^{1}

^{*}

^{2}

^{3}

^{4}

^{1}

^{1}

The authors have declared that no competing interests exist.

Conceived and designed the experiments: KG. Performed the experiments: KG DG YF. Analyzed the data: KG BK SM. Contributed reagents/materials/analysis tools: SK. Wrote the paper: KG BK SK.

In tropical and subtropical regions of eastern and South-eastern Asia, dengue fever (DF) and dengue hemorrhagic fever (DHF) outbreaks occur frequently. Previous studies indicate an association between meteorological variables and dengue incidence using time series analyses. The impacts of meteorological changes can affect dengue outbreak. However, difficulties in collecting detailed time series data in developing countries have led to common use of monthly data in most previous studies. In addition, time series analyses are often limited to one area because of the difficulty in collecting meteorological and dengue incidence data in multiple areas. To gain better understanding, we examined the effects of meteorological factors on dengue incidence in three geographically distinct areas (Ratnapura, Colombo, and Anuradhapura) of Sri Lanka by time series analysis of weekly data. The weekly average maximum temperature and total rainfall and the total number of dengue cases from 2005 to 2011 (7 years) were used as time series data in this study. Subsequently, time series analyses were performed on the basis of ordinary least squares regression analysis followed by the vector autoregressive model (VAR). In conclusion, weekly average maximum temperatures and the weekly total rainfall did not significantly affect dengue incidence in three geographically different areas of Sri Lanka. However, the weekly total rainfall slightly influenced dengue incidence in the cities of Colombo and Anuradhapura.

Dengue fever (DF) and dengue hemorrhagic fever (DHF) outbreaks occur in most tropical and subtropical regions and are the most important emerging arboviral diseases worldwide. The endemic area for dengue extends over 60 countries

In Sri Lanka, although dengue is endemic, the case fatality ratio (CFR) is below 1%; the number of adult cases have increased recently

At present, the causes and influencing factors of dengue epidemics are unknown in Sri Lanka. Previous studies demonstrate statistically significant associations between infectious diseases and meteorological variations such as rainfall and temperature. The effects of climate change on the endemics of infectious diseases such as cholera, malaria, and plague have been recognized

Time series analyses are often used in studies of the relationship between meteorological factors and disease and are most successful when data have been accumulated over long periods. However, it is extremely difficult to collect such meteorological and health data in developing countries. Although daily outcome data are desirable for time series analysis, obtaining such data from most developing countries is impossible

Fortunately, in Sri Lanka, the number of dengue cases is reported from all over the country, and meteorological data are collected and made readily available. Importantly, both these databanks contain weekly data. Thus, in the present study, we examined the effects of meteorological factors on dengue outbreak in Sri Lanka using time series analysis. Studies of dengue in Sri Lanka are few

We aimed to quantify in detail the association between meteorological variables and the frequency of notified cases of dengue in three geographically distinct areas (Ratnapura, Colombo, and Anuradhapura) of Sri Lanka using time series analyses of weekly data.

The climate of Sri Lanka is characterized as tropical and is traditionally divided into three climatic zones. In a large number of previous studies, time series analyses were performed in one study area owing to difficulties in data collection. However, it is desirable to compare several study areas when investigating the effect of meteorological factors on infectious diseases. In this study, using existing surveillance data, we quantified the association between meteorological variables and dengue incidence in three climatically different areas, namely Ratnapura, Colombo, and Anuradhapura districts (

Ratnapura district is located in the South-western part of Sri Lanka and is 101 km from Colombo in the Sabaragamuwa Province, which has a tropical rainforest climate and a population of 1 million. The average annual precipitation is approximately 4,000–5,000 mm in the valley (21 m above sea level) of the River Kalu Ganga, and the average temperature varies from 24°C to 35°C. Colombo district is the largest in Sri Lanka and is the administrative capital of the province located in the country's west coast. This region has a tropical monsoon climate and a population of 2.3 million. The average annual precipitation is approximately 2,400 mm, and the average temperature varies 28°C. Anuradhapura district is the capital of the North Central Province and one of the ancient capitals of Sri Lanka. It is located 206 km from Colombo. This district has a hot tropical climate and a population of 0.7 million. The average annual precipitation is approximately 1,300 mm, and the average temperature varies from 20°C to 30°C.

