Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

The effect of climate change on cholera disease: The road ahead using artificial neural network

  • Zahra Asadgol,

    Roles Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Department of Environmental Health Engineering, School of Public Health, Iran University of Medical Sciences, Tehran, Iran

  • Hamed Mohammadi ,

    Roles Data curation, Formal analysis, Methodology, Validation, Visualization

    gholamim@iums.ac.ir (MG); ham19@zums.ac.ir (HM)

    ‡ MG and HM are joint corresponding authors on this work.

    Affiliation Department of Environmental Health Engineering, School of Public Health, Zanjan University of Medical Sciences, Zanjan, Iran

  • Majid Kermani,

    Roles Data curation, Formal analysis, Visualization, Writing – review & editing

    Affiliations Department of Environmental Health Engineering, School of Public Health, Iran University of Medical Sciences, Tehran, Iran, Research Center for Environmental Health Technology, Iran University of Medical Sciences, Tehran, Iran

  • Alireza Badirzadeh,

    Roles Conceptualization, Formal analysis, Investigation, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Department of Parasitology and Mycology, School of Medicine, Iran University of Medical Sciences, Tehran, Iran

  • Mitra Gholami

    Roles Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Writing – review & editing

    gholamim@iums.ac.ir (MG); ham19@zums.ac.ir (HM)

    ‡ MG and HM are joint corresponding authors on this work.

    Affiliations Department of Environmental Health Engineering, School of Public Health, Iran University of Medical Sciences, Tehran, Iran, Research Center for Environmental Health Technology, Iran University of Medical Sciences, Tehran, Iran

The effect of climate change on cholera disease: The road ahead using artificial neural network

  • Zahra Asadgol, 
  • Hamed Mohammadi, 
  • Majid Kermani, 
  • Alireza Badirzadeh, 
  • Mitra Gholami
PLOS
x

Abstract

Climate change has been described to raise outbreaks of water-born infectious diseases and increases public health concerns. This study aimed at finding out these impacts on cholera infections by using Artificial Neural Networks (ANNs) from 2021 to 2050. Daily data for cholera infection cases in Qom city, which is located in the center of Iran, were analyzed from 1998 to 2016. To determine the best lag time and combination of inputs, Gamma Test (GT) was applied. General circulation model outputs were utilized to project future climate pattern under two scenarios of Representative Concentration Pathway (RCP2.6 and RCP8.5). Statistical downscaling was done to produce high-resolution synthetic time series weather dataset. ANNs were applied for simulating the impact of climate change on cholera. The observed climate variables including maximum and minimum temperatures and precipitation were tagged as predictors in ANNs. Cholera cases were considered as the target outcome variable. Projected future (2020–2050) climate in previous step was carried out to assess future cholera incidence. A seasonal trend in cholera infection was seen. Our results elucidated that the best lag time was 21 days. According to the results of downscaling tool, future climate in the study area by 2050 will be warmer and wetter. Simulation of cholera cases indicated that there is a clear trend of increasing cholera cases under the worst scenario (RCP8.5) by the year 2050 and the highest cholera cases observe in warmer months. The precipitation was recognized as the most effective input variable by sensitivity analysis. We observed a significant correlation between low precipitation and cholera infection. There is a strong evidence to show that cholera disease is correlated with environment variables, as low precipitation and high temperatures in warmer months could provide the swifter bacterial replication. These conditions in Iran, especially in the central parts, may raise the cholera infection rates. Furthermore, ANNs is an executive tool to simulate the impact of climate change on cholera to estimate the future trend of cholera incidence for adopting protective measures in endemic areas.

Introduction

Studies have elucidated the effects of climate change on infectious diseases, including water, vector and food-borne diseases [13]. There is the evidence of relationship between climatic parameters and the burden of water-borne diseases such as cholera, particularly in undeveloped/developing countries [46].

The cause of cholera is Vibrio cholerae (V. cholerae), which is categorized as a gram-negative bacterium. Its natural habitat is brackish water. The disease is a severely contagious acute bacterial infection, which is caused by colonization and multiplication of V. cholerae inside the intestine. High risk individuals get the infection when they ingest an infective dosage of bacteria from contaminated water, vegetables and foods [7]. According to the World Health Organization (WHO) reports, cholera is one of the health threats in some developing countries. In 2017, a total of 1227391 cases of cholera were reported by 34 countries including 5654 deaths [8]. Cholera infection has emerged in association with seasonality, travel, natural catastrophes, warfare and distinct conditions which ends to insufficient sanitation and poverty [9]. With growing concerns about global climate change, the role of climatic factors in cholera incidence has been investigated in the last decades. Weather conditions such as increasing ambient temperature were known as a key parameter for cholera incidence [10, 11]. Several studies have shown a significant correlation between either high or low rainfall and incidence of cholera cases. So, either increasing or decreasing the average rainfall could lead to flood and droughts which can affect the concentration of bacteria and also human health [1214]. Rainfall can also effect on nutrient concentrations, salinity and pH of water resources which effect on bacterial survival [15].

General Circulation Models (GCMs) have been applied to evaluate the effect of climate change on a wide region. The Intergovernmental Panel on Climate Change (IPCC) has published the latest sets of scenarios (Representative Concentration Pathway (RCP)) in the its fifth Assessment Report (AR5) in 2014. These scenarios include RCP8.5, RCP6, RCP4.5, and RCP2.6 [16]. To link between outputs of GCMs (spatial resolution of 100–300 km) and the local climatic processes (spatial resolution of 10–20 km), downscaling technique was investigated. This technique was applied to project future climate pattern at the local scale under RCP scenarios [17].

