The drivers of West Nile virus human illness in the Chicago, Illinois, USA area: Fine scale dynamic effects of weather, mosquito infection, social, and biological conditions

West Nile virus (WNV) has consistently been reported to be associated with human cases of illness in the region near Chicago, Illinois. However, the number of reported cases of human illness varies across years, with intermittent outbreaks. Several dynamic factors, including temperature, rainfall, and infection status of vector mosquito populations, are responsible for much of these observed variations. However, local landscape structure and human demographic characteristics also play a key role. The geographic and temporal scales used to analyze such complex data affect the observed associations. Here, we used spatial and statistical modeling approaches to investigate the factors that drive the outcome of WNV human illness on fine temporal and spatial scales. Our approach included multi-level modeling of long-term weekly data from 2005 to 2016, with weekly measures of mosquito infection, human illness and weather combined with more stable landscape and demographic factors on the geographical scale of 1000m hexagons. We found that hot weather conditions, warm winters, and higher MIR in earlier weeks increased the probability of an area of having a WNV human case. Higher population and the proportion of urban light intensity in an area also increased the probability of observing a WNV human case. A higher proportion of open water sources, percentage of grass land, deciduous forests, and housing built post 1990 decreased the probability of having a WNV case. Additionally, we found that cumulative positive mosquito pools up to 31 weeks can strongly predict the total annual human WNV cases in the Chicago region. This study helped us to improve our understanding of the fine-scale drivers of spatiotemporal variability of human WNV cases.

region in order to evaluate relationships between the factors on fine temporal and spatial scale and identify the drivers that potentially affect the presence of human WNV illness and may act as early warning predictors. The paper is well written with a well-organized text, the data were analyzed using multilevel statistical modeling approaches and the findings are sufficiently documented and the results are valuable for a better understanding of the fine-scale drivers of spatiotemporal variability of WNV human case prevalence in an urban environment such as in the study area. Although numerous published studies that have shed light on factors that affect WNV transmission in an area, the knowledge regarding the influence of climatic variables in correlation with the data from the entomological surveillance and the number of WNV human cases, is still limited. For that reason, the paper makes a substantial contribution to the literature and is therefore recommended for publication in PLOS-ONE after minor revision taking into account the following general or specific comments. •Thank you for your comments and your feedback!

General comments
The study uses and analyzes the 10-year data (2005 to 2014) from Cook and DuPage counties in the Chicago, Illinois region and the accuracy of the predictions of the developed model tested with the data of the same specific area. However, according to the literature, it is well known that models predicting the WNV transmission and human WNV infections do not always have the same accuracy when applied to other areas with different mosquito fauna, weather conditions and/or geomorphological and demographic characteristics. Therefore, we consider that the study area should also be mentioned in the title. •Thank you for the suggestion, we have made that change Please comment and, if necessary, provide an adequate justification in the manuscript, for the reason that in this work were note included data from passive or active monitoring of WNV presence in birds and equids, which are considered by several authors as important prediction factors of the presence and spread of WNV virus in an area.
•We have added a statement (145-147) that the avian and equid surveillance programs were not consistent across the time period, and added a discussion section (467-474) about the point.

Specific comments
Line 170 of the manuscript: If available, please provide information on the species of Culex mosquitoes that have been tested for WNV presence as the vectorial competence of different species may vary significantly for WNV transmission to humans. •We agree that is an important point; we have added some information as to common species in the region.
Line 179 of the manuscript: Please add a bibliographical reference in the reference section for the MIR estimation tool by Biggerstaff, 2006. •Thank you, corrected Line 188 of the manuscript: Please provide a definition and some additional information about the category of "probable cases of WNV" that were also included in the study along with the "confirmed cases" because the symptoms of infection by the West Nile vary in severity, with the mild forms can be easily confused with flu symptoms and usually go unreported.
•We have added the information. The difference between probable and confirmed cases is confirmatory testing by either IDPH or CDC; all cases had positive diagnostic results and clinical signs during the likely transmission season.
