Enhancing Type 2 diabetes risk communication with message framing and tailored risk feedback: an online randomised controlled trial

ABSTRACT

Objective

Type 2 diabetes (T2D) risk communication may help individuals better understand their risk and motivate behavioural changes. There is a wealth of research in health risk communication which suggest the effectiveness of message framing and tailored risk feedback; however, little is known about their potential utility when used concurrently and in high-risk population approaches to T2D prevention.

Methods

This study evaluated the effects of message framing and tailored risk feedback on T2D risk perception and behavioural intentions, and if these effects were varied by level of alcohol consumption. Three hundred and forty-seven online participants were stratified by levels of alcohol consumption and subsequently randomised to receive T2D information, risk estimates, and lifestyle recommendations that were subjected to four different message framing and tailoring manipulations.

Results

No significant differences were observed in T2D risk perceptions or behavioural intentions by study arm. However, T2D risk perception scores and accuracies, and behavioural intentions significantly increased post-intervention across all conditions.

Conclusions

Despite the lack of impact of message framing or message tailoring, this study suggests that a brief online T2D risk communication can help to correct risk perceptions and increase behavioural intentions. These preliminary findings are encouraging and support the continued development of online risk assessment and communication to help combat the current T2D epidemic.

KEY POINTS

What is already known about this topic:

• (1) Most individuals at risk of Type 2 diabetes do not engage in risk-reducing behaviours.

• (2) Risk communication may help to correct Type 2 diabetes risk perception and lead to healthy behavioural changes.

• (3) Message framing and tailored risk feedback have been shown to be effective ways of communication, though no studies have examined them in combination.

What this topic adds:

• (1) Preliminary support for the brief online Type 2 diabetes risk communication intervention in correcting risk perceptions and increasing behavioural intentions.

• (2) Effectiveness of manipulation did not differ based on message manipulation.

• (3) Preliminary support for the use of the risk communication intervention in high-risk populations.

KEYWORDS:

• Type 2 diabetes
• risk communication
• risk perception
• alcohol consumption
• behavioural intentions

Diabetes is one of the leading causes of death in the world (World Health Organization, 2016). The number of adults living with diabetes has almost quadrupled since 1980 to 422 million adults, and the dramatic increase is largely due to the rise in Type 2 diabetes (T2D; World Health Organization, 2016). As a “modifiable disease”, up to 58% of T2D cases can be delayed or prevented by making positive changes to one’s lifestyle (Diabetes Australia, 2015). However, despite the strong association between modifiable lifestyle factors (e.g., physical activity and diet) and T2D risk, most at-risk individuals do not engage in these T2D risk-reducing behaviours (Geiss et al., 2010). Additionally, it is unknown how many of these individuals who do engage actually achieve the targets shown to be of benefit in reducing their risk of T2D. One potential barrier could be gaps in people’s knowledge and awareness about T2D symptoms and risk factors, which has resulted in discrepancies between reported awareness, motivation, and behaviour (Kayyali et al., 2019). The identification and communication of modifiable risk factors, via T2D risk assessment, presents a viable intervention that may help at-risk individuals better understand their T2D risk and motivate risk-reducing behavioural changes.

T2D risk assessments are increasingly accessible to the public. Online tools such as the Australian Type 2 Diabetes Risk Assessment Tool (AUSDRISK; Chen et al., 2010) and the Finnish Diabetes Risk Score (FINDRISC; Lindström & Tuomilehto, 2003) provide T2D risk estimates within minutes using several questions about diabetes risk factors and straightforward anthropometric measurements. This form of risk communication is categorised as personalised/tailored risk feedback, in which the risk communicated is based on the recipient’s individual characteristics. Compared to general health warnings, providing tailored risk feedback is more relevant to the individual and is therefore thought to be better processed, understood and more likely to lead to behavioural changes (Edwards et al., 2000). Tailored risk feedback has been shown to correct subjective risk perception (Hovick et al., 2014), improve rational decision-making (Hembroff et al., 2004), ensure adherence to recommended screening and health behaviours (Edwards et al., 2013), and identify those who may benefit from health interventions (Chen et al., 2010). With the feasibility, benefits, and ease of access to these online T2D risk assessment tools, health agencies and governments have embraced and widely utilised them in public health campaigns for primary prevention against T2D (Johnson et al., 2015).

