Statistical Analyses
Incidence rates for stroke are calculated per 1000 person-years of
follow-up. The significance of differences in risk factor means or
proportions between high and low depressive symptoms categories
was tested by means of unadjusted least-squares estimates using the
SAS general linear models procedure (35). To control for all risk
factors simultaneously and to account for unequal lengths of followup,
Cox proportional hazards regression models (SAS procedurePHREG) were used to model time to the event and to calculate
estimates of the relative risk of stroke and associated 95% confidence
intervals (36, 37). PHREG performs regression analysis of
survival data based on the Cox proportional hazards model. The Cox
model is the preferred model for analyzing NHEFS data because it
takes into account different lengths of follow-up and does not require
assumptions about the distribution of survival time (38).
Length of follow-up was calculated as the time from the date of
examination to the date of a stroke (cases) or to the date of the last
follow-up interview or death (noncases). Participants were followed
up for a mean length of 16.0 years (maximum length 5 21.8 years).
The SAS-callable SUDAAN survival procedure was used to check
for changes in stroke risk after using sampling weights (39). Because
the weights and complex survey design variances did not change the
overall conclusions of the study, only unweighted results are presented