Recently, the prediction of effect of climate change on health aspects has been widely studied [18, 19]. Prediction of an infectious disease such as cholera by reliable modeling can help managers to perform preventive actions such as preparation of adequate manpower, pharmaceutical and logistic resources [20]. Artificial neural networks (ANNs), as nonlinear statistical modeling, can be applied for creating models of disease spreading and prediction of epidemic outcomes. Moreover, ANNs can gain elucidation of data and nonlinear forecast analysis of variables for evaluation of biological and environmental data [21]. Several studies applied ANNs to forecast the incidence of cholera over an area and illustrated advantages and disadvantages of this method [21, 22].

Many undeveloped/developing countries such as Iran still suffer from frequent epidemics; therefore, the objectives of present study were evaluation of the association between the cholera cases and climate variables and also finding the most effective parameters. In addition, we developed ANNs to simulate the climate change impact on cholera cases by the year 2050 in Qom, Iran, with the highest cholera incidences, within two scenarios of RCP2.6 and RCP8.5. In considering importance of controlling cholera in Qom, it can help to develop a system for warning incidence of cholera which lead to promotion of planning and decision-making for public health and reducing the health effects of climate change.

Material and methods

Study area

The study area was Qom (50.88°N 34.64°E), located in the center of Iran and southwest of Tehran (capital of Iran) (Fig 1). Qom is one of the biggest, industrialized and densely populated cities in Iran, with an urban population of over 1 million. The region’s physical geography is dry and warm with an average annual maximum temperature of 40.3°C in July and the average annual minimum temperature of -1.6°C in January. Annual rainfall average is 125 mm which occurs between October and May [23, 24].

thumbnail
Fig 1. Study area: Qom, Iran (created by Arc GIS version 10.2).

https://doi.org/10.1371/journal.pone.0224813.g001

Data source and collection

Cholera.

Daily cholera prevalence data from January 1998 to December 2016 were collected from the records of "The Centers for Disease Control and Prevention" at Qom University of Medical Sciences. Systematic cholera surveillance has been started working from 1998 epidemic in Qom, Iran; therefore, there was not available any cholera data for previous year. The suspected cholera cases were diagnosed by the conventional culture technique and confirmed as cholera case-counts.

Meteorological data.

Meteorological data were collected from Iran Meteorological Organization (IMO) for the period of 1976 to 2016. Data were collected regarding daily minimum temperatures (Tmin), daily maximum temperatures (Tmax) (in °C) and daily precipitation (pr) (in mm) from Qom meteorological stations.

Preprocessing and optimization processes

The Gamma Test (GT) was known as a non-linear modeling analysis tool which estimated the minimum Mean Square Error (MSE). GT can be attained when modeling the unseen data using any continuous nonlinear models. This tool was firstly introduced by Koncar [25] and Agalbjörn [26] and then improved by other researchers [27, 28]. GT is an index of strength association between two variables measured at the ordinal level. This index estimated that part of variance of output data which cannot be accounted for any smooth data model and calculate by vertical intercept of the regression line [29]. Furthermore, slope or gradient is the slope of regression line and could provide information on complexity of the system. GT is known as a non-parametric method and its results used to build a model regardless of the techniques. The result can standardize by V-ratio, which determines as Gamma/Var (output) and scale between 0 and 1. A V-ratio equal or close to zero demonstrates a high degree of predictability of given output. Furthermore, if the standard error (SE) value tends to zero, there is a high confidence in the value of gamma statistic as an estimate for the noise variance on the output [30]. To estimate the reliability of gamma statistic, a series of GT (for a definite number of unique data points (M)) was run to determine the size of the data set required to create a stable asymptote. The M-test helps us to find out how many data points are adequate for building a model with a proximate MSE to the estimated noise variance (when the M-test plot be-comes flat) [31].

In the present study, different combinations of input data were explored to assess their influence on the cholera disease. GT evaluated the best combination of the candidate inputs by Genetic Algorothim (GA). The parameters of GA to find good embedding were the population size (100 chromosomes), the mutation rate (0.01), the crossover rate (0.5), the gradient fitness (0.1), intercept fitness (0.8), and length fitness (0.1) [16]. The non-linear analysis and modeling tool which used to analyze the data was winGamma that developed by the Department of Computer Science, Cardiff University [29].

In the first step, five time-series data including cholera's data with no lag time (NL), cholera's data with seven days lag (7DL), cholera's data with 14 days lag (14DL), cholera's data with 21 days lag (21DL) and cholera's data with 30 days lag (30DL) were prepared to be examined. In the next step, standard normalization was applied to give an equal chance to contribute to an output prediction for each data group. Then, different smooth data modeling techniques were built and examined. Three principal factors of GT including gamma statistics, model complexity, and the required number of inputs were used to determine the fitness of particular mask or feature set. Finally, according to the results of GT, GA and M-test, the best lag time was chosen between five time-series.

Projecting future climate patterns and climate scenario

Simulated climate change data on monthly scales in baseline experiments (1976 to 2005) and future experiments (2021 to 2050) were obtained from Intergovernmental Panel on Climate Change (IPCC)-data distribution center (DDC) website. The future experiments were included four RCP scenarios (RCP8.5, RCP6, RCP4.5, and RCP2.6), which in the present study, we used two of the RCP scenarios, RCP8.5 and RCP2.6 (with a relatively high and low future pathway). In this study, ten global climate change model outputs (France IPSL-CM5A-LR, Japan MIROC5, Canada CanESM2, USA CESM1-CAM5, Australia CSIRO-Mk3-6-0, China FIO-ESM, South Korea HadGEM2-AO, UK HadGEM2-ES, Germany MPI-ESM-MR, China bcc-csm1-1) were applied to project the future climate. ArcGIS10.2 was utilized to display and transfer Net-CDF (network common data form) files to excel. Monthly average of all variables (baseline and future) were calculated and graphed separately. Then, the above-mentioned data of the two selected groups in 10 models were compared. Furthermore, change factor method was used to select the most appropriate GCMs.