Providing risk estimates and correcting risk perceptions may not be enough to significantly change behavioural intentions and drive health behaviours (Holmberg & Parascandola, 2010). Research in the cognitive and decision sciences has suggested that effective risk communication could be dependent on the way health messages are framed, because different frames can influence perceptions of risk and people’s decisions (Glare et al., 2018). Originating out of work on prospect theory, message framing posits that people’s decisions are sensitive to the way information is presented (Tversky & Kahneman, 1985). The theory states that when the messages are gain-framed (e.g., positive consequences of performing a behaviour), people are risk-averse, but when the messages are loss-framed (e.g., negative consequences of not performing the behaviour), they are risk-seeking. Accordingly, gain-framed appeals are argued to be more effective in promoting health-affirming (prevention) behaviours (e.g., physical activity; O’Keefe & Jensen, 2011) while loss-framed messages tend to be more effective in promoting illness-detecting (screening) behaviours (e.g., breast cancer screening; Gallagher et al., 2011). Hence, the use of gain-framed messages, in addition to personalised/tailored risk feedback, could be advantageous in spurring individuals to make lifestyle changes that reduce their T2D risk.

Currently, no study has sought to assess the combined effects of tailored risk feedback and message framing, though research on both types of message manipulations suggest promising results should they be used in combination. Meta-analyses and systematic reviews of either message framing or message tailoring tend to indicate a significant, though small, effect on health-related intentions and behaviours (French et al., 2017; Gallagher & Updegraff, 2012). With the combination of both message manipulations, individuals may achieve a more accurate risk perception (through personalised/tailored risk feedback) and feel more motivated to make positive lifestyle changes (from receiving a gain-framed message). This may lead to a larger effect on health-related intentions and behaviours than just using either manipulation alone.

People at high risk of T2D include individuals who have a substance use disorder (SUD), with a recent study indicating that 48% high-risk diabetes adults had SUD recorded in their medical records (Wu et al., 2018). It is surprising that T2D has been understudied in people with SUD (Walter et al., 2016), given that they commonly report unhealthy lifestyle behaviours that puts them at higher risk (Kelly et al., 2012; Vancampfort et al., 2019). People with alcohol use disorders (AUD) are of particular significance. A recent meta-analyses of T2D in people with AUD found that the T2DM prevalence rate observed in people with AUDs is similar to the T2DM prevalence observed in people with severe mental illness, which was double the relative risk for T2DM found in a matched background general population (Vancampfort et al., 2016). Heavy alcohol consumption can lead to negative consequences such as weight gain and high blood pressure which are risk factors of T2D (National Diabetes Services Scheme, 2020). Furthermore, a longitudinal population study has found heavy alcohol consumption in early adulthood to be significantly associated with increased risk of T2D and higher levels of its biomarkers throughout adulthood in men (T. Han et al., 2019). More importantly, the study suggested that the gradual reduction of alcohol consumption to moderate levels over the years may not necessarily lead to a decrease in T2D risk. This stresses the importance of early intervention in T2D prevention efforts and strategies for people with AUD.

This is the first in a series of studies aimed at developing a risk communication intervention that will support individuals who are at high risk of T2D, particularly those with SUD or AUD. The aim of this study was to assess the effects of the T2D risk communication intervention on T2D risk perceptions and behavioural intentions among an online sample of participants. Specifically, we examined whether tailored risk feedback (i.e., personalised vs. generalised) and message framing (i.e., gain vs. loss frame) have an effect on risk perception and behavioural intentions. We also assessed whether these effects varied based on levels of alcohol consumption. Lastly, we gathered feedback from participants to improve the intervention. We hypothesized that:

H1: There will be an increase in levels of risk perceptions and behavioural intentions and greater accuracy of risk perception across all conditions post-intervention.

H2A: There will be an interaction between message framing and message tailoring on behavioural intentions (i.e., physical activity and diet) and risk perception accuracy. Participants who receive the gain personalised manipulation will report a greater increase in behavioural intentions and greater accuracy in risk perception than participants who receive the other three conditions (i.e., gain generalised, loss personalised and loss generalised).

H2B: The aforementioned interaction effect will be similarly observed in participants who report high levels of alcohol consumption.

Method

Participants and design

Participants were recruited via Amazon’s Mechanical Turk (MTurk; see Mason & Suri, 2012) website and directed to the online survey software Qualtrics to complete the study. MTurk (mturk.com) is an online crowdsourcing platform where researchers recruit participants, otherwise known as “workers”, for intellectual tasks and workers voluntarily choose tasks to perform (see Buhrmester et al., 2011 for information about the reliability of data provided by MTurk samples). After completing tasks, workers receive a small amount of money as compensation.

