North to south gradient and local waves of influenza in Chile

Abstract

Influenza seasonality is caused by complex interactions between environmental factors, viral mutations, population crowding, and human travel. To date, no studies have estimated the seasonality and latitudinal patterns of seasonal influenza in Chile. We obtained influenza-like illness (ILI) surveillance data from 29 Chilean public health networks to evaluate seasonality using wavelet analysis. We assessed the relationship between the start, peak, and latitude of the ILI epidemics using linear and piecewise regression. To estimate the presence of incoming and outgoing traveling waves (timing vs distance) between networks and to assess the association with population size, we used linear and logistic regression. We found a north to south gradient of influenza and traveling waves that were present in the central, densely populated region of Chile. Our findings suggest that larger populations in central Chile drive seasonal influenza epidemics.

Figure 1

Hospital location and ILI rates, Chile 2010–2016. Left panel. Location of hospitals included through the Chilean territory. Right panel: Y axis, represent Hospitals ordered by latitude corresponding to the grouped zones marked by horizontal lines. X axis, time in days from 2010 to 2016. Gradient colors represent rates of ILI per 100,000 population. Z-score of log10 transformed shown for better visualization^,.

Figure 2

Detrended ILI incidence rate, wavelet reconstruction, local wavelet power spectrum and average power per period, Chile 2010–2016. (a) Detrended ILI incidence rate time series and wavelet reconstructed time series (red). Wavelet reconstruction using periods between 46 − 53 weeks, for Chile 2010–2016. (b) Local wavelet power spectrum of ILI rates for periods and time in weeks in Chile between 2010–2016. Significant power against white noise presented inside black contour. Ridge shown in white. A significant and high power was present through the complete time series for periods between 40 − 60. (c) Average power for periods, representing the average of local power from panel b with significance levels < 0.01 (red) and < 0.05 (blue).

Figure 3

Power per period in hospitals and 95th percentile of power z-score across period and hospitals, Chile 2010–2016. (a) Power per each of the 65 hospitals presented across periods. Dark red represent higher power, yellow represents low power and fit, white spaces denote non-significant power compared to white noise. (b) Cumulative power z-score to calculate the periods with best fit across hospitals, represented as the 95th percentile of cumulative power. Power for each hospital were normalized, log transform, subtracted from the mean, and divided by the standard deviation. The sum of z-score for each period was plotted with the 95th percentile (dotted lines in both a and b)^,.

Figure 4

Start and peak of ILI vs latitude, 29 health networks in Chile, 2010–2016. Linear model for start day and peak day vs latitude. (a) and (b). all health networks; c and d health networks classified by zone: north, center, and south. Red lines: p-value < 0.05, black: p-value > 0.05. Vertical grey zones denote 95% CI, horizontal grey: range of start days for de period for each health network. Black circles size represents population size and black circle position, median start (panels a and c) or peak day (panels b and d). (a) Linear model for start day versus latitude denotes a slight, but significant north to south gradient: β = 0.99, P = 0.003, R² = 0.04. (b) Linear model for peak day versus latitude denotes a slight, but significant north to south gradient: β = 0.8, P = 0.002, R² = 0.04. (c) Linear model for start day versus latitude by zone shows a significant north to south gradient in the center zone of the country. North zone linear regression (top): β = 0.5, P = 0.75, R² = 0.003. Center zone linear regression (middle): β = 3.49, P < 0.01, R² = 0.09. Southern zone linear regression (bottom): β = 1.49, P = 0.33, R² = 0.02. D) Linear model for peak day versus latitude by zone shows a significant north to south gradient in the center zone of the country. North zone linear regression (top): β = 0.8, P = 0.554, R² = 0.01. Center zone linear regression (middle): β = 3.65, P < 0.01, R2 = 0.11. Southern zone linear regression (bottom): β = − 1.8, P = 0.08, R² = 0.05^,.

Figure 5

Distance vs pairwise Spearman's correlation and travelling waves of influenza in health networks. Chile 2010–2016. (a) Spearman’s correlation of phase angles versus distance between all pairs of health networks. Dashed lines represent the average correlation. The model used a loess model with α = 0.75 (red line) and 95% CI (grey). There is a change in the descending relation of correlation and distance at 1250 km that we used as the upper distance limit to define local wave. (b) Phase difference in weeks vs distance (km) for Health Networks with incoming (red) and outgoing (blue) waves. In grey linear models with a significance level of 0.05. In red and blue, linear models with a Bonferroni corrected significance level < 0.001. The distance sign was inverted for negative lag times for a more intuitive display of incoming waves. (c) Geographic location of health networks with significant traveling waves. Incoming and outgoing waves are clustered in the center of the country. Colors correspond to panel (b)^,.