Study design and participants
Between 2006 and 2010, UK Biobank recruited over 500,000 middle- and old-aged UK adults [15]. Individuals registered with the National Health Service and living within 25 miles of a UK Biobank assessment center were invited to participate voluntarily. Of the 9 million individuals invited to participate, 5.5% were ultimately enrolled. All participants attended 1 of 22 assessment centers where they completed questionnaires, took physical measurements, and provided biological samples [15]. All participants provided informed consent through electronic signature at baseline assessment, and the UK Biobank study was approved by the North West-Haydock Research Ethics Committee (16/NW/0274). Participants who withdrew from UK Biobank (n =46) and had cancer (n =56,029) or CVD (n =31,637) at baseline and those without data on salt added at the table (n =956) and those who died from coronavirus disease 2019 during pandemic (n =728) were excluded from the analysis. The final analytical sample included 413,109 participants (Additional file 1: Fig. S1).
Exposure assessment
At the baseline, the frequency of adding salt at the table was recorded. Participants were asked to report their daily salt use habits, with the question “Do you add salt to your food? (Do not include salt used in cooking)”. Possible responses include “Never/rarely,” “Sometimes,” “Usually,” or “Always.” Discretionary salt is usually added during cooking or at the dinner table [3]. Therefore, adding salt to food other than during cooking was defined as adding salt at the table. A quadratic weighted kappa statistic was used to evaluate the long-term reliability of adding salt at the table at the baseline and first repeated assessment (mean 4.3 years after baseline) or second repeated assessment (mean 9.0 years after baseline). Moderate reliability was identified with a kappa coefficient of 0.60 for the first repeated assessment and 0.53 for the second (Additional file 1: Table S3).
Outcome ascertainment
The main outcomes were incident CVD and all-cause mortality, and the secondary outcomes included the main types of CVD (coronary heart disease, heart failure, stroke) and CVD mortality. CVD was identified based on the 9th revision of the International Statistical Classification of Diseases (ICD-9), ICD-10, and Office of Population, Censuses and Surveys-4 (OPCS-4) codes and self-reported data fields with the choice-, disease-, or procedure-specific codes (Additional file 1: Table S1) [16]. We compared the date of the first CVD diagnosis and the baseline assessment visit. Participants with CVD before or at the baseline were ascertained as prevalent CVD cases, while those diagnosed with CVD after baseline were regarded as incident CVD cases. At the time of this analysis, the inpatient record data were available as of March 31, 2021. We censored follow-up at this date or the date of the first incident of CVD or death, whichever came first. As mortality data were available up to 28 February 2021, we censored follow-up at this date or the date of death, whichever occurred first.
Covariate assessment
Information on sociodemographic factors, lifestyle factors, dietary intake, medical history, and medication use was collected using a touch-screen, self-completed questionnaire at the baseline assessment. Height (m) and body weight (kg) were measured by trained nurses at the baseline. Ethnicity was categorized as White, Mixed, Asian, Black, Chinese, and others. Education was categorized as college or university, vocational, upper secondary, lower secondary, and others. Household income was categorized as <£18,000, £18,000–£30,999, £31,000–£51,999, £52,000–£100,000, and >£100,000. Smoking status was defined as never, former, and current. The frequency and volume of current alcohol consumption were self-reported, and alcohol consumption was considered as a continuous variable. People in the UK are advised not to drink more than 14 units a week (equivalent to 16 g/day) [17]. Socioeconomic status was derived from Townsend deprivation index scores and categorized into quintiles (a higher score denoted a higher degree of deprivation) [18]. Physical activity level over a typical week was self-reported using the validated International Physical Activity Questionnaire, and the total metabolic equivalent of task (MET) in a week was categorized into quintiles [19]. Body mass index (BMI) was calculated as the weight in kilograms (kg) divided by the square of the height in meters (m2) and was considered a continuous variable. Medical history (hypertension, diabetes, dyslipidemia, family history of stroke, and heart disease) was classified as yes or no. We constructed a healthy diet score concerning the dietary priorities for cardiometabolic health recommended by the American Heart Association [20]. Definitions of each component of the healthy diet score are shown in Additional file 1: Table S2. The scale of the healthy diet score ranged from 0 to 10, and a higher score denoted a healthier dietary pattern. The healthy diet score was categorized into quintiles [16].
Statistical analysis
Baseline characteristics were presented as the number (percentage) for categorical variables and the mean (standard deviation) for continuous variables. General linear models were used to evaluate the association between the frequency of adding salt at the table and the estimated 24-h urinary sodium excretion. Multivariable-adjusted Cox proportional hazards models were performed to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations of adding salt at the table with incident CVD and mortality. The proportionality assumption was checked by using the Schoenfeld residual test. Three sequential models were used. Model 1 adjusted for age, sex, ethnicity, education, Townsend deprivation index, and household income; model 2 further included smoking status, alcohol consumption, total physical activity, and healthy diet score; model 3 further included BMI, diabetes, dyslipidemia, family history of stroke, and heart disease. Because the formula for calculating estimated 24-h urinary sodium included age, sex, height, and weight [21,22,23], we did not adjust for age, sex, and BMI in the general linear model. The missing values of covariates were treated as dummy variables.
Subgroup analyses were performed by the following variables: age (<60, ≥60 years), sex (man or woman), education (College or university/Vocational, upper secondary or lower secondary), ethnicity (white, non-white), household income (≤£30,999, >£30,999), Townsend deprivation index (below the median, above the median), smoking status (current, previous, never), alcohol consumption (≤16, >16 g/d), BMI (<30.0, ≥30.0 kg/m2), diabetes (yes or no), and dyslipidemia (yes or no). As for hypertension, we stratified it into diagnosed and undiagnosed. Diagnosed hypertension was defined based on ICD-9, ICD-10, a self-reported history of hypertension, or the use of antihypertensive drugs. Given that elevated blood pressure might be the main pathway by which salt affects CVD, rather than a source of reverse causality, screen-detected participants with baseline systolic or diastolic blood pressure ≥140 mmHg or 90 mmHg and those without hypertension were defined as undiagnosed hypertension. The effect modification was assessed by including multiplicative interaction terms with the frequency of adding salt at the table in the models.
Four sensitivity analyses were conducted to test the robustness of the main findings. First, to minimize the potential reverse causation bias, participants who developed CVD or died during the first 2 years of follow-up were excluded and re-ran the main analyses. Second, participants with chronic kidney disease were excluded because they have diminished capacity to excrete sodium and a higher risk of cardiovascular disease [24]. Third, dietary variation may alter the habit of adding salt at the table, and participants whose diets frequently changed week to week were excluded. Fourth, major diet changes before the baseline may contribute to changes in the habit of adding salt at the table, and those who had major dietary changes in the previous 5 years were excluded.
A 2-tailed P-value of less than 0.05 was used to determine the statistical significance. Results from subgroup analyses were considered exploratory due to multiple testing. All analyses were performed by SAS version 9.4 software (SAS Institute, USA).