Study design and participants

The UK Biobank is a prospective cohort study that recruited 502 368 individuals aged 40–69 years from 22 assessment centers between 2006 and 2010. The baseline assessment comprised of a touchscreen questionnaire, a face‑to‑face interview, a range of physical measurements, and collection of biological samples (blood, urine, and saliva).

Our study included the participants who reported AP at baseline. AP was identified using the International Classification of Diseases 10th Revision (ICD‑10) codes K85.0, K85.1, K85.2, K85.3, K85.8, and K85.9, as well as ICD‑9 code 5770. The excluding criteria were concomitant chronic pancreatitis and MACCE prior to baseline.

Covariates

Information on age and sex was obtained from self‑reported baseline characteristics. Ethnicity was also self‑reported by the participants and categorized as white, Asian, black, or other. The educational attainment was gauged by the age at which the participants left full‑time education. For those who only reported a degree or qualification status, an equivalent age at completion of the education was determined. Income levels were calculated from average total household income before tax and grouped into level 1 (less than 18 000 GBP), level 2 (18 000 to 30 999 GBP), level 3 (31 000 to 51 999 GBP), and level 4 (over 52 000 GBP). Employment was categorized as employed / self‑employed or retired / unemployed. The Townsend deprivation index was calculated immediately before the participant joined the UK Biobank to assess the impact of socioeconomic factors on health outcomes. Physical activity was defined by the International Physical Activity Questionnaire as low, moderate, or high. Diet was classified as healthy or unhealthy, which was evaluated using a more recent definition of the ideal intake of dietary components for cardiovascular health.¹⁰ If the participants achieved the intake goal, they were considered to have an adequate intake of the diet component. Adequate intake of at least half of all diet components was considered as a healthy diet, less than half was considered an unhealthy diet (field identifiers and serving sizes used per diet component in the UK Biobank are provided in Supplementary material, Table S1). Smoking and drinking status were self‑reported and categorized as never, former, or current. BMI was obtained from physical measurements, which included height and weight measured at the initial visit to the assessment center. The presence or absence of hypertension, dyslipidemia, and diabetes were self‑reported at baseline, and were categorized as yes or no. Information on medication for high cholesterol level, high blood pressure, and diabetes was obtained from the “Health and medical history” section of the touchscreen questionnaire initially completed at the assessment center. More detailed information on covariates was provided in the Supplementary material, Table S1.

Cardiovascular outcomes and follow‑up

The primary outcome of this study was MACCE, defined as a composite of all‑cause death, myocardial infarction (MI), and stroke. The secondary outcomes included all‑cause death, MI, and stroke. The date of the earliest reported MI or stroke for a participant was algorithmically defined by the UK Biobank, as it was linked to relevant hospital inpatient records (Hospital Episode Statistics for England, Morbidity Records for Scotland). Date of death was acquired from death register dataset (National Health Service [NHS] Digital, NHS Central Register, and National Records). For all participants, follow‑up started at inclusion in the UK Biobank and ended either when MI, stroke, or all‑cause death occurred or on the study’s conclusion on March 30, 2023, whichever occurred first.

Statistical analysis

To present the baseline data, mean (SD) was used for continuous variables, and counts and percentages were used for categorical variables. To address the missing data, we applied multiple imputation with chained equations to impute the missing values, and conducted ten imputations (baseline demographic comparison in the imputation cohort, nonimputation cohort, and excluded cohort was shown in Supplementary material, Table S2). The median (interquartile range [IQR]) follow‑up time was calculated using the reverse Kaplan–Meier method. Crude incidence rate (IR) per 1000 person‑years was calculated with the Poisson regression and associated 95% CIs using the exact Poisson method. Cumulative rates for MACCE and its components according to sex were exhibited on the Kaplan–Meier curve, and differences between the 2 groups were compared by the log‑rank tests.

