Related Experiment Video
Updated: May 9, 2026
Competing-Risk Nomogram for Predicting Cancer-Specific Survival in Multiple Primary Colorectal Cancer Patients after Surgery
Published on: September 27, 2024
Local recurrence after surgery for non-small cell lung cancer: a recursive partitioning analysis of
Chris R Kelsey1, Kristin A Higgins, Bercedis L Peterson
1Department of Radiation Oncology, Duke Cancer Institute, Durham, NC.
Objective:
To define subgroups at high risk of local recurrence (LR) after surgery for non-small cell lung cancer using a recursive partitioning analysis (RPA).
Methods:
This Institutional Review Board-approved study included patients who underwent upfront surgery for I-IIIA non-small cell lung cancer at Duke Cancer Institute (primary set) or at other participating institutions (validation set). The 2 data sets were analyzed separately and identically. Disease recurrence at the surgical margin, ipsilateral hilum, and/or mediastinum was considered an LR. Recursive partitioning was used to build regression trees for the prediction of local recurrence-free survival (LRFS) from standard clinical and pathological factors. LRFS distributions were estimated with the Kaplan-Meier method.
Results:
The 1411 patients in the primary set had a 5-year LRFS rate of 77% (95% confidence interval [CI], 0.74-0.81), and the 889 patients in the validation set had a 5-year LRFS rate of 76% (95% CI, 0.72-0.80). The RPA of the primary data set identified 3 terminal nodes based on stage and histology. These nodes and their 5-year LRFS rates were as follows: (1) stage I/adenocarcinoma, 87% (95% CI, 0.83-0.90); (2) stage I/squamous or large cell, 72% (95% CI, 0.65-0.79); and (3) stage II-IIIA, 62% (95% CI, 0.55-0.69). The validation RPA identified 3 terminal nodes based on lymphovascular invasion (LVI) and stage: (1) no LVI/stage IA, 82% (95% CI, 0.76-0.88); (2) no LVI/stage IB-IIIA, 73% (95% CI, 0.69-0.80); and (3) LVI, 58% (95% CI, 0.47-0.69).
Conclusions:
The risk of LR was similar in the primary and validation patient data sets. There was discordance between the 2 data sets regarding the clinical factors that best segregate patients into risk groups.
