Uncertainty Analysis in Population-Based Disease Microsimulation Models

Table 2

Proposed steps for uncertainty analysis in population-based microsimulation models and corresponding steps for POHEM-OA uncertainty analysis example.

Steps for performing uncertainty analysis in population-based microsimulation models	Steps for uncertainty analysis in POHEM-OA
Step  1. Selecting the outcome	Sex-specific prevalence in (2001–2021)
Step  2. Establishing the UA list of parameters	(1) Hazard ratios for each BMI-category; (2) BMI progression
Step  3. Assigning a probability density function to parameters (or use bootstrap sampling)	(1) Lognormal distribution for hazard ratios with sex-specific correlation matrices; (2) Using 8 alternative regression models using bootstrap weights from NPHS^§ (1996–2006)
Step  4. Applying Monte Carlo method
(i) Select a sampling approach (or use bootstrap sampling).	(1) Latin hypercube sampling; (2) Random sampling with replacement (among eight alternate sets of parameters)
(ii) Implement calibration into MC method.	Calibrate on incidence by age and sex; implement the calibration algorithm into the MC method, using squared error criteria for convergence.
(iii) Calculate 𝑛^∗, 𝑚^∗ for precision level 𝑑.	Result: 𝑚0=20,000,000 population size; 𝑛0=5 (for the initial run); 𝑚^∗=500,000 people; 𝑛^∗=785 MC-runs for 𝑑=0.01; for 12-hour run of a PC with 12 GB memory, CPU = i7-980 Intel, 3.3 GHz.
Step  5. Constructing the outcome distribution	Please see Figure 3.

^§NPHS: National Population Health Survey, a longitudinal household survey by statistics Canada with a sample of size of over 17,000 persons that started in 1994 [49].