Lines 578-580 of the manuscript: Please, correct Reference no 39 by adding the name of the journal, volume number and pages numbers. Messina  This is a well written paper that deals with the determination of factors affecting the spatiotemporal variability of WNV cases in humans through identification of the fine scale drivers of WNV transmission in an urban area with a repeated history of WNV outbreaks. The findings are very interesting since they include multi-level modeling of weekly data from over a decade and they extend our knowledge in the correlation of variables related to temperature, precipitation, mosquito infection, land cover, and demographic characteristics with the probability of an area having a WNV case or not. •Thank you Further down please consider some comments of minor importance that may benefit the manuscript.
It seems that the infection status of avian population, as primary reservoirs of WNV, and equids, as dead-end hosts, were not included among the tested variables for modeling structure. Please note that these are critical factors implicated in the WNV transmission in order to develop predictive models. As mentioned in the introduction, public health surveillance for WNV involves collection and testing of dead birds suspected to have died of WNV, testing of sentinel chickens or of wild birds captured for this purpose and reporting of cases of equine illness. Could you please justify this data gap in the model structuring? Is there any surveillance system for infected avian and equids population in the study area? In the "Introduction" you may add any relevant literature data where bird and/or equine infection rate were used for development of models predicting WNV transmission in humans. Also, in lines 440-459 of the manuscript, you could mention the fact that avian and equids infection status was not considered as a factor for prediction of WNV cases in humans in the study area.
•We have added a statement as to the inconsistent application of avian and equid surveillance in this region (145-147), and given more information about that surveillance in the discussion (467-474), including references to models using these data types.
According to the best multivariable model that was used, the proportion of open water was negatively associated with the probability of WNV cases. Also, as mentioned in the discussion, a negative association of precipitation and WNV cases was observed and this indicates that dry and hot weather conditions would increase the probability of an area being positive for a WNV case. Instead, it is supposed that high rainfall and high percentage of water bodies in an area may favor mosquito population by increasing their breeding sites, and therefore may lead to increased WNV cases in humans. Hence, a positive correlation between precipitation and water bodies with WNV cases in humans is anticipated. Please comment.
•Open water is classified as areas in which any aquatic vegetation is submerged, as opposed to woody or herbaceous wetlands. This is not likely to be stagnant water of the type used by Culex mosquitoes for breeding. Therefore, the negative association between proportion of open water and WNV cases is most likely due to the fact that open water, as defined, does not favor the mosquito population. We have noted this in the discussion (434-438).   of Culex vector mosquitoes, collection and testing of dead birds suspected to have died of 50 WNV, testing of sentinel chickens or of wild birds captured for this purpose, and reporting of 51 cases of human and equine illness [9]. The ultimate goal of these surveillance data is to target 52 mosquito control, and thereby reduce illness through the reduction of the number of infected 53 vector mosquitoes, and to target educational messages to warn citizens to reduce individual 54 exposure. One additional advantage of having a strong surveillance system in place is that the 55 long-term data generated can be integrated with publicly available weather, landscape, and 56 socioeconomic data and can be used effectively to identify the important drivers of WNV 57 [24]. Each of these spatial scales has its own inherent biases, as these political boundaries do 98 not necessarily correspond to the ecological processes of the disease in question [34]. 99 Alternatively, dividing the area into equal spaces, such as rectangular bins or hexagons, has 100 been used to reduce some of these biases (e.g. variables to assess the effects of rainfall, temperature and the WNV mosquito infection rate 113 [39]. This analysis determined that white populations and housing from the 1950s were 114 associated with increased illness in some years, but this was not consistent. Interestingly, 115 census tracts with lower rainfall had higher rates of WNV illness, but the mosquito infection 116 rate was not an important variable in any of the models [39]. 