Figure 1 provides a pictorial representation of the study procedure. On Qualtrics, participants gave informed consent and completed a pre-screen survey to ensure that they met the following inclusion criteria: (1) currently not diagnosed with diabetes, (2) score at least a moderate risk (<5 points) on the AUSDRISK, (3) people living in Australia and the United States, and (4) understand English. Participants meeting eligibility requirements in the pre-screen were offered the opportunity to complete the main study.

Figure 1. Study procedure

In the main study, eligible participants were stratified into either the low or high alcohol use group based on their level of alcohol consumption (measured using the Alcohol Use Disorder Identification Test) and within each group randomised to one of the four conditions by Qualtrics’s built-in randomiser. Participants completed a series of questions on T2D risk perception and behavioural intentions pre- and post-intervention. At the end of the survey, participants were asked to provide feedback regarding the risk communication intervention. Following the completion of the pre-screen survey and the main study, participants received US$0.50 and another US$0.50, respectively. Additionally, participants were given the option to download a personalised version of the risk communication intervention. Both the pre-screen survey and main study took approximately 15 minutes in total.

This study followed a number of recommendations to minimize participant misrepresentation from the use of MTurk and improve response/data quality (Aust et al., 2013; MacInnis et al., 2020; Wessling et al., 2017). This included accurate description of the study, blocking duplicate IP addresses and duplicate/suspicious geotag locations, ensuring fair payment (e.g., paying all participants rather than only those meeting screening criteria), and utilising a 2-step recruitment process. An instructional manipulation item (IMC) was also included to check for attention and reliable responding (Oppenheimer et al., 2009). The IMC reads, “Please click on the blue arrow at the bottom right of the screen. Do not move the scale.“ and is followed by a Likert Scale (from 0 to 10) with endpoints of “very rarely” to ”very frequently”. Participants will be excluded if they moved the scale (i.e., scored any number).

The research protocol was reviewed and approved by the University Human Research Ethics Committee. This trial was registered with the Australia and New Zealand Clinical Trials Registry (ANZCTR; ACTRN12619001421123).

Sample size and power calculation

Using “G-Power” (Faul et al., 2007), a priori power analysis indicated a sample size of 280 to be sufficient to attain power of .80 to detect a small effect size (f = .10), p-value of .05. This is based on meta-analyses of message framing and tailoring which have found significant but small effect sizes (O’Keefe & Jensen, 2007).

Health risk communication intervention (message manipulation)

Four different versions of the T2D risk communication intervention (i.e., personalised gain, personalised loss, generalised gain and generalised loss) were developed based on message framing and message tailoring manipulations that have been trialled online and across other populations (O’Connor et al., 2009; Zikmund-Fisher et al., 2008). The four versions were similar in length and structure, and included visual aids to promote greater recall and understanding of health and risk information (Garcia-Retamero & Cokely, 2017). The T2D health risk communication intervention consisted of three section:s

1. The general fact sheet on T2D. This section was standardised across all four versions and it discussed diabetes and its risk, risk factors and complications (Diabetes Australia, 2015; International Diabetes Federation, 2020).

2. The T2D risk section. This section consisted of either the personalised or generalised-framed message. The generalized-framed risk message provided the T2D risk category of the individual (e.g., moderate risk); the personalised-framed risk message not only provided the T2D risk category, but also shared the specific risk estimate (e.g., score 8 points, moderate risk, approximately one person in every 50 will develop diabetes) of the individual in text and graphic.

3. The lifestyle recommendation section. This section was constructed using clinical guidelines for T2D prevention that focused on the effects of health behaviour change (National Institute for Health and Care Excellence, 2017). Participants were randomised to receive either one of the four message manipulations. The generalised message provided general lifestyle advice to individuals to reduce the risk of T2D. The personalised messages further discussed specific steps to take to lower the risk (e.g., lose weight, get active and healthier diet). The gain-framed message discussed the positive impact on T2D risk by living a healthier lifestyle (e.g., “If you lose weight and keep it off, you may be able to prevent or delay diabetes.”). The loss-framed message discussed the negative impact on T2D risk of not living a healthier lifestyle (e.g., “If you do not lose weight and do not keep it off, you may not be able to prevent or delay diabetes.”). More details about the message manipulations can be found in the Supplementary Materials.

Measures

Alcohol use

To measure alcohol use, the 3-item Alcohol Use Disorder Identification Test (AUDIT-C) was used (Bush et al., 1998). The three items measure frequency of alcohol use, number/quantity of drinks, and binge drinking behaviour. Responses are rated on a five-item scale: 0 = never; 1 = less than monthly; 2 = monthly; 3 = weekly; and 4 = daily or almost daily. The AUDIT-C performs well as a brief screening tool in general population surveys (Aalto et al., 2009). The AUDIT-C can be used as a marker for high alcohol consumption and predicts hazardous drinking (Fujii et al., 2016). The widely used cut-off total score of 4 and above was used to indicate high alcohol use in both men and women.

Type 2 diabetes risk

T2D risk was examined using the Australian Type 2 Diabetes Risk Assessment Tool (AUSDRISK), a questionnaire developed specifically for the Australian population (Chen et al., 2010). It identifies individuals at high risk of developing T2D and consists of 11 items which assess demographic and diabetes risk factors: age, gender, country of birth, ethnicity, family history of diabetes, history of high blood glucose, hypertension, smoking status, fruit and vegetable intake, physical activity levels and waist circumference (in centimetres or inches). As the AUSDRISK was originally developed for an Australian sample population, the responses for “country of birth” and “ethnicity” were modified (based on how specific race/ethnic groups in America had been defined in Golden et al., 2012; Spanakis & Golden, 2013) to fit American participants. The maximum AUSDRISK score is 38, and under Australian guidelines, a score of ≥12 is considered high risk, a score of 6–11 is considered moderate risk and anything ≤5 is considered low risk. The AUSDRISK identifies both incident and prevalent undiagnosed diabetes, with the area under the receiver operating characteristic curves .783 and .781, respectively (Chen et al., 2010).

Perceived risk of Type 2 diabetes

Consistent with previous studies (e.g., Amason et al., 2016), T2D risk perception was assessed using two items from the Risk Perception Survey for Developing Diabetes questionnaire (RPS-DD). The RPS-DD is a validated questionnaire that measures the perception of risk for developing diabetes and factors that may modify perception of risk (Walker et al., 2003). The two items are: (1) What do you think your risk or chance is for getting diabetes over the next 10 years?; and (2) If you don’t change your lifestyle behaviours, such as diet or exercise, what is your risk or chance of getting diabetes over the next 10 years?. Responses are scored on a Likert-type scale of 0 (almost no chance) to 10 (high chance; Michigan Diabetes Research Training Center, 2010). The scale was scored as the average of both items and a higher score is interpreted as a higher diabetes perceived risk. An average score of >7 was considered high risk, a score of 3–7 was considered moderate risk, and anything <2 was considered low risk.

Accuracy of Type 2 diabetes risk

Dichotomous measures of accuracy were created for T2D risk by comparing participants’ actual and perceived T2D risk pre- and post-intervention. Risk perception is deemed to be accurate if perceived risk and actual risk are concordant. Participants were considered to have either improved (i.e., inaccurate to accurate), stayed the same, or worsened (i.e., accurate to inaccurate).

Behavioural intentions

According to the clinical guidelines (National Institute for Health and Care Excellence, 2017), both physical activity and diet are targeted as key health behaviours in Type 2 diabetes prevention in people at high risk. Based on widely used and recommended measures of behavioural intentions (Prestwich et al., 2003), behavioural intentions for physical activity and diet were measured, respectively, using three items each: e.g., In the next month: (i) “I intend to exercise more/eat healthier”, (ii) “I expect to exercise more/eat healthier”, (iii) “I will try to exercise more/eat healthier”. The items were rated on a 7-point scale ranging from (1) very unlikely to (7) very likely and combined into a sum score (Cronbach’s α = .96), with the average score used to indicate behavioural intentions.

Data analysis

Descriptive statistics were used to describe participants’ demographics and their feedback regarding the usefulness of the risk communication intervention. A 2 × 4 mixed model analysis of variance test (ANOVAs) using General Linear Model (GLM) was used to examine the main and interaction effects of message manipulations and alcohol use on T2D risk perception, physical activity and diet behavioural intentions. To explore between- and within-group differences, a series of post hoc analyses were performed using paired samples t-tests and McNemar’s tests (Adedokun & Burgess, 2011). Additionally, chi-squared tests of association was used to test for differences between groups in Type 2 diabetes risk perception accuracy at post-intervention The open-ended feedback were analysed via iterative categorization (Neale, 2016). Tests were two-tailed with p < 0.05. All analyses were performed using Statistical Package for the Social Sciences (SPSS) Version 25.

Results

Sample

The pre-screen survey on Qualtrics received 1,280 responses. After removing problematic responses, including 565 with duplicate IP addresses and 27 indicating their location to not be in either Australia or the United States, there were 688 responses remaining. Of these, 469 responses met inclusion criteria and were provided with a link to proceed on to the main study. A total of 347 completed surveys were collected and had correctly responded to the IMC, resulting in a 74.0% completion rate among those who were eligible for participation.

Fifty-five percentage of the sample were male (n = 189). Participants were mostly in the under 35 years old age group (n = 158; 46%), followed by 81 participants aged 35–44 years old (23%), 60 participants aged 45–54 years old (17%), 42 participants aged 55–64 years old (12%) and 6 participants aged 65 years old and above (2%). A large majority of participants were born in USA (n = 320; 92%) and the rest were either born in Asia (n = 13; 4%), Australia (n = 7; 2%) or other parts of the world (n = 7; 2%). One-third of participants (n = 114) had previously received treatment for mental health problems, while 39% of participants (n = 135) met criteria for high alcohol use (i.e., scores ≥4) based on the AUDIT-C. Overall, the sample’s average score on the AUSDRISK was 11.61 points (SD = 4.61), bordering between the moderate and high-risk category. Table 1 details the participants’ characteristics as stratified by intervention groups.

Table 1. Participant’s characteristics stratified by intervention groups

	Personalised Gain (N = 85)	Personalised Loss (N = 77)	Generalised Gain (N = 91)	Generalised Loss (N = 94)
Gender
Male	48 (56.5%)	43 (55.8%)	46 (50.5%)	52 (55.3%)
Female	37 (43.5%)	34 (44.2%)	45 (49.5%)	42 (44.7%)
Age
Under 35 years old	41 (48.2%)	45 (58.4%)	35 (38.5%)	37 (39.4%)
35–44 years old	25 (29.4%)	7 (9.1%)	23 (25.3%)	26 (27.7%)
45–54 years old	8 (9.4%)	11 (14.3%)	22 (24.2%)	19 (20.2%)
55–64 years old	9 (10.6%)	12 (15.6%)	22 (24.2%)	10 (10.6%)
65 years old and above	2 (2.4%)	2 (2.6%)	11 (12.1%)	2 (2.1%)
Country
Australia	1 (1.2 %)	1 (1.3%)	2 (2.2%)	3 (3.2%)
United States	80 (94.1%)	73 (94.8%)	81 (89.0%)	86 (91.5%)
Others	4 (4.7%)	3 (3.9%)	8 (8.8%)	5 (5.3%)
Education
High school diploma or less	19 (22.4%)	18 (23.4%)	25 (27.5%)	24 (25.5%)
Associate’s/ technical degree	13 (15.3%)	14 (18.2%)	16 (17.6%)	19 (20.2%)
Bachelor’s degree	42 (49.4%)	37 (48.1%)	39 (42.9%)	39 (41.5%)
Postgraduate degree	11 (12.9%)	8 (10.4%)	11 (12.1%)	12 (12.8%)
AUDIT-C Category
Low	51 (60%)	48 (62.3%)	56 (61.5%)	57 (60.6%)
High	34 (40%)	29 (37.7%)	35 (38.5%)	37 (39.4%)
AUSDRISK Category
Intermediate	55 (64.7%)	42 (54.5%)	58 (63.7%)	54 (57.4%)
High	30 (35.3%)	35 (45.5%)	33 (36.3%)	40 (42.6%)

Type 2 diabetes risk perception, physical activity and diet

A 2 × 4 mixed model ANOVA did not reveal a significant interaction between message manipulation, alcohol use, and time on T2D risk perception, behavioural intentions for physical activity or diet (Table 2). There were also no significant interaction effects between message manipulation and alcohol use, message manipulation and time, or alcohol use and time. Additionally, no significant main effects were found for message manipulation and alcohol use. However, there was a statistically significant main effect for time on T2D risk perception, physical activity, and diet. Paired-samples t-test analyses indicated that T2D risk perception, physical activity and diet scores significantly increased post-intervention across all intervention groups (Table 3).

