Background and objectives: Preventive treatment of unruptured intracranial aneurysms (UIAs) requires assessment of treatment risks vs expected benefit. Although established scores exist to estimate rupture and growth risk, currently, no externally validated tools exist to estimate the risks of microsurgical treatment of UIAs. Clinical prediction models based on machine learning enable generation of personalized risk estimates for each individual patient based on their specific patient and aneurysm characteristics. Methods: Using data from 20 international centers from the prediction of adverse events after microsurgery for intracranial unruptured aneurysms study on patients treated microsurgically for UIAs, we developed and externally validated clinical prediction models for 3 outcomes measured at hospital discharge: poor neurological outcome (modified Rankin Score ≥3), new sensorimotor neurological deficits, and all-cause adverse events (Clavien-Dindo Grade ≥1). Results: A total of 3705 patients were included. Data from 13 centers (2881, 78%) were used for model development. Fully trained models were evaluated on 824 patients (22%) from 7 additional centers. Average age was 56 ± 12 years, and 1049 (28%) were male. At discharge, poor neurological outcome was seen in 514 patients (14%). New sensorimotor deficits were observed in 534 patients (14%), and 894 patients (24%) experienced adverse events until discharge. At external validation, prediction of poor neurological outcome was achieved with good calibration and an area under the curve (AUC) = 0.70 (95% CI: 0.63-0.75). Similarly, new neurological deficits were predicted with good calibration and with an AUC = 0.69 (95% CI: 0.63-0.74). Prediction of all-cause adverse events only achieved an AUC = 0.59 (95% CI: 0.55-0.64) with fair calibration. The prediction model was integrated into a web application accessible at https://neurosurgery.shinyapps.io/PRAEMIUM/. Conclusion: The developed models for prediction of poor neurological outcome and new sensorimotor neurological deficits at discharge exhibit good calibration and fair discrimination based on a multinational external validation, indicating that the predicted probabilities correspond well to real-world risks and may thus be clinically useful in more objectively estimating the risk of microsurgical treatment.
Staartjes et al. (Fri,) studied this question.