^{1}

^{2}

^{3}

^{4}

^{5}

^{5}

The authors have declared that no competing interests exist.

Visceral leishmaniasis (VL) is an important neglected disease caused by a protozoan parasite, and represents a serious public health problem in many parts of the world. It is zoonotic in Europe and Latin America, where infected dogs constitute the main domestic reservoir for the parasite and play a key role in VL transmission to humans. In Brazil this disease is caused by the protozoan

Visceral leishmaniasis [VL] is an important but neglected tropical disease that occurs worldwide. In areas where the disease is zoonotic, it is considered a serious public and animal health problem. In Brazil, despite the existing prevention and control programs, the disease is spreading, and in São Paulo State its dispersion presents a distinct temporal-geographic pattern. The goal of our study was to understand the landscape, climatic and economic factors that influence the dispersion of visceral leishmaniasis in Sao Paulo State, and then use these findings to predict its spread over time and space. To this end, we integrated data on the sand fly vector of the causative parasitic agent of VL, infected dogs and human cases. We find that landscape and climate were more important than economic factors in predicting competent vector, infected human and infected dog distributions, and the presence of the competent vector and infected dogs strongly influenced the dispersion of infected humans. Our study represents the first integrated investigation of vector and infected host invasion potential for individual municipalities, contributing to VL disease prevention and control planning in São Paulo State.

Visceral leishmaniasis (VL), also known as kala-azar, is characterized by irregular bouts of fever, weight loss, enlargement of the spleen and liver, and anemia [

Where the disease is zoonotic, infected dogs constitute the main domestic parasite reservoir and play a key role in VL transmission to humans [

The Brazilian State of São Paulo has experienced rapid spread of human and dog VL over the last two decades, mainly associated with the expansion of the vector’s geographic range. In 1998, the sand fly that transmits VL was found in only two municipalities, but since then it has been recorded in 164 municipalities by 2014, principally in the western part of the State [

The expansion of VL in São Paulo occurred along a major axis extending from the northwest to the southeast towards the Bauru region, following both the Bolivia-Brazil gas pipeline and the Marechal Rondon Highway [

Infectious disease outbreaks have both spatial and temporal dimensions [

São Paulo State is comprised of 645 municipalities (^{2}. Based on digitized municipal boundary maps produced by the Brazilian Institute of Geography and Statistics [

For each Municipality, we obtained data on the presence and absence of (1) VL sand fly vectors, (2) autochthonous VL human cases, and (3) VL seropositive dogs, from 1999 to 2013. Data on the distribution of the vector

To investigate factors driving the spread of sand fly vectors, VL infected dogs and infected human dispersion across São Paulo State, we examined the following covariates at the municipal level: invasion pressure, climatic features, economic factors, and the distance between each municipality’s centroid and (1) the Marechal Rondon Highway and (2) the Bolivia-Brazil gas pipeline.

To account for the spatial contagion of the vectors, infected dogs and humans, we created an invasion pressure index for each population and each municipality that takes into account the spatial configuration of already invaded municipalities. More specifically, the invasion pressure index for municipality _{ij} are distance-decay weights and _{j(t−1)} is the invasion status (0 = not invaded, 1 = invaded) of municipality _{ij} (the Euclidean distance between the geographic centroids of two municipalities

We included annual average temperature (

Data on São Paulo municipalities’ Annual Gross Domestic Product (

Location data for the Bolivia-Brazil gas pipeline and Marechal Rondon Highway were obtained from the Brazilian Transport Ministry [

Bayesian models for disease mapping have primarily utilized a generalized linear model framework, with regression parameters modeled as random effects with spatial (or space-time) covariance matrix [

Our observations consist of _{it}, which is a binary outcome indicating whether municipality _{it} If a municipality has already been invaded by a competent vector (or by an infected dog or human), we assume it will keep its invasion status throughout the study period. This assumption can be stated as:

If the municipality has not been previously invaded (i.e., _{i(t-1)} = 0), we assume that:

We base our investigations of municipality “invasion” on the following premises, related to the competent vector and infected hosts:

The competent vector dispersion does not depend on the presence of infected hosts. Therefore, we assume that the probability of vector invasion for municipality

