Burden of Disease and Health Status Among Hurricane Katrina–Displaced Persons in Shelters: A Population-Based Cluster Sample

Introduction 

Results 

Two clusters had inadequate sample size and had to be excluded from the data analysis. The resulting 18 clusters yielded 549 head-of-household respondents. Among these, 50 (10.4%; design effect=3.12) declined to participate, leaving 499 adult respondents, who as heads of household represented 1,455 individuals. The age distribution of the sheltered evacuee population (Figure) tended to be less uniform than the US population distribution, with peaks at young and middle-age groups. Table 1 describes the demographic of the sample, with the percentages reflecting the calculated influence of design effect and not simple percentages. The majority was female, and more than two thirds of the sheltered residents were single, widowed, or divorced. The racial mix of the sample matched the racial profile of the city of New Orleans.13 Nearly 40% were receiving some form of benefits; of those, nearly one quarter were Medicaid recipients. More than half (52.9%) were unemployed or underemployed, and 66.2% (design effect=2.18) rented rather than owned their homes (inversely proportional to a national home ownership percentage of 68.9%).14 The percentage of the sample without health insurance was more than 3 times that of the national average (47.0% versus 15.7%),15 and nearly one third lacked a regular health care provider.

View Large Image Download to PowerPoint

Figure. Age-sex population pyramid of 1,455 shelter residents compared to US population.

Table 1. Characteristics of 499 respondents.

	Characteristics	Categories	No. (%)	Design Effect
	Sex	Female	291(57.6)	3.49
		Male	198(39.1)	2.68
		Not answered	10(3.2)	2.78
	Race	Black	360(76.4)	4.11
		White	90(15.2)	5.03
		Hispanic	25(4.8)	1.68
		Asian	6(1.4)	1.13
		Other	8(1.2)	1.21
		Not answered	10(1.0)	0.84
	Marital status	Single	259(52.5)	0.62
		Married	148(28.9)	0.78
		Divorced	61(12.7)	1.03
		Widowed	26(5.6)	1.91
		Not answered	5(0.4)	0.82
	Employment	Full time	229(46.6)	1.04
		Part time	54(12.0)	1.00
		Retired	58(9.2)	2.54
		Unemployed	155(31.7)	0.39
		Not answered	3(0.5)	1.38
	Health insurance before hurricane	No	224(47.0)	2.27
		Yes	269(51.9)	2.29
		Not answered	6(1.0)	2.58
	Health care provider before hurricane	No	158(33.2)	3.90
		Yes	337(65.9)	3.45
		Not answered	4(0.9)	1.34
	Receiving assistance/benefits before hurricane	No	267(56.9)	1.90
		Yes	209(38.5)	2.49
		Not answered	23(4.5)	0.76
	Type of assistance/benefits	WIC	32(6.0)	1.31
		Medicaid	117(23.2)	0.52
		Medicare	64(12.0)	0.69
		AFDC welfare	18(2.4)	2.21
		SSDI disability	64(11.5)	3.77

In a cluster sample (as opposed to a simple random sample), the respondents may be more similar to each other for a given variable or outcome measure than if they had been independently selected in a simple random sample. The design effect is a statistic that accounts for this level of intracluster correlation and is calculated for each variable or outcome measure. The statistical formula is the ratio of the variance in the cluster sample to the variance that would have been obtained from a simple random sample of the same size. This ratio is then multiplied by the sample size of a simple random sample to obtain the sample size for a cluster sample. A higher intracluster correlation results in a higher design effect and thus a larger sample size.

The prevalence of sheltered evacuees with known chronic disease is listed in order of frequency (Table 2). Hypertension, hypercholesterolemia, diabetes, and lung diseases were the most commonly identified chronic diseases. The prevalence of psychiatric and mental health illness was notably high as well. Among the sampled population, those reporting a medical history of stroke, HIV, dialysis dependence, or seizure disorder were rare.

Table 2. Prevalence of medical problems in 499 respondents, in order of frequency.

	Medical Problems	No.	Prevalence, % (95% CI)	Design Effect
	High blood pressure	170	34.8(30.4–39.2)	0.96
	High cholesterol	86	16.8(12.0–21.6)	1.87
	Diabetes	69	14.3(12.5–16.1)	0.29
	Psychiatric/emotional	70	13.5(10.4–16.6)	0.94
	Lung disease (including asthma)	66	12.9(8.1–17.8)	2.32
	Gastrointestinal or liver	50	11.6(6.5–16.7)	2.84
	Heart disease	48	10.4(6.9–14.0)	1.53
	Substance abuse	18	4.3(2.9–5.7)	0.54
	Dialysis	6	1.6(0.5–2.6)	0.83
	Sickle cell disease	5	0.8(0–1.9)	1.79
	Tuberculosis	3	0.6(0–1.4)	1.27
	HIV/AIDS	2	0.5(0–1.3)	1.42
	Any medical problems listed above	276	55.6(51.9–59.4)	0.63

