Table 1 Overview of modelling approaches for subnational TB
From: Subnational burden estimates to find missing people with tuberculosis: wrong but useful?
Modelling paradigm | Type of reasoning | Use of empirical data | Role of chance | Examples of models |
|---|---|---|---|---|
Mathematical: Produces definite TB outcomes based on pre-specified inputs and relationships with other TB or non-TB data | Deductive: relies on known relationships and specific inputs to yield TB outcomes | No empirical data is necessary about the TB indicator to be modelled as long as relationships with other TB or non-TB data can be formulated | Variation can be introduced in the model simulations to account for inherently random effects and reflect variability in model inputs | Deterministic and stochastic compartmental models Individual-based deterministic and stochastic models |
Statistical: Uses existing empirical TB data to estimate probabilities and make predictions for unobserved cases | Inductive: Draws general conclusions about a TB outcome from specific observations and associations | Empirical data for the TB indicator to be modelled is necessary for certain areas for model building (the same areas for which risk factors are also available) | Variation inherent to model building as associations between the TB outcome and risk factors are estimated based on probability theory | Frequentist regression models Bayesian approaches Machine learning |