Abstract

Dengue is a viral disease caused by dengue viruses (DENVs) which are transmitted through the bite of an infected mosquito, primarily Aedes aegypti and Aedes albopictus, which are container-inhabiting species commonly found in urban areas. There are four distinct dengue serotypes which can cause separate, repeat infections in individuals (Wilder-Smith et al., 2019). Dengue is the most prominent arbovirus worldwide, infecting tens of millions of people every year. Globalization has expanded the geographic range of Ae. aegypti and frequent international travel has led to the reintroduction of dengue into regions where the disease was previously eliminated, such as the southern United States. Florida is uniquely vulnerable and routinely experiences dengue fever cases as a result of frequent international travel occurring between dengue endemic countries in Central and South America and the Caribbean. This project aims to address the following research questions: First, what is the spatial distribution of dengue incident rates at the county level in Florida? Do county-level incidence rates for travel and locally-acquired dengue transmission exhibit spatial clustering or randomness? Finally, what is the relationship between locally-acquired dengue incidence and travel-related incidence for the past 25 years in Florida?

Introduction

Dengue is a persistent global health threat due to global travel patterns and abundance of the mosquito vector, Ae. aegypti, responsible for its transmission in populated, urban areas. Dengue can only spread to people through the bite of an infected mosquito. Likewise, mosquitoes can become infected with dengue virus when biting an infected person. Historically, the mosquito vector was introduced into the Americas by sailing ships in water containers which contained viable eggs (Wilder-Smith & Gubler 2008). In the United States, Florida is uniquely vulnerable because of its subtropic climate, frequent travel between and proximity to dengue endemic countries, and presence competent vectors, Ae. aegypti and Ae, albopictus. Previous research has shown that temperature, humidity, and precipitation are strong predictors for dengue occurrence in Florida (Stephenson et al., 2022). The sub-tropical climate of Florida represents an ideal habitat for dengue-competent vector species such as Ae. aegypti. In addition, the number of travel-related cases has been associated with an increase in local transmission of dengue.

Although the majority of dengue cases in Florida occur when individuals travel and contract the disease abroad in dengue-endemic countries, Florida possess a unique combination of sub-tropical climate, urbanization, abundance of Ae. aegypti, and frequent international travel, creating suitable conditions for periodic local transmission. Local transmission of dengue occurs primarily in Florida’s southernmost counties (Rey, 2014). Although vertical transmission of dengue, where an infected female mosquito can infect her progeny has been documented (Ferreira-de-Lima & Lima Camara, 2018), local transmission of dengue in Florida is most dependent upon the number of imported cases which increase the likelihood of horizontal transmission from an infected person to a mosquito occurring. Understanding the spatial and temporal patterns associated with the transmission of dengue is vital towards safeguarding against its continued spread and establishment in local mosquito populations.

Consequently, this research project aims to answer the following questions: What is the spatial distribution of dengue incident rates at a county level in Florida? Do incidence rates for travel and locally-acquired dengue transmission exhibit spatial clustering or randomness? Finally, what is the relationship between locally-acquired dengue incidence and travel-related incidence at the county-level?

Data

Dengue case information was obtained from the Florida Department of Health and aggregated by year, month, county, serotype detected and categorized as either travel-related or locally acquired from 2009 to 2024. The dates associated with travel-related cases correspond to when the onset of symptoms began, whereas locally-acquired cases correspond to when the patient received care—Cases include both Florida and non-Florida residents. For travel cases, the reported county indicates the county of residence of the patient, whereas local cases represent the county where care was sought. Confirmed cases in the data set were determined by the Council of State and Territorial Epidemiologist (CSTE) case definitions which provide uniform reporting standards and criteria for infectious diseases. County-level population estimates for 2023 were obtained from the American Community Survey (ACS) five-year estimates retrieved using the tidycensus package (Walker & Herman, 2025). The unit of analysis for this project is at the county-level due to the aggregation provided by the Florida Department of Health. The two primary data sets utilized for this project are displayed in Table 1. The workflow for data processing, reshaping, and analysis are illustrated in Figure 1.

