Randomized trial of planning tools to reduce unhealthy snacking: Implications for health literacy

Objective Guidance to address health literacy often focuses on health education rather than tools to facilitate action, despite action being important for self-management. This study evaluated an online intervention informed by health literate design principles and behavior change theory to reduce unhealthy snacking. Methods 440 participants were recruited online and randomized to an intervention: 1) Health-literate action plan (guided implementation intention); 2) Standard action plan (self-guided implementation intention); 3) Education (healthy snacking fact-sheet). The primary outcome was self-reported unhealthy snacking. Follow-up was at 1 month. Results 373 participants (84.8%) completed follow-up. Half the sample had adequate health literacy (52%), and the other half had low (24%) or possibly low (25%) health literacy, as measured by Newest Vital Sign (NVS). At follow-up, lower health literacy was associated with more unhealthy snacks and there was no overall difference between intervention groups. However, participants with lower health literacy who used the health-literate action plan reported less unhealthy snacking compared to the standard action plan; the reverse was true for those with higher health literacy scores (b = 1.7, p = 0.03). People scoring 2 points below the mean NVS (M = 3.4, SD = 2.0) using the health-literate action plan reported eating 8 fewer serves of unhealthy snacks, whereas people scoring 2 points above the mean NVS reported eating 6 more serves of unhealthy snacks using the same tool. Conclusions These findings suggest that the universal precautions approach currently recommended for health information may be less effective for facilitating action than tailoring to health literacy level. Trial registration ANZCTR identifier: ACTRN12617001194358.

However, it can be difficult to create effective implementation intentions. Gollwitzer, Wieber [37] identified several criteria for effective implementation intentions, including strong behavioral intentions, an easily recognizable and frequently occurring cue, and a behavioral response that contributes to the goal and is neither too challenging nor too easy. The 'volitional help sheet' [38] is a specific type of implementation intention tool that can increase the likelihood that these criteria are met. This tool guides the user to select the most relevant 'if' and 'then' statements from a predetermined list of cues and responses. Studies have also shown that this tool is effective for lifestyle behavior change [39][40][41][42][43][44].
The current study investigated the effectiveness of an online action plan intervention to reduce unhealthy snacking. High intake of discretionary foods (i.e. 'snacks' including cakes, biscuits, chocolate and confectionary) has been identified as an important contributor to unhealthy diets in Australia and worldwide [45]. This intervention was designed to meet current best practice guidelines for online health literacy interventions, such as simple language and use of white space [19]. Plan formation was facilitated through an online volitional help sheet. This was compared to two controls: a standard action plan (a standard implementation intention) and education (a fact sheet on healthy snacking). It was hypothesized that health literacy would moderate the effect of the interventions on unhealthy snacking with lower literacy participants showing better outcomes using the health-literate action plan design

Study design
This between-subjects randomized trial compared the effects of three interventions on unhealthy snacking over one month. This study was approved by the University of Sydney Human Research Ethics Committee [2017/622]. The study was prospectively registered with ANZCTR (www.anzctr.org.au) (ACTRN12617001194358).

Participants
The market research company Research Now Survey Sampling International independently recruited Australian adults (N = 440) to participate in an online survey in return for points that can be redeemed for gift vouchers. Quota sampling ensured representation by lower educated participants with 50% of the sample with less than university education, and equal numbers of men and women. Inclusion criteria were fluency in English and age over 30 years to maintain relevance to chronic and lifestyle conditions such as type 2 diabetes. Participants were not selected on the basis of current snacking behavior or BMI.

Procedure
Surveys were hosted by Qualtrics. After completing informed consent, demographics and baseline measures, participants were automatically randomized evenly to an intervention group (see Box 1 and Figs 1 and 2) using the Qualtrics 'randomizer,' and asked post-intervention to make a digital or paper copy of their plan. Participants were blind to the intervention to which they were randomized. Participants received emails containing their plan at baseline, 1 and 2 weeks. Participants completed the follow-up survey at 4 weeks.

