Model Predicts Vision-Threatening Anterior Segment Disease

Optometrist examing patient's eyes
Optometrist examing patient’s eyes
The model could help reduce the number of patients referred to ophthalmology, researchers suggest.

A prediction model can increase the positive predictive value for anterior segment vision-threatening diseases (asVTD) compared with referral based on prevalence probabilities, according to the results of a study published in Cornea.

The investigators developed the decision-support tool to aid primary care physicians (PCPs) with patient triage and referral for asVTD.

The team reviewed University of Michigan electronic health record (EHR) data of patients who presented with anterior eye symptoms to a PCP who then visited an ophthalmologist within 30 days, between January 1, 2016, and May 31, 2019. The investigators included diagnosis of corneal ulcer, iridocyclitis, hyphema, anterior scleritis, or scleritis with corneal involvement by an ophthalmologist as asVTD. Predictors evaluated in logistic regression modelling and cross-validation included patient demographics and PCP notes. 

In total, 2942 patients met the inclusion criteria, and 133 patients (4.5%) had asVTD. Patients with asVTD had significantly lower median age than those without asVTD (42 vs 53 years; P =.001). No differences in sex or race were observed between the groups.

The team’s final prediction model had an area under the curve of 0.72 (95% confidence interval, 0.67–0.77), and at a threshold that achieved a sensitivity of 90%, the it had a specificity of 30%, positive predictive value of 5.8%, and negative predictive value of 99%. 

“As compared with the 22 patients currently evaluated by ophthalmologists for every case of asVTD (2942/133; the number of patients seen by a PCP who had an anterior segment eye complaint and went on to see an ophthalmologist divided by the number of patients who were diagnosed with asVTD), the use of the model could reduce this to 17 patients per case with 90% sensitivity or 15 patients per case with 80% sensitivity,” wrote the authors.

Limitations of the study included potential errors and missing data in the EHR, data from a single center and its affiliated primary care practices, variability in clinical documentation among practices, and the relatively low number of asVTD cases in the study.

“The use of the prediction model increased the positive predictive value for asVTD compared with referral based on prevalence probabilities,” researchers report. “A prediction algorithm has potential to improve triage and initial management decision-making for PCPs because it performs better than probabilities in the absence of such a tool.”


Singh K, Thibodeau A, Niziol LM, et al. Development and validation of a model to predict anterior segment vision-threatening eye disease using primary care clinical notes. Cornea. Published online October 6, 2021. doi:10.1097/ICO.0000000000002877