Dogs can become infected by the vector at the focal municipality (Municipality _{it} as well as the other covariates:

Infection in humans depends on vector presence, and when dogs and vectors occupy the same space, the disease becomes established [_{it} and the presence of infected dogs _{it} as well as the other covariates:

Finally, we assume no false positives or false negatives. The models formulated above can be thought of as generalizations of the geometric distribution (number of trials before the first success/invasion), where the individual success/invasion probabilities are not constant and spatial and temporal dependence are taken into account by evaluating the status of other municipalities in the previous time step, as captured by the invasion pressure covariate for vectors, infected dogs and infected humans (

We fit this model in a Bayesian framework using JAGS, which enables us to coherently represent uncertainty when creating predictions of invasion probability from 2014 to 2020. Our Bayesian model is likely to be very useful for researchers interested in understanding disease spread, and for this reason, we include a detailed tutorial-style appendix that describes how to implement the model (

The model described above was also used to predict the “invasion” of VL-free municipalities from 2014 to 2020 by using the posterior distribution of the parameters and forward simulation of the invasion process. For each sample of the posterior distribution, we sequentially created a forward simulation for the vector, for the infected dogs, and finally for infected humans, using 2013 as the starting point. We then summarized these synthetic invasion scenarios by calculating the probability that each municipality has been invaded. Our annual predictions of these invasion probabilities were mapped (using QGIS version 2.8), and maps were compiled into a time-loop video (available online) that incorporates both actual vector and infected host presence/absence data from 1999–2013 as well as predicted probability of invasion from 2014–2020.

For model validation, we obtained 2015 data on municipalities invaded by the vector and VL infected humans. Model validation was accomplished by comparing the 2015 predicted invasion probabilities with actual 2015 observations. Although only five and 15 new municipalities were invaded by infected humans or vectors in 2015, respectively, these data were still useful in evaluating the model’s predictive power.

A summary of our invasion risk factor analysis for vectors, dogs and humans is given in

Param. | Covariate | VECTOR | DOGS | HUMANS | |||
---|---|---|---|---|---|---|---|

Means (CI 95%) | |||||||

b0 | -5.46 | (-5.83/-5.12) | -6.79 | (-7.41/-6.25) | -7.21 | (-8.03/-6.551) | |

b1 | Invasion pressure | 5.51 |
(4.11/6.89) | 0.11 | (-1.58/1.87) | 0.19 | (-1.51/1.870) |

b2 | Gas pipeline | -0.67 |
(-1.01/-0.34) | -0.17 | (-0.66/0.32) | -0.41 | (-0.99/0.147) |

b3 | Highway | -0.25 | (-0.65/0.14) | -1.02 |
(-1.56/-0.51) | -0.14 | (-0.81/0.49) |

b4 | GDP | -0.13 | (-0.34/0.06) | -0.10 | (-0.39/0.17) | -0.30 | (-0.66/0.047) |

b5 | Altitude | -0.24 | (-0.85/0.41) | 0.56 | (-0.34/1.45) | -0.22 | (-1.40/1.013) |

b6 | Temperature | 0.86 |
(0.25/1.46) | 1.54 |
(0.67/2.44) | 0.35 | (-0.67/1.354) |

b7 | Rainfall | -0.28 | (-0.80/0.12) | 0.40 | (-0.15/0.86) | -0.94 |
(-1.92/-0.084) |

b8 | Vector | --- | --- | 3.69 |
(3.1/4.32) | 1.45 |
(0.52/2.303) |

b9 | Dog | --- | --- | --- | --- | 1.18 | (-0.35/2.563) |

b10 | Vector |
--- | --- | --- | --- | 1.46 |
(0.03/2.973) |

Param.: parameter symbols. GDP: Gross Domestic Product. 95% credible intervals (CI) are given between parentheses.

*Statistically significant slope parameters (i.e., parameters for which the 95% credible interval range did not include zero). Negative parameter estimates indicate a negative association between the covariate and the probability of invasion (e.g., greater distance to the gas pipeline decreases the invasion probability for vectors). Positive parameter estimates indicate a positive association between the covariate and the probability of invasion (e.g., higher temperatures increase the invasion probability by infected dogs).