Statistical downscaling establishes the relationships between the selected GCM outputs and observed regional climatic processes [32]. LARS-WG (Long Ashton Research Station Weather Generator, version 5–5) was applied as a successful downscaling tool, to both generate or simulated future daily meteorological variables in our study area [33]. The process of creating daily site-specific climate data were carried out in three steps as follows:

1. Site analysis (model calibration): In this step, observed daily weather data from 1976 to 2005 (30 years) recorded in Qom synoptic stations was first used to calculate their statistical parameters. These parameters were applied by LARS-WG in next step to generate synthetic data series. The Kolmogorov-Smirnov (K-S) were used to determine the best copula fitting to data, t-test for monthly means, and F-test for standard deviation [34].

2. Model validation: Synthetic weather series (generated by the WG) were methodologically tested against observed data to find out the presence of any statistically significant difference in the statistical parameters. This method led to increasing the confidence in the models’ predictions [33].

3. Generation of future daily synthetic weather data: LARS-WG baseline parameters which are derived from observed weather data (during the process of model calibration) are applied to generate synthetic weather by using the output of GCMs (as predictors). It derived changes in precipitation and temperature and statistically downscale the conditioning of parameters. The LARS-WG version of 5–5 has not covered the latest sets of scenarios which released in the IPCC-AR5. Thus, to produce the RCP scenario fills and generate future weather data, the observed climate data in the study area was compared with the GCM outputs and then applied to the LARS scenario fill.

Simulation

The ANNs were applied to simulate the climate change impacts on cholera disease. ANNs is a powerful mechanism and an important computational data-driven model which can learn a correlation between inputs and outputs (training) and then represents the relationship between input and output parameters by building a model. One of the most commonly and effective used types of neural network is Back Propagation Neural Network (BPNN) approach, i.e. Multilayer Perceptron (MLP). Using MLP, a learning model created by baseline data and then it uses to produce outputs for new inputs [35].

In the present study, the software package Neuro-Solutions for Excel, version 7 (Neuro-Dimension, Inc. 3701 NW 40th Terrace, Suite 1 Gainesville, FL 32606, NeuroSolutions for Excel) was applied. The observed climate datasets in the study area were tagged as inputs and the number of cholera disease was tagged as desired in an ANNs model for training, cross-validation, and testing. MLPs algorithm was used to create a neural network model and predict cholera disease number, due to climate change. MLPs are structured of three layers of input, hidden, and output and two steps of input feed-forward network and back propagation. After randomization of the datasets, for training, cross-validation, and testing the network, 60, 15 and 25% of the inputs and outputs datasets were applied, respectively. In this study, to evaluate the prediction success, we assessed the accuracy of the estimate by using MSE, correlation coefficients (R-values), root mean square error (RMSE), normalized mean square error (NMESE), and mean absolute error (MAE). Finally, the future (2021–2050) trends of cholera disease were estimated under two RCP scenarios (RCP2.6 and RCP8.5) using the projected climate datasets in the optimized ANNs.

Statistical analysis

For statistical analysis of collected data, SPSS software (version 20) was applied and the normality of all data was evaluated by Kolmogorov-Smirnov test. Descriptive statistics were used for analysis of meteorological data and the number of cholera cases. Furthermore, the relationship between the number of cholera cases and climate variables was determined by Pearson correlation coefficient. For all tests, P-values of ≤0.05 are embraced as statistically significant. GraphPad Prism 8.0.1 was used to design and graph in the current study. Also, ArcMap 10.2 Geographical Information System (GIS) (ESRI, Redlands, CA) was applied as a suitable tool for mapping the study area and transferring Net-CDF (network common data form) files to excel.

Results

Overview of effect of climate variability on cholera

From March 1998 to December 2016, in total, 1243 patients were identified with cholera by the surveillance system in Qom province. The highest number of annual cholera cases were recognized in five years 1998, 2001, 2005, 2008 and 2011 with 836, 22, 167, 26 and 137, respectively, and the other years have few or zero number of cholera cases. The minimum, maximum and mean of monthly variables for each year are presented in Table 1. The maximum monthly of cholera was 21 in August 1998, as a year with the highest number of cholera cases. No case was recognized in some years observed in Table 1. The monthly average of minimum temperature varied between 10.5°C in 2000 and 11.54°C in 2015 which shows an increase for minimum temperature. The lowest temperature was recorded in January 2008 with a monthly average of -23°C after a catastrophic weather in Qom. An increasing trend was observed for monthly average of maximum temperature. So, the maximum temperature was elevated from 43.5°C in August 1998 to 45.2°C in June 2016 and also during 19 years, the highest temperature was recorded in July 2010 with 47°C. The maximum precipitation was varied between 9.5 mm in December 2014 and 35 mm in November 1999. Furthermore, a decreasing trend was observed for precipitation during the years. Time series of cholera and climate variables were applied to demonstrate the relationship between climatic factors and number of cholera cases (Fig 2). For better understanding the figure, we presented the monthly values of cholera cases and climate variables in just 5 years with the highest number of choleras and ignored the years with negligible or zero cases of cholera. A seasonal trend in cholera cases and climate variables is shown during 1998 to 2016 (Fig 3). Highest numbers of cholera were observed during the warm season from June to September with highest maximum and minimum temperature and lowest precipitation. Analysis of the correlation between monthly average of climate variables and number of cholera cases were carried out by Pearson test (Table 2). The maximum and minimum temperatures as independent variables had a positive correlation and also precipitation has a negative correlation with number of cholera cases (R2 = 0.211, R2 = 0.204 and R2 = -0.226, respectively). Moreover, the seasonal pattern of cholera disease was confirmed by correlation coefficient of 0.332 between months and the number of cholera cases.

thumbnail
Fig 2. Variation in the number of cholera cases and climate variables in five years (1998, 2001, 2005, 2008 and 2011).

https://doi.org/10.1371/journal.pone.0224813.g002

thumbnail
Fig 3. Monthly variation in the number of cholera cases and climate variables, 1998–2016.

