Continuous Likelihood Ratios for Glaucoma Diagnosis Using the Combined Index of Structure and Function

Abstract

 

Purpose

 

To evaluate the ability of a combined index of structure and function (CSFI) to diagnose glaucoma using a new methodology for calculating continuous likelihood ratios (LRs).

 

Methods

 

This was an observational cohort study that included 937 eyes of 624 subjects recruited from the Diagnostic Innovations Glaucoma Study (DIGS) and the African Descent and Glaucoma Evaluation Study (ADAGES). Eyes were classified as having glaucoma if they had previous progressive optic nerve change and/or reproducible visual field damage. The control group was composed of eyes from healthy subjects recruited from the general population, regardless of visual field and optic disc status. All eyes underwent retinal nerve fiber layer imaging with Spectralis spectral domain optical coherence tomography (SDOCT) and standard automated perimetry (SAP). Estimated number of retinal ganglion cells (RGCs) were obtained from SDOCT and SAP and a weighted average (WRGC) was used to obtain the final estimate of number of RGCs for each eye. The CSFI was calculated as the percent loss of RGCs obtained by subtracting estimated from age-expected RGC numbers. Continuous LRs for glaucoma diagnosis were estimated for WRGC estimates and CSFI using a new methodology based on estimating the tangents to the Receiver Operating Characteristic (ROC) curve.

 

Results

 

Three hundred ninety nine eyes of 240 subjects were included in the control group and 538 eyes of 384 patients in the glaucoma group. The median of mean deviation was -3.5 dB (range -27.5 to 2.0 dB) for the glaucoma group. The number of WRGC was 1,078,000±195,000 in the control group and 596,000±217,000 in the glaucoma group (P<0.0001). CSFI was -3.7±12.3 in the control group and 34.7±23.3 in the glaucoma group (P<0.001). Continuous LRs were calculated for each specific value of WRGC and CSFI. Lower WRGC estimates and larger CSFI values were associated with larger LRs for glaucoma diagnosis. CSFI values larger than 26 were associated with large effects on post-test probability of disease, that is, LRs greater than 10.

 

Conclusions

 

The methodology allowed estimation of continuous LRs for glaucoma diagnosis for specific values of WRGC estimates and CSFI values. Calculation of continuous LRs may facilitate the incorporation of results of diagnostic tests into clinical decision-making for glaucoma diagnosis.

 

 

Graph of Likelihood Ratios against Combined Index of Structure and Function

 

Keywords: 468 clinical research methodology • 501 discrimination • 758 visual fields