Artificial intelligence and machine learning-based approaches are increasingly finding their way into everyday clinical practice. They may prove helpful in predicting future treatment intervals in the management of patients with neovascular AMD, according to Maximilian Pfau MD, who presented his research at the EURETINA 2021 Virtual Congress.
“Our study showed that anti-VEGF therapy demand in choroidal neovascularization (CNV) due to AMD can be predicted with reasonable accuracy and achieve the same prediction as classical machine learning-based models. The advantage of this approach is that it can help physicians consider the model output thoughtfully or use standard treatment as a fallback alternative,” he said.
Using such AI models may help physicians select patients for whom long-term treatment options beyond conventional anti-VEGF injections may provide a favourable risk/benefit ratio, added Dr Pfau.
In the study carried out at the University of Bonn, Dr Pfau and co-workers conducted a retrospective analysis of 138 visits of 99 eyes with neovascular AMD using scans from two clinical centres.
The team extracted 270 spectral-domain-OCT-based features using deep learning-based AI segmentation. Retinal thickness and reflectivity values were extracted for the central and the four inner Early Treatment Diabetic Retinopathy Study (ETDRS) subfields for six retinal layers. Machine-learning models were probed to predict the anti-VEGF treatment frequency within the next 12 months.
Probabilistic forecasting was performed using natural gradient boosting (NGBoost), which outputs a full probability distribution. The primary outcome measure was the mean absolute error (MAE) between the predicted versus actual anti-VEGF treatment frequency.
In terms of results, the prediction of future anti-VEGF treatment frequency was observed with an accuracy of 2.60 injections per year using random forest regression and 2.66 injections per year using NGBoost. The prediction intervals were well calibrated and reflected the genuine uncertainty of NGBoost-based predictions, said Dr Pfau.
Unlike classic point prediction, which determines how many injections are required for each patient, probabilistic forecasting includes an estimate of each prediction’s uncertainty, explained Dr Pfau.
With this knowledge, predictions with high certainty would allow physicians to use the prediction for clinical decision-making about upcoming anti-VEGF therapy approaches. On the other hand, uncertain predictions would enable them to opt for a manual fallback alternative such as conventional PRN or treat-and-extend protocol with established anti-VEGF agents.
Responding to a question concerning the possibility of using the AI model to predict beyond a 12-month window, Dr Pfau said that while it was technically possible to do so, the model would need much larger data sets to reduce uncertainty in its predictions.
Going forward, Dr Pfau said that future work would focus on validating the uncertainty estimates for NGBoost to ensure they truly highlight out-of-distribution samples, especially in conjunction with external data sets.
END
Contact: [email protected]