Table 2. Mixed model ANOVA analysis results

	F-value (df1, deferrer)	p-value	ηp^2
Time
Type 2 Diabetes Risk	111.87 (1, 339)	.000*	.248
Physical Activity	60.83 (1, 339)	.000*	.152
Diet	96.60 (1, 339)	.000*	.222
Message Manipulation (MM)
Type 2 Diabetes Risk	.189 (3, 339)	.904	.002
Physical Activity	1.092 (3, 339)	.353	.010
Diet	.575 (3, 339)	.632	.005
Alcohol Use (AU)
Type 2 Diabetes Risk	1.986 (1, 339)	.160	.006
Physical Activity	.318 (1, 339)	.573	.001
Diet	.309 (1, 339)	.579	.001
Interaction between MM & AU
Type 2 Diabetes Risk	.769 (3, 339)	.512	.007
Physical Activity	.075 (3, 339)	.973	.001
Diet	.400 (3, 339)	.753	.004
Interaction between time & MM
Type 2 Diabetes Risk	1.37 (3, 339)	.252	.012
Physical Activity	.625 (3, 339)	.599	.005
Diet	.202 (3, 339)	.895	.002
Interaction between time & AU
Type 2 Diabetes Risk	.393 (1, 339)	.531	.001
Physical Activity	1.387 (1, 339)	.240	.004
Diet	.409 (1, 339)	.523	.001
Interaction between time, MM & AU
Type 2 Diabetes Risk	.160 (3, 339)	.923	.001
Physical Activity	.962 (3, 339)	.411	.008
Diet	.255 (3, 339)	.858	.002

*p-value < 0.001

Table 3. Differences in Type 2 diabetes risk and behavioural intention scores post-intervention (N = 347)

	Type 2 Diabetes Risk					Physical Activity					Diet
Intervention Group	Pre Mean (SD)	Post Mean (SD)	t (df)	sig	Effect size (d)	Pre Mean (SD)	Post Mean (SD)	t (df)	sig	Effect size (d)	Pre Mean (SD)	Post Mean (SD)	t (df)	sig	Effect size (d)
Gain & Personalised (n = 85)	3.00 (.29)	4.17 (.31)	6.35 (84)	.000	.689	4.88 (.22)	5.25 (.22)	4.43 (84)	.000	.481	4.74 (.22)	5.24 (.22)	5.51 (84)	.000	.597
Loss & Personalised (n = 77)	3.13 (.30)	4.47 (.33)	5.86 (76)	.000	.668	4.70 (.24)	5.02 (.23)	2.88 (76)	.005	.328	4.51 (.23)	4.99 (.23)	4.96 (76)	.000	.566
Gain & Generalised (n = 91)	3.32 (.28)	4.16 (.30)	3.87 (90)	.000	.406	4.55 (.22)	4.93 (.21)	4.60 (84)	.000	.482	4.73 (.21)	5.16 (.21)	4.63 (90)	.000	.486
Loss & Generalised (n = 94)	2.83 (.27)	4.26 (.30)	6.36 (93)	.000	.656	4.29 (.21)	4.79 (.21)	4.89 (93)	.000	.505	4.45 (.21)	4.86 (.21)	4.90 (93)	.000	.506

Accuracy of Type 2 diabetes risk

Table 4 displays participants’ accuracy of Type 2 diabetes risk from two perspectives: (1) the number and percentage of participants who improved (i.e., inaccurate to accurate risk perception), stayed the same, or worsened (i.e., accurate to inaccurate risk perception), and (2) whether these changes in accuracy were statistically significant. McNemar’s test indicated that the proportion of participants who improved their T2D risk perception was significantly greater than the proportion of participants who worsened and this result was consistent across all groups. Overall, approxmiately 25% of participants reported an improvement from an inaccurate to an accurate T2D risk perception while 56% of participants maintained an inaccurate perception of T2D risk. Chi-square analyses did not reveal any significant association between intervention group and accuracy in T2D risk perception (Χ²(9) = 9.17, p = 0.42. All four manipulations had similar effects on accuracy of T2D risk perception and none was significantly superior.

