Study design and population
Since 1999, the NHANES survey transitioned into a continuous program, with findings released biennially. To gather information on prevalence and severity of psoriasis, the present work used data from four cycles (2003-2006, 2011-2014), encompassing 40,228 general participants(17,18). The NHANES study protocol was approved by the NCHS research ethics review board, with all included participants providing written informed consent at enrollment.
The National Death Index documented the vital status of corresponding participants from the NHANES survey. The follow-up period for each study participant spanned from the NHANES baseline interview to either the participant’s death or the final date of follow-up (December 31, 2019). The cause of death was defined by ICD-10 codes.
This study was designed to explore the association between psoriasis and CVD. Considering the small number of participants with psoriasis, relatively lenient exclusion criteria were applied: participants younger than 20 years old (n = 18,089); missing data for psoriasis status (n = 1,928); missing data for psoriasis area (n = 6); missing data for mortality follow-up (n = 44); and pregnant participants (n = 420). After applying these criteria, a total of 19,741 participants, including 542 with psoriasis and 19,199 without, were incorporated into the study. The participant selection procedure was further elucidated in a flowchart in Figure 1.
Data collection
In a standard NHANES data collection procedure, demographic data and standardized questionnaires were collected by trained interviewers. Baseline information, including age, sex, race/ethnicity, family poverty income ratio (PIR), employment status, marital status, education, smoking status, drinking status, and medical condition, was self-reported without any leading questions. Subsequently, participants underwent physical examinations in mobile examination units, where anthropometric measurements were taken. Serum specimens were also sampled, stored and detected for biochemical profiles.
Data preprocess
Participants with psoriasis were identified based on a positive response to the question "Ever told had Psoriasis?". The severity of psoriasis was categorized according to the extent of the lesions. Participants with "little or no psoriasis" or "only a few patches" were classified as having mild psoriasis, while those with "scattered patches" or "extensive psoriasis" were considered moderate/severe cases. Likewise, cardiovascular symptoms or diseases were defined by the self-reported health condition questionnaires. In addition, we defined overweight as BMI ≥ 25 kg/m2 and obesity as ≥ 30 kg/m2, in accordance with the WHO classification(19). Due to sample size limitations, borderline diabetes population was classified as normal rather than a separate category.
In reference to previous studies, we recoded certain variables for the convenience of subsequent analyses. Race was categorized into Mexican, Hispanic, Black, White and others. Education levels were divided into three ordered levels: less than high school, high school or equivalent, more than high school. For marital status, widowed and divorced participants were considered as separated, and living with partner was merged with married. The economic status of participants was evaluated based on the annual family PIR, partitioned at 1.3 and 3.5(20). Additionally, we included multiple laboratory data files, using their respective concentration units. For indicators with international standardized unit format, we opted for their standardized format for further analysis
Statistical analyses
Following data collection and cleaning, variables with more than 30% missing values were removed, with the exception of the primary cause of death. Multiple imputation by chained equations was employed to complement missing values, and the interpolated results were validated.
We calculated differences in demographic variables, medical conditions, and biochemistry profiles. The Shapiro-Wilk test was performed on continuous variables to determine their distribution. Variables with a non-normal distribution were described using medians and interquartile ranges, and exact tests were applied for comparisons between participants with and without psoriasis. Categorical variables were reported using counts and percentages and compared using Pearson's Chi-squared tests.
The associations between psoriasis and the prevalence of hypertension, hypercholesterolemia, diabetes, obesity, CHF, CAD, angina pectoris, MI and stroke were evaluated using logistic regression. The risks attributable to psoriasis were presented as odds ratios (ORs) and 95% confidence intervals (CIs). Potential confounders such as age, sex, race, education and smoking status were adjusted during analysis.
In our prospective study, psoriasis patients were stratified into mild and moderate/severe subgroups. All-cause mortality and cause-specific mortality from CVD were calculated using the Kaplan–Meier (KM) method and represented as cumulative hazards. Differences between groups were estimated using the log-rank test. The proportional hazards assumption was tested by calculating Schoenfeld residuals. Cox regression models were fitted based on univariate and multivariate conditions and hazard ratios (HRs) were respectively calculated. Adjusted survival curves were visualized based on the multivariate Cox model, accounting for sex, age, race, education, marital status, family PIR, smoking and drinking status.
Subgroup analyses were conducted using univariate regression to balance potential clinical heterogeneity between participants with moderate/severe psoriasis and those without psoriasis. Subgroup analysis factors included sex, age brackets (<65, ≥65 years old), race, education, marital status, family PIR, drinking, smoking, hypertension, hypercholesterolemia, diabetes and obesity, which are known CVD risk factors. Interaction effects were assessed statistically between psoriasis and the above variables. Interaction effects between psoriasis and these variables were statistically assessed.
Potential causal associations from psoriasis to CVD mortality via serum biochemical indexes were explored using multiple mediation analysis. Using the "mma" package, Mediators and their mediation effects were statistically inferred, adjusting for baseline CVD risk factors(21). The bootstrapping method was employed to estimate effect sizes and 95% CIs.
All analyses were performed using R version 4.2.2. All statistical tests were two-sided, with a pre-set significance threshold of p < 0.05.