Clinical implications of intra- and inter-reader agreement in four different automated visual fields

This study was designed and conducted to examine the degree of inter- and intra-reader agreement when four readers evaluate visual fields in patients known to have glaucoma and patients known to not have glaucoma.

Fifty-three patients known to have glaucoma and 60 patients known not to have glaucoma were selected randomly from a population at the University of Alabama at Birmingham School of Optometry. Four visual fields were performed on each patient, non-dilated, with the proper correction in place. The fields were performed on three different instruments by technicians who were unaware of the diagnostic status of the patient. Four independent (blinded) clinicians read the visual fields and judged them as: "glaucoma field defect," "no glaucoma field defect," and "poor reliability-cannot judge."

Statistical analysis revealed that a single reader's interpretation of four different types of visual fields are very different from, are poorer than, and are more consistent than are the four readers' interpretations of a single type of visual field. The analysis also shows that the readers consistently underestimated the number of glaucoma patients with one instrument and overestimated the number with another.

This results suggest that a single visual field will discriminate patients with glaucoma with an accuracy in the 70 to 75 percent range. For one of the instruments, the shorter protocol was almost as effective in discriminating glaucomatous fields as the longer protocol. This study also suggests that repeating the visual fields or analysis of a single field by two readers improves reliability.