Comparing statistical methods to predict leptospirosis incidence using hydro-climatic covariables

TL;DR

This study compares semiparametric and classic time series methods, including ARIMA and ARIMAX, to predict leptospirosis outbreaks in Argentina using hydro-climatic data, aiming to improve outbreak forecasting accuracy.

Contribution

It introduces a comparison of SFPLR with ARIMA and ARIMAX for leptospirosis prediction, highlighting the potential of nonparametric methods with covariates.

Findings

SFPLR outperforms ARIMA in predictive accuracy.

ARIMAX effectively incorporates hydro-climatic covariates.

Nonparametric methods improve outbreak prediction models.

Abstract

Leptospiroris, the infectious disease caused by the spirochete bacteria Leptospira interrogans, constitutes an important public health problem all over the world. In Argentina, some regions present climate and geographic characteristics that favors the habitat of the bacteria Leptospira, whose survival strongly depends on climatic factors. For this reason, regional public health systems should include, as a main factor, the incidence of the disease in order to improve the prediction of potential outbreaks, helping to stop or delay the virus transmission. The classic methods used to perform this kind of predictions are based in autoregressive time series tools which, as it is well known, perform poorly when the data do not meet their requirements. Recently, several nonparametric methods have been introduced to deal with those problems. In this work, we compare a semiparametric method, called Semi-Functional Partial Linear Regression (SFPLR) with the classic ARIMA and a new alternative ARIMAX, in order to select the best predictive tool for the incidence of leptospirosis in the Argentinian Litoral region. In particular, SFPLR and ARIMAX are methods that allow the use of (hydrometeorological) covariables which could improve the prediction of outbreaks of leptospirosis.