To identify dietary patterns (DPs) characterized by a set of nutrients of concern and their association with incident type 2 diabetes (T2D).
A total of 120,343 participants from the U.K. Biobank study with at least two 24 h dietary assessments were studied. Reduced rank regression was used to derive DPs explaining variability in energy density, free sugars, saturated fat, and fiber intakes. We investigated prospective associations with T2D using Cox proportional hazard models.
Over 8.4 years of follow-up from the latest dietary assessment, 2,878 participants developed T2D. Two DPs were identified that jointly explained a total of 63% variation in four nutrients. DP1 was characterized by high intakes of chocolate and confectionery, butter, low-fiber bread, and sugars and preserves, and low intakes of fruits and vegetables. DP1 was linearly associated with T2D in multivariable models without BMI adjustment (per z score, hazard ratio [HR] 1.11 [95% CI 1.08–1.14]) and after BMI adjustment (HR 1.09 [95% CI 1.06–1.12]). DP2 was characterized by high intakes of sugar-sweetened beverages, fruit juice, table sugars and preserves, and low intakes of high-fat cheese and butter, but showed no clear association with T2D. There were significant interactions between both DPs and age, with increased risks among younger people in DP1 (HR 1.13 [95% CI 1.09–1.18]) and DP2 (HR 1.10 [95% CI 1.05–1.15]), as well as with DP1 and BMI, with increased risks among people with obesity (HR 1.11 [95% CI 1.07–1.16]).
A DP characterized by high intakes of chocolate and confectionery, butter, low-fiber bread, and added sugars, and low in fresh fruits and vegetables intake is associated with a higher incidence of T2D, particularly among younger people and those with obesity.
Introduction
The prevalence of diabetes, predominantly type 2 diabetes, has increased from 463 million in 2019 and is projected to reach 700 million by 2045 worldwide (1). Type 2 diabetes is associated with higher morbidity and mortality risks from other noncommunicable diseases such as cardiovascular disease (2). In addition, observations during the COVID-19 pandemic have revealed that people with diabetes have at least twice the risk of severe COVID-19 outcomes (3).
The risk of type 2 diabetes can be reduced by modifying unhealthy behaviors, such as a poor diet (4). According to the 2019 Global Burden of Diseases, 29.7% of disability-adjusted life-years of diabetes are attributable to dietary factors (5). In previous research focused on nutrients, associations were reported between poor diet quality and the incidence of type 2 diabetes and its comorbidities, particularly for diets high in saturated fat (6) and low in fiber (7). Evidence of the role of free sugars is less consistent (8); although high intakes have been associated with increased risk of type 2 diabetes, it is unclear whether the detrimental effects are caused by free sugars per se or by their contribution to excess energy intake. Increased free sugars intake is associated with reduced insulin sensitivity (9,10), but it has been reported that a high intake of free sugars without excess energy may not have any detrimental impact on health (11). There is also some evidence for positive associations between single foods and an increased risk of developing type 2 diabetes, such as higher intakes of unprocessed and processed meat (12,13), fruit juices, sugar-sweetened beverages (SSBs), refined grains, sweets, and desserts (8,14) and low intakes of fresh fruit and vegetables (15).
However, the impact of individual nutrients or foods on health outcomes may not adequately reflect the health effects of a whole diet in which foods are eaten in combination. Therefore, the observed associations could be driven by one or more nutrients, foods or other aspects of the diet that co-occur simultaneously. Reduced rank regression (RRR) is a data-driven approach that derives dietary patterns (DPs) by using a priori knowledge to select nutrients hypothesized to be on the causal pathway to examine how specific combinations of foods are associated with disease outcomes (16,17). This method allows specific investigations into certain diet-disease pathways (18), as opposed to the more exploratory approaches of principal components analysis or factor analysis (19). RRR has previously been used to study the effect of DPs on pathways of glucose homeostasis (20), inflammatory biomarkers (21), plasma circulating fatty acids (22), and blood lipids (23) as biomarkers of type 2 diabetes risk, although this evidence is not particularly strong, because of smaller sample sizes and low explained variance for the intermediate biomarkers. Here, we use detailed dietary data from the UK Biobank participants to identify DPs explaining high variability in known dietary risk factors, such as energy density, free sugars, saturated fat, and low fiber intakes and to assess the association between DPs and incident type 2 diabetes.
Research Design and Methods
Study Population
UK Biobank is a population-based cohort comprising 502,536 participants (aged 37 to 73 years) from 22 sites across England, Wales, and Scotland, with baseline measures collected between 2006 and 2010 and data linked to hospital and mortality records. Extensive information on sociodemographics, health behavior, and medical history was obtained via a touch-screen questionnaire and face-to-face interviews. Physical measures (e.g., height, weight) and blood and urine samples were collected by trained personnel conforming to a standardized protocol. All participants provided written informed consent (24,25).
Measurement of Dietary Intake
A web-based 24 h dietary instrument (Oxford WebQ) was used to collect detailed dietary data. The Oxford WebQ was validated against an interviewer-administered 24 h recall questionnaire (26) and collected information on the quantities of up to 206 widely consumed food items and 32 types of drinks consumed over the previous day (27). Participants with valid e-mail addresses were invited to complete a dietary questionnaire at baseline and were followed up to four times between April 2009 and June 2012 (cycle 1: February 2011 to April 2011; cycle 2: June 2011 to September 2011; cycle 3: October 2011 to December 2011; cycle 4: April 2012 to June 2012). Participants with at least two assessments were retained for analysis to better reflect usual intakes (28), and their mean dietary intake was calculated (17).
Using methods described previously (17,27), food intake data were aggregated into 50 main groups aligned to the U.K. National Diet and Nutrition Survey and according to the similarity of their nutritional composition and culinary use. Total energy and nutrient intake data were automatically estimated by multiplying the number of portions consumed by the set quantity of each food portion size and its nutrient composition according to the UK Nutrient Databank food composition tables (2012–2013 and 2013–2014) (29). Energy density, saturated fatty acid (SFA), free sugars, and fiber density were selected because of their significant roles in the development of obesity and type 2 diabetes and their high frequency of intake in daily life (4,9,30). Energy density (kJ/g) was calculated by dividing total food energy (in kilojoules) by total food weight (grams); all beverages were excluded because of their disproportionate influence on total energy density value (31). The percent total energies of SFA and free sugars (% total energy) were calculated by dividing daily energy from saturated fat or free sugars by total daily energy intake. Fiber density (g/MJ) was calculated by absolute intake of fiber (grams per day) divided by total daily energy intake (in megajoules), then multiplying by 1,000. Free sugars were defined as such on the basis of the U.K. Scientific Advisory Committee on Nutrition definitions; fiber was calculated using the Englyst method (29).