‎

DATASET SOURCE PURPOSE
Dengue Case Information Florida Department of Health Travel and local dengue cases
Estimated County Population U.S Census Bureau ACS Survey County-level population estimates

Table 1. Data sets used for dengue analysis

Methods

Figure 1. Flowchart illustrating data processing, transformation, and analysis workflow

Processing

The processing, analysis, and visualization of dengue case data were conducted using the R programming language, utilizing a variety of libraries. The tidyverse suite of packages was used for data transformation, along with the here and janitor packages for simplifying data file directories and data cleaning tasks. Retrieval of demographic and population data from the U.S. Census Bureau was accomplished by leveraging the tidycensus library. For spatial analysis, sf was used for handling county-level vector data sets and spdep facilitated the construction of spatial weight matrices for assessing spatial autocorrelation. Map figures were generated using the tmap, leaflet, and leafpop libraries. Statistical summaries and data tables were created using knitr and kableExtra for ease of formatting.

# load library
library(tidyverse)     # tidyverse 
library(tidycensus)    # acquire census data
library(here)          # path references
library(janitor)       # data cleaning
library(sf)            # handle simple features, vector spatial data
library(tmap)          # another mapping package
library(spdep)         # for spatial autocorrelation, setting spatial matrix
library(kableExtra)    # table formatting
library(plotly)        # interactive plots
library(ggiraph)       # interactive graphics
library(leaflet)       # leaflet maps
library(leafpop)       # create pop-ups for leaflet maps

# cache TIGRIS data for offline use
options(tigris_use_cache = TRUE)

The dengue dataset contains 1,063 records across five variables capturing the spatial and temporal distribution of dengue cases in Florida from 2009 to 2024. The dataset was provided at the county-level and grouped by year, month, and county, with travel and locally-acquired cases separately summarized. For data preparation, the CSV file was imported, cleaned, and reformatted. Variable names and values were standardized to be consistent between case types—records where county of residence or infection could not be determined were removed from the dataset. In the final, transformed dataset, dengue.w, the travel variable represents the number of travel-related cases for each county, year, and month. Whereas the local variable indicates the number of locally acquired cases. Counties and months without reported cases were assigned zero values to maintain consistency between case types. The dataset structure is illustrated in Table 2. Table 3 present the dengue case data summarized by county for the 25-year period.

# load dengue data from csv
dengue <- read_csv(here('data', 'dengueCasesFL.csv'))

# clean & transform dengue data set
clean_dengue <- dengue %>%
  # clean names to camel case
  clean_names(., "lower_camel") %>%
  # replace 'unknown' value with 'Unknown' for consistency between travel & imported cases
  mutate(serotype = str_replace(serotypeDetected, 'unknown', 'Unknown')) %>%
  # drop duplicate stereotype column
  select(-serotypeDetected) %>%
  # filter out unknown county rows
  filter(county != 'Unknown County of Exposure')

# pivot wider
dengue.w <- clean_dengue %>%
  # group by year, month, county
  group_by(year,month,county,caseType) %>%
  # summarize
  summarize(cases = sum(numberOfCases)) %>%
  # get names from case type
  pivot_wider(names_from = caseType, 
              # get values from number of cases
              values_from = cases,
              # set fill value to 0 (for NA values)
              values_fill = 0)
Variable Type Description
year double The year when the dengue cases were recorded.
month character The month when the dengue cases were recorded (e.g., ‘May’).
county character County in Florida where the dengue cases were reported
travel double The number of dengue cases associated with travel (imported cases)
local double The number of locally transmitted dengue cases

Table 2. Dengue data set structure and data types

# summarize dengue data set by county
summary.county <- dengue.w %>%
  # group by county
  group_by(county) %>%
  # sum local cases
  summarize(sumLocal = sum(local),
            # sum travel cases
            sumTravel = sum(travel),
            # sum all cases
            sumAll = sumLocal+sumTravel) %>%
  # arrange by descending order
  arrange(-sumAll)

# create clean, formatted table with top 10 counties
summary.county %>%
  # get top 10 rows for all cases
  top_n(10, sumAll) %>%
  # arrange by desc order
  arrange(-sumAll) %>%
  # create kable table
  kable(col.names = c("County","Local","Travel",'All'), # set column names
        # set table caption
        caption = "Top 10 Florida Counties with Highest Dengue Cases (2009 to 2024") %>%
  # set kable styling 
  kable_styling(full_width = TRUE,   # full width table 
                # font size
                font_size=14)
Top 10 Florida Counties with Highest Dengue Cases (2009 to 2024
County Local Travel All
Miami-Dade 321 2026 2347
Broward 11 373 384
Hillsborough 6 239 245
Palm Beach 6 219 225
Monroe 164 22 186
Orange 2 173 175
Lee 0 102 102
Osceola 1 64 65
Collier 1 49 50
Duval 0 48 48

Table 3. Florida counties with highest reported dengue cases

Between 2009 and 2024, a total of 4,322 dengue cases were reported in forty-nine of Florida’s sixty-seven counties, with 3,751 travel-related cases and 571 as a result of local transmission. The top ten counties with the highest dengue incidence are displayed in Table 3. Although travel-related cases were more abundant and widespread, local transmission was limited to a small subset of counties, primarily Miami-Dade and Monroe counties in South Florida. Miami-Dade County reported the greatest number of both dengue cases, far eclipsing all other counties with 2,347 reported cases.

While the absolute values indicate where dengue cases were most reported by county, to accurately compare case abundance between counties, each county’s population are are used for determining the incidence rate. The dengue incidence rate per 100,000 residents are computed for each county for both dengue case types. The tidycensus library is utilized to acquire county-level population data for each Florida county. The county names are cleaned and standardized in order to match and join with the dengue case data frame. The U.S. Census Bureau’s 5-Year ACS Community Survey population estimates are joined with the dengue case summary in order to calculate county-level 25-year incidence rates for both travel-related and locally acquired dengue cases from 2009 to 2024. Incidence rates are expressed per 100,000 people and were calculated using the following formula:

\[ \text{Incidence Rate} = \left( \frac{\text{Number of Cases}}{\text{Population}} \right) \times 100,000 \] The incidence rate represent how many residents in each county, over 25 years, would acquire dengue either from travel or local transmission per 100,000 people. Table 4 presents the top ten counties which exhibited the greatest overall dengue incidence over the 25-year period. Monroe County, which contains the Florida Keys and Everglades National exhibited the highest combined incidence rate at 227.27 per 100,000 residents, with the majority of cases (88%) a result of local transmission (200.39 per 100,000). These results are consistent with reported historical outbreaks of dengue which have occurred in the Florida Keys, especially from 2009 to 2011 (Radke et al., 2012). Miami-Dade has the second largest incidence rate at 87.40 per 100,000, however with a majority of cases (86%) imported as a result of travel. Although less populous, Hardee County possessed a local incidence rate of 78.1 per 100,000 with no travel-related cases, suggesting significant local transmission occurring despite it’s small county population size. Many other counties, such as Hendry, Broward, Martin, Hillsborough, Osceola, Palm Beach, and Collier exhibited lower incidence rates, with the vast majority of cases imported with no significant local transmission. Figure 2 illustrates the spatial distribution by counter for total dengue incidence as a result of travel and local transmission. The highest incidence for dengue are observed in south Florida, primarily Miami-Dade, Monroe counties.

# get population data 
fl_pop <- get_acs(
  geography = "county",      # county-level
  variables = "B01003_001",  # total population
  state = "FL",              # Florida
  year = 2023,               # 2023 estimates
  survey = "acs5",           # ACS 5-year survey
  geometry = TRUE) %>%       # get geometry
  # replace County, Florida suffix
  mutate(county = str_remove(NAME, ' County, Florida'),
         # replace 'DeSoto' with 'Desoto' for join
         county = str_replace(county, 'DeSoto','Desoto')) %>%
  # drop redundant NAME column
  select(-NAME)