https://doi.org/10.1371/journal.pone.0224813.g003

thumbnail
Table 1. Descriptive statistics of the monthly climatic variables and cholera cases.

https://doi.org/10.1371/journal.pone.0224813.t001

thumbnail
Table 2. Correlation between the number of cholera cases and climate variables during 1998–2016.

https://doi.org/10.1371/journal.pone.0224813.t002

The results of GT

In the first step of analysis, GT results including gamma statistics and other relevant measures were calculated for 3 inputs and one output in five time series data (NL, 7DL, 14DL, 21DL, 30DL). Furthermore, different combinations were analyzed to assess their influence on the cholera disease modeling (Table 3). As shown in the Table, the minimum value of Gama (0.123) was observed when we used 21DL dataset with just precipitation (pr) as input combination. The best input data combination has a set of low values of gradient (0.22), standard error (0.03) and V-ratio (0.37) and selected as a best lag time for simulation in the next step. Interesting findings of the present study shows that the three input variables combination has a higher gamma statistic, compared with the other combinations.

Regarding the size of dataset, the near neighbors (the number of Pmax) were estimated in 10 which shows that produces the most accurate estimate. The variation of gamma statistic and the SE versus the Pmax has been illustrated in Fig 4(A). Moreover, the sufficient data for building a smooth data model was estimated by M-test. From Fig 4(B), it can be seen that the sufficient data point to construct a smooth data model without overtraining were at least 4000–4500.

thumbnail
Fig 4.

(a) Variation of gamma and SE as the number of near neighbors (b) M-test graph.

https://doi.org/10.1371/journal.pone.0224813.g004

Predictions of climate variability under future climate change scenarios

Statistical downscaling using LARS-WG was applied to define an empirical relationship between observed regional climatic (1976–2005) and GCM output models for 2021–2050 under two scenarios of RCP2.6 and RCP8.5 for the study area. Fig 5 demonstrates the downscaled monthly mean values of three parameters of precipitation, Tmax and Tmin under RCP2.6 and RCP8.5 scenarios, compared to the baseline values. As can be seen in this figure, RCP2.6 and RCP8.5 predicted that the minimum and maximum temperatures will rise in the study area by the year 2050. The highest increase in Tmin and Tmax will happen in July (1.44°C) and May (1.81°C) for RCP8.5, respectively. In addition, the lowest amount of precipitation will occur in Mar (-7.89) but in overall, RCP2.6 and RCP8.5 scenarios predicted that precipitation will decrease in the first 6 months of the year, while it will increase in the three months of winter during October, November and December.

thumbnail
Fig 5.

Future monthly mean of (a) minimum temperature (Tmin), (b) maximum temperature (Tmax) and (c) precipitation.

https://doi.org/10.1371/journal.pone.0224813.g005

Cholera simulation based on ANNs

The simulation of effect of climate change on cholera disease was carried out by using a feed-forward MLP type of ANNs. The schematic of the optimized NN is shown in Fig 6. This figure demonstrates that the network has one hidden layer and Tan Axon and Momentum were set as layer’s transfer function and learning rule, respectively. It is worth to mention that the results are based on data with 21-days lag (21DL). According to the results, the average final MSEs were 0.011 and 0.008 with standard deviations of 0.004 and 0.003, respectively for training and cross-validation.

thumbnail
Fig 6. Optimized ANNs structure for simulation in the study.

https://doi.org/10.1371/journal.pone.0224813.g006

In evaluation of future trend of cholera disease, Fig 7(A) was prepared to show the trends of cholera cases for years of 2021–2050 under the two scenarios of RCP2.6 and RCP8.5. Also, Fig 7(B) shows monthly average of cholera cases during 2021–205. Accordingly, there is an increasing trend in cholera cases under RCP8.5 by the year 2050. The seasonal trend of cholera will change in the future qua the highest cholera cases will observe in spring and summer over next 30 years, while the highest monthly average of cholera cases in baseline period was in August. These conditions will occur when social factors such as availability of health and sanitation facilities and also safe water do not either change or become worse through the time. Modeling by ANNs and sensitivity analysis of the mean of three inputs show that precipitation had the highest effect on the incidence of cholera with sensitivity of 0.54, while Tmin and Tmax had sensitivity of 0.07 and 0.018, respectively (Fig 8).

thumbnail
Fig 7.

(a) Trend of cholera cases by 2050, (b) Monthly average of cholera cases.

https://doi.org/10.1371/journal.pone.0224813.g007

Discussion

In the present study, the effects of climate variables on cholera disease in Qom, were evaluated. Since Qom is one of the main religious tourist cities, control of cholera is a strategic and important health issue. From July 1998 and after a huge outbreak in cholera in Qom with 836 cases, the surveillance system began to identify and document all positive cases officially. The health center of Qom province assumed responsibility for finding out the source of infection, monitoring, surveillance and controlling the disease. In the 19-year time period, five waves with a 3–4 year life cycle were happened, as has been recognized in India and Bangladesh with 3-year cycles and 3–6 year cycles, respectively [36, 37]. Cholera infection can be transmitted by direct person-to-person contacts, using contaminated containers of water, unhealthy process of food preparation, or contact with contaminated environmental reservoir with persistence bacteria. However, pathways of transmission can be affected by climate conditions [38]. For example, high temperature can accelerate the growth and proliferation of pathogens in their habitats (e.g., raw food, water or environment). Rainfall can have a direct influence on transmission of cholera in two ways: first, high rainfall can increase the risk of contamination of either raw or treated water with wastewater (transmission of person-to-environment) and second, low rainfall can also increase the concentration of pathogens in water media (transmission of environment-to-person) [37]. According to the results of correlation between monthly average of climate variables and the number of cholera cases, it is obvious that high temperature with low precipitation in dry season was the ideal climatic conditions for cholera infection. In the current study, a seasonal trend was apperceived, so that the highest number of cholera cases were observed during the summer to early fall (June to October). We hypothesized that an increase in air temperatures during summer created proper environmental conditions for bacterial growth and their survival through increasing salinity. On the other hand, considerable decreasing in precipitation and increasing evaporation during dry months could lead to reducing the amount of water in surface and groundwater sources. In overall, these changes in climatic conditions could lead to increasing salinity, organic matter and aqueous temperatures and subsequent growth and proliferation of pathogens, decreasing raw water quality and increasing bacterial concentration in open water wells and the other water resources. In this regard, using contaminated and untreated water for residential consuming or irrigating vegetation could lead to increasing the risk of cholera disease incidence in high risk people. In different recent studies, environmental variables were known as most important factors in incidence of cholera. Indeed, seasonality was introduced as a stronger parameter which leads to increasing in cholera outbreaks generally during warm months [39]. The positive correlation between monthly average of temperature and cholera infection was observed in this study, which well proved by findings from East Africa, Zambia and South China [10, 11, 40]. In the case of precipitation, other studies found that heavy rainfall as an important parameter on cholera incidence can lead to flooding and affect water quality and sanitation systems [41, 42]. Although, the effect of low precipitation in increasing the risk of cholera disease described in this study, was supported by recent findings in Bangladesh and Iran [12, 21].