Of the 276 respondents with diagnosed chronic disease, 224 (81.4%; design effect=0.19) were taking prescribed medications. The concern that displacement would interrupt their care prompted us to investigate whether the 224 diagnosed with chronic disease had arrived at the shelters with their medications (Table 3); at least 95 (48.4%) had not. However, when the survey concluded more than 2 weeks later, the percentage without medications had improved to 19.8%. Lack of a primary care provider was the sole factor associated with lacking medications at the survey (odds ratio 5.62; 95% CI 1.91 to 16.55; P=.002). In the multivariate regression adjusting for health benefits and assistance, male sex and lacking health insurance were risk factors associated with lacking medications for chronic disease on arrival at the shelter (Table 4).

Table 3. Lack of prescribed medication when arriving at the shelter and at survey by chronic medical disease in 224 respondents who took medicine for their medical problems.

	Medical Problems	No.	No., (%, Design Effect) of Lack of Medicine When Arriving at the Shelter	No., (%, Design Effect) of Lack of Medicine at Survey
	High blood pressure	151	61(47.3,0.65)	29(22.7,1.27)
	High cholesterol	76	25(42.2,0.64)	15(19.5,1.37)
	Diabetes	65	25(43.6,0.83)	8(11.1,1.36)
	Psychiatric/emotional	59	23(42.4,0.80)	11(19.2,1.54)
	Lung disease (including asthma)	55	28(58.8,1.28)	14(30.5,0.28)
	Heart disease	45	18(47.0,2.14)	6(12.2,0.49)
	Gastrointestinal or liver	43	22(55.2,1.42)	10(27.0,2.46)
	Any medical problems⁎	224	95(48.4,1.04)	39(19.8,1.29)

⁎ See Table 2 for a list.

Table 4. Factors associated with lack of prescribed medication when arriving at shelter, adjusting for receiving benefits or assistance among 206⁎ subjects with prescribed medication.†

	Variables	Categories	OR (95% CI)
	Sex	Male	1.58(0.96–2.59)
		Female	1.00
	Receiving assistance/benefits	No	1.50(0.71–3.19)
	Before hurricane	Yes	1.00
	Having health insurance	No	2.25(1.21–4.20)
		Yes	1.00
	Employment	Unemployed	0.68(0.37–1.24)
		Retired	0.35(0.16–0.78)
		Full time or part time	1.00

OR, Odds ratio.

⁎ Eighteen subjects were excluded because of lack of information on sex, assistance/benefits, health insurance, and employment. † R2=0.95.

Given the degree to which the evacuee population was exposed to environmental hazards during flight, a focus of disaster health services during the Hurricane Katrina response was assessment of their condition on arrival to the shelters. One third of the arriving evacuees had symptoms that would prompt immediate treatment interventions (Table 5). Among these, dehydration and generalized weakness, shortness of breath, chest pain, and injury were the most common presenting complaints of evacuees on arrival to the shelter. Nearly 1 in 10 evacuees arrived with chest pain. Approximately 15% of evacuees arrived with symptoms of communicable disease potential (fever, diarrhea) that might prompt shelter managers to segregate them from the general shelter population or to initiate infectious control mechanisms. Arriving evacuees with fever (87.9% versus 59.3%; P<.001; design effect=0.77), shortness of breath (76.8% versus 60.3%; P=0.031; design effect=0.77), and infected wounds (93.8% versus 65.1%; P=0.003; design effect=0.77) were more likely than others to seek immediate medical care for their symptoms. Evacuees without health insurance or those afflicted with chronic disease requiring medications (whether they had them or not) were nearly twice as likely to present with acute symptoms on arrival to shelters (Table 6). Those with comorbidities were also 3 times more likely to be acutely ill on arrival compared with those with no other conditions.

Table 5. Prevalence of symptoms in 499 respondents when arriving at the shelter, in order of frequency.

	Symptoms	No.	Prevalence, % (95% CI)	Design Effect
	Dehydration or general weakness	59	12.0(9.9–14.2)	0.49
	Shortness of breath	61	11.5(9.4–13.7)	0.51
	Chest pain	46	9.7(7.8–11.6)	0.46
	Injury	48	9.4(6.1–12.8)	1.48
	Any fever	40	8.0(5.2–10.7)	1.15
	Fever with cough	25	5.1(2.7–7.5)	1.35
	Fever with stiff neck/headache	21	4.2(2.5–5.9)	0.81
	Fever with rash	5	1.8(0.3–3.4)	1.53
	Diarrhea	34	6.9(5.6–8.1)	0.26
	Thoughts of self-harm	16	3.7(2.1–5.3)	0.80
	Thoughts of harming others	4	1.3(0.3–2.4)	0.96
	Infected wounds	7	1.1(0.2–1.9)	0.76
	Any symptoms listed above	170	33.4(27.0–39.8)	2.05

Table 6. Factors associated with acute symptoms when arriving at shelter in 493⁎ respondents, multivariate regression.†

	Variables	Categories	OR (95% CI)
	Having health insurance	No	1.83(1.10–3.02)
		Yes	1.00
	Known chronic disease ± medication on arrival to shelter	Has medication	2.60(1.98–3.43)
		Lacks medication	2.22(1.36–3.63)
		Not prescribed medication	1.25(0.64–2.43)
		No known chronic disease	1.00

OR, odds ratio.