Individual estimated energy requirements (EERs) were calculated by the Schofield equation from the 1985 Food and Alcohol Organization/World Health Organization/United Nations University Expert Consultation Report on Human Energy Requirements. Dietary misreporting was calculated by using the ratio of energy intake (EI) to EER and its 95% CI (32). Dietary underreporters (EI:EER < 95% CI EI:EER) and overreporter (EI:EER >95% CI EI:EER) also were excluded.
Ascertainment of Outcomes
The outcomes for this study were primary or secondary diabetes events (excluding type 1 diabetes) ascertained from hospital episode statistics and death registry data linked to the UK Biobank. We defined incident diabetes events as a hospital admission or death with the following International Classification of Diseases, 10th Revision, codes in the hospital or death records: E11–E14. Hospital admission data were available up to 31 January 2021 in England and Scotland and 28 February 2018 in Wales. Deaths were ascertained via linkage to the death certificates and were available starting from baseline up to 31 January 2021 in England, Wales, and Scotland. Therefore, we censored diabetes analysis at the date of first incident diabetes or death or 31 January 2021, whichever occurred first.
Statistical Analyses
Identification of DPs
RRR was used to identify DPs that explained the maximum variation in a set of nutrient-response variables hypothesized to be on the causal pathway between food groups and incident type 2 diabetes. Energy density, SFA, free sugars, and fiber density were all used as response variables in the RRR model. The number of extracted patterns is in accordance with the number of response variables in the model. Respondents were scored for each DP (z score) representing the degree to which their dietary intake reflected each DP relative to that of other respondents. Increasing intakes of foods with positive factor loadings increase the DP z score; increasing intakes of foods with negative factor loadings decrease the DP z score. A larger factor loading value indicates that the food group makes a greater contribution to the DP. The associations between DPs and nutrient-response variables were evaluated by correlation coefficients (Supplementary Table 2). DPs that individually explained >20% of variation in response variables were retained for subsequent analyses.
Prospective Association of DPs With Incident Type 2 Diabetes
We used multivariable Cox proportional hazard models stratified for sex and regions (England, Scotland, and Wales) with age (years) as a timescale variable to obtain hazard ratios (HRs) with 95% CIs per unit increase in DP z scores, with sequential adjustment for race or ethnicity (White race, non-White race, missing data), Townsend Index of Deprivation (quintiles 1–5, with higher scores representing greater deprivation), education (higher degree [college or university degree, or professional qualifications], any U.K. school degree [advanced levels, advanced subsidiary levels, ordinary levels, General Certificate of Secondary Education, Certificate of Secondary Education, or equivalent], vocational qualifications [National Vocational Qualification, Higher National Diploma, Higher National Certificate, or equivalent], other [none of the listed qualifications], missing data), smoking status (never, current, previous, missing data), physical activity (low [<600 metabolic equivalent (MET)-minutes/week], moderate [≥600 and <3,000 MET-minutes/week], high [≥3,000 MET-minutes/week], missing data), log-transformed energy intake, family history of diabetes (yes, no), hypertension (yes, no), cardiovascular disease (yes, no), high cholesterol (yes, no), menopause in women (yes, no, not applicable [i.e., men]) and BMI group (underweight [<18.5 kg/m2], healthy weight [18.5 to <25 kg/m2], overweight [25 to <30 kg/m2], obese [≥30 kg/m2], missing data). Detailed information on variables are provided in Supplementary Table 1. The proportional hazards assumption was assessed by Schoenfeld residuals. Trend tests were performed by including the median score of each pattern quintile as a continuous variable in the models; the lowest quintile was used as the referent. Restricted cubic splines models with the same covariate specification were computed with 5 knots to examine nonlinear associations between DP z scores and incident type 2 diabetes.
Likelihood ratio tests were used to examine the heterogeneity of the associations of the DPs with risk of incident type 2 diabetes by age group (<60, ≥60 years), sex (female, male), smoking status (never, previous, current), physical activity (low, moderate, high), BMI group (underweight/healthy [<25], overweight [25 to <30], obesity [≥30]).
Sensitivity and Exploratory Analyses
To investigate potential bias in relation to random variation in individual intakes, we repeated the RRR analysis to derive DPs among participants providing at least three (n = 75,003), at least four (n = 34,644), or five (n = 5,504) online dietary assessments, which may better reflect usual intakes. A second sensitivity analysis excluded participants who had a diabetes event within 2 years after completing their last 24 h online dietary assessment.
RRR analysis was performed using SAS statistical software (version 9.4; SAS Institute). The rest of the analyses were performed in Stata/MP 14.0.
Results
Of the 502,536 UK Biobank participants recruited at baseline, participants were excluded for the following reasons: a diabetes event occurred before baseline (n = 26,845); pregnancy (n = 375); did not complete any validated dietary assessment (n = 273,434) or completed only one 24 h online dietary assessment (n = 79,995); diabetes occurred before the latest dietary questionnaire (n = 172); missing nutrient data (n = 21); implausible energy intake (overreporters: n = 130; underreporters: n = 996; or missing BMI data (n = 225) (Supplementary Fig. 1). In this analysis, 120,343 participants who provided complete data using the 24 h online dietary assessments on at least two occasions were included.
Four DPs were derived (Supplementary Table 2). DP1 (43%) and DP2 (20%) (Fig. 1) jointly explained 63% of variation in all response variables; DP3 and DP4 only contributed to 10% and 4%, respectively, which were retained in subsequent analyses. DP1 was characterized by high consumption of chocolate and confectionery, butter and other animal-fat spreads, and low-fiber bread, and low consumption of fresh fruit, vegetables, and high-fiber breakfast cereals (Fig. 1). DP2 was characterized by high consumption of SSBs, fruit juice, table sugars, and preserves, and low consumption of high-fat cheese and butter and other animal-fat spreads (Fig. 1). In sensitivity analyses, we derived new DPs from RRR using subsamples of people who provided three, four, or five dietary assessments or excluding those who experienced the event within 2 years of completing their latest assessment (Supplementary Figs. 2–5), yielding consistent results.