In the first step of analysis, to determine the foremost lag time, the best combination of inputs and to evaluate the effects of climate variables on cholera disease modeling, GT was applied. A modeler needs much time to calibrate and test different built models for all input combinations and also, there is no idea, how many data points required for calibration. The results of GT and GA can considerably reduce the model workload and help to focus on the best selection of inputs. The GA search and optimization techniques were used to evaluate the best combination of input variables for predicting the target which was chosen by smallest the asymptotic gamma statistic (the best MSE) between 2"-1 meaningful combination for a particular output [29]. The gradient and V-ratio also indicate the model complexity and the degree of predictability of given outputs [43]. Thus, a model with low gamma and smaller gradient and V-ratio are assumed as the best scenario for modeling. We indicated the impact of climate variables on cholera disease were lagged by 21 days (21DL). Our reason for choosing different lag times for variables was evaluation of the sensitivity of the cholera disease to delay in the effect of climate events. These lag times between the changes in climatic variables and the disease emergence may acquire, due to (a) time required to grow pathogens, (b) exposure to infections, (c) incubation period, (d) disease detection and (e) report delayed. Bacterial diseases have a shorter delay time than other pathogens and react more quickly to changes in temperature and precipitation [38, 44]. For example, maximum correlation coefficients between malaria incidence and sea surface temperatures in Colombia was happened with a 6 to 8 month lag [45], while the findings of Constantin de Magny in Bangladesh were shown the short time lag (one month) between environmental variables and cholera epidemics [18].

According the results of our study, just precipitation (pr) was known as the most effective input combination which means a built model by a combination of all input variables was not suitable to estimate cholera disease. On other words, the most important weather factor was only precipitation. Wang et al., used six weather input variables to estimate the evapotranspiration and their model using the combination with four input variables (maximum temperature, relative humidity, wind speed, solar radiation). They have shown that these variables have a better performance to estimate evapotranspiration and this model with more input variables could be developed more easily with less model complexity [43]. A study has reported that three contributing weather factors including daily wind speed (W), relative humidity (RH) and daily saturation vapor pressure deficit (Ed) were the best input data combination between four weather factors based on the gamma value (0.0216) [30]. If the gamma value and other three factors (i.e., gradient, Standard error and V-ratio) have low values, it is possible to build a mathematical model with high quality.

For generation of local-scale daily climate scenarios, baseline parameters of the LARS-WG (calculated from observed weather data for the period 1976–2005) modified for 2021–2050 under two RCP scenario files predicted by selected GCM model. In other words, by using this technique, the future values of local precipitation and temperature were generated for the Qom. To calibrate WG, synthetic daily weather data were generated by the parameters file in the Qom sites and then the observed and synthetic data were compared. In the current study, 10 GCMs experimental outputs were used and compared with the 30-year recorded climatic values and also, LARS-WG model applied for statistical downscaling to analyze and project future climate pattern under two scenarios of RCP2.6 and RCP8.5. Several studies applied GCMs to estimate the changes in disease in their area [46, 47]. Furthermore, some studies were used LARS-WG to generate daily site-specific climate scenarios for the future [16, 48]. The findings of this study are in agreement with Mohammadi's findings that simulated the impact of climate change on emergency medical services clients in Tehran, Iran. In fact, they predicted a wetter and warmer climate for their study area by the year 2050 [16]. In another study, the effect of climate change in the Saguenay watershed in Canada was investigated by LARS-WG downscaling technique. Their results have shown a clear increase in monthly average of both minimum and maximum temperatures (2.5 °C between current and 2080s climate) but there was not any significant changes in the daily average of precipitation in the study area [49].

The impacts of climate change on cholera disease were simulated using a feed-forward MLP type of ANNs. To build a network, the trial-and-error approach was used and to find the optimum network, training multiple times was applied. In addition, cross-validation was carried out and performance of the network was tested to protect it against overtraining. The optimized NN has one hidden layer and Tan Axon and Momentum were set as layer’s transfer function and learning rule, respectively. Different studies applied ANNs to predict the diarrheal and cholera outbreaks in distinct areas of the world [21, 22, 50, 51]. Pezeshki et al. used a multilayer perception ANNs to create a model and predict cholera disease in Chabahar, Iran. They applied monthly average of temperature, humidity and rainfall as climatic variables and their results illustrated that cholera cases were significantly related to humidity and there was a significant relationship between the cholera incidence and lack of rainfall [21]. Considering the importance of controlling the infectious diseases, many studies were done to estimate the effect of climate changes on infectious diseases in human societies in the future and prepare helpful prediction tools [52]. Accordingly, in another study, a new BPNN Model was established to predict the number of infectious diarrheas in Shanghai of China with climatic variables as input. Moreover, the results of BPNN Model were compared with SVR, RFR and MLR models. Their results indicated that BPNN model creates the best prediction results compared to the MLR, RFR and SVR models. In addition, sensitivity analysis defined that temperature-related variables (Tmax, Tmin, and Tavg) were the effective climatic factors on infectious diarrhea while rainfall had minimum effect [51].

