BACKGROUND & AIMS:

Most patients with pedunculated T1 colorectal tumors referred for surgery are not found to have lymph node metastases, and were therefore unnecessarily placed at risk for surgery-associated complications. We aimed to identify histologic factors associated with need for surgery in patients with pedunculated T1 colorectal tumors.

METHODS:

We performed a cohort-nested matched case-control study of 708 patients diagnosed with pedunculated T1 colorectal tumors at 13 hospitals in The Netherlands, from January 1, 2000 through December 31, 2014, followed for a median of 44 months (interquartile range, 20-80 months). We identified 37 patients (5.2%) who required surgery (due to lymph node, intramural, or distant metastases). These patients were matched with patients with pedunculated T1 colorectal tumors without a need for surgery (no metastases, controls, n = 111). Blinded pathologists analyzed specimens from each tumor, stained with H&E. We evaluated associations between histologic factors and patient need for surgery using univariable conditional logistic regression analysis. We used multivariable least absolute shrinkage and selection operator (LASSO; an online version of the LASSO model is available at: http://t1crc.com/calculator/) regression to develop models for identification of patients with tumors requiring surgery, and tested the accuracy of our model by projecting our case-control data toward the entire cohort (708 patients). We compared our model with previously developed strategies to identify high-risk tumors: conventional model 1 (based on poor differentiation, lymphovascular invasion, or Haggitt level 4) and conventional model 2 (based on poor differentiation, lymphovascular invasion, Haggitt level 4, or tumor budding).

RESULTS:

We identified 5 histologic factors that differentiated cases from controls: lymphovascular invasion, Haggitt level 4 invasion, muscularis mucosae type B (incompletely or completely disrupted), poorly differentiated clusters and tumor budding, which identified patients who required surgery with an area under the curve (AUC) value of 0.83 (95% confidence interval, 0.76-0.90). When we used a clinically plausible predicted probability threshold of ≥4.0%, 67.5% (478 of 708) of patients were predicted to not need surgery. This threshold identified patients who required surgery with 83.8% sensitivity (95% confidence interval, 68.0%-93.8%) and 70.3% specificity (95% confidence interval, 60.9%-78.6%). Conventional models 1 and 2 identified patients who required surgery with lower AUC values (AUC, 0.67; 95% CI, 0.60-0.74; P = .002 and AUC, 0.64; 95% CI, 0.58-0.70; P < .001, respectively) than our LASSO model. When we applied our LASSO model with a predicted probability threshold of ≥4.0%, the percentage of missed cases (tumors mistakenly assigned as low risk) was comparable (6 of 478 [1.3%]) to that of conventional model 1 (4 of 307 [1.3%]) and conventional model 2 (3 of 244 [1.2%]). However, the percentage of patients referred for surgery based on our LASSO model was much lower (32.5%, n = 230) than that for conventional model 1 (56.6%, n = 401) or conventional model 2 (65.5%, n = 464).

CONCLUSIONS:

In a cohort-nested matched case-control study of 708 patients with pedunculated T1 colorectal carcinomas, we developed a model based on histologic features of tumors that identifies patients who require surgery (due to high risk of metastasis) with greater accuracy than previous models. Our model might be used to identify patients most likely to benefit from adjuvant surgery.