Both vector and infected dog dispersions were strongly influenced by temperature. On the other hand, none of the dispersion patterns were affected by altitude variations or low economic productivity (GDP). Interestingly, results regarding proximity to the gas pipeline and the highway were inconsistent, suggesting potentially different dispersion mechanisms for vectors versus dogs.

Maps of vector, infected dog, and infected human presence/absence for 2013 and prediction maps for 2016, 2018, and 2020 displaying “invasion” probabilities are shown in

The probability of “invasion” was calculated using the posterior distribution of the parameters (described in the section “Covariates”) from the Bayesian model and forward simulations, as described in the section “Predictions”.

The majority of municipalities currently affected or predicted to be affected by vectors or hosts lie in the western part of São Paulo State. The top 20 municipalities with the highest invasion probabilities for vectors, infected dogs and infected humans are displayed in

Numbers assigned to municipalities are described in the

As expected, our models tended to assign higher invasion probabilities to “invaded” municipalities than to “non-invaded” municipalities, indicated by the taller boxes for the “invaded” municipalities in

Boxes represent the 25% and 75% percentiles, with the central horizontal line representing the median, while whiskers extend to the minimum and maximum values.

One of the most important risk factors for VL around the world is the migration of people from endemic to non-endemic regions [

We found a significant influence of the gas pipeline on VL distribution, specifically on vector dispersion, and an effect of the Highway on the dispersion of infected dogs (see

We found no significant effect of altitude on vector or infected host dispersion in São Paulo State. Vector and infected hosts occurred in municipalities with elevation between 274m and 804m, but preferentially between 274m and 539m, representing 85.3% and 40% of the municipalities altitudes at the State, respectively (

Factors such as temperature, relative humidity, and rainfall can influence sand fly population density [

A map of average rainfall (

In Brazil, other environmental factors have been positively correlated with anti-

We originally hypothesized that the invasion pressure index would play a key role in dispersion of vectors, infected dogs and infected humans, but our results only showed an impact on vector dispersion, suggesting that spatial contagion is a key process only for vectors. The spatial contagion (as captured by our invasion pressure covariate) was not significant, given the presence of the vector, likely due to the relatively high mobility of dogs and humans, as compared to the vector. These results highlight the importance of simultaneously accounting for vectors and multiple hosts when assessing the spatial and temporal distribution of human VL cases. Presence of the vector itself is a poor predictor of disease transmission, because infected reservoirs or/and vectors must be present to establish parasite circulation in transmission foci [

In Brazil some studies found that the occurrence of VL in humans is associated with the presence of infected dogs in the same area [

Considering this zoonotic scenario, and in an effort to reduce or even eliminate human cases of VL, the Brazilian Leishmaniasis Control Program treats human cases and invests in the reduction of vector densities (spraying insecticide) and canine control (culling seropositive dogs) [

Other models used to predict zoonotic VL [

We used a Bayesian model to better define the risk factors for the spread of VL in São Paulo State, integrating data on its vector and hosts, and to predict the distribution and dispersion of the disease over time and space. As we already cited above, zoonotic VL has been spreading in other regions of Brazil outside of São Paulo [

Landscape features, such as the gas pipeline and Marechal Rondon Highway and annual temperature, rather than economic factors, were the most important risk factors in predicting VL dispersion in São Paulo State, Brazil. Since the dispersion of infected humans is affected by the spatial distribution of vectors and infected dogs, strategies to block the spread of the disease in humans and dogs need to address all three components of the VL dynamic cycle. Prevention and control measures should not only focus on vector control (e.g., use of residual insecticide), but also employ measures to block vector-human contact (e.g., insecticide spraying) and vector-dog contact (e.g., insecticide impregnated collars and vaccines). We suggest that these measures be prioritized for areas that have no current record of vectors or hosts infected with

Values represent Brazilian currency, Reais (R$), classified according to quantile intervals, rounded to the nearest thousand.

(TIF)

(TIF)

(TIF)

(TIF)

(JPG)

(MP4)

(PDF)

Municipality superscript numbers are represented on the map in

(PDF)