Prognostic insights after surgery for advances in understanding signet ring cell gastric cancer: a machine learning approach.

Background: Signet ring cell (SRC) gastric carcinoma is traditionally associated with a poor prognosis. However, the literature has presented contradictory results. Linear models are the standard statistical tools typically used to study these conditions. However, machine learning (ML) models have the potential to replace or even outperform linear models in terms of predictive performance.

Methods: This study analyzed 608 patients diagnosed with gastric cancer at our institution. The analysis compared traditional linear models and ML models. Variables examined included demographic data, presence of an SRC component, lymph nodes (LNs) resected and affected (ratio), stage of the disease, body mass index, pathologic features, type of surgery, tumor location, and carcinoembryonic antigen levels to evaluate their influence on 5-year mortality and 2-year recurrence rates.

Results: SRC carcinoma was associated with poorer prognosis in terms of 5-year overall survival than non-SRC carcinoma. In addition, SRC exhibited higher rates of LN metastasis and a higher LN ratio (resected/affected) and was more prevalent in younger patients (<65 years). However, SRC was not an independent factor in the multivariate analysis. Linear models showed worse predictions for 5-year mortality and 2-year recurrence than ML models. The ML models did not consider the presence of the SRC component as an important variable.

Conclusion: SRC gastric carcinoma continues to present an uncertain prognosis. ML models can evaluate prognosis more accurately than traditional linear models. Large-scale studies using ML algorithms are necessary to elucidate the predictive potential of such models.