During 1,350,644 person-years of follow-up (11.2 years of median follow-up from baseline, 8.4 years of follow-up after the latest dietary assessment), 2,878 participants developed incident type 2 diabetes (including 83 fatal cases). Baseline characteristics including demographics, socioeconomic status, behavior risk factors, health history and conditions, dietary intake, main food groups across quintiles of DP scores are given in Table 1. Higher quintiles of DP1 z scores included people with the following characteristics: a higher proportion of men, younger age, higher Townsend index, lower attained education level, current smokers, less physically active, and higher prevalence of obesity or hypertension. Higher quintiles of DP2 z scores included a higher proportion of younger participants, lower attained education level, never smokers, more physically active, higher prevalence of hypertension, cardiovascular disease, and high cholesterol.
Baseline characteristics of the study population and average dietary intake
. | . | DP1 quintiles . | DP2 quintiles . | ||||||
---|---|---|---|---|---|---|---|---|---|
Total (N = 120,343) . | 1 (n = 24,069) . | 3 (n = 24,068) . | 5 (n = 24,068) . | P value* . | 1 (n = 24,069) . | 3 (n = 24,068) . | 5 (n = 24,068) . | P value* . | |
Demographics | |||||||||
Female sex (%) | 56.5 | 70.5 | 59.0 | 35.9 | <0.001 | 56.2 | 60.8 | 47.5 | <0.001 |
Age (years)† | 56.1 (7.8) | 57.4 (7.3) | 56.3 (7.7) | 54.1 (8.1) | <0.001 | 56.0 (7.7) | 56.3 (7.8) | 55.5 (8.1) | <0.001 |
White race (%) | 96.8 | 96.6 | 96.9 | 96.7 | 0.089 | 97.7 | 97.2 | 94.7 | <0.001 |
Socioeconomic status: Townsend index quintile | <0.001 | <0.001 | |||||||
1 | 20.0 | 20.3 | 20.6 | 18.4 | 19.3 | 20.5 | 19.2 | ||
3 | 20.0 | 19.8 | 20.7 | 19.9 | 19.8 | 20.4 | 19.8 | ||
5 | 19.9 | 19.2 | 18.6 | 22.5 | 20.9 | 18.6 | 21.4 | ||
Missing data | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | ||
Education group (%) | |||||||||
Vocational qualification (NVQ, HND, HNC, or equivalent) | 12.4 | 10.4 | 12.4 | 16.1 | 11.7 | 11.9 | 14.0 | ||
Any school degree (A level, AS level, O level, GCSE, CSE, or equivalent) | 29.2 | 27.0 | 28.9 | 31.6 | 28.8 | 29.9 | 28.8 | ||
Higher degree (college, university, or professional degree or qualification) | 51.8 | 56.5 | 52.5 | 44.6 | 53.6 | 51.6 | 50.0 | ||
None of the preceding groups | 6.3 | 5.8 | 5.9 | 7.3 | <0.001 | 5.6 | 6.4 | 6.7 | <0.001 |
Missing data | 0.3 | 0.3 | 0.3 | 0.3 | 0.2 | 0.3 | 0.3 | ||
Behavior risk factors | |||||||||
Smoking status (%) | <0.001 | <0.001 | |||||||
Never | 57.6 | 60.3 | 59.3 | 52.2 | 54.3 | 58.3 | 59.4 | ||
Previous | 35.3 | 35.7 | 35.0 | 34.9 | 37.0 | 35.5 | 33.0 | ||
Current | 6.9 | 3.8 | 5.5 | 12.8 | 8.5 | 6.0 | 7.3 | ||
Missing | 0.2 | 0.2 | 0.2 | 0.2 | 0.2 | 0.2 | 0.2 | ||
Physical activity (IPAQ) (%) | <0.001 | <0.001 | |||||||
Low | 19.1 | 14.0 | 19.4 | 23.8 | 19.4 | 19.6 | 18.1 | ||
Moderate | 45.0 | 43.0 | 45.9 | 44.2 | 45.8 | 45.4 | 43.4 | ||
High | 35.8 | 43.0 | 34.6 | 32.0 | 34.8 | 34.9 | 38.5 | ||
Missing data | <0.1 | <0.1 | <0.1 | <0.1 | <0.1 | <0.1 | <0.1 | ||
Health history and conditions | |||||||||
Family history of diabetes (%) | 15.7 | 15.4 | 15.7 | 16.1 | 0.32 | 15.5 | 15.9 | 16.0 | 0.45 |
Hypertension (%) | 23.3 | 22.8 | 23.1 | 24.3 | <0.001 | 22.3 | 23.1 | 24.6 | <0.001 |
Cardiovascular disease (%) | 5.2 | 5.3 | 5.3 | 5.2 | 0.86 | 4.4 | 5.0 | 6.1 | <0.001 |
High cholesterol (%) | 13.5 | 13.3 | 13.6 | 13.3 | 0.76 | 11.4 | 13.7 | 15.2 | <0.001 |
Menopause in women (%) | 23.2 | 24.1 | 24.3 | 19.4 | <0.001 | 22.8 | 24.7 | 20.8 | <0.001 |
BMI (kg/m2) | 26.5(4.4) | 26.0(4.4) | 26.4(4.3) | 27.3(4.6) | <0.001 | 26.7(4.7) | 26.5(4.4) | 26.5(4.4) | <0.001 |
BMI group (%) | <0.001 | <0.001 | |||||||
Underweight (<18.5) | 0.6 | 0.9 | 0.5 | 0.5 | 0.7 | 0.5 | 0.5 | ||
Healthy weight (18.5 to <25) | 40.1 | 45.4 | 41.3 | 32.1 | 39.3 | 40.6 | 39.1 | ||
Overweight (25 to <30) | 41.1 | 38.1 | 41.2 | 44.0 | 39.9 | 41.0 | 42.7 | ||
Obese (≥30) | 18.2 | 15.5 | 17.0 | 23.3 | 20.0 | 17.8 | 17.6 | ||
Missing data | <0.1 | 0.1 | 0.0 | <0.1 | <0.1 | <0.1 | <0.1 | ||
