Patient specific predictions of visual acuity and inflammation in uveitis

Abstract

Purpose : Accurate prediction of VA is highly relevant for uveitis patients as uveitis is the leading cause of legal blindness in the working population in the western world. To this end, we developed an interactive tool for patient specific predictions of visual acuity (VA) and inflammation (IN).

Methods : Data consisted of both eyes from 366 uveitis patients who visited the Rotterdam Eye Hospital in the period from 2000 to 2015. Mean follow-up (FU) time and mean number of visits were 2.5 years and 15 visits, respectively. A statistical model that jointly models the IN process (multi-state model with random effects for each patient and eye) and the VA (linear mixed model) was employed. The model provides predictions of VA and IN episodes after entering the patient’s disease and treatment history. The model was evaluated using the mean absolute error (MAE) of observed vs predicted VA. MAE was calculated for all patients at different FU times and at different time points in the future.

Results : Figure 1 shows an example of the prediction model applied to a patient with 1 year of FU, consisting of 11 visits after onset and almost all with an active IN. Decreasing vision was predicted for both eyes and OD was expected to be below driving level (Snellen<0.6) within a year. In general, the probability of IN was high (around 0.45) for both eyes.
MAEs for the VA predictions were approximately 0.10-0.15 (LogMAR). With longer FU, and thus more data on each patient, the MAE decreased, while it mostly increased when predicting further into the future (Figure 2). The predictions were considerably better than assuming no change in VA, although the difference decreased with longer FU.

Conclusions : Prediction models are useful tools to help support clinical decision making and inform the patient about the disease prognosis. Our model showed good predictive ability of VA, but a more extensive validation is needed before applying it in a clinical setting.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

 

Figure 1: The vertical dotted line indicates the current visit. Past: Observed VA and IN status (dots). Future: Predictions of VA (blue line), their prediction intervals (shaded blue area) and the probability of IN (red line).

 

Figure 2: MAE of predicted vs observed VA at different time points during FU (shape) and into the future. The colours indicate the model predictions (blue) vs VA at the last FU visit (red).