AboutSubmit my research

Health

Research articles · Dec 31, 2016

NCDMod: A microsimulation model projecting chronic disease and risk factors for Australian adults

Sharyn Lymer, Deborah Schofield, Crystal M Y Lee, Stephen Colagiuri

Faculty of Pharmacy, University of Sydney, Australia; The Boden Institute of Obesity, Nutrition, Exercise & Eating Disorders, University of Sydney, Australia

Volume 9(3) Winter 2016

Cite as: S. Lymer, D. Schofield, C. Lee, S. Colagiuri; 2016; NCDMod: A microsimulation model projecting chronic disease and risk factors for Australian adults; International Journal of Microsimulation; 9(3); 103–139. doi: 10.34196/ijm.00145

Abstract

Chronic disease and the associated risk factors are an ongoing public health concern. They represent a major burden of disease and result in significant health expenditure. The purpose of this paper is to describe the development and attributes of the microsimulation model NCDMod, an Australian, purpose-built model that simulates multiple chronic diseases and associated risk factors. The model offers flexibility in perspective and population of interest. The model projects out to 2025 in 5-year increments using a combination of static and dynamic methods. Transition equations are sourced from the literature and in-house analysis. NCDMod is able to capture the impacts of population ageing and shifts in incidence of specific chronic diseases and risk factors. Results from NCDMod will help to fill gaps in current Australian evidence on the population level impacts of interventions to prevent/delay chronic health conditions particularly diabetes, heart disease and obesity. Further, in combination with Health&WealthMod2030 this model will be able to offer evidence regarding the potential impact of health interventions on the long-term costs (both from an individual and government perspective) of ill health leading to unplanned retirement. In the consideration of disease interventions, the additional workers who may be able to continue to work due to delayed/prevented chronic disease can be modelled.

This article contains 2 figures and 8 tables.

Figures

Figure 1.
Figure 2.

Tables

Table 1.
Model Module Variable
NCDMod Basefile
Demography Age
Sex
Ethnicity
Family type
Socio-economic status
Chronic Illness Diabetes
Cardiovascular disease
Peripheral vascular disease
Arthritis
Cancer
Asthma
Renal failure
Depression
Dementia
Risk Factors BMI category
Reported high cholesterol
Reported high blood pressure
Reported high blood sugar
Smoking status
Alcohol consumption
Physical activity level
Population Projection
Demographics Age
Sex
Health Risk Factor and Disease Transition Modelling
Risk Factors BMI
Total cholesterol
Systolic blood pressure
Smoking status
Chronic Illness Diabetes incidence
Heart disease events
Stroke events
Mortality CVD deaths
Outcomes
Health Outcomes YLL (Years of life lost)
DALY
Expenditure Intervention costs
Health System costs
Scenario Comparator ICER (Incremental cost effectiveness ratio)
Table 2.
2010 Baseline Characteristics Gender
Obese
Male (%) Female (%) Total (%)
BMI < 30 kg/m2 72.0 73.6 72.8
BMI ≥ 30 kg/m2 28.0 26.4 27.2
Known T2DM
Known T2DM 5.7 6.7 6.3
Not 94.3 93.3 93.7
Ever had a heart attack
Yes 3.3 1.4 2.3
No 96.7 98.6 97.7
Ever had a stroke
Yes 1.6 1.4 1.5
No 98.4 98.6 98.5
Overall 100.0 100.0 100.0
Total Population (in thousands) 8,194 8,764 16,958

Source: NCDMod version June 2016.

Notes: BMI = Body mass index; T2DM = Type 2 diabetes mellitus.

Table 3.
Baseline Distribution Gender
Male Female Total Population
Age 2010 (in years)
Mean 46.5 48.7 47.6
Median 45.0 47.0 46.0
Standard Deviation 77.29 82.10 79.97
BMI2010 (kg/m2)
Mean 27.9 26.9 26.4
Median 27.1 25.4 26.4
Standard Deviation 25.91 29.02 27.65
Systolic Blood Pressure 2010 (mmHg)
Mean 130.0 127.4 128.7
Median 125.8 121.0 123.7
Standard Deviation 56.86 74.51 66.84
Total Cholesterol 2010 (mmol/L)
Mean 5.5 5.6 5.6
Median 5.4 5.5 5.4
Standard Deviation 2.66 2.53 2.60

Source: NCDMod version June 2016.

