The Crowd Sourced Glaucoma Study (CSGS) sought to derive an objective definition of glaucoma for research based on combined assessments by global expert evaluators.

Cross-sectional diagnostic evaluation study.

(1) Investigators from 22 centers in 15 countries who provided data that made up the CSGS dataset; (2) 1,234 patients whose optic disc photographs, visual field (VF) and optical coherence tomography (OCT) examinations were included in the dataset and (3) 531 expert evaluators from 74 countries who assessed the glaucoma likelihood of the patients.

Evaluators provided a glaucoma likelihood score (from 0% to 100%) using a web-based application that interactively displayed the VF and OCT results. Evaluators were asked to evaluate 40 to 200 patients, with a range of disease severity. A mean glaucoma likelihood score was computed from the multiple evaluations for each patient.

We analyzed the relationship among all available global and sectoral parameters from VF and OCT examinations and mean likelihood score. The parameters were generalized for the most used VF and OCT devices. A decision tree analysis, with 80% of the dataset for training, was used to derive criteria yielding the best diagnostic performance based on mean likelihood score. The diagnostic performance of the criteria was then tested independently in the remaining 20% of the dataset.

Of the 531 evaluators, 491 (93%) were ophthalmologists among whom 412 (78%) had specialized training in glaucoma. Four hundred ninety-two (93%) evaluators were from the Americas, Europe, or Asia. Each patient had a mean (SD) of 35 (4) evaluations. The mean glaucoma likelihood score ranged from 1.4% to 99.6%, with inter-evaluator variability highest for mid-range likelihood scores and lowest for either very low or very high likelihood scores. The mean (SD) intra-evaluator difference in likelihood score was 0.8 (18.9)%. A definition of glaucoma based on whether any retinal nerve fiber layer sector was abnormal, followed by whether the Glaucoma Hemifield Test was borderline or outside normal limits, had sensitivity to detect glaucoma and no glaucoma of 85% and 90%, respectively.

The CSGS criteria represent parameter-defined consensus-based definition of glaucoma for clinical research.