Bénédicte Delcoigne1, Edoardo Colzani2, Michaela Prochazka2, Giovanna Gagliardi3, Per Hall2, Michal Abrahamowicz4, Kamila Czene2, Marie Reilly2. 1. Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, P.O. Box 281, SE-17177 Stockholm, Sweden. Electronic address: benedicte.delcoigne@ki.se. 2. Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, P.O. Box 281, SE-17177 Stockholm, Sweden. 3. Section of Radiotherapy Physics and Engineering, Department of Medical Physics, Karolinska University Hospital, P.O. Box 260, Stockholm SE-17176, Sweden. 4. Department of Epidemiology and Biostatistics and Occupational Health, McGill University, Purvis Hall, 1020 Pine Avenue West, Montreal, Quebec H3A 1A1, Canada.
Abstract
OBJECTIVE: To demonstrate the advantage of using weighted Cox regression to analyze nested case-control data in overcoming limitations encountered with traditional conditional logistic regression. STUDY DESIGN AND SETTING: We analyzed data from 1,051 women who were sampled in a case-control study of lung cancer nested within a cohort of breast cancer patients. We investigated how lung cancer risk is associated with radiation therapy and modified by smoking, with both conditional logistic regression and weighted Cox regression models. RESULTS: In contrast to logistic regression, weighted Cox regression exploited the information regarding radiation dose received by each individual lung. The weighted method also mitigated a problem of overmatching apparent in the data and revealed that the risk of radiotherapy-associated lung cancer was modified by smoking (P = 0.026) with a hazard ratio of 4.09 (2.31, 7.24) in unexposed smokers and 8.63 (5.04, 14.79) in smokers receiving doses >13 Gy. The cumulative risk of lung cancer increased steadily with increasing radiotherapy dose in smokers, whereas no such effect was found in nonsmokers. CONCLUSION: The weighted Cox regression makes optimal and versatile use of the information in a nested case-control design, allowing dose-response analysis of exposure to paired organs and enabling the estimation of cumulative risk.
OBJECTIVE: To demonstrate the advantage of using weighted Cox regression to analyze nested case-control data in overcoming limitations encountered with traditional conditional logistic regression. STUDY DESIGN AND SETTING: We analyzed data from 1,051 women who were sampled in a case-control study of lung cancer nested within a cohort of breast cancerpatients. We investigated how lung cancer risk is associated with radiation therapy and modified by smoking, with both conditional logistic regression and weighted Cox regression models. RESULTS: In contrast to logistic regression, weighted Cox regression exploited the information regarding radiation dose received by each individual lung. The weighted method also mitigated a problem of overmatching apparent in the data and revealed that the risk of radiotherapy-associated lung cancer was modified by smoking (P = 0.026) with a hazard ratio of 4.09 (2.31, 7.24) in unexposed smokers and 8.63 (5.04, 14.79) in smokers receiving doses >13 Gy. The cumulative risk of lung cancer increased steadily with increasing radiotherapy dose in smokers, whereas no such effect was found in nonsmokers. CONCLUSION: The weighted Cox regression makes optimal and versatile use of the information in a nested case-control design, allowing dose-response analysis of exposure to paired organs and enabling the estimation of cumulative risk.