Dietary intake | |||||||||
Energy intake (MJ/day) | 8.7 (2.2) | 8.2 (2.1) | 8.3 (2.0) | 10.0 (2.5) | <0.001 | 9.3 (2.4) | 8.3 (2.1) | 9.0 (2.3) | <0.001 |
Dietary energy density (kJ/g) | 6.5 (1.6) | 4.8 (0.8) | 6.4 (0.8) | 8.4 (1.4) | <0.001 | 7.1 (1.6) | 6.3 (1.5) | 6.4 (1.6) | <0.001 |
Saturated fat (% E) | 11.7 (3.2) | 9.7 (2.6) | 11.8 (2.8) | 13.4 (3.3) | <0.001 | 14.4 (2.9) | 11.3 (2.7) | 10.0 (2.8) | <0.001 |
Free sugars (% E) | 11.5 (5.2) | 8.9 (4.1) | 11.4 (4.5) | 14.7 (6.1) | <0.001 | 7.7 (3.3) | 10.6 (3.6) | 17.4 (5.3) | <0.001 |
Fiber (g/day) | 18.1 (6.2) | 23.3 (6.7) | 17.1 (5.0) | 15.1 (5.3) | <0.001 | 18.2 (6.2) | 17.8 (6.0) | 18.5 (6.8) | <0.001 |
Fiber density (g/MJ) | 2.1 (0.6) | 2.9 (0.6) | 2.1 (0.4) | 1.5 (0.4) | <0.001 | 2.0 (0.6) | 2.2 (0.7) | 2.1 (0.7) | <0.001 |
Main food groups (g/day) | |||||||||
Chocolate and confectionery | 11.9 (21.4) | 6.0 (12.4) | 9.7 (16.0) | 23.2 (33.5) | <0.001 | 8.3 (16.0) | 10.4 (17.9) | 18.9 (30.8) | <0.001 |
Butter and other normal animal-fat spreads | 5.1 (8.7) | 1.9 (4.8) | 4.3 (7.2) | 10.5 (12.4) | <0.001 | 12.1 (12.6) | 3.5 (6.4) | 2.0 (5.0) | <0.001 |
Added sugars and preserves | 8.9 (13.0) | 5.4 (8.4) | 7.6 (10.2) | 15.7 (19.4) | <0.001 | 6.0 (9.4) | 7.4 (10.4) | 15.1 (18.6) | <0.001 |
SSBs and other sugary drinks | 87.3 (166.1) | 45.1 (107.4) | 74.1 (134.8) | 163.1 (245.5) | <0.001 | 30.8 (77.6) | 58.4 (105.1) | 219.5 (267.7) | <0.001 |
Fruit juice | 110.0 (137.6) | 104.8 (139.9) | 114.0 (135.4) | 108.1 (144.3) | <0.001 | 52.7 (87.6) | 92.7 (108.0) | 205.7 (188.9) | <0.001 |
Low-fiber bread | 28.1 (42.3) | 9.2 (22.1) | 22.2 (32.2) | 61.0 (58.7) | <0.001 | 36.6 (50.4) | 25.3 (38.2) | 25.9 (41.0) | <0.001 |
Processed meat | 18.2 (28.0) | 10.3 (19.7) | 16.8 (24.5) | 29.4 (37.5) | <0.001 | 25.9 (35.6) | 16.8 (25.1) | 14.0 (24.3) | <0.001 |
High-fat cheese | 14.8 (18.1) | 10.9 (15.4) | 14.4 (16.7) | 19.4 (22.3) | <0.001 | 28.0 (24.3) | 12.1 (14.1) | 7.9 (12.4) | <0.001 |
High-fiber bread | 49.0 (47.4) | 53.8 (50.0) | 51.2 (45.5) | 39.3 (47.1) | <0.001 | 55.8 (52.6) | 47.2 (44.2) | 46.5 (47.7) | <0.001 |
Vegetables | 191.0 (146.0) | 325.6 (185.2) | 170.7 (105.8) | 107.1 (92.6) | <0.001 | 196.0 (147.7) | 191.3 (140.6) | 184.2 (154.9) | <0.001 |
Fresh fruit | 197.0 (153.3) | 342.7 (181.7) | 177.9 (111.6) | 97.4 (97.4) | <0.001 | 169.9 (142.6) | 200.8 (148.5) | 216.3 (170.0) | <0.001 |
. | . | DP1 quintiles . | DP2 quintiles . | ||||||
---|---|---|---|---|---|---|---|---|---|
Total (N = 120,343) . | 1 (n = 24,069) . | 3 (n = 24,068) . | 5 (n = 24,068) . | P value* . | 1 (n = 24,069) . | 3 (n = 24,068) . | 5 (n = 24,068) . | P value* . | |
Demographics | |||||||||
Female sex (%) | 56.5 | 70.5 | 59.0 | 35.9 | <0.001 | 56.2 | 60.8 | 47.5 | <0.001 |
Age (years)† | 56.1 (7.8) | 57.4 (7.3) | 56.3 (7.7) | 54.1 (8.1) | <0.001 | 56.0 (7.7) | 56.3 (7.8) | 55.5 (8.1) | <0.001 |
White race (%) | 96.8 | 96.6 | 96.9 | 96.7 | 0.089 | 97.7 | 97.2 | 94.7 | <0.001 |
Socioeconomic status: Townsend index quintile | <0.001 | <0.001 | |||||||
1 | 20.0 | 20.3 | 20.6 | 18.4 | 19.3 | 20.5 | 19.2 | ||
3 | 20.0 | 19.8 | 20.7 | 19.9 | 19.8 | 20.4 | 19.8 | ||
5 | 19.9 | 19.2 | 18.6 | 22.5 | 20.9 | 18.6 | 21.4 | ||
Missing data | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | ||
Education group (%) | |||||||||
Vocational qualification (NVQ, HND, HNC, or equivalent) | 12.4 | 10.4 | 12.4 | 16.1 | 11.7 | 11.9 | 14.0 | ||
Any school degree (A level, AS level, O level, GCSE, CSE, or equivalent) | 29.2 | 27.0 | 28.9 | 31.6 | 28.8 | 29.9 | 28.8 | ||
Higher degree (college, university, or professional degree or qualification) | 51.8 | 56.5 | 52.5 | 44.6 | 53.6 | 51.6 | 50.0 | ||
None of the preceding groups | 6.3 | 5.8 | 5.9 | 7.3 | <0.001 | 5.6 | 6.4 | 6.7 | <0.001 |
Missing data | 0.3 | 0.3 | 0.3 | 0.3 | 0.2 | 0.3 | 0.3 | ||
Behavior risk factors | |||||||||
Smoking status (%) | <0.001 | <0.001 | |||||||
Never | 57.6 | 60.3 | 59.3 | 52.2 | 54.3 | 58.3 | 59.4 | ||
Previous | 35.3 | 35.7 | 35.0 | 34.9 | 37.0 | 35.5 | 33.0 | ||
Current | 6.9 | 3.8 | 5.5 | 12.8 | 8.5 | 6.0 | 7.3 | ||