Notes: BMI = Body mass index.

Table 4.
Population Characteristics NCDMod AHS % Difference
% 65 years and over 19.4 17.8 9.0
% male 48.3 49.4 -2.2
% type 2 diabetes (self-reported) 6.3 5.8 8.6
% obese (BMI ≥ 30 kg/m2) 27.2 28.8 -5.6
Population Size 16,957,458 16,393,424 3.4

Source: NCDMod version June 2016.

Notes: BMI = Body mass index; AHS = Australian Health Survey 2011.

Table 5.
Outcome 2010 2015 2020 2025
BMI (kg/m2)
Total 27.4 28.1 28.7 29.4
Female 26.9 27.7 28.4 29.2
Male 27.9 28.5 29.0 29.7
SBP (mmHg)
Total 128.7 130.3 130.9 131.9
Female 127.4 129.0 129.9 131.2
Male 130.0 131.6 132.0 132.7
Total Cholesetrol (mmol/L)
Total 5.6 5.6 5.5 5.4
Female 5.6 5.6 5.5 5.5
Male 5.5 5.5 5.5 5.4

Source: NCDMod version June 2016.

Notes: BMI = Body mass index; SBP = systolic blood pressure.

Table 6.
Variable 2011–2015 2016–2020 2021–2025
New Known Type 2 Diabetes 261,496 288,503 325,301
CVD Events (excl. CVD deaths) 461,566 600,715 697,050
Years Lived in Disability 1,380,228 1,826,728 2,085,673
CVD Deaths 258,948 378,091 445,225
Years of Life Lost 2,527,629 3,276,962 3,669,757
New Known Type 2 Diabetes Rate (per 100,000 population) 287.6 285.0 296.3
CVD Events (excl. CVD deaths) Rate (per 100,000 population) 507.6 593.4 635.0
CVD Deaths Rate (per 100,000 population) 284.8 373.5 405.6

Source: NCDMod version June 2016.

Notes: CVD = cardiovascular disease.

Table 7.
Population Characterstics NCDMod 2015 NHS 2014/15 Absolute % point difference %Difference
% Ever Smoke 20.9 17.4 3.5 20.1
% Obese (BMI ≥ 30 kg/m2) 26.1 27.9 1.8 -6.5
% High Blood Pressure 18.4 21.5 3.1 -14.4
% High Total Cholesterol 34.3 32.8 1.5 4.6
% T2DM 5.5 5.1 0.4 7.8
% CVD prevalence 3.2 5.2 2.0 -38.5

Source: NCDMod version June 2016 and ABS NHS 2014/15 (Australian Bureau of Statistics, 2015b).

Notes: Adults aged 18 years and over, aligning with ABS definitions used in reporting. T2DM = Type 2 diabetes mellitus. CVD = Cardiovascular disease.

Model Element HealthAgeingMod NCDMod
Basefile NHS 2005 NHS 2005
Cloning of Records No Yes
Number of records in basefile Approximately 26,000 records Approximately 1.15 million records
Basefile reweighting No SAS macro GREGWT (Bell, 2000) was used to reweight the basefile to 2010 based on benchmarks of ABS ERP 2010 in 5-year age group-sex numbers; ABS 2011 in 5- year age group-diabetes numbers, 5-year age group- obesity numbers, 5- year age group high blood pressure numbers and 5- year age group-smoker numbers.
Projection reweighting Reweighted 2005 to 2010 using ABS population projections series B(Australian Bureau of Statistics, 2008) routinely. Reweighted 2010 to 2015, 2020 and 2025 using ABS population projections series B (Australian Bureau of Statistics, 2013b) routinely.
Alignment Alignment of CVD deaths according to the trend of ABS causes of death data 2000–2006, such that in 2005 there were approximately 35,000 CVD deaths (Walker et al., 2013) Alignment of number obese by age-sex groups in 2015 using GREGWT and benchmarks from 2014/15 ABS NHS data.
Age: imputation of single years Monte Carlo simulation with allocation across each single year within the 5 year age group, with 20% going into each year Monte Carlo simulation by gender, with allocation across each year within the 5 year age group based on ABS ERP 2010 as the distribution of age within 5 year group.
Imputation of parental history of diabetes No Allocated by Monte Carlo simulation. Age based probability from AUSDIAB00.
Ethnicity Based on NHS05 country of birth variable made binary Australian born or not. Additional flag of ethnicity to mark “Southern European, Asian, ATSI” is a random allocation of 0.33 those not “Australian born”.
Education level Binary variable – post school or not based on NHS05 education data Same as HealthAgeingMod.
BMI: imputation of continuous values Imputed randomly using uniform distribution with 8 BMI bands on NHS05 Same as HealthAgeingMod.
BMI transition Models BMI at end of 5 year period using OLS: 3.225 – 0.021 * (AGE) -0.135 * HBA1C * HBA1C scale + 1.000 * BMIvalue – 0.005 * SBP – 0.147 * cholesterol + 0.129 * female Based on USDIAB 00–05 Models BMI change across 5 year period: 2.694 – 0.022 * (aget+1) + 0.151 * sex -0.047 * cholesterol -0.005 *SBP + 0.192 * smoker -0.354 * diab2 -0.092 * ED level
Based on AUSDIAB 00–05.
Smoking status transition Nil Models smoking status at end of 5 year period using logistic regression to determine probability.
If smoker: 1.429 – 0.012 * age10 -0.012 * male +0.092 * EDlevel
If non-smoker: 2.116 + 0.056 * age10 -0.010 * male -0.372 * EDlevel Monte Carlo simulation used to allocate state.
Total cholesterol: impute continuous value OLS modelling to allocate a total cholesterol value.
If high cholesterol: 5.339 + 0.002 * AGE + 0.012 * BMI value + 0.048 * hba1c Value + 0.003 * BP value -0.008 * ED level
If normal cholesterol: 4.038 + 0.005 * AGE + 0.012 * BMI value -0.036 * hba1c Value + 0.002 * BP value -0.037 * Exercise_3+0.005 * ED level Based on AUSDIAB 2000.		 If self-reported high cholesterol:
Exp(1.635 – 0.0026*male + 0.003 * age + 0.007 * BMI value -0.0001 * BMIvalue2 + 0.007 * smoke + 0.015 * HB -0.003 * diabetes).
If not self-reported high cholesterol: Exp(1.262 + -0.028*male + 0.001 *age + 0*age2 + 0.023 * BMI value -0.0003 * BMI value2 + 0.032 * smoke + 0.039 * High BP -0.206 * diabetes) Based on AHS2011 data.
HDL Cholesterol: impute continuous value No Modelled using OLS with exponential transformation: Exp(0.599 + 0.001*age – 0.088 * male -0.074 * smoker + 0.033 * e3 -0.016 * BMI value + 0.036 * Cholesetrol value + 0.040 * male_smoke – 0.004*male_bmi).
Systolic blood pressure: impute continuous variable If High BP: 128.185+0.346*AGE + 1.007*female + 0.139*BMI value+1.665*E1
If normal BP: 99.183+0.212*AGE-5.521*female+0.582*BMI value+1.153*Hi Cholesterol – 0.749*Exercise_2-1.093*SMOK- 0.071*ED level
Based on AUSDIAB 2000.		 Same as HealthAgeingMod.
Systolic blood pressure transition Model SBP at end of 5 year period: 23.551 + 0.201*(AGE)+0.437*BMI value * BMI scale +0.618 *BP value * BP scale – 2.703 * female
Based on AUSDIAB 00–05.		 Annual change in SBP modelled as:
If aged less than 35 years there is no change.
If female aged 35 and over: 35–45 years.
:2.445+0.107*(SBP -117.891
45–55 years : 2.278 +0.039*(SBP -127.312)
55–65 years :2.052+- 0.003*(SBP -137.656);
65y.o.+ : 2.296 + 0.024*(SBP -125.177);
If male aged 35 and over: 35–45 years: 1.321 + 0.015*(SBP -124.825);
45–55 years: 1.1915 + 0.064*(SBP -127.948);
55–65 years: 2.860 + 0.058*(SBP -128.675);
65 years + : 1.85 + 0.048*(SBP-126.722);
Based on Wu et al. (1980). If on antihypertensive medication then SBP remains unchanged.
Systolic blood pressure medication Nil If SBP greater than 160 allocated to start antihypertensives.
Diabetes duration Allocated such that:
If 35<=AGE<55 then duration is 7.5 years
If 55<=AGE<65 then duration is 10 years
if aged 65 years and over then duration is 15 years		 Same as HealthAgeingMod.
Impute has undiagnosed diabetes Monte Carlo simulation amongst those without diabetes
Probability based on logistic regression model: -29.242+0.038*AGE-0.073*female+3.783*hba1c Value+0.030*BMI value +0.021*BP value-0.048*Hi Cholesterol -0.164*E1- 0.143*Exercise_2-(-0.164- 0.143)*Exercise_3 – 0.057*SMOK+0.070*ED level		 Same as HealthAgeingMod.
Transition to self- reported diabetes Logistic regression model: – 23.200 – 0.020*(AGE) +0.835*female + 3.344*hba1c + 0.041*BMI + 0.011*SBP +0.403*SMOK
Based on AUSDIAB Same as HealthAgeingMod.
Transition to known diabetes Logistic regression model: – 28.826 + 3.887 * hba1c + 0.078 * BMI + 0.015 * SBP Based on AUSDIAB Random effects model of probability develop known diabetes across 5 year period:
-9.429 -0.063 * age10 – 0.717 * sex + 0.022 * BMI + 0.015 * SBP -0.449 * ed level – 0.657 * ethnic + 0.092 * previous Hx Diab – 0.344 * total cholesterol + 0.005 * age * bmi + 0.841* sex * previous Hx Diab Based on AUSDIAB00-05-11.
Year diabetes commenced Random allocation so 50% in year 2 and 50% in year 4 of five year transition period. Random allocation into single years across the 5 year period.
Impute has IFG Monte Carlo simulation amongst those without diabetes or undiagnosed diabetes.
Probability based on logistic regression model : -15.915 + 0.017*AGE-0.947 * female + 2.172 * hba1c Value + 0.063 * BMI value -0.003 * BP value – 0.111 * Hi Cholesterol – 0.027 * Exercise_1 + 0.064 * Exercise_2-(-0.027+0.064) * Exercise_3 -0.109 * SMOK + 0.038 * ED level		 Same as HealthAgeingMod.
Impute pre-diabetes IGT
Monte Carlo simulation amongst those without diabetes or undiagnosed diabetes.
Probability based on logistic regression model : -16.650 + 0.029 * AGE + 0.308 * female + 1.924 * hba1c Value + 0.050 * BMI value + 0.014 * BP value – 0.031 * Hi Cholesertol-0.094 * Exercise_1 – 0.087 * Exercise_2-(- 0.094 * E1 – 0.087) * Exercise_3 – 0.065 * SMOK -0.138 * ED level		 Same as HealthAgeingMod.
Life expectancy to calculate YLL By gender single years based on Australian Government Actuary Life Tables 2005–07 By gender single years based on ABS life tables 2010–12.
HbA1c impute continuous value If has self-reported high sugar model hba1c Value = 7.172 -0.0235* AGE + 0.013 * BP value – 0.408 *
Hi Cholesterol
If self-report not having high sugar levels then hba1c Value = 4.455 + 0.007 * AGE -0.070 * female + 0.0113 * BMI value +0.0003 * BP value + 0.032 * Hi Cholesterol + 0.035 * Exercise_1 + 0.029 * Exercise_2 + 0.013 * SMOK -0.004.		 Same as HealthAgeingMod.
HbA1c transition OLS to calculate the new HbA1c value at the end of 5 year projection 1.545 + 0.002 * (AGE in 2010) + 0.691 * HBA1C value + 0.008 * BMI value
Based on AUSDIAB 00–05.		 Same as HealthAgeingMod.
Impute Left Ventricular Hypertrophy (LVH) status Nil Logistic regression model to model probability: -7.167 + 0.054 * age + 6.043 * male -0.156 * BMI value + 0.027 * BP value- 0.075 * male_age Based on AUSDIAB 05. Monte Carlo simulation to allocate condition.