117 Despite the identification of some of these potential risk factors, accurate prediction 118 of human illness cases from WNV remains elusive at the local scale, especially as it is related 119 to dynamic weather and mosquito infection status. Using long-term data on human WNV 120 illness and intensive mosquito surveillance for the Chicago region, we can identify the fine 121 scale drivers of spatiotemporal variability of human WNV epidemic in an urban environment. 122 The overall goal of this study is to determine factors affecting the spatiotemporal variability  The two Illinois counties of Cook and DuPage, comprising Chicago and its suburbs, 136 were included in this study. The total area covered by these two counties is nearly 5,100 137 square kilometers, and the total population in 2010 was 6.1 million. These areas were 138 selected because of the relatively high incidence of human West Nile virus illness reported 139 from these two counties and the long-term intensive mosquito surveillance data available for 140 this region. The temporal window included in this study was the 24-week time period from 141 late May to late October (weeks 22 to 45), which corresponds to the timing of mosquito hexagons were included in the analysis. All independent variables related to weather, land 154 cover, mosquito infection and demography were calculated for each hexagon, as described 155

Mosquito data 157
Mosquito testing data from 2005 to 2016 were obtained from the Illinois Department 158 of Public Health (IDPH) through a user agreement. The IDPH collates the data from local 159 public health agencies and mosquito abatement districts across Illinois and maintains a 160 statewide database for the results from WNV mosquito testing. The IDPH developed a 161 mosquito surveillance protocol that local health and mosquito abatement districts are 162 expected to follow in order to standardize the mosquito collection and testing across the state. 163 In general, the local agencies collect vector mosquitoes with gravid traps, identify the sex and 164 species of the mosquitoes, and make pools of up to 50 mosquitoes of a single species from 165 those captured in each trap to test for the presence of WNV infection. When fewer then 50 166 mosquitoes are captured, a pool will consist of fewer than 50 mosquitoes. During the study 167 period, the common tests used to identify WNV in mosquitoes included antigen assays, 168 VecTest or the Rapid Analyte Measurement Platform (RAMP) test. Some pools were also 169 tested by Real Time reverse transcriptase polymerase chain reaction (RT-PCR). In instances 170 when a pool was tested using more than one type of test, only the RT-PCR results were used 171 in the analysis. Our analysis used only the test results from pools of female Culex 172 mosquitoes. Not all mosquitoes were identified to species prior to testing; however, the 173 majority of Culex collected in this region belong to the species Cx. pipiens or Cx. restuans 174 To determine the location of the mosquito traps, we used the existing latitude and 176 longitude recorded in the IDPH database. In cases where the spatial data were missing, we 177 and 1989, and houses built after 1990. These demographic data were processed in ArcGIS 213 using the intersection tool to calculate a parameter for each hexagon. 214

Landcover data 215
Landcover data for the entire United States was obtained from the national landcover University, http://prism.oregonstate.edu). The PRISM daily data are available as spatial grids 233 of 4 km resolution, which are calculated through interpolation and statistical techniques using 234 point data from weather monitoring networks across the country combined with topographic 235 data. These daily data were used to calculate the weekly temperature and precipitation. For 236 our analysis, the weekly mean temperature was calculated by taking the average of the seven 237 daily averages for that week, and the weekly precipitation was calculated as a sum of the 238 daily precipitation for that week. Finally, the weekly temperature and precipitation for each 239 year and week for each hexagon was calculated by using the zonal statistics as table function 240 in ArcGIS 10.1. We also calculated average January temperature for each hexagon for each 241 year from the daily data as a proxy for the winter temperature. 242

Statistical methods 243
