Background: Current eligibility criteria for lung cancer (LC) screening are derived from randomized controlled trials and largely based on age and smoking history. However, the individualised benefits of screening are highly variable and may be mediated by the presence of co-existing comorbid disease, including diabetes mellitus (DM). This study examines in detail screening outcomes for those reporting a prior diagnosis of DM.
Methods: This was a secondary analysis of 53,452 high-risk subjects from the National Lung Screening Trial (NLST), and compared outcomes following screening with computed tomography (CT) or chest x-ray (CXR) stratified according to DM status. Models of LC mortality were derived after adjustment and LC rate ratios (per 1000 person years), including 95% Confidence Intervals (95% CI), were examined according to screening arm and DM status.
Findings: Compared to those without DM, DM subjects (N=5,174, 9.7%) had a 2-fold greater baseline cardiovascular comorbidity (p<0.0001), 2-fold greater non-LC mortality (p<0.0001) and greater LC lethality (p=0.02), with more late-stage lung cancer (p=0.03). We found comparable stage shift and surgical rates, favouring those randomised to CT relative to CXR, for both DM and non-DM subgroups. However, we found no reduction in LC mortality for DM subjects favouring CT (2.2% vs 2.1% respectively, Rate ratio per 1000 person years (RR)=1.03, 95%CI 0.71-1.49, p=0.89), contrasting with non-DM subjects (RR=0.83, 95%CI 0.73-0.95, p=0.006)(p for interaction 0.28). In a modified Cox-Proportional Hazard model for dying of lung cancer by screening arm, adjusting for relevant co-variables, DM was associated with a hazard ratio (HR)=1.03, 95%CI=0.71-1.50, p=0.88) compared to non-DM (HR=0.82, 95%CI=0.72-0.94, p=0.003). LC mortality for those randomised to CT was greater for DM vs non-DM subjects (2.2% vs 1.6%, RR=1.35 (95% CI 1.02-1.79, p=0.033) but no different for CXR (2.1% vs 2.0%, RR=1.09 (95% CI 0.82-1.44, p=0.55).
Interpretation: The significant reduction in LC mortality favouring CT-based screening found in non-DM subjects was not observed in those reporting DM. While study design (under-powering), collider/confounder effects (bias) and newer treatment modalities remain possible limitations, the findings from this clinical trial data support simulation studies suggesting LC screening outcomes may be attenuated by comorbidity such as DM.