Effectiveness of behavioural economics-based interventions to improve colorectal cancer screening participation: A rapid systematic review of randomised controlled trials

Highlights • We searched PubMed, PsycInfo and EconLit for RCTs that evaluated BE interventions in CRC screening.• We identified 1027 papers for title and abstract review. 30 studies were eligible for the review.• The most frequently tested BE intervention was incentives, followed by default principle and salience.• Default-based interventions were most likely to be effective. Incentives had mixed evidence.• BE remains a promising field of interest in relation to influencing CRC screening behaviours.


Background
Colorectal cancer (CRC) is a leading cause of death in Europe, Asia and North America (Bowel cancer, 2017;Schreuders et al., 2015;Sung et al., 2021). Survival is greatly improved when CRC is diagnosed early (Bowel cancer statistics, 2015; Koo et al., 2017) and, as a result, many countries throughout the world have implemented organised screening programmes for the early detection of CRC (Schreuders et al., 2015).
Despite the availability of several effective CRC screening tests, uptake is often low compared with other organised screening programmes, such as those for breast and cervical screening (Koo et al., 2017). Participation rates routinely fall short of national and international guidelines (e.g. the European Guidelines for Quality Assurance in Colorectal Cancer Screening and Diagnosis, which has a minimum participation rate of 65%) (Koo et al., 2017;Duffy et al., 2017;Directorate-General for Health and Consumers, 2010). As a result, there is currently much interest in the development of effective interventions to improve CRC screening uptake.
Historically, a wide range of interventions to improve CRC screening uptake have been investigated, including advance notification letters (also referred to as: 'pre-screening reminders'), mailed reminders and telephone navigation calls (Duffy et al., 2017). Such strategies have focussed on a cognitive model of behaviour change, aiming to influence behaviour via reflective, rational thought processes known as System II thinking (Dolan et al., 2012).
More recently, researchers have begun testing interventions grounded in behavioural economics (BE). Such interventions seek to exploit the automatic (learned) responses of System I thinking, called 'heuristics' (Dolan et al., 2012;Purnell et al., 2015; The rise of behavioural economics, 2017), and usually involve small changes, known as 'nudges', to the context in which the decision is being made (Thaler and Sunstein, 2009).
Several types of nudge have been described, each of which can be leveraged differently to try and improve CRC screening participation (Purnell et al., 2015). Choice architecture, for example, refers to the way in which choices are presented, and can be optimised to encourage participation for certain test options (Purnell et al., 2015). Similarly, perceived social norms, default options, financial incentives and other principles can all be leveraged to unconsciously influence screening behaviours in a positive manner (Purnell et al., 2015).
While the concept of BE has gained traction over the past two decades, it is still a relatively new and emerging field in relation to cancer screening, and CRC screening more specifically (Dolan et al., 2012). Indeed, to date, the evidence base for BE interventions, in relation to CRC screening, remains uncertain, with no formal review of the literature having been published. Previous studies have explored some BErelated interventions, such as GP endorsement (Duffy et al., 2017) and incentives (Facciorusso et al., 2021); however, none has attempted to perform a review of the plethora of BE interventions that exist.
The present study, therefore, seeks to address this dearth in the literature, by synthesising data from existing studies to investigate whether interventions using BE principles are effective at influencing CRC screening behaviours in individuals eligible for population-based screening.

Search strategy and study design
We searched PubMed (May 2021) for randomised controlled trials (RCTs) that tested the effectiveness of BE interventions to improve participation in CRC screening. To be eligible, a full text English article had to be available. Studies were excluded if they did not use a RCT design, or measured intentions instead of behaviour (see Table 1 for a detailed overview of the inclusion / exclusion eligibility criteria).
Due to resource restrictions (e.g. funding, staff, etc.), a rapid review of the literature was performed. Therefore, rather than running a single search with the full, comprehensive list of search terms, we performed an initial search, using a narrow selection of search terms (guided by the PICOS frameworksee Table 2). The string was then expanded, successively, by adding a small number of additional search terms (to each PICOS component), using the Boolean operator 'OR', from a pool of search terms identified from the previous literature (Duffy et al., 2017;Purnell et al., 2015).
The exact combination and order in which search terms were added to the search string were determined by running multiple searches in PubMed, with the combination providing the largest number of results being the one selected for the expansion at each stage (for transparency, the individual searches and number of results received for each is available from Open Science Framework: https://osf.io/tqsmc/).
After each expansion, title and abstract review was performed for the combination that received the most results. This process of identifying the optimal combination of search terms, expanding the search string, and performing title and abstract review was continued until no new publications were eligible upon title and abstract review. At this point, it was assumed that further expansion of the search string was unlikely to yield additional publications eligible for inclusion.
To minimise the risk of excluding eligible studies not available on PubMed, the final search was also performed on PsycInfo and EconLit (see Table 3).

