27-year mortality in the Western Collaborative Group Study: construction of risk groups by recursive partitioning

The relationship of selected biological and behavioral characteristics measured at baseline examination to 27-year mortality due to coronary heart disease (CHD), cancers of all sites, and total mortality in the 3154 men that form the Western Collaborative Group Study was investigated using tree-structured survival analysis or recursive partitioning (RP). Intake (1960-61) characteristics included in the present analyses were age, serum cholesterol, systolic blood pressure (SBP), cigarette smoking, body mass index (BMI), Type A/B behavior, and behavioral hostility. Tree-structured survival analysis for CHD mortality partitioned the cohort into six groups and identified five groups with distinct survival experience. Exceptionally high CHD mortality rates (17.3 and 14.6 per thousand) were experienced by 89 older men with elevated hostility ratings and SBP less than or equal to 150, and 238 men whose initial SBP was greater than 150 mmHg. Younger men (age less than or equal to 48) with SBP less than or equal to 150 and with serum cholesterol levels greater than 227 had a death rate of 4.8 per thousand, compared with a rate of 1.7 in similar men with lower cholesterol levels. Applied to 27-year cancer mortality, the RP algorithm partitioned the cohort into four distinct survival groups. Younger (age less than 45) Type B men had superior survival compared with Type A men of similar ages, and the proportion of ever cigarette smokers in these two groups was not statistically different. The results obtained by tree-structured survival analyses were compared with results obtained by Cox regression survival analyses.