Importance of events per independent variable in proportional hazards regression analysis. II. Accuracy and precision of regression estimates

摘要: The analytical effect of the number events per variable (EPV) in a proportional hazards regression analysis was evaluated using Monte Carlo simulation techniques for data from randomized trial containing 673 patients and 252 deaths, which seven predictor variables had an original significance level p < 0.10. deaths 7 correspond to 36 analyzed full set. Five hundred simulated analyses were conducted these at EPVs 2, 5, 10, 15, 20, 25. For each simulation, random exponential survival time generated patients, results compared with their counterparts. As EPV decreased, coefficients became more biased relative true value; 90% confidence limits about values did not have coverage large sample properties hold variance estimates model, Z statistics used test lost validity under null hypothesis. Although single boundary avoiding problems is easy choose, value = 10 seems most prudent. Below this EPV, should be interpreted caution because statistical model may valid.