# join dengue data set to fl_pop
dengue.pop <- summary.county %>%
  # perform right join (to retain all counties)
  right_join(fl_pop) %>%
  # fill NA values for counties with no dengue case information
  mutate(across(everything(), ~ replace_na(., 0))) %>%
  # set geometry
  st_set_geometry('geometry') %>%
  # set coordinate reference system (otherwise will get warning on leaflet map)
  st_transform(crs = 4326) %>%
  # calculate incidence rate per 100,000 people
  mutate(inc_travel = (sumTravel/estimate) * 100000,
         inc_local = (sumLocal/estimate) * 100000,
         inc_all = (sumAll/estimate) * 100000) %>%
  # arrange by incidence
  arrange(-inc_all)

# create clean, formatted table with top 10 counties by incidence rate
dengue.pop %>%
  # drop geometry
  st_drop_geometry %>%
  # subset columns
  select(county,inc_local,inc_travel,inc_all) %>%
  # round incidence rates
  mutate(across(where(is.numeric), ~ round(.x,2))) %>%
  # get top 10 rows for all cases
  top_n(10, inc_all) %>%
  # arrange by desc order
  arrange(-inc_all) %>%
  # create kable table
  kable(col.names = c("County","Local","Travel",'All'), # set column names
        # set table caption
        caption = "Top 10 FL County Incidence Rates per 100,000 (2009 to 2024)") %>%
  # set kable styling 
  kable_styling(full_width = TRUE,   # full width table 
                # font size
                font_size=14)
Top 10 FL County Incidence Rates per 100,000 (2009 to 2024)
County Local Travel All
Monroe 200.39 26.88 227.27
Miami-Dade 11.95 75.45 87.40
Hardee 78.41 0.00 78.41
Broward 0.57 19.17 19.73
Hendry 0.00 19.61 19.61
Martin 15.58 3.12 18.70
Hillsborough 0.40 16.04 16.45
Osceola 0.25 15.73 15.97
Palm Beach 0.40 14.53 14.93
Collier 0.26 12.64 12.90

Table 4. *Florida counties with highest dengue incidence rates per 100,000**

# create color palette  
paletteDengue <- colorNumeric(palette = "Reds",  # reds
                    # set possible values
                    domain = dengue.pop$inc_all[dengue.pop$inc_all > 0],
                    na.color = "gray")
# map incidence rates
dengue.pop %>%
  # round incidence rates
  mutate(across(where(is.numeric), ~ round(.x,2)),
         # set fill color to distinguish 0's from rest of incidence rates
         fillcol = ifelse(inc_all == 0, "gray", paletteDengue(inc_all))) %>%
  # create leaflet map for incidence rate (all)
  leaflet() %>%
  # add basemap (dark matter)
  addProviderTiles(providers$CartoDB.DarkMatter) %>%
  # add pop.dengue data
  addPolygons(
    # set fill color to red palette
    fillColor = ~fillcol,
    # set border color
    color = "black",
    # set line thickness
    weight = .75,
    # set opacity
    fillOpacity = .75,
    # set highlight options (when mouse hovers)
    highlightOptions = highlightOptions(
      # set outline weight
      weight = 4,
      # set color
      color = "white",
      # bring county to front (looks slightly better)
      bringToFront = TRUE
    ),
    # create tooltip label with County: Incidence Rate
    label = ~paste0(county, ": ", inc_all)
  ) %>%
  # add legend
  addLegend(
    # set color palette
    pal = paletteDengue,
    # set values to be incidence rate (greater than 0)
    values = dengue.pop$inc_all[dengue.pop$inc_all > 0],
    # set legend title
    title = "Dengue Incidence per 100,000 (All)",
    # set legend position
    position = "bottomleft",
    # format legend values with %
    labFormat = labelFormat(suffix = " cases")
  )

Figure 2. Local dengue incidence per 100,000 by county (2009-2024)

Analysis

Global Spatial Autocorrelation

To investigate the spatial distribution of dengue incidence in Florida, spatial autocorrelation was assessed using Moran’s l and Local Indicators of Spatial Association (LISA) which were computed using the spdep package (Bivand & Wong, 2018). Given the potential for imported cases to be a contributing factor for local transmission, understanding whether incidences rates exhibit spatial clustering or randomness is important for facilitating state-wide public health interventions and collaboration between neighboring county vector control programs. An assessment of spatial autocorrelation will provide insights into whether incidence rates are geographically proximate between counties or if dengue transmission dynamics are more dispersed. First, global spatial autocorrelation was examined using Moran’s I to determine if total dengue incidence exhibited spatial clustering, with statistical significance evaluated through a Monte Carlo simulation with 999 permutations. To construct a spatial weight matrix, neighboring counties were defined along contiguous boundaries. With the neighborhood list and spatial weight matrix defined, Moran’s l scatter plot is constructed by calculating the lagged values for each county’s travel case incidence rates. The spatially lagged values are the weighted average of the neighboring values for each county. The scatter plot is created by regressing the spatially lagged travel incidence rate on the actual travel case incidence rate. The slope of this regression is Moran’s I coefficient, which quantifies the degree of spatial autocorrelation present in the dengue data set. Moran’s I coefficient was also calculated without explicitly computing the lagged values by fitting a regression model using the moran() function, where the neighbor list and spatial matrix are provided as parameters. After computing global spatial correlation for travel-related dengue incidence using Moran’s I, the resulting coefficient was 0.429, indicating moderately positive spatial autocorrelation. Counties which exhibited high or low travel-related dengue incidence from 2009 to 2024 tended to cluster geographically.

A Monte Carlo permutation test (n =999) was performed to evaluate the statistical significance of the spatial autocorrelation. Monte Carlo simulation is a technique which assesses statistical significance by simulating possible outcomes under the null hypothesis. Specifically, the travel-related dengue incidence rates were randomly re-assigned across the counties 999 times and Moran’s l coefficient is re-calculated, generating a distribution of Moran’s 1 coefficients that would be expected if the null hypothesis were true (travel-related dengue incidence rates are randomly distributed across Florida counties). The computed Moran’s I coefficient was at the far end of the Monte Carlo simulated distribution (rank = 1000) as illustrated in Figure 4, with a pseudo p-value of 0.001. This result strongly suggest that we should reject the null hypothesis that travel-related dengue incidence are randomly distributed across Florida counties. Travel-related dengue incidence exhibits statistically significant spatial clustering by county.

The results of Moran’s I and Monte Carlo simulation indicate strong evidence that travel-related dengue incidence in Florida are not randomly dispersed but exhibit statistically significant spatial clustering. The computed Moran’s I coefficient of 0.429 suggests a moderately positive spatial autocorrelation, indicating that neighboring counties possess similar travel-associated dengue incidence rates, potentially due to shared patterns of international travel or population dynamics. This finding has several important implications for public health and vector control agencies, namely that clustering suggests that some counties may act as the primary ports of entry for imported dengue cases, possibly due to the presence of international travel hubs like airports, which could contribute to local transmission.

# create neighboring polygons
# queen = False: at least one edge be shared, True would mean at least one vertex
# construct neighbors list from polygon list
nb <- poly2nb(dengue.pop, queen=FALSE)

# assign weights to each neighboring polygon
# weight is calculated by 1/(# of neighbors) aka a proportion, set "binary" using B 
lw <- nb2listw(nb, style="W", zero.policy=TRUE) # zero policy true allows for lists of non-neighbors

# generate Moran's l scatter plot
# extract lagged values from lw object
dengue.pop$lag_local <- lag.listw(lw, dengue.pop$inc_local)
dengue.pop$lag_travel <- lag.listw(lw, dengue.pop$inc_travel)

# create a regression model for lagged travel vs travel incidence
lm_travel <- lm(lag_travel ~ inc_travel, dengue.pop)

# create a regression model for lagged local vs local incidence
lm_local <- lm(lag_local ~ inc_local, dengue.pop)

# plot Moran's 1 scatter plot for travel incidence rate
plot(lag_travel ~ inc_travel, 
     dengue.pop, 
     main = "Moran's l Scatter Plot (Travel Dengue Incidence)",
     xlab = "Incidence Rate (Travel)",
     ylab = "Lagged Incidence Rate (Travel)",
     # symbol style
     pch=21, 
     asp=1, 
     las=1, 
     # set color
     col = "grey40", 
     bg="grey80")
# add the regression line from model lm_travel
# the slope of the regression model is the moran's l coefficient
abline(lm_travel, col="blue") 
abline(v = mean(dengue.pop$inc_travel), lty=3, col = "grey80")
abline(h = mean(dengue.pop$inc_travel), lty=3, col = "grey80")

Figure 3. Moran’s I Scatter Plot for Travel Dengue Incidence

# compute Moran's l coefficient (without needing to compute the lagged values and fitting a regression model)
moran(dengue.pop$inc_travel,
      listw = lw, # weights list
      n = length(nb), # total number of features having at least one neighbor
      S0 = Szero(lw)) # sum of all weights with non-zero neighbors (sZero function does this)

# assessing statistical significance (the slope in the scatter plot)
# 1. Randomly permute the incidence rates across all polygons (counties)
# 2. Compute a Moran's l coefficient for each permuted set of values
#     - this gives the distribution of moran's l values expected under the null hypothesis
#     - H0: The incidence rates are randomly distributed across all counties
MC <- moran.mc(dengue.pop$inc_travel,  # travel dengue incidence rate
               lw,                     # spatial weight matrix
               nsim=999)               # # of permutations

# view results (including pseudo p-value)
#MC
# plot the distribution (this is a density plot instead of a histogram)
plot(MC, 
     main = "Monte Carlo Simulation",
     xlab = "Dengue Travel Incidence per 100,000",
     ylab = "Density",
     las=1)

Figure 4. Monte Carlo Simulation Density Plot for Travel Dengue Incidence

Local Moran’s l

To identify localized spatial clustering and outliers, Local Indicators of Spatial Association (LISA) or Local Moran’s l was calculated using a Monte Carlo simulation (n = 9999). This statistic measures local spatial autocorrelation, identifying statistically significant county clusters and spatial outliers, where counties were classified as either High-High (hot-spots), Low-Low (cold-spots), and High-Low/Low-High outliers from their neighboring counties. High-High clusters represent counties surrounded by similarly high dengue incidence counties, whereas Low-Low indicates clusters of low incidence counties surrounded by other low incidence counties. Spatial outliers, High-Low and Low-High, indicate counties where local incidence differs significantly from surrounding counties. Pseudo p-values were assigned to each county to assess the statistical significance of spatial clustering. The results are visualized using the tmap library, highlighting counties with significant local clustering (p-value < 0.05). The results indicate that Collier, Monroe, Broward, and Miami-Dade counties formed a statistically signifcant High-High cluster, indicating a persistent dengue hotspot with elevated travel-related incidence. Miami-Dade and Broward counties operate as major international travel hubs with a high volume of travelers, especially from dengue-endemic areas. Monroe County, which includes the Florida Keys, has as history of Dengue outbreaks and periodic local transmission events in the past decade. Collier County, although less prominent in international travel volume than Miami-Dade or Broward counties, exhibited consistently heightened travel incidence of dengue, potentially due to its geographic proximity to other hotspot counties, such as Miami-Dade.

# calculate Local Moran's l
MCi <- localmoran_perm(dengue.pop$inc_travel,
                       lw,
                       nsim = 9999)
# convert to data frame
MCi.df <- as.data.frame(MCi)
# add pseudo p-values to dengue.pop (one-sides test)
dengue.pop$p <- MCi.df$'Pr(folded) Sim'

#### Map significant clusters ####
# identify significant clusters using hotspot function (cutoff will be 0.05)
dengue.pop$Ii <- hotspot(MCi,
                         Prname = "Pr(folded) Sim",
                         cutoff = 0.05,
                         p.adjust = "none")
#Ii factor re-leveled
dengue.pop$Ii <- factor(dengue.pop$Ii,
                        levels = c("High-High",
                                   "Low-Low",
                                   "Low-High",
                                   "High-Low",
                                   ">0.05"))
# replace NA with >0.05
dengue.pop$Ii[is.na(dengue.pop$Ii)] <- ">0.05"

