LAUSR

Purpose: The aim of this study was to identify characteristics of infectious keratitis predictive of poor outcome to develop a web-based predictive calculator. Method: A retrospective chart review was performed at the Duke Eye Center. Two hundred fifteen adult patients with culture-proven infectious keratitis presenting between January 1, 2016, and December 31, 2020, were separated into a derivation set (136 patients, 53 positives; 83 controls) and a temporal validation set (79 patients, 26 positives; 53 controls). The poor outcome group consisted of patients requiring penetrating keratoplasty for visually significant scarring, penetrating keratoplasty for ulcer progression, or evisceration/enucleation for endophthalmitis. Univariable analysis was performed followed by stepwise multivariable logistic regression to obtain a predictive model in the derivation data set. Culture-naïve and postculture models were constructed. Discrimination and calibration were assessed using the area under the curve (AUC) and calibration plots, respectively. Results: The culture-naïve model consisted of corticosteroid drop use postsymptom onset [Odds Ratio (OR) = 2.3, P = 0.054], decreased vision (OR = 2.4, P = 0.001), and increased ulcer area (OR = 1.017, P = 0.017). The postculture model additionally included fungal keratitis (OR = 5.4, P = 0.006) and elapsed time from symptoms to organism-sensitive therapy (OR = 1.027, P = 0.014). The models were summarized by the acronym C-DU(KE). The AUCs for the culture-naïve model were 0.794 in the derivation set and 0.850 in the validation set. The AUCs for the postculture model were 0.898 in the derivation set and 0.946 in the validation set. Calibration plots indicated goodness of fit in the data sets for both models. The calculator was deployed under the URL: https://duke-eye-calculator.shinyapps.io/Corneal_Ulcers/. Conclusions: The C-DU(KE) calculator permits a data-driven prediction of outcome in infectious keratitis that can supplement clinical judgment.