The UK Biobank included 502,409 participants from England, Scotland and Wales between March 2006 and July 2010. Inclusion criteria included age 40–69 years and living within a reasonable travel distance (10 miles) from one of the 22 assessment centers. Participants provided biological samples, physical measurements and baseline information in the assessment centers. After 6 participants dropped out of the UK Biobank project, 502,409 were obtained. Furthermore, 1526 participants without information for milk type used, 1283 participants lost to follow-up and 49,093 participants diagnosed with CVD at baseline were excluded. Then, 450,507 participants without CVD at baseline were enrolled in this study (Fig. 1).

Flowchart of participant selection

All the enrolled participants attended one of 22 assessment centers across the UK and finished a questionnaire using a touch screen device. One of the questions asked, “What type of milk do you mainly use?” The participants can choose their answer from a list of milk, including skimmed, semi-skimmed, full cream, soy and other milk types.

The study’s main outcomes were all-cause mortality, CVD mortality and CVD events. The secondary outcomes were stroke and myocardial infarction incidence and mortality. The date and cause of death were determined by linking to the Death Registry of the National Health Service (NHS) Information Centre for the UK and Welsh participants and the Death Registry of the Scottish NHS Central Registry for Scottish participants. Additionally, dates and reasons for hospitalization were identified by linking to Scottish morbidity records for Scottish participants and health event statistics for England and Wales participants. Detailed information can be found at http://content.digital.nhs.uk/services. This date was used as the end date for follow-up unless death or hospital admission occurred first for CVD outcomes. The CVD events were considered as hospitalization or death due to the following ICD-10 codes according to the hospital or death record: CVD death codes: I00–I99, CVD codes: I20–I25 and I60–I64, myocardial infarction codes: I21, I22, I23, I24.1 or I25.2 and stroke codes: I60–I64 [15,16,17,18].

The baseline questionnaires were used to evaluate some potential confounding variables: sociodemographic factors (age, sex, ethnic background and household income), socioeconomic status (Townsend Deprivation Index), lifestyle habits (smoking status, alcohol consumption, tea intake, processed meat intake obesity, dietary intake (fruits and vegetables). Furthermore, medication use (blood pressure drugs and cholesterol-lowering drugs use), vitamin supplements (vitamin A, vitamin B, vitamin C, vitamin D, vitamin E, multivitamins, or folic acid), minerals supplements (calcium, iron, zinc or selenium) and comorbidities (hypertension, diabetes, high cholesterol and long-term illness) were evaluated. The Townsend Deprivation Index was provided in the UK Biobank. The information for medical history (hypertension, diabetes, high cholesterol and long-term illness) was obtained through self-report at baseline. The body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. In this study, obesity was considered as BMI ≥ 30, which consists with the European guidelines for obesity management in adults [19]. Details of these evaluations could be obtained from UK Biobank (www.ukbiobank.ac.uk).

Lilliefors test was adopted to detect if the data have a normal distribution [20, 21]. Continuous variables were indicated as mean ± standard deviation for normally distributed data; otherwise, they were shown as median and interquartile range. All categorical variables were presented as counts with percentages. Cox regression models were adopted to analyze the connection between milk consumption and all-cause mortality and CVD outcomes (CVD mortality, myocardial infarction mortality, stroke mortality, CVD events, myocardial infarction and stroke). Two models were constructed, and the basic model was adjusted for baseline age (years) and sex (male or female). The multivariable model was further adjusted for Townsend Deprivation Index, ethnic background (white or others), household income (< £18,000, £18,000–£30,999, £31,000–£51,999, £52,000–£100,000, or > £100,000), obesity (yes or no), fruit consumption, vegetable consumption, smoking status (current, never or previous). In addition, alcohol intake (< 1, 1–2, 3–4, or > 4 times/week), tea intake (< 2.0, 2.0–3.9, or ≥ 4.0 servings/day), processed meat intake (< 2, 2–4, 5–6, or > 6 times/week), vitamin use (yes or no), minerals use (yes or no), blood pressure drugs use (yes or no), cholesterol-lowering drugs use (yes or no), hypertension (yes or no), diabetes (yes or no), high cholesterol (yes or no) and longstanding illness (yes or no).

Additionally, several subgroup analyses were conducted by age (≥ 60 vs < 60 years), ethnic background (white vs others), sex (male vs female), current smoking status (yes vs no), diabetes (yes vs no), high cholesterol (yes vs no), hypertension (yes vs no), obesity (yes vs no), longstanding illness (yes vs no), blood pressure drugs use (yes vs no), cholesterol-lowering drugs use (yes vs no) and minerals use (yes vs no).

Several sensitivity analyses were conducted to determine the stability of cox regression model results. First, the participants who had undergone an outcome event during the first 2 years of follow-up were removed. Second, the participants who use vitamin were excluded. Third, all missing covariate data were imputed using multiple imputations. Fourth, the participants diagnosed with cancer at baseline were removed. All results were indicated as HR and 95% CI. R version 4.1.2 (www.r-project.org) was adopted for the analysis, and a two-sided P value < 0.05 was set as statistically significant.