ORIGINAL PAPER

Direct and indirect mortality in Florida during the 2004hurricane season

Nathan McKinney & Chris Houser &

Klaus Meyer-Arendt

Received: 9 April 2010 /Revised: 14 September 2010 /Accepted: 20 September 2010# ISB 2010

Abstract Previous studies have shown that natural disas-ters, and hurricanes in particular, have led to more deathsthan those usually documented in short post-storm surveys.Such indirect deaths, thought to be related to dietary, stressor pre-existing medical conditions, can exceed the numberof direct deaths and may persist for weeks or even monthsbeyond the event itself. In the present study, cumulativesum of deviations plots are used to quantify the number ofdirect and indirect deaths resulting from HurricanesCharley, Frances, Ivan and Jeanne that made landfall inFlorida in 2004. Results suggest that there was an elevatedmortality for up to 2 months following each storm, resultingin a total of 624 direct and indirect deaths attributable to thestorm. Trauma-related deaths that can be associated directlywith the storm account for only ∼4% of the total storm-related mortality, while indirect mortality accounts for moststorm-related deaths. Specifically, a large percentage of theelevated mortality was associated with heart (34%) andcancer-related deaths (19%), while diabetes (5%) andaccident-related deaths (9%) account for a smaller but stillsignificant percentage of the elevated mortality. The resultsfurther suggest that the elevated mortality was the result ofadditional deaths that would not have otherwise occurredwithin that 5 month period, and not simply a clustering ofdeaths that were inevitable between 1 August and 31

December 2004. The elevated mortality identified in thisstudy is significantly greater than the official count of 31direct and 113 indirect deaths resulting from the fourhurricanes combined. This suggests a need for improvedmortality counts and surveillance in order to better evaluateand identify effective prevention policies, and to identifypreventable deaths.

Keywords Hurricane .Mortality . Direct and indirect death

Introduction

Floods, earthquakes, tropical storms and other high magni-tude environmental events have been referred to historicallyas disasters because they are often associated withcatastrophic damage to human lives and property. Increas-ingly in hazard research, the events are not themselvesviewed as disasters as the term disaster is used to describethe social situation resulting from both a hazardous eventand a population susceptible to damage from that event(Combs et al. 1999). An increased understanding of theseevents and the damaging effects associated with them hasallowed the developed world to dramatically decrease theirimpact (Cardona 2003). This begins with the identificationand accurate evaluation of the processes that initiated thedisaster, the vulnerable systems that allowed the disaster tooccur, and the form and extent of the damages caused bythe disaster. The damage is measured and reported usingtwo approaches: economic damage and impact on humanlife. The economic impact, estimated from insurancereports and then doubled to account for uninsured losses,measures the amount of damage and value loss tobuildings, infrastructure, property and other assets (UNDP2004; Pasch et al. 2005). Because low-lying coastal areas

N. McKinney :K. Meyer-ArendtDepartment of Environmental Studies, University of West Florida,11000 University Parkway,Pensacola, FL 32514, USA

C. Houser (*)Department of Geography, Texas A&M University,810 O&M Building,College Station, TX 77843-3147, USAe-mail: chouser@tamu.edu

Int J BiometeorolDOI 10.1007/s00484-010-0370-9

are most vulnerable to hurricane effects, the rapid popula-tion growth in these areas has caused the economic impactof tropical storms and hurricanes to increase exponentiallyover the last century, even when adjusted for inflation(Riebsame et al. 1986). Two-thirds of the top 30 costliesthurricanes on record occurred in the past 25 years, a periodover which the US Atlantic and Gulf Coasts gained50 million new residents (Blake et al. 2007).

Despite the influx of population to vulnerable coastalareas and an increasing number of deaths due to other typesof natural hazards, the average number of deaths causeddirectly and indirectly by tropical storms and hurricanes hasdecreased dramatically over the course of the twentiethcentury (Rappaport 2000; Riebsame et al. 1986; UNDP2004: http://www.undp.org/annualreports/2004/english/).Of the 51 hurricanes between 1851 and 2006 causing 25or more deaths, only 8 occurred since 1960. This decreaseis thought to be the result of better preparation practices andtechnological advancement. Early warning systems com-bined with better communications and transportationinfrastructure have allowed at-risk populations to evacuatefrom areas susceptible to surge and flooding whileimproved building regulations and materials have madebuildings less hazardous in high winds (Riebsame et al.1986). Prior to 1970, the majority of deaths related tohurricane impacts were attributed to direct contact withstorm surge, especially drownings in salt water caused bystorm surge (Rappaport 2000). Improvements in prepara-tion practices in recent decades have been so successful thatstorm surge initiated drownings are now considered rare.While exceptionally successful by any measure, currentpreparation and mitigation practices have not been able tocompletely eliminate tropical storms as a hazard to humanlife. Arguez and Elsner (2001) note that, despite thedecreases in overall death count and average storm relatedmortality, the median number of deaths occurring due tocontact with hurricanes and tropical storms has increasedover the last century, since it is now nearly certain that astorm making landfall will impact a populated area. The1,800-plus deaths associated with Hurricane Katrina dem-onstrates that widespread casualties are still possible despitethe near 80-year period in which no storm was associatedwith more than 500 deaths (Blake et al. 2007; Knabb et al.2006).

In the United States, compiling the death count forhurricanes and tropical storms is the responsibility of thelocal coroner or medical examiner’s office. Medicalexaminer districts may cover a single county or spanseveral contiguous counties, such as Florida’s 67 countiesthat are covered by 24 medical examiner districts. Becausea hurricane impact is certain to affect multiple districts,hurricane mortality data gathered at the district level arecompiled by a state level agency, such as the Florida

Medical Examiners Commission (Ragan et al. 2008). Alldeaths occurring under unusual or suspicious circumstances,trauma or by a continuing public health threat, includingtropical storms, are required by Florida law to be investigat-ed (Ragan et al. 2008). Medical examiners inspect thecircumstances and evidence surrounding each death and usea case-classification method to determine whether the deathis a result of the storm. (Combs et al. 1996; Shultz et al.2005). This classification procedure works well for creating ahigh precision count of flood- and surge-related drownings,trauma from falling trees, structures or wind-propelled debrisand other events involving contact with wind or water duringthe storm impact period. Hurricanes, however, as with otherhazardous environmental events, can cause human deaths bymeans other than by contact with the event’s physical forcesand can create conditions that contribute to increasedmortality well after the wind and rain have ceased (Combset al. 1996).

