Evaluating the effect of individually-targeted food incentives on grocery purchases: The smart cart study protocol for a randomized controlled cross-over trial

Participant Characteristics Associated with High Responsiveness to Personalized Healthy Food Incentives: a Secondary Analysis of the Randomized Controlled Crossover Smart Cart Study

BACKGROUND

Personalized dietary behavioral interventions could be enhanced by understanding factors accounting for individual variability in dietary decisions.

OBJECTIVE

This study was a secondary analysis of the Smart Cart randomized controlled trial to determine whether participant characteristics predicted high responsiveness to personalized grocery incentives for purchasing healthy food.

METHODS

This secondary analysis of a 9-mo crossover study included 192 regular shoppers (86%) from a Rhode Island supermarket. To analyze whether health, behavioral, and/or sociodemographic characteristics predicted responsiveness to a personalized grocery incentive intervention, participants were divided into 3 categories [high (n = 47), moderate (n = 50), and unresponsive (n = 95)] based on percentage changes in their Grocery Purchase Quality Index scores during the intervention versus control period calculated from sales data. We tested whether participant characteristics, including individual, household, and intervention-related factors, differed across responsiveness groups using ANOVA and whether they predicted the likelihood of being high responsive versus unresponsive or moderate responsive using logistic regression.

RESULTS

Participants had a mean (SD) age of 56.0 (13.8) y and were 89% female. Education, self-reported BMI, income, diet-related medical condition, food insecurity, cooking adequacy, and value consciousness differed across responsiveness categories (P < 0.1). High versus moderate and unresponsive participants increased their percentage of spending on targeted foods (P < 0.0001) and purchased fewer unique items (P = 0.01). In multinomial adjusted models, the odds of being high versus unresponsive or moderate responsive were lower for participants with a BMI (in kg/m2) <25 versus ≥25 (OR: 0.41; 95% CI: 0.19, 0.90) and higher with a diet-related medical condition present (OR: 3.75; 95% CI: 1.20, 11.8). Other characteristics were not associated with responsiveness.

CONCLUSIONS

Findings demonstrated that a BMI ≥25 and having a diet-related medical condition within the household predicted high responsiveness to a personalized grocery purchasing intervention, suggesting that personalized dietary interventions may be particularly effective for households with higher health risk. This trial is registered at www.clinicaltrials.gov as NCT03748056.

The utility of household Grocery Purchase Quality Index scores as an individual diet quality metric

The Grocery Purchase Quality Index (GPQI) reflects concordance between household grocery purchases and US dietary recommendations. However, it is unclear whether GPQI scores calculated from partial purchasing records reflect individual-level diet quality. This secondary analysis of a 9-month randomised controlled trial examined concordance between the GPQI (range 0-75, scaled to 100) calculated from 3 months of loyalty-card linked partial (≥50 %) household grocery purchasing data and individual-level Healthy Eating Index (HEI) scores at baseline and 3 months calculated from FFQ (n 209). Concordance was assessed with overall and demographic-stratified partially adjusted correlations; covariate-adjusted percentage score differences, cross-classification and weighted κ coefficients assessed concordance across GPQI tertiles (T). Participants were middle aged (55·4 (13·9) years), female (90·3 %), from non-smoking households (96·4 %) and without children (70·7 %). Mean GPQI (54·8 (9·1) %) scores were lower than HEI scores (baseline: 73·2 (9·1) %, 3 months: 72·4 (9·4) %) and moderately correlated (baseline r 0·41 v. 3 months r 0·31, P < 0·001). Correlations were stronger among participants with ≤ bachelor's degree, obesity and children. Scores showed moderate agreement (κ = 0·25); concordance was highest in T3. Participants with high (T3) v. low (T1) GPQI scores had 7·3-10·6 higher odds of having HEI scores >80 % at both time points. Household-level GPQI was moderately correlated with self-reported intake, indicating their promise for evaluating diet quality. Partial purchasing data appear to moderately reflect individual diet quality and may be useful in interventions monitoring changes in diet quality.

Effect of Personalized Incentives on Dietary Quality of Groceries Purchased: A Randomized Crossover Trial

Importance

Many factors are associated with food choice. Personalized interventions could help improve dietary intake by using individual purchasing preferences to promote healthier grocery purchases.

Objective

To test whether a healthy food incentive intervention using an algorithm incorporating customer preferences, purchase history, and baseline diet quality improves grocery purchase dietary quality and spending on healthy foods.

Design, Setting, and Participants

This was a 9-month randomized clinical crossover trial (AB-BA) with a 2- to 4-week washout period between 3-month intervention periods. Participants included 224 loyalty program members at an independent Rhode Island supermarket who completed baseline questionnaires and were randomized from July to September 2018 to group 1 (AB) or group 2 (BA). Data analysis was performed from September 2019 to May 2020.

Intervention

Participants received personalized weekly coupons with nutrition education during the intervention period (A) and occasional generic coupons with nutrition education during the control period (B). An automated study algorithm used customer data to allocate personalized healthy food incentives to participant loyalty cards. All participants received a 5% grocery discount.

Main Outcomes and Measures

Grocery Purchase Quality Index-2016 (GPQI-16) scores (range, 0-75, with higher scores denoting healthier purchases) and percentage spending on targeted foods were calculated from cumulative purchasing data. Participants in the top and bottom 1% of spending were excluded. Paired t tests examined between-group differences.

Results

The analytical sample included 209 participants (104 in group 1 and 105 in group 2), with a mean (SD) age of 55.4 (14.0) years. They were predominantly non-Hispanic White (193 of 206 participants [94.1%]) and female (187 of 207 participants [90.3%]). Of 161 participants with income data, 81 (50.3%) had annual household incomes greater than or equal to $100 000. Paired t tests showed that the intervention increased GPQI-16 scores (between-group difference, 1.06; 95% CI, 0.27-1.86; P = .01) and percentage spending on targeted foods (between-group difference, 1.38%; 95% CI, 0.08%-2.69%; P = .04). During the initial intervention period, group 1 (AB) and group 2 (BA) had similar mean (SD) GPQI-16 scores (41.2 [6.6] vs 41.0 [7.5]) and mean (SD) percentage spending on targeted healthy foods (32.0% [10.8%] vs 31.0% [10.5%]). During the crossover intervention period, group 2 had a higher mean (SD) GPQI-16 score than group 1 (42.9 [7.7] vs 41.0 [6.8]) and mean (SD) percentage spending on targeted foods (34.0% [12.1%] vs 32.0% [13.1%]).

Conclusions and Relevance

This pilot trial demonstrated preliminary evidence for the effectiveness of a novel personalized healthy food incentive algorithm to improve grocery purchase dietary quality.

Trial Registration

ClinicalTrials.gov Identifier: NCT03748056.