The projected scenario’s datasets were utilized on the optimized NN data-driven model to evaluate the future trend of cholera cases. These datasets included daily values of cholera cases and climate variables for 30 years (2021–2050). According to the present results, a slight increase was shown in cholera cases under the two scenarios of RCP2.6 and RCP8.5. Our data identified 4–5 years cycles between 2021 and 2050 under the scenario of RCP8.5. In one important study, the ‘cyclical periodicity’ of cholera disease was demonstrated in relation to the cycles of rainfall by the director of public health in Bengal. They reported that insufficient rainfall under drought conditions could lead to epidemics of cholera [53, 54]. In fact, changing the seasonality of cholera can be due to changes in rainfall patterns over the years. It should be noted that all of the conditions will happen when the authorities do not accomplish any preventive actions and social factors (availability of healthy water especially in suburb and development of wastewater systems) remain either the same or worse during the time. Considering extensive changes in global and local climate conditions in last decades and its impact on the spread of different diseases, it is a great necessity to find proper algorithms to predict the vastness of these diseases based on climate conditions. In this regards, ANNs are known as one of the appropriate tools for climate modeling and weather prediction [21]. In ANNs models, the sensitivity analysis was carried out for enhancing network performance and determination of the relative importance and contribution of input parameters on the output [55].

According to investigation of Qom Health Center, the main cause of cholera in Qom was determined as consumption of contaminated vegetables and untreated water in suburb. The justification of this phenomena can be found in the summer time. As mentioned, climate in Qom is dry and warm; therefore, lack of precipitation, drought and water shortage in the summer led to increasing bacterial load and contaminate groundwater by waterborne infection, V. cholerae. Water scarcity, made people to use untreated water (with a higher risk of contamination) from wells with lower depths for drinking and other usages in suburb. Irrigation of agricultural products by contaminated water and untreated wastewater and their consumption can transmit V. cholerae to high risk people. V. cholerae is the major cause of human diarrheal infection of cholera, which cause economic loss and mortality in developing countries. It is nominated as "the disease of poverty" because it has shown in areas with no access to adequate water treatment systems and sewage collection lines [56]. The best environmental conditions for V. cholerae are 30°C water temperature, 15% salinity, and almost alkaline environments (pH = 8.5) [57]. Human movement also is another factor which play a noticeable role in the epidemiology and transmission of infectious diseases. As mentioned, Qom is a religious city and hosts lots of tourists from endemic countries such as Iraq, Afghanistan and Pakistan annually. All of these suitable conditions were provided in Qom and it could be resulted intense growth of bacteria and increase incidence rate of cholera infection. Therefore, the promotion of any control measures such as the development of sewage collection lines, the use of treated wastewater for irrigation of farms, monitoring on the production/distribution of foods and raw vegetables can help to reduce the predicted number of cholera incidence by the model in the future.

One limitation of the present study was lack of long-term monitoring data on cholera and diarrhea disease especially in details of patient data such as age and sex categories and social and economic classes. Another limitation of our study was lack of hygienic, social and demographic information in the modeling process. Addition of some determinants affecting cholera such as availability to safe water, wastewater collection system, sanitation access, consumption of raw sewage in irrigation, hygiene in restaurants and other food providers and etc. can promote the potential of model to predict incidence of cholera with more accuracy. However, it may also require more effort, cost, and time. For further work, these impact factors can add to model and it may be helpful to use this approach to other waterborne infectious diseases.

Conclusion

Cholera outbreaks and diarrhea disease can be linked to climatic variables and climate change. After establishing a surveillance system and recording cholera data in Qom, the information showed that the first cholera epidemic started in July 1998 and continued over the 15 years. To project the future climate using outputs from GCMs under two scenarios of the RCP2.6 and RCP8.5, we used statistical downscaling by LARS-WG as an efficient tool. Results of this study demonstrated a warmer and wetter climate pattern in our studied area by 2050. Based on the feedforward MLP type ANN technique, this study indicated that annual cholera cases will increase under both RCP scenarios during 2021–2050.

Our results showed that ANNs is a very useful and executive tool for simulation of infectious diseases such as cholera to perform preventive actions and decrease fatalities and also limit unwanted social results. But, because of some mentioned limitations of our dataset, sensitivity of simulation is not high enough. Therefore, it is important to get comprehensive and influential data including hygienic, social and demographic parameters. These results were obtained based on the present social conditions and surveillance systems. Promotion in surveillance system, increasing in the number of epidemiologic studies and identification of climatic and social risks factors can help to enhance control and prevention interventions.

Acknowledgments

Authors would like to express their gratitude to the authorities of Iran meteorological organization (IRIMO). The authors also wish to thank of Department of Communicable Disease, Deputy of Health, Qom University of Medical Sciences, Qom, Iran for providing their daily records. This article has been extracted from the Ph.D. thesis written by Zahra Asadgol in the Department of Environmental Health Engineering, School of Public health at Iran University of Medical Sciences.

References

  1. 1. Semenza JC, Suk JE, Estevez V, Ebi KL, Lindgren E. Mapping climate change vulnerabilities to infectious diseases in Europe. Environmental health perspectives. 2011;120(3):385–92. pmid:22113877
  2. 2. Greer A, Ng V, Fisman D. Climate change and infectious diseases in North America: the road ahead. Canadian Medical Association Journal. 2008;178(6):715–22. pmid:18332386
  3. 3. Mills JN, Gage KL, Khan AS. Potential influence of climate change on vector-borne and zoonotic diseases: a review and proposed research plan. Environmental health perspectives. 2010;118(11):1507–14. pmid:20576580
  4. 4. Tonnang HE, Kangalawe RY, Yanda PZ. Predicting and mapping malaria under climate change scenarios: the potential redistribution of malaria vectors in Africa. Malaria journal. 2010;9(1):111.
  5. 5. González C, Wang O, Strutz SE, González-Salazar C, Sánchez-Cordero V, Sarkar S. Climate change and risk of leishmaniasis in North America: predictions from ecological niche models of vector and reservoir species. PLoS neglected tropical diseases. 2010;4(1):e585. pmid:20098495
  6. 6. de Magny GC, Thiaw W, Kumar V, Manga NM, Diop BM, Gueye L, et al. Cholera outbreak in Senegal in 2005: was climate a factor? PLoS One. 2012;7(8):e44577. pmid:22952995
  7. 7. Emch M, Feldacker C, Islam MS, Ali M. Seasonality of cholera from 1974 to 2005: a review of global patterns. International Journal of Health Geographics. 2008;7(1):31. pmid:18570659
  8. 8. Weekly Epidemiological Record [Internet]. 21 September 2018.
  9. 9. Morens DM, Fauci AS. Emerging Infectious Diseases: Threats to Human Health and Global Stability. PLOS Pathogens. 2013;9(7):e1003467. pmid:23853589
  10. 10. Olago D, Marshall M, Wandiga SO, Opondo M, Yanda PZ, Kangalawe R, et al. Climatic, socio-economic, and health factors affecting human vulnerability to cholera in the Lake Victoria basin, East Africa. AMBIO: A Journal of the Human Environment. 2007;36(4):350–8.
  11. 11. Luque Fernández MÁ, Bauernfeind A, Jiménez JD, Gil CL, Omeiri NE, Guibert DH. Influence of temperature and rainfall on the evolution of cholera epidemics in Lusaka, Zambia, 2003–2006: analysis of a time series. Transactions of the Royal Society of Tropical Medicine and Hygiene. 2009;103(2):137–43. pmid:18783808
  12. 12. Hashizume M, Armstrong B, Hajat S, Wagatsuma Y, Faruque AS, Hayashi T, et al. The effect of rainfall on the incidence of cholera in Bangladesh. Epidemiology. 2008;19(1):103–10. pmid:18091420
  13. 13. Ruiz-Moreno D, Pascual M, Bouma M, Dobson A, Cash B. Cholera seasonality in Madras (1901–1940): dual role for rainfall in endemic and epidemic regions. EcoHealth. 2007;4(1):52–62.
  14. 14. Hashizume M, Faruque AS, Wagatsuma Y, Hayashi T, Armstrong B. Cholera in Bangladesh:" Climatic Components of Seasonal Variation". Epidemiology. 2010:706–10. pmid:20562706
  15. 15. Lipp EK, Huq A, Colwell RR. Effects of global climate on infectious disease: the cholera model. Clinical microbiology reviews. 2002;15(4):757–70. pmid:12364378
  16. 16. Mohammadi H, Ardalan A, Bavani AM, Naddafi K, Talebian MTJHS. Simulation of Climate Change Impact on Emergency Medical Services Clients Caused by Air Pollution. 2018;7(2).
  17. 17. Nasr-Azadani F, Khan R, Rahimikollu J, Unnikrishnan A, Akanda A, Alam M, et al. Hydroclimatic sustainability assessment of changing climate on cholera in the Ganges-Brahmaputra basin. Advances in Water Resources. 2017;108:332–44.
  18. 18. de Magny GC, Murtugudde R, Sapiano MR, Nizam A, Brown CW, Busalacchi AJ, et al. Environmental signatures associated with cholera epidemics. Proceedings of the National Academy of Sciences. 2008;105(46):17676–81.
  19. 19. Pascual M, Chaves L, Cash B, Rodó X, Yunus M. Predicting endemic cholera: the role of climate variability and disease dynamics. Climate Research. 2008;36(2):131–40.
  20. 20. Akanda AS, Jutla AS, Gute DM, Evans T, Islam S. Reinforcing cholera intervention through prediction-aided prevention. Bulletin of the World Health Organization. 2012;90:243–4. pmid:22461722
  21. 21. Pezeshki Z, Tafazzoli-Shadpour M, Nejadgholi I, Mansourian A, Rahbar M. Model of cholera forecasting using artificial neural network in Chabahar City, Iran. Int J Enteric Pathog. 2016;4(1):1–8.
  22. 22. Penna MLF. Use of an artificial neural network for detecting excess deaths due to cholera in Ceará, Brazil. Revista de saude publica. 2004;38(3):351–7. pmid:15243663
  23. 23. climate-data. Available from: https://en.climate-data.org/asia/iran/qom/qom-956725/.
  24. 24. weather-atlas. Available from: https://www.weather-atlas.com/en/iran/qom-climate.
  25. 25. Koncar N. Optimisation methodologies for direct inverse neurocontrol: University of London; 1997.
  26. 26. Stefánsson A, Končar N, Jones AJJNC, Applications. A note on the gamma test. 1997;5(3):131–3.
  27. 27. Chuzhanova NA, Jones AJ, Margetts SJB. Feature selection for genetic sequence classification. 1998;14(2):139–43. pmid:9545445
  28. 28. Tsui APM. Smooth data modelling and stimulus-response via stabilisation of neural chaos: University of London; 1999.
  29. 29. Jones AJCUoW, Cardiff. The WinGamma User Guide. 1998;2001.
  30. 30. Moghaddamnia A, Gousheh MG, Piri J, Amin S, Han DJAiWR. Evaporation estimation using artificial neural networks and adaptive neuro-fuzzy inference system techniques. 2009;32(1):88–97.
