Abstract
Purpose: :
To determine the sensitivity and specificity of the Glaucoma Probability Score (GPS) in a large working population.
Methods: :
4234 of 13037 employees of the Evonik Industries between 40 and 65 y were screened for glaucoma. Glaucoma suspects were identified by evaluation of the optic nerve head, frequency doubling technology, IOP and central corneal thickness by a glaucoma expert. The subjects were categorized in non-glaucoma suspects (G0) and glaucoma suspects, differentiating between possible (G1) and probable (G2) glaucoma cases. Furthermore, each subject underwent HRT imaging using the GPS algorithm. For analysis, the GPS global classification according to the manufacturers recommendation was used: inside normal limits (INL,<0.29), borderline (BL,<0.65), and outside normal limits (ONL,≥0.65). The clinical diagnosis of each subject was compared with the results of GPS (specificity (Sp), sensitivity (Se), positive (PPV) and negative (NPV) predictive values). Three different analyses were performed: INL vs. BL+ONL against G0 vs. G1+G2 (analysis 1); INL+BL vs. ONL against G0 vs. G1+G2 (analysis 2); INL+BL vs. ONL against G0+G1 vs. G2 (analysis 3).
Results: :
Good quality HRT images (standard deviation <30µm) were achieved for 4059 (95.9%) subjects for both eyes. On the basis of clinical diagnosis 3952 (97.4%) subjects were categorized as G0, 98 (2.41%) as G1 and 9 (0.22%) as G2.Further 85 subjects (analysis 1) and 99 (analyses 2 and 3) had to be excluded due to undefined GPS. Analysis 1: In 80 of 107 subjects GPS was in agreement with the clinical diagnosis, thus Se=75%. GPS graded 1446 of 3867 subjects as positive (Sp=63%). The PPV was 5.2%, the NPV 98.9%. Analysis 2: GPS categorized 39 subjects as positive among 107 G1+G2 subjects and 426 among 3853 G0 subjects (Se=36.5%, Sp=88.9%, PPV=8.4%, NPV=98.1%). Analysis 3: Among 11 G2 subjects GPS categorized 6 as ONL and 459 ONL subjects were identified among 3949 G0+G1 subjects (Se=55%, Sp=88.4%, PPV=1.3%, NPV=99.9%).
Conclusions: :
When considering BL and ONL as glaucoma suspect, we found a sensitivity of 75% for the GPS at best. However, with positive predictive values of less than 10%, GPS seems not be an adequate single screening unit for glaucoma.
Keywords: imaging/image analysis: clinical • optic disc • clinical (human) or epidemiologic studies: health care delivery/economics/manpower