Spatiotemporal dynamics of dengue epidemics, southern Vietnam

Abstract

An improved understanding of heterogeneities in dengue virus transmission might provide insights into biological and ecologic drivers and facilitate predictions of the magnitude, timing, and location of future dengue epidemics. To investigate dengue dynamics in urban Ho Chi Minh City and neighboring rural provinces in Vietnam, we analyzed a 10-year monthly time series of dengue surveillance data from southern Vietnam. The per capita incidence of dengue was lower in Ho Chi Minh City than in most rural provinces; annual epidemics occurred 1-3 months later in Ho Chi Minh City than elsewhere. The timing and the magnitude of annual epidemics were significantly more correlated in nearby districts than in remote districts, suggesting that local biological and ecologic drivers operate at a scale of 50-100 km. Dengue incidence during the dry season accounted for 63% of variability in epidemic magnitude. These findings can aid the targeting of vector-control interventions and the planning for dengue vaccine implementation.

Keywords: Dengue; Vietnam; arthropod vectors; epidemiology; heterogeneity; spatial distribution; surveillance; transmission; vector-borne infections; viruses.

Figure 1

Vietnam and the southern 19 provinces included in this analysis. The map shows current administrative boundaries; for our analysis, we aggregated 2 provinces (Can Tho and Hau Giang) to reflect the administrative boundaries before 2004. A, Lam Dong; B, Binh Phuoc; C, Dong Nai; D, Binh Duong; E, Tay Ninh; F, Ho Chi Minh City; G, Ba Ria – Vung Tau; H, Long An; I, Dong Thap; J, Tien Giang; K, Ben Tre; L, Tra Vinh; M, Vinh Long; N, Soc Trang; O, Can Tho; P, An Giang; Q, Kien Giang; R, Bac Lieu; S, Ca Mau.

Figure 2

Dengue time series from the 19 provinces and 159 districts in southern Vietnam, 2001–2010. A) Monthly aggregate time series of dengue cases reported from provinces. B) Monthly dengue incidence in each province; boldface line indicates Ho Chi Minh City. C) Monthly dengue incidence in each district. Data have been square-root transformed and normalized to zero mean and unit variance. Districts are ordered from north (top) to south (bottom) by first ordering provinces north to south, then ordering districts within each province, according to latitude of district centroid.

Figure 3

Wavelet analysis of dengue periodicity, 2001–2010. A) Left panel: wavelet power spectrum (WPS) of the aggregate monthly dengue time series for southern Vietnam (square-root transformed, normalized, and trend suppressed). Colors code for increasing spectrum intensity, from blue to red; dotted lines show statistically significant area (threshold of 95% CI); the black curve delimits the cone of influence (region not influenced by edge effects). Right panel: Mean spectrum (solid line) with its threshold value of 95% CI (dotted line) for the aggregate time series. B) WPS and mean spectrum for Binh Duong Province. C) WPS and mean spectrum for Bac Lieu Province. D) WPS and mean spectrum for Ca Mau Province. The wavelet power spectra for Binh Duong, Bac Lieu, and Ca Mau Provinces are shown because they were the only 3 provinces in which a dominant multiannual cycle was detected.

Figure 4

Correlation across provinces (A) or districts (B) between annual dengue incidence and variation in epidemic timing. Epidemic timing represents the pairwise interprovince or interdistrict delay between wavelet transformed annual dengue time series. The variation in epidemic timing is significantly correlated with the overall magnitude of transmission in that year; there is less variation (i.e., more synchrony) in the timing of dengue epidemics across southern Vietnam in high-incidence years than in low-incidence years

Figure 5

Spatiotemporal patterns in annual dengue epidemics in southern Vietnam. The phase interval (days) between the dengue time series in each province (A) relative to Ho Chi Minh City (HCMC) and each district (B) relative to District 1 in HCMC is shown by year. The largest negative values (dark purple) indicate the earliest locations for the annual dengue epidemics, zero (gray hatched) represents synchrony with the HCMC time series, and positive values (green) indicate dengue epidemics that occurred later than in HCMC.

Figure 6

Spatial coherence in the magnitude (A and B) and timing (C and D) of dengue epidemics in southern Vietnam. District data are shown in panels A and C, and province data in panels B and D. Solid lines represent the correlation between provinces/districts as a function of the distance between the centroids of those provinces/districts, in kilometers. Dashed lines represent 95% CIs, and the horizontal line is the overall correlation across southern Vietnam. Coherence in the magnitude and timing of epidemics was measured by pairwise correlation between provinces/districts in their standardized square root–transformed annual incidence and monthly phase series, respectively.

Figure 7

Dry season dengue incidence as a predictor of the magnitude of the subsequent dengue epidemic. Plots show the association between annual epidemic incidence (April–December) and the preceding dry season dengue incidence (January–March). For Ho Chi Minh City (HCMC), these definitions were a priori shifted 1 month later (May–January and February–April, respectively) because of the consistently later occurrence of the dengue epidemic season in HCMC. Each point represents 1 province (A) or district (B) and year, correlating the standard deviation from mean incidence in the rainy season against the standard deviation from mean incidence in the preceding dry season, in the same province or district. The solid line shows fitted values from a linear model of epidemic incidence against dry season incidence. We excluded 71 data points from the district analysis (B) because there were no dengue cases during the dry season.