Globally, flooding is the most common natural disaster and the most frequent cause of natural disaster–related mortality. From 1994 to 2013, flooding affected 2.4 billion people and caused 750 000 deaths.1 The impacts on human health associated with flood events can be direct or indirect. Direct effects include risk of drowning and injury. Indirect effects include increased risk of food-, water-, and vector-borne diseases, as well as adverse mental health outcomes.
Flooding may have negative impacts on health by increasing the risk of certain bacterial and protozoan diseases that are typically associated with fecal-oral transmission or contaminated food and water. These include salmonellosis, campylobacteriosis, cryptosporidiosis, giardiasis, and vibriosis. Flooding can lead to sewer system overflows. Contaminated floodwaters can spread and expose many people. For example, Campylobacter has been detected in floodwaters following urban flood events.2 Flooding of the home has been shown in international studies to increase the odds of diseases such as cryptosporidiosis and paratyphoid fever by 200% to 350%.3 The risk of these diseases after flood events in the United States appears to be lower but is still important to consider. Extreme rainfall events (>90th percentile) preceded 51% of waterborne disease outbreaks in the United States (1948-1994).4 Increased risk of gastrointestinal illness with flooded homes or yards has also been demonstrated in more recent US-based studies.3
Flooding may also have impacts on the mental health of individuals, both in the short term and the long term. The flood itself can be a cause of significant stress, and stress during the recovery period may also lead to mental health issues. Some of the psychosocial issues that may arise after a flood include grief or bereavement leading to depression, economic problems, behavioral issues in children, posttraumatic stress, increased substance use or abuse, increased domestic violence, and exacerbation of preexisting mental health conditions.5
Overall, extreme flood events have significant impacts on human populations. We sought to examine adverse health outcomes and policy implications that may be related to an extreme flooding event using the Health Impact Assessment Framework. Specifically, we sought to determine the types of health effects associated with a severe nontropical flood event and how that compared with a similar period of time without a flood, as well as determine whether there were any policy changes that could result in fewer health effects during a similar future event. For this study, we focused on those impacts in a single county, using a specific event. To place the human health impacts in context, it is helpful to understand the localized magnitude of the flooding.
From April 29 to May 3, 2014, the Escambia County area of Florida received an enormous amount of rainfall. We identified 17 different weather stations with precipitation records for each of the 5 days in the 2014 event (see Supplemental Digital Content Figure 1, available at http://links.lww.com/JPHMP/A330). From these 17 data points, we used inverse distance weighting to interpolate a continuous surface of estimated rainfall across the county for each day (see Supplemental Digital Content Figure 2, available at http://links.lww.com/JPHMP/A331). We considered Kriging for the spatial interpolation, but with relatively few recording stations (n = 17), inverse distance weighting produced maps that were more interpretable, while Kriging did not eliminate or substantially minimize the relative concentration of high estimates near the recording stations.6 For the Pensacola Airport, the longest and most complete precipitation record in the county, we computed long-term (1960-2014) descriptive statistics of daily and monthly rainfall totals for only the months of April and May, including mean and maximum monthly precipitation and the percentile value of each. This approach allowed us to characterize the historical climatological significance of the 2014 flood.
Extreme rainfall and flooding
For April 29, the maximum estimated rainfall was concentrated in the city of Pensacola near the International Airport and generally in the southeastern part of the county. The peak rainfall estimates range up to 15.5 in, which is an underestimate due to the known failure of the airport rainfall-monitoring station for some period on the afternoon of April 29. In spite of this equipment failure, the recorded rainfall at the airport station for April 29 was 15.5 in, the highest single-day rainfall total in the airport station record, nearly 9 in greater than the second highest daily total since 1960.
For the same day, rainfall estimates in the central and southwestern parts of the county were less than 3 in. Thus, the bulk of the severe precipitation on April 29 was spatially confined to a relatively small area around Pensacola and the airport, including zip codes 32514, 32504, 32503, 32505, 32502, and 32501. On April 30, the highest rainfall totals (ranging to 18.9 in) were in the extreme southwestern part of the county and in a belt circling the northern part of Pensacola, including the central portion of the county. Relatively low rainfall totals were recorded in the south-central and far northern parts of the county. Much of the city of Pensacola received less than the center of the county, but rainfall totals in the city still ranged from 5.5 to 7.5 in. For this day, the main rainfall was concentrated in zip codes 32533, 32506, and 32507. Rainfall totals on May 1 to 3, 2014, ranged from about ½ in to 1 in in different parts of the county. While this is a comparatively small amount, the rain was falling on very saturated ground, which exacerbated flooding.
For the purpose of this study, the extreme rainfall event was defined as April 29 to May 3, 2014. This multiday event resulted in the wettest April-May period in the historical record and was nearly 11 in greater than the second highest total (2005).