Table 4. Change of accuracy in T2D risk perception from pre-post T2D risk message (N = 347)

	Improved	Stayed the Same		Worsened
	Pre-Post	Pre-Post	Pre-Post	Pre-Post	Change
Intervention Group	Inaccurate to Accurate	Both Accurate	Both Inaccurate	Accurate to Inaccurate	p-value^a
Gain & Personalised (n = 85)	20 (23.5%)	16 (18.8%)	44 (51.8%)	5 (5.9%)	.004
Loss & Personalised (n = 77)	18 (23.4%)	11 (14.3%)	45 (58.4%)	3 (3.9%)	.001
Gain & Generalised (n = 91)	21 (23.1%)	8 (8.8%)	53 (58.2%)	9 (9.9%)	.043
Loss & Generalised (n = 94)	29 (30.9%)	10 (10.6%)	51 (54.3%)	4 (4.2%)	.000

^ap-value based on McNemar Test statistic (Chi-square Test)

Feedback

Participants were asked to provide comments about the usefulness of and level of “surprise” with the T2D risk message, and any other open-ended feedback (not mandatory). Overall, across all intervention groups, 74% of participants rated the message to be either “somewhat useful” or “extremely useful”. Additionally, most participants either had rated the information to be “not surprising” (48.6%) or were “surprised at how high their risk was than expected” (48.6%). The study received 89 open-ended responses and the main themes are summarised in the Supplementary Materials. A third of the comments were positive feedback on the health message and the rest were mostly suggestions centred around having more information on individual diabetes risk, risk factors, and lifestyle recommendations.

Discussion

This study is the first to examine the interaction effects of tailored risk feedback and message framing on T2D risk perceptions and behavioural intentions. There was a main effect for time indicating that accuracy of T2D risk perceptions, diet, and physical activity intention scores significantly increased from pre- to post-intervention across all groups. However, these changes were not significantly greater for the gain/personalised group. Taken together, the findings suggested that the brief online T2D risk communication intervention was effective in correcting participants’ accuracy of risk and increasing intentions to engage in healthier lifestyle behaviours, although the effectiveness did not differ based on the type of message manipulation.

The null findings of main and interactions effects were unexpected but not surprising, considering that past research have also found similar results for either message tailoring or message framing alone (e.g., Gallagher & Updegraff, 2012). The null results corroborate a recent study (Lipkus et al., 2019) which similarly did not find a significant main or interaction effects for message framing and “tailored risk”, though the authors had defined “tailored risk” into two categories (i.e low-risk estimate vs. high-risk estimtes) instead of risk being personalised (or not) to the individual as in this study. Overall, the lack of significant effects could be attributed to a few reasons. Firstly, the framing and tailoring manipulations being too brief and subtle to have had an impact on participants’ scores. It may be possible that participants need a longer time or repeated messages to fully comprehend or internalise the risk information (Suka et al., 2020). Secondly, it is possible that being a brief online intervention, the degree to which messages were personalised was limited. Lastly, it is noted that effect sizes tend to be larger when measures of behaviour rather than attitudes or intentions are used to assess the persuasive impact of framed messages (Gallagher & Updegraff, 2012). It is recommended that future studies adopt a longitudinal approach to allow for examination of behavioural change over time.

T2D risk perception scores reported post-intervention were significantly higher in all study arms. Additionally, there were significantly greater accuracies in T2D risk perception, with about 25% of participants showing improvements. This finding was similar to the results of another study (Silarova et al., 2018) and adds to the body of literature demonstrating improved risk perception accuracy after risk assessment feedback. Overall, it reflects the practicality of the risk communication intervention in improving levels and accuracy of T2D risk perceptions. More importantly, the results lend support to the adequacy of utilising a brief online risk assessment tool to communicate T2D risk and other chronic diseases. With increasing reliance on electronic communication over face-to-face consulations secondary to COVID-19 pandemic, it is vital that viable tools are available to aid medical professionals in enhancing self-management among patients. Research has shown that an online risk assessment tool can encourage greater involvement in decision-making and promote an active role in care (Manuel et al., 2018). As such, the use of online risk assessment tools can add value to telehealth options by allowing people to monitor and self-manage their risk of chronic diseases (McCoy et al., 2005).

It is important to highlight that the majority of participants still had inaccurate risk perceptions. Qualitative studies in cancer research suggest possible explanations that include personal or lay theories of disease and risk (Heiniger et al., 2015), differences between laypersons’ understanding of risk information and clinical risk information (P. K. J. Han et al., 2009), and past experiences, expectations and beliefs (Holmberg et al., 2015). Further research is needed to explore the factors identified in aforementioned qualitative studies when communicating risk to participants with different baseline risk perception.