Measures
Health literacy. Health literacy measures included the Newest Vital Sign (NVS) [47], a 6-item measure of functional health literacy. NVS scores of 0-1 indicate a high likelihood of

Box 1: Description of interventions
Health-literate action plan Description: 1. Participants chose 3 scenarios they found most difficult to eat healthily (Panel a, Fig 2). The list of scenarios was based on previous studies [43,44], adapted for unhealthy snacking. Participants then selected the scenario they were most motivated to address (Panel b, Fig 2).
2. Participants dragged an action from a list and placed it underneath the scenario (Panel c, Fig 2). This created the implementation intention i.e. "If [scenario], then I will [action]." Participants were asked to imagine the plan (Panel d, Fig 2) and rate how difficult they would find it over the next month using a 10-point scale (Very easy to very hard). Those scoring 7 or more were asked to revise their action.
Active elements: • Implementation intention (focus on cues for action) • High quality plan (Cue is likely to be high frequency and personally relevant, solution is not too challenging) • Health-literate design for comprehension, including minimal and simple text, illustrative images and white space [19].

Standard action plan
Description: Participants read instructions similar to those used by Armitage [40]: "We want you to plan how you will change your unhealthy snacking behavior each day because forming plans has been shown to improve snacking habits. You are free to choose how you do this but we want you to formulate your plans in as much detail as possible. Please pay attention to the situations in which you will implement these plans. Focus on situations when you are not hungry but find yourself snacking." Active elements: • Implementation intention (focus on cues for action)

Education
Description: The fact sheet was based on the Australian National Diabetes Services Scheme 'Healthy snacks' fact sheet [46], with references to diabetes, carbohydrates and blood glucose removed.
Active elements: • Information about healthy snacking options Snack scores (previous month). A 7-item measure of snacking was based on a diet score developed and validated by a well-established and highly regarded Australian scientific organization (CSIRO) [45]. Items were drawn from the 'discretionary foods' category which the Australian Guidelines to Healthy Eating define as 'not considered necessary for a healthy diet'. Alcohol and sugar sweetened beverages were excluded from the assessment in this study as the focus was on 'snacks.' Participants answered how many serves of unhealthy snacks they ate in the past month. Participants could answer according to the number of serves per day, week or month. Average weekly servings of unhealthy snacks were calculated from these scores.
Cognitive variables. Cognitive variables with theoretical importance for the action phase of behavior change were included [7]. Items were based on existing measures of selfefficacy [7], planning [7] and adapted to refer specifically to unhealthy snacking. These were measured at baseline and follow-up, with the exception of habit strength (baseline only) and action control (follow-up only). Responses were recorded on 7-point Likert scales (strongly disagree to strongly agree).
Difficulty using the planning tool. A single item asked participants to rate how hard it was to use the planning tool (1 = not at all hard, 5 = extremely hard).

Analysis
Sample size. A sample size estimate of 261 was based on 90% power at α = 0.05 to detect a moderate effect size (Cohen's f = 0.25, corresponding to ANCOVA effects and interaction as small as ηp 2 = 0.06) in the primary outcome (unhealthy snacking score). With an anticipated attrition rate of 40%, this totaled 435 participants. Effect size estimate was based on a previous analysis a of volitional help sheet intervention [40].
Analysis methods. Randomization at baseline was assessed using ANOVA (or Kruskal-Wallis H test for non-normal data) for continuous variables and chi-square tests for categorical variables. Multiple linear regression was used to predict snacking scores and perceived difficulty using the plan at follow-up. Contrasts compared participants allocated to a) the healthliterate and standard action plans; and b) the latter two plans and education. Important correlates of health literacy (age, level of education, language spoken at home) [1], baseline snacking and health literacy (as NVS scores) were controlled for in the model, including an intervention group × health literacy interaction term. NVS scores were examined both continuously and categorically. To assist with interpretation of the effect on snacking behavior, the effect of NVS scores were expressed in 2-point increments as this corresponds to a meaningful shift in health literacy category [47].
Each cognitive variable (baseline habit strength; follow-up intention, action control, planning and self-efficacy) was independently assessed as a potential mediator of the effect of intervention group and health literacy on snacking behavior. Mediation was evaluated using Hayes [52]' mediation method (PROCESS V3.0). Each assessment of mediation for follow-up cognitive variables included the corresponding baseline measure as a covariate where available), as well as baseline snacking, age, level of education and language spoken at home. Model 4 was employed for main effects (a mediation model testing the direct effect of intervention group and health literacy on snack scores, the effect of the potential mediator on snack scores, and the indirect effect of intervention group and healthy literacy on snack scores via the potential mediator), and Model 8 was employed for interaction effects, similar to Model 4 but with health literacy included as the moderator of the effect of intervention group on snack scores. Mediation was evaluated using 95% confidence intervals generated from 5,000 bootstrapped samples. For contrasts, only the relevant groups were included in the mediation analysis.
Two researchers independently coded standard action plans to indicate the extent that participants followed standard action plan instructions and the extent that plans differed from the pre-determined options presented in the health-literate action plans. Coders were blind to the health literacy level of participants. Any disagreements were resolved through discussion.