CVA incidence Logistics regression model: -9.550 + 0.050 * AGE + 0.027 * BMIvalue
Based on AUSDIAB 00–05.
Monte Carlo allocation of events.		 Non-proportional hazards Weibull accelerated failure time model to calculate probability of CVA event: 26.512 + 0.202 * female – 2.374 * log(age) -2.464 * log(SBP) – 0.391 * SMOK -0.023 * log(total chol/HDL) -0.309 * diab2 – 0.263 * female * Diab2 – 0.236 * LVH
Based on Framingham ( Anderson et al., 1991). Monte Carlo simulation to allocate event
Allocate year of CVA incident Random allocation so 50% in year 2 and 50% in year 4 of five year transition period. Random allocation into single years across the 5 year period
Heart disease incidents Probability of event in 5 year period using logistic regression model: -10.021 + 0.046 * AGE – 1.031 * female + 0.051 * BMI + 0.016 * SBP – 0.478 * CVD history + 0.346 * ED level Based on AUSDIAB 00–05. Monte Carlo allocation of events Non-proportional hazards Weibull accelerated failure time model: 11.471 + 10.511 * female – 0.797 * log(age)-5.422 * log(age)*female + 0.710 * log(age)*log(age)*female – 0.662 * log(SBP)- 0.268 * SMOK -0.428 * log(total chol/HDL)-0.153 * diab2 -0.117 * female * diab2 -0.159 * LVH*male;
Based on Framingham (Anderson et al., 1991). Monte Carlo simulation to allocate event
Allocate year of heart disease event Random allocation so 50% in year 2 and 50% in year 4 of five year transition period. Random allocation into single years across the 5 year period
CVD deaths Used Anderson (1991) Equation. Adjustment of population weights to reflect total population after these records removed for the next time period due to their “death”. Used Anderson(1991) Equation to flag those records to reflect the deaths in the 5 year transition period. The records are reused with the population weights to reflect the whole population for the end of transition period
DALY weights Used Begg et al. (2007) and Lopez et al. (2006) as Source. Used GBoD study 2010 (Salomon et al., 2012) as basis of estimates
Imputation of children’s risk factors i.e. the records that become adults across the projection period Imputes BMI to new 18+ year olds in proportion to NHS 2005 distribution of BMI categories for 18–19 years. Individual given mid- point for continuous value of BMI Total Cholesterol level allocated as 4.7
HbA1c allocated as 4.8 Systolic blood pressure allocated as 120		 BMI imputed by gender, allocate BMI category based on distribution of AHS 2011. Then randomly allocated continuous variable using uniform distribution across group BMI band they in which they were allocated. Allocated exercise as 180 minutes for boys and 140 minutes for girls
Used ABS AHS 2011 age 18–34 years with 0.5 prevalence of diabetes to allocate T2 diabetes
Future adult education was randomly allocated using probabilities from logistic regression model :-1.347 + 0.033*sex + 0.651 *obese Rest of risk factors as per HealthAgeingMod.
Annual disease costs Models the direct health costs associated with diabetes and CVD incidents. Both the cost of diabetes and the cost of CVD were Sourced from the literature. Persons with both diabetes and a CVD event in the time period had the costs added together and given 20% extra cost.
Direct health costs based on if has diabetes and/or CVD.
Costs Sources mainly from Clarke et al. (2008)
0.20 factor added to those with both CVD and diabetes.		 Based on AUSDIAB 05 for health service usage and costs based on MBS, PBS and AIHW hospital costs data. The equation was estimated using a linear regression model. The predictive variables included were: BMI, diabetes status and CVD status. This method provides a cost across the complete Australian population, calculating individual annual health system costs: 7599.91 + 3112.04*diab2 + 9330.60*CVD – 1830.25*BMI_normal – 1456.55*BMI_overweight – 4762.96 * age25 -3141.65 * age45
Language/software used to write the code and run the scenarios SAS v9.2 SAS v9.4