The impact of calorific screening thresholds and weight status when validating UK supermarket transaction records in...

Supermarket transaction data has the potential to provide a wide-scale understanding of population dietary behaviours, but its relationship with consumption is unclear as purchases are typically made at the household level while consumption occurs at the individual level. These behaviours may differ by weight status. This study assessed whether calorific screening thresholds improved the agreement between objective consumer purchase data and self-reported dietary intake for people living with (PLWOw/Ob_with) and without (PLWOw/Ob_without) overweight/obesity.

Participants (N = 642) from a retailer's loyalty card database completed a 170-item food frequency questionnaire, shared transaction records, height, weight, and household composition data for this study. Nutrients (energy, sugar, total fat, saturated fat, protein, and sodium) from supermarket transaction purchases were allocated to the study individual proportionally based on their household composition. Bland-Altman analyses were used to assess the agreement and bias between objective consumer purchase data and self-reported dietary intake globally, and between PLWOw/Ob_with and PLWOw/Ob_without, across a range of calorific screening thresholds.

No agreement was identified between objective consumer purchase data and self-reported dietary intake for any of the nutrients when the data were analysed without screening thresholds. However, agreement was identified when screening thresholds were employed at ≥1000 Kcal/day (energy, sugar, total fat and saturated fat) or ≥1500 Kcal/day (protein and sodium). PLWOw/Ob_with consumed greater energy (19%), sugar (36%), total fat (22%) and saturated fat (25%) than they were estimated to have purchased at the retailer. PLWOw/Ob_without only consumed more sugar (19%).

The application of screening thresholds based on estimated individual calories purchased may provide a valuable preprocessing step within the analysis of consumer purchase data, allowing agreement to be achieved for absolute nutrient values. Differences in bias between PLWOw/Ob_with and PLWOw/Ob_without show that insights into purchase and consumption patterns can be identified using consumer purchase data.