Ultra-processed food intake and incident venous thromboembolism risk: Prospective cohort study

Background & aims: Ultra-processed food (UPF) intake has been associated with multiple health outcomes, but data on the association between UPF intake and venous thromboembolism (VTE) risk are lacking. We conducted this study to examine the association between UPF intake and the risk of incident VTE. Methods: This prospective cohort study was based on 186,323 participants free of baseline VTE from the UK Biobank. UPF intake was assessed by 24-h recall questionnaires. Data on incident VTE came from the nationwide inpatient and primary care datasets and the death registry. Cox proportional hazards regression was used to estimate the association between UPF intake and incident VTE risk. Multiplicative interactions and strati ﬁ ed analyses by age, sex, and body mass index were performed. Results: During a 10.5-year (median) follow-up, 4235 incident VTE cases were diagnosed. After adjusting for covariates, the hazard ratio of VTE among individuals with the highest quintile of UPF intake was 1.05 (95% con ﬁ dence interval [CI] 0.94, 1.17) for UPF in servings, 1.12 (95% CI 1.01, 1.24) in grams, 1.10 (95% CI 1.00, 1.22) in grams %, 1.21 (95% CI 1.10, 1.33) in energy, and 1.15 (95% CI 1.05,1.27) in energy % compared to those in the lowest quintile. Age, sex, and body mass index did not modify the associations ( P interaction > 0.05). Conclusions: Higher UPF intake was associated with a moderately increased risk of VTE.

Ultra-processed foods (UPFs) as a product of industrial processing technique development are widely consumed worldwide nowadays and affect more and more people due to intensive marketing and the comparatively low price [1].It generates a huge impact on habitual diet.A systematic review indicated that over half of the daily energy intake in Western countries comes from UPFs [1].
The classification of UPFs may not be straightforward, yet UPFs are often energy-dense with a higher content of added sugar, saturated and trans fatty acids, and sodium as well as lower content of essential nutrients and dietary fiber compared with nonprocessed foods [2], and may thus confer detrimental health effects.Available studies have found that a high UPF intake is associated with an increased risk of all-cause mortality [3], cardiovascular disease [4,5], inflammatory bowel disease [6e8], and depression [9].In addition, a high UPF intake has been associated with obesity [10], in particular abdominal obesity [11], inflammation [12,13], and impaired kidney function [14].These factors have been identified as risk factors for venous thromboembolism (VTE) [15e18].VTE is a common disorder of the circulatory system and causes a heavy disease and economic burden worldwide [19].Even though the evidence speculates links between an excessive UPF intake and several risk factors for VTE, no studies have been conducted to examine the association between UPF intake and VTE risk.Here, we performed a prospective cohort analysis to investigate the association between UPF intake, classified according to the NOVA system, and the risk of incident VTE in a large-scale British population.

Study population
The present study was based on data from the UK Biobank, a large-scale ongoing population-based cohort study that recruited over 500,000 individuals, aged between 40 and 69 years, in 2007e2010 from 22 health centers [20].In this analysis, we included 191,897 participants who filled in a 24-h dietary recall questionnaire at least once and reported reasonable energy intake [8].After removing 5574 individuals with baseline VTE, 186,323 participants were included in the analysis.The UK Biobank study had been approved by the North WesteHaydock Research Ethics Committee (REC reference: 21/NW/0157), and all participants had signed an electronic consent.

Measurements of ultra-processed food intake
Dietary intakes from the 24-h recall of UK Biobank participants were assessed by the Oxford WebQ questionnaire that contains 206 food and 32 alcohol and beverage items consumed over the past 24 h.Compared to the food frequency questionnaire, this 24-h recall questionnaire had been well-validated to reduce bias by multiple measurements [21,22].Informed about the size of one portion, participants were asked to select how many portions they consumed for each food item.Dietary intakes in grams were estimated by multiplying the number of portions by grams of standard portion size.Based on this, we further calculated intakes of energy and nutrients by multiplying grams of consumed food by its energy and nutrient composition, respectively according to The Composition of Foods 6th edition (2002) [23].The dietary Web questionnaire was sent out to participants five times in 2011e2012 [20].For those who filled in the questionnaire more than once, the mean value of repeated 24-h dietary recalls per participant was calculated to present his/her intake.We defined the baseline time as the time when the participant filled in the dietary Web questionnaire for the first time.
UPFs were defined by the NOVA system, a food classification system based on the nature, extent, and purpose of industrial food processing [24].In this study, UPFs included foods and beverages that are classified NOVA4 together with ham and bacon [25].UPF intake was estimated based on information on portion size as well as nutrient and energy composition for each food item.The food items included in the UPF intake estimation are reported in Table S1.Total UPF intake was calculated as the sum of all these dietary items.We used five measures of quintiles of UPF consumption in the analysis: 1) numbers of UPF servings per day; 2) UPF intake in grams per day; 3) proportion of grams of food intake from UPFs per day; 4) energy intake from UPFs per day; and 5) proportion of energy intake from UPFs per day.