⁎ Six individuals were excluded because of lack of information on health insurance status. † R2=0.8691.

Limitations 

We acknowledge limitations of the study to represent the larger displaced population that had no need for shelters or was displaced to shelters in other states. The shelter population is arguably unique and more vulnerable, representing the level of safety net that must be in place to maintain the health of the population as a whole. Cluster samples by their design have a tendency to select for respondents who have similar experiences and demographic profiles. This tendency is not a significant concern for this study, because the population and outcome measures of interest assume a similar health profile and demographic and similar experiences during the displacement; outcome measures chosen do not depend on specific comparisons. Our inability to analyze 2 clusters could enhance this design effect and miss unintended variability; they were not in the same locations. Our study had a short recall period that allowed for more precision. The sample does not reflect the evacuee population with special needs who have higher acuity and more labor-intensive long-term health problems.

Discussion 

There is a consensus among disaster medical personnel and epidemiologists that natural and human-generated disasters will become more frequent and intense and will affect greater numbers of people because of increased geographic vulnerability, climate change, poverty, and urbanization.16 Having a population-based understanding of the disaster displaced, including social and economic demographics, burden of disease, and baseline level of health care, is critical for a society to prepare and respond effectively. Anecdotal reports from the media and existing or ad hoc medical facility records are not sufficient to make intelligent focused decisions on the broader public health issues in acutely displaced populations. This epidemiologic study attempts to characterize the largest population ever displaced to shelters by a natural disaster in the United States.

Disasters affect the poor and vulnerable disproportionately; Hurricane Katrina’s effect on the Mississippi delta and the city of New Orleans unmasked the lack of reserve the less fortunate members of society have when facing a disaster of this magnitude. Our sample was representative of the larger displaced population because it closely resembled that of New Orleans: the majority was black (76.4%), with single heads of households that fell well below the national average for health care access (insurance, regular health care provider, having medications) and socioeconomic indicators (home ownership, limited income-generating activity, and reliance on benefits/assistance). For instance, home ownership of the sheltered evacuee population was 33.8% compared with 68.9% for the United States and 70.6% for Louisiana; 31.7% of the sheltered evacuee population were unemployed compared to national and Louisiana unemployment rates of 5.1% and 5.6%, respectively.17, 18

Also representative was the prevalence of chronic disease in the sample. The prevalence of hypertension among the sheltered evacuees (34.8%), although above the national average of 25.6%, was in line with the national average for that of blacks (36.9% men, 39.5% women); the prevalence of diabetes in our sample (14.3%) was similarly above the national average (9.4%) but more representative of the US black prevalence for diabetes (11.2%).19

The baseline lack of health care access and the burden of acute and chronic disease in displaced populations are critical issues to the effective delivery of health care during disaster response efforts. That nearly half our sampled population with known chronic disease was lacking treatment at arrival to shelters represents the kind of vital information that directly affects how disaster health providers prepare for and implement a health response. Stockpiled common medications for chronic disease can be readily available for shelter residents if this requirement is realized early in the response phase. In addition, the prevalence of known psychiatric illness further complicated by a recent stress event will necessitate access to early mental health services either in the shelter or in the local community. A network of skilled local providers can be in place in advance to address the needs of the recently displaced.

A population with a significant burden of chronic disease that has limited health care access at baseline and is lacking medications will be at risk for acute exacerbations of those diseases. In addition, the risk of communicable diseases significantly increases when living quarters become more densely populated. Shelter managers would benefit from instituting a registration mechanism that has both a medical intake (for chronic disease) and a screening triage (for acute disease) component. Real-time information on the health of the shelter population would allow the shelter health providers to implement communicable disease prevention strategies, manage resources for diagnostic and treatment interventions more effectively, and plan for contingencies. In the international arena, humanitarian relief agencies are following evidence-based, consensus-driven guidelines that rely on applied epidemiology, such as this study, to respond to displaced populations in emergencies.20

In summary, the challenges to meeting the health needs of marginalized displaced populations are immense. A population-based approach that provides an evidence base for preparation and response will help ensure adequate access for a given burden of disease and address the level of disease acuity as it arises in the immediate postdisaster period.