Article Text

Download PDFPDF

Maternal exposure to hurricane destruction and fetal mortality
  1. Sammy Zahran1,2,
  2. Ian M Breunig3,
  3. Bruce G Link4,
  4. Jeffrey G Snodgrass5,
  5. Stephan Weiler6,
  6. Howard W Mielke7
  1. 1Department of Economics, Colorado State University, Fort Collins, Colorado, USA
  2. 2Department of Epidemiology, Robert Wood Johnson Health and Society Scholar, Columbia University, Mailman School of Public Health, New York, New York, USA
  3. 3Pharmaceutical Health Services Research, University of Maryland School of Pharmacy, Baltimore, Maryland, USA
  4. 4Department of Epidemiology, Columbia University, Mailman School of Public Health, New York, New York, USA
  5. 5Department of Anthropology, Colorado State University, Fort Collins, Colorado, USA
  6. 6Department of Economics, Colorado State University, Fort Collins, Colorado, USA
  7. 7Department of Pharmacology, Tulane University, School of Medicine, New Orleans, Louisiana, USA
  1. Correspondence to Dr Sammy Zahran, Department of Economics, Colorado State University, C132A Clark Building, Fort Collins, CO 80523-1771, USA; szahran{at}colostate.edu

Abstract

Background The majority of research documenting the public health impacts of natural disasters focuses on the well-being of adults and their living children. Negative effects may also occur in the unborn, exposed to disaster stressors when critical organ systems are developing and when the consequences of exposure are large.

Methods We exploit spatial and temporal variation in hurricane behaviour as a quasi-experimental design to assess whether fetal death is dose-responsive in the extent of hurricane damage. Data on births and fetal deaths are merged with Parish-level housing wreckage data. Fetal outcomes are regressed on housing wreckage adjusting for the maternal, fetal, placental and other risk factors. The average causal effect of maternal exposure to hurricane destruction is captured by difference-in-differences analyses.

Results The adjusted odds of fetal death are 1.40 (1.07–1.83) and 2.37 (1.684–3.327) times higher in parishes suffering 10–50% and >50% wreckage to housing stock, respectively. For every 1% increase in the destruction of housing stock, we observe a 1.7% (1.1–2.4%) increase in fetal death. Of the 410 officially recorded fetal deaths in these parishes, between 117 and 205 may be attributable to hurricane destruction and postdisaster disorder. The estimated fetal death toll is 17.4–30.6% of the human death toll.

Conclusions The destruction caused by Hurricanes Katrina and Rita imposed significant measurable losses in terms of fetal death. Postdisaster migratory dynamics suggest that the reported effects of maternal exposure to hurricane destruction on fetal death may be conservative.

  • FERTILITY
  • DISASTER RELIEF
  • Environmental epidemiology
  • FETAL
  • MATERNAL HEALTH

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

The majority of research documenting the public health impacts of natural disasters focuses on the well-being of adults and their living children.1–5 But could it be that negative effects also occur in the unborn exposed to disaster stressors when critical organ systems are developing and when the consequences of exposure are particularly lethal? We explore this question using large samples of objectively ascertained data on infant survival collected before and after two devastating hurricanes in Louisiana, Hurricanes Katrina and Rita (‘HKR’) (29 August and 24 September 2005).

Katrina was the costliest natural disaster in American history, and Rita was the fourth most intense hurricane ever recorded.6 ,7 HKR not only destroyed vast amounts of property and infrastructure, but also imposed enormous costs in physical and/or psychological stress and injury.5 Compared with residents in Alabama and Mississippi, residents in damaged parishes in New Orleans were more likely to suffer the death of a loved one, property and income loss, housing adversity and/or physical illness or injury leading to higher prevalence of DSM-IV anxiety-mood disorders.8 Also, unlike other disasters where symptoms of psychological distress and injury retreat in time, unresolved hurricane-related stressors in New Orleans led to increases in serious mental illness, post-traumatic stress disorder and suicidal ideation long after HKR occurred.9

