Machine learning models predict delayed hyponatremia post-transsphenoidal surgery using clinically available features
Authors' conclusion
The proposed machine learning models with pre- and post-resection features predicted DHN after the resection of PitNETs.
Study context
- Population
- Patients with pituitary neuroendocrine tumors (PitNETs) undergoing endoscopic transsphenoidal surgery (eTSS); predictors available preoperatively and on the first postoperative day
- Participants
- 193 patients
- Exposure
- Four machine learning models for predicting delayed hyponatremia
- Outcome
- Prediction of delayed hyponatremia (DHN): serum sodium < 135 mmol/L at ≥ 1 time between postoperative days 3 and 9
- Follow-up
- Postoperative days 3–9
Methodological detailssingle-centre, machine learning prediction model development
- Single-centre
- Machine learning prediction model development
Main finding
As reported in the abstract
ROC-AUC for DHN prediction
| Group | Value | AUC |
|---|---|---|
| Random forest (RF) | 0.759 ± 0.039 | 0.759 ± 0.039 |
| Support vector machine | 0.747 ± 0.034 | 0.747 ± 0.034 |
| LGBM | 0.738 ± 0.026 | 0.738 ± 0.026 |
| Logistic regression | 0.710 ± 0.028 | 0.710 ± 0.028 |
Value
Random forest (RF)0.759 ± 0.039
Support vector machine0.747 ± 0.034
LGBM0.738 ± 0.026
Logistic regression0.710 ± 0.028
AUC
Random forest (RF)0.759 ± 0.039
Support vector machine0.747 ± 0.034
LGBM0.738 ± 0.026
Logistic regression0.710 ± 0.028
Observational data: associations, not effects.
Further findings
| Outcome | LGBM |
|---|---|
| Accuracy | 0.746 ± 0.029 |
| Accuracy | 0.746 ± 0.029 |
- Accuracy — Highest accuracy observed in LGBM model
- 0.746 ± 0.029
Features of best-performing RF model
- Features in RF model
- 24
- Features clinically available preoperatively
- 9
Original publication
Machine learning models predict delayed hyponatremia post-transsphenoidal surgery using clinically available features. Pituitary. Published 30 March 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →