The Costs of Respiratory Illnesses Arising from Florida Gulf Coast Karenia brevis Blooms

Background Algal blooms of Karenia brevis, a harmful marine algae, occur almost annually off the west coast of Florida. At high concentrations, K. brevis blooms can cause harm through the release of potent toxins, known as brevetoxins, to the atmosphere. Epidemiologic studies suggest that aerosolized brevetoxins are linked to respiratory illnesses in humans. Objectives We hypothesized a relationship between K. brevis blooms and respiratory illness visits to hospital emergency departments (EDs) while controlling for environmental factors, disease, and tourism. We sought to use this relationship to estimate the costs of illness associated with aerosolized brevetoxins. Methods We developed a statistical exposure–response model to express hypotheses about the relationship between respiratory illnesses and bloom events. We estimated the model with data on ED visits, K. brevis cell densities, and measures of pollen, pollutants, respiratory disease, and intra-annual population changes. Results We found that lagged K. brevis cell counts, low air temperatures, influenza outbreaks, high pollen counts, and tourist visits helped explain the number of respiratory-specific ED diagnoses. The capitalized estimated marginal costs of illness for ED respiratory illnesses associated with K. brevis blooms in Sarasota County, Florida, alone ranged from $0.5 to $4 million, depending on bloom severity. Conclusions Blooms of K. brevis lead to significant economic impacts. The costs of illness of ED visits are a conservative estimate of the total economic impacts. It will become increasingly necessary to understand the scale of the economic losses associated with K. brevis blooms to make rational choices about appropriate mitigation.

Harmful algal blooms (HABs) of Karenia brevis, a marine dinoflagellate, occur almost annually in the Gulf of Mexico off the west coast of Florida (Heil and Steidinger 2009). K. brevis cells produce potent poly ether neuro toxins known as brevetoxins (Baden et al. 1995;Poli et al. 1986), which are released into the ocean when the cells are lysed by wind and waves (Pierce et al. 2001). At high concentrations, K. brevis blooms may cause mortality of fish, marine mammals, and sea birds. Humans who consume shellfish contaminated with brevetoxins are at high risk of developing neuro toxic shellfish poison ing (NSP) (Watkins et al. 2008). Further, there have been anecdotal reports of skin ail ments resulting from contact with brevetoxin contaminated water (Kirkpatrick et al. 2004).
HAB cell count generally correlates with brevetoxin concentration, and residual breve toxins in seawater have been observed after blooms have diminished (Pierce RH, personal communication). Bubbles from breaking waves transport brevetoxins to the sea sur face, where they may be released to the air as jet drops when the bubbles burst (Blanchard 1989;Pierce et al. 1990Pierce et al. , 2003. In the air, brevetoxins become incorporated into marine aerosols, which are built around charged salt particles. Aerosol transport is highly influenced by wind speed and direction, and aerosolized brevetoxins can travel as much as 1 mile inland (Fleming et al. 2005;Kirkpatrick et al. 2008). Woodcock (1948) reported the first accounts of respiratory irritation in humans during severe red tide (HAB) conditions on the Florida Gulf Coast. Subsequent studies have linked inhalation of aerosolized breve toxins with adverse health effects, including rhinorrhea, non productive cough, and severe broncho constriction (Asai et al. 1982;Backer et al. 2003;Cheng et al. 2005;Fleming et al. 2005Fleming et al. , 2007Kirkpatrick et al. 2004Kirkpatrick et al. , 2001Music et al. 1973). Epidemiologic studies, animal experiments, and anecdotal reports now suggest that aerosolized brevetoxins are linked to both upper and lower respiratory ill nesses in humans (Backer et al. 2003;Fleming et al. 2005Fleming et al. , 2007Kirkpatrick et al. 2004). In addition to significantly increased respiratory symptoms, asthmatic individuals have shown small but statistically significant changes in lung function immediately after 1hr visits to the beach during Florida blooms (Backer and McGillicuddy 2006;Fleming et al. 2005Fleming et al. , 2007. After only 1 hr of exposure to aero sols at the beach, respiratory complaints may last up to 5 days in asthmatics (Kirkpatrick et al. 2008). Finally, research using data for respiratory visits to the emergency department (ED) demonstrated a significantly increased risk of visits for pneumonia, bronchitis, and asthma during a Florida red tide period com pared with a similar period without red tide (Kirkpatrick et al. 2006). This risk is particu larly relevant for coastal residents.