Study selection process
Title and abstract reviews were performed by two reviewers (LT and RK). Each reviewer assigned publications a value of 1 ('include') or 0 ('exclude'). Any paper that received a score of 1, from either reviewer, following title and abstract review, underwent full paper review. As with title and abstract review, two reviewers assigned the full text of publications a value of 1 ('include') or 0 ('exclude). Unlike title and abstract review, however, disagreements regarding full-text review were resolved through discussion with a neutral arbitrator (SS).
Finally, the reference lists of publications that passed full paper review were searched for further publications (as well as a relevant review published by Duffy and colleagues (2017); (Duffy et al., 2017) not available on PubMed, PsycInfo and EconLit. As with the database searches, publications reported in the reference lists underwent title, abstract and full paper review by two authors (with disagreements resolved through discussion with a third during full paper review).

Data extraction
Data on the author, year of publication, title, primary aim, intervention delivered, method of randomisation (individual or cluster), population characteristics (e.g. age, sex, ethnicity, etc.), timepoint at which participation in screening was measured and the statistical method used (univariate or multivariate) was extracted for each study using customised Excel templates.

Data categorisation
As the field of BE contains many synonymous terms for the different BE principles (for example the term 'choice architecture' may be used to describe multiple types of intervention), RCTs were categorised according to the type of BE intervention tested, using the popular MIND-SPACE framework (Dolan et al., 2012). MINDSPACE is a mnemonic framework comprised of nine behavioural science principles that can be used to influence behaviour through automatic processes, as shown in Table 4 (Dolan et al., 2012). It was developed specifically to aid in policy making and is, therefore, relevant to the real-life application of this research (MINDSPACE.pdf, 2021). Studies, that tested a combination of interventions, were categories according to their primary BE intervention. Note that in the results section the categories are presented in order of number of studies found and not the order of Table 4.

Data analysis
Descriptive statistics were used to quantify study characteristics and the number and percentage of RCTs that reported a positive effect, negative effect, or no effect of BE interventions on CRC screening participation (collectively and stratified by MINDSPACE component). No meta-analysis was conducted due to the high degree of overlaps between different components and differences in mode of employment. Furthermore, according to the Cochrane Rapid Reviews Methods Group, meta-analysis should only be conducted where studies are able to be appropriately pooled (Garritty et al., 2021).

Risk of bias
Eligible papers were assessed for bias using version 2 of the Cochrane risk-of-bias tool for randomised trials (RoB 2) (RoB 2, 2021). A test version of the RoB 2 tool for cluster-randomised (Inadomi et al., 2012;Tinmouth et al., 2015;Nisa et al., 2019;Kullgren et al., 2014;Dacus et al., 2018;Cole et al., 2002;Zajac et al., 2010;Coronado et al., 2020;Stoffel et al., 2021a) trials or crossover trials (Slater et al., 2018;Van Roosbroeck et al., 2012) was used where appropriate. RoB 2 is recommended when assessing bias in randomised trials and consists of signalling questions around several domains where bias can occur, including randomisation process, missing outcome data, and reporting of results (RoB 2, 2021). Trials can be classified as 'low risk' of bias, 'high risk' of bias or raising 'some concerns' (RoB 2, 2021). This was conducted by the primary reviewer (LT) with a secondary reviewer (SS) completing risk of bias assessment for 20% of studies.

