Time series analysis of dengue incidence in Guadeloupe, French West Indies: forecasting models using climate variables as predictors

Abstract

Background: During the last decades, dengue viruses have spread throughout the Americas region, with an increase in the number of severe forms of dengue. The surveillance system in Guadeloupe (French West Indies) is currently operational for the detection of early outbreaks of dengue. The goal of the study was to improve this surveillance system by assessing a modelling tool to predict the occurrence of dengue epidemics few months ahead and thus to help an efficient dengue control.

Methods: The Box-Jenkins approach allowed us to fit a Seasonal Autoregressive Integrated Moving Average (SARIMA) model of dengue incidence from 2000 to 2006 using clinical suspected cases. Then, this model was used for calculating dengue incidence for the year 2007 compared with observed data, using three different approaches: 1 year-ahead, 3 months-ahead and 1 month-ahead. Finally, we assessed the impact of meteorological variables (rainfall, temperature and relative humidity) on the prediction of dengue incidence and outbreaks, incorporating them in the model fitting the best.

Results: The 3 months-ahead approach was the most appropriate for an effective and operational public health response, and the most accurate (Root Mean Square Error, RMSE = 0.85). Relative humidity at lag-7 weeks, minimum temperature at lag-5 weeks and average temperature at lag-11 weeks were variables the most positively correlated to dengue incidence in Guadeloupe, meanwhile rainfall was not. The predictive power of SARIMA models was enhanced by the inclusion of climatic variables as external regressors to forecast the year 2007. Temperature significantly affected the model for better dengue incidence forecasting (p-value = 0.03 for minimum temperature lag-5, p-value = 0.02 for average temperature lag-11) but not humidity. Minimum temperature at lag-5 weeks was the best climatic variable for predicting dengue outbreaks (RMSE = 0.72).

Conclusion: Temperature improves dengue outbreaks forecasts better than humidity and rainfall. SARIMA models using climatic data as independent variables could be easily incorporated into an early (3 months-ahead) and reliably monitoring system of dengue outbreaks. This approach which is practicable for a surveillance system has public health implications in helping the prediction of dengue epidemic and therefore the timely appropriate and efficient implementation of prevention activities.

Figure 1

Dashed red line: Weekly incidence rates of dengue (per 100,000) in Guadeloupe from January 2000 to December 2007 compared to crude meteorological variables for the same period: A) minimum temperature (blue diamond), maximum temperature (yellow cross) and average temperature (green square); B) relative humidity (blue area); C) weekly cumulated rainfall (blue solid line).

Figure 2

A and B) Autocorrelation function (ACF) and Partial ACF (PACF) plot of original dengue incidence. C and D) ACF and PACF plot of integrated dengue incidence. E and F) ACF and PACF of residuals after applying a SARIMA (0, 1, 1) (0, 1, 1)₅₂ model. The X-axis gives the number of lags in weeks and, the y-axis, the value of the correlation coefficient comprised between -1 and 1. Dotted lines indicate 95% confidence interval.

Figure 3

Natural logarithm of dengue incidence in Guadeloupe for 2007. Solid line (filled circle): observed values during the period. Dashed lines: univariate SARIMA (0,1,1) × (0,1,1)52 model; blue (circle): 52 weeks ahead values; red (triangle): 13 weeks ahead values; green (square): 4 weeks ahead values.

Figure 4

Cross correlation functions between dengue fever cases and meteorological variables after applying SARIMA models. The x-axis gives the number of lags in weeks. Dotted lines indicate 95% confidence interval. Only positive lags are taken into account a) Accumulated rainfall b) Minimum temperature c) Maximum temperature d) Average temperature e) Relative humidity.

Figure 5

Natural logarithm of dengue incidence in Guadeloupe for 2007. Solid line (filled circle): observed values during the period. Red (triangle): univariate SARIMA (0,1,1) × (0,1,1)52 model's 13 weeks ahead values. Blue (circles): multivariate SARIMA (0,1,1) × (0,1,1)52 model's 13 weeks ahead values with minimum temperature lag-5.

Figure 6

Natural logarithm of dengue incidence in Guadeloupe for the 2000-2007 period. Solid line: observed values during the period. Red (crosses): Fitted values from 2000 to 2006 and green (circles): predicted values for 2007 with multivariate SARIMA (0,1,1) × (0,1,1)52 model's 13 weeks ahead values with minimum temperature lag-5 as an external regressor and their 95% prediction intervals (Blue dashed lines).