Measurements of covariates
We extracted data on age in years, sex (women and men), ethnicity (white and others), education attainment (below college, and college and above), smoking status (current, never, and past), alcohol consumption status (current, never, and past), and physical activity level (high, moderate, low, and missing) from the baseline questionnaires.Body mass index (BMI) was calculated by body weight in kilograms divided by the square of body height in meters measured by trained nurses during physical checks.The Townsend deprivation index (TDI) was estimated as an indicator of socioeconomic status and the method used for TDI estimation was described elsewhere [26].Total energy intake was treated as a covariate which was estimated using the method described in the previous part.We additionally obtained data on tea and coffee intake, sedentary time, air pollution, self-reported fracture, major operation, aspirin use, the oral contraceptive pill and hormonereplacement therapy use among women, and comorbidities that were represented by the Charlson Comorbidity Index.Detailed definitions of included covariates are shown in Table S2.

Ascertainment of outcomes and follow-up
VTE and its subtypes (deep vein thrombosis [DVT] and pulmonary embolism [PE]) were defined by the International Classification of Disease (ICD)-9 and 10 codes (Table S3) with data from the nationwide inpatient dataset and primary care dataset and death registry.We also used information on self-report diagnoses reviewed by nurses to remove individuals with baseline VTE.Individuals were followed up until the date of first diagnosis, death, loss to follow-up, or end of the follow-up whichever came first.

Statistical analysis
Missing rates of covariates ranged from 0.1% for alcohol consumption to 15.1% for physical activity (Table S4).Missing values were replaced by the mean and mode values in the study population for continuous and categorical variables, respectively.Given that the missing rate was high for physical activity, we created a group named "missing" for individuals without data on physical activity.We compared the differences in dietary intake of UPF components between quartiles 1 and 5 based on different UPF measures.
The associations between UPF intake and the risk of incident VTE and its subtypes were estimated using Cox proportional hazards regression.The proportional hazards assumption was examined by the Schoenfeld residual test and was found to be satisfied (minimal P > 0.078).Two models were used: model 1 adjusted for age, sex, and ethnicity, and model 2 was additionally adjusted for BMI, TDI, educational attainment, smoking and alcohol drinking status, physical activity, and energy intake.To detect the nonlinear association between UPF intake and VTE risk, we used a Cox proportional hazards model where the UPF intake score was entered in the model as a restricted cubic spline with three knots placed at the 25th, 50th, and 75th percentiles.We also examined multiplicative interactions of age (< and 60 years), sex, and BMI ( and >30 kg/ m 2 ) with UPF intake on VTE risk and conducted corresponding stratified analyses to estimate the associations in subgroups.To differentiate the associations of UPF food and beverage intake with VTE, we performed analyses to examine the associations of UPF beverages and UPF without beverages with VTE risk.
Several sensitivity analyses were performed to examine the robustness of results: 1) the analysis with additional adjustment for tea and coffee intake, sedentary time, air pollution exposure, fracture, major operation and aspirin use, the Charlson Comorbidity Index, and oral contraceptive pill and hormone-replacement therapy for women, which may be confounders; 2) the analysis confining to the population with at least two surveys of dietary information; 3) the analysis excluding incident cases within the first year of follow-up; 4) the analysis based on covariates with missing data imputed by chained equations; 5) the sensitivity analysis to further adjust for diet quality (measured by Alternative Healthy Eating Index [27]) to examine whether the UPF-VTE association is driven by diet quality; and 6) an analysis stratified by the Charlson Comorbidity Index to examine whether the UPF-VTE association is influenced by conditions predisposing to VTE.Since some food items containing bacon and ham are classified as ultraprocessed accordingly to the coding criteria for UPF in the U.K. National Diet and Nutrition Survey [25], this study included bacon and ham as UPFs in a conservative way.Given that it is controversial to labeling these two food items as UPF components, we further performed a sensitivity analysis excluding bacon and ham from UPFs.The association with a P-value <0.05 was deemed significant.All tests were two-sided and performed using R software, version 4.2.1.

