EXTERNAL VALIDATION OF A MODEL OF GLAUCOMA PROGRESSION USING INDIVIDUAL LEVEL DATA

Purpose:   Evaluation of the cost-effectiveness of treatment for chronic disease requires development and validation of a model of the progression of disease validated using epidemiological evidence.  We constructed a Markov model to predict the progression of glaucoma using three large prospective randomized treatment trials and evaluated the modelÕs internal and external predictive validity.

Method: Glaucoma severity and disease progression were defined clinically in terms of visual field loss in mean deviation (MD) as decibels (dB). Patient level data for the model came from the Collaborative Initial Glaucoma Treatment Study (n=607), the Ocular Hypertension Treatment Study (n=1,636), and the Advanced Glaucoma Intervention Study (n=560). We modeled the change in MD over seven years as a function of four variables at baseline: age (3 categories), race (2 categories), starting MD (30 levels) and intraocular pressure (IOP, 5 categories, also included longitudinally). Transition matrices were calculated for all combinations of these risk factors. The microsimulation model was estimated using TreeAge software using. Internal validation was conducted by comparing the predicted value of hypothetical participants with the same characteristics as the actual study participants to the actual participantÕs results at seven years.  Preliminary external validation was conducted with a longitudinal sample of 116 people with glaucoma not in the studies.

Result: The results of internal validation are shown in the Table below. Nearly 70% of all participants were classified within 3 dB (minimum significant difference) of their final result at seven years. Of those outside this range, over 60% were overpredicted. External validation results from our preliminary sample found that at seven years over 73% of patients were within 3 dB. At five years, the model classifies over 80% of the external sample patients within 3 dB.        

Conclusion: Our initial results indicate that the model properly predicts the result of disease progression in over 70% of ÒparticipantsÓ. External validation with additional samples is ongoing. If the model results are found to be robust, the model will provide an important tool for clinicians, investigators and policy makers.