As a result of this tremendous rainfall, 392 road damage reports and 15 bridge damage reports were filed with Escambia County. Much like the rainfall totals, damages to roads and bridges were abundant and widespread in the southern part of the county. Conversely, roads and bridges in the central and northern parts of the county were essentially unaffected. Major reported transportation failures included the partial or total destruction of a 2-mile stretch of US 90 (Scenic Highway) in the southeastern part of the county and the destruction of Old Corry Road bridge over Jones Swamp in West Pensacola (south-central part of the county). Although flooding was widespread, we could not find any data source(s) from which to map or specifically confirm the spatial extent of standing water during and after the event.
A sewage lift station is a pump that moves human waste toward a treatment facility. The failure of a sewage lift station is normally an unusual event, but flooding emergencies create conditions in which such stations can be inundated with floodwaters (lose electrical power) and subsequently stop pumping. In Escambia County, some sewage lift stations failed as a result of this flooding event. In some cases, the electrical panel for the sewage lift station was flooded, which led to a loss of power and pumping ability. This created a sewage backup that led to many overflows from surrounding manholes and other connections. We obtained records for 357 individual sewage lift stations in Escambia County. These included primarily public (Emerald Coast Utilities Authority) lift stations and one private lift. No lift stations were located in zip codes 32535, 32568, or 32577 (northern rural half of the county). Thus, all lift station failures and related impacts were concentrated in the southern half of the county. Of the 357 stations, 31 failed during the 5-day period, giving a countywide lift station failure rate of 8.7% (see Supplemental Digital Content Figure 3, available at http://links.lww.com/JPHMP/A332).
Florida data sources used to characterize health impacts
The Florida Agency for Health Care Administration has been collecting hospital discharge and emergency department (ED) data since 1988 and 2005, respectively. These data sources contain a detailed record of each hospitalization and ED visit, and each record lists the primary and contributing diagnoses, patient demographics, and billing information (such as zip code). Agency for Health Care Administration data were used for the following health conditions: injuries from flooding and cleanup activities, asthma and other respiratory diseases, and mental health conditions. We included primary and all secondary diagnoses among Florida residents for the health conditions of interest.
We included hospital and ED visits for all-cause injury (unintentional or undetermined cause) among Florida residents. International Classification of Diseases, Ninth Revision (ICD-9) and E-codes used for defining injuries included
* fractures, dislocations, sprains, and strains (800-848);
* intracranial injuries (850-854);
* internal injuries (860-869, 900-904, 950-957);
* open wounds (870-897);
* superficial injuries/contusions (910-924);
* crushing injuries (925-929);
* burns (940-949);
* other/unspecified injuries (959); and
* unintentional/unspecified causes of injury (E800-E848, E850-E869, E880-E929).
Hospitalizations and ED visits for chronic respiratory problems can be considered indicators of poorly controlled disease rather than of total prevalence or incidence. However, changes in the rate of ED visits and hospitalizations can be used as a proxy to track changes in the severity of these diseases over time. Adhering to proper medication regimens can be more difficult during times of natural disaster. The ED visits and hospitalizations for respiratory diseases were based on either a primary or a secondary diagnosis of the following ICD-9 codes:
* Diseases of the respiratory system (460-519).
We relied on ED visits only for mental health and behavioral disorders. Only principal diagnoses were considered to improve specificity. We considered the following conditions that may be associated with weather-related impacts:
* Organic psychotic conditions (ICD-9: 290-294)
* Depression (ICD-9: 311)
* Stress-related disorders (ICD-9: 308-309)
* Substance-related disorders (ICD-9: 291-292, 303-305)
* Neuroses (ICD-9: 300)
* Other psychoses (ICD-9: 295-299)
* All mental health conditions (ICD-9: 290-319).
We also examined data from Florida's Notifiable Disease Surveillance System, called Merlin. Merlin data were used for enteric diseases (ie, Campylobacter, Giardia, and Salmonella infections). We include all case classifications (confirmed, probable, and suspect) and only cases acquired in Florida among Florida residents.
Campylobacteriosis and salmonellosis are bacterial diseases commonly associated with foodborne transmission. Giardiasis is a protozoan disease commonly associated with waterborne transmission and contaminated water. Affected individuals often have mild or asymptomatic infections, and many cases do not seek treatment and are not reported.
All-cause mortality data were acquired from the Department of Health Bureau of Vital Statistics. This data system collects a variety of data on all deaths, including the underlying and contributing causes based on International Classification of Diseases, Tenth Revision (ICD-10) coding with up to 20 contributing causes of death currently available. All death certificates in which the manner of death code was not listed as suicide, homicide, or pending investigation were included regardless of the listed underlying or contributing causes of death.
Study design and analysis
We used an ecological study design to examine the health impacts associated with this flood event, comparing health outcomes of interest that occurred in Escambia County during and immediately after the flood event (impact period) to the same outcomes in a year without a flood event (control period).