In the present study we assessed the relation ship between K. brevis blooms and respiratory illnessrelated visits to hospital EDs while con trolling for environmental factors and disease that also may function as significant risk factors for respiratory ailments. Further, we developed estimates of the costs of illness resulting from medical treatments for acute brevetoxinrelated respiratory illnesses. This work is an important step toward understanding one component of the economic effects of K. brevis blooms. It may help guide the selection and implementation of management actions to mitigate public health effects and illness costs for brevetoxins and possi bly other HABs. Further, it may provide insight to health care administrators and providers as to human resource and medical supply needs as a result of K. brevis blooms.
Background: Algal blooms of Karenia brevis, a harmful marine algae, occur almost annually off the west coast of Florida. At high concentrations, K. brevis blooms can cause harm through the release of potent toxins, known as brevetoxins, to the atmosphere. Epidemiologic studies suggest that aerosolized brevetoxins are linked to respiratory illnesses in humans. oBjectives: We hypothesized a relationship between K. brevis blooms and respiratory illness visits to hospital emergency departments (EDs) while controlling for environmental factors, disease, and tourism. We sought to use this relationship to estimate the costs of illness associated with aerosolized brevetoxins. Methods: We developed a statistical exposure-response model to express hypotheses about the relationship between respiratory illnesses and bloom events. We estimated the model with data on ED visits, K. brevis cell densities, and measures of pollen, pollutants, respiratory disease, and intraannual population changes. results: We found that lagged K. brevis cell counts, low air temperatures, influenza outbreaks, high pollen counts, and tourist visits helped explain the number of respiratory-specific ED diagnoses. The capitalized estimated marginal costs of illness for ED respiratory illnesses associated with K. brevis blooms in Sarasota County, Florida, alone ranged from $0.5 to $4 million, depending on bloom severity. conclusions: Blooms of K. brevis lead to significant economic impacts. The costs of illness of ED visits are a conservative estimate of the total economic impacts. It will become increasingly necessary to understand the scale of the economic losses associated with K. brevis blooms to make rational choices about appropriate mitigation.

Methods.
Aerosolized brevetoxins may result in respiratory illnesses of varying severities. Many individuals with respiratory symptoms caused by brevetoxins may respond initially by purchasing overthecounter medicines or taking time off from work or leisure. At pres ent, data on visits to physicians and out patient facilities are unavailable for unreported dis ease. For lowseverity illnesses, medical costs are uncertain but are expected to be minor and unlikely to involve significant productiv ity losses. In this study, we focused on visits to EDs for which we have credible data. There also may be cases of further hospitalizations or mortalities, but we expect that the number of patients experiencing these severities is small. In particular, we are unaware of any deaths caused by aerosolized brevetoxins.
We focused on the relationship between HAB cell counts and respiratory hospital ED visits from 2001 to 2006 in Sarasota, Florida. Sarasota experiences nearshore K. brevis blooms on an almost annual basis; in some years, there are multiple or even continuous blooms. These events have been documented through monitoring efforts by researchers at Mote Marine Laboratory (MML). We focused on Sarasota Memorial Hospital (SMH), the hospital located closest to the coastline of the four hospitals in Sarasota County. SMH is the largest acutecare facility in Sarasota County, serving 63.3% of the county's population (Kirkpatrick et al. 2006).
In our analysis we concentrated on two main types of respiratory conditions: all respi ratory diseases taken together and a combi nation of upper airway disease (UAD) and chronic/acute bronchitis. Previous work by Kirkpatrick et al. (2006) demonstrated that UAD and bronchitis (as well as pneumonia and asthma) are particularly likely to be exac erbated by aerosolized brevetoxins. Adverse respiratory conditions may be attributable to HABs or a combination of HABs with other ambient factors.
