Cholera in Bangladesh is closely linked to climate variability, particularly El Niño–Southern Oscillation. Warm sea surface temperature anomalies in the tropical Pacific during winter amplify post-monsoon autumn outbreaks. In Dhaka, the capital, incidence has been low over the past five years, but the strong 2015–2016 El Niño resembled the 1998 event, when the city experienced its largest outbreak in two decades. This work sits at the intersection of epidemiology and climatology: alongside it in the archive lies a study that treats the city as an organism and applies mathematics to dengue fever, addressing the same theme of climate-sensitive infections but with a different pathogen and approach. The authors test whether a climate signal can be used for advance forecasting, which for Dhaka has so far been done only statistically, without accounting for transmission mechanisms.
The researchers applied two models. The first, a mechanistic model, describes the dynamics of susceptible, infected, and recovered individuals (SIRS) in the city core. The force of infection includes a long-term trend, secondary transmission, and a constant environmental source. Seasonality is captured by splines, and the El Niño influence enters through the January temperature anomaly in the Niño-3.4 region, transformed by a sigmoid function. The model is stochastic, with gamma noise, and accounts for incomplete case reporting. Parameters were estimated by maximum likelihood using iterated particle filtering on data from 1995 to 2010. The second model, a statistical one, divides the city into districts (thanas) and groups them into core and periphery. Incidence in each district is coded into three levels: no cholera, low, and high. Transition probabilities between levels depend on season, the maximum level of neighbors, and the climate index. For the 2016 forecast, parameters estimated earlier on data up to 2008 were used. Forecasts were generated by running 1000 simulations from January of each year and updating parameters as new data arrived.
Both models reproduced the low incidence of 2011–2015 well: in none of those years did the probability of exceeding the 75th percentile rise above 29%, consistent with the absence of major outbreaks. For the August–December 2016 season, the mechanistic model gave an 86.5% probability of exceeding the 75th percentile (a major outbreak), while the statistical model gave 97.2%. For an extreme outbreak (exceeding the 95th percentile, as in 1998), the probabilities were 50.1% and 26.5%, respectively. The models also predicted exceeding the median with probabilities of 99% and 99.2%. Comparison with data from 2011–2015 showed that median forecasts were close to observations, although the models slightly overestimated the autumn peaks of 2012 and 2015. The statistical model overestimated more, likely due to the absence of a long-term trend. Analysis of June 2016 rainfall showed a mixed picture: above normal in northwest India and northern Bangladesh, below normal in the central part of the country, leaving uncertainty in the monsoon's development.
This work demonstrates that mechanistic transmission models incorporating climate signals can provide useful forecasts with long lead times. Unlike the neighboring study on dengue fever, which treats the city as an organism and uses mathematics to describe outbreaks, here the focus is on long-term forecasting of a specific disease with transmission mechanisms considered. This is the first application of a process-based model for forecasting cholera in Dhaka, whereas previously only statistical approaches were used for the city. The results highlight the value of sea surface temperature as an early indicator of risk, which could be extended to other climate-sensitive infections.
The topic could evolve toward refining the link between El Niño and regional rainfall in Bangladesh. Different El Niño events with the same temperature anomaly can affect the monsoon differently, so future models should incorporate additional event characteristics, such as its spatial structure or temporal evolution. Also promising is combining the long-term El Niño-based forecast with short-term forecasts based on local rainfall in July–August. Accumulating data on actual outcomes after forecasts will allow model calibration and improved accuracy.
The results directly affect Bangladesh's public health system, enabling advance preparation for a potential cholera outbreak. They are also useful for other countries where cholera is climate-sensitive and for researchers working on infectious disease forecasting.
Immediate next steps include monitoring the monsoon in July–August 2016 and comparing the forecast with actual incidence after the season ends. If the forecast does not materialize, it will be necessary to determine whether this is related to the characteristics of the current El Niño, public health interventions, or changes in the pathogen's ecology.
This work addresses a fundamental problem in epidemiology: how predictable outbreaks of infectious diseases are under the influence of external factors. For cholera, the role of the environmental reservoir and its response to climate events remains an open question. It is also unclear why, after a long period of low incidence, the system may respond differently to similar climate forcings. These questions relate to the broader issue of nonlinearity and threshold effects in infection dynamics.
The sea surface temperature anomaly in the Niño-3.4 region in January 2016 was 2.60°C, almost matching the 2.56°C anomaly in January 1998, when Dhaka experienced its largest cholera outbreak in 20 years and a flood that submerged more than half the city.