Construction and validation of a prognostic nomogram for locally recurrent rectal cancer: a population-based study.

Abstract

Postoperative recurrence was a life-threatening condition for patients with rectal cancer. Due to the heterogeneity of locally recurrent rectal cancer (LRRC) and controversy of the optimal treatment for patients, it was difficult to predict the prognosis of LRRC. This study aimed to develop and validate a nomogram that could accurately predict the survival probability of LRRC. Patients diagnosed with LRRC between 2004 and 2019 from the Surveillance, Epidemiology, and End Results (SEER) database were included in the analysis. Multiple imputations with chained equations were used for missing values. These patients were further randomized into training set and testing set. Cox regression was used for univariate and multivariate analysis. Potential predictors were screened by the least absolute shrinkage and selection operator (LASSO). The Cox hazards regression model was constructed and it was visualized by nomogram. C-index, calibration curve, and decision curve were used to evaluate the model's predictive ability. Then X-tile was used to calculate the optimal cut-off values for all patients and the cohort was divided into three groups. A total of 744 LRRC patients were enrolled and allocated to the training set (n=503) and the testing set (n=241). Cox regression analysis of the training set yielded meaningfully clinicopathological variables. A survival nomogram was created based on the identification of ten clinicopathological features in the LASSO regression analyses of the training set. The C-index of 3-, 5-year survival probabilities were 0.756, 0.747 in training set, and 0.719, 0.726 in testing set, respectively. The calibration curve and decision curve both demonstrated the satisfactory performance of the nomogram for prognosis prediction. Moreover, the prognosis of LRRC could be well distinguished according to the grouping of risk scores (P<0.001 in three groups). This nomogram was the first prediction model to preliminarily evaluate the survival of LRRC patients, which could provide more accurate and efficient treatment in clinical practice.