Exposure to the Indian Ocean Tsunami shapes the HPA-axis resulting in HPA “burnout” 14 years later

Significance

We provide causal evidence on the long-term effects of traumatic exposures on HPA-axis function, using longitudinal data from survivors of the 2004 Indian Ocean tsunami in Indonesia. Fourteen years after the disaster, women who were exposed to the tsunami’s direct impacts present with levels of hair cortisol 30% lower than counterparts not similarly exposed. We distinguish short-term from longer-term levels of elevated post-traumatic stress symptoms and show that effects are larger among individuals for whom post-traumatic symptoms remain elevated for 2 y after the tsunami which likely reflects “burnout” since low cortisol is also associated with contemporaneous indicators of poor general and psychosocial health. The impacts of large-scale traumatic events on women’s physiology through the HPA-axis endure for many years.

Abstract

Despite significant research on the effects of stress on the hypothalamic–pituitary–adrenal (HPA) axis, questions remain regarding long-term impacts of large-scale stressors. Leveraging data on exposure to an unanticipated major natural disaster, the 2004 Indian Ocean tsunami, we provide causal evidence of its imprint on hair cortisol levels fourteen years later. Data are drawn from the Study of the Tsunami Aftermath and Recovery, a population-representative longitudinal study of tsunami survivors who were living along the coast of Aceh, Indonesia, when the tsunami hit. Annual rounds of data, collected before, the year after and 2 y after the disaster provide detailed information about tsunami exposures and self-reported symptoms of post-traumatic stress. Hair samples collected 14 y after the tsunami from a sample of adult participants provide measures of cortisol levels, integrated over several months. Hair cortisol concentrations are substantially and significantly lower among females who were living, at the time of the tsunami, in communities directly damaged by the tsunami, in comparison with similar females living in other, nearby communities. Differences among males are small and not significant. Cortisol concentrations are lowest among those females living in damaged communities who reported elevated post-traumatic stress symptoms persistently for two years after the tsunami, indicating that the negative effects of exposure were largest for them. Low cortisol is also associated with contemporaneous reports of poor self-rated general and psychosocial health. Taken together, the evidence points to dysregulation in the HPA axis and “burnout” among these females fourteen years after exposure to the disaster.

Understanding the long-term impacts of exposure to stressful events on human health and well-being is a pressing scientific imperative (1). Insults to the hypothalamic–pituitary–adrenal (HPA) axis in particular have the potential to accumulate over time and affect multiple biological systems, but there is a paucity of evidence on the longer-term impacts of stressors on HPA axis function. This research contributes to filling this gap.

Exposure to an unanticipated disaster, the 2004 Indian Ocean tsunami, is harnessed to identify its causal impact on hair cortisol levels 14 y later. Combining population-representative longitudinal data collected before and after the tsunami, we compare individuals directly exposed to the tsunami with otherwise similar individuals not directly exposed. We document lower levels of cortisol 14 y after the tsunami among females exposed to the disaster relative to females who were not exposed. Differences are particularly large for females with elevated levels of post-traumatic stress reactivity 2 y after the tsunami. Low cortisol, in turn, is linked to poorer contemporaneous physical and psychosocial health. Our work advances understanding of the enduring consequences of large-scale adverse events including events arising from climate change, environmental degradation, and pandemics.

We address several key challenges that complicate interpretation of evidence on long-term health outcomes after trauma (2, 3). First, exposure to adverse events typically depends on individual, family, and environmental characteristics that are hard to measure and confound analyses. Second, few studies draw comparisons between individuals who are comparable apart from exposure to the adverse event. Third, studies of long-term effects of exposure usually rely on recalling events years before, the accuracy of which is unclear. This research leverages plausibly random exposure to an unanticipated stressor, the 2004 Indian Ocean tsunami. Using longitudinal data collected in Aceh, Indonesia, before and after the tsunami, we compare those living in coastal communities hit directly by the waves with those living in nearby coastal communities that were spared.

Cortisol is central to the body’s stress response. Elevated cortisol is a healthy physiological response to a stressor but should deactivate once the stressor has ceased (4). Elevated cortisol has been linked to many stress-related experiences, including long-term unemployment (5), early-life psychosocial stress (6), and stressful life events (7, 8).

Whereas elevated cortisol is a normal response to stress, repeated consistent elevation and concomitant increases in allostatic load may, in the long run, lead to HPA axis dysregulation in the form of hypocortisolism, i.e., reduced levels of cortisol and other regulatory hormones (1, 9, 10). Hypocortisolism has been documented in animal models (11), humans working highly stressful jobs (12, 13), humans with low early-life SES (14), and among those exposed to early-life trauma or repeated traumatic experiences (15–19).

Although most studies have relatively short-term horizons, some evidence suggests immediate and longer-term outcomes differ (17, 20). Among survivors of the 2008 Wenchuan earthquake, hair cortisol was elevated immediately afterward but declined among some survivors after a year (21). Housing damage due to the 2011 tsunami in Japan has been linked to lower serum cortisol levels 5 to 8 mo later (22). Longer-term evidence is sparse and heterogeneity in how individuals process stressors is not well understood (23, 24). We examine cortisol levels after 14 y and investigate mechanisms underlying heterogeneity in those levels.

Study of the Aftermath and Recovery (STAR) respondents were first interviewed before the 2004 Indian Ocean tsunami and reinterviewed multiple times afterward. Cortisol is assessed 14 y after the tsunami using scalp hair, providing a retrospective measure of several months’ cumulative HPA axis activity (25, 26). This approach avoids measurement and modeling complications that arise in cortisol levels from saliva, urine, or blood, for which diurnal variation necessitates repeat observations during the day and, ideally, over multiple days (27–29).

The 2004 tsunami was a major disaster spawned by a massive megathrust earthquake 160 km off the coast of the island of Sumatra that killed over 160,000 people living in coastal areas of Aceh province (30, 31). Two facts are key. First, the tsunami was completely unanticipated: Evidence from sediment cores indicates the last tsunami that hit mainland Sumatra occurred some 600 y ago (32). Second, whether a specific coastal community was directly affected by the tsunami depended on idiosyncratic features of the land and sea-floor topography in combination with wave direction (33). Thus, among people living in coastal communities in Aceh at the time of the tsunami, exposure to large-scale stressors from the experience of the tsunami and the resulting damage were plausibly random. We leverage this exogenous shock to identify the causal effects of exposure to those stressors on HPA activation over the long term.

We test whether hair cortisol levels 14 y after the tsunami are significantly different for adults who were exposed to the tsunami relative to those not directly exposed, based on residential location at the time of the tsunami. Focusing on the exposed group, we assess whether cortisol levels are related to a measure of post-traumatic stress immediately after the tsunami and over the following 2 y. We also examine associations between low cortisol and contemporaneous self-reported health to investigate the evidence for HPA axis “burnout” among those exposed to the tsunami and those who experienced elevated levels of post-traumatic stress.

Results

Cortisol levels in hair are analyzed for 615 adult respondents. The cortisol distributions are displayed in Fig. 1A. The sample average is 6.5 pg per milligram of hair with females about 30% higher than males.^* In multivariable ordinary least squares regression models, we use the logarithm of cortisol as the dependent variable because the distribution is right skewed. (Log normality is not rejected: Shapiro–Wilks test P-value = 0.35).

Fig. 1.

Distribution of hair cortisol, and relationship with PTSR, by gender. (A) Distributions of hair cortisol levels. (B) Hair