LDCT features and timing of LDCT-detected recurrence after radical lung cancer surgery

To investigate the low-dose computed tomography (LDCT) imaging characteristics of LDCT-detected postoperative recurrence after radical lung cancer resection, analyze their spatiotemporal distribution patterns, and identify associated clinicopathological risk factors. The primary estimand is the distribution and imaging appearance of recurrence first identified during institutional LDCT surveillance, not the overall incidence of recurrence after curative lung cancer surgery. A retrospective cohort study. Electronic medical records and the institutional Picture Archiving and Communication System (PACS) from a tertiary hospital database between January 2018 and December 2023. This study analyzed a cohort of 150 patients who underwent curative lung cancer resection and completed at least 2 years of postoperative LDCT follow-up, with first recurrence detected by LDCT (isolated brain or bone recurrences detected by other modalities were excluded). This is therefore a cohort of LDCT-surveillance completers with LDCT-detected first recurrence, and the findings describe recurrence patterns within this specific cohort rather than estimating overall post-resection recurrence. Patients were divided into recurrence (n=45) and non-recurrence (n=105) groups. LDCT features (95th percentile CT value, mean CT value, volume), recurrence timing, and patterns (locoregional vs. distant) were analyzed. Fine-Gray competing-risk models were used to estimate cumulative incidence of recurrence, and penalized ridge regression with bootstrap internal validation was used to identify predictors of recurrence. Among 45 recurrence patients, most recurrences occurred between 3 and 12 months post-surgery (55.55%). At the patient level, distant metastasis alone (25/45, 55.6%) was more common than locoregional recurrence alone (17/45, 37.8%), with synchronous locoregional and distant recurrence in 3/45 (6.7%). Recurrent lesions predominantly presented as solid nodules with lobulated margins. In the recurrence group, the 95th percentile CT value was significantly higher (68.2 ± 18.7 HU vs. 41.5 ± 8.3 HU, P < 0.001), while mean CT value was lower (368.41 ± 90.17 HU vs. 572.58 ± 165.42 HU, P < 0.001), and nodule volume was larger (2.08 ± 0.52 cm³ vs. 1.47 ± 0.35 cm³, P < 0.001) compared with the non-recurrence group. In Fine-Gray models, the 48-month cumulative incidence of recurrence was 30.2% (95% CI: 22.1%–38.9%), with death without recurrence as a competing event. Higher 95th percentile CT value (sHR = 2.98, 95%CI:1.62–5.48), larger nodule volume (sHR = 3.42, 95%CI:1.72–6.81), preoperative lymph node metastasis (sHR = 2.67, 95%CI:1.48–4.82), and advanced TNM stage (sHR = 2.38, 95%CI:1.39–4.07) were independent predictors of recurrence (all P < 0.05). The optimism-corrected concordance index (C-index) for the Fine-Gray model was 0.815 (95% CI: 0.75–0.87) after 1000 bootstrap resamples. Postoperative lung cancer recurrence shows distinct LDCT features and temporal clustering within the first 3 years in this selected cohort of LDCT-surveillance completers. Higher preoperative nodal metastasis burden, advanced TNM stage, and poor differentiation are independently associated with recurrence. LDCT is effective for detecting intrathoracic and some intra-abdominal (e.g., adrenal, liver) recurrences, but its inherent limitation in visualizing brain and bone metastases must be recognized. These findings support the hypothesis that risk-stratified, time-adapted surveillance strategies merit prospective evaluation, but they should not be used as prescriptive clinical recommendations. Symptom- or risk-triggered dedicated imaging of the brain and spine should be considered for detecting metastases outside the chest LDCT field; in patients with suspected or identified vertebral disease, assessment should follow SINS/NOMS-style multidisciplinary evaluation rather than routine dedicated spine imaging being inferred from this cohort. Not applicable. This retrospective study utilized existing clinical data without direct patient or public involvement in the design or conduct of the research.