To assess the temporal relationship between human illness and MIR, we calculated 244 the Spearman rank correlation between the weekly MIR of 1-6 weeks lag and human cases. To visualize the spatial patterns of human illness over time, we first developed 257 choropleth maps of WNV cases. Then, we used local Moran's I method using an inclusive 258 second order queen contiguity weight matrix in the spatial analysis software GeoDa to further 259 identify the spatial clusters of cumulative human WNV cases from 2005 to 2016. We also 260 examined differences in results using neighboring cells and rook contiguity weight matrix, 261 but the results did not vary. 262 For the spatiotemporal statistical model, the outcome variable was the presence/ 263 absence of a human WNV case in each hexagon for each year and week. The predictors 264 included 32 variables related to weather, land cover, mosquito infection and demography 265 (Table 1)  We found a strong temporal relationship between the MIR of previous weeks and 300 human WNV cases in the study region (Table 3, Fig 1). The strongest correlation (r= 0.837) 301 was with MIR at a one-week lag ( Table 3) (Table 3). When evaluated for only 2012, when case counts were 304 highest, the correlation between MIR and human WNV cases was also the highest (r= 0.899). 305 In both high and low years, the strength of the correlation gradually declined with the number 306 of weeks lagged and there was almost no correlation with MIR after lags of four weeks. 307 308 309 We found that the MIR of mid-summer (weeks 28-33) was able to explain 93% of the 317 variability in total annual human cases (Table 4, Fig 2). The model predicted 44.8 human 318 cases for 2015, compared to 35 actual cases, and 142.7 human cases for 2016 compared to 319 108 actual cases. Likewise, the cumulative number of positive pools also strongly explained 320 and predicted the total annual human cases (Table 4, Fig 3); the cumulative number of 321 positive mosquito pools by week 31 explained 93% of the variability in total annual human 322 cases, similar to that explained by mid-summer MIR (  The spatial pattern of human WNV cases in Cook and DuPage counties showed that 339 cases were distributed throughout most areas of the study region at some point during the 340 study period, with some pockets of higher numbers of cases (Fig 4). Out of the total 5,345 341 hexagons in the study area, 750 hexagons had experienced at least one case of human WNV 342 case during the years 2005 to 2016. Cumulatively, 123 hexagons had more than one human 343 WNV case, with the maximum number of cases in a hexagon being five (Fig 4). The local 344 Moran's I identified some spatial clusters of human WNV cases in Cook and DuPage 345 counties (Fig 5): 92 hexagons with higher numbers of cases were also near to others with 346 higher numbers of cases. variables that included temperature, MIR, land cover, and demographic characteristics was 359 the best model (Table 5). The final multivariable model indicated that higher temperatures 360 two, three, and four weeks earlier and warmer average January temperature were associated 361 with a higher probability of a hexagon being positive for human WNV case ( Table 6). The 362 lagged mosquito infection rates of one to four weeks earlier were also positively associated 363 with the outcome variable (Table 6). Among the land cover variables, the proportion of open 364 water, grassland, and deciduous forests were negatively associated with the probability of a 365 WNV case while the proportion of low intensity developed areas was positively associated 366 (Table 6). Among the demographic variable, total population was found to be positively 367 associated with the probability of a WNV case, while the proportion of housing built after 368 1990 was negatively associated ( Table 6). The area under the ROC curve was 0.948, which 369 indicates that model performance was excellent (Fig 6). 370 used long-term data on human illness, mosquito surveillance, weather, landscape, and 385 demographic data. We found significant spatial clusters of human WNV cases within this 386 urban environment. We also found a strong correlation between the weekly MIR of earlier 387 weeks and weekly human WNV cases, and further developed predictive temporal models 388 using mid-summer average MIR and cumulative positive mosquito pools which can be used 389 to estimate the total annual human WNV cases. 390 The temporal variation in the weekly human WNV cases was strongly correlated with 391 MIR of one to four weeks earlier, with a correlation of one week earlier being the strongest. 392 This finding was similar to our earlier model based on Illinois climate divisions, in which 393 Division 2 includes our current study area [42]. The similarity in the correlation may be due 394 to the fact that the data for Climate Division 2 were dominated by the data from Cook and 395 DuPage, as these counties have more intensive surveillance compared to other Illinois 396 counties. However, similar observations were also found in Ontario, Canada, where MIR of 397 one week earlier was most strongly correlated with the weekly variation in human WNV 398 cases [16]. In our study, we also found that the correlations between weekly MIR and human 399 cases increased in high WNV years, which was also observed in a study conducted in Long 400 Island, New York [18]. This is understandable, as stochastic variability decreases with 401 increased numbers of cases, allowing for more precise estimation. 402 The temporal models we developed using mid-summer average MIR and cumulative 403 mosquito positive pools were both able to explain more than 90% of the variability in the week 34 was suggested as an action threshold potential to estimate the total annual human 413 cases [16]. In Chicago, we obtained this signal three weeks earlier, which can be crucial to 414 the ability to intervene in the upcoming potential human WNV outbreak. 415 We found spatial clustering of human WNV cases within the study area, indicating 416 that some areas were more likely than others to have a WNV human case. A spatial clustering 417 pattern of human WNV cases in Chicago area was also observed in the 2002 WNV outbreak 418 year [12]. Several factors might play a role in the observed spatial clustering pattern, 419 including differences in the fine-scale variation in the local landscape structure that affects 420 mosquito population, fine-scale weather variation, demographic characteristics, access of 421 people to health care system, and spatially variable mosquito abatement practices 422 [12, 39,45,46]. 423 In this study, through multilevel modeling, we identified several dynamic factors that 424 are possibly driving the fine scale spatiotemporal variation in the human WNV cases 425 occurrence in the Chicago region. We found that the higher temperature in the previous 426 weeks increases the probability of an area being positive for a WNV case. The association 427 between higher temperature and WNV human illness has also been observed in other studies 428 conducted at different spatial scales [15,17,20]. This is possibly due to the dynamic effect of We also found increased MIR up to four weeks earlier will increase the probability of 443 an area being positive for a WNV human case. The temporal association between lagged 444 MIR and human WNV cases is relatively well established [10,16,52]. However, it was 445 interesting to find the positive association of MIR when spatiotemporal variabilities of human 446 cases were considered. In our current analysis, we found that areas with a higher percentage 447 of white population had a higher probability of being positive for WNV, which has also been 448 observed in a previous study of this [12]. This may be a function of access to the health care 449 system and likelihood of seeking medical treatment and testing [12,27], or may simply be due 450 to high proportions of white population in areas of the study region where environmental 451 conditions are also conducive to increased mosquito activity. 452 This study also found that the probability of a hexagon being a positive for WNV case 453 decreased in developed medium and high-intensity urban areas and increased in developed 454 low-intensity urban areas, indicating that the suburban areas of Chicago are more at risk than 455 the highly developed urban centers. The lack of mosquito breeding grounds and bird activity 456 in the high-intensity urban areas might be responsible for this. Previous studies conducted in 457 the same area have also indicated that sub-urban region in Chicago is at more risk from the 458 WNV [12,27]. This is probably due to the poor sanitation system in the older houses 459 compared to new houses. 460 In this study, we did not consider prior seasonal differences in the weather conditions, 461 which we recommend be incorporated in future studies. In addition, the calculation of MIR 462 for hexagons may be biased as the IDW interpolation technique used to develop continuous 463 surface maps is affected by the uneven distribution of mosquito traps across the study area. 464 Alternatively, other interpolation methods such as kriging might be used to develop 465 continuous surface maps for MIR, as this method takes into account spatial autocorrelation 466 and also creates an error map. In this study, we did not distinguish between neuroinvasive 467 and non-neuroinvasive WNV cases. Separate analysis for only neuroinvasive cases might 468 help us to identify what conditions drive the occurrence of the severe form of WNV infection 469 and should also help to reduce diagnostic bias. Also, in future studies, we might consider 470 using different spatial scales to identify if the geographic scale has affected the results. We 471 were also unable to use data from avian or equid surveillance in this study, despite its 472 usefulness in other modeling approaches [53][54][55], due to the lack of consistent data across the 473 time period. Bird surveillance in Illinois is limited to passive surveillance of a small number 474 of dead birds tested in each county per year, and is generally suspended after WNV is known 475 to be circulating in the area, while equid surveillance is based entirely on passive self-476 reporting [56]. This lack of consistent data on avian mortality has been noticed by others 477 [10], and remains an issue for the use of data on the primary host in WNV forecasting. 478 In conclusion, our analysis helped to better understand the fine-scale dynamic drivers           vector mosquitoes, and to target educational messages to warn citizens to reduce individual 54 exposure. One additional advantage of having a strong surveillance system in place is that the 55 long-term data generated can be integrated with publicly available weather, landscape, and 56 socioeconomic data and can be used effectively to identify the important drivers of WNV 57 transmission and to develop predictive models [10,11]. [24]. Each of these spatial scales has its own inherent biases, as these political boundaries do 98 not necessarily correspond to the ecological processes of the disease in question [34]. 99 Alternatively, dividing the area into equal spaces, such as rectangular bins or hexagons, has 100 been used to reduce some of these biases (e.g. variables to assess the effects of rainfall, temperature and the WNV mosquito infection rate 113 [39]. This analysis determined that white populations and housing from the 1950s were 114 associated with increased illness in some years, but this was not consistent. Interestingly, 115 census tracts with lower rainfall had higher rates of WNV illness, but the mosquito infection 116 rate was not an important variable in any of the models [39]. 117 Despite the identification of some of these potential risk factors, accurate prediction 118 of human illness cases from WNV remains elusive at the local scale, especially as it is related 119 to dynamic weather and mosquito infection status. Using long-term data on human WNV 120 illness and intensive mosquito surveillance for the Chicago region, we can identify the fine 121 scale drivers of spatiotemporal variability of human WNV epidemic in an urban environment. 122 The overall goal of this study is to determine factors affecting the spatiotemporal variability The two Illinois counties of Cook and DuPage, comprising Chicago and its suburbs, 136 were included in this study. The total area covered by these two counties is nearly 5,100 137 square kilometers, and the total population in 2010 was 6.1 million. These areas were 138 selected because of the relatively high incidence of human West Nile virus illness reported 139 from these two counties and the long-term intensive mosquito surveillance data available for In general, the local agencies collect vector mosquitoes with gravid traps, identify the sex and 164 species of the mosquitoes, and make pools of up to 50 mosquitoes of a single species from 165 those captured in each trap to test for the presence of WNV infection. When fewer then 50 166 mosquitoes are captured, a pool will consist of fewer than 50 mosquitoes. During the study 167 period, the common tests used to identify WNV in mosquitoes included antigen assays, 168 VecTest or the Rapid Analyte Measurement Platform (RAMP) test. Some pools were also 169 tested by Real Time reverse transcriptase polymerase chain reaction (RT-PCR). In instances 170 when a pool was tested using more than one type of test, only the RT-PCR results were used 171 in the analysis. Our analysis used only the test results from pools of female Culex 172 mosquitoes. Not all mosquitoes were identified to species prior to testing; however, the 173 majority of Culex collected in this region belong to the species Cx. pipiens or Cx. restuans 174 To determine the location of the mosquito traps, we used the existing latitude and 176 longitude recorded in the IDPH database. In cases where the spatial data were missing, we 177  (Biggerstaff, 2006  data. These daily data were used to calculate the weekly temperature and precipitation. For 236 our analysis, the weekly mean temperature was calculated by taking the average of the seven 237 daily averages for that week, and the weekly precipitation was calculated as a sum of the 238 daily precipitation for that week. Finally, the weekly temperature and precipitation for each 239 year and week for each hexagon was calculated by using the zonal statistics as table function 240 in ArcGIS 10.1. We also calculated average January temperature for each hexagon for each 241 year from the daily data as a proxy for the winter temperature. 242

Statistical methods 243