Pregnant mothers exposed to hurricane-related stressors are known to suffer negative labour outcomes,10 ,11 and scholars have questioned whether unborn children of pregnant mothers exposed to postdisaster wreckage and social disorder suffer negatively.12–20 While postdisaster negative effects on fetal health are partly explained by compromised maternal access to healthcare before and after a hurricane event,21 ,22 researchers postulate that disaster-induced trauma and distress experienced by pregnant mothers may impair the development of an unborn child.23 Observational studies have found associations between maternal stress and newborn health.24–29 Studies have linked increased risk of spontaneous abortion, preterm births, low birth weight and other negative birth outcomes with mothers’ perceived stress in pregnancy,30–32 maternal depression, anxiety and trauma,30–33 as well as physiological stressors such as poor maternal nutrition.34–36 Although the literature on pathways linking maternal stress with fetal development and birth outcomes is far from convincing, a constellation of mechanisms involving neuroendocrine systems and maternal stress hormones are implicated.27 ,28 ,36

In a recent review of research, Zotti et al23 found uneven empirical evidence that negative fetal outcomes increase in maternal exposure to postdisaster destruction, maintaining that differences in research design, measurement and sample sizes may account for inconsistent findings across studies.17–19 Our study aims to extend understanding of the in utero costs of natural disasters by investigating whether maternal exposure to destruction caused by HKR increased the risk of fetal death. Methodologically, we exploit spatial and temporal variation in hurricane behaviour as a quasi-experimental design to assess whether fetal death is dose-responsive in the extent of hurricane damage. Data on births and fetal deaths are merged with Parish-level housing wreckage data, with the average causal effect of maternal exposure to postdisaster destruction captured by difference-in-differences analyses.

Methods

Infant birth and fetal death data

Infant birth and fetal death data are from Louisiana's Office of Public Health, State Center for Health Statistics, from 1 January 1999 to 31 December 2009. Our dependent variable, fetal death, includes both incidences of spontaneous abortion (gestation length <20 weeks) and fetal demise (>20 weeks of gestation). Of the 5194 fetal death events observed, 4566 occurred after 20 weeks of gestation, with 2945 fetal deaths occurring between 20 and 30 weeks.

Each birth or death record contains information on maternal, congenital and placental conditions, as well as other risk factors known to influence the likelihood of fetal death.38–40 Maternal conditions analysed include diabetes, hypertension, haemoglobinopathy and excessive bleeding. The presence of congenital abnormalities was analysed, as were placental variables: cord accident, abruption, ruptured membrane and uterine bleeding. Other potential risk factors included mothers’ race/ethnicity, advanced age (>35), prior preterm or small for gestational age birth, tobacco use and alcohol use during pregnancy. In addition to the above control variables, we screened other demographic characteristics including, but not limited to, maternal marital status and educational attainment. Summary statistics on fetal deaths by independent variables examined are reported in online supplementary table S1.

Housing unit damage data

The office of the Federal Coordinator for Gulf Coast Rebuilding at the Department of Homeland Security, in cooperation with the Federal Emergency Management Agency (FEMA), Small Business Administration, and Department of Housing and Urban Development, has compiled data assessing housing damage caused by HKR at the parish level.37 Damage estimation was determined from direct assessment by FEMA contract inspectors for all occupied housing units. A property is coded as incurring major/severe damage if damage exceeded $5200, personal property loss exceeded $5200 or (in the absence of direct inspection) remote sensing analyses found water depth of over 1 foot (30 cm) in the most significantly damaged portions of Orleans, St. Bernard and Jefferson Parishes.

Overall, 38 of 64 parishes in Louisiana suffered measureable damage associated with HKR, with 515 251 occupied housing units incurring some damage, of which 37 incurred some level of major/severe damage, affecting 204 682 occupied housing units. Figure 1 shows the geographic distribution of the major/severe housing damage suffered by each parish, detailed in online supplementary table S2. Parishes were assigned to four categories corresponding to distinct breaks in the distribution of housing stock damaged. Parishes that suffered the greatest amount all had >50% of local housing stock damaged: St. Bernard (78%), Cameron (72%), Plaquemines (58%) and Orleans (56%). Three other parishes had 10–50% of local housing stock damaged: St. Tammany (25%), Jefferson (20%) and Vermilion (13%). The remainder of the parishes were categorised as either having incurred 1–10% or <1% of damage to local housing stock.

Figure 1

Spatial distribution of severe housing damage estimates by Louisiana Parish, as of April 2006.

Empirical models for fetal death

We exploited the spatial and temporal variation in hurricane behaviour as a quasi-experimental design to calculate effects of local hurricane damage on fetal death. Since HKR did not strike all parishes in Louisiana, mothers residing in undamaged areas (<1% damage) presented a natural control group. We compared fetal outcomes among mothers in damaged versus undamaged parishes strictly before (1 January 2004–28 August 2005) and any time after (29 August 2005–31 December 2007) the hurricane events. As a test for the robustness of our results, we also expanded the observation period to be 1 January 1999 through 31 December 2009. A multivariate logistic estimator was used to model the probability of fetal death.

