PREDICTING BIRTH TRENDS IN COLOMBIA POST-COVID-19 USING TIME SERIES MODELS
Author(s)
Andreina Jose Alamo, BSc Stat, Jair A. Arciniegas, MSc, Omar Escobar, MD MSc, Jorge LaRotta, MD, Juan Manuel Reyes, MSc.
Pfizer Colombia, Bogotá, Colombia.
Pfizer Colombia, Bogotá, Colombia.
OBJECTIVES: Population projections estimated by official entities in Colombia did not consider the most recent trends and exogenous economics variables. This study aims to forecast the number of newborns in Colombia for the period of 2025-2029 using historical data and sociodemographic variables.
METHODS: This analysis employed three statistical approaches: exponential smoothing (ETS and Theta models) and autoregressive integrated moving average models (ARIMA). The newborns monthly time series was extracted from the Colombian Statistics Bureau. The series were segmented into a training dataset (June 2012 to July 2024) and a test dataset (August 2024 to July 2025). Shapiro-Wilk normality test was conducted to assess the residuals distribution and Ljung-Box test were used to detect serial correlation. Considering the population dynamics framework, co-variables such as rate of women of childbearing age, unemployment rate, and covid period were included and were extracted from official entities. Exploratory and descriptive analysis were made to characterize the series. To evaluate goodness-of-fit, metrics such as the RMSE, MAPE and the Winkler score were calculated. The computation of yearly forecasts and its prediction intervals (PI) were developed through bootstrap.
RESULTS: By accuracy the best model was ETS with multiplicative errors, seasonality and damped additive trend. RMSE/MAPE/Winkler score were 518, 1.44 and 4,401. Nevertheless, only ARIMA models satisfied the residuals’ assumptions. The former suggests a forecast for 2025 of 427,545 births (PI95% 418,179-436,397), with a decrease of 5.8% from the previous year. This trend continues through 2029, when 379,212 births were projected (PI95% 276463.3- 501294.1). A SARIMA model [p=8,d=1,q=1, D=1] forecasted 422,141 births for 2025 (PI95% 418,179- 436,397) and (PI95% 276,463-501,294) for 2029.
CONCLUSIONS: Our model projected a decrease in the newborn population over the next 5 years but much sharper than official estimates. Additional considerations may be warranted to the impact of social factors when making future population projections.
METHODS: This analysis employed three statistical approaches: exponential smoothing (ETS and Theta models) and autoregressive integrated moving average models (ARIMA). The newborns monthly time series was extracted from the Colombian Statistics Bureau. The series were segmented into a training dataset (June 2012 to July 2024) and a test dataset (August 2024 to July 2025). Shapiro-Wilk normality test was conducted to assess the residuals distribution and Ljung-Box test were used to detect serial correlation. Considering the population dynamics framework, co-variables such as rate of women of childbearing age, unemployment rate, and covid period were included and were extracted from official entities. Exploratory and descriptive analysis were made to characterize the series. To evaluate goodness-of-fit, metrics such as the RMSE, MAPE and the Winkler score were calculated. The computation of yearly forecasts and its prediction intervals (PI) were developed through bootstrap.
RESULTS: By accuracy the best model was ETS with multiplicative errors, seasonality and damped additive trend. RMSE/MAPE/Winkler score were 518, 1.44 and 4,401. Nevertheless, only ARIMA models satisfied the residuals’ assumptions. The former suggests a forecast for 2025 of 427,545 births (PI95% 418,179-436,397), with a decrease of 5.8% from the previous year. This trend continues through 2029, when 379,212 births were projected (PI95% 276463.3- 501294.1). A SARIMA model [p=8,d=1,q=1, D=1] forecasted 422,141 births for 2025 (PI95% 418,179- 436,397) and (PI95% 276,463-501,294) for 2029.
CONCLUSIONS: Our model projected a decrease in the newborn population over the next 5 years but much sharper than official estimates. Additional considerations may be warranted to the impact of social factors when making future population projections.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
EPH31
Topic
Epidemiology & Public Health
Topic Subcategory
Public Health
Disease
SDC: Pediatrics