Towards a leptospirosis early warning system in northeastern Argentina

Abstract

Leptospirosis is a zoonotic disease with a high burden in Latin America, including northeastern Argentina, where flooding events linked to El Niño are associated with leptospirosis outbreaks. The aim of this study was to evaluate the value of using hydrometeorological indicators to predict leptospirosis outbreaks in this region. We quantified the effects of El Niño, precipitation, and river height on leptospirosis risk in Santa Fe and Entre Ríos provinces between 2009 and 2020, using a Bayesian modelling framework. Based on several goodness of fit statistics, we selected candidate models using a long-lead El Niño 3.4 index and shorter lead local climate variables. We then tested predictive performance to detect leptospirosis outbreaks using a two-stage early warning approach. Three-month lagged Niño 3.4 index and one-month lagged precipitation and river height were positively associated with an increase in leptospirosis cases in both provinces. El Niño models correctly detected 89% of outbreaks, while short-lead local models gave similar detection rates with a lower number of false positives. Our results show that climatic events are strong drivers of leptospirosis incidence in northeastern Argentina. Therefore, a leptospirosis outbreak prediction tool driven by hydrometeorological indicators could form part of an early warning and response system in the region.

Keywords: Bayesian modelling; El Niño; climate; early warning system; leptospirosis; outbreak prediction.

Figure 1.

Schematic representation of the transmission and reporting of leptospirosis. Leptospira infection in humans can be direct (resulting from direct contact with an infected animal) or indirect (via an environment contaminated by Leptospira bacteria). Meteorological variables (including precipitation and temperature) can influence environmental contamination. Once humans have become infected, this usually results in asymptomatic infection. However, around 10–30% of infections result in disease. The majority of those who do experience symptoms have a mild self-limiting disease, although in a small percentage of people (approximately 10%), clinical disease can be severe and result in hospitalization. Depending on the clinical and laboratory capabilities, a proportion of cases will be reported in the surveillance system. Early indicators of leptospirosis risk could feed into a climate-based EWS to produce short- and medium-term forecasts of leptospirosis risk.

Figure 2.

Leptospirosis cases in Entre Ríos and Santa Fe between January 2009 and December 2020. (a) Location of Entre Ríos and Santa Fe. Paraná River, one of the most important water bodies of South America, flows in the limit between the provinces. Meteorological information was obtained from stations belonging to the National Meteorological Service (points on the map). (b) Number of confirmed leptospirosis cases per 100 000 inhabitants per month recorded in each province and reported to the SIVILA system.

Figure 3.

Fitted versus observed leptospirosis cases per month. Posterior median (dashed lines) and posterior 95% credible intervals (shaded area) for number of leptospirosis cases per month in (a) Entre Ríos and (b) Santa Fe between January 2009 and December 2020. Observed values (solid line) were recorded by the national surveillance system. Estimates are presented for the random effects-only model, and the best-fitting ENSO and local climate models (dashed lines).

Figure 4.

ROC curves. Performance of candidate models in detecting outbreaks in (a) Entre Ríos and (b) Santa Fe. The points for which the probability thresholds correspond to the optimal balance between HR and FAR, i.e. closest to the point of perfect classification (0, 1), are circled and the values are displayed on the graph. The diagonal dashed lines represent the point at which the HR equals the FAR, indicating a model where the performance is no better than a coin toss.

Figure 5.

Schematic timeline demonstrating the two-stage computation of outbreak probability for March 2010 in Entre Ríos. Public health officials with access to surveillance reports and data from the national meteorological service would compute the probability of an outbreak ahead of its occurrence. In this example, (a) a first prediction with the Niño 3.4 index from December 2009 was computed after leaving data between January and December 2010 out of sample. Then, (b) a second prediction with precipitation and river height measurements from February 2010 was computed by excluding data from March 2010 until February 2011. Grey dashed line: outbreak threshold computed as the 75th percentile of cases observed in the target month excluding the number of cases in the year of interest.