Missing | 0.2 | 0.2 | 0.2 | 0.2 | 0.2 | 0.2 | 0.2 | ||
Physical activity (IPAQ) (%) | <0.001 | <0.001 | |||||||
Low | 19.1 | 14.0 | 19.4 | 23.8 | 19.4 | 19.6 | 18.1 | ||
Moderate | 45.0 | 43.0 | 45.9 | 44.2 | 45.8 | 45.4 | 43.4 | ||
High | 35.8 | 43.0 | 34.6 | 32.0 | 34.8 | 34.9 | 38.5 | ||
Missing data | <0.1 | <0.1 | <0.1 | <0.1 | <0.1 | <0.1 | <0.1 | ||
Health history and conditions | |||||||||
Family history of diabetes (%) | 15.7 | 15.4 | 15.7 | 16.1 | 0.32 | 15.5 | 15.9 | 16.0 | 0.45 |
Hypertension (%) | 23.3 | 22.8 | 23.1 | 24.3 | <0.001 | 22.3 | 23.1 | 24.6 | <0.001 |
Cardiovascular disease (%) | 5.2 | 5.3 | 5.3 | 5.2 | 0.86 | 4.4 | 5.0 | 6.1 | <0.001 |
High cholesterol (%) | 13.5 | 13.3 | 13.6 | 13.3 | 0.76 | 11.4 | 13.7 | 15.2 | <0.001 |
Menopause in women (%) | 23.2 | 24.1 | 24.3 | 19.4 | <0.001 | 22.8 | 24.7 | 20.8 | <0.001 |
BMI (kg/m2) | 26.5(4.4) | 26.0(4.4) | 26.4(4.3) | 27.3(4.6) | <0.001 | 26.7(4.7) | 26.5(4.4) | 26.5(4.4) | <0.001 |
BMI group (%) | <0.001 | <0.001 | |||||||
Underweight (<18.5) | 0.6 | 0.9 | 0.5 | 0.5 | 0.7 | 0.5 | 0.5 | ||
Healthy weight (18.5 to <25) | 40.1 | 45.4 | 41.3 | 32.1 | 39.3 | 40.6 | 39.1 | ||
Overweight (25 to <30) | 41.1 | 38.1 | 41.2 | 44.0 | 39.9 | 41.0 | 42.7 | ||
Obese (≥30) | 18.2 | 15.5 | 17.0 | 23.3 | 20.0 | 17.8 | 17.6 | ||
Missing data | <0.1 | 0.1 | 0.0 | <0.1 | <0.1 | <0.1 | <0.1 | ||
Dietary intake | |||||||||
Energy intake (MJ/day) | 8.7 (2.2) | 8.2 (2.1) | 8.3 (2.0) | 10.0 (2.5) | <0.001 | 9.3 (2.4) | 8.3 (2.1) | 9.0 (2.3) | <0.001 |
Dietary energy density (kJ/g) | 6.5 (1.6) | 4.8 (0.8) | 6.4 (0.8) | 8.4 (1.4) | <0.001 | 7.1 (1.6) | 6.3 (1.5) | 6.4 (1.6) | <0.001 |
Saturated fat (% E) | 11.7 (3.2) | 9.7 (2.6) | 11.8 (2.8) | 13.4 (3.3) | <0.001 | 14.4 (2.9) | 11.3 (2.7) | 10.0 (2.8) | <0.001 |
Free sugars (% E) | 11.5 (5.2) | 8.9 (4.1) | 11.4 (4.5) | 14.7 (6.1) | <0.001 | 7.7 (3.3) | 10.6 (3.6) | 17.4 (5.3) | <0.001 |
Fiber (g/day) | 18.1 (6.2) | 23.3 (6.7) | 17.1 (5.0) | 15.1 (5.3) | <0.001 | 18.2 (6.2) | 17.8 (6.0) | 18.5 (6.8) | <0.001 |
Fiber density (g/MJ) | 2.1 (0.6) | 2.9 (0.6) | 2.1 (0.4) | 1.5 (0.4) | <0.001 | 2.0 (0.6) | 2.2 (0.7) | 2.1 (0.7) | <0.001 |
Main food groups (g/day) | |||||||||
Chocolate and confectionery | 11.9 (21.4) | 6.0 (12.4) | 9.7 (16.0) | 23.2 (33.5) | <0.001 | 8.3 (16.0) | 10.4 (17.9) | 18.9 (30.8) | <0.001 |
Butter and other normal animal-fat spreads | 5.1 (8.7) | 1.9 (4.8) | 4.3 (7.2) | 10.5 (12.4) | <0.001 | 12.1 (12.6) | 3.5 (6.4) | 2.0 (5.0) | <0.001 |
Added sugars and preserves | 8.9 (13.0) | 5.4 (8.4) | 7.6 (10.2) | 15.7 (19.4) | <0.001 | 6.0 (9.4) | 7.4 (10.4) | 15.1 (18.6) | <0.001 |
SSBs and other sugary drinks | 87.3 (166.1) | 45.1 (107.4) | 74.1 (134.8) | 163.1 (245.5) | <0.001 | 30.8 (77.6) | 58.4 (105.1) | 219.5 (267.7) | <0.001 |
Fruit juice | 110.0 (137.6) | 104.8 (139.9) | 114.0 (135.4) | 108.1 (144.3) | <0.001 | 52.7 (87.6) | 92.7 (108.0) | 205.7 (188.9) | <0.001 |
Low-fiber bread | 28.1 (42.3) | 9.2 (22.1) | 22.2 (32.2) | 61.0 (58.7) | <0.001 | 36.6 (50.4) | 25.3 (38.2) | 25.9 (41.0) | <0.001 |
Processed meat | 18.2 (28.0) | 10.3 (19.7) | 16.8 (24.5) | 29.4 (37.5) | <0.001 | 25.9 (35.6) | 16.8 (25.1) | 14.0 (24.3) | <0.001 |
High-fat cheese | 14.8 (18.1) | 10.9 (15.4) | 14.4 (16.7) | 19.4 (22.3) | <0.001 | 28.0 (24.3) | 12.1 (14.1) | 7.9 (12.4) | <0.001 |
High-fiber bread | 49.0 (47.4) | 53.8 (50.0) | 51.2 (45.5) | 39.3 (47.1) | <0.001 | 55.8 (52.6) | 47.2 (44.2) | 46.5 (47.7) | <0.001 |
Vegetables | 191.0 (146.0) | 325.6 (185.2) | 170.7 (105.8) | 107.1 (92.6) | <0.001 | 196.0 (147.7) | 191.3 (140.6) | 184.2 (154.9) | <0.001 |
Fresh fruit | 197.0 (153.3) | 342.7 (181.7) | 177.9 (111.6) | 97.4 (97.4) | <0.001 | 169.9 (142.6) | 200.8 (148.5) | 216.3 (170.0) | <0.001 |
A level, advanced level; AS, advanced subsidiary; CSE, Certificate of Secondary Education; E, energy; GCSE, General Certificate of Secondary Education; HNC, Higher National Certificate; HND, Higher National Diploma; IPAQ, International Physical Activity Questionnaire; NVQ, National Vocational Qualification; O, ordinary.
