Abstract:
Objectives: The purpose of the current study is to develop an
accurate supervised ML model able to predict the risk of intraoperative
CSF leakage by comparing different machine learning methods.
Design: Retrospective diagnostic-prognostic analysis.
Setting : Endoscopic transsphenoidal surgery for pituitary
adenomas (PAs).
Participants: A cohort of patients consecutively treated via
E-TNS for PAs was selected. Clinical, radiological and endocrinological
preoperative data were reviewed and elaborated through a feature
selecting algorithm. A customized pipeline of several ML models was
programmed and trained in parallel for CSF leakage prediction; the best
five models were included for further analyses. Selected risk factors
were then used for training and hyperparameters optimization.
Main outcome measures : The main outcome metrics were: accuracy,
sensitivity, specificity, PPV, NPV and F1-score of each ML model.
Results: Intraoperative CSF leak occurred in 54 (22,6%) of 238
patients. Best risk’s predictors were: non secreting status, older age,
x-, y- and z-axes diameters, ICD and R ratio. The random forest (RF)
classifier outperformed other models, with an AUC of 0,84, high
sensitivity (87%) and specificity (82%). Positive predictive value and
negative predictive value were 69% and 93% respectively. F1 score was
0,87.
Conclusion : A supervised machine learning prediction model able
to identify patients at higher risk of intraoperative CSF leakage was
trained and internally validated. The random forest classifier showed
the best performance across all models selected by the authors. RF
models might predict surgical outcomes in heterogeneous multimorbid and
fragile populations outperforming classical statistical analyses and
other machine learning models.
Keywords :
Machine learning;
Supervised machine learning;
Pituitary adenoma;
Intraoperative CSF leakage;
CSF fistula;