Two groups were compared on baseline characteristics,
using Chi-squared test, Independent t-tests and Mann-Whitney test based on the
normal distribution of data.
The event date for the incident of
each outcome was defined as mid-time between the date of follow-up visit at
which the outcome was diagnosed for the first time, and the most recent
follow-up visit prior to the diagnosis and for those with negative event
(censored subjects). The time was the interval between the first and the last
observation dates. Incidence density of each outcome was calculated per 1000
person-years of follow up among the PCOS and control groups.
Univariate and multiple cox
proportional hazards regression with age as the time scale was used to estimate
the adjusted hazard ratio (HR) of each outcome in relation to PCOS. In this
model, the baseline hazard assumes that each subject’s observed risk period was
started at birth and was not adjusted for age-truncation.
The proportional hazards
assumption of each model was assessed via the Schoenfeld residual test and
rechecked graphically by log (? log) that suggested departures from the
proportionality in univariate and multivariate models (Figure 2). Therefore,
the extended cox proportional hazards regression model with age as the time
scale was used to compute the hazard ratios (HR) separately for age-varying
exposure of the PCOS. This model allowed step functions and provided constant
hazard ratios within different age intervals. In this respect, the hazard
ratios were computed separately for ? 40 years and > 40 years with heavyside
functions. The age of forty were selected based on the cumulative hazard curves
of each outcome. These curves showed that the difference in hazard between the
PCOS and controls were diluted or had crossing pattern after the age of forty.
The baseline BMI and physical
activity were evaluated for confounding covariates for all events. In addition,
baseline SBP, family history of hypertension (HTN) and smoking history for HTN
events, baseline lipid profile (TC, TG, and HDL-C) for dyslipidemia events and
family history of HTN, family history of diabetes for metabolic syndrome were
included in adjusted models as special confounders for each outcome separately.
All statistical tests were 2-tailed, and p-value < 0.05 was considered statistically significant. The Statistical Package for Social Sciences (SPSS ® version 16; SPSS Inc., Chicago, IL, USA) and STATA software (Version 12; STATA Inc., College Station, TX, USA) were used for data analysis.