Cluster-Based Structural Phenotypes of Chronic Central Serous Chorioretinopathy

To identify structural phenotypes of chronic central serous chorioretinopathy (CSC) by unsupervised clustering of baseline optical coherence tomography and functional variables, and to relate them to photoreceptor staging.

Multicenter, retrospective, clinical cohort study with an independent comparison cohort.

139 eyes of 126 patients with chronic CSC treated with verteporfin photodynamic therapy at two referral centers. K-means clustering (k = 3) of baseline photoreceptor stage, central subfield thickness, subfoveal choroidal thickness (SCT) and best-corrected visual acuity (BCVA) defined the phenotypes, which were expressed as clinical rules. The primary outcome was change in BCVA after subretinal fluid resolution. Phenotype was compared with six-stage photoreceptor grading, and its distribution with 55 chronic eyes from an independent cohort.

138 eyes were phenotyped as pachychoroid-dominant (36 [26%]), transitional (79 [57%]) or degenerative (23 [17%]). Out-of-bag accuracy was 0.87 (95% CI, 0.74-0.96). Without access to the staging system, clustering placed 22 of the cohort's 24 eyes at stage 4 or 5 into the degenerative group, which gained 0.33 logMAR less than pachychoroid-dominant eyes after adjusting for baseline BCVA (95% CI, 0.23-0.44; P < .001). Complete SRF resolution occurred in 83%, 76% and 57% (HR for degenerative, 0.39; 95% CI, 0.20-0.77; log-rank P = .013). Phenotype added nothing to photoreceptor stage for visual recovery (marginal R², 0.575 to 0.590; P = .23) or for resolution (P = .45). Pachychoroid-dominant and transitional eyes, 83% of the cohort, were indistinguishable on photoreceptor grading (P = .68) yet differed in SCT (583 versus 397 µm) and in age (45.5 versus 52.0 years; P = .01), a variable outside the clustering. In the independent chronic cohort the phenotype proportions and the age gradient were both reproduced (51, 54 and 65 years; P = .41 and P = .01), whereas the choroidal separation was attenuated (50 versus 186 µm).

Chronic CSC has a two-dimensional structure. Unsupervised clustering recovered the photoreceptor threshold that governs visual recovery and resolved a second, choroidal dimension that staging cannot represent, separating eyes staging treats as equivalent by choroidal thickness and age. Visual prognosis follows the photoreceptor axis alone. The 520-µm threshold is calibrated to pachychoroid-confirmed disease.