We developed an exposure-response model to express hypotheses about the relationship between respiratory illnesses, HAB events, and other potential explanatory variables. The model is formulated as follows: where the subscript t indexes the relevant week; ED is a measure of the number of respi ratory illness ED visits at SMH; H is a lagged measure of in situ K. brevis cell counts near the Sarasota coast; W is a vector of environmental and weather conditions; D is a meas ure of regional respiratory disease outbreaks; and T is a measure of tourist visits to Sarasota County. Equation 1 is estimated using auto regressive error models. Data. We investigated a large number of environmental, disease, and tourism vari ables (Polansky et al. 2007). We present two parsimonious models here. Variables inves tigated but not included in the final model specifications are mentioned briefly below. Descriptive statistics for model variables appear in Table 1.
The total number of daily SMH ED vis its for respiratory diagnoses were compiled from October 2001 through September 2006. Access to anonymous medical data was pro vided by SMH after its Institutional Review Board's approval of the use of data for our study. Using codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD9CM), these diagnoses were categorized as respiratory ill nesses (e.g., pneumonia, bronchitis, asthma, UAD) or all other primary diagnoses (NCHS 2009). We calculated for each week the daily average number of ED visits for two catego ries of respiratory diagnoses (all four diseases and a combined UAD/bronchitis).
In situ K. brevis cell counts served as a proxy for aerosolized brevetoxin concentra tions along the coast. Cheng et al. (2005) developed an empirical relationship between brevetoxin concentration in the ocean and in the atmosphere as a function of wind speed and wind direction. Aerosolized breve toxins have been measured as much as 1 mile inland from the coast (Kirkpatrick et al. 2008); thus, the affected population may include those who are not right on the coast or do not visit the beach or shoreline. Measurements of breve toxin concentrations in the ocean are time consuming and costly to make. Consequently, longterm brevetoxin concentration data sets are unavailable. We assumed that because higher HAB cell counts would yield increased breve toxin concentrations in water, and sub sequently in air, they were a reasonable proxy for aerosolized toxin.
K. brevis cell counts were sampled at two Sarasota Bay locations [New Pass (27.19°N,82.34°W) and the MML Bay Dock (27.33°N, 82.58°W)]. Water samples were analyzed weekly during non bloom conditions and daily during blooms. We averaged the data from both sampling stations to obtain a measure of in situ cell count. To create a consistent data set, we compiled for each week the daily average of K. brevis cell counts across both sampling sta tions. Because of the large range of cell counts, we transformed the data for scale purposes. The HAB measure we employed in the model is the square root of onethousandth of the weekly average of daily average K. brevis cell counts. Pierce et al. (2001) determined that toxins are released into the water as cells rupture, thereby increasing the amount of extra cellular toxins as a bloom progresses. After release into marine waters, the toxin subsequently may become aerosolized and transported by winds onshore, where humans may be exposed . We expected to find a lag between the detection of cells in offshore waters and onshore aerosolized brevetoxin exposures because of the time required for intra and extra cellular brevetoxin metabo lism and transformation, as well as subsequent aeolian transport. A latency period associated with the observed health effects of K. brevis blooms has not been established, however.
In the model, we tested for a 1week lag for the in situ HAB cell count. We were unable to find correlations between ED visits and measures of wind speed, wind direction, or wind velocity. Information on air temperature was col lected from a monitor at the MML New Pass Weather Station, located at 27. 19°N, 82.34°W (MML 2009). We expected lower temperatures would increase the susceptibility of individuals to respiratory illnesses, possibly with a lagged effect. We compiled the data as a weekly average temperature (°C). We were unable to find correlations between ED visits and a measure of relative humidity.
Pollen is a common allergen that can produce histaminergic responses in sensitive individuals, exacerbating existing respiratory conditions or initiating new ones (Andersson et al. 2003;Atkinson and Strachan 2004;Burge and Rogers 2000). Daily pollen counts (pollen grains per cubic meter of air) were pro vided by the National Allergy Bureau (Jelks M, personal communication). Counts were obtained using a Burkard 7day pollen collec tor at the regional sampling station, located approxi mately 160 miles northeast of Sarasota at the University of Florida in Gainesville. Counts are given as the square root of pollen grains per cubic meter of air × 10 -2 . We also investigated mold counts and local pollution measures (ozone, two sizes of airborne particu lates, forest fires, and acres of forest burned). None of these variables were found to correlate with ED visits.
