Risk prediction models for lymph node metastasis in early gastric cancer patients: a systematic review and meta-analysis.

BACKGROUND

The number of risk prediction models for lymph node metastasis in early gastric cancer is increasing, but the quality and applicability of these models in clinical practice and future research remain unknown.

OBJECTIVE

To systematically review studies published on prediction models for the risk of lymph node metastasis in early gastric cancer patients.

DESIGN

Systematic review and meta-analysis of observational studies.

METHODS

A search was conducted in databases. Data from selected studies were extracted, including study design, data sources, outcome definitions, sample size, predictive factors, model development, and performance indicators. The risk of bias in prediction models was assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST) checklist.

RESULTS

Fifty articles were included in this meta-analysis. Most studies used Logistic Regression (LR) to establish risk prediction models. The training model's overall c-statistic was 0.85 (95% CI (0.81-0.89)), whereas the validation model's overall c-statistic was 0.82 (95% CI (0.80-0.83)). The overall pooled accuracy rate for the training group model was 0.80 [95% CI (0.72-0.87)], and the overall accuracy rate for model validation was 0.71 [95% CI (0.61-0.79)]. Tumor size was the most common risk predictive factor. All studies had a high risk of bias, primarily due to inappropriate data sources.

CONCLUSION

Based on the results of the PROBAST analysis, it was determined that all of the studies were highly biased. Models with bigger samples, more stringent research methods, inclusion of multicenter samples, and external validations should be the focus of future studies.