Machine learning-based survival prediction nomogram for postoperative parotid mucoepidermoid carcinoma

Machine learning-based survival prediction nomogram for postoperative parotid mucoepidermoid carcinoma

Screening and traits of the sufferersThis examine examined 882 sufferers with stage I–IVA P-MEC, who met the inclusion–exclusion standards, from the SEER database between 2004 and 2015. Figure 1 illustrates the affected person choice course of, whereas Table 1 summarizes sufferers’ demographic and clinicopathological traits. The lymph node ratio (LNR) cut-off was decided utilizing X-tile evaluation, with a resultant cut-off of 1.15%. The median (95% CI) follow-up time was 99 (92–105) months, and the median (IQR) age at prognosis was 52 (37–66) years. A majority of the sufferers had been white (661, 74.9%), with most tumors being grade II (396, 44.9%), stage I (353, 40%), T1-stage (381, 43.2%), N0-stage (685, 77.7%), and LNR0 (686, 77.8%) in accordance with the AJCC sixth stage. All variables, besides for chemotherapy (94.2% vs 5.8%), had proportions exceeding 10%. The examine encompassed 12 variables, together with age, gender, grade, stage, tumor (T) stage, node (N) stage, radiation, chemotherapy, laterality, marriage, and LNR. Nine elements—age, gender, grade, stage, T stage, N stage, radiation, chemotherapy, and LNR—had been chosen based mostly on univariate Cox regression. Multivariate Cox regression revealed that 4 elements (age, grade, T stage, and chemotherapy) had been impartial danger elements, every with P-values lower than 0.05. In the multivariate evaluation, people aged 60–70 years (HR = 5.936, 95% CI = 3.016–11.681, P < 0.001), these over 70 years previous (HR = 11.962, 95% CI = 6.303–22.703, P < 0.001), Grade III (HR = 2.324, 95% CI = 1.235–4.375, P = 0.009), Grade IV (HR = 3.148, 95% CI = 1.710–5.795, P < 0.001), T2 (HR = 3.162, 95% CI = 1.059–9.440, P = 0.039), T3 (HR = 4.300, 95% CI = 1.501–12.316, P = 0.007), T4 (HR = 4.414, 95% CI = 1.439–13.535, P = 0.009), and chemotherapy (HR = 1.721, 95% CI = 1.096–2.703, P = 0.018) emerged as impartial danger elements for total survival (OS). Nevertheless, radiation(HR = 0.750, 95% CI = 0.525–1.072, P = 0.114), LNR (HR = 0.868, 95% CI = 0.114–6.602, P = 0.891), and different variables demonstrated no prognostic worth (Table 2).Table 1 Demographic and medical traits of sufferers with P-MEC.Table 2 Univariate and multivariate analyses for OS in sufferers with P-MEC.Variable choiceFigure 2A shows the connection between the LASSO coefficients and the regularization parameter, lambda (λ), and demonstrates the variable choice course of and the impact of λ on the coefficients. The “lambda.min” worth, which represents the lambda worth comparable to the minimal probability deviation or the best C-index, was utilized for choosing tuning parameters in LASSO regression. Another vertical line was “lambda.1se,” which corresponds to essentially the most regularized mannequin inside one normal error of the minimal (Fig. 2B). The “λ.min” (λ = 0.0050724) was chosen for the very best predictive efficiency. A ten-fold cross-validation was employed. Ten variables had been chosen via the LASSO regression algorithm, together with age, gender, grade, T stage, N stage, radiation, chemotherapy, laterality, marriage, and LNR. Employing the adjusted R-squared most of the BSR, we chosen eight variables: age, grade, stage, T stage, N stage, radiation, chemotherapy, and marriage(Fig. 3). In the RF mannequin and XGBoost, we independently extracted the highest 10 variables, excluding laterality, radiation (RF), and LNR (XGBoost) (Fig. 4). We assessed the important thing efficiency of machine studying and conventional statistics utilizing AUC and AIC. Multivariate Cox stepwise backward regression reconfirmation recognized LASSO, BSR, and XGBoost as the very best of the 5 screening strategies based mostly on each AUC (AUC = 88.4) and AIC (AIC = 2118.9) standards (Table 3).Figure 2Predictor Screening: the least absolute shrinkage and choice operator (LASSO) regression and fivefold cross-validation.Figure 3Predictor Screening: A SHAP plot and a function significance plot are visualizations used to interpret XGBoost mannequin outcomes.Figure 4Predictor Screening: (A) Random Forest significance plot; (B) Best Subset Regression (BSR), it chosen the very best subset of predictor variables to precisely mannequin a response variable.Table 3 Screening outcomes of 5 strategies for figuring out predictors.Nomogram constructionConsequently, we constructed a nomogram with seven variables from the three algorithms (LASSO, BSR, and XGBoost), together with age, grade, tumor stage, node stage, chemotherapy, radiation, and marriage. We developed an OS-nomogram able to predicting a affected person’s 3-, 5-, and 10-year OS charges utilizing these variables (Fig. 5). By changing medical, pathological, and therapeutic elements into factors, the nomogram precisely predicted OS. The complete danger level rating, calculated by summing all factors, considerably correlated with 3-, 5-, and 10-year OS. We utilized a 5-year ROC curve to find out the optimum danger rating cut-off level. Kaplan–Meier curves revealed that low-risk group sufferers (danger rating < 80.29) had higher survival prognosis in comparison with high-risk group sufferers (danger rating ≥ 80.29, log-rank check, P < 0.001) (Fig. S1).Figure 5A survival nomogram for predicting total survival (OS) for sufferers with P-MEC. (1) When utilizing the nomogram, seven predictors had been quantified as “level” based mostly on patient-specific elements after which the sum of the “level” corresponded to the “complete level” beneath, which corresponded to the three, 5, 10 yr OS ; (2) The optimum cut-off complete level was 80.29 (the median of sufferers’ level), which divided the sufferers into high-risk group and low-risk group.Predictive capability analysisWe evaluated the predictive capability of our nomogram by setting up time-dependent receiver working attribute (ROC) curves at 3, 5, and 10 years. The ROC curves demonstrated glorious discriminative capability of our mannequin, with areas beneath the curves (AUCs) of 86.9 (95% CI = 83.3–90.6), 88.4 (95% CI = 83.5–91.4), and 87.7 (95% CI = 84.1–91.3) (Fig. 6). This signifies that our mannequin has excessive accuracy in predicting total survival in parotid MEC sufferers.Figure 6(A–C) The calibration curves. The calibration curves of the nomogram predicting (A) 3-years, (B) 5-years, and (C) 10-years OS. (D–F) Time dependent ROC curve. (D) ROC curves for 3-year, (E) 5-year, and (F) 10-year total survival charges. (G–I) Decision curve evaluation (DCA) plot. (G) DCA plot for 3-year, (H) 5-year, and (I) 10-year total survival charges.We additionally carried out 1000 bootstrap resampling analyses on the dataset and generate calibration plots for the prediction mannequin. The calibration plots confirmed that the curves intently aligned with the 45-degree line, indicating a well-calibrated mannequin in sensible use (Fig. 6). Furthermore, the 1000 bootstrap resamplings indicated good concordance between precise and predicted values in each the coaching and validation datasets, as evidenced by C-index (3-year, 0.8499, 0.775–0.914; 5-year 0.8557, 0.793–0.911; 10-year, 0.8375, 0.772–0.897) and AUC (3-year, 0.8670, 95 CI% = 0.787–0.935; 5-year, 0.8879, 95 CI% = 0.82–0.945; 10-year, 0.8767, 95 CI% = 0.792–0.947). (Fig. 7). These outcomes additional help the reliability and accuracy of our prediction mannequin.Figure 7This determine presents a bootstrap evaluation of a dataset, displaying the 3-year and 5-year AUC and C-index values. The evaluation was carried out utilizing 1000 bootstrap replicates. The determine demonstrates the accuracy and predictive energy of the mannequin for the desired time intervals.Clinical utility analysisTo decide the medical utility of our prediction mannequin, we utilized the choice curve evaluation (DCA) plot. The DCA plot illustrates the online advantage of the prediction mannequin throughout a spectrum of threshold possibilities. Our mannequin demonstrates medical utility, as evidenced by its web profit curve lies above each two strains throughout the vary of threshold possibilities (Fig. 6). This means that our prediction mannequin is simpler than TNM stage or grade and may support in making medical choices for P-MEC sufferers.In abstract, our nomogram exhibited glorious predictive capability and calibration, in addition to medical utility, indicating its potential usefulness in medical follow.
https://www.nature.com/articles/s41598-024-58329-8

Recommended For You