The Mayo Lung Project (MLP) and non-small cell lung cancer (NSCLC): Does overdiagnosis (OD) explain improved survival?

Abstract

7181 Background: Screening for lung cancer (LC) is not recommended. While randomized trials consistently demonstrate significant survival advantages in populations randomized to screening, survival advantages have been interpreted as indicating that CXR screening leads to the OD of LC. The objective of this analysis is to consider whether MLP data are consistent with OD. Methods: In MLP, all participants had a negative prevalence screen, consisting of chest x-ray (CXR) and sputum cytology (SC). Subsequently, 4,607 men were randomized to an experimental group (EG), which underwent CXR and SC every 4 months, and 4,585 to a control group (CG). CG participants were advised to undergo annual CXR and SC. EG was screened for 6 yrs, followed by 3 yrs of observation, while CG was followed for 9 yrs. Results: After 9 yrs, 158 NSCLC were detected in EG and 115 in CG (RR=1.37, 95% CI: 1.08–1.73). In EG, 107 cases (68%) were screen-detected compared to 40 (35%) in CG (p<0.001). CXR detected 88% of all screen-detected cases. Resection was performed in 88 (56%) EG and 48 (42%) CG cases (p=0.027). Six yr survival was 39% in EG vs 17% in CG (p=0.0095). However, there were 86 LC deaths in EG and 76 in CG. Accordingly LC mortality was slightly higher in EG (RR 1.13; 95% CI: 0.83–1.53). The survival/mortality discrepancy occurred because NSCLC incidence was 37% higher in EG. These findings led to the hypothesis that OD was responsible for a spurious survival advantage. However, if screening detects pseudodisease, some NSCLC patients should be long-term survivors despite conservative treatment, and this was not observed. Six yr survival was 58% in 136 resected NSCLC and 0% in 137 unresected NSCLC (p<0.0001). Among 117 resected patients with screen-detected NSCLC, 6-yr survival was 60%, compared to 0% among 28 unresected patients (p<0.0001). Multivariate analysis indicated that resection was the only significant predictor of survival. Screen-detection was only important insofar at it predicted resection. Among NSCLC patients, LC was responsible for 89% of all deaths. Conclusions: OD does not explain survival/mortality discrepancies in MLP. Survival advantages reflect a benefit for screening, and support that CXR screening is effective. No significant financial relationships to disclose.