To assess the temporal relationship between human illness and MIR, we calculated 244 the Spearman rank correlation between the weekly MIR of 1-6 weeks lag and human cases.  We found a strong temporal relationship between the MIR of previous weeks and 300 human WNV cases in the study region (Table 3, Fig 1). The strongest correlation (r= 0.837) 301 was with MIR at a one-week lag ( Table 3) (Table 3). When evaluated for only 2012, when case counts were 304 highest, the correlation between MIR and human WNV cases was also the highest (r= 0.899). 305 In both high and low years, the strength of the correlation gradually declined with the number 306 of weeks lagged and there was almost no correlation with MIR after lags of four weeks. 307 308 309 We found that the MIR of mid-summer (weeks 28-33) was able to explain 93% of the 317 variability in total annual human cases (Table 4, Fig 2). The model predicted 44.8 human 318 cases for 2015, compared to 35 actual cases, and 142.7 human cases for 2016 compared to 319 108 actual cases. Likewise, the cumulative number of positive pools also strongly explained 320 and predicted the total annual human cases (Table 4, Fig 3); the cumulative number of 321 positive mosquito pools by week 31 explained 93% of the variability in total annual human 322 cases, similar to that explained by mid-summer MIR (  The spatial pattern of human WNV cases in Cook and DuPage counties showed that 339 cases were distributed throughout most areas of the study region at some point during the 340 study period, with some pockets of higher numbers of cases (Fig 4). Out of the total 5,345 341 hexagons in the study area, 750 hexagons had experienced at least one case of human WNV 342 case during the years 2005 to 2016. Cumulatively, 123 hexagons had more than one human 343 WNV case, with the maximum number of cases in a hexagon being five (Fig 4). The local 344 counties (Fig 5): 92 hexagons with higher numbers of cases were also near to others with 346 higher numbers of cases. variables that included temperature, MIR, land cover, and demographic characteristics was 359 the best model (Table 5). The final multivariable model indicated that higher temperatures 360 two, three, and four weeks earlier and warmer average January temperature were associated 361 with a higher probability of a hexagon being positive for human WNV case (Table 6). The 362 lagged mosquito infection rates of one to four weeks earlier were also positively associated 363 with the outcome variable (Table 6). Among the land cover variables, the proportion of open 364 water, grassland, and deciduous forests were negatively associated with the probability of a 365 WNV case while the proportion of low intensity developed areas was positively associated 366 (Table 6). Among the demographic variable, total population was found to be positively 367 associated with the probability of a WNV case, while the proportion of housing built after 368 1990 was negatively associated (Table 6). The area under the ROC curve was 0.948, which 369 indicates that model performance was excellent (Fig 6). 370 WNV cases in Chicago region, Illinois, an area of ongoing WNV transmission. Our analysis 384 used long-term data on human illness, mosquito surveillance, weather, landscape, and 385 demographic data. We found significant spatial clusters of human WNV cases within this 386 urban environment. We also found a strong correlation between the weekly MIR of earlier 387 weeks and weekly human WNV cases, and further developed predictive temporal models 388 using mid-summer average MIR and cumulative positive mosquito pools which can be used 389 to estimate the total annual human WNV cases. 390 The temporal variation in the weekly human WNV cases was strongly correlated with 391 MIR of one to four weeks earlier, with a correlation of one week earlier being the strongest. 392 This finding was similar to our earlier model based on Illinois climate divisions, in which 393 Division 2 includes our current study area [42]. The similarity in the correlation may be due 394 to the fact that the data for Climate Division 2 were dominated by the data from Cook and 395 DuPage, as these counties have more intensive surveillance compared to other Illinois 396 counties. However, similar observations were also found in Ontario, Canada, where MIR of 397 one week earlier was most strongly correlated with the weekly variation in human WNV 398 cases [16]. In our study, we also found that the correlations between weekly MIR and human 399 cases increased in high WNV years, which was also observed in a study conducted in Long 400 Island, New York [18]. This is understandable, as stochastic variability decreases with 401 increased numbers of cases, allowing for more precise estimation. 