This study covers dengue incidence and meteorological data from 2005 to 2011 (7 years). However, complete data were only collected until the 39^{th} week, 52^{nd} week, and 48^{th} week of 2011 from Ratnapura, Colombo, and Anuradhapura, respectively. Meteorological data were collected by the Department of Meteorology in Sri Lanka and included the daily maximum temperature and total rainfall, which were acquired from Ratnapura (6.68N, 80.40E, 34.4 m), Colombo (6.90N, 79.87E, 7.3 m), and Anuradhapura (8.35N, 80.38E, 92.5 m) weather stations managed by the Department of Meteorology. We obtained these archived data from them. Because time series analyses were performed using weekly disease data, daily meteorological data were converted to weekly data.

Dengue incidence is reported in Sri Lanka through a national network that covers the whole country. These values are published as weekly epidemiological reports (WERs) by the Epidemiology Unit, Ministry of Health, Sri Lanka. In this study, weekly dengue incidence data were obtained from clinically diagnosed cases at this unit and excluded laboratory surveillance. In addition, data included both DF and DHF and were not divided into the four viral serotypes DV-1, DV-2, DV-3, and DV-4.

The ordinary least squares (OLS) method and the vector autoregressive model (VAR) were used in this study to examine the association between meteorological variables and the incidence of dengue from 2005 to 2011. In this study, OLS regression analyses were performed, and if serial correlations were revealed, these analyses were followed by VAR. VAR is one of the most flexible models for analyses of multivariate time series. The main advantage of VAR is that multivariate variables are both explained and explanatory variables. Hence, this model performs more accurate predictions using the relations between multiple variables

In addition, we also used impulse response function (IRF) to identify shock reactions to the maximum temperature and total rainfall. IRF tracks the impact of all variables on the others in the system

All analyses were performed using STATA version 12 (StataCorp. LP, College Station, USA).

Characteristics of meteorological variables and dengue incidence differed between study areas (

(A) Ratnapura (B) Colombo (C) Anuradhapura.

Total numbers of dengue cases recorded between 2005 and 2011 were 2720, 22231, and 2090 in Ratnapura, Colombo, and Anuradapura, respectively. Outbreaks occurred in 2009 and 2010 in all three areas, although in Colombo, the outbreak was most remarkable and continued in 2011.

