo effect, whereby expectation that an intervention will cause side-effects is self-fulfilling. [6][7][8][9]90 Identifying psychosocial factors associated with side-effect expectations enables these factors to be targeted Interestingly, only a minority of studies investigating the objective frequency of side-effects (eg comparing 'uncommon' to 'common'; or '1 in 100' to '1 in 10') found that side-effect expectations increased in line with objective descriptors. This may be due to strong preconceptions about medication side-effects which were not influenced by study information, or, where information was presented numerically, because people did not understand the information presented to them due to poor numeracy. 93 Decreased numeracy is often associated with having less accurate perceptions about the risk of medical interventions 94 and being more easily influenced by the way numerical information is framed. 95,96 While only investigated by few studies in this review, poorer numeracy and health literacy were associated with increased side-effect expectations.
Changing the phrasing of current patient information leaflets may be one of the cheapest ways to alter side-effect expectations. 97 Consistent with other research, we found that side-effect expectations were higher when incidence was described verbally rather than numerically. 98 However, there was no clear evidence for the type of numerical descriptor (eg percentage or natural frequency) which generated the lowest side-effect expectations. 99 Studies investigating the accuracy of side-effect expectations arising from information presented in different formats have found that using simple infographics, such as pictographs, increases accuracy of estimates of incidence of side-effects. 100,101 Pictographs are also perceived as being more trustworthy and helpful than information presented in tables and text. 100 In addition to presenting information numerically, improving the readability of patient information leaflets, by making the font larger, using simple language and including more pictures, 102,103 might also increase accuracy of understanding of information about medical interventions.
Very little research has investigated the role of sources of information on side-effect expectations, with mixed findings. While other research has focused on the negative role of the media on side-effect reporting, 104,105 one study included in the review found that consulting newspapers as a source of information was associated with decreased side-effect expectations. 52 Research has indicated that side-effect expectations mediate the association between increased suggestion of side-effects from different sources, and later perception of side-effects. 9 It is therefore important to quantify the role of suggestions from different sources such as online searches, social media, news and the influence of friends, family and health-care practitioners across different treatments for different illnesses.
There was very little evidence for the influence of psychological traits or state on side-effect expectations. Increased health anxiety was associated with increased side-effect expectations, although only one study investigating this factor was good quality. In line with a systematic review finding weak evidence for an association between state and trait anxiety and the nocebo effect, this review found no evidence for an association with increased side-effect expectations. 5 Believing that medicines are overused and that you were more sensitive to medicines, were also associated with increased side-effect expectations; however, few studies investigated these factors. More research is needed to understand how influential wider beliefs about medicines are in the formation of side-effect expectations.
Evidence from the review indicates that personal characteristics do not systematically influence side-effect expectations, with studies reporting associations with both increased and decreased side-effect expectations for multiple factors (eg age, education).
Rather than using personal characteristics to target populations for interventions aiming to decrease side-effect expectations, results of this review suggest that clinical characteristics may be more useful. In particular, clinicians should take care with patients with lower pre-treatment quality of life and well-being, those who are currently experiencing symptoms, and those who have previously experienced side-effects from the treatment. Given the growing influence of the media and social media, more research investigating their influence on side-effect expectations is also needed.

| Limitations of the studies included in the review
Most studies included in the review were poor quality. Studies scored particularly poorly for external validity, with only a small number being appropriately powered. Few studies investigated the same predictors; this was particularly notable for studies investigating presentation format. Outcome measures and statistical tests used were also heterogeneous. Studies investigated hypothetical and actual scenarios, with some studies including both people who were due to receive the intervention and those who were not in the same sample. People who were about to receive a medical intervention may have paid more attention to the information given to them about that intervention, or may have interpreted risks differently given the potential for personal experience. 106

| Limitations of the review
Limitations of the review should also be considered. First, studies investigated many side-effect expectations for many different medical interventions (eg chemotherapy and pills) and in different populations (eg healthy and unwell). We were unable to investigate whether factors were differentially associated with side-effect expectations for different medical interventions or populations, meaning that we are unable to draw fine-grained conclusions about whether factors affecting side-effect expectations differed by medical intervention or study populations. The ecological validity of results, and ability to extrapolate findings to other populations or medical interventions, should be considered when interpreting findings.
Second, few studies investigated the same factors, leading to a lack of replication across studies. Therefore, our interpretation and conclusions for some predictors are based on limited results and should be taken with caution.
Third, we did not search MeSH terms, meaning that we may have missed some studies which were eligible for inclusion.
Fourth, only 19 studies in the review were identified through our search, with the majority coming from reference and forward citation tracking. This may have impacted the results of, and conclusions drawn from, the review.
Fifth, we are aware that any heuristic used in this review to aggregate data (eg counting the number of studies finding significant and non-significant associations between predictors and side-effect expectations) are susceptible to bias. More robust methods of reviewing the evidence, such as meta-analyses, would be preferred to minimize this bias. However, in this case, studies were too heterogeneous to carry out a meta-analysis.

| CON CLUS ION
Clinical characteristics and presentation format may impact side-effect expectations; there is less evidence for a role of personal characteristics, psychological traits or states, and information sources. There was some evidence that patients who are currently experiencing symptoms; have lower quality of life; and who have previously experienced mild side-effects from the medical intervention may have heightened side-effect expectations. Clinicians should take particular care with these patients. Using verbal descriptors of risk, such as 'common' or 'rare', was associated with greater side-effect expectations than numerical descriptors, such as percentages or natural frequencies. There was no evidence that a particular type of numerical descriptor was as- Research should also attempt to replicate findings, to ensure they are robust.

CO N FLI C T O F I NTE R E S T
The authors have no conflicts of interest relevant to this article to disclose.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analysed in this study.