Results
During a median follow-up of 10.5 (interquartile range 1.4) years, 4235 incident VTE cases (1855 DVT and 2380 PE cases) were diagnosed.The baseline characteristics of participants by incident VTE status are shown in Table 1.In general, compared to individuals with no incident VTE, those who developed VTE were more likely to be men and current smokers and had a lower socioeconomic status, a higher BMI, and a lower physical activity level.Compared to the group with the lowest intake of UPF (the lowest quintile of UPF measures), the group in the highest quintile had a higher intake of both UPF drinks and foods (Table S5).
Higher intake of UPF was associated with an increased risk of VTE using different measures of UPF consumption in both models (Table 2).In the fully adjusted model (model 2), the hazard ratio (HR) of VTE among individuals in the highest quintile was 1.05 (95% confidence interval [CI] 0.94, 1.17) for UPF in serving, 1.12 (95% CI 1.01, 1.24) in grams, 1.10 (95% CI 1.00, 1.22) in grams %, 1.21 (95% CI 1.10, 1.33) in energy, and 1.15 (95% CI 1.05, 1.27) in energy % compared to those in the lowest quintile of UPF intake.These associations remained stable in the analyses treating UPF as a continuous variable per standard deviation increment with HR of VTE ranging from 1.04 (95% CI 1.01, 1.08) in grams to 1.06 (95% CI 1.03, 1.10) in energy, but not in serving (HR 1.01, 95% CI 0.98, 1.05).The associations were overall consistent for the two subtypes of VTE albeit with larger CIs and a clearer association for PE (Table 3).Among different measures of UPF intake, higher UPF intake in energy was robustly associated with an elevated risk of DVT and PE.Compared with those in the lowest quintile, individuals in the highest quintile had an HR of 1.20 (95% CI 1.05, 1.36) for DVT and 1.23 (95% CI 1.07e1.42)for PE (Table 3).When differentiating the associations of UPF food and beverage intake with VTE, we found a decreased magnitude of the association for UPF beverage intake compared to UPF food intake with VTE risk (Table S6).When measuring UPF in energy or energy %, we found positive associations between UPF without beverages with VTE risk and the associations were stronger than that for UPF beverages.
There was an indication of nonlinearity of the association between UPF intake and VTE risk when assessing UPF intake in servings (P ¼ 0.007) and grams (P ¼ 0.047).The HR of VTE appeared to dramatically increase at UPF intake of 400e800 g (around 6e14 servings) per day (Fig. 1).Beyond a UPF intake of 800 g or 14 servings per day, the association between UPF intake and VTE risk became positively linear (Fig. 1).We did not detect any evidence of nonlinear associations with VTE for other UPF intake measures (Fig. 1).
No clear interactions were observed with age, sex, or BMI, and the associations between UPF and VTE risk were similar between subgroups defined by these factors (Tables S7eS9).The positive association between UPF intake and the risk of incident VTE was stable in a series of sensitivity analyses (Tables S10eS12) as well as in the analysis with further adjustment for diet quality (Table S13).The UPF-VTE association was also consistent in individuals with and without comorbidities (Table S14) and the analysis excluding bacon and ham from UPFs (Table S15).

