The Mayo Lung Cohort: A Regression Analysis Focusing on Lung Cancer Incidence and Mortality

Publisher Website
Google Scholar
Abstract
PURPOSE: The Mayo Lung Project has been interpreted as negative because it failed to demonstrate a significant mortality reduction among those randomized to chest x-ray and cytology. In contrast, survival suggests that screening is highly effective. This report was undertaken to analyze the trial as a closed cohort study, in an effort to identify predictors of lung cancer incidence and mortality, and to determine whether survival or mortality was unbiased. PATIENTS AND METHODS: The Mayo Lung Cohort comprised all 9,192 randomized individuals. Cox proportional hazards regression was used both to determine predictors of incidence and mortality in the population and to identify predictors of mortality among cases. Survival analyses using intent-to-treat principles and measuring survival from randomization were used to evaluate length bias and lead-time bias. Multivariate Cox regression was used to investigate the extent to which the data are consistent with overdiagnosis. RESULTS: Cox regression demonstrates that, in addition to age and smoking, randomization to screening predicted increased lung cancer incidence (hazard ratio, 1.30; 95% confidence interval [CI], 1.06 to 1.60). Predictors of mortality were similar, except randomization to screening was not significant (hazard ratio, 1.06; 95% CI, 0.83 to 1.37). Among cases, survival was significantly superior in the experimental population. Higher incidence in the experimental group accounts for the mortality/survival discrepancy. Both lead-time and length biases can be excluded, because survival from randomization was superior in the experimental population. Overdiagnosis is eliminated because resection was the only significant multivariate predictor of survival. Overall, 50% of resected and 0% of unresected cases were cured. CONCLUSION: Survival was superior in the screened population, and this advantage was not attributable to lead-time bias, length bias, or overdiagnosis bias. Mortality was biased, because incidence differences confounded the ability of mortality to reflect the true effect of screening. Indeed, survival provided an unbiased surrogate for cure in the Mayo Lung Cohort.

This publication has 19 references indexed in Scilit: