Development and validation of a predictive ecological model for TB prevalence.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Nationally representative tuberculosis (TB) prevalence surveys are crucial for measuring disease burden but are resource-intensive.
• Developing alternative methods to estimate TB prevalence is essential, particularly for low- and middle-income countries.
• Existing survey data can be leveraged to create predictive models for indirect TB burden estimation.

Purpose of the Study:

• To develop and validate an ecological statistical model for indirectly estimating TB prevalence.
• To predict bacteriologically confirmed TB prevalence in low- and middle-income countries lacking survey data.

Main Methods:

• Utilized data from 30 national and 2 district-level TB prevalence surveys (50 data points) for model development.
• Employed ecological predictors including TB notification rates, co-morbidities, and socio-environmental factors.
• Developed a random-effects multivariable binomial regression model, validated using cross-fold techniques.

Main Results:

• A final model with five predictors was identified: continent, retreated TB case percentage, TB notification rates, population density, and proportion of the population under 15.
• The model demonstrated a very good average fit with R-squared of 0.92 in cross-fold validation.
• The model was used to predict TB prevalence in 63 low- and middle-income countries in Africa and Asia for 2015.

Conclusions:

• Predictive ecological modeling serves as a valuable complementary method for estimating TB burden.
• This approach can aid countries with limited empirical TB prevalence data.
• The model offers a practical alternative for assessing TB prevalence where direct surveys are not feasible.