We first identified categorical thresholds for observing a significant dose–response of the likelihood for fetal death to the severity of damage in a parish. A series of analyses was conducted with mothers grouped by hurricane-caused damage levels as exposed groups (Embedded Image, where c=1–10%, 10–50% or >50%, respectively) while always using mothers that resided in undamaged parishes (<1% damage) as the reference group. The average causal effect of each level of hurricane damage was identified using the difference-in-difference estimator (Equation 1). The exogenous impact of exposure to hurricane damage in the aftermath of HKR was captured by the estimated coefficient Embedded Image on the interaction of the indicator for whether the gestation occurred in the period after the hurricane events, Embedded Image, and the indicator for the exposed group in question, Embedded Image. If the risk of fetal death is dose-responsive to exposure to HKR damage, we expect the magnitude of Embedded Image to be statistically significant and rising in the level of damage. Embedded Image(1)

We estimate the overall as well as the adjusted risk of fetal death at each level of damage exposure. Equation 1 was estimated with and without controlling for known modifiers, Embedded Image, of the relationship between stressors due to the damage left behind by HKR and recorded fetal death (ie, maternal, congenital, and placental conditions and other risk factors associated with fetal death),38–40 including a yearly time trend, Embedded Image. Covariates in the final model were selected through a stepwise process and considered to be potential effect modifiers if p<0.10.

We also aimed to compute the estimated fraction of observed fetal deaths in impacted parishes that may be attributable to the destruction and disorder caused by HKR. Equation 2 adapted our identification strategy to a continuous measure of the percentage of major damage to the housing stock in the mother's parish,Embedded Image. To allow the underlying probability of fetal death to vary non-linearly over all parishes, Embedded Image represented the vector of mutually exclusive categorical dummies for damage. However, by using the exact reported percentage of housing damage, the coefficient Embedded Image more precisely estimated the fetal death toll of maternal exposure to hurricane damage. Note that all mothers in gestation after the hurricane events who resided in parishes with housing damage above the extensive margin identified in the analysis above (>10%) were considered as exposed to hurricane damage, that is, Embedded Image. Embedded Image(2)

We computed the number of fetal deaths attributable to hurricane exposure in impacted parishes by using the OR corresponding to Embedded Image in Equation 2 and the formula, Embedded Image, where Embedded Image is the observed number of fetal deaths and Embedded Image is the observed percentage of housing stock damaged in Parish j.

Results

Table 1 reports ORs and 95% CI for eight models predicting fetal death between 2004 and 2007, with and without control variables. Statistical models 1–6 are divided into intervals of 1–10%, 10–50% and >50% of housing stock damaged. In models 1–2, we find that mothers residing in parishes experiencing 1–10% of major/severe damage to housing stock were no more likely to experience a fetal death than mothers visited by no major/severe damage. In models 3–4, we find an increase in the overall (OR=1.398, CI 1.096 to 1.784) and adjusted (OR=1.396, CI 1.067 to 1.827) risks of fetal death for mothers in parishes with 10–50% of damage versus mothers with no major damage in their parish. In models 5–6, we find that the overall and adjusted risks of fetal death were 79.6% (OR=1.796, CI 1.323 to 2.438) and 136.7% (OR=2.367, CI 1.684 to 3.327) greater for mothers exposed to >50% of destroyed housing, respectively.

Table 1

ORs predicting fetal death obtained from multivariate logistic regressions, Louisiana, 2004–2007

Online supplementary table S3 reproduces the logic of table 1 but expands the time period examined to 1999–2009. Results show a mild deflation in the size of ORs as more years after the event are added to the analysis, perhaps reflecting the slow return to normalcy in severely affected areas. Figure 2 graphically summarises results from tables 1 and online supplementary table S3. It supports the notion that the risk of fetal death is dose-responsive, rising in hurricane damage and stress incurred.

Figure 2

ORs (95% CI) of DD estimator by housing damage and time. DD, difference-in-differences.

In table 1, models 7–8, we regress fetal death on the per cent of housing stock in a parish suffering major/severe damage. The OR of interest is %Damage. Adjusting for relevant risk factors (model 8), we find that a 1% increase in housing stock destroyed (over the 10% threshold) increases the odds of fetal death by 1.7% (OR=1.017, CI 1.011 to 1.024). With the time period extended (see online supplementary table S3), results behave near identically (OR=1.016, CI 1.012 to 1.019).