OLS regression analyses were initially performed for each area and are shown in

Source | SS | df | MS | Number of observations = 352 |

Model | 5056.87238 | 2 | 2528.43619 | F (2, 349) = 6.04 |

Residual | 146061.446 | 349 | 418.514171 | Prob>F = 0.0026 |

Total | 151118.318 | 351 | 460.536519 | R-squared = 0.0335 |

Adj R-squared = 0.0279 | ||||

Root MSE = 20.458 |

dengue | Coeff. | Std. Err. | t | P>| t | | 95% Conf. | Interval |

Temp | −1.735021 | 0.7685898 | −2.26 | 0.025 | −3.246672 | −0.2233708 |

Rainfall | 0.0261063 | 0.0179327 | 1.46 | 0.146 | −0.0091634 | 0.061376 |

_cons | 67.12854 | 24.89586 | 2.70 | 0.007 | 18.16375 | 116.0933 |

Source | SS | df | MS | Number of observations = 365 |

Model | 69816.8803 | 2 | 34908.4401 | F (2, 349) = 7.96 |

Residual | 1588380.42 | 362 | 4387.79123 | Prob>F = 0.0004 |

Total | 1658197.30 | 364 | 4555.48710 | R-squared = 0.0421 |

Adj R-squared = 0.0368 | ||||

Root MSE = 66.24 |

dengue | Coeff. | Std. Err. | t | P>| t | | 95% Conf. | Interval |

Temp | −12.65323 | 4.052452 | −3.12 | 0.002 | −20.62253 | −4.683925 |

Rainfall | −0.1823936 | 0.0544296 | −3.35 | 0.001 | −0.2894315 | −0.0753558 |

_cons | 464.0644 | 125.5453 | 3.70 | 0.000 | 217.1746 | 710.9541 |

Source | SS | df | MS | Number of observations = 361 |

Model | 391.735052 | 2 | 195.867526 | F (2, 349) = 1.87 |

Residual | 37561.8328 | 358 | 104.921321 | Prob>F = 0.1561 |

Total | 37953.5679 | 360 | 105.426577 | R-squared = 0.0103 |

Adj R-squared = 0.0048 | ||||

Root MSE = 10.243 |

dengue | Coeff. | Std. Err. | t | P>| t | | 95% Conf. | Interval |

Temp | −0.4547149 | 0.2908249 | −1.56 | 0.119 | −1.026655 | 0.117225 |

Rainfall | −0.0209214 | 0.0124478 | −1.68 | 0.094 | −0.0454015 | 0.0035586 |

_cons | 21.32583 | 9.640052 | 2.21 | 0.028 | 2.308586 | 40.34308 |

Because positive serial correlations between the maximum temperature and total rainfall values were found in all areas using Durbin–Watson statistics, we performed OLS regression analyses using differences between variables without constant terms. Serial correlations were detected using the Breusch–Godfrey test (Ratnapura, Prob>chi2 = 0.000; Colombo, Prob>chi2 = 0.000; Anuradhapura, Prob>chi2 = 0.000) and Durbin's alternative test (Ratnapura, Prob>chi2 = 0.000; Colombo, Prob>chi2 = 0.000; Anuradhapura, Prob>chi2 = 0.000). Likewise, in OLS regression analysis using the lag model, both the Breusch–Godfrey test (Ratnapura, Prob>chi2 = 0.000; Colombo, Prob>chi2 = 0.000; Anuradhapura, Prob>chi2 = 0.001) AND Durbin's alternative test (Ratnapura, Prob>chi2 = 0.000; Colombo, Prob>chi2 = 0.000; Anuradhapura: Prob>chi2 = 0.001) identified serial correlations. Thus, OLS regression analyses were inappropriate for this study.

To test the assumption that time series data represent a stationary process, a test for stationary processes was performed before time series analysis. The Dickey–Fuller GLS unit root test indicated that the original series of each variable were non-stationary processes in all three areas, with the exception of the total rainfall at Ratnapura and Colombo. In addition, as shown in

(A) Logarithm of dengue incidence (B) Logarithm of maximum temperature (C) Logarithm of total rainfall.

(A) Logarithm of dengue incidence (B) Logarithm of maximum temperature (C) Logarithm of total rainfall.

(A) Logarithm of dengue incidence (B) Logarithm of maximum temperature (C) Logarithm of total rainfall.

To determine the appropriate number of lags to be used in VAR, the final prediction error (FPE) and the Akaike Information Criterion (AIC) were used as common selection criteria. Both FPE and AIC selected a lag of four in Ratnapura (FPE = 0.0001179; AIC = 1077038), a lag of four in Colombo (FPE = 0.000315; AIC = 0.452156), and a lag of three in Anuradhapura (FPE = 0.004152; AIC = 3.02952).

As shown in

Equation | Excluded | Chi2 | Prob>chi2 | ||||

Ratnapura | Colombo | Anuradhapura | Ratnapura | Colombo | Anuradhapura | ||

The Number of Dengue | Maximum Temperature | 1.31980 | 0.61225 | 0.09922 | 0.251 | 0.434 | 0.753 |

The Number of Dengue | Total Rainfall | 0.45196 | 3.79430 | 3.58700 | 0.501 | 0.051 | 0.058 |

The Number of Dengue | All | 0.33810 | 3.79760 | 3.64450 | 0.512 | 0.150 | 0.162 |

Maximum Temperature | The Number of Dengue | 0.10739 | 0.06836 | 2.85560 | 0.743 | 0.794 | 0.091 |