  31. 31. Piri J, Amin S, Moghaddamnia A, Keshavarz A, Han D, Remesan RJJoHE. Daily pan evaporation modeling in a hot and dry climate. 2009;14(8):803–11.
  32. 32. Nasr-Azadani F, Unnikrishnan A, Akanda A, Islam S, Alam M, Huq A, et al. Downscaling river discharge to assess the effects of climate change on cholera outbreaks in the Bengal Delta. 2015;64(3):257–74.
  33. 33. Semenov MA, Stratonovitch PJCr. Use of multi-model ensembles from global climate models for assessment of climate change impacts. 2010;41(1):1–14.
  34. 34. Haris AA, Khan M, Chhabra V, Biswas S, Pratap A. Evaluation of LARS-WG for generating long term data for assessment of climate change impact in Bihar. 2010.
  35. 35. Yomwan P, Cao C, Rakwatin P, Suphamitmongkol W, Tian R, Saokarn AJG, Natural Hazards, et al. A study of waterborne diseases during flooding using Radarsat-2 imagery and a back propagation neural network algorithm. 2015;6(4):289–307.
  36. 36. HW B. The history of cholera in India from 1862 to 1881: Tru¨bner & Company; 1885.
  37. 37. Pascual M, Bouma MJ, Dobson AP. Cholera and climate: revisiting the quantitative evidence. Microbes and Infection. 2002;4(2):237–45. pmid:11880057
  38. 38. Iacono GL, Armstrong B, Fleming LE, Elson R, Kovats S, Vardoulakis S, et al. Challenges in developing methods for quantifying the effects of weather and climate on water-associated diseases: A systematic review. PLoS neglected tropical diseases. 2017;11(6):e0005659. pmid:28604791
  39. 39. Martinez ME. The calendar of epidemics: Seasonal cycles of infectious diseases. PLoS pathogens. 2018;14(11):e1007327. pmid:30408114
  40. 40. Yue Y, Gong J, Wang D, Kan B, Li B, Ke C. Influence of climate factors on Vibrio cholerae dynamics in the Pearl River estuary, South China. World Journal of Microbiology and Biotechnology. 2014;30(6):1797–808. pmid:24442820
  41. 41. Mendelsohn J, Dawson T. Climate and cholera in KwaZulu-Natal, South Africa: the role of environmental factors and implications for epidemic preparedness. International journal of hygiene and environmental health. 2008;211(1–2):156–62. pmid:17383231
  42. 42. Ramírez IJ. Cholera resurgence in Piura, Peru: examining climate associations during the 1997–1998 El Niño. GeoJournal. 2015;80(1):129–43.
  43. 43. WANG Z, Yao L, Guo Z, editors. The estimation of reference evapotranspiration based on gamma test and gene expression programming using the weather data set from different climatic zones in China. 2015 ASABE Annual International Meeting; 2015: American Society of Agricultural and Biological Engineers.
  44. 44. Bandyopadhyay S, Kanji S, Wang L. The impact of rainfall and temperature variation on diarrheal prevalence in Sub-Saharan Africa. Applied Geography. 2012;33:63–72.
  45. 45. Poveda G, Rojas W, Quiñones ML, Vélez ID, Mantilla RI, Ruiz D, et al. Coupling between annual and ENSO timescales in the malaria-climate association in Colombia. Environmental health perspectives. 2001;109(5):489–93. pmid:11401760
  46. 46. Nasr-Azadani F, Unnikrishnan A, Akanda A, Islam S, Alam M, Huq A, et al. Downscaling river discharge to assess the effects of climate change on cholera outbreaks in the Bengal Delta. Climate Research. 2015;64(3):257–74.
  47. 47. Hartman J, Ebi K, McConnell KJ, Chan N, Weyant J. Climate suitability: for stable malaria transmission in Zimbabwe under different climate change scenarios. Global Change and Human Health. 2002;3(1):42.
  48. 48. Semenov MA, Stratonovitch P. Use of multi-model ensembles from global climate models for assessment of climate change impacts. Climate research. 2010;41(1):1–14.
  49. 49. Dibike YB, Coulibaly P. Hydrologic impact of climate change in the Saguenay watershed: comparison of downscaling methods and hydrologic models. Journal of hydrology. 2005;307(1–4):145–63.
  50. 50. Yomwan P, Cao C, Rakwatin P, Suphamitmongkol W, Tian R, Saokarn A. A study of waterborne diseases during flooding using Radarsat-2 imagery and a back propagation neural network algorithm. Geomatics, Natural Hazards and Risk. 2015;6(4):289–307.
  51. 51. Wang Y, Li J, Gu J, Zhou Z, Wang Z. Artificial neural networks for infectious diarrhea prediction using meteorological factors in Shanghai (China). Applied Soft Computing. 2015;35:280–90.
  52. 52. Organization WH. Protecting health from climate change: report for World Health Day 2008. 2008.
  53. 53. Bryden JL. Report on the General Aspects of Epidemic Cholera in 1869: A Sequel to A Report on the Cholera of 1866–68: Office of Superintendent of Government Print.; 1870.
  54. 54. Bryden JL. Epidemic Connection of the Cholera of Madras and Bombay with Cholera Epidemics of the Bengal Presidency: Office of Superintendent of Government Print.; 1871.
  55. 55. Acherjee B, Mondal S, Tudu B, Misra D. Application of artificial neural network for predicting weld quality in laser transmission welding of thermoplastics. Applied soft computing. 2011;11(2):2548–55.
  56. 56. V PS. Influence of environmental factors on the presence of Vibrio cholerae in the marine environment: a climate link. J Infect Dev Ctries. 2007;1:224–41. https://doi.org/10.3855/jidc.359. pmid:19734600
  57. 57. Chowdhury FR, Nur Z, Hassan N, von Seidlein L, Dunachie S. Pandemics, pathogenicity and changing molecular epidemiology of cholera in the era of global warming. Annals of Clinical Microbiology and Antimicrobials. 2017;16(1):10. pmid:28270154