ANOVA or χ2 test where appropriate.
Intakes of chocolate and confectionery, butter and other animal-fat spreads, table sugars and preserves, SSBs and other sugary drinks, low-fiber bread, processed meat, and high-fat cheese were higher across quintiles of DP1 z scores, along with lower intakes of high-fiber bread, vegetables, and fresh fruit. In DP2, intakes of chocolate and confectionery, added sugars and preserves, SSBs and other sugary drinks, fruit juice, and fresh fruit were higher across quintiles of DP2 z scores, with lower intakes of low-fiber bread, processed meat, high-fat cheese, and vegetables.
There was positive association between adherence to DP1 and incident type 2 diabetes after sequential adjustment for demographics, sociodemographics, behavior risk factors, and health history and conditions (Ptrend < 0.001), and remained statistically significant after adjustment for BMI (Ptrend < 0.001; Table 2, Supplementary Fig. 6). There was no evidence of an association between adherence to DP2 and the risk of type 2 diabetes. The exclusion of participants who had a diabetes event within 2 years of completing their last 24 h online dietary assessment showed that the associations between DPs and the risk of diabetes events were largely unchanged (Supplementary Table 4).
Sequentially adjusted HRs of total incident diabetes associated with each z score increase in DP
. | No. of events . | Adjusted for sex, age, and region . | Plus sociodemographics* . | Plus behavior risk factors† . | Plus health history and conditions‡ . | Plus BMI group . |
---|---|---|---|---|---|---|
DP1 | ||||||
Total | 2,878 | 1.16 (1.13–1.19) | 1.14 (1.11–1.18) | 1.11 (1.08–1.14) | 1.11 (1.08–1.14) | 1.09 (1.06–1.12) |
Q1 | 452 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
Q2 | 513 | 1.15 (1.05–1.25) | 1.15 (1.05–1.26) | 1.11 (1.02–1.21) | 1.10 (1.01–1.20) | 1.13 (1.04–1.23) |
Q3 | 549 | 1.23 (1.13–1.33) | 1.22 (1.12–1.33) | 1.16 (1.06–1.26) | 1.15 (1.06–1.25) | 1.19 (1.10–1.30) |
Q4 | 621 | 1.41 (1.31–1.53) | 1.39 (1.29–1.51) | 1.28 (1.18–1.38) | 1.27 (1.17–1.37) | 1.25 (1.16–1.35) |
Q5 | 743 | 1.73 (1.60–1.86) | 1.65 (1.53–1.78) | 1.46 (1.35–1.59) | 1.45 (1.34–1.57) | 1.38 (1.27–1.49) |
Ptrend | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | |
DP2 | ||||||
Total | 2,878 | 1.05 (1.01–1.08) | 1.03 (0.99–1.06) | 1.03 (1.00–1.07) | 1.02 (0.98–1.05) | 1.03 (0.99–1.06) |
Q1 | 610 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
Q2 | 545 | 0.98 (0.90–1.06) | 0.97 (0.89–1.06) | 0.96 (0.89–1.05) | 0.96 (0.88–1.04) | 0.97 (0.89–1.05) |
Q3 | 567 | 1.02 (0.94–1.11) | 1.02 (0.94–1.10) | 1.01 (0.93–1.10) | 0.99 (0.91–1.07) | 1.01 (0.93–1.10) |
Q4 | 524 | 0.93 (0.85–1.01) | 0.91 (0.83–0.99) | 0.91 (0.84–1.00) | 0.88 (0.81–0.96) | 0.90 (0.82–0.98) |
Q5 | 632 | 1.09 (1.01–1.18) | 1.04 (0.96–1.13) | 1.06 (0.98–1.14) | 1.01 (0.93–1.09) | 1.04 (0.96–1.12) |
Ptrend | 0.234 | 0.735 | 0.476 | 0.840 | 0.818 |
. | No. of events . | Adjusted for sex, age, and region . | Plus sociodemographics* . | Plus behavior risk factors† . | Plus health history and conditions‡ . | Plus BMI group . |
---|---|---|---|---|---|---|
DP1 | ||||||
Total | 2,878 | 1.16 (1.13–1.19) | 1.14 (1.11–1.18) | 1.11 (1.08–1.14) | 1.11 (1.08–1.14) | 1.09 (1.06–1.12) |
Q1 | 452 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
Q2 | 513 | 1.15 (1.05–1.25) | 1.15 (1.05–1.26) | 1.11 (1.02–1.21) | 1.10 (1.01–1.20) | 1.13 (1.04–1.23) |
Q3 | 549 | 1.23 (1.13–1.33) | 1.22 (1.12–1.33) | 1.16 (1.06–1.26) | 1.15 (1.06–1.25) | 1.19 (1.10–1.30) |
Q4 | 621 | 1.41 (1.31–1.53) | 1.39 (1.29–1.51) | 1.28 (1.18–1.38) | 1.27 (1.17–1.37) | 1.25 (1.16–1.35) |
Q5 | 743 | 1.73 (1.60–1.86) | 1.65 (1.53–1.78) | 1.46 (1.35–1.59) | 1.45 (1.34–1.57) | 1.38 (1.27–1.49) |
Ptrend | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | |
DP2 | ||||||
Total | 2,878 | 1.05 (1.01–1.08) | 1.03 (0.99–1.06) | 1.03 (1.00–1.07) | 1.02 (0.98–1.05) | 1.03 (0.99–1.06) |
Q1 | 610 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
Q2 | 545 | 0.98 (0.90–1.06) | 0.97 (0.89–1.06) | 0.96 (0.89–1.05) | 0.96 (0.88–1.04) | 0.97 (0.89–1.05) |
Q3 | 567 | 1.02 (0.94–1.11) | 1.02 (0.94–1.10) | 1.01 (0.93–1.10) | 0.99 (0.91–1.07) | 1.01 (0.93–1.10) |
Q4 | 524 | 0.93 (0.85–1.01) | 0.91 (0.83–0.99) | 0.91 (0.84–1.00) | 0.88 (0.81–0.96) | 0.90 (0.82–0.98) |
Q5 | 632 | 1.09 (1.01–1.18) | 1.04 (0.96–1.13) | 1.06 (0.98–1.14) | 1.01 (0.93–1.09) | 1.04 (0.96–1.12) |
Ptrend | 0.234 | 0.735 | 0.476 | 0.840 | 0.818 |
Adjusted HRs and 95% CIs of total DP scores obtained using Cox proportional hazard regression. Age at risk adjusted for by using age during the study as the underlying timescale for Cox regression. All the models were stratified by sex and regions (England, Scotland, and Wales). Q, quintile.
Sociodemographic characteristics: Race or ethnicity (White, others, missing), Townsend Index of Deprivation (Q1–Q5, with higher scores representing greater deprivation), education (higher degree [college or university degree, or professional qualifications], any school degree [advanced levels, advanced subsidiary levels, ordinary levels, General Certificate of Secondary Education, Certificate of Secondary Education], vocational qualifications [National Vocational Qualification, Higher National Diploma, Higher National Certificate], other [none of the listed qualifications], missing data).
