Establishment and validation of a nomogram for predicting IDH-wildtype glioblastomas in nonenhancing adult-type diffuse gliomas

Abstract

Background: Approximately 8.94%-44.44% of nonenhancing adult-type diffuse gliomas are identified as glioblastomas. Our purpose is to develop a nomogram that can predict glioblastomas from nonenhancing adult-type diffuse gliomas.

Methods: Nonenhancing adult-type diffuse gliomas were collected from Beijing Tiantan Hospital and TCIA public database. Univariate and multivariate logistic regression were performed to screen features on the training set. The features with P < .05 in multivariate logistic regression were used to establish the prediction model. The testing and validation sets were used to test the model.

Results: A total of 557 and 67 nonenhancing adult-type diffuse gliomas were collected from Beijing Tiantan Hospital and TCIA, respectively. The T2-FLAIR mismatch sign exhibited 100% specificity but low sensitivity (<30%) in ruling out glioblastoma. Age, tumor location, rADC_(kurtosis), and rADC_(median) were identified as independent predictors and employed for developing the prediction model. The AUC of the model was 0.901, 0.861, and 0.945 in the training, testing, and validation set, respectively. The best cutoff value of nomoscore was 138.5, which achieved sensitivity of 0.935, 0.714, and 0.895, specificity of 0.777, 0.782, and 0.8775 in the training, testing, and validation sets, respectively. Survival analysis shown that patients with nomoscore above 138.5 had significantly poorer survival time than those with scores below 138.5.

Conclusions: Positive T2-FLAIR mismatch sign can effectively rule out glioblastoma in nonenhancing adult-type diffuse gliomas with high specificity. Nonenhancing adult-type diffuse gliomas with nomoscore above 138.5 are highly suspicious for glioblastoma or nonglioblastoma with a poor prognosis.

Keywords: adult-type diffuse glioma; glioblastoma; nomogram; nonenhancing; prognosis.

Figure 1.

Patient selection flowchart. Astro & Oligo: astrocytoma and oligodendroglioma.

Figure 2.

Feature importance evaluation and comparison of nomogram prediction models. (A) feature importance evaluation using recursive feature elimination; (B) the area under curve (AUC) of the prediction model established using different features; (C) the score of each feature measured by nomogram.

Figure 3.

Evaluate the performance of the nomogram prediction model. (A) the receiver operating characteristic (ROC) curve; (B) the nomogram score distribution between astrocytoma & oligodendroglioma and glioblastoma (GBM) groups; (C) the calibration curve of the prediction model; (D) the decision curve of the prediction model.

Figure 4.

Survival analysis based on nomoscore. (A) survival analysis between patients (underwent GTR or STR) with nomoscore above or below 138.5 (best cutoff value); (B) survival analysis of patients (underwent GTR or STR) grouped according to nomoscope and pathological categories; (C) survival analysis between patients (underwent STR) with nomoscore above or below 138.5; (D) survival analysis of patients (underwent STR) grouped according to nomoscope and pathological categories; GTR: gross total resection; STR: subtotal resection; Astro & Oligo: astrocytoma and oligodendroglioma.