Pediatric obstructive sleep apnea (OSA) is a prevalent breathing disorder linked to cardiovascular diseases. While polysomnography is the diagnostic gold standard, its complexity, high cost, and limited accessibility often lead to underdiagnosis. Therefore, automated methods based on simplified signals, such as single-lead electrocardiogram (ECG), have been developed. While deep learning (DL) models using ECG signals have demonstrated high diagnostic performance, their "black-box" nature hinders clinical adoption. To address this challenge, we propose a novel quantitative explainable artificial intelligence (XAI) framework based on SHAP (SHapley Additive exPlanations) to characterize the cardiac response to pediatric OSA at the morphological level. Building upon an interpretable DL stacking model evaluated on 3,320 nocturnal pediatric ECG recordings, we computed a quantitative SHAP analysis at the ECG segment level using the test subset. Specifically, we calculated Shapley values for ECG signals from the test subset and mapped them to specific waveform components, including QRS complex, ST segment, TP segment, and QT interval. This enabled quantifying the contribution of each component to the estimated apnea hypopnea index (AHI) across the four OSA severities. Results demonstrated a severity-dependent increase in the contribution of the QRS complex and QT and ST intervals to model predictions (p<0.01), supporting their role as key cardiac biomarkers in pediatric OSA. Importantly, an analysis of model misclassifications revealed that wrong predictions in the healthy group exhibited significantly higher Shapley values in the QRS and QT regions (p<0.05), suggesting the detection of subclinical cardiac stress unrecognized by standard AHI criteria. Conversely, misclassified mild OSA cases showed significantly lower values (p<0.05), suggesting an absence of cardiac abnormalities despite apneic events or that the cardiac patterns are similar in both groups. This segment-level SHAP framework could translate DL predictions into clinically meaningful cardiac biomarkers, paving the way for personalized risk stratification in pediatric OSA.