Introduction

Urbanization increases disaster risk because of the high concentrations of population and assets that are vulnerable when a major disaster occurs. In lower- and middle-income countries, continual social and demographic change also complicate issues of urban disaster risk management. Concentrations of economically poor slum residents, especially in cases where slum settlements expand to hill slopes or flood plains, increase the hazard profile of the city. Lack of urban planning capacity, implementation, and enforcement of building codes and poorly regulated development are common problems across many urban centers in lower- and middle-income countries that increase disaster risk [1] In Haiti, the capital of Port-au-Prince is known to be at risk for hydro-meteorological hazards, in particular hurricanes, with little concern for geologic hazards because of the lack of significant historical earthquakes [2]. This changed with the magnitude 7.0 earthquake that struck on January 12, 2010, and devastated the city with a reported 222,750 deaths, 300,000 injured, 1.5 million displaced, and more than 3 million affected. Relief and recovery operations were historic, and by October 2010, the response was one of the largest in history with an estimated cost of $4.5 billion [3]. This study sought to estimate mortality from the 2010 Haiti earthquake and identify risk factors for mortality in major earthquakes in low-income settings.

Methods

A cross-sectional cluster survey of the earthquake-affected population in metropolitan Port-au-Prince was conducted in January 2011 to assess the impact of the 2010 earthquake on Haitian households and to characterize perceptions of humanitarian assistance at one year postearthquake; the mortality data presented in this paper were collected as part of this larger study. The proportion of the affected populations residing in camps and neighborhoods could not be accurately estimated from available government and United Nations information; thus, a stratified design was used to enable comparison between camp and noncamp (neighborhood) populations. A stratified cluster survey with 60 clusters of 20 households, including 30 camp and 30 neighborhood clusters, was used to characterize earthquake impact and receipt of humanitarian assistance in camp and neighborhood populations. Sample size was calculated based on the broader study objective of comparing the status of camp and noncamp populations. Calculations were based on the most conservative prevalence rate of 50% and a hypothesized difference of 10% between camp and noncamp populations for measures of impact, inadequate living conditions, and receipt of humanitarian assistance; other parameters included 80% power, alpha = 0.05, and an anticipated cluster sample design effect of 1.5. Households were included in the survey only if the dwelling had been damaged, income or livelihood affected, or a household member was injured or died as a result of the earthquake. The survey instrument was developed in English and translated into Creole and conducted by trained Haitian nationals.

Clusters were assigned using probability proportional to size (PPS) sampling. Camps were sampled using the PPS methodology from a list of planned and spontaneous settlements obtained from the Camp Coordination and Management Cluster [4]. The starting point in each camp cluster was a randomly selected intersection, and every third shelter was sampled until 20 households were completed. For neighborhoods, cluster allocation was based on a remote sensing building damage assessment [5], under the assumption that the number of moderately and heavily damaged residential structures was proportional to the earthquake-affected population within an administrative unit. The proportion of moderately to heavily damaged residential buildings from the damage assessment and 2009 population estimates were used to estimate the affected population within each commune (similar to a district) in Port-au-Prince. Clusters were then assigned to communes proportionate to the estimated affected population. Within each commune, clusters were assigned to sections (similar to a subdistrict or neighborhood) proportionally based on 2009 population estimates [6]. In each section, geographic coordinates were randomly selected and the nearest intersection was used as the cluster start. Then, a randomly selected number and direction were generated among the streets or pathways meeting the intersection to select one for the survey start location. From there, every third residential entrance was sampled; in buildings with multiple households, one household was randomly sampled. Data were collected using questionnaire-based interviews by a team of local Haitian interviewers from Port-au-Prince. Any adult household member was eligible to serve as a respondent; however, the head of household usually acted as the respondent if he or she was present at the time of the interview. If absent, second priority was given to the head of household’s spouse or the caretaker of young children. Only pre-earthquake household members were used for mortality estimation; individuals that joined the household after the earthquake were excluded because they were originally part of other households, and a separate interview would have been needed. This would have been difficult given that many of them were children. The response rate was high (>95%) in households in which adults were present; the number of households in which no adult was present and those that did not meet inclusion criteria were not recorded.

Data analysis was performed with Stata version 12 (College Station, TX) using simple logistic regression, multivariable logistic regression, and mixed-effects logistic modeling methods, as well as chi-squared and t-tests. The design effect was calculated at 1.21; consequently, odds of mortality and related confidence intervals (CIs) were adjusted to reflect the cluster survey design. Multiple variable logistic regressions were conducted with a mixed-effect model (xtmelogit command of Stata). Mixed-effects simple and multiple logistic regressions were used to measure the total unadjusted and adjusted odds of mortality. The final model was selected based on Akaike Information Criterion and Pearson’s goodness of fit test.