Results
Three hundred and seventy-three of the 440 participants (84.8%) completed the follow-up survey (Fig 1). Descriptive statistics are shown in Tables 1-3. There were no significant differences in age, gender, BMI, intention, habit strength, NVS score and baseline snacking scores across intervention groups at baseline.
At baseline, participants scored around the midpoint of all cognitive variables (habit strength, planning and self-efficacy). Intention scores were on average slightly higher, at approximately 5 out of 7, indicating a moderately positive intention (Table 3). Table 4 shows the frequency of selected health-literate action plan scenarios and solutions. Twenty-eight participants (22%) revised their solution after being advised to choose an easier plan (See S2 File for examples).

Standard action plan
Average plan length was 23.3 words (SD = 20.0). This varied by health literacy level. Plan length was 24.0 words (SD = 19.9) for participants with adequate health literacy, and 10.3 words (SD = 12.0) for participants with high likelihood of limited health literacy. One third of participants with high likelihood of limited health literacy did not create a plan, compared to 13% of participants with adequate health literacy (Table 5).

Impact of interventions on snacking behavior
The overall model was significant (F (9, 363) = 9.0, p<0.001), explaining 18.3% of variance in snack diary scores. Of the explained variance, 15.5% can be attributed to health literacy and intervention group (ΔR 2 from base model = 0.03, F (5, 363) = 2.5, p = 0.03). Overall this represented a small effect on unhealthy snacking [53]. There was no main effect of intervention group. The model suggested that NVS score significantly predicted snack scores, controlling for age, language spoken at home, education and baseline snacking score, such that for each point increase on the NVS (i.e. higher health literacy), unhealthy snacks per week were expected to decrease by 1.65 serves (SE = 0.67, p = 0.01) ( Table 6). alcohol', 'it is part of a celebration or special event', 'I have arrived home', 'someone offers me a snack', and 'I am about to go to bed.' †'Other solutions with frequency <5%' include: 'listen to music,' 'take the food out of the packet and put it on a plate,' 'move the snack into the cupboard,' 'politely say "No thank you",' and 'take a break.' There was a significant interaction between NVS scores and the health-literate and standard action plan groups (b = 1.74, SE = 0.78, p = 0.03). As shown in Fig 3, this suggests that participants with lower health literacy benefited more from the health-literate action plan compared to those using the standard action plan. For example, a person scoring 2 points below the mean on the NVS would be expected to eat 7.7 fewer serves of unhealthy snacks per week using the health-literate action plan rather than the standard version. By contrast, participants with higher health literacy had significantly lower snack scores when they used the standard action plan, such that a person who scored 2 points above the mean on the NVS would eat 6.2 fewer serves of unhealthy snacks per week using the standard action plan, as compared to the health-literate action plan.
Cognitive variables were assessed as potential mediators for: 1) the main effect of health literacy on snacking; and 2) the interaction between intervention group contrast 1 (health-literate action plan vs standard action plan) and health literacy. Neither health literacy nor the intervention group contrast predicted scores on cognitive variables at follow-up. Furthermore, there was no statistical evidence that the cognitive variables mediated the effects of health literacy and intervention group on snacking scores.

