The microsimulation model

The Future Adult Model (FAM) projects 30-year impact of education campaigns on cardiometabolic outcomes among adults aged 25 years and older. It simulates life trajectories through transition probabilities across health states (e.g., incidence of diabetes or hypertension) and economic states in two-year increments. These transition equations are estimated primarily using the Panel Study of Income Dynamics (PSID), a longitudinal survey (2001–2015) representative of US households. The type 2 diabetes and hypertension transition model estimates are shown in Table 1.

Two-year Type 2 diabetes and hypertension transition estimates, 2001–2015 PSID

Standard errors in parentheses

We begin simulations with a cohort of adults aged 25 years and over in 2009 (we start simulations before the assumed interventions to allow the simulation model to stabilize rather than depend on the starting cohort), and age them until death. A replenishing cohort of adults aged 25–26 years old is added every period (two years) until 2049. These replenishing cohorts are synthetically generated to match trends based on population demographics from the census and behavioral changes based on the National Health Interview Study (NHIS) and the National Health and Nutrition Examination Survey (NHANES). Additionally, we categorized BMI and education attainment at the beginning of each 2-year period to align with those in NHANES. The Future Adult Model is a full life-cycle model for adults, but not for childhood, adolescence. Later life outcomes can be better predicted starting in adulthood (particularly, once educational achievements are completed). Efforts to create a full life-cycle microsimulation model have been less successful – meaningful outcomes for youth and young adults tend to be on shorter time scales and result in different transition functions than for adults.^17–19

Figure 1 illustrates the model schematic. A validation study found that BMI and diabetes projections compared to well PSID and Behavioral Risk Factors Surveillance System for men and women, except for the highest BMI levels.²⁰ Simulations are implemented in C++ and run on Linux, Windows, and Mac OS X. Graphs were created with Stata. The replenishing cohort model is estimated using the Stata CMP package.²¹ Transition and cross-sectional models are estimated in Stata (Stata Statistical Software: Release 16. College Station, TX: StataCorp LLC). The timeline of data used for transition equations, the base population, the intervention baseline, and projected future period are presented in eFigure1 (Appendix). A complete description of the model, including the transition equations and parameters, is available in the Technical Documentation. ²²

Model schematic