Defining the control period
Using daily precipitation data for April and May covering the period 1960-2014, we calculated descriptive statistics, including mean, median, and standard deviation. We computed percentile values for daily observations and running 5-day sum. For both the daily observations (specifically April 30) and the 5-day totals (April 29 to May 3), the 2014 event ranked first among all April and May days since 1960. To identify the control (“normal rainfall”) period for comparison, we examined the daily and 5-day running totals for only the period April 15 to May 15, for only the most recent 10 years (2005-2014; limited by the availability of the health data). We selected 2008 as the control period. The April 15 to May 15, 2008, precipitation record from the Airport station is not well above average (<80th percentile of all comparable periods since 1960), nor is it zero (∼70% of all daily observations in the full record are zeros).
Defining the follow-up periods of interest for analysis
The follow-up periods of interest for each of these health outcomes is variable (see Table 1) due to the nature of the health outcome, and we defined 3 separate follow-up periods for analysis. For injury, we were interested in an impact period that encompassed postflood cleanup activities (including the 5-day event plus 14 days after the end of the event). Thus, we compared injuries that occurred during a 19-day total time frame (April 29 to May 17) in the impact (2014) compared with the control period (2008). For enteric diseases, the follow-up period included a longer postflood period (the 5-day event plus 30 days after the end of the event; April 29 to June 2) to account for variable incubation periods for our outcomes of interest. For mental health conditions and mortality, the impact period was similar (5-day event plus 30 days postevent). This allowed us to capture the mental health effects associated with loss of property, displacement, injury, and death of loved ones and to capture deaths considered both directly and indirectly related to the flood event. Finally, for asthma and other respiratory diseases, we included the 5-day event plus 60 days postevent (April 29 to July 2), as we were interested in the respiratory effects primarily from mold. This longer time frame allowed mold to develop, people to be exposed, and then develop symptoms associated with this exposure.
The unit of health impact analysis was the count or proportion of health care visits or disease reports related to a specific health condition during the impact or control period of interest. Measures were assessed for the county as a whole and separately for each zip code in Escambia County. This study was reviewed by the Department of Health IRB Committee. No direct interaction with study participants occurred as part of this project.
Proportions were compared between the impact (2014) and control (2008) period for certain health outcomes to look for any increases from the earlier year to the latter, for each zip code and for all zip codes combined depending on the outcome of interest. For ED visits and hospitalizations, the denominator was the total count of ED visits or hospitalizations during that same period and in that same geographic area. This was done to reduce the bias caused by increased utilization of EDs and hospitals over time. For mortality data, the denominator was the total population for the entire county for the impact and control periods. We used a difference of proportions test and a 2-tailed t test to identify significant differences in the proportions of hospital or ED visits for each cause category between the 2 periods.7 An important caveat is that each of these health outcomes has a different follow-up period of interest (Table 1).
Descriptive and comparative results were completed for the impact (2014) and control period (2008) for each of the health outcomes considered in this assessment.
During the impact period, there were 231 hospitalizations and 1970 ED visits related to injury in Escambia County during the follow-up period (5-day event plus 14 days postevent). These visits accounted for 10.9% of all hospitalizations and 24.3% of all ED visits during the exposure window. During the control period, there were 166 hospitalizations and 1665 ED visits, accounting for 8.2% of hospitalizations and 24.0% of ED visits during the follow-up period. Supplemental Digital Content Table 2, available at http://links.lww.com/JPHMP/A339, presents the proportions of visits by zip code, and Supplemental Digital Content Figure 4A, available at http://links.lww.com/JPHMP/A333, and Supplemental Digital Content Figure 4B, available at http://links.lww.com/JPHMP/A334, show the percent change in proportions from 2008 to 2014 for all-cause injury hospitalizations and ED visits. Supplemental Digital Content Figure 5, available at http://links.lww.com/JPHMP/A335, displays the percent change in proportions for all hospital and ED visits combined. The ED and hospitalization data for all-cause injury showed increased visits comparing 2008 (control year) with 2014 (flooding year) for the entire county and for many of the zip codes, with several of these differences being statistically significant (Table 2).
During the follow-up period (5-day event plus 30 days postevent), 254 deaths were reported in the impact period and 300 in the control period countywide. This yielded a significantly lower crude mortality rate during the impact (84.0 deaths per 100 000) compared with the control period (110.1 per 100 000, P < .05). Few differences in underlying causes of death were noted during the 2 periods, except that there was a greater proportion of deaths due to injury during the impact than control period (7.4% vs 4.0%). Because of the small numbers, differences in all-cause mortality were not assessed by zip code.
Asthma and other respiratory effects
For respiratory diseases, we were interested in the 5-day event plus a 60-day postevent follow-up. Countywide, 2285 (30.7%) respiratory hospitalizations and 5163 (18.3%) ED visits occurred during the impact period (2014); while 1909 (28.3%) hospitalizations and 3953 (16.6%) ED visits occurred during the control period (2008). Both ED visits and hospitalizations were greater after the flood event than during a non–flood control period countywide. This same trend held true for many zip codes, especially for respiratory-related ED visits (see Supplemental Digital Content Table 2, available at http://links.lww.com/JPHMP/A339; Supplemental Digital Content Figure 6A, available at http://links.lww.com/JPHMP/A336; Supplemental Digital Content Figure 6B, available at http://links.lww.com/JPHMP/A337; and Supplemental Digital Content Figure 7, available at http://links.lww.com/JPHMP/A338), with some differences being statistically significant.
Few cases of the enteric diseases of interest were reported during the follow-up periods (5-day event plus 30 days postevent): 17 cases during the impact period and 10 cases during the control period. Among all cases, 55.6% were salmonellosis, 40.7% were campylobacteriosis, and 3.7% were giardiasis. The number of enteric diseases was too low to analyze by zip code.
Mental health referrals
For mental health–related ED visits, 340 (2.2% of all ED visits) occurred during the impact period and 295 (2.3% of all ED visits) occurred during the control period. The majority of ED visits for both periods were for substance-related disorders (28.7%) and neuroses (23.8%).
Overall, these data show a countywide increase in the proportion of ED visits and hospitalizations for injuries and respiratory conditions, which could be flood related. This was a historic flood event, with severe impacts for the Pensacola area that lasted many months. The Health Impact Assessment Framework allowed us to examine how flooding overwhelmed public and private infrastructure, resulting in environmental and public health impacts. The data on hospitalization and ED visits for multiple health outcomes at the zip code level showed mixed results. Some Escambia County zip codes showed significant increased proportions of hospitalizations and ED visits that may have been associated with flood impacts, and other zip codes showed nonsignificant decreases. That conclusion is based on a descriptive analysis comparing the time period of the flood (2014) to the same days during a control year (2008). For the county as a whole, there was a significant increase in the proportion of ED visits and hospitalizations due to respiratory problems and injuries. For other health impacts that could be caused by flooding, there were simply not enough cases reported in Escambia County to determine whether there was a measurable change. However, if future flooding occurs in a larger metropolitan area, then it may be feasible to consider such impacts.
This analysis was limited by the small scale (ie, 1 county) of the flood event and reduced sample sizes for the health outcomes of interest that resulted. We considered a variety of other health outcomes; however, many of these outcomes had no observations during the follow-up periods for either the impact or control years. For example, we considered carbon monoxide poisoning, but only one case was reported during the impact and none during the control period. In addition, we considered vector-borne and other enteric diseases (eg, vibriosis and cryptosporidiosis) but no cases were reported during the impact or control follow-up periods.
Other limitations relate to the types of surveillance systems available. In Florida, there is no surveillance system available for mental health visits, nor good estimates of the prevalence of these conditions. Therefore, we relied on ED visits as a proxy. Merlin is a passive surveillance system, placing the burden of reporting on health care providers and incomplete reporting of cases may have occurred. In addition, the relatively mild nature of many enteric diseases and the availability of over-the-counter medications means that many cases were likely not reported in Merlin. Finally, the use of ED and hospitalization data to characterize widespread public health impacts is imperfect because it captures only a portion of the burden (typically the most severe cases). Furthermore, the zip codes used in this analysis were collected by Agency for Health Care Administration for billing purposes and may not represent where the persons were actually exposed.
As a final limitation, there are competing needs after a natural disaster that further limit our ability to analyze associated health effects. For example, many of the resources that are utilized to collect health and environmental data in Florida are the same resources that are deployed for disaster response (eg, health department employees and other county officials), and priorities are different in the aftermath of such an event.
Implications for Policy & Practice
* One of the findings of this analysis was the failure of sewage lift stations during this flooding event. To reduce the number of sewage lift stations that fail (stop pumping), it is recommended that electrical panels be lifted as high as reasonably possible on the structures that house the lift stations. It is not known how many of the failures during this event could have been prevented by such actions, as that would depend on the height of the floodwaters relative to the location of the electrical panel. However, it is reasonable to assume that raising the electrical panels will result in a reduced likelihood of inundation during future flood events, and this could potentially result in continued operation of lift stations during natural disasters.
* The benefits to the community are reduced exposure to human sewage on the ground and reduced sewage in the surrounding environment (which also has benefits for wildlife and potential recreational activities).
* Additional policy recommendations include the use of proper flood cleanup procedures recommended by the Centers for Disease Control and Prevention and the proper disposal of flood-damaged materials into adequately managed Construction and Demolition Landfills. See the longer report available on the Florida Environmental Public Health Tracking Web site at http://www.floridatracking.com.