Discussion
This study found that lower health literacy scores predicted greater consumption of unhealthy snacks after 1 month of using an online intervention. This effect was moderated by the type of intervention. People with lower health literacy ate fewer serves of unhealthy snacks when they used a health-literate action plan rather than the standard version. Conversely, people with higher health literacy ate fewer unhealthy snacks when they used the standard action plan rather than health-literate version. This study did not find evidence of mediation by cognitive variables (habit strength, intention, self-efficacy, planning or action control) of the effect of health literacy and intervention group on self-reported snacking scores at follow-up. This study opens up an important discussion about how the universal precautions approach to health literacy (that is, using a lower health literacy design for all consumers) might apply to self-management behavior change interventions. Whilst numerous studies have found that the universal precautions approach offers benefits to people with lower health literacy without compromising outcomes for people with higher health literacy [1][2][3][4], this research has largely focused on information comprehension, recall, acceptability of the intervention and implications for service use and cost [1,2]. Comparatively few studies have investigated the effect of health-literate design on the action phase of behavior change interventions. Furthermore, no studies have specifically investigated the effects of health-literate design on action plans which seek to reduce the intention-behavior gap.
The findings from the present study suggested that the universal precautions approach may adversely impact outcomes for people with higher health literacy. This highlights that education and action may not operate via the same mechanisms despite both being common components of health interventions. The universal precautions approach that is applied to education is underpinned by the principle of reducing cognitive demand. By contrast, action plans are inherently cognitively demanding because of the high level of self-reflection required [18,23]. Goal setting theories also argue that cognitive effort can actually facilitate engagement as long it is not too challenging [23].
In the present study the health-literate action plan facilitated plan formation with the aim of reducing cognitive demand compared to the standard version. For example, the planning process was broken down into steps, users were restricted to a single plan, and there was no option for free text. Participants with lower health literacy benefited from this approach. Not only were their reported snacking scores lower using the health-literate action plan, but one third of participants with a high likelihood of inadequate health literacy did not create a plan, and almost one quarter created plans that lacked specificity.
By comparison, participants with higher health literacy may have felt restricted by the options available in the health-literate action plan and were able to adequately generate their own effective plans using the standard version. For example, many of these plans included strategies that were not explicitly included in the health-literate action plan, such as preparing healthy snacks ahead of time so that they would be ready and available to eat. Together, these results for participants with lower and higher health literacy support the idea that tailoring intervention components that focus on the action phase may be a more effective way to promote lifestyle behavior change in diverse populations (Fig 4). and health literacy levels, providing evidence that online self-management tools can be appropriate for target populations for chronic disease prevention and self-management.

Limitations
A limitation was that the snack measure asked participants to self-report food consumed over the previous month. This was mitigated by using a validated dietary measure [45] that asked about specific categories of foods to assist recall. Randomization also reduced the risk of differential self-reported snacking across intervention groups.
It should also be noted that all participants who used the health-literate action plan had to create a plan, whereas participants using the standard action plan could submit text that did not follow the instructions. As such, it is possible that the mandatory creation of plans may also have contributed to the effectiveness of the health-literate action plan.

Future directions
This study provides a starting point for investigating how health literate design can best be applied to action plans for lifestyle change. For example, future research should investigate whether simply incorporating additional options into the health-literate action plan would improve its effectiveness for people with higher health literacy. It would also be useful to investigate whether individuals can reliably self-select the type of action plan for their health literacy level, or whether a screening item for health literacy could adequately fulfil this purpose.

Conclusions
This study addressed an important gap in the literature by examining how an action plan intervention can be adapted for a lower health literacy audience to improve self-management (reducing unhealthy snacking). The results indicated that the effectiveness of the intervention was dependent on the health literacy of the participant; participants with lower health literacy benefited from health-literate design in which cognitive demand was reduced, whereas users with higher health literacy benefited from a design that gave them freedom to generate their own detailed and highly personalized plans. These findings have two key implications: (1) health literacy interventions that target people with lower health literacy should aim to reduce the cognitive demand required to create action plans; and (2) people with higher and lower health literacy may need different strategies for creating effective action plans.