Behavior risk factors: smoking status (never, current, previous, missing data), physical activity level (low [<600 metabolic equivalent (MET)-minutes/week], moderate [≥600 and <3,000 MET-minutes/week], high [≥3,000 MET-minutes/week], missing data), log-transformed energy intake.
Health history and conditions: family history of diabetes, menopause in women, hypertension, cardiovascular disease, high cholesterol. Full model was stratified by sex and regions (England, Scotland, and Wales) and adjusted for demographics, socioeconomic status, behavior risk factors, and health history and conditions with BMI group included.
There was a significant interaction between both DPs and age, with significantly higher risks among younger people (<60 years; DP1: HR 1.13 [95% CI 1.09–1.18], DP2: HR 1.10 [95% CI 1.05–1.15]; Fig. 2). A significant interaction was also found for DP1 and BMI, with higher risks among people with obesity at baseline (HR 1.11; 95% CI 1.07–1.16). There was no evidence of any effect modification by sex, smoking status, or level of physical activity.
The association between DPs and risk of diabetes by age, sex, smoking status, and BMI group (N = 120,343).
The association between DPs and risk of diabetes by age, sex, smoking status, and BMI group (N = 120,343).
Conclusions
In this large cohort of British adults, we identified two main DPs. The main DP (DP1) explained almost half of the variance in nutrient intakes (43%) and was characterized by high intakes of chocolate and confectionery, butter and other animal-fat spreads, and low-fiber bread, and low intakes of fresh fruit, vegetables, and high-fiber breakfast cereals, contributing to excess energy intake, SFAs, free sugars, and low intake of fiber. There was a positive linear association between this DP and type 2 diabetes events, with stronger associations found among people aged <60 years and those who had obesity at baseline. A second DP (DP2), explaining much less of the variance in nutrient intakes (20%) and characterized by a very high intake of free sugars from higher intakes of SSBs, fruit juice, table sugars, and preserves, together with lower intakes of high-fat cheese and butter and other animal-fat spreads, was not associated with the incidence of type 2 diabetes.
This study contributes new observational evidence to help understand the associations of a whole diet with incident type 2 diabetes. In practice, nutrients are not consumed separately and they may exert effects on health in a synergistic way. Analyses of DPs through RRR, which use nutrients of concern as response variables in the causal pathway, may help detect stronger associations due to the cumulative effects of different dietary components. The main DP derived in this study is characterized by food groups (and underlying nutrients) for which associations with type 2 diabetes are consistent with findings from previous studies (19,33). For example, excessive consumption of chocolate and confectionery (34), red and processed meat (35), added sugars and preserves (36), and low consumption of vegetables, fresh fruit (37), and high-fiber bread (38) have been associated with adverse metabolic effects and higher risk of diabetes. In a recent systematic literature review and meta-analysis of 16 cohorts, authors found three patterns derived by RRR showing refined grains, SSBs, and processed meat were all significantly associated with diabetes risk independent of the biomarkers used as response variables and study populations (39). A recent clinical trial in overweight men provided evidence that compared with a diet enriched in free sugars, a diet enriched in SFAs led to increased intrahepatic triacylglycerol levels and postprandial glycemia, but a diet enriched in free sugars did not influence intrahepatic triacylglycerol levels and led to only minor metabolic changes (40).
The second DP was characterized by foods that contributed high intake of free sugars (e.g., SSBs, fruit juice) but in the context of low SFA levels and energy density and adequate amounts of fiber due to high intake of fruits and vegetables. People in the highest quintile of this pattern consumed, on average, 17.4% of dietary energy from free sugars (more than three times the U.K. dietary guideline) but with only 10% from SFAs (the recommended maximum level) with no significant increase in energy intake. Here there was no evidence of an overall association with incidence of diabetes. Accordingly, this analysis, which controls for differences in BMI, supports the hypothesis that free sugars increase the risk for type 2 diabetes because of their contribution to excess energy intake rather than having an direct, independent effect (8,41). In short- or medium-term isoenergetic intervention trials, free sugars showed no effect on body weight or blood pressure and the effects on blood lipid profile were not conclusive (11). Our findings are also consistent with the evidence that diets low in saturated fat reduce the risk of type 2 diabetes (42).
