Table 3.
Results of the lasso logistic regression toxicity prediction model construction process with and without using K-Means clustering to identify radiosensitive patients, from expansion-response quantified at the end of radiation therapy, for a total of 126 study patients. S.D. – Standard Deviation.
| Model Type | AUCTraining (S.D.) | AUCTest (S.D.) | Brier Score | Scaled Brier (%) |
|---|---|---|---|---|
| No Clustering | 0.842 (±0.065) | 0.693 (±0.099) | 0.175 (±0.020) | 18.2 (±12.0) |
| K-Means Clustering | 0.907 (±0.055) | 0.753 (±0.094) | 0.151 (±0.019) | 12.1 (±11.2) |
| Model Type | Top Recurring Predictors | |||
| No Clustering | MED, LE50Gy100%, Left Medial, LE60Gy100%, Smoking Status | |||
| K-Means Clustering | RS Tag, MED, LE60Gy100%, Smoking Status, Left Medial, Age | |||
The highest recurring predictors from all 1000 iterations of the model construction process are listed from highest to lowest recurring. Standard deviation of AUC values are listed in parentheses.