A lasso-logit prediction model
Increased total bilirubin, direct bilirubin and chemotherapy phase of
induction are screened out as independent risk factors for pancreatitis
by lasso regression. The variation of the variable coefficients in the
lasso-logit regression model is visualized in Figure 2A. To optimize
model performance, we employed a 10-fold cross-validation procedure,
identifying an optimal model with both high performance and minimal
variable count when the regularization parameter (lambda) was set to
0.015 (Figure 2B). This selected set of variables was then used to
construct a logistic regression model (Table 2). The resulting
predictive model demonstrated robust performance, achieving a
c-statistic of 0.862. The calibration curve (Figure 3) further indicates
good predictive accuracy, with a low mean absolute error (MAE) of 0.029,
signifying a close agreement between predicted and observed
probabilities. As can be seen from table2 and the nomogram (Figure 4),
patients with higher total and direct bilirubin levels during induction
phase are more likely to develop pancreatitis.