Maximum Temperature | Total Rainfall | 0.35354 | 0.01394 | 0.38072 | 0.532 | 0.906 | 0.537 |

Maximum Temperature | All | 0.47130 | 0.08630 | 3.03240 | 0.790 | 0.958 | 0.220 |

Total Rainfall | The Number of Dengue | 0.14717 | 1.33500 | 0.30285 | 0.701 | 0.248 | 0.582 |

Total Rainfall | Maximum Temperature | 0.04101 | 2.59390 | 2.53710 | 0.840 | 0.107 | 0.111 |

Total Rainfall | All | 0.18020 | 3.64130 | 2.95590 | 0.914 | 0.162 | 0.228 |

IRF analyses presented in

This manuscript defines the influence of meteorological factors on dengue incidence using time series analysis of the weekly average maximum temperature and total rainfall from 2005 to 2011 in three geographically distinct areas of Sri Lanka: Ratnapura, Colombo, and Anuradhapura. In this study, we conducted time series analyses using OLS regression followed by VAR in each of the three areas. To the best of our knowledge, this is the first study to examine the impact of meteorological variables on dengue incidence in Sri Lanka using time series analyses based on VAR. In addition, such analyses of weekly data from three geographically distinct areas are extremely rare.

The analyses in this study led to the conclusion that the weekly average maximum temperature and total rainfall do not significantly affect dengue incidence in Ratnapura, Colombo, or Anuradhapura. However, the total weekly rainfall slightly influenced dengue incidence in Colombo and Anuradhapura (Colombo, p = 0.051; Anuradhapura, p = 0.058).

The results of this study differ from those of previous studies that indicate an association between meteorological variables and dengue incidence

Furthermore, whereas monthly data have been used in most previous time series studies, the weekly data used in the present VAR method provided more detailed associations between variables. Nonetheless, the present data indicate that meteorological variables do not affect dengue incidence. Presumably, meteorological data are insufficient to explain regional and other complex factors that influence dengue incidence.

A disadvantage of this study is the absence of data corresponding to the four viral serotypes DV-1, DV-2, DV-3, and DV-4, which may have differential influences on population immunity. In Sri Lanka, DV-2 and DV-3 are currently the most common serotypes. Further time series studies are required to decipher the combined effects of serotype and climate on dengue incidence. In this study, we used time series analysis and developed statistical approaches to determine the impact of meteorological variables on dengue incidence in Sri Lanka. Further time series studies may include other complex factors such as population density, forest cover rate, and socio–economic status. We were unable to add the data of population density or the immigration and emigration ratio to this time series analyses because migration data were not reported at weekly intervals. Although the usage of data related to demography has been attempted in the study by time series analysis, most of these studies gave up this use owing to difficulty in obtaining this type of demographic data in a short interval.

In Sri Lanka, census is conducted only once for approximately 10 years, and the population of other years is estimated. According to the last two censuses (2001 and 2012) by the Department of Census and Statistics in Sri Lanka, the average annual growth rate in Ratnapura, Colombo, and Anuradhapura from 2001 to 2012 is 0.59%, 0.35%, 1.33%, respectively. The highest annual growth rate in Sri Lanka between the 2001 to 2012 period was reported from Anuradhapura. In contrast, the annual growth rate in Ratnapura and Colombo is below 1%. Kalutara district (1.23%) and Gampaha district (1.02%) of the Western Province, including Colombo, have also reported annual population growth rates of more than 1%. It is appears that people migrate from the urban areas of Colombo to these two neighboring districts for residence, which explains the higher annual growth rates. Therefore, social demographic change in each area must be considered as the analyzing data in time series analysis. Meanwhile, we need to give a great deal of thought to the difficulty in collecting demographic data in the case of short-interval time series analysis such as that in this study.

We would like to thank all public health professionals, Medical Officers of Health, and clinicians for their assistance in collecting and compiling the data related to dengue incidence. We would also like to thank the meteorological professionals for collecting and compiling the meteorological data.