Weekly influenza virus outbreak data for the South Atlantic Region were compiled by the U.S. Centers for Disease Control and Prevention (CDC). These data are a measure of the percentage of specimens testing posi tive for influenza within a particular week, as measured by the World Health Organization and the National Respiratory and Enteric Virus Surveillance System laboratories. Data were provided from October through May (weeks 40 through 52 or 53 and weeks 1-20), as this time frame is known to correspond to epidemic periods (Simonsen et al. 1997). The CDC does not track influenza during minimal outbreak periods (weeks 21-39). We assumed that these gaps in the data cor respond to periods of little or no influenza cases, and we replaced them with zeroes.
We obtained data on Sarasota County monthly hotel/motel occupancy rates for October 2001-September 2006 and the number of units by lodging type (e.g., hotel/ motel, campsite, mobile home, apartment, condominium, house) from the Sarasota Convention and Visitors Bureau (Haley V, personal communication). We assumed that the percent occupancy of all lodging types approximates that of hotels/motels. We derived a total number of occupants by lodg ing type by making the assumptions of two people per hotel/motel, condominium, or apartment unit and four people per camp site, mobile home, or house. We obtained a monthly estimate of the temporary resi dent population by summing the number of occupants in all units. The tourism variable is measured as persons × 10 -4 . Note that the weekly estimates used in the models remained constant for the weeks within each month. Table 2 presents YuleWalker estimates of two different auto regressive error models corrected for third order auto correlation. Each model considers a different dependent variable. In model I, the dependent variable is the natural logarithm of SMH ED visits for all respiratory ailments (lnED tot ). In model II, the dependent variable is the natural logarithm of SMH ED visits for the combination of UAD/bronchitis (lnED ub ). Note that the latter measure is included in the former, which also includes ED visits due to pneumonia and asthma.

Exposure-response model.
All parameter estimates have the expected signs and are significant at the 10% level or better in both models. The model results show that ED visits for respiratory ailments are correlated with lagged K. brevis cell counts, regional pollen counts, regional influenza outbreaks, and Sarasota County tourist vis its, and inversely correlated with temperature and temperature lagged 1 week. Both models explained about threefourths of the varia tion in their respective dependent variables. Figure 1A illustrates how well the model pre dicts ED visits compared with the actual data. The marginal effects (the incremental change in ED visits with an incremental change in the variable of interest) and the elasticities (percentage change in ED visits with a per centage change in the variable of interest) are shown in Table 3. Both measures reveal the importance of temperature and tourist num bers as predictors of ED visits.
The inverse relationship between tempera ture and respiratory ED visits may result from the interaction of other agents responsible for respiratory ailments (e.g., bacteria and viruses). Like influenza, these triggers tend to affect more individuals when low ambient temperatures prevail (Simonsen et al. 1997).
Importantly, our measure of aerosolized brevetoxins is a significant predictor of ED visits for respiratory ailments. Figure 1B depicts the number of SMH UAD/bronchitis ED visits due to K. brevis blooms predicted by model II. Because of variable transformations, the parameter value cannot be interpreted per se as the marginal effect of changes in cell counts on ED visits. The marginal effect is itself a function of the cell count, and there fore its value will depend on the state of a K. brevis bloom.
In Table 4, we present model predictions of yearly ED visits for three different bloom levels: low, medium, and high. The 4year period from October 2001 to December 2004   Referring to the results of model II in Table 4, the model's predictions yield annual estimates of the likely number of ED visits for low (39 visits), mean (76 visits), and high (218 visits) bloom levels in Sarasota County. We developed our estimates using data for SMH, which treats 63.3% of the patients in Sarasota County. We have adjusted the esti mates for Sarasota County accordingly. These estimates are made holding all other model variables constant at their respective means.
Cost-of-illness estimates. A measure of the economic impact of respiratory illnesses caused by aerosolized brevetoxins is the sum of the costs of medical services and lost productivities during the illness period. When considering the costs of a specific illness, especially one that does not involve unusual technologies, treatments, or medical skills and one that does not comprise a large portion of the total number of visits, esti mates of the marginal costs of treatment are appropriate. In the case of respiratory ailments that are the consequence of exposure to aero solized breve toxins, the additional admissions per week (on average only 1.5% of the total) are unlikely to lead to the need to expand capacity in respiratory ED services at SMH.
In many cases, hospitals must charge patients more than average or marginal costs for specific medical services. The excess of charges over costs is generally used to cover the burden to a hospital of the provision of medical services that go unpaid because of the lack of insurance coverage or the inability or unwillingness of patients to pay for treatment. Hospitals often crosssubsidize medical services within their facilities using the net revenues of services with a high inelasticity of demand, such as emergency services, to subsidize those important services that may have a more elastic demand or a lower net revenue margin. Cost ofillness estimates based on average charges likely exaggerate the true costs of ED visits. In the present study, we applied an estimate of the ratio of marginal to average charges for general ED visits to approximate the marginal costs of ED visits for respiratory ailments due to aerosolized brevetoxins. Williams (1996) studied the costs of ED visits in six Michigan community hospitals and developed estimates of both average and marginal costs of ED visits. The author esti mated the marginal costs of ED visits for three different illness severities: $24 for non urgent treatments, $67 for semi urgent treatments, and $148 for urgent treatments. Marginal costs averaged only about 23% of average costs.
The Florida Agency for Health Care Administration (AHCA) compiles data on charges levied by Florida hospitals for indi vidual and total medical services for primary diagnosed illnesses. Most of the charges for UAD/bronchitis illnesses involve emergency treatments, radiology services, and laboratory tests. Focusing on the relevant ICD9CM codes for UAD/bronchitis illnesses, we calcu lated the 25th and 75th percentile estimates for the total ED visit charges and found that they ranged from $252 to $1,045 per visit. We applied the Williams (1996) ratio of mar ginal to average costs of 23% to calculate a range of marginal charges from the AHCA data of $58-$240 as our preferred range of marginal cost estimates.
Where illnesses are severe enough to result in lost time at work or lost leisure opportu nities, an estimate of the cost of illness also should include an estimate of lost productiv ity. We assumed that individuals valued their leisure time at the margin at approximately the same rate as their employment wages. We used annual median personal income in 2006 as a measure of lost productivity (U.S. Census Bureau 2008). This measure was converted into daily median personal income.
There are few extant estimates of the total number of days of lost productivity to a patient as a consequence of a respiratory illness treated at an ED. The effects of severe asth matic attacks could incapacitate a patient for weeks or even months [National Heart, Lung, and Blood Institute (NHLBI) 2007]. Even so, Rodrigo et al. (2004) relied on a British study (Hoskins et al. 2000) that estimated that indi rect costs (lost productivities) accounted for only 10% of the costs of asthma illnesses. Here we refer to the U.S. and British guidelines for the treatment of moderate asthma illnesses (British Thoracic Society 2008;NHLBI 2007), potentially requiring an ED visit. These guide lines suggest, on average, 1 day for treatment and up to 2 days for recuperation, for a total of 3 days of lost productivity per illness.
The population that is adversely affected by aerosolized brevetoxins is composed of Florida residents as well as tourists or longer term visitors (snowbirds) from other states and other countries. The AHCA compiles data on the state of residence of patients visiting the SMH ED for respiratory ailments. We used the residential distribution of these patients to weight annual median personal income by state to determine a weighted average median personal income of $38,589. Three days of lost productivity therefore has an estimated value of approximately $335 (2008 dollars).
In Table 5, we calculated ranges of costs of illness for Sarasota County and for the Florida Gulf Coast. We used the predicted number of ED visits from model II in combination with the range of marginal costs of ED respira tory visits for UAD/bronchitis illnesses from the AHCA data. We estimated annual costs, 5year costs, and capitalized costs using a dis count rate of 3%. We chose this discount rate as reflective of the historical rate of return on a safe asset, such as mediumterm government bonds. The latter two estimates imply that the low, medium, or highbloom levels persist during the period of interest. We found that the capitalized costs of illness can range from $0.5 to $4 million in Sarasota County alone. We expect that the costs of illness should be much greater for the entire Florida Gulf Coast.

Discussion
We investigated whether HABs can explain respiratory ED visits while controlling for relevant environmental factors that also may explain these visits. We found that K. brevis cell counts (lagged 1 week), low temperatures, a high incidence of influenza outbreaks, high pollen levels, and large numbers of tourist visits (including the snowbirds) all explained the number of respiratoryspecific ED diagno ses. An individual's ED visit also could be the result of multi day exposure to numerous, inter acting conditions (e.g., aerosolized brevetoxins combined with environmental particulates, disease, local weather conditions) rather than the influence of one factor over a short time period. Statistical tests of our data failed to identify such inter actions, however, suggesting an avenue for future research.  Analysis of our data revealed that ED visits for UAD and bronchitis are explained by the K. brevis cell count, but ED visits for pneumo nia and asthma are not. However, Asai et al. (1982) observed that 80% of the 15 asthmatic patients exposed to red tide aerosols at the beach complained of asthma attacks, and, more recently, Fleming et al. (2005Fleming et al. ( , 2007 found statistically significant changes in respiratory function in asthmatic residents of Sarasota after spending only 1 hr at a Florida beach dur ing a red tide. One explanation for our find ing is that asthmatics may be conditioned to recognize their symptoms, responding with pre existing prescriptions, overthecounter medications, or outpatient doctor visits, instead of visits to the ED.
Estimates of the costs of illness are only one component of the total economic impact of K. brevis blooms along the Florida Gulf Coast. Notably, other costs of illness relating to illness severities and different effects, such as shellfish poisoning, are likely to exist. Additional costs may be associated with accessing primary care physicians, allergists, or pulmonologists, as well as prescriptions and overthecounter medi cations. These data were unavailable for this study, however. Although there is no obvious evidence of mortalities as a consequence of respiratory disease, chronic respiratory prob lems may be triggered or exacerbated by aero solized breve toxins. Here, we did not attempt to estimate the costs of additional medi cal services required to treat chronic illness. Consequently, our costofillness estimates should be considered conservative.
Further, we did not account for the non market costs associated with pain and suffer ing when an individual experiences respiratory problems. When a bloom occurs, people often avoid visiting the beach; thus, additional non market economic losses are associated with this behavior. Morgan et al. (2009) recently measured economic impacts in the coastal restaurant trade during HAB events and indi cated that other tourismrelated losses, such as declines in snowbird visits, could have broader impacts on the regional economy.
As the human migration to the coast of the Gulf of Mexico shows little evidence of abating, it will be necessary to consider alterna tive means of mitigating the public health and economic impacts of this natural hazard. It is unclear at this juncture whether human activi ties have contributed to the observed increase in frequency and intensity of K. brevis blooms (cf., Gilbert et al. 2005). Preliminary mitiga tion strategies involve risk communication (poison information center hotlines, signage, informational brochures and materials), natural resource management (shellfish bed closures), and nutrient controls (municipal fertilizer ordi nances). All of these strategies entail costs.

Conclusions
We found that the number of respiratory specific ED diagnoses at SMH for UAD/bron chitis were correlated with lagged K. brevis cell counts, low air temperatures, influenza outbreaks, high pollen counts, and tourist visits. Extrapolating to Sarasota County, we estimated that the capitalized costs of illness for ED respiratory illnesses associated with K. brevis blooms in Sarasota County alone ranged from $0.5 to $4 million. The costs of illness of ED visits are a conservative estimate of the total economic impacts of K. brevis blooms in Sarasota County and for the Florida Gulf Coast. Because this is only a portion of the total economic impact of these events, it will become increasingly necessary to under stand the full scale of the economic losses asso ciated with K. brevis blooms in order to make rational choices about appropriate mitigation.