The Archimedes Diabetes Model
Based on information supplied for 2014 Mt Hood Challenge meeting.
The Archimedes Diabetes Model is a component of the Archimedes Model and as such shares some of its features. The Archimedes Model is a continuous-time discrete event simulation model. Probabilities of the occurrence of health outcomes are functions of bio-medical information (continuous variables such as age, blood pressures, lipid panel, etc , and categorical variables such as gender, family history of diabetes, previous MI, etc.) that evolve as a simulated person ages, and interventions that may reduce risks. Based on these probability functions, random uniform draws are used to determine the time of event for a particular simulated person. Two additional features of the Archimedes Model are: (i) the simulated people are based on real people in NHANES, and (ii) healthcare processes are included to mimic typical care in the United States.
The Archimedes Model has a separate probability function for type 1 and type 2 diabetes. The risk variables for the incidence of type 2 diabetes are age, body mass index, and whether a first degree relative has had type 2 diabetes. In addition, the probability of getting type 2 diabetes depends on a person’s gender, race and ethnicity. Fasting plasma glucose in the Archimedes Model evolves over time and becomes greater than 126 mg/dl when a person develops diabetes. FPG increases further as the disease progresses. The diagnosis of diabetes, which is performed by the simulated healthcare system, takes place if a simulated person develops diabetic-related symptoms and is forced to seek medical care or if a blood test is performed in a possibly routine office visit.
If a patient develops diabetes, then the patient may incur diabetes-related complications, including diabetic retinopathy, neuropathy and nephropathy. A patient diagnosed with type 2 diabetes is advised by the healthcare system to diet and exercises. Patients with diabetes are also offered medications such as metformin, sulfonylurea, TZDs, GLP-1 agonists, DPP-4 Inhibitors and insulin. Depending on the combination of anti-diabetes medication use, a patient may experience hypoglycaemia events. The retinopathy, neuropathy and nephropathy models include distinct probability functions, which like the diabetes model, are functions of various risk variables. Systolic blood pressure and HbA1c are among the risk variables for all three microvascular complications. These microvascular models have adequate level of complexity to capture the progression of the disease. For example, the nephropathy model includes urinary albumin to creatinine ratio and glomerular filtration rate as basic variables and microalbuminuria, macroalbuminuria, chronic kidney disease stage 3 and end-stage-renal disease (ESRD) as its principal outcomes. Enhanced death due to ESRD, dialysis treatment and kidney transplants are also modelled.
When a patient develops type 2 diabetes, their lipid panel worsens in agreement with what is observed in real patients. The driving factor is insulin resistance in fat cells, which is captured in the Model through elevated triglycerides that cause total cholesterol to go up and HDL to go down. Patients with diabetes are at elevated risk for myocardial infarction, heart failure, and stroke. Part of this elevated risk is due to poorer biomedical variables associated with diabetes and part of this is attributed to the diabetes disease itself.
Simulation output consists of demographic characteristics, biomarker progressions, prescribed treatments and adherence, incidence of microvascular complications and macrovascular outcomes (myocardial infarction, stroke, congestive heart failure), rate of hypoglycaemia and mortality.
Eddy DM, Schlessinger L. Validation of the Archimedes diabetes model. Diabetes Care. 2003; 26:3102-10.
Kahn R, Alperin P, Eddy D, Borch-Johnsen K, Buse J, Feigelman J, Gregg E, Holman RR, Kirkman MS, Stern M, Tuomilehto J, Wareham NJ. Age at initiation and frequency of screening to detect type 2 diabetes: a cost-effectiveness analysis. Lancet. 2010; 75:1365-74.