Delineating the Seasonality of Varicella and Its Association With Climate in the Tropical Country of Colombia

Abstract Background Varicella causes a major health burden in many low- to middle-income countries located in tropical regions. Because of the lack of surveillance data, however, the epidemiology of varicella in these regions remains uncharacterized. In this study, based on an extensive dataset of weekly varicella incidence in children ≤10 during 2011–2014 in 25 municipalities, we aimed to delineate the seasonality of varicella across the diverse tropical climates of Colombia. Methods We used generalized additive models to estimate varicella seasonality, and we used clustering and matrix correlation methods to assess its correlation with climate. Furthermore, we developed a mathematical model to examine whether including the effect of climate on varicella transmission could reproduce the observed spatiotemporal patterns. Results Varicella seasonality was markedly bimodal, with latitudinal changes in the peaks' timing and amplitude. This spatial gradient strongly correlated with specific humidity (Mantel statistic = 0.412, P = .001) but not temperature (Mantel statistic = 0.077, P = .225). The mathematical model reproduced the observed patterns not only in Colombia but also México, and it predicted a latitudinal gradient in Central America. Conclusions These results demonstrate large variability in varicella seasonality across Colombia and suggest that spatiotemporal humidity fluctuations can explain the calendar of varicella epidemics in Colombia, México, and potentially in Central America.


S1.1. Varicella data
In Colombia, healthcare institutions report clinically confirmed varicella cases to the national surveillance system (Sistema Nacional de Vigilancia en Salud Pública, SIVIGILA), which in turn publishes online the freely available unidentifiable individual data [1,2].The SIVIGILA works under the Colombian National Health Institute (NHI) in cooperation with the Pan American Health Organization (PAHO) and collects weekly data from public and private healthcare provider institutions.According to "Law 1712 of 2014 for Transparency and the Right to Access Public Information", the SIVIGILA database is available publicly online.The database provides 1) aggregated data, consisting of municipality-level, weekly time series of varicella cases, and 2) individual data without personal identification information from 2007 to date [3,4].We aggregated the individual data by week and municipality, and all the analyses were performed using the aggregated data, which is available in the Edmond repository.
A varicella case is clinically defined by the SIVIGILA as an acute onset illness that initiates with moderate fever and small erythematous macules that evolve into papules, water-clear vesicles, yellowish pustules, and finally, crusts.The cases are evaluated by a healthcare professional and can be epidemiologically linked to another case.A varicella case is then stored under the 831 code in the SIVIGILA database.

S1.2. Demographic data
We obtained municipality-level data on the population using demographic estimates from the Colombian National Administrative Department of Statistics (DANE) [5].With these estimates and the area of each municipality, we estimated the population density and population density of children ≤10 years old from 2011-2014 per municipality.To assess migration and urbanization in each municipality, we examined the fraction of the total varicella cases reported in children who are migrants and the fraction of varicella cases reported in children residing in urban areas.

S1.3. Spatial resolution and coordinates
The spatial resolution was defined at the level of the municipalities for Colombia, considering that the varicella data is reported at the same level.For México and Central America, the analyses were performed for the capital city of the first-level administrative divisions of each country.For Colombia, coordinates of the centroid for each municipality were obtained from the Colombian National Institute for Hydrology, Meteorology, and Environmental Studies (IDEAM) database [6].For México and Central America, coordinates of the capital cities were obtained from Google Maps.

S1.4. School terms and school holidays data
We obtained the school information from 2014 for Colombia from the Education Ministry website [7].Data for Mexico and other Central American countries (Panamá, Costa Rica, Nicaragua, Honduras, El Salvador, Guatemala, and Belize), were obtained likewise [8-14].

S2.1. Transmission model formulation
We formulated a Susceptible-Exposed-Infected-Recovered (SEIR) model of varicella transmission, including school terms and climate as sources of seasonality for the transmission rate.The term-time forcing driven by the alternation between school terms and school holidays was modeled using a square wave: Where Term(t) equals 1 during school terms and -1 during school holidays.The parameter A 1 represents the amplitude of term-time forcing, fixed to 0.25 based on the empirical observation that schoolchildren make 40% fewer contacts during holidays than during school terms ((1 − A 1 )/(1 + A 1 ) = 0.6) [15].The denominator represents a correction factor, used to ensure that the square wave has a seasonal mean of 1.The climate forcing was modeled as: where Climate(t) is the weekly standardized specific humidity, and A 2 represents the amplitude of the climate forcing, fixed to 0, -0.04 or -0.08.
The model is then represented using the following system of ordinary differential equations: The seasonal force of infection is given by: Briefly, individuals started at the susceptible to varicella compartment (S).Next, susceptible individuals entered the exposed compartment E with a seasonal force of infection λ (per susceptible rate of infection, reproduction number, 10) [16].Then, exposed individuals become infectious at a rate σ (1/σ = average latent period, 14 days) [17], enter the infectious compartment I, and then move to the R recovered compartment at a rate γ (1/γ = average recovery period, 7 days) [17].
For all simulations, the seasonal model was initialized at the values of the endemic equilibrium of the corresponding seasonally unforced model.The model was then simulated for 200 years, and the last year of the simulation was used for assessing the seasonality of weekly cases.

Supplementary figure 1 .
Selection of municipalities according to varicella reports and signalto-noise ratio.For definiteness, we selected the municipalities with a signal-to-noise ratio (mean to standard deviation ratio of the weekly reports) over one and an average of at least five cases/week.(A) 25 municipalities met the criterion and collectively included (B) 67.4% of the total reports of varicella, (C) 41.2% of the total study population (children ≤10 years of age).(D) Map of included municipalities, the 25 municipalities included 6 of the 17 Köppen-Geiger climatic classifications most commonly found in the tropics.

Supplementary figure 9 .
Predicted seasonality of varicella across Colombia and Mexico.The lines represent the weekly incidence of varicella (rescaled to have a yearly average of 1) predicted from a transmission model incorporating term-time forcing and different levels of specific humidity forcing.

. Data flow chart on varicella reports.
We accessed varicella cases for the study period (2011-2014) from the Sistema Nacional de Vigilancia en Salud Pública (SIVIGILA).(A) Data flow chart.(B) For 60,257 cases that reported age but no birthdate information, we assumed that the age was not misclassified as their distribution was not different from the cases with birthdate data.(C) The age distribution of varicella infections across municipalities in Colombia in children ≤10 years of age.The vertical white line represents the mean age at infection.