Comprehensive Summary
This study presented by Hogg et al. established an improvement upon current treatment monitoring methods for neovascular age-related macular degeneration (nAMD) through the recognition of AI-enabled decision making software. The research was performed via conducting a dual-site external validation study utilizing data from The Moorfields Eye Centre and Newcastle Eye Centre. Under these circumstances, Hogg et al. were able to analyse a total of 521 paired retinal optical coherence tomography (OCT) images curated from 468 distinct individuals. As a result, a comparison of AI-led and real-world clinical assessments of diagnostic accuracy for nAMD displayed a rNPV (relative negative predictive value) of 1.17 (1.12-1.22), as well as a rPPV (relative positive predictive value) of 0.942 (0.75-1.19). In translation, this means that across both datasets, the AI-led model statistically performed more favorably than the real-world clinical diagnostics. Furthermore, error analysis was performed on the quantitative data such that assessments were evaluated across age, sex, and ethnic subgroups.
Outcomes and Implications
The results which Hogg et al. presents lay out a blueprint for increased quality in autonomous assessment of nAMD disease in the future via application and integration of AI into the diagnostic process. Due to the large range of demographic consideration in this study, the external validity of the research confirms that the AI system is already valuable. However, Hogg et al. notes that further replication of studies would be preferred, and proves true as the human-computer cognitive biases will apply more influence on the reality of the AI system’s clinical use. A further note is considered in terms of how patient acceptance plays into clinical application, as autonomous decisions stemming from AI may not be agreed upon with certain patients. As a result, while the technology developed currently shows exceptional applicational use, there are more qualitative factors which may have to be considered and tested for before being translated into the clinical workspace.