We used the Cox proportional hazards regression models to assess the sex differences in cardiovascular outcomes, the hazard ratios (HRs) and 95% CIs, followed by separate analyses with MACCE components of all‑cause death, MI, and stroke. In the sequentially‑adjusted Cox proportional hazards models, we adjusted all covariates step by step. Model 1 included only sex; model 2 included age, ethnicity, education level, income levels, employment status, and the Townsend deprivation index, and was further adjusted on the basis of model 1; in model 3, physical activity, diet quality, smoking status, and drinking status were further adjusted on the basis of model 2; in model 4, BMI, hypertension, dyslipidemia, and diabetes were further adjusted on the basis of model 3; and in model 5 (fully‑adjusted model), medications including antihypertensive, lipid‑lowering, and hypoglycemic drugs were further adjusted on the basis of model 4. Female sex was used as the reference group in all models. The Schoenfeld residuals were used to assess the proportional hazard assumption, and no violation was found. The interaction between sex and age was evaluated in model 5.

Considering that the primary outcome of this study was a composite end point, the results for the outcome were further analyzed with the win ratio method, and the inverse probability of treatment weighting (IPTW) was used to adjust for confounding and selection bias at baseline,¹¹ including sex, age, ethnicity, education level, income levels, employment status, the Townsend deprivation index, physical activity, diet quality, smoking status, drinking status, BMI, hypertension, dyslipidemia, diabetes, and medication. Notably, both inverse probability weighted average treatment effect (IPW‑ATE) and stabilized IPTW (SIPTW)-ATE, were similarly adjusted for the aforementioned indicators. Every male patient was compared with every female patient during a shared follow‑up time defined as the minimum of their follow‑up times. The pairs were classified as winners for men if all‑cause death during follow‑up occurred first in women, and losers if all‑cause death occurred first in men. If both participants in a pair completed or exited the study before all‑cause death, they were classified according to who experienced MI first in a hierarchical order. If those completed or exited the study before MI, they were classified according to who experienced stroke first. A pair was tied if a decision could not be made on whether it was a winner or a loser. The win ratio was defined as the total number of winner pairs divided by the total number of loser pairs. Therefore, the benefit of male sex is provided by a win ratio above 1. The 95% CIs were estimated for the win ratio effect measures. To facilitate the comparison of the estimates provided by the win ratio with those of the HR, the HR as 1/HR and respective 95% CIs are also reported. 1/HR represents the rate of events in the female group vs the rate in the male group, whereby a 1/HR above 1 indicates a higher event rate in the female group than in the male group (ie, a benefit for the men).

Several sensitivity analyses were conducted to test the robustness of our findings. First, we only included participants with no missing values. Then, within this subset, the individuals were stratified into either single or recurrent AP cohort, based on the frequency of AP episodes, facilitating a renewed examination of the correlation between sex and MACCE. All statistical analyses were performed using R project version 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria). A 2‑sided P value below 0.05 was considered significant.

Ethics

The UK North West Multi‑Centre Research Ethics Committee approved the UK Biobank (11/NW/0382). All participants provided their written informed consent for the questionnaire and collection of biological data. This research was conducted under UK Biobank application number 152047.

Characteristics of the study participants

Of the 1963 participants diagnosed with AP at baseline, 485 were excluded due to coexisting chronic pancreatitis, and 108 due to a history of MACCE. This left a total of 1371 participants for final analyses. Figure 1 shows the flowchart of the study sample selection.

Mean (SD) age of the study participants at baseline was 59 (7.3) years, 788 were women (57.5%) and 583 were men (42.5%). Age distribution did not differ between sexes (P = 0.06). The baseline characteristics of the sample according to sex are provided in Table 1. In general, men achieved higher educational level, had higher income, and higher employment rate. Men also demonstrated reduced BMI, increased blood pressure, and a higher prevalence of current smokers and current drinkers. Additionally, a higher percentage of men than women had hypertension, dyslipidemia or diabetes, and were taking antihypertensive, lipid‑lowering, and hypoglycemic drugs.