The study was certified exempt by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board and approved by the Haitian Ministry of Public Health and Population. This work was supported by the Johnson and Johnson Foundation and the Johns Hopkins Bloomberg School of Public Health Center for Refugee and Disaster Response.

Results

In January 2011, when the survey was conducted, the 1,197 households surveyed had a combined population of 6,696 individuals, of which 6,547 (97.9%) were reported as household members on the day of the earthquake. A total of 149 individuals were born or moved into households after January 12, 2010 and were not included in the earthquake-exposed population that served as a denominator for mortality calculations. Descriptive characteristics of the sample population are summarized in Table 1. The average household size was 5.3, and household size was similar in the camp and neighborhood populations (p = 0.474). The population was 52.5% female, and camps had a significantly higher proportion of females than neighborhoods (53.9% female vs. 51.0% female; p = 0.017). Age composition of the sample population was as follows: 0–17 years, 35.2%; 18–49 years, 56.44%; 60+ years 8.39%. The average age was 26.1 years (standard deviation [SD] = 16.1), and camp residents were significantly younger than neighborhood residents (23.8 years vs. 27.4 years; p < 0.001). The highest educational level attained by any household member (categorized as none, primary school or less, secondary school, or higher education) showed significant differences between camps and neighborhoods, with camp residents having lower education levels (p < 0.001). As compared to neighborhood residents, camp residents were significantly less likely to own their own home (53.2% vs. 22.7%) and more likely to live in crowded conditions (24.9% vs. 39.1%, defined as ≥4 people/room) prior to the earthquake, suggesting that households of lower socioeconomic status were more likely to be displaced.

Table 1 Comparison of baseline characteristics by postearthquake residence location
Full size table

A total of 159 deaths were reported in the pre-earthquake population between the January 12, 2010 earthquake and the January 2011 survey (of 6,536 individuals with reported survival status). Of these, 153 were direct deaths that occurred on the day of the event (95%) or within a month as result of earthquake-sustained injuries (5%). The overall earthquake mortality rate is estimated to be 24 deaths/1,000 (CI: 20–28) (Table 2). Elevated mortality was observed among households residing in camps, at 28 deaths/1,000 (CI: 22–34), as compared those in neighborhoods where the death rate was 19 deaths/1,000 (CI: 15–25) (p = 0.030). Age- and sex-specific mortality rates are presented in Figure 1. No significant difference in mortality was observed between males and females (p = 0.786). When assessed by age, mortality rates were as follows: children 0–17 years, 16 deaths/1,000 (CI: 11–22); adults 18–49 years, 24 deaths/1,000 (CI: 20–31); older adults 50+ years, 44 deaths/1,000 (CI: 28–64).

Table 2 Mortality estimates and odds of death by individual and household characteristics
Full size table
Figure 1
figure 1

Age and sex specific morality rates.

Full size image

Variables related to socioeconomic status, such as highest household education level attained, whether or not the pre-earthquake home had multiple levels, and crowding, were associated with mortality outcomes. The highest mortality rate was found among households in which no member had completed primary school (38 deaths/1,000, CI: 19–67), although as a categorical variable, household education was not significant in predicting mortality. Crowding can serve as a proxy for socioeconomic status and was associated with increased mortality risk; however, this finding was of marginal statistical significance (p = 0.082). Residence in a multistory building was associated with increased risk of death; mortality rates among households in single and multistory residences were 19 deaths/1,000 (CI: 15–24) and 33 deaths/1,000 (CI: 26–42) (p < 0.001).

In simple logistic regression, age and housing characteristics were significantly associated with increased mortality risk (Table 3). In the multiple logistic regression model, age and housing characteristics remained significantly associated with increased odds of death. As compared to children 0–17 years, middle-age (18–49 years) and older (50+ years) adults were 1.34 (CI: 0.89-2.01) and 2.50 (CI: 1.41-4.39) times more likely to experience death, respectively. Members of households in residing in multistory buildings had a mortality risk of 1.98 (CI: 1.38-2.83) times greater than residents of single level buildings. Crowding was a marginally significant predictor of mortality (p = 0.082) but was included in the final model based on log rank tests. Within category comparisons indicated that odds of death were 1.7 times (CI: 1.06-2.69) higher in populations with more than four people per room as compared to populations with two or fewer people per room (p = 0.027). The mortality rate among populations displaced to camps was 1.5 (CI: 0.95-2.38) times greater than populations in neighborhoods; however, this finding was only marginally significant (p = 0.080).