# create color pallette for HH, LL, HL, and LH clusters
pal1 <- c( "#FF0000", "#0000FF", "#a7adf9", "#f4ada8","#ededed") # create palette

# map local Moran's I results
tm_shape(dengue.pop) + tm_polygons(style="cat", 
                                   # border color
                                   border.col = "grey", 
                                   # line width
                                   lwd = .75,
                                   # value col
                                   col = "Ii", 
                                   # palette
                                   palette=pal1,
                                   # legend title
                                   title = "Cluster Type") +
  # legend properties
  tm_legend(outside = TRUE, 
            text.size = .75) +
  # layout properties
  tm_layout(frame = FALSE)

Linear Regression

Given that travel-related cases can serve as the primary source for local transmission in local Ae. aegypti, the statistical relationship between incidence rates for travel-related cases and locally acquired cases was investigated using a simple linear regression model. Assessing the correlation between imported and local dengue cases is important for assessing the relative risk posed by travel-related dengue incidence for local transmission. This approach aims to quantify the influence of imported dengue cases on the incidence for local transmission. If travel-related cases are statistically significant and have explanatory power for the local incidence, greater efforts should be made for the detection and prevention of imported dengue. The model was specified by the following formula:

\[ \hat{Y} = \beta_0 + \beta_1 X + \epsilon \]

Where the response variable is the incidence rate per 100,000 residents of locally acquired dengue cases and the predictor variable (β1) is the incidence rate for travel-related cases. The model was fit using the ordinary least squares (OLS) for the incidence rate of travel-related dengue cases inc_travel on the incidence rate of locally-acquired dengue inc_local.

# fit linear model of local dengue incidence and travel incidence
dengue.lm <- lm(inc_local ~ inc_travel, data = dengue.pop)

The model result with the estimated coefficients:

\[ \text{Local Incidence} = 0.862 + 0.652 \times \text{Travel Incidence} + \epsilon \]

where:

  • \({Y}\) represents the predicted local dengue incidence rate per 100,000 residents.

  • \(\beta_0 = 0.862\) (p = 0.8104) is the intercept, which is not statistically significant, suggesting that there is no baseline for local incidence in the absence of travel-related cases.

  • \(\beta_1 = 0.652\) (p = 0.0358) is the estimated effect of travel-related incidence on local incidence, indicating that for every additional travel-related dengue case per 100,000 residents, the expected local incidence increases by approximately 0.65 cases.

  • \(\epsilon\) represents the residual error for the model.

The model explains a small proportion of the variation in local incidence (\(R^2 = 0.066\), Adjusted \(R^2 = 0.052\)), suggesting that additional factors contribute to dengue transmission beyond travel-related cases. The residual standard error of 25.5 indicates considerable variability in local incidence unexplained by the model. The coefficient for travel-related cases (β1 = 0.652, p-value = 0.0358) was statistically significant, with a positive relationship where each additional travel-related dengue case is associated with an estimated 0.65 case increase for local incidence in each county. The regression intercept (β0 = 0.066, p-value = 0.8104) was not statistically significant, suggesting that there was no baseline incidence for locally transmitted dengue in Florida from 2009-2024

The model explained a small proportion (Adjusted R-Squared = 0.052) of local cases, indicating that travel-related cases alone are not a strong predictor for local dengue incidence. Furthermore, the residual standard error (25.5) suggests that there is unexplained variability for local dengue incidence, thus there are likely other important factors which contribute to the the spread of local transmission. Local dengue transmission is highly dependent upon the population of Ae. aegypti mosquitoes in a given area. Future modeling will be enhanced by integrating vector species population, including Ae, aegypti and Ae. albopictus present in each county and climatic factors, such as temperature, humidity, and precipitation.