Leveraging ORs in model 8, table 2 reports estimated fractions of observed fetal deaths in impacted parishes that may be attributable to hurricane destruction. In the discussion below, we provide various reasons pertaining to postdisaster migratory dynamics that necessitate treating derived attributions of fetal death as rough approximations. Although, of the 410 officially recorded fetal deaths in Jefferson, Orleans, Plaquemines, St. Bernard, St. Tammany and Vermillion parishes recorded in the aftermath of HKR, we estimate that between 117 and 205 may be attributable to hurricane destruction and consequent distress imposed on the resident population. This estimated fetal death toll is between 17.4% and 30.6% of the human death toll of 671 persons killed in these six parishes when the hurricane struck. The fetal death toll is disproportionate relative to population size, but consistent with the well-established bimodality of mortality risk in age.

Table 2

Estimated fetal death attributable to hurricane damage*

Discussion

Given that HKR caused considerable destruction and post-traumatic stress in resident populations in Louisiana,5 ,8 ,9 and given a sizable literature suggesting that maternal stress during pregnancy may increase the risk of negative fetal outcomes, we analysed whether unborn fetuses conceived by mothers residing in parishes visited by severe wreckage and housing adversity were at higher risk of fetal death. Exploiting spatial and temporal variation in hurricane devastation as a quasi-experimental design strategy, our results indicate that the destruction to housing stock caused by HKR imposed significant measurable losses in terms of fetal survival. We found that the odds of fetal death was about 1.4× and 2.4× higher in parishes suffering severe wreckage to housing stock of 10–50% and >50%, respectively.

While destruction of housing stock is a reasonable operationalisation of the local devastation caused by HKR, it incompletely captures the full extent of the loss, dispossession and anguish imposed on communities in Louisiana and is insufficiently sensitive to the experiences of individuals coping with the aftermath of HKR. Moreover, while results do show that the risk of fetal death increases monotonically in hurricane wreckage, the forced migration of population subgroups during HKR complicates easy conclusions. A major analytic challenge is distinguishing between disaster exposure and compositional effects.

In parishes suffering >50% severe/major damage to housing stock, we observed dramatic decreases in the count of live births following HKR. The count of live births in the hardest hit parishes was 40.44% lower in 2007 (N=4919) as compared with 2004 (N=8259), while the risk of fetal death increased from 4.48 to 7.52 per 1000 live births. The joint behaviour of live birth and fetal death counts in time is more suggestive of a disaster exposure as opposed to a compositional effect. Online supplementary appendix figure 1A provides more detail, showing the monthly count of live births in time in parishes grouped by severe damaged incurred. With the landfall of Hurricane Katrina on 29 August 2005, we observe a significant decline in the count of live births in areas incurring >50% destruction to housing stock. The combined count of live births in St. Bernard, Cameron, Plaquemines and Orleans parishes (areas with >50% of housing stock severely damaged) declined an astonishing 79.2% in the 3 months following Katrina. This precipitous decline in the reproductive force reflects the large and well-documented out-migration of resident populations from the hardest hit coastal parishes of Louisiana.

With respect to the compositional effect as possibly explaining the observed increase in fetal death risk, insofar as persons forcibly displaced by HKR were from demographic subgroups predisposed to higher risk of fetal death, then our reported disaster effects are likely underestimated. In such a case, the change in our disaster exposure group will bias our ORs downward, as mothers remaining behind are of lower underlying risk of fetal death. If however persons displaced by HKR were from demographic subgroups predisposed to lower risk of fetal death, our reported ORs are likely biased upward.

In online supplementary appendix 1B, we observe the proportion of live births delivered by African-American mothers in time in parishes grouped by severe damage incurred. According to regression models, African-American mothers are between 1.47× and 1.67× more likely to suffer a fetal death. In panel B, and similar to the monthly count of live births, we observe a sudden contraction in the proportion of infants born to African-American mothers in parishes with >50% severe damage to housing stock. To be precise, the proportion of live births had by African-American mothers declined from 0.68 to 0.46 in the 3 months following HKR, suggesting that persons forcibly displaced by HKR were from demographic subgroups at higher risk of fetal death. Moreover, in parishes visited by >50% severe destruction to housing stock, we also found increases in the percentage of married mothers and mothers with college education. While these variables are uncorrelated with fetal death, their postdisaster movement suggests that forced displacement was selective of mothers of lower socioeconomic status.