Transparency
The review was registered prospectively with PROSPERO (reference number: CRD42021253534) and written in accordance with PRISMA guidelines (see Appendix A for completed checklist).

Search results
In total, the database and reference list searches identified 1041 articles (524 from PubMed, 11 from PsycInfo, 0 from EconLit [confirmed to be a 'true zero' by searching the database for eligible papers identified through PubMed and PsycInfo] and 506 from the reference lists). After removing 14 duplicates, 1027 papers were eligible for title and abstract review, of which 49 passed and underwent full paper review. A total of 29 studies were deemed eligible and were included in the review. This number was increased to 30, following the inclusion of an additional paper, authored by RK and SS (Stoffel et al., 2021a), which was accepted for publication, but not published, after the database searches and reference list searches were performed. An overview of the search results is provided in Fig. 1. Table 5 presents a summary of the studies included in the review (a detailed overview of each study is provided in Table 6). The majority of studies were conducted in the USA (N = 17, 57%) (Inadomi et al., 2012;Kullgren et al., 2014;Dacus et al., 2018;Slater et al., 2018;Mehta et al., 2018;Church et al., 2004;Huf et al., 2021;Bakr et al., 2020; Gupta et al., 2016), examined the effectiveness of BE interventions to promote the uptake of gFOBT or FIT (N = 27, 90%) and had a sample size of ≤ 5,000 participants.

Risk of bias
The results of the RoB 2 analysis are presented in Table 5 (the completed worksheets for each study are available from Open Science Framework: https://osf.io/tqsmc/). The majority of studies had some bias (N = 16, 53%), which was largely due to missing data or contamination of the intervention groups.
The next most frequently tested BE intervention was the default principle (N = 7, 23.3%). There, three studies (42.8%) tested opt-out approaches, two (28.6%) offered a choice between different screening tests and the remainder either reminder or pre-alert messages (N = 2, 28.6%). The majority of studies (N = 5, 71.4%) testing this style of intervention found a statistically significant positive effect on participation (Tinmouth et al., 2015;Van Roosbroeck et al., 2012;Mehta et al., 2018;Church et al., 2004;Huf et al., 2021), while the remainder (N = 2, 28.6%) found no statistically significant positive or negative effect on participation (Inadomi et al., 2012;MACS, 2006). Employing the default component using an opt-out approach with direct mailing of the screening test kit was always effective in increasing uptake (Tinmouth et al., 2015;Van Roosbroeck et al., 2012;Mehta et al., 2018). Similarly, pre-alert and reminder messages were also always found to be effective (Church et al., 2004;Huf et al., 2021). Offering choice, however, never increased uptake compared to recommending specific tests (Inadomi et al., 2012;Tinmouth et al., 2015;MACS, 2006). There was no difference in the risk of bias scores for those that found a positive effect and those that found no effect.
As with defaults, seven papers tested the effectiveness of salience on CRC screening participation. Three studies (42.8%) investigated messages from social psychology and BE Stoffel et al., 2021a;Bakr et al., 2020), two studies (28.6%) alternative choice framings (Mehta et al., 2019b;Stoffel et al., 2021b) and two studies (28.6%) concepts from psychology, namely: anticipated regret and the questionbehaviour effect (O'Carroll et al., 2015;Hagoel et al., 2016). Only three studies (42.8%), two on BE messages and one on the question-behaviour effect found a statistically significant positive effect Bakr et al., 2020;Hagoel et al., 2016), with most studies (N = 4, 57%) finding no statistically significant positive or negative effect on participation (Stoffel et al., 2021a;Mehta et al., 2019b;O'Carroll et al., 2015;Stoffel et al., 2021b). Moreover, the risk of bias scores for studies that reported a positive effect, compared to those reporting no effect, were higher, indicating that the papers reporting no effect were of better quality.
Few interventions tested the effects of norms (N = 2, 7%) and results were divided equally between a significant positive effect for a study that employed an injunctive norm, stating that the GP endorses the screening programme and that participation is desired  and no significant positive or negative effect on behaviour for a study that used descriptive norms, communicating that the majority of people get screened . Both studies have some concern for biases.
The messenger component was also tested in two studies, both of which reported a statistically significant positive effect (Cole et al., 2002;Zajac et al., 2010). In both studies, the messenger component consisted of an invitation letter with the practice letterhead and signature of the GP or practice partner. The quality of the studies, however, was low as one study had risk of bias (Zajac et al., 2010) and the other raised some concerns (Cole et al., 2002).
One study tested the effectiveness of primers and found a significant positive effect (Coronado et al., 2020). The study had a low risk of bias and tested the effectiveness of pre-notification phone calls three weeks before the anticipated receipt of the test kit.
No studies testing the effectiveness of affect, commitment or ego were identified (see Table 8).

