Using a dengue temporal model to predict outbreaks in Singapore
A set of statistical models running on the Least Absolute Shrinkage and Selection Operator (LASSO) method is used to forecast the incidence of dengue in Singapore and to manage potential outbreaks, reveals an analyst.
The three key characteristics of the forecasting model is its ability to forecast dengue case trend and incidence up to 12 weeks ahead, it’s uncertainty in forecast is represented and it can be run in ‘real time’, said Janet Ong, an analyst with the National Environmental Agency (NEA), Singapore. As the system has the ability to run in ‘real time’, future automation of the data-handling process will allow forecasts to be established on a daily basis.
There are other prediction models using classical approaches such as the Autoregressive Integrated Moving Average (ARIMA), distributed lag non-linear model and time series multivariate regression model, which is a commonly used prediction model. Compared to the Seasonal Autoregressive Integrated Moving Average (SARIMA) and step-down linear regression models, the dengue temporal prediction model using the LASSO method has the best predictive accuracy for dengue incidence, said Ong.
Prior to developing the predictive model, key risk factors that are associated with dengue transmission were analyzed to determine which should be built into the model. The numerous risk factors were classified into six groups: climatology, environmental, entomological, virological, population and demographics.
Further elaborating on climatology, Ong said in the first 6 months of 2016, which had the highest average daily temperatures compared to past years, the population of Aedes aegypti mosquitoes was reduced. Upon further analysis, they discovered a complex, non-linear relationship between the sustained high temperature and decreased mosquito population.
As for virological risk factors, virus surveillance enables NEA to know which serotype is circulating among the Singaporean population at any point in time and to detect any switch in the predominant dengue virus serotype. This is important because a change in the predominant serotype is a precursor for dengue epidemic in Singapore, said Ong.
After assessing the risk factors, we then proceed to construct a predictive model that can provide early warnings of outbreaks, facilitate preparedness for public health response and aid in planning of long-term strategy.
Data is gathered from the Singaporean Ministry of Health (MOH), Department of Statistics and NEA to generate forecast using the dengue prediction model. The MOH provides information on the number of dengue cases, the Department of Statistics provides population data, and the NEA provides climate and vector data.
The NEA’s objective is to keep the incidence of dengue incidences low by keeping the mosquito population low and breaking the transmission chain. One of the ways to achieve the objective is via surveillance, and this is where the LASSO prediction model comes into play.