While, on balance, observed changes in the demographic composition of birth mothers (ie, compositional effects) following HKR likely function to down-bias calculated postdisaster exposure effects, we emphasise caution with respect to estimated fractions of officially recorded fetal deaths attributable to maternal exposure to hurricane destruction. It should be noted that fetal death data are imperfect, involving routine underestimation of fetal deaths among unborn children <20 weeks in gestational age.41 Given the significant social and economic disorder caused by HKR in the hardest hit parishes, it is likely that reporting errors on fetal death certificates worsened in the aftermath of HKR. Although one cannot discern precisely the behaviour of this measurement bias following HKR, it is more likely than not that the problem of undercounting fetal death increased in the hardest hit parishes relative to unscathed parishes.

The demand for public health science and practice in pursuit of detailing and preventing the in utero costs of natural disasters will likely rise in time. Climate scientists maintain that the frequency, duration and intensity of North Atlantic tropical cyclones like HKR are rising in anthropogenic climate change.42–44 The expected economic impacts of more frequent, longer lasting and more intense hurricanes are substantial.45 Insofar as our empirical findings meaningfully generalise in time, the health risks to the unborn and their perinatal development will likely increase with more frequent and intense hurricanes.45

From a public health science standpoint, many granular questions on the link between maternal exposure to catastrophic events and fetal death remain unanswered. Postdisaster studies carefully inventory the destruction to property and infrastructure, job loss, homelessness, injuries and deaths, the psychic toll on resident populations and the overrun of social welfare organisations (like hospitals and voluntary associations).5–9 ,46 With respect to postdisaster reproductive outcomes, it is unclear which of these proximate sources of hurricane-induced distress exercises the greatest influence over fetal outcomes. Zotti et al23 maintain that improved public health surveillance can resolve such questions, advancing understanding of exposure effects on maternal risk factors, behaviours and fetal outcomes. Specifically, they suggest appending a disaster-specific module to the Pregnancy Risk Assessment Monitoring System (PRAMS), a population-based surveillance survey collecting vital information on maternal and child health before, during and after pregnancy. With more granular information, public health interventions, from the calibration of warning systems 47 to targeted investments in disaster-prone areas,10 can be improved to minimise the in utero costs of disasters and advance social welfare.

What is already known about this subject?

  • In a recent systematic review of research on the subject, Zotti et al (2013) find inconsistent evidence for the idea that negative fetal/infant health outcomes increase with maternal exposure to destruction caused by natural disasters, concluding that: “Effects on outcomes were not consistently demonstrated, and study methodologies varied widely. Even so, these studies suggest an association between disasters and reproductive health and highlight the need for further studies of clarify associations.”

What this study adds?

  • With availability of geo-referenced housing wreckage data, we advance prior research in terms of measurement of postdisaster impacts and develop a unique quasi-experimental design strategy to assess the association between maternal exposure to postdisaster destruction and fetal death risk. By exploiting spatial and temporal variation in hurricane caused destruction of housing stock as a quasi-experimental design, we assess whether fetal death is dose-responsive in the extent of hurricane damage, with the average causal effect of maternal exposure to postdisaster destruction captured by difference-in-differences analyses. We find that the odds of fetal death increase significantly in the severity of housing wreckage imposed on localities in Louisiana. Our identification strategy allowed estimation of the number of fetal deaths attributable to the destruction and local disorder caused by Hurricanes Katrina and Rita. We also consider how our reported effects are likely biased downward by postdisaster out-migration of residents in severely damaged Parishes, and raise questions for future research on the proximate sources of maternal distress to inform practice. 

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:

  • Press release

    Files in this Data Supplement:

Footnotes

  • Acknowledgment The authors wish to thank the Robert Wood Johnson Health and Society Scholars program for their financial support.

  • Contributors SZ conceived the project, designed and executed the statistical analysis, and wrote the majority of text. IMB assisted with design and execution of statistical analysis and provided text throughout the article. BGL assisted with design and provided text throughout the article. JGS provided text on literature on maternal stress, disasters and negative fetal outcomes. SW assisted with research design and policy implications and provided text throughout. HWM assisted with geographic analysis and provided text throughout.

  • Competing interests None.

  • Ethics approval Colorado State University Institutional Review.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Data were provided to the lead author, SZ, by the Louisiana Department of Public Health. According to the data agreement, only SZ is permitted to analyse the data.