Sex differences in incidence rates of cardiovascular outcomes

Over a median (IQR) follow‑up duration of 13.9 (13–14.7) years, encompassing 18 030.9 person‑years, 226 cases of MACCE occurred in 1371 individuals, including 122 in men (IR, 16.44; 95% CI, 13.77–19.64) and 104 in women (IR, 9.8; 95% CI, 8.09–11.88) (Table 2). The IR for all‑cause death was significantly higher in men than in women (11.39; 95% CI, 9.24–14.04 vs 7.59; 95% CI, 6.11–9.42). IR of MI and stroke was also greater in men (4.68; 95% CI, 3.41–6.43 vs 2.05; 95% CI, 1.36–3.08 and 2.42; 95% CI, 1.56–3.75 vs 1.33; 95% CI, 0.8–2.2, respectively; Table 2). For women as controls, the IR ratio for MACCE, all‑cause death, MI, and stroke was 1.68, 1.5, 2.29, and 1.82, respectively. Even after stratifying the population by age (<⁠60 years old and ≥60 years old), IR for MACCE and its components was still higher for men in both age groups (Supplementary material, Table S3).

The Kaplan–Meier survival curves showed the cumulative risk of MACCE and its components in men and women. The log‑rank test demonstrated significantly higher long‑term rates of cumulative MACCE, all‑cause death, and MI in men (Figure 2). After stratification according to age, the Kaplan–Meier curves showed the same results (Supplementary material, Figure S1).

Risk for men in multivariate models

The Cox regression analysis adjusting for multiple variables consistently demonstrated that men were at a significantly increased risk for MACCE (HR, 1.8; 95% CI, 1.36–2.38), all‑cause mortality (HR, 1.49; 95% CI, 1.08–2.06), and MI (HR, 2.75; 95% CI, 1.57–4.81; Figure 3). However, the risk of stroke did not show significant sex‑dependent differences (HR, 1.67; 95% CI, 0.79–3.52).

The HR for sex differences for MACCE and all‑cause mortality increased after adjusting for demographic characteristics, and then dropped after adjusting for lifestyle and metabolic factors and medication. For MI, the HR for sex difference rose after addition of demographic characteristics, lifestyle factors, and metabolic factors, and it reached its maximum value after addition of medication. The same trend was observed for stroke, but no significant sex difference was found after step‑by‑step adjustment (Figure 3). Additionally, no significant interaction between age and sex was observed in the fully‑adjusted model (Supplementary material, Table S4).

Win ratio for the analysis of the primary composite outcome

The summary statistics of the baseline characteristics of patients by pre- and post‑IPTW adjustment are presented in Supplementary material, Table S5. To better assess the homogeneity of baseline characteristics before and after IPTW adjustment, the standardized mean difference in each of the baseline covariates was calculated (Supplementary material, Figure S2).

Table 3 displays the unadjusted and adjusted win ratio statistics for analyses of the composite outcome of AP patients by sex. The unadjusted win ratio was 0.61 (95% CI, 0.47–0.8; P <⁠0.001), the IPTW‑adjusted win ratio from the IPTW‑ATE and SIPTW‑ATE was 0.55 (95% CI, 0.42–0.72; P <⁠0.001). The IPTW‑adjusted win ratio closely aligned with the results of the adjusted Cox models regarding the point estimate, interval estimate of 1/HR, and P value.

Sensitivity analyses

As a sensitivity analysis, we assessed the sex‑depending risk for all outcomes in the population without missing values (Supplementary material, Table S6), and the results remained essentially unchanged (Supplementary material, Tables S7 and S8). Then, we stratified the whole cohort into single and recurrent AP cohorts, and the results were still consistent with our primary outcome (Supplementary material, Tables S9–S11).

Conclusions

In this large cohort study of the UK participants with long‑term follow‑up, we found a higher risk of MACCE, all‑cause death, and MI in men than women. Further investigations are necessary to better understand the mechanisms behind this increased risk in men, and further actions are required to prevent and manage cardiovascular risk in the male population.