Prediction of Poor Visual Outcomes at Idiopathic Intracranial Hypertension Diagnosis Using a Supervised Machine Learning Algorithm.

Abstract

Background: Idiopathic intracranial hypertension (IIH) is a vision-threatening disorder mainly affecting women of a reproductive age. Prompt diagnosis and intervention are vital to prevent vision loss, but validated tools to predict visual outcomes are lacking. The purpose of this study was to create a machine learning algorithm predicting poor visual outcomes at the time that the diagnosis of IIH is established, and stratifying risk among those with and without poor visual acuity at presentation.

Methods: Using electronic health records, a retrospective cohort study was conducted between June 1, 2012 and September 30, 2023. Any patient aged 0-70 years who was diagnosed with IIH and met the revised diagnostic criteria was included in the analysis. In total, 391 patients with IIH had final outcomes available and were included in this analysis. Final visual outcomes were reported between 3 months and 1 year after diagnosis. Poor visual outcomes served as the model outcome and was defined as a visual field mean deviation (VFMD) worse than -7 dB or a visual acuity of 20/80 or worse. Both logistic regression and decision trees were used to build predictive models. Models were evaluated using multiple parameters including accuracy, sensitivity, specificity, and area under the curve. The best performing models were validated using a k-fold cross-validation.

Results: The decision tree models performed the best and 4 prognostic risk groups were created: critical, high, medium, and low. In the critical risk group, patients who had both high baseline VFMD (worse than -12.59 dB) and identified as non-White had a poor visual outcome risk of 92.6%. A baseline VFMD worse than -9.1 dB resulted in a critical risk of a poor visual outcome at 69.8%. Any patient with a baseline VFMD better than -3.39 dB had a risk of a poor visual outcome at 1.04%.

Conclusions: Our study provides clinicians with valuable prognostic markers to assist in identifying patients who are at critical risk for significant vision loss. Patients with a VFMD worse than -9.1 dB have a critical risk of a poor visual outcome, and this further increased if they identified as a minority patient.