Non-linear isocaloric substitution analysis: an example using the UK Biobank prospective cohort study

Abstract

BackgroundDiet is one of the major modifiable lifestyle risk factors of mortality. When total energy intake is held constant, isocaloric substitution of energy sources can also affect mortality risk. The assumption in conventional methods that this exposure-outcome relationship is linear has been found to be untrue in many scenarios. This paper explores non-linear isocaloric substitution analysis using the UK Biobank data. MethodsThis is a UK Biobank prospective cohort study including 195 658 participants from England, Scotland, and Wales with complete information in at least one dietary questionnaire. Baseline data were collected between 2007 and 2010. Participants who reported non-typical diets were excluded. The exposure variables were intakes of carbohydrates (including sugar, starch, fibre), fat (including monounsaturated, polyunsaturated, and saturated fat) and protein, assessed using a 24 h dietary recall. Prospective all-cause mortality, the outcome variable, was derived from linkage to death records. We did a non-linear isocaloric substitution analysis using penalised cubic splines in Cox proportional hazard models. This analysis provides effect sizes conditional on the current intake, whereas conventional analysis provides an average over the whole range of intake. Analyses were adjusted for sociodemographic and other lifestyle factors. This study was performed under generic ethics approval obtained by UK Biobank from the National Health Service National Research Ethics Service. Written consent was obtained from all participants. FindingsThe majority (187 549) of the participants were White and 55·9% were female. Five of nine of the associations between macronutrient intake (as percentage of total energy, %E) and all-cause mortality were non-linear. Conventional isocaloric substitution showed no difference in mortality risk when replacing sugar with starch. However, the non-linear method revealed that replacing sugar with starch was associated with approximately 10% higher risk per 5%E when the current starch intake was greater than 30% of total energy, and with approximately 7% lower risk per 5%E when current intake was less than 25%. InterpretationWhen the underlying exposure-outcome association is non-linear, a relevant non-linear isocaloric substitution analysis technique should be used. The approach described here could provide more meaningful individualised dietary advice in line with precision public health. FundingNone.