Summary of main findings
This review sought to synthesise the available evidence from RCTs Table 4 Summary of the components included in the MINDSPACE behavioural science framework.

Messenger:
The source of the information being communicated to us influences our automatic reaction. This may depend on the level of authority the messenger possesses, e.g. GP endorsement of CRC screening ( Dolan et al., 2012;Purnell et al., 2015 Jan).

Incentives:
Financial incentives and vouchers can be used in various ways to encourage CRC screening uptake. Predictable heuristics dictate how we are likely to respond to the size and timing of incentivisation, e.g. using loss-framing as opposed to equivalent gains (Dolan et al., 2012).

Norms:
How others behave can influence individual behaviour through the concept of sociocultural norms (Dolan et al., 2012). Descriptive norms which highlight the behaviour of others and injunctive norms, highlighting what others believe one ought to do, have both been used to examine cancer-related behaviours (Smith-McLallen and Fishbein, 2008).

Defaults:
Structuring screening invitations such that the default represents the most beneficial option can improve associated behaviours, as individuals often resort to default options over active choices, e.g. opt-out instead of opt-in screening (Dolan et al., 2012).

Salience:
The most relevant information is generally what attracts our attention. Therefore, increasing the salience of CRC in relation to an individual's personal circumstances may improve related behaviour (Dolan et al., 2012).

Priming:
Priming refers to pre-activation of knowledge with cues that may unconsciously impact subsequent behaviour (Dolan et al., 2012), e. g. asking a certain question prior to screening invitation may impact the response.

Affect:
Emotional responses are often automatic and may be acted upon before rational decision making occurs (Dolan et al., 2012;Purnell et al., 2015 Jan).

Commitment:
The conscious act of pre-commitment to a behaviour may subconsciously improve ensuing behaviours, as people strive to deliver on public commitments (Dolan et al., 2012). In CRC screening, asking for commitment such as pledging could be used to increase the likelihood of participation.

Ego:
The notion of self-image may motivate individuals to act in ways that facilitate positive self-evaluation (Dolan et al., 2012).
for using BE interventions to improve CRC screening participation. A total of 30 studies were identified, the majority of which were conducted in the USA and examined the effectiveness of BE interventions to promote the uptake of gFOBT and FIT screening. Importantly, most studies found that BE interventions resulted in a statistically significant increase in CRC screening participation suggesting that BE interventions have the potential to improve participation. The differences observed between studies that reported a positive effect on CRC screening uptake, compared to those that reported no effect, are unlikely to be attributable to bias, as the papers are of similar quality. Default-based interventions had the best evidence to support their use in terms of the proportion that found a positive effect and a low risk of bias. Employing the default component using an opt-out approach with direct mailing of the screening test kit was more likely to be effective than recommending specific tests. Similarly, when using the messenger component, presenting the screening offer on practice letterhead and including the signature of a practice partner appears more effective than simply indicating GP support (Cole et al., 2002;Zajac et al., 2010). Incentives, meanwhile, had mixed evidence to support their use; evidence is particularly uncertain outside of the USA, where all but one study tested their effectiveness. Additionally, as US studies were not conducted as part of organised screening programmes, the generalisability of their findings may be limited for organised programmes. Although the types of interventions utilising the salience component were varied, active choice and anticipated regret interventions were seemingly less effective than use of other salienceinformed mailings. Only a single study leveraged the primer component with pre-notification phone calls (and despite the fact that it was of low risk of bias), it is difficult to draw conclusions about the overall efficacy of priming to improve CRC screening participation.
The results of this review also suggest that the quality of the literature is mixed, with 60% of the included studies raising some degree of concerns for bias.

Comparisons with the previous literature.
The results of this review are consistent with the findings of studies examining the effectiveness of BE interventions to modify other behaviours. For example, the finding that defaults are effective at improving CRC screening participation are consistent with studies assessing their effectiveness to improve organ donation (Johnson and Goldstein, 2003). The results of this review are also consistent with those of a previous review of interventions to promote cancer screening participation, which also found that GP-endorsement (messenger) and mailing kits with the invitation (default) were effective. Finally, the results of this review are also consistent with a recent review of the evidence for use of incentives to improve participation in CRC screening, specifically (Facciorusso et al., 2021). Not only this, but the present review identified all the papers included in the incentives review (as well as several additional ones), suggesting that the rapid search strategy used was effective.

Implications for policy and future research.
The finding that most BE interventions have no effect or a positive effect on CRC screening participation suggests that they can be adopted into formal screening programmes with minimal risk to adversely affecting participation. Moreover, as recent cost-effectiveness analyses have shown that BE interventions can be cost-effective in increasing vaccination and improving antibiotic prescribing for acute respiratory infections (Benartzi et al., 2017;Gong et al., 2019). Most interventions (aside from those leveraging financial incentives) involve relatively low costs, so that they are unlikely to significantly influence the finances of the impact screening programmes (RoB 2, 2021), further justifying their use in line with APEASE criteria. The results of this review do, however, suggest that further research into specific MINDSPACE components is required to determine their effectiveness in promoting participation in CRC screening, namely: norms, messenger, and primer-based interventions. In addition, the results of this review suggest that further research is needed to determine the effectiveness of BE interventions in a wider range of settings and contexts. For example, no studies conducted in Asian countries were identified, despite the fact that many of these countries do offer organised screening programmes for CRC and operate differently to the USA (Chiu et al., 2017). Finally, the present review suggests that further high quality research is needed in this area, as most studies had some or high risk of bias.

Strengths and limitations
This review has several strengths. First, no restrictions were placed in terms of when studies were published, meaning no literature was excluded on this basis. Second, the review focussed solely on CRC screening, making the results highly specific to CRC screening behaviours. Finally, a relatively high number of papers were included, considering BE in cancer screening is a relatively new field of research. Note that alternative empirical studies, such as online experiments, were not included in this study as they do not measure real behaviour von Wagner et al., 2019;Stoffel et al., 2020).
This review also has several limitations. Most importantly, the search strategy used was not comprehensive; it was limited to peer-reviewed articles available on PubMed, PsycInfo and EconLit. Furthermore, several search terms were omitted, due to zero articles being identified by the final search. As such, it is possible that our review did not include several relevant studies. This is a common limitation with rapid reviews, one which is often accepted in favour of reviewing the literature in a shorter period, usually because the time and resources required for a  Dacus et al., 2018;Green et al., 2019;Inadomi et al., 2012;MACS group, 2006;Mehta et al., 2017;Mehta et al., 2019a;Mehta et al., 2020a;Slater et al., 2018.

Conclusion
This review shows that the effectiveness of BE interventions to promote CRC screening participation is mixed. The majority of studies have had a positive effect on screening participation, but a considerable proportion have found no effect. Several areas remain in need of future research, including investigation into BE interventions in programmes outside of the USA. Overall, BE remains a promising field of interest in relation to influencing CRC screening behaviours.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence  FOBT (Kullgren et al., 2014)

Table 8
Results stratified by MINDSPACE component.

MINDSPACE Component:
Studies reporting a positive effect (N ¼ 18): Studies reporting no effect (N ¼ 12): Total RoB 2 score Low risk Some concern High risk RoB 2 score Low risk Some concern High risk RoB 2 score Messenger (N = 2) 1 (50%) 1 (50%) 3/6 (50%) N/A 3/6 (50%) Note: Studies with low risk of bias were given 3 points, those with some concern 2 point and high-risk studies received 1 point. The RoB 2 score summarises the points of the overall studies and for those reporting a positive effect or no effect, separately, as well as by MINDSPACE component.
the work reported in this paper.