Abstract
BACKGROUND: To assess longitudinal changes in the in vivo confocal microscopy (IVCM) features during Acanthamoeba keratitis (AK) treatment and develop a prognostic model. METHODS: This retrospective study included 59 AK patients who underwent IVCM at baseline and 1 and 3 months. Fourteen morphological features covering pathogen-related characteristics, cyst arrangement patterns, and inflammatory markers were compared between good and poor prognosis groups, which were defined based on clinical outcomes including corneal perforation, the need for therapeutic keratoplasty, or final best-corrected visual acuity (BCVA) ≤ 0.05. Prognostic modeling was performed exclusively using baseline IVCM features and applied univariable and Firth-corrected multivariable logistic regression with collinearity assessment and clinical filtering, followed by 5-fold cross-validation. RESULTS: Among 59 AK patients, 45 (76.3%) had a good prognosis and 14 (23.7%) had a poor prognosis. Poor prognosis eyes showed a higher prevalence of double-walled cysts, trophozoites, and clustered cysts, along with higher cyst density and deeper stromal invasion. In contrast, good-prognosis eyes had more target-like cysts, immature dendritic cells, and mature dendritic cells. Clustered cysts independently predicted poor prognosis (OR = 2.98), whereas target-like cysts (OR = 0.26) and mature dendritic cells (OR = 0.37) were protective (AUC = 0.883; all p < 0.05). CONCLUSIONS: IVCM provides a quantitative tool for early outcome prediction and individualized management. Higher cyst burden, clustered cysts, and persistent stromal involvement indicated poorer prognosis, whereas target-like cysts and mature dendritic cells indicated better prognosis.