In order to improve healthcare systems, we must know what they are capable of. "Simulation Modeling for Implementation Analysis" is a pilot project to bring complex systems methodologies of discrete event simulation and agent based modeling to the analysis of healthcare delivery in the VA, specifically in the provision of care for diabetic retinopathy.
Diabetic retinopathy is the leading cause of new blindness in Americans aged 20-70. It is a common co-morbidity of diabetes mellitus, which currently afflicts 25% of the US Veteran population. Diabetic retinopathy is estimated to cost more than $500 million annually to treat in the United States.
1) Design and create a hybrid agent-based/discrete event simulation comprised of the agent-based population and a discrete event simulation of the St. Louis VAMC Ophthalmology Clinic DR services, whereby the population may seek and receive treatment for DR according to current practices
2) Evaluate the effect of implementation of biannual screening of patients with no or background DR on the prevalence of DR-caused blindness in the simulated population of diabetic eye clinic patients over the next ten years.
3) Evaluate the effect of aggregate population level improvements of BMI, and A1c on demand for pan-retinal photocoagulation laser surgery as treatment for incident diabetic retinopathy over the next ten years.
An agent-based population of diabetic veterans with diabetic retinopaty will be created, and coupled to a discrete event simulation of an ophthalmology clinic. The agents comprising the population will be able to seek care at the simulated clinic, exactly as a real world population seeks care at a real world clinic.
In this way, we can measure the effects of clinical changes at the level of the population health, by observing changes in the rate of progression of diabetic retinopathy in response to changes made to the clinic. And conversely, we can measure how changes to population demography and healthy influence demand and access at the clinic.
We find that there was no statistically significant increase in diabetic retinopathy (DR) associated vision loss over a simulated ten year period associated with increasing the screening interval for Veterans with no or background DR from one year to two years. However, significant differences were found between screening intervals of one and three years (P<0.01), and between two and three years (P=0.0106). This supports the scientific consensus that the recent adoption of biannual screening intervals for DR is safe. However, additional increases in screening interval duration are not recommended by this research.
We also found that adopting the biannual screening policies saved a significant number of eye clinic resources, with the annual screening group (n=501) averaged 4206 (94.1) appointments over a 10 year period, while the biannual screening group (n=501) averaged 3441 (85.7). P<0.0001.
We have provided supporting evidence of the safety and effectiveness of the biannual screening policy for diabetic eye exams in patients without DR. Additionally, we conclude that eye clinic resources may be freed up to serve other Veterans as a result of this policy.
Additionally, we report on the development of a new hybrid AB/DES model architecture which can model and predict the interactions of a population with the clinical systems that serve it. This shows broad potential to model and optimize many different systems within VHA.
- Day TE, Ravi N, Xian H, Brugh A. Sensitivity of diabetic retinopathy associated vision loss to screening interval in an agent-based/discrete event simulation model. Computers in biology and medicine. 2014 Apr 1; 47:7-12.
- Day TE, Ravi N, Xian H, Brugh A. An Agent-Based Modeling Template for a Cohort of Veterans with Diabetic Retinopathy. PLoS ONE. 2013 Jun 21; 8(6):e66812.