Behavioural intention scores were also significantly increased in all study arms. This finding suggests that the online T2D risk communication intervention may be useful in increasing motivation or readines to change, which is widely seen as the first step towards lifestyle behavioural changes in the long run (DiClemente et al., 2004). This would be particularly important for people with SUD, who often report a lack of motivation or readines to change their unhealthy lifestyle behaviours (Myers et al., 2016). Indeed, studies have shown that by targeting these factors, it could lead to a change of health behaviours, such as alcohol reduction (Bertholet et al., 2009; Collins et al., 2012). Based on this result, it may be that personalised risk communications may be best suited to motivating people to engage in effective behaviour change programmes, by motivating attempts to change behaviour. Given this, it would be beneficial to compare these risk communication strategies in terms of whether they promote uptake of evidence-based behaviour change programmes, since it appears unlikely that sustained behaviour change will be brought about solely by communicating personalised risk.

One of the aims of the current study was to determine whether effects of risk communications would generalise to those with higher levels of alcohol consumption, prior to engaging in studies where those with SUD/AUD were targeted. There was no significant interaction effect between alcohol consumption, message manipulation, and time. This indicates that the level of alcohol consumption or type of message manipulation did not have an impact on any of the outcomes measures. Though the hypothesis was not supported, this also suggests that regardless of the level of alcohol consumption, the online risk communication intervention can help to correct participants’ risk perception and improve behavioural intentions (as mentioned above). This preliminary result strengthens the argument of using online risk assessment tools to drive behavioural changes in people with SUD, a high-risk population who are traditionally reluctant to seek help until forced to do so or until their problems become severe (Cunningham et al., 1993; Luitel et al., 2017). Assuming that the online T2D risk communication intervention leads to the average small-medium effect size behavioural change (Kohl et al., 2013; Webb & Sheeran, 2006), it would be considered a significant improvement for this high-risk population particularly when the intervention is digitalised and automated. Future studies are warranted to investigate the feasiblity of the online risk communication intervention in people with SUD and other high-risk populations, particularly in the reductions of risk factors.

The intervention was generally well received, with a majority of participants reporting that the intervention was useful and one-third of the open-ended responses being positive feedback. Furthermore, 46% of participants were surprised by how high their T2D risk was. This might explain the significant increase in behavioural intentions post-intervention, as participants correct their T2D risk perception and felt the need to make lifestyle changes.

Limitations

The study has a few limitations which indicate caution when intepreting the results. Recruitment via MTurk may not result in participants that are representative of general community samples. Additionally, the young study population is not indicative of those at higher risk of T2D, which is generally advised to be of those aged 45 and above. Therefore, future studies are recommended to employ recruitment methods which are the most suitable to reach the target population (e.g., people with SUD or at higher risk of T2D) and allow generalisability of these results.

Responses regarding behavioural intent do not necessarily translate into actual behavioural change and the cross-sectional nature of these data do not allow us to ascertain if participants acted upon their intentions. Further, the design did not include a no risk feedback/no framing arm and therefore changes in risk perceptions and intentions cannot be definitively attributed to the intervention. Despite this limitation, pragmatically risk assessment and communication is advocated in a wide variety of contexts and this study indicates that improved risk perceptions and health behaviour intentions coincided with the intervention. However, it is still recommended for future research to utilise a longitudinal and control group design to strengthen the credibility and validity of the findings (Buch, 2016; Kinser & Robins, 2013).

Clinical implications

Despite the positive results, the intervention is still in its preliminary stages and clearly needs follow-up to assess for actual behavioural changes. The intervention could be partnered with existing behaviour change programmes(e.g., Keane et al., 2016) that are being trialled in drug and alcohol services to enhance its effectiveness. Furthermore, the easiness and feasibility of using the risk communication intervention could help to address the lack of T2D risk screening in the healthcare sector. Anyone entering rehabilitation, outpatient services, or in waiting rooms could be screened for T2D risk quickly and cost-efficiently, with minimum to no staff required.

Conclusion

In summary, study findings highlight the potential and advantages of leveraging an online risk assessment tool to communicate personalised T2D risk. The utility and benefits of the tool were also endorsed by participants in this study. These preliminary findings are encouraging and support the continued development of online risk assessment and communication to help combat the current T2D epidemic.

Disclosure statement

No potential conflict of interest was reported by the author(s).