402 The temporal models we developed using mid-summer average MIR and cumulative 403 mosquito positive pools were both able to explain more than 90% of the variability in the 404 annual number of human cases. This similarity of the results was not surprising, as positive 405 mosquito pools are used to calculate the MIR. However, the cumulative positive pools up to 406 week 31 better predicted the annual human cases compared to mid-summer average MIR for 407 408 variability of the MIR calculation depending on the mosquito pool size [43,44]. Taking the 409 most extreme possibility, when there was only one mosquito in a pool and it tested positive, 410 this would yield a MIR of 1000 in contrast to MIR of 20 when a pool with 50 mosquitoes was 411 tested positive. In Ontario, Canada, the cumulative number of positive mosquito pools up to 412 week 34 was suggested as an action threshold potential to estimate the total annual human 413 cases [16]. In Chicago, we obtained this signal three weeks earlier, which can be crucial to 414 the ability to intervene in the upcoming potential human WNV outbreak. 415 We found spatial clustering of human WNV cases within the study area, indicating 416 that some areas were more likely than others to have a WNV human case. A spatial clustering 417 pattern of human WNV cases in Chicago area was also observed in the 2002 WNV outbreak 418 year [12]. Several factors might play a role in the observed spatial clustering pattern, 419 including differences in the fine-scale variation in the local landscape structure that affects 420 mosquito population, fine-scale weather variation, demographic characteristics, access of 421 people to health care system, and spatially variable mosquito abatement practices 422 [12, 39,45,46]. 423 In this study, through multilevel modeling, we identified several dynamic factors that 424 are possibly driving the fine scale spatiotemporal variation in the human WNV cases 425 occurrence in the Chicago region. We found that the higher temperature in the previous 426 weeks increases the probability of an area being positive for a WNV case. The association 427 between higher temperature and WNV human illness has also been observed in other studies 428 conducted at different spatial scales [15,17,20]. This is possibly due to the dynamic effect of 429 higher temperature on mosquito breeding and virus replication [35,[47][48][49]. The unique 430 feature of our study is that by considering the dynamic nature of weather, we allowed the 431 temperature and precipitation to vary both temporally and spatially to capture the better role 432 Reviewer #1: Peer review report on PLOS ONE manuscript " The drivers of West Nile virus human illness: fine scale dynamic effects of weather, mosquito infection, social, and biological conditions", (Manuscript number PONE-D-19-34216).

Recommendation: Minor Revision
Comments to Authors: This manuscript analyzes the available long-term data of mosquito infection rates, West Nile virus human cases and weather variables from 2005 to 2016 combined with landscape and demographic characteristics of two Illinois counties of the Chicago region in order to evaluate relationships between the factors on fine temporal and spatial scale and identify the drivers that potentially affect the presence of human WNV illness and may act as early warning predictors. The paper is well written with a well-organized text, the data were analyzed using multi-level statistical modeling approaches and the findings are sufficiently documented and the results are valuable for a better understanding of the fine-scale drivers of spatiotemporal variability of WNV human case prevalence in an urban environment such as in the study area. Although numerous published studies that have shed light on factors that affect WNV transmission in an area, the knowledge regarding the influence of climatic variables in correlation with the data from the entomological surveillance and the number of WNV human cases, is still limited. For that reason, the paper makes a substantial contribution to the literature and is therefore recommended for publication in PLOS-ONE after minor revision taking into account the following general or specific comments.
 Thank you for your comments and your feedback!

General comments
The study uses and analyzes the 10-year data (2005 to 2014) from Cook and DuPage counties in the Chicago, Illinois region and the accuracy of the predictions of the developed model tested with the data of the same specific area. However, according to the literature, it is well known that models predicting the WNV transmission and human WNV infections do not always have the same accuracy when applied to other areas with different mosquito fauna, weather conditions and/or geomorphological and demographic characteristics. Therefore, we consider that the study area should also be mentioned in the title.
 Thank you for the suggestion, we have made that change Please comment and, if necessary, provide an adequate justification in the manuscript, for the reason that in this work were note included data from passive or active monitoring of WNV presence in birds