Discussion
Based on data from a prospective cohort of 186,323 middle-aged adults, our study as the first cohort study revealed a positive association between UPF intake and the risk of incident VTE.The association was stable when measuring UPF intake in servings, grams, and especially in energy that may comparatively stably reflect habitual intake.The association was similar across subgroups defined by age, sex, and BMI.
Our finding of a positive association between UPF intake and VTE risk is in line with the potentially detrimental effects of an excessive UPF intake on a wide range of health issues [2e7, 9,28], in particular on cardiovascular disease [4,5] that shares certain etiological bases with VTE.In the NutriNet-Sant e cohort including 105,159 individuals with a median follow-up of 5.2 years, the HR comparing individuals with high to low UPF intake was 1.23 (95% CI 1.04, 1.45) for any cardiovascular disease, 1.18 (95% CI 0.96, 1.45) for coronary heart disease, and 1.23 (95% CI 1.00, 1.53) for the cerebrovascular disease after controlling for relevant confounders [4].Another cohort study including 3003 individuals with baseline cardiovascular disease found that one additional daily serving of UPF was associated with an HR of 1.05 (95% 1.02, 1.08) for overall cardiovascular disease and 1.09 (95% 1.04, 1.15) for coronary heart disease after an 18-year follow-up [5].Given atherosclerosis is a major risk factor for cardiovascular disease, these consistent findings may inform a general link between UPF intake and atherosclerotic development [29].Our study identified a similar association between a high UPF intake and VTE risk, which might add population-level evidence of a link between UPF intake and thrombosis formation and partly indicated a negative influence of UPF on the whole circulatory system along with the findings on atherosclerotic outcomes.Even though the observed association was moderate, this finding has great public health implications for VTE prevention given that UPF is commonly consumed in many high-income areas [1].In addition, our study found a possible Jshaped association between UPF intake and the risk of VTE at the interval of consuming UPF 400e800 g per day.Considering that the mean intake of UPF was more than 400 g in our study and other studies [2,5], most of the population may suffer from an increased risk of VTE caused by a high UPF intake.
Several mechanisms may explain the positive association between UPF intake and VTE risk.First, a high intake of UPF has been associated with obesity, particularly abdominal adiposity [10,11], which is a causal risk factor for VTE [15].In our analysis, the positive association between UPF intake and VTE risk persisted after adjusting for BMI, which indicates other alternative pathways.Second, UPF intake has been associated with inflammation, for example, high levels of interleukin-6 [13] and high-sensitivity Creactive protein [12].Increased levels of interleukin-6 and C-reactive protein have been associated with the risk of incident VTE [30].Third, excessive consumption of UPF has been reported to associate with renal dysfunction [14].Reduced estimated glomerular filtration rate, an indicator of kidney impairment, has been found to be a risk factor for VTE [16].Forth, a high intake of UPF may be correlated with an unideal dietary quality that was associated with an increased risk of VTE [31].However, the Moli-sani Study revealed that the association between UPF intake and mortality was not explained by the poor quality of diet [32].Our study also found a robust positive UPF-VTE association after additional adjustment for diet quality, which indicates that this association may be not driven a The continuous variables (age, BMI, total energy intake, and ultra-processed food intake) were expressed in mean (standard deviation) and the categorical variables were in number (percentage).by poor diet quality either.In addition, our study did not differentiate the associations of UPF foods and beverages on VTE given that both were associated with VTE risk when measured in energy or energy %.
This study has several advantages, including a large sample size with a long follow-up period, multiple surveys on food intake for most of the population, adjustment for many important covariates, and robust results from a series of sensitivity analyses.Limitations need to be specified when interpreting our findings.First, even though we minimized confounding by adjusting for confounders as possible, this study is an observational cohort study that cannot infer the causality of the association between UPF intake and VTE risk.Second, dietary habits and lifestyle factors might be prone to be changed during this long follow-up.Thus, using baseline The associations were adjusted for age, sex, ethnicity, body mass index, Townsend's deprivation index, educational attainment, smoking and alcohol drinking status, physical activity, and energy intake.Abbreviations: CI, confidence interval; HR, hazard ration; SD, standard deviation; UPF, ultra-processed food.
Fig. 1.Associations of various estimates of UPF intake with incident venous thromboembolism risk in restricted cubic spline analyses.The spline analyses included three knots at the 25th, 50th, and 75th percentiles.P for nonlinearity was 0.007 for A, 0.047 for B, 0.970 for C, 0.685 for D, and 0.164 for E. Abbreviations: CI, confidence interval; HR, hazard ration; SD, standard deviation; UPF, ultra-processed food.
information only might generate misclassification.However, given the prospective cohort design of the study, any misclassification of these factors should be randomly distributed between groups with and without incident VTE diagnosis, which might attenuate the association between UPF intake and VTE risk in a conservative way.In addition, using dietary data from one or two 24-h recall surveys might not reflect habitual intake.However, our results were robust when measuring UPF intake in energy which is a more stable indicator of habitual dietary intake compared to servings and grams.Third, we cannot rule out the possibility that participants adjusted their dietary patterns due to early symptoms of VTE even though this possibility should be minimal.However, the association remained stable in the sensitivity analysis where we removed incident VTE cases diagnosed within the first year of follow-up.Fourth, we did not have access to data on coagulation factors, which hinders the exploration of mechanistic pathways.Future studies are worthy of examining UPF intake and the levels of coagulation factors in relation to VTE development.Fifth, we included bacon and ham in UPF in the main analysis in a convective way since some food items contain bacon and ham are classified as ultra-processed accordingly to the coding criteria for UPF in the U.K. National Diet and Nutrition Survey even though bacon and ham per se are defined processed [25].If any bias by adding the two food items in UPF, this influence should be minimal due to the daily average energy intake of only 8.8% from processed food and even much less from bacon and ham in the UK population [25].In addition, we have added a sensitivity analysis excluding bacon and ham from UPF and found consistent results.
In conclusion, this prospective cohort study found that a high UPF intake was associated with an increased risk of incident VTE.This finding suggests that it may be a dietary strategy to prevent VTE by reducing UPF intake via informing customers about the health adversities of UPF, increasing taxes on UPF, etc.

Table 1
Baseline characteristics and UPF intake of participants by incident outcome status.

Table 2
Associations between UPF intake and risk of incident venous thromboembolism.

Table 3
Associations of UPF intake with risk of incident pulmonary embolism and deep venous thrombosis.