Application of machine learning algorithm in distant metastasis prediction of T1 gastric cancer

Machine Learning

Patients with GC with distant metastases have a poor prognosis, with a 5-year survival rate of less than 5% and a median survival of 11-18 months.FiveExtensive evidence indicates that approximately 40% of patients have distant metastases at initial diagnosis of GC, and the incidence increases as the tumor progresses.15,16Due to the high 5-year survival rate of T1 patients, many scholars have ignored the possibility of distant metastases in T1 patients.9,17A recent study showed an 8.17% chance of distant metastasis in stage T1 patients.18Therefore, risk factors and prognosis for distant metastasis in T1 gastric cancer should be investigated. Significantly, this is the first study to build a model for predicting distant metastasis in stage T1 gastric cancer by machine learning and to analyze its survival and prognosis.

Previous studies have demonstrated that distant metastases are rare in T1 GC, indicating a favorable prognosis for most patients with early gastric cancer.9Strikingly, the present study found an elevated risk of distant metastases of 11.64% in patients with T1 GC. Therefore, there is an urgent need to determine whether T1 patients have distant metastases upon initial diagnosis. While conventional imaging studies (such as magnetic resonance imaging and computed tomography) can detect significant diffuse lesions, positron emission tomography (Positron Emission Tomography) is used to examine distant metastases of GC, especially in detecting micrometastases. It’s a more reliable method.However, limited by its effectiveness and real cost19Establishing a simple and effective predictive model can therefore help clinicians identify high-risk patients for further testing and diagnosis.

Machine learning algorithms are a new class of techniques that can accurately process raw data, analyze relationships between important data, and make accurate decisions. One of the best features of machine learning algorithms is their excellent performance in predicting outcomes on large databases. This is an advantage over traditional regression techniques.20In this study, we analyzed predictive models established by seven ML algorithms, including logistic regression (LR), random forest (RF), LASSO, support vector machines (SVM), k-nearest neighbors (KNN), and naive Bayesian models. and compared. (NBC), and artificial neural networks (ANN). First, the training set was used to build prediction models, AUC, sensitivity, specificity, F1 score, and accuracy were used to evaluate the efficacy values ​​of seven prediction models, and finally the RF model was the best predictive efficacy (AUC: 0.941, precision: 0.917, recall: 0.841, specificity: 0.927, F1 score: 0.877). Further validation of the results using the test set showed that the RF model was the best predictive model for predicting DM in stage T1 GC, with the best predictive efficacy (AUC = 0.825). rice field. The ability of the RF model to accurately predict DM in stage T1 gastric cancer was also confirmed by an external validation set (AUC = 0.750). RF is likely one of the most widely used and accurate machine learning models in applied clinical research. Perhaps because RF models use more advanced classification decisions and different weight ratios compared to other models, Random Forest models outperform other algorithms when dealing with data with a large number of features and highly nonlinear data. reported to be better than21,22This study confirmed that the random forest prediction model can accurately predict the high-risk group with distant metastases in T1 patients. This will facilitate further clinical examination of this population and help develop better diagnostic and therapeutic strategies.

In this study, the six most important characteristics of age, T stage, N stage, tumor size, grade, and tumor location were included in the final RF prediction model. As a result, younger patients (< 60 歳) の DM の割合は、高齢患者 (> 60 years old). Previous studies have reported a high rate of lymph node metastasis in young GC patients.13,23,24The prevalence of lymph node involvement in young patients may be one reason for distant metastasis. Recently, accumulating research has found that tumor biology plays an important role in disease development. This may be closely related to the occurrence and development of distant metastasis.twenty fiveAdditional studies have shown that tumor size, depth of invasion, and lymph node metastasis are significantly associated with advanced gastric cancer.26Nevertheless, our study found that the N and T stages are closely associated with distant metastasis. Interestingly, the rate of distant metastases in stage T1a patients was significantly higher than in stage T1b patients. This may be attributed to lymph node metastasis occurring first in submucosal patients and hematogenous metastasis occurring later during invasion into deeper layers in mucosal patients. According to Japanese guidelines for the treatment of GC, patients with tumors larger than her 2 cm are at significantly higher risk of metastases and should undergo radical resection for clean removal.12Furthermore, although the risk of distant metastases was found to increase significantly with tumor enlargement, this risk for patients with tumor sizes greater than 5 cm was 8–9 times lower than that for patients with tumor sizes less than 2 cm. It was double. In our study, tumor site was one of the independent risk factors affecting distant metastasis in T1 GC patients. Fundus tumors are prone to distant metastasis, which may be attributed to their abundant blood vessels. Abundant blood vessels are closely associated with hematogenous metastasis. Furthermore, our results showed that patients with moderately and poorly differentiated GCs were more likely to develop distant metastases than undifferentiated and well-differentiated patients. It has the ability to grow faster. This seems to deviate from our previous understanding and requires further validation.

Subsequently, we also performed a prognostic survival analysis for patients with distant metastasis. Consequently, surgery (HR = 3.620, 95% CI 2.164–6.065) and adjuvant chemotherapy (HR = 2.637, 95% CI 2.067–3.365) were independent risk factors for survival and prognosis in patients with T1 distant metastases. became clear.This is consistent with previous research27Surgery of the primary tumor may reduce the potential burden of immunosuppressive tumors and eliminate sources of further metastasis.28Thus, for patients with T1 distant metastases, aggressive surgery combined with adjuvant chemotherapy significantly improves patient prognosis and increases survival.

This study is the first to use ML algorithms to predict DM in stage T1 GC, establishing an accurate predictive model to help identify individuals at high risk for DM in the early stages of the clinic. increase. However, this study still has some limitations. First, as a retrospective study, her sample size of 2698 patients from 2010 to 2017 was relatively small. Second, because of the finite number of variables included in our study and the lack of other similar potential risk factors such as tumor markers, nutritional index and inflammatory index, the model with more variables was further developed. Using it can improve prediction accuracy.

In conclusion, we built and validated a predictive model for DM in T1 GC patients via ML algorithms. The RF model has the highest predictive efficiency and can accurately screen high-risk groups, which is useful for further clinical metastasis screening. On the other hand, our study also found that aggressive surgery and adjuvant chemotherapy can improve survival in DM patients.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *