Follow-up-dependent calendar-period associations with cardiac mortality after lung cancer radiotherapy: a competing-risks SEER analysis.

OBJECTIVES: Calendar-period comparisons of cardiac mortality after thoracic radiotherapy can be distorted by competing cancer death and by unequal follow-up, because more recently diagnosed patients are observed for less time. We aimed to determine whether a reported recent-era decline in cardiac mortality after lung radiotherapy reflects changing practice or analytic artifact. METHODS: We identified 56,241 patients with lung and bronchus cancer diagnosed in 2009-2021 in the SEER database (28,246 beam radiotherapy without surgery; 27,995 surgery without radiotherapy; chemotherapy-recorded patients excluded). Cardiac death was the event of interest and lung cancer death the competing event. In each cohort we estimated Fine-Gray subdistribution hazard ratios (sHRs) for diagnosis period under (i) full follow-up, (ii) full follow-up restricted to diagnoses through 2018, and (iii) a fixed 36-month window for that same 2009-2018 cohort. We additionally estimated period sHRs across fixed windows of 12-60 months, repeated all models treating other-cause deaths as competing events and using cause-specific Cox regression, and tested the proportional-subdistribution-hazards assumption. RESULTS: In the radiotherapy cohort, more recent diagnosis was associated with a lower subdistribution hazard of cardiac death under full follow-up (sHR 0.761; 95 % CI 0.678-0.854; p<0.001), but the association was not significant at any equalized follow-up window (12-month sHR 0.864; 24-month 0.918; 36-month 0.994 [0.853-1.157]; 48-month 0.944; 60-month 1.026; all p≥0.14), and the period effect was non-proportional over follow-up time (p=0.003). The attenuation persisted when other-cause deaths were treated as competing events (36-month sHR 1.000) and under cause-specific regression (36-month HR 0.898; p=0.167), and median follow-up differed markedly by period (108 vs. 38 months). By contrast, the surgery cohort retained a significantly lower recent-period estimate through 36 months under the same matched-window analysis (restricted-full sHR 0.734; 36-month sHR 0.770; p=0.017), attenuating only at 48-60 months. CONCLUSIONS: A recent-period reduction in the surgery cohort persisted through a 36-month matched window, showing that matched-window analysis does not mechanically null a calendar-period signal; by contrast, the radiotherapy cohort showed no such association at any comparable follow-up. Registry data therefore do not support attributing a cardiac-mortality decline to changes in radiotherapy practice, and temporal registry comparisons must use comparable follow-up windows, account for competing death, and verify proportionality before such attribution.