We found that the positive associations between DP1 and risk of type 2 diabetes were slightly attenuated but remained statistically significant after adjusting for BMI, suggesting that BMI may only partly explain the variability in type 2 diabetes risk. However, in subsequent stratified analyses, DP1 was much more strongly associated with type 2 diabetes among people with obesity at baseline. Therefore, these results suggest that residual confounding or mediation by adiposity cannot be ruled out and might account for some of the observed associations.
Adherence to both DP1 and DP2 was a particular risk factor for the development of type 2 diabetes in younger people (<60 years), suggesting that when all other things are held constant, a poor diet could increase the risk of type 2 diabetes in younger people. Also, there are also age-related differences in healthy food consumption such as fruits and vegetable, with older adults consuming more fruit and vegetables than do younger adults—a pattern that has been observed in several U.K. national surveys. In the National Diet and Nutrition Survey (2016–2017 to 2018–2019), adults (aged 19–64 years) consumed, on average, 4.3 portions of fruit and vegetables per day and older adults (≥65 years) consumed 4.5 portions per day (43). Reports in the wider literature also have suggested that older people tend to consume more fruits and vegetables and other healthy foods (44), which may be the case among the UK Biobank participants. In our middle-aged population, 55% of type 2 diabetes events were observed in people aged ≥60 years. Our study did not show differential associations by sex, although a previous study using the RRR approach found that a high-fat, high glycemic index, low-fiber DP was associated with increased type 2 diabetes risk in middle-aged British women but not men (45), though the mechanistic basis for such differences is unclear. Also, differential associations across physical activity groups were not observed. DP1 was consistently associated with higher risk of type 2 diabetes regardless of the amount of physical activity reported; however, this association was slightly less strong in people with a high level of physical activity.
Strengths of this study include a large sample size and the empirical hypothesis-based DP analysis that was used to identify DPs in this cohort of middle-aged British adults and their association with type 2 diabetes. Some limitations of this study should be acknowledged. First, DPs were identified and validated from at least two 24 h, online, self-reported dietary assessments, which may be affected by recall bias or misreporting (e.g., people at higher risk of diabetes or aware of their diabetes biomarkers may report foods differentially), and DPs were derived using data from people with higher willingness to report their diet (which may be affected by selection bias). However, our results were reassured by the sensitivity analyses, which used at least 2, 3, 4, or 5 repeated assessments to better capture usual intakes, showing consistent results. Second, some uncommon foods and other important covariates were not captured by the questionnaires, hence some residual confounding may remain. Third, although the identified DPs are specific to the U.K. population, assuming that the relationships between dietary factors and health outcomes would be similar among individuals, outcome-dependent methods, such as RRR, are expected to be relatively reproducible across different study populations (46,47). Fourth, incident diabetes cases were ascertained based on hospital records, but this may not fully capture all cases of type 2 diabetes, because many of these are reported in primary care. Finally, as usually happens with the RRR approach, the emerging DPs did not explain all of the variability in the nutrient-response variables that were included in the RRR model, and any remaining variability may be explained by other nutrients that may also be involved in the disease pathway. In exploratory mediation analyses (Supplementary Table 4), we investigated the role of the nutrient-response variables in the association between DPs and the risk of diabetes. After adjustment for the nutrients with the highest correlations with each DP, the associations for DP1 and DP2 were largely unchanged, which means that the derived DPs were associated with diabetes independent of those nutrient-response variables. This can be partly explained by the fact that both the DP and the nutrient-response variables may explain the same variability. But this also suggests that there might be other nutrients in the causal pathway explaining the observed risk with type 2 diabetes, for which future research is warranted. Residual confounding due to unmeasured variables (e.g., blood glucose (48), impaired fasting glycemia (49,50) is another limitation of observational research that can also explain the observed results.
Current dietary guidelines in the U.K. and many other countries are still based predominately on nutrient recommendations. Although this reflects the underlying evidence base, especially experimental and mechanistic research, it does not reflect the way people eat. Food-based dietary guidelines may help accelerate behavior change compared with nutrient-based recommendations by providing targeted and simpler advice on foods, which make it more, or less, likely to achieve an overall healthy diet (51). Moreover, this may help reduce the risk of conflicting messages regarding the relative importance of one or another nutrient, particularly saturated fat and free sugars, and recognize that there are many foods that are important sources of both.
In conclusion, this large population-based cohort study has shown clear evidence that diets high in chocolate and confectionery, butter and other animal-fat spreads, and low-fiber bread, and low intakes of fresh fruit, vegetables, and high-fiber breakfast cereals are associated with a higher risk of developing type 2 diabetes in this cohort of middle-aged British adults. The effects of a poor diet are especially pronounced in younger people and those living with obesity.
This article contains supplementary material online at https://doi.org/10.2337/figshare.19209750.
Article Information
Acknowledgments. This research was conducted using the UK Biobank resource under application number 14990.
Funding. C.P. received a British Nutrition Foundation pump-priming award that paid for the access to the data. S.A.J. and C.P. are funded by the Oxford and Thames Valley National Institute of Health Research (NIHR) Applied Research Centre. S.A.J. and P.A. are NIHR senior investigators. S.A.J. and J.C. are funded by NIHR Oxford Biomedical Research Centre. J.C. is supported by core grants to Clinical Trial Service Unit from the Medical Research Council and the British Heart Foundation (grant CH/1996001/9454).
The funders had no role in designing the study, the analysis, or the decision to submit the paper. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. M.G. and C.P. conceived and designed the research question and wrote the first draft of the manuscript. M.G. analyzed the data and G.L.A. contributed to dietary pattern analysis. G.L.A., A.P.C., K.P., J.C., and C.P. contributed to the data analysis. All authors contributed to interpretation of the data, revised the manuscript critically for important intellectual content, and read and approved the final manuscript. M.G. and C.P. are guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.