# plot local dengue incidence rate
plot(dengue.pop$inc_local, 
     pch = 16, 
     col = "lightgrey",
     main ="",
     xlab = "Travel Dengue Incidence per 100,000",
     ylab = "Local Dengue Incidence per 100,000")
# add regression line f,rom linear model fit
abline(dengue.lm, col = "red")

Reflection

In conclusion, spatial analysis confirmed statistically significant clustering of travel-related dengue cases, with Moran’s I coefficient (0.429, p-value = 0.001) indicating moderately positive spatial autocorrelation between counties. Local Indicators of Spatial Association or Local Moran’s I further revealed county hot-spots for travel-related dengue incidence in southern Florida, namely in Monroe, Miami-Dade, Collier, and Broward counties, which have historically reported the highest number of dengue cases. This underscores the importance of coordinated dengue surveillance and regional collaboration between vector control programs for detecting and mitigating local transmission risks. However, the neighborhoods defined for assessing spatial autocorrelation using county boundary contiguity presents a limitation, as shared borders do not necessarily indicate epidemiological relevance for dengue transmission. For example, Monroe County includes extensive uninhabited areas, such as the Everglades National Park, which do not contribute to dengue risk due to the desired habitat of Ae. aegypti being urban areas. Future research should redefine and narrow the spatial weight matrices by incorporating urban land-use cover, transportation networks, or human mobility information to more accurately capture dengue transmission dynamics. Additionally, this analysis has demonstrated the statistical relationship between local dengue incidence and travel-related incidence in Florida from 2009 to 2024. While the regression model revealed a statistically significant positive relationship between the case types, the low explanatory power (Adjusted R-Square = 0.052) suggest that there may be additional environmental and social factors influences for local transmission of dengue in Florida which should be explored.

Despite these limitations, these preliminary findings underscore the role of travel-related dengue for local transmission in Florida and the importance of continued vector management practices and public health monitoring. Further integration of climatic and socioeconomic factors could also enhance predictive modeling for local dengue incidence toward improving public health preparedness and outbreak response.

References

Bivand, R. S., & Wong, D. W. S. (2018). Comparing implementations of global and local indicators of spatial association. Test (Madrid, Spain), 27(3), 716–748.

Ferreira-de-Lima, V. H., & Lima-Camara, T. N. (2018). Natural vertical transmission of dengue virus in Aedes aegypti and aedes albopictus: A systematic review. Parasites & Vectors, 11(1). https://doi.org/10.1186/s13071-018-2643-9

Radke, E. G., Gregory, C. J., Kintziger, K. W., Sauber-Schatz, E. K., Hunsperger, E. A., Gallagher, G. R., Barber, J. M., Biggerstaff, B. J., Stanek, D. R., Tomashek, K. M., & Blackmore, C. G. (2012). Dengue outbreak in Key West, Florida, USA, 2009. Emerging Infectious Diseases, 18(1), 135–137. https://doi.org/10.3201/eid1801.110130

Rey, J. R. (2014). Dengue in Florida (USA). Insects, 5(4), 991-1000. https://doi.org/10.3390/insects5040991 Stephenson, C., Coker, E., Wisely, S., Liang, S., Dinglasan, R. R., & Lednicky, J. A. (2022). Imported Dengue Case Numbers and Local Climatic Patterns Are Associated with Dengue Virus Transmission in Florida, USA. Insects, 13(2), 163. https://doi.org/10.3390/insects13020163

Walker K, Herman M (2025). tidycensus: Load US Census Boundary and Attribute Data as ‘tidyverse’ and ‘sf’-Ready Data Frames. R package version 1.7.1, https://walker-data.com/tidycensus/.

Wilder-Smith, A., Ooi, E.-E., Horstick, O., & Wills, B. (2019). Dengue. The Lancet, 393(10169), 350–363. https://doi.org/10.1016/s0140-6736(18)32560-1

Wilder-Smith, A., & Gubler, D. J. (2008). Geographic expansion of dengue: The impact of international travel. Medical Clinics of North America, 92(6), 1377–1390. https://doi.org/10.1016/j.mcna.2008.07.002