Worker healthy effect in cardiovascular disease prospective cohorts. Does it still matter? Comparing survival curves of northern Italian population-based and factory-based cohorts.

Purpose: When comparing results of prospective cohort studies with cardiovascular disease as primary endpoint, the potential bias due to worker healthy effect (WHE) is a matter of concern. In particular, the concern rises when population-based and factory-based studies are included in meta-analysis or in pooled-analysis. We aim to estimate the WHE in Northern Italian cohorts for major cardiovascular endpoints. Methods: Four population-based North Italian cohorts (three MONICA surveys and the PAMELA Study) and a large factory-based cohort (the Surveillance of Employees of the Municipality of Milan SEMM Study) were examined between 1986 and 1994 and in 1991-94. Participation rate was higher in the factory-based sample: 76% vs. 67%. Cardiovascular risk factors were investigated adopting standardized MONICA or MONICA-like procedures. The population-based cohorts account for 2997 women and 2964 men and the factory-based cohort for 3991 women and 2115 men, at baseline 25-65 year-old, CVD free and with valid data on CVD risk factors. All cohorts were followed-up to ascertain the first coronary or ischemic stroke event (fatal and non-fatal; all MONICA validated) up to the end of 2008, adopting the same standardized procedures including probabilistic record linkages to trace suspected events from hospital discharge records. We compared Kaplan-Meyer survival curves and used the log-rank test to assess differences between the two types of cohorts. To account for age, the study population was then stratified into three groups (<40, 40-49 and 50-64 years). Results: Among 25-64 year-old subjects free from CHD or stroke at baseline, in populations based cohorts in 14.9 median years of follow-up, we registered 236 fatal or not-fatal first CHD events and 76 first ischemic strokes. In the SEMM cohort in 14.1 median years of follow-up, the numbers of the above mentioned endpoints were 100 and 30, respectively. Unadjusted KM curves showed differences in survival between population-based and factory based cohorts, for cardiovascular death as well as for the incidence of coronary and stroke events. Likewise, the log-rank tests were statically significant. When we stratified for age classes, the log-rank tests were no more statistically significant in all age groups and for all the considered endpoints. Conclusions: When age is controlled for, the WHE for major CVD events, fatal and non-fatal, is negligible. This may be in part attributed to the increased proportion of white collars and the higher participation rates in more recently recruited factory-based samples.