Deaths not directly resulting from storm forces, but fromthe ensuing disaster are referred to as indirect deaths todistinguish them from drownings and trauma occurringduring the storm (Combs et al. 1996). While indirectmortality is not caused directly by the forces of the event, itis a product of the conditions resulting from the event’scontact with a human population and should still be viewedas a consequence of the hurricane. Counting indirect deathsis a relatively recent development in disaster epidemiologyand is still overlooked completely in some studies. Studiesthat have attempted to take indirect deaths into account andhave reported them separately have consistently identified agreater number of deaths occurring due to indirectly relatedcauses than from the physical hurricane forces (Combs etal. 1996; Jani et al. 2006; Shultz et al. 2005; Zachria andPatel 2006). Because the immediate or proximate cause ofan indirect hurricane death is not limited to contact withwater, wind or storm-driven objects, the list of possiblecircumstances surrounding these deaths is endless. However,this also makes it more difficult to assess whether a deathwas triggered or in some way dependent on the storm(Bourque et al. 2006; Combs et al. 1996). Despite the highpercentage of indirect deaths cited in recent studies, manyresearchers were found to be highly critical of current studydesigns of post-storm health and mortality assessments(Dominici et al. 2005; Shultz et al. 2005).

Among the more often cited mechanisms of indirecthurricane death are electrocutions due to downed powerlines, carbon monoxide poisoning from improper use ofelectric generators, accidents during repair and cleanupactivity, candle started fires, and automobile accidents dueto wet, damaged, or obstructed roads. These typically work-related deaths occur up to 37 days after landfall of the storm(Fayard 2009) and lead to an increase in the number of callsto Poison Control Centers (Forrester 2009). Recent studies

Int J Biometeorol

also suggest that storms may influence the mortality ratethrough other less observable mechanisms (Combs et al.1996; Hendrickson and Vogt 1996; Jonkman and Kelman2005). Abramson et al. (2008) suggests that the increase inmortality and chronic illness following Hurricane Katrinastems from an increase in mental health distress that variedwith the number of children in the household, fatalisticsentiments and poor self-efficacy. Post-storm distress isfurther exacerbated by the loss and displacement of familymembers, friends, pets and social networks (Combs et al.1999; Lavie et al. 2009), and possibly the displacement ofprimary care physicians (Madamala et al. 2009). Thedistress often leads to an increase in cardiovascular disease,chronic lower-respiratory disease and diabetes/glucoseabnormalities in both men and women, and an increase inobstetric/gynecological disease in women (Sharma et al.2007). Recent evidence from Joseph et al. (2009) suggeststhat a large number of those recently diagnosed with cancerin parts of Louisiana, Alabama and Mississippi were fromthe population displaced by Hurricane Katrina. As shownby Gautam et al. (2009), the increase in chronic illness andmortality can persist for several years in response tocontinued unemployment, lack of medical insurance andliving in temporary housing, in addition to noncompliancewith medication, smoking and substance abuse. Whilefurther removed from the storm’s physical forces, Kelman(2004) argues that indirect disaster deaths are no less aproduct of the event than those occurring by direct means.

Stephens et al. (2007) attempted to evaluate mortalityfollowing Hurricane Katrina in New Orleans by countingnewspaper death notices post-storm and comparing theresults to previous years. The study found an increase inmortality between January and June 2006 almost 47%higher than over the same period between 2002 to 2004.Because this study focused on Katrina, which led to thelong-term displacement of a large part of the population andextensive damage to the City of New Orleans, the resultsare not likely to be representative of most hurricanelandfalls in the United States. Nonetheless, the study doessuggest that an elevated mortality rate is not just restrictedto the immediate post-impact phase (see also Bennet 1970;Baxter et al. 2001; Galea et al. 2005; Shultz et al. 2005),although there remains skeptism whether there is arelationship between post-storm symptoms and mortalityrates (van den Berg et al. 2005; Yzermans et al. 2005). Inpractice, researchers have limited the scope of post-disastermortality studies to an arbitrary period of 2 weeks to amonth (Combs et al. 1996). Despite greater recognition andfocus on indirect mortality, there have been only a handfulof studies (following Hurricane Katrina) to consider the‘long-term’ impact of a storm on direct and indirectmortality. The purpose of the present study is to quantifythe number of direct and indirect deaths resulting from the

four hurricanes that impacted Florida in 2004 and todetermine how long the increased mortality persistedfollowing the storm.

2004 storm season in Florida

In 2004, a record four hurricanes (Charley, Frances, Ivanand Jeanne) and Tropical Storm Bonnie made landfall inthe State of Florida (Fig. 1). Bonnie made landfall south ofApalachicola, Florida on 12 August as a relatively weaktropical storm that resulted in three fatalities in thesoutheastern United States and one in New Brunswick,Canada (NOAA 2004: http://www.nhc.noaa.gov/2004atlan.shtml). Hurricane Charley made landfall near Punta Gorda,Florida on 14 August 2004 as a Category 4 storm, causing∼US $19 billion in damage (Table 1). It was estimated thatthe storm was responsible for 37 deaths in Florida.Hurricane Frances made landfall on 5 September 2004 asa Category 2 storm between Fort Pierce and West PalmBeach, and again in the Florida Panhandle as a tropicalstorm. Approximately 3 weeks later, Hurricane Jeannemade landfall at almost the exact same location asHurricane Frances as a Category 3 storm. Combined,Frances and Jeanne were responsible for 68 deaths andUS $16 billion in property damage. Hurricane Ivan madelandfall near the Florida-Alabama border, putting Escambiaand Santa Rosa counties in the northeast quadrant of thisCategory 3 storm. The storm was blamed for 68 deaths inthe United States (39 in Florida) and US $13 billion inproperty damage (Ragan et al. 2008). Population estimatesbased on house counts (Florida Office of Economic andDemographic Research 2010) suggest that year-to-yearpopulation growth between 2001 and 2006 in the affectedareas was unaffected by the storms, although there was nosignificant change in population between 2004 and 2005 inthe area affected by Hurricane Ivan. However, it isimportant to note that these are simply gross estimatesbased on the number of occupied houses and the averagepersons per household (2.46 for 2000), and do notnecessarily reflect the true population nor the short-termchanges in population immediately following the storm andduring the rebuilding period.

Methods

Mortality data

Mortality records were acquired for Florida countiesmaking up three study areas corresponding to eachhurricane (see Fig. 1). Because Frances and Jeanne madelandfall in essentially the same area, the area corresponding

Int J Biometeorol

to these hurricanes were combined. The study area for eachhurricane was based on those counties: that were (1)directly impacted by the hurricanes, (2) ordered evacuatedregardless of the level of damage that actually occurred, and(3) adjacent to the impact zone where direct deaths werereported. Due to the prohibitive costs of acquiring allmortality data, the analysis was limited to the months ofAugust through December, when direct and indirectmortality from the storms was expected. Data werepurchased for the years 2001 through 2006 with the yearsother than 2004 needed as comparison data. The source ofthe data is the State of Florida Department of Health Officeof Vital Statistics. Since only gross estimates of annualpopulation are available for the selected region (FloridaOffice of Economic and Demographic Research 2010), the

analysis is based on the mortality count and not themortality rate (mortality count/population).

Records were organized into daily mortality counts foreach affected region and plotted as a time series. In addition tothe overall daily death count, the data were organized toidentify increases in specific causes of death. A cause analysiswas based on the RECODE_DEATH_GROUP field, whichrepresents a grouping of the International Classification ofDiseases (ICD, version 10) codes.While the dataset containedvery specific causes using ICD-10 coding, it was decided thatthe 12,000 possible causes using ICD-10 were too specificand would produce insufficient sample sizes. The recodedfield we used contains 113 possible death cause groups.

For the purposes of this study, some cause codes werefurther grouped. Deaths that will be referred to as “heart-

Fig. 1 Map of Florida showing tracklines and storm-affected areas for Hurricane Ivan and Hurricanes Charley, Frances, Ivan and Jeanne duringthe 2004 Hurricane Season along with the spatial extent of hurricane force winds

Date of landfall Damage (billion US $) Official death count

Hurricane Charley 14 August 19 37

Hurricane Frances 5 September 9 46

Hurricane Ivan 16 September 13 39

Hurricane Jeanne 25 September 6.8 22

Table 1 Dates, damageestimates and official deathcounts for hurricanes to makelandfall in the State of Floridaduring the 2004 hurricaneseason

Int J Biometeorol

related” encompass 11 groups of diseases of the circulatorysystem including stroke. The class referred to as “cancers”includes malignant neoplasms of various organs, lymphomaand leukemia. “Accidents” refers to falls, accidentalpoisoning, and drowning but not to deaths caused by motorvehicle crashes or transport accidents, which were analyzedseparately. These groupings were created (1) to increasesample size to a level where analyses would be valid, and(2) because many of the 113 death codes are thought not tobe sufficiently specific as to have different hurricane-relatedrisk factors, such as cancers of different organs. In additionto these groups, the code for diabetes mellitus wasconsidered in response to the findings of Hendrickson andVogt (1996).

Data analysis

Subtle changes in mortality following a hurricane or otherextreme event can be identified by examining the cumula-tive sum of deviations from the (long-term) averagemortality for a given area. An increase in mortality willbe visible as an increase in the cumulative sum ofdeviations, while a decrease in mortality relative to theaverage will be visible as a decrease in the cumulative sumof deviations. No increasing or decreasing trend in thecumulative sum of deviations occurs when the number ofdeaths fluctuates around the long-term average. It isimportant to note that the exact start and end dates for aperiod of elevated mortality cannot be separated fromstorm-related deaths, since it is statistically possible to haveseveral days in a row with elevated mortality. What can bedefined as the start of an elevated period depends on thedistribution of daily mortality counts for a specific causeduring non-storm periods.

The long-term average mortality is calculated for alldeaths within the storm-affected counties and for eachmortality group in those counties over the same 5-monthperiod for years between 2001 and 2006. Plots of thecumulative sum of deviations between 1 August and 31December are then prepared for the ensemble average of allnon-hurricane years between 2001 and 2006, the storm-affected counties during the 2004 hurricane season, and forthe mortality groups individually. A period of elevatedmortality following a storm can be identified by anextended period with an increasing cumulative sum ofdeviations that is initiated at or close to storm landfall, andends when the cumulative deviation stabilizes (i.e., mortal-ity returns to the long-term average). The deviation of dailymean mortality for all counties in the storm-affected areasin non-hurricane years between 2001 and 2006 is presentedin Fig. 2a, and the corresponding cumulative sum ofdeviations plot is presented in Fig. 2b. The plot shows adecrease in mortality through the late summer, a stable

mortality near the long-term average in the fall, followed byan increase in mortality through the winter holidays (seePhillips et al. 2004). Use of cumulative sum allows theidentification of a non-arbitrary study period that isdifferent for each study area or each cause group. Oncethe period is identified, an estimate of the hurricane’s totalimpact on deaths can be derived from the difference of theobserved deaths in this period in 2004 and the meannumber of deaths expected from the same period duringnon-hurricane years between 2001 and 2006.

Where an extended period of elevated mortality isidentified following a storm, Grubb’s test is used todetermine if the number of deaths in total and for eachmortality type is significantly greater than in non-hurricaneyears. The test statistic (Z) is calculated as the differencebetween the mortality count for 2004 (x) and the averagemortality x for all years divided by the standard deviation(σ) of non-hurricane years:

Z ¼ x� x2004s

Since the standard deviation is affected by the inclusionof the 2004 data, there is a maximum to the test statisticsdefined by:

N � 1ffiffiffiffi

Np

where N is the number of samples (Grubb 1969). Thecritical test statistic at the 95 and 99% confidence levels is1.96 and 2.59, respectively, such that values greater thanthese critical values have less than 5% and 1% chance thatthe data from 2004 is not an outlier and is statisticallysimilar (at the 95% confidence level) to the non-hurricaneyears. Grubb’s test assumes that the mortality count, oversome defined period, is normally distributed. The mortalitycounts for non-hurricane years at daily, weekly andmonthly intervals were examined for normality using aKolmogorov-Smirnoff test. Total mortality counts for non-hurricane years between 2001 and 2006 are not significantlydifferent at the 95% confidence level from a normaldistribution at daily (d=0.032, P>0.05), weekly (d=0.049,P>0.05) and monthly (d=0.066, P>0.05).

Results

Hurricane Charley

The cumulative deviation time series are presented inFig. 3 for all deaths in counties affected by HurricaneCharley (Fig. 3a), for each of the major mortality codes(Fig. 3b–e) and for the non-hurricane years between 2001

Int J Biometeorol

and 2006 (Fig. 3f). Similar to Fig. 2, the plot for non-hurricane years shows a an extended period of below-average mortality through the late summer, a stablemortality near the long-term average in the fall, followedby an extended period of above-average mortality throughthe winter holidays. In contrast, there is a 62-day period ofelevated mortality beginning ∼3 days following stormlandfall, in which there were 113 more deaths than thenon-hurricane average over the same period (Table 2).Grubb’s test suggests that the increased mortality over that62 day period is a statistically significant outlier relative tonon-hurricane years at the 95% confidence level (P>0.05,d=1.98). The greater mortality during the storm-impactperiod is associated with a statistically significant increasein heart, cancer and accident-related deaths. Heart-related

deaths were elevated significantly for 62 days followingstorm landfall (Fig. 3b), in which there were approxi-mately 72 more deaths than during the same period innon-hurricane years. Cancer-related deaths remain ele-vated for only 24 days after storm landfall (Fig. 3c), withapproximately 21 more deaths than in non-hurricane yearsfollowed by a significant decrease in cancer-related deathsthrough the remainder of 2004. Diabetes-related deathswere also elevated 15 days following storm landfall for atotal of six more deaths than during the non-hurricaneyears (Fig. 3d). As shown in Fig. 3e, accident-relateddeaths started to increase 6 days before landfall, but thefirst 3 days of this increase occurred before a state ofemergency was declared and the storm was forecast tomake landfall in western Florida. The initial increase is

Fig. 2 Time series of adeviations from daily meanmortality, and b cumulative sumof deviations during non-hurricane years for all Floridacounties affected by the 2004hurricanes between 1 Augustand 31 December. As annotated,the cumulative time seriesexhibits an extended periodof below-average mortalitythrough the late summer(1 August–19 September),average mortality through thefall (19 September–1December), and an extendedperiod with above-averagemortality through the winterholidays (1–31 December)

Int J Biometeorol

most likely part of the background variability and cannotbe separated from the storm-related increase. In thisrespect, the increase in accident-related deaths can onlyhave started 3 days before landfall and is responsible for at

most an extra 21 deaths over 24 days. Combined, heart,cancer, diabetes and accident-related deaths account for∼89% of the elevated mortality during the post-stormperiod.

Fig. 3 a–f Time series and cumulative time series of mortality incounties impacted by Hurricane Charley. Time series: a all 2004deaths, b 2004 heart-related deaths, c 2004 cancer-related deaths, d2004 diabetes-related deaths, e 2004 accident-related deaths, f average

deaths for non-hurricane years between 2001 and 2006. Sections of thecumulative time series showing elevated mortality in response toHurricane Charley are highlighted

Int J Biometeorol

Hurricane Ivan

The cumulative deviation time series are presented in Fig. 4for all deaths in counties affected by Hurricane Ivan(Fig. 4a), for each of the major mortality codes (Fig. 4b–e)and for the non-hurricane years between 2001 and 2006(Fig. 4f). As shown in Fig. 4a, there is a statisticallysignificant (P>0.05, d=2.05) increase in mortality over the47-day period following landfall, in which there were 90more deaths than over the same period in non-hurricaneyears (Table 3). Unlike the counties affected by HurricaneCharley, the increase in mortality is also associated with anincrease in heart, cancer and accident-related deaths; nostatistically significant increase in diabetes-related deaths isobserved. Almost one-third of the elevated mortality isassociated with the 29 cancer-related deaths that occurredover a 37-day period starting 6 days after landfall(Fig. 4c). Five days after landfall, heart-related deathswere elevated for 29 days for 18 more deaths compared tonon-hurricane years (Fig. 4b). Accident-related deathsbecame elevated 5 days before landfall and remainedelevated for 22 days, for a total of 12 more deaths than innon-hurricane years (Fig. 4d). Combined, heart, cancerand accident-related deaths account for∼66% of themortality during the post-storm period of elevated mortal-ity. Other causes that were significantly elevated (at the95% confidence level) during the post-storm periodinclude fetal and newborn deaths (4 deaths, Z=2.7),cirrhosis of the liver (5 deaths, Z=4.7.), drowning (5deaths, Z=3.2) and ill-defined/unknown causes (17deaths, Z=5.4). While relatively small, results of theGrubb’s analysis indicates that there is less than 1%chance that these increases are not significantly differentfrom the non-hurricane years examined in this study. Dueto the limited sample size and small increases, thesesecondary causes of mortality will not be discussedfurther.

Hurricanes Frances and Jeanne

The cumulative deviation time series are presented in Fig. 5for all deaths in counties affected by Hurricanes Francesand Jeanne (Fig. 5a), for each of the major mortality codes(Fig. 5b–e) and for the non-hurricane years between 2001and 2006 (Fig. 5f). Combined, the storms accounted forapproximately 401 more deaths than over the same 59-dayperiod in non-hurricane years—a statistically significantdifference at the 95% confidence level (P>0.05, d=2.08).Consistent with Hurricane Charley, the increase in mortalityis related largely to an increase in heart, cancer, accidentand diabetes-related death, which combined account for60% of the increased mortality. Other causes that weresignificantly elevated (at the 95% confidence level) duringthe post-storm period include bronchopneumonia (23deaths, Z=7.8), pancreatitis (8 deaths, Z=3.8), HIV-related deaths (4 deaths, Z=4), Alzheimer’s (31, Z=2.2)and ill-defined/unknown causes (21 deaths, Z=13.9). Thesedeaths were concentrated largely in Brevard and IndianRiver counties. As noted with Hurricane Ivan above, theresults of the Grubb’s analysis suggests that these relativelysmall increases during the post-storm period have less thana 1% chance of occurring by random chance alone andappear to result from the storm. However, due to limitedsample size and the small increases, these secondary causesof mortality will not be discussed further.

Heart-related deaths were significantly elevated for the58 days following landfall by Hurricane Frances (Fig. 5b),in which there were approximately 122 more deaths thanduring the same period during the non-hurricane years.Accident-related deaths started to increase before landfalland remained elevated for 20 days with an extra 24 deaths.Cancer-related deaths started to increase 3 days followinglandfall of Hurricane Frances and remained elevated over16 days, with 22 more deaths than in non-hurricane years.There was a second increase in cancer-related deaths in the

Hurricane All Heart Cancer Diabetes Accident

Charley Start Day -3 -3 3 0 -3

Length 62 62 24 15 24

Deaths 133 72 21 6 20

Ivan Start Day -4 5 6 -5

Length 47 29 37 22

Deaths 90 18 29 12

Frances/Jeanne Start Day -1 -3 3/19 1/9 -1

Length 59 58 16/20 12/6 20

Deaths 401 122 22/45 14/13 24

Average Start Day -1 0 8 3 -3

Length 54 50 24 11 22

Total Deaths 624 212 117 33 56

Table 2 Mortality statisticsfor each hurricane and theaverage of the storms. The startday (relative to hurricanelandfall), length of elevatedmortality and the number ofdeaths associated with eachstorm are presented

Int J Biometeorol

19 days following landfall of Hurricane Jeanne, in whichthere were 45 deaths over a 20-day period. Diabetes-relateddeaths also exhibited separate peaks for both storms, withan initial increase of 14 deaths for 12 days following

Hurricane Frances. Following Hurricane Jeanne, there are13 more diabetes-related deaths over a 6-day period. Bothcancer and diabetes-related deaths are stable around thelong-term average between storms, suggesting that the

Fig. 4 a–f Time series and cumulative time series of mortality incounties impacted by Hurricane Ivan. Time series: a All 2004 deaths,b 2004 heart-related deaths, c 2004 cancer-related deaths, d 2004diabetes-related deaths, e 2004 accident-related deaths, f average of

deaths for non-hurricane years between 2001 and 2006. Sections of thecumulative time series showing elevated mortality in response toHurricane Ivan are highlighted

Int J Biometeorol

increase in mortality following the storms is associated withadditional deaths, rather than a clustering of those whowould have died during that same period.

Discussion

Tropical storms and hurricanes can exact considerableeconomic and social impact on a vulnerable population.To some degree, the social impact of a storm can bemitigated (with varying levels of success) through betterpreparation and technological advancements. The ability toforecast storm landfall location and timing, along withimproved communication and transportation infrastructureallows for at-risk populations to better prepare and evacuateareas subject to storm-force winds and surge. This hasresulted in a significant decrease in the number of deathsdirectly attributed to the storm at landfall. However, theever-increasing population along the coast means that it isstill possible for a large number of deaths to be directlyattributable to a storm that makes landfall near a large urbancenter such as New Orleans. While official counts hadHurricane Katrina directly responsible for 1836 deaths,recent studies estimate that there were significantly moredeaths resulting indirectly from the storm for an extendedperiod after landfall (Stephens et al. 2007). Results from thepresent study on the impact of Hurricanes Charley, Frances,Ivan and Jeanne on mortality in Florida suggest that thenumber of deaths have been underestimated by a factor ofmore than four. Specifically, we estimate that there were624 deaths in response to the four hurricanes that struckFlorida in 2004, while only 144 deaths were officiallyreported (Table 3).

The low official count can be attributed partly to thelimited scope of most post-disaster mortality studies, whichtend to be only 2-weeks to a month in length (Combs et al.1996). Results of the cumulative sum analysis suggest thatthe period of elevated mortality ranged from 35 days forHurricane Charley to 44 days for Hurricane Ivan. Hurri-canes Frances and Jeanne combined were responsible forelevated mortality for 59 days following landfall of the firststorm. For Hurricane Charley, the period of elevatedmortality started within ∼3 days before the storm made

landfall in response a greater number of accident-relateddeaths as residents prepared and evacuated ahead of thestorm. Heart-related deaths started to increase, relative tothe non-hurricane average, 3 days prior to HurricaneFrances making landfall, and remained elevated for 60 days.There was no pre-storm increase for Hurricanes Ivan andJeanne, although in the latter case it is difficult todistinguish between deaths after Hurricane Frances fromthose in response to the arrival of Hurricane Jeanne.Regardless, these results suggest the need for a reassess-ment of the methods and the period over which storm-related deaths are counted. Moreover, the results suggest aneed to reevaluate the absolute and relative number ofdirect and indirect deaths associated with past storms.Indirect deaths are no less a product of the storm as thoseoccurring by direct means, and strategies to reduce post-storm mortality depend on an understanding of the primaryforms and reasons.

Not only are the cumulative sum of deviation plotssensitive to small and persistent increases in mortality, theysuggest that for all storms the elevated mortality is theresult of additional deaths not a clustering of those deathsthat most likely would have occurred within the sameperiod. A clustering of inevitable deaths would be visibleby a decreasing trend in the cumulative sum of deviationsplot following the post-storm period of elevated deaths. Therelatively stable deviation plot between Hurricanes Francesand Jeanne for cancer and diabetes-related deaths (Fig. 5c,d) indicates that there is a return to the non-hurricaneaverage between storms. In other words, the elevatedmortality is associated with additional (new) deaths andnot a clustering of inevitable deaths between 1 August and31 December. This is not to suggest that the storms madenon-lethal cases of cancer and diabetes lethal, but ratherthat the additional deaths during the post-storm periods (of62, 47, 59 and 58 days respectively) came from outside the5-month period examined in this study. It is not possiblewith the available data to determine if there was asignificant increase in the number of deaths sometimebetween January and August 2005, which would not be anoticeable or represent a statistically significant decrease inmortality if these additional deaths were taken from that212-day period. In contrast, there is a decrease in the

Charley Ivan Frances/Jeanne

Period of increased mortality (days) 62 47 59

Average deaths/day during 2004 36.3 12.5 70.5

Expected deaths per day 32.5 10.3 63.7

Deaths above expected 133 90 401

Official death count 37 39 68

% explained by official death count 28 40 17

Table 3 Comparison ofobserved storm-dependentmortality and official mortalityfor the 2004 hurricane season inFlorida

Int J Biometeorol

deviation plot for cancer and diabetes-related deathsfollowing Hurricane Charley (Fig. 3c,d), suggesting thatthe elevated mortality is associated with a clustering ofcancer and diabetes-related deaths from within that 5-monthperiod. A similar pattern is observed for cancer-related

deaths following Hurricane Ivan (Fig. 4c). However,following Hurricane Charley, the mortality for all deathscombined is stable after the elevated post-storm period,indicating that a large part of the elevated mortality ispossibly the result of additional deaths from outside the

Fig. 5 a–f Time series and cumulative time series of mortality incounties impacted by Hurricanes Frances and Jeanne. Time series: aAll 2004 deaths, b 2004 heart-related deaths, c 2004 cancer-relateddeaths, d 2004 diabetes-related deaths, e 2004 accident-related deaths,

f average of deaths for non-hurricane years between 2001 and 2006.Sections of the cumulative time series showing elevated mortality inresponse to Hurricanes Frances and Jeanne are highlighted

Int J Biometeorol

5-month period and not a clustering of inevitable deathsbetween 1 August and 31 December.

For all storms, the period of elevated mortality wasassociated with a statistically significant increase in heart,cancer and accident-related deaths. While these causes ofdeath were all elevated for some time after the storm, theperiod of elevated mortality was different for each cause.Accidents were elevated at, or just prior to, the stormmaking landfall, and remained elevated for an average of22 days (see Figs. 3e, 4e). Among the most often citedmechanisms of indirect hurricane death are electrocutionsdue to downed power lines, carbon monoxide poisoningfrom improper use of electric generators, accidents duringrepair and cleanup activity, candle-started fires, andautomobile accidents due to wet, damaged, obstructed andcongested roads. Accidental deaths are elevated during aperiod in which residents prepare and protect their propertyfor the coming storm, evacuate, return to their storm-damaged properties and begin repairs. For the 2004hurricanes, trauma accounted for 49% of all accidentaldeaths, followed by drowning (11%), electrocution (5%),CO poisoning (7%), burns/inhalation (3%). However, thesetrauma-related deaths account for only 4% of all deathsattributable to the storms. In comparison, heart- and cancer-related deaths account for 34% and 19% of all deathsattributable to the storms, and over a period of up to2 months following landfall (Table 3). It is reasonable toexpect that the number of storm-related deaths and thecause of mortality will be different between the urban andrural areas impacted by the storm (see Borden and Cutter2008), but unfortunately the data used in the present studydo not provide this level of detail. Further research isrequired not only to refine cause of death and the circum-stances/stressors responsible, but also the socio-economicsetting in which it occurred.

While readily observable as being related to the storm,the results of this study and those following HurricaneKatrina (e.g., Stephens et al. 2007) suggest that accidentaldeaths account for only a small percentage of the total anddo not persist for an extended period after the storm. Theresults of this study suggest that the stress of the pre-stormpreparation, evacuation and post-storm repairs can alsoexacerbate pre-existing chronic illnesses such as heartdisease, cancer and diabetes. Heart-related deaths becameelevated several days following storm landfall andremained elevated for an average of 50 days, whilecancer-related deaths also became elevated an average of24 days after storm landfall. While it is difficult to identifythe exact stressors, previous studies suggest that post-stormdistress is exacerbated by the loss and displacement offamily members, friends, pets and social networks (Combset al. 1999; Lavie et al. 2009). Armenian et al. (1998) also

found that most significant increases in heart disease riskfollowing earthquakes correlate to the loss of familymembers and property. It is also reasonable to expect thatextended power outages and the lack of air conditioning forseveral weeks following a storm can lead to heat stresssimilar to that observed following extreme heat waves(Kalkstein and Davis 1989; Whitman et al. 1997; Davis etal. 2003).

It is important to note that the elevated mortalitydescribed in the present study is based on mortality countsand not a population adjusted mortality rate. Populationchanges, whether associated with the storms or not, willaffect the number of deaths and may lead to an overestimateof hurricane-related mortality if 2004 had the largestpopulation between 2001 and 2006. This is particularlyimportant to the elevated mortality estimated for cancer,which typically has not shown sensitivity to short-termenvironmental or seasonal factors, but rather increases indirect response to population change. A post-storm reduc-tion in population in the years following a hurricane ispossible if the storm causes extensive damage and residentsare displaced for an extended period. A significant changein population was observed in New Orleans for severalyears following Hurricane Katrina (Kates et al. 2006), butdamage during the 2004 hurricane season in Florida wasnot as extensive, and the population estimates published byFlorida Office of Economic and Demographic Research(2010) suggest that the population continued to increase in2005 and 2006 and did not peak in 2004. While it isreasonable to expect that the most heavily damaged areas(such as Pensacola Beach and Gulf Breeze in Escambia andSanta Rosa Counties, respectively) would experience apost-storm drop in population, the population remained inthe storm-affected region and population growth wasobserved in all regions following the 2004 storm season.Without an accurate population count for each year between2001 and 2008, and in the immediate period following thestorm, the degree to which the elevated mortality is anartifact of population is not clear and requires furtherinvestigation. Despite this potential limitation, the presentedevidence clearly indicates that the hurricanes had ameasurable impact on mortality, consistent with studiesthat show an increase in mortality and chronic illnessfollowing Hurricane Katrina (Abramson et al. 2008; Josephet al. 2009), albeit over a longer post-storm period.

The results of the present study suggest that the period ofelevated mortality (across all types) persisted well after thestorm, with an average of 54 days (Table 2). In comparison,the increase in chronic illness and mortality followingHurricane Katrina has persisted for several years, althoughnot immediately following the storm (Gautam et al. 2009).The disparate periods of elevated mortality reflects the

Int J Biometeorol

much greater level and extent of storm damage duringHurricane Katrina, which displaced a large part of thepopulation for an extended period with disruption of normalcare. Following Hurricane Charley, a large percentage ofhouses with at least one older adult experienced exacerbat-ed or prevented normal care for a pre-existing condition(CDC 2004), but for a relatively short period of time. Thedisplacement during Katrina was exacerbated by the loss offamily, friends and social networks, unemployment, andliving in temporary housing (Combs et al. 1999; Lavie et al.2009). While these stressors were present after HurricanesCharley, Frances, Ivan and Jeanne, they did not affect aslarge a population for as long a time. Regardless of thedifferent levels of storm impact, the results of the presentstudy provide further evidence that there is a persistence ofelevated mortality and most likely chronic illness followingstorm landfall. Improvements in mortality counts andsurveillance are needed to better evaluate disaster-relatedhealth impacts, to evaluate and identify effective preventionpolicies, and to identify preventable deaths.

Conclusions

Cumulative sum of deviations plots are used to quantify thenumber of direct and indirect deaths resulting fromHurricanes Charley, Frances, Ivan and Jeanne that madelandfall in Florida in 2004. Results suggest that there wasan elevated mortality for up to 2 months following eachstorm, resulting in a total of 624 direct and indirect deathsattributable to the storm. The elevated mortality identifiedin this study is significantly greater by a factor of four thanthe official 143 deaths attributed to the four hurricanes. Alarge percentage of the elevated mortality was associatedwith heart- and cancer-related deaths, while diabetes andaccident-related deaths account for a smaller but stillsignificant percentage of the elevated mortality. Moreimportantly, the results further suggest that the elevatedmortality was the result of additional deaths from outsidethe 5-month study period, and not simply the clustering ofdeaths that were inevitable between 1 August and 31December 2004. Further study is required to determine ifthis also translates to an increase in the mortality rate orwhether the estimates of elevated mortality are confoundedby population trends and variability immediately followingthe storm and over the much longer recovery and rebuildingperiod.

Acknowledgment Funding for this study was provided from theNational Weather Service Program Office (Award NumberNA04NWS4680010) through the Florida Hurricane AllianceResearch Program led by Stephen Leatherman of Florida Interna-tional University.

References

Abramson D, Stehling-Ariza T, Garfield R, Redlener I (2008)Prevalence and predictors of mental health distress post-Katrina: Findings from the Gulf Coast child and family healthstudy. Disaster Med Public Health Prep 2:77–86

Armenian HK, Melkonian AK, Hovanesian AP (1998) Long termmortality and morbidity related to degree of damage followingthe 1988 earthquake in Armenia. Am J Epidemiol 148(11):1077–1084

Arguez A, Elsner JB (2001) Trends in US tropical cyclone mortalityduring the 20th century. Florida Geographer 32:28–37

Baxter PJ, Möller I, Spencer T, Spence RJ, Tapsell S (2001) Floodingand climate change. Health effects of climate change. Section 4.6.UK Department of Health Document

Bennet G (1970) Bristol floods 1968—controlled survey disaster.BMJ 3:454–458.

Blake ES, Rappaport EN, Landsea CW (2007) The deadliest, costliest,and most intense United States tropical cyclones from 1851 to2006 (and other frequently requested hurricane facts). USDepartment of Commerce, National Weather Service, NationalHurricane Center, Miami, FL. NOAA Technical MemorandumTPC-5. http://www.nhc.noaa.gov/pdf/NWS-TPC-5.pdf

Borden KA, Cutter SL (2008) Spatial patterns of natural hazardsmortality in the United States. Int J Health Geogr 7(64):1–13.doi:10.1186/1476-072X-7_64

Bourque LB, Siegel JM, Kano M, Wood MM (2006) Weathering thestorm: The impact of hurricanes on physical and mental health.Ann Am Acad Polit Soc Sci 604:129–151

Cardona OD (2003) The need for rethinking the concepts ofvulnerability and risk from an holistic perspective: a necessaryreview and criticism for effective risk management. In: BankoffG, Frerks G, Hilhorst D (eds) Mapping vulnerability: disasters,development and people. Earthscan, London, pp 37–51

CDC (2004) Rapid assessment of the needs and health status of olderadults after Hurricane Charley- Charlotte, DeSoto, and Hardeecounties, Florida, August 27–31, 2004. MMWR 36:837–840

Combs DL, Parrish RG, McNabb SJN, Davis JH (1996) Deathsrelated to Hurricane Andrew in Florida and Louisiana, 1992. Int JEpidemiol 25(3):537–544

Combs DL, Quenomoen LE, Parrish RG, Davis JH (1999) Assessingdisaster-attributed mortality: development and application of adefinition and classification matrix. Int J Epidemiol 28:1124–1129

Davis RE, Knappenberger PC, Michaels PJ, Novicoff WM (2003)Changing heat-related mortality in the United States. EnvironHealth Perspect 111(14):1712–1718

Dominici F, Levy JI, Louis TA (2005) Methodological challenges andcontributions in disaster epidemiology. Epidemiol Rev 27:9–12

Fayard GM (2009) Fatal work injuries involving natural disasters,1992–2006. Disaster Med Public Health Prep 3:201–209

Florida Office of Economic & Demographic Research (2010)Population Based Information. http://edr.state.fl.us/population.htm.

Forrester MB (2009) Impact of Hurricane Ike on Texas Poison ControlCenter Calls. Disaster Med Public Health Prep 3:136–137

Galea S, Nandi A, Vlahov D (2005) The epidemiology of post-traumatic stress disorder after disasters. Epidemiol Rev 27:78–91

Gautam S, Menachem J, Srivastav SK, Delafontaine P, Irimpen A(2009) Effect of Hurricane Katrina on the incidence of acutecoronary syndrome at a primary angioplasty center in NewOrleans. Disaster Med Public Health Prep 3:144–150

Grubb F (1969) Procedures for detecting outlying observations insamples. Technometrics 11:1–21

Int J Biometeorol

Hendrickson LA, Vogt RL (1996) Mortality of Kauai residents in the12-month period following Hurricane Iniki. Am J Epidemiol 144(2):188–191

Jani AA, Fierro M, Kiser S, Ayala-Simms V, Darby DH, Juenker S,Storey R, Reynolds C, Marr J, Miller G (2006) Hurricane Isabel-related mortality—Virginia, 2003. J Public Health Manage Pract12:97–102

Jonkman SN, Kelman I (2005) An analysis of the causes andcircumstances of flood disaster deaths. Disasters 29:75–97

Joseph DA, Wingo PA, King JB, Pollack LA, Richardson LC, Wu X,Chen V, Austin HD, Rogers D, Cook J (2009) Use of state cancersurveillance data to estimate the cancer burden in disasteraffected areas- Hurricane Katrina, 2005. Prehosp Disaster Med22:282–290

Kalkstein LS, Davis RE (1989) Weather and human mortality: Anevaluation of demographic and interregional responses in theUnited States. Ann Assoc Am Geogr 79(1):44–64

Kates RW, Colten CE, Laska S, Leatherman SP (2006) Reconstructionof New Orleans after Hurricane Katrina: a research perspective.Proc Natl Acad Sci USA 103(40):14653–14660

Kelman I (2004) Philosophy of flood fatalities. FloodRiskNetNewsletter 1:3–4

Knabb RD, Rhome JR, Brown DP (2006) Tropical Cyclone Report:Hurricane Katrina. U.S. Department of Commerce, NationalWeather Service, National Hurricane Center, Miami, FL.[http://www.nhc.noaa.gov/pdf/TCR-AL122005_Katrina.pdf]

Lavie CJ, Gerber TC, Lanier WL (2009) Hurricane Katrina: theinfarcts beyond the storm. Disaster Med Public Health Prep3:131–135

Madamala K, Campbell CR, Hsu EB, Hsieh YH, James J (2009)Characteristics of physician relocation following HurricaneKatrina. Disaster Med Public Health Prep 1:21–26

Pasch RJ, Brown DP, Blake ES (2005) Tropical Cyclone Report:Hurricane Charley. US Department of Commerce, NationalWeather Service, National Hurricane Center, Miami, FL. [http://www.nhc.noaa.gov/2004charley.shtml]

Phillips DP, Jarvinen JR, Abramson IS, Phillips RR (2004) Cardiacmortality is higher around Christmas and New Year’s than at anyother time. Circulation 110:3781–3788

Rappaport EN (2000) Loss of life in the United States associated withrecent Atlantic tropical cyclones. Bull Am Meteorol Soc 81(9):2065–2073

Ragan P, Schulte J, Nelson SJ, Jones KT (2008) Mortality surveillance2004 to 2005: Florida hurricane-related deaths. Am J ForensicMed Pathol 27(2):148–153

Riebsame WE, Diaz HF, Moses T, Price M (1986) The social burdenof weather and climate hazards. Bull Am Meteorol Soc 67(11):1378–1388

Sharma AJ, Weiss EC, Young SL, Stephens K, Ratard R, Straif-Bourgeois S, Sokol TM, Vranken P, Rubin CH (2007) Chronicdisease and related conditions at emergency treatment facilities inthe New Orleans area after Hurricane Katrina. Disaster MedPublic Health Prep 2:27–32

Shultz JM, Russell J, Espeinel Z (2005) Epidemiology of tropicalcyclones: the dynamics of disaster, disease and development.Epidemiol Rev 27:21–35

Stephens KU, Grew D, Chin K, Kadetz P, Greenough G, Burkle FM,Robinson SL, Franklin ER (2007) Excess mortality in theaftermath of Hurricane Katrina: a preliminary report. DisasterMed Public Health Prep 1(1):21–26

van den Berg B, Grievink L, Yzermans J, Lebret E (2005) Medicallyunexplained symptoms in the aftermath of disasters. EpidemiolRev 27:92–106

Whitman S, Good G, Donoghue ER, Benbow N, Shou W, Mou S(1997) Mortality in Chicago attributed to the July 1995 heatwave. Am J Public Health 87:1515–1518

Yzermans CJ, Donker GA, Kerssens JJ, Dirkzwager AJE, SoetemanRJH, ten Veen PMH (2005) Health problems of victims beforeand after disaster: a longitudinal study in general practice. Int JEpidemiol 34:820–826

Zachria A, Patel B (2006) Deaths related to Hurricane Rita and massevacuation. Chest 130:134S

Int J Biometeorol