A study on Statistical Method for Controlling the Effect of Intermediate Events: Application to the Control of the Healthy Worker Effect.
- Author:
Chung Mo NAM
1
;
Jinheum KIM
;
Dae Ryong KANG
;
Yeon Soon AHN
;
Hoo Yeon LEE
;
Dae Hee LEE
Author Information
1. Department of Preventive Medicine and Public Health, Yonsei University, Korea. cmnam@yumc.yonsei.ac.kr
- Publication Type:Original Article
- Keywords:
Healthy worker effect;
Intermediate;
Length bias sampling;
Simulation;
Level
- MeSH:
Bias (Epidemiology);
Confounding Factors (Epidemiology);
Employment;
Epidemiology;
Healthy Worker Effect*;
Mortality;
Proportional Hazards Models
- From:Korean Journal of Epidemiology
2002;24(1):7-16
- CountryRepublic of Korea
- Language:Korean
-
Abstract:
PURPOSE: The healthy worker effect is an important issue in occupational epidemiology. This study was conducted to propose a new method to test the relation between exposure and mortality in the presence of the healthy worker effect. METHODS: In this study, the healthy worker hire effect was assumed to operate as a confounding variable of health status at the beginning of employment and healthy worker survival effect as a confounding and intermediate variable of employment status. In addition, the proposed method reflects the length bias sampling caused by changing of an employment status. Simulation studies were also carried out to compare the proposed method with Cox's time dependent covariates models . RESULTS: The theoretical development of the healthy worker survival effect is based on the result that an observation with change of an employment status requires that the survival time without intermediate event exceeds the waiting time for the intermediate event. According to our simulation studies, both the proposed method and Cox's time dependent covariates model which includes the change of employment status as time dependent covariates seem to be satisfactory at 5% significance level. However, Cox's time dependent covariates models without or with the change of employment status as time fixed covariate are unsatisfactory. The proposed test is superior in power to tests based on Cox's model. CONCLUSIONS: The healthy worker effect may not be controlled by classical Cox's proportional hazards models. The proposed method performed well in the presence of healthy worker effect in terms of level and power