Table 3 Crude and adjusted Odds ratios for mortality
Full size table

Mortality projections using survey-determined rates and methods approximating those used in the Kolbe et al. [7] and Schwartz et al. [8] studies are presented in Table 4. Using direct extrapolation, estimated total mortality for the affected population was 74,190 (range: 63,061-86,555). Extrapolation by damage level yielded a total estimate of 63,901 deaths (range: 49,033-81,862); a significant difference was observed between mortality rates of households living in homes that were mildly or moderately damaged compared to those that were significantly damaged or destroyed (p < 0.001)a.

Table 4 Earthquake mortality projections*
Full size table

Discussion

There is substantial variation in earthquake mortality outcomes [9]. Low levels of economic development have been associated with higher earthquake mortality, suggesting that poorer countries face increased risk due to a variety of characteristics of the built environment [9–11]. Building collapse accounts for almost all earthquake-related death, with a majority of deaths occurring indoors [9, 11–13]. Construction materials that have been associated with increased mortality or injury risk include unreinforced masonry[14], mud and stone walls [15], concrete [16], panel construction [17], and wood [18–20].Worldwide, the greatest risk of collapse and resulting deaths is from unreinforced masonry structures (including mud brick) [21]. Other features of the built environment and individual location (at the time of the earthquake) associated with increased risk for earthquake mortality include being on the ground floor [22], or upper floors [11, 23] of multistory buildings, earthquake intensity, and distance to the epicenter [24–27]. Individual-level characteristics associated with increased risk for earthquake mortality and injury include both age and sex. Higher rates of death among older populations are common [17, 23, 26–33] and several studies also show that children face an increased risk of death [15, 25, 27, 31]. Female sex was significantly associated with increased risk of death in at least three earthquakes [10, 29, 32]; however, a similar number of studies found no significant difference in mortality by sex [24, 25, 28, 32].

This analysis is one of several attempts to characterize mortality following the 2010 Haitian earthquake. The results from the data collected in this study found an earthquake mortality rate of 24 (CI: 20–28) deaths per 1,000 in the sample population. This observation suggests that mortality was nearly four-fold lower than officially reported figures. At one year after the earthquake, UN revised mortality figures estimated that 222,750 deaths occurred [34]. With a population of 3.1 million in the affected areas [6], this equates to a mortality rate of approximately 72 deaths per 1,000. The methods by which the official mortality estimates were obtained are undocumented, and their accuracy has been questioned, with some critics suggesting they are inflated, in particular given the substantially lower casualty figures reported in the immediate aftermath [8]. However, our study could underestimate the total deaths because the sampling did not account for households with high or complete mortality that could not be sampled. Other scientific studies have estimated Haitian earthquake mortality at 111,794 (CI: 93,273–130,316) deaths during or immediately after the earthquake, with a mortality rate of 41 (CI: 35–48) deaths per 1,000 [7] and between 46,190 and 84,961 deaths, with a mortality rate of 22 deaths per 1,000 [8].

Extrapolation of survey-based rates to large urban populations in instances such as Haiti where damage and mortality are highly variable and geographically concentrated presents numerous methodological challenges, particularly in the context of widespread displacement and mortality. While data from three epidemiologic studies conducted to date clearly indicate lower levels of mortality than the official estimates, there are noteworthy differences in study methods that likely contribute to the observed variation in mortality estimates. The Kolbe et al. study [7] used a pre-earthquake random sample of Port-au-Prince households and then sought to follow up the population postearthquake, whereas this survey and the Schwartz study [8] intended to capture representative samples of the affected population. The extrapolation methods also differed with Kolbe et al., using direct extrapolation of the survey rate to the affected population, whereas the Schwartz et al. and the present study determined mortality rates by building damage level to account for variation in impacts.

While both extrapolation methods were applied in the current paper, the authors believe the methodology that takes damage level into account is more accurate. When damage levels are accounted for, the results of the present study yielded an estimate of 63,901 deaths (range: 49,033-81,862), which is consistent with the Schwartz estimate of 46,190 to 84,961 deaths. Extrapolating from structure damage levels could underestimate fatalities because there is no distinction within the “red” category (uninhabitable) between damaged but standing and collapsed buildings, despite structural collapse being the greatest predictor of mortality [20, 24]. While these values provide another independent projection of the number of earthquake fatalities, it is important to note that other more advanced methods of mortality modeling, including those that incorporate Geographic Information Systems (GIS) and which are based on a larger sample size, would likely yield more accurate estimates of total mortality.

