Abstract
Purpose: :
PVR is still a major cause of failure in the RD surgery, occurring in 5% to 10% of patients with RD. Several attempts have been made to predict the risk of having this complication based on the identification of the most relevant clinical characteristics of RD patients. Then, formulas have been developed based on these clinical factors but none of them has been validated by using it on an external sample.The purpose of this work was to perform an external validation of the already published formulas for predicting the development of PVR, by using the data collected in the Retina 1 project.
Methods: :
Retina 1 project collected 1047 consecutive RD between October 2004 and February 2008 from 17 centers in Spain and Portugal admitted for surgery.Data from these patients were used for external validation. Four formulas reported in the literature were found, including one reported by us in 2005 (R1). For validation of formulas 1 (F1) and 2 (F2), 791 patients who meet the same criteria of the original samples, were considered. For formula 3 (F3) data from 615 patients were used. To validate the quality of formulas receiver operating characteristics (ROC) curve was used. Measures of discrimination were calculated for each formula. Cohen’s Kappa index was used for analysing the consistency.
Results: :
Area under the ROC curves were: R1 0.58 (95% IC: 0.52-0.64), F1 0.60 (95% IC: 0.54-0.66), F2 0.64 (95% IC: 0.59-0.70) and F3 0.41 (95% IC: 0.34-0.48). Best sensitive and accurate model was R1, 74.2% (70.71%-77.71%). Almost 19% of PVR cases were correctly classified by R1 compared to 13%, 15% and 10% for F1, F2 and F3. There was moderate consistency only between F1 and F2.
Conclusions: :
After external validation none formula was accurate enough for its routine clinical use. Therefore some other factors, besides the clinical ones, should be incorporated into future formulas.
Keywords: proliferative vitreoretinopathy • clinical (human) or epidemiologic studies: risk factor assessment • retinal detachment