Predicting relapse in patients with medulloblastoma by integrating evidence from clinical and genomic features

Abstract

Purpose: Despite significant progress in the molecular understanding of medulloblastoma, stratification of risk in patients remains a challenge. Focus has shifted from clinical parameters to molecular markers, such as expression of specific genes and selected genomic abnormalities, to improve accuracy of treatment outcome prediction. Here, we show how integration of high-level clinical and genomic features or risk factors, including disease subtype, can yield more comprehensive, accurate, and biologically interpretable prediction models for relapse versus no-relapse classification. We also introduce a novel Bayesian nomogram indicating the amount of evidence that each feature contributes on a patient-by-patient basis.

Patients and methods: A Bayesian cumulative log-odds model of outcome was developed from a training cohort of 96 children treated for medulloblastoma, starting with the evidence provided by clinical features of metastasis and histology (model A) and incrementally adding the evidence from gene-expression-derived features representing disease subtype-independent (model B) and disease subtype-dependent (model C) pathways, and finally high-level copy-number genomic abnormalities (model D). The models were validated on an independent test cohort (n = 78).

Results: On an independent multi-institutional test data set, models A to D attain an area under receiver operating characteristic (au-ROC) curve of 0.73 (95% CI, 0.60 to 0.84), 0.75 (95% CI, 0.64 to 0.86), 0.80 (95% CI, 0.70 to 0.90), and 0.78 (95% CI, 0.68 to 0.88), respectively, for predicting relapse versus no relapse.

Conclusion: The proposed models C and D outperform the current clinical classification schema (au-ROC, 0.68), our previously published eight-gene outcome signature (au-ROC, 0.71), and several new schemas recently proposed in the literature for medulloblastoma risk stratification.

Fig 1.

Bayesian cumulative log-odds model (probabilistic network) that integrates clinical and high-level genomic information to predict the probability of relapse. The submodel of model C corresponding to subtype determination is shown in a different color because it is applied separately and in advance for any unlabeled sample. RSV, respiratory syncytial virus.

Fig 2.

(A) Receiver operating characteristic (ROC) plots (empirical, dashed line; binormal, solid line) and area under the ROC (au-ROC) curve performance of the current clinical schema and models A to D in the training set. (B) ROC plot and au-ROC curve performance of the current clinical schema and models A to D in the independent test set. (C) ROC plots and au-ROC curve performance for model C in the independent test set for the standard-risk and (D) high-risk patient groups as defined by the current clinical schema. Note that only 73 of the test samples had corresponding clinical annotation that allowed categorization as standard risk or high risk. The model will still make a predictive call based on the genomic data. The 95% CIs in Figures 2B and 2C are estimates based on bootstrap sampling and are affected by small sample size. N, no; Y, yes.

Fig 3.

Bayesian nomogram showing the amount of evidence that each feature (risk factor) provides in the context of a specific patient's feature values. The arms of the nomogram represent the values of the posterior log odds ratio evidence, Ev(r | x), for each feature's values. The actual values taken by each feature are shown in blue or red according to the sign of Ev(r | x): positive magnitude to the right side is evidence for relapse (red) and negative magnitude to the left is evidence for no relapse (blue). The final sum of Ev(r | x) provides the final probability of relapse, which is 0.89 (95% CI, 0.65 to 0.97). OR, odds ratio; Lods, log-odds; Prob, probability; RSV, respiratory syncytial virus; H, high; L, low; DESMO, desmoplastic; LCA, large-cell/anaplastic (lymphoma); N, no; Y, yes.

Fig 4.

Heat map of 38 samples showing the Genomic Identification of Significant Targets in Cancer (GISTIC) amp(8q24.21/c-Myc) status, the GISTIC amp(2p24.3/N-Myc) status, the expression of the disease-independent c-Myc pathway, the del(6q/monosomy 6) GISTIC status, and the expression of the beta-catenin pathway.

Fig A1.

Nomogram entry for histology feature. DESMO, desmoplastic; LCA, large-cell/anaplastic (lymphoma).