Age was important in predicting the odds of death in this study, with increased mortality risk observed among older adults, a finding common to many earthquake settings [17, 22, 25–32]. However, this observation is contrary to the Kolbe et al. study [7], which found that children were 5.8 (CI: 4.0-8.3) times more likely to have been killed in the earthquake than adults, and that children accounted for the majority of earthquake deaths. Unfortunately, no other findings on age or sex as a risk factor for death in the Haitian earthquake have been reported, which limits further comparison. In this study, the population displaced to camps had a higher mortality rate compared to those remaining in neighborhoods; however, the statistical significance of this finding was marginal. Number of levels in the housing structure and crowding, a proxy for both socioeconomic status and population density, were both associated with mortality risk. In our analysis, residential construction characteristics such as roof or wall construction material, housing type (apartment, detached, or attached home), and number of levels, had varying levels of association with mortality in simple regression, and only the number of levels was included in the final multiple regression model. Associations between single versus multilevel homes, crowding, and increased mortality risk support the hypothesis that earthquake vulnerability is related to the built environment and that households of lower socioeconomic status with high levels of population density face increased risk of both displacement and death.

The high levels of mortality observed in Haiti were largely attributable to substandard and unsupervised construction practices [35]. Given that most mortality and injury following earthquakes is mediated by building damage, rebuilding structures with improved earthquake-proof construction is critical for future disaster preparedness efforts. Substantial investment in infrastructure, urban planning, and building code enforcement are required if reconstruction efforts are to yield sound earthquake-resilient structures. However, mitigating the risk posed by poor construction is especially challenging and perhaps unfeasible in resource-poor settings where retrofitting buildings and code enforcement are both cost-prohibitive and often not a priority for local authorities and at risk populations.

Limitations

Sample size calculations for the survey were based on initial mortality reports from the Haitian government; however, observed mortality was three-fold lower than rates used for sample planning. A larger sample size would have improved exploration of risk factors, detection of significant differences, and led to greater precision. One limitation of cluster survey designs in earthquake settings is that areas of concentrated deaths, such as collapsed apartment buildings, are likely to be missed, which could result in the underestimation of mortality. Another limitation that likely contributed to the underestimation of mortality is survivor bias, in which households where no members survived or where few members survived were not included in the retrospective mortality estimates. Furthermore, it is possible that the sample was not representative of the affected population because replacement sampling was used, when households that were not at home were systematically excluded. With respect to assessing risk factors for death, improved descriptions of buildings in which deaths occurred and location of individuals at the time of the earthquake would have been useful in analyzing the built environment’s mediating effect on mortality outcomes. However, the long recall period, potential inaccuracies in reporting relevant environmental details for the pre-earthquake residence location of all household members, and the inability to systematically confirm reported building material data for the majority of individuals prohibited the collection of more detailed information on the built environment and subsequent risk analysis. Finally, more research is needed to investigate how built environment characteristics can or cannot serve as proxies for socioeconomic status; a clear understanding of these relationships would have enabled stronger conclusions.

Conclusion

Estimation of mortality and morbidity in natural disasters and conflicts is controversial, and inconsistencies in the reported impacts of these events are not uncommon. The importance of death and injury counts for humanitarian response planning, documentation of human rights violations, and longer-term public policy responses to emergencies is increasingly accepted. However, the methods used to arrive at these estimates are often poorly documented, widely variable, and often the subject of intense debate criticism. Mortality in the 2010 Haitian earthquake is no exception. Results of epidemiologic surveys suggest a substantially lower mortality burden than figures reported by the Haitian government and United Nations. Findings from this study indicate an earthquake mortality rate of 24 deaths per 1,000 (CI: 20–28) in the metropolitan Port-au-Prince survey population. The most important predictors of mortality risk were older age, residence in a multilevel structure, high levels of crowding, and location at the time of the earthquake. A better understanding of risk factors associated with earthquake mortality and the risk attributed to the built environment is essential for humanitarian response planning and to implement risk reduction strategies that are feasible in resource-poor settings.

Consent

Respondents were read a statement describing the survey and requesting their participation, and oral consent was obtained prior to initiating the interview.

Endnotes

aMinistry of Transportation and Public Works Building Damage Assessment result for the residence as reported by respondents in the household survey.