The Diabetes Model, and its Scenarios

The Diabetes Model

Model Structure

Diabetes Models creates several scenarios: Null_Diabetes, D1, D2, D3, D4, D5

The Diabetes Model refers to a "Model architecture": A structure of states and transitions, which can be used to run different scenarios. A scenario is when the structure has a different set of transition rates between the states.

The Diabetes model is used to run two scenarios of treatment coverage: D1, D2, D3, D4 and D5 In addition, the Diabetes model is also used to run a "Null Scenario": Called Diabetes_Null. We will explain the Null Scenario later.

Structure of the Diabetes Model

Diabetes has several key states, DsFreeSus, DiabetesEpsd, DiabeticRetinopathy, LowerExtrmAmpt, DiabeticNephropathy and Deceased. DsFreeSus means "Disease free, susceptible", and this refers to the majority of the population. DiabetesEpsd means "Diabetes Episode", and generally refers to people who have been diagnosed with diabetes. DiabeticRetinopathy means "Diabetic Retinopathy" and refers to people who have retinopathy caused by diabetes. LowerExtrmAmpt means "Lower Extremity Amputation" and refers to people who have had a lower limb, or part of that lower limb (e.g. a foot) amputated because of diabetes. DiabeticNephropathy means "Diabetic Nephropathy", and refers to people who have nephropathy caused by diabetes. Deceased refers to the people in the model who have died, either through background mortality (DsFreeSus -> Deceased) or through the Diabetes episode (DiabetesEpsd -> Deceased).

In addition to these states, there are other "states" which are used to perform calculations, or collect useful statistics about the model. This is something we have chosen to do in our model structure, so it's visible to users, but it is not strictly necessary. For example, we have states for Disability, which collects information about the stock of key states and multiplies them against some disability weight. We also have states to calculate births, migration, and the effects of interventions on disability and mortality. Once again, we made these design decisions so that users can see how these work, but they aren't strictly necessary. They can be done elsewhere, and simply rendered as a transition rate.

The Model has two key components

The Diabetes model is large, but can be broken down into two components.

The main component moves people between states

The main component has the states we've introduced: DsFreeSus, DiabetesEpsd, DiabeticRetinopathy, LowerExtrmAmpt, DiabeticNephropathy, Deceased, Disability, Births. Importantly there are some other states which sit between states:

• DsFreeSus Disability sits between DsFreeSus and Disability
• DiabetesEpsd Disability sits between DiabetesEpsd and Disability
• DiabeticRetinopathy Disability sits between DiabeticRetinopathy and Disability
• LowerExtrmAmpt Disability sits between LowerExtrmAmpt and Disability
• DiabeticNephropathy Disability sits between DiabeticNephropathy and Disability
• DiabetesEpsd Mortality sits between DiabetesEpsd and Mortality
• DiabeticRetinopathy Mortality sits between DiabeticRetinopathy and Mortality
• LowerExtrmAmpt Mortality sits between LowerExtrmAmpt and Mortality
• DiabeticNephropathy Mortality sits between DiabeticNephropathy and Mortality
• DiabeticRetinopathy Incidence sits between DiabetesEpsd and DiabeticRetinopathy
• LowerExtrmAmpt Incidence sits between DiabetesEpsd and LowerExtrmAmpt
• DiabeticNephropathy Incidence sits between DiabetesEpsd and DiabeticNephropathy

These states aren't really states in a true sense. Rather, these states set the value of the transition rates around them. So, for example, DsFreeSus Disability is really the transition rate for DsFreeSus -> Disability. In the nomenclature of the Botech protocol, we call this a "Surrogate node". This is a structural decision we have made, but it doesn't change the results. Rather, we do this, so we can show how the calculations work to determine the disability and mortality effects.

The calculation component sets the transition rates

The "Surrogate Nodes" mentioned above, are set by a series of calculations in the model. These calculations follow the common order of operations described in the introduction.

The Diabetes model scenarios

Null

In the null scenario, the coverage of all treatments is set to its baseline in the first year of the projection, then 0% afterwards.

D1

Foot care to prevent amputation in people with diabetes (including educational programmes, access to appropriate footwear, multidisciplinary clinics)

In D1:

• NeuropathyScr is set at its baseline coverage for the first year (2019) of the projection, and then to 95% coverage for the rest of the projection.
• RetinopathyScrn continues at its baseline coverage for the entirety of the run
• NephropathyScr continues at its baseline coverage for the entirety of the run
• StdGlycControl continues at its baseline coverage for the entirety of the run
• IntsvGlycControl continues at its baseline coverage for the entirety of the run

D2

Diabetic retinopathy screening for all diabetes patients and laser photocoagulation for prevention of blindness

In D2:

• NeuropathyScr continues at its baseline coverage for the entirety of the run
• RetinopathyScrn is set at its baseline coverage for the first year (2019) of the projection, and then to 95% coverage for the rest of the projection.
• NephropathyScr continues at its baseline coverage for the entirety of the run
• StdGlycControl continues at its baseline coverage for the entirety of the run
• IntsvGlycControl continues at its baseline coverage for the entirety of the run

D3

Glycaemic control for people with diabetes, along with standard home glucose monitoring for people treated with insulin to reduce diabetes complications

In D3:

• NeuropathyScr continues at its baseline coverage for the entirety of the run
• RetinopathyScrn continues at its baseline coverage for the entirety of the run
• NephropathyScr continues at its baseline coverage for the entirety of the run
• StdGlycControl is set at its baseline coverage for the first year (2019) of the projection, and then to 95% coverage for the rest of the projection.
• IntsvGlycControl is set at its baseline coverage for the first year (2019) of the projection, and then to 95% coverage for the rest of the projection.

D5

Control of blood pressure in people with diabetes

In D5:

• NeuropathyScr continues at its baseline coverage for the entirety of the run
• RetinopathyScrn continues at its baseline coverage for the entirety of the run
• NephropathyScr is set at its baseline coverage for the first year (2019) of the projection, and then to 95% coverage for the rest of the projection.
• StdGlycControl continues at its baseline coverage for the entirety of the run
• IntsvGlycControl continues at its baseline coverage for the entirety of the run

Missing Diabetes Scenarios

The diabetes scenarios not mentioned here are actually modelled using the CVD model, and are mentioned there.

Order of Operations

The Diabetes model follows the common order of operations used by all disease models. The key Diabetes-specific variations are:

Treatment Effect Calculations (Steps 4-7)

Diabetes treatments primarily affect incidence rates rather than mortality. The model calculates effects for five treatments: StdGlycControl, IntsvGlycControl, RetinopathyScrn, NeuropathyScr, and NeprhopathyScr.

Step 4 example: RetinopathyScrn_PIN × RetinopathyScrn_DiabeticRetinopathy_Incidence_Impact × RetinopathyScrn_Calculated_Coverage

Step 6: Blended disability calculated for: DiabetesEpsd, DiabeticRetinopathy, LowerExtrmAmpt, and DiabeticNephropathy

Main State Transitions (Step 10)

Diabetes has a progressive, no-remission structure:

• DsFreeSus → DiabetesEpsd (diabetes onset)
• DiabetesEpsd → DiabeticRetinopathy, LowerExtrmAmpt, DiabeticNephropathy (complications)
• No transitions back to healthier states
• Prevalent populations moved in first year only

Interventions

Intervention Table

The modelled treatments for Diabetes

For Diabetes, there are five treatments. An intervention is something that has an effect on the main components of the model, such as disability, or mortality.

• StdGlycControl
  • Name in Spectrum: Standard Glycemic control
• IntsvGlycControl
  • Name in Spectrum: Intensive Glycemic control
• RetinopathyScrn
  • Name in Spectrum: Retinopathy Screening + photocoagulation
• NeuropathyScr
  • Name in Spectrum: Neuropathy screening and preventive foot care
• NeprhopathyScr
  • Name in Spectrum: Nephropathy screening and treatment

While treatments are always present in the structure of the Diabetes model, their coverage differs depending on the scenario.

Treatment Impacts

NOTE - These figures imply a modification of effect sizes. E.g. StdGlycControl reduces the transition to DiabeticRetpathy by 75% (0.75).

Population in Need

NOTE - Refers to the proportion of people in DiabetesEpsd who are "in need" of this treatment. e.g. 90% of DiabetesEpsd are "in need" of StdGlycControl

Assumptions

NOTE - A document is a difficult place to put entire lists of assumptions, as many of the assumptions we have change over time, and many of the assumptions are arrays of values, which apply to males and females differently, as well as different ages.

Therefore, please look at ./data/diabetes.csv as a reference guide for some assumptions. Values for disability weights have come from ./data/Diabetes.xlsx which is taken from Spectrum. Furthermore, even though measures of incidence, prevalence, and mortality may appear in this document, the final values were taken from ./data/GBD_Country_DATA.xlsx.

The baseline scenario is the default scenario

The baseline scenario has a coverage rate that is static, and continues from the start year to the end year. This is important, because it completely removes the effect of the Calculation Component. This is because, in essence, the effect of treatments is governed by the calculation: effect = impact * coverage * population in need. However, coverage is no the current coverage, but the difference between the current coverage, and the starting coverage. Therefore: effect = impact * (current_coverage - starting_coverage) * population in need. Because current_coverage - starting_coverage = 0, there is no effect to add to the default values for disability and mortality.

The null scenario reduces the coverage, and therefore the impacts

In the null scenario, all treatments are reduced from the baseline coverage to zero. For example, in Afghanistan, it is assumed that the baseline coverage rate is 5%. Therefore: effect = impact * (0 - 0.05) * population in need = impact * -0.05 * population in need. Therefore, in this country, the null scenario implies a 5% reduction in the impact of the four treatments.

The scale-up scenario increases the coverage, and therefor the impacts

In the scale-up scenario, select treatments are increased from baseline to 95% for the projection, starting in the second year. For the treatments that aren't selected, they are left at the baseline level, and thus do not contribute to effect calculations. Therefore, for a select treatment in Afghanistan: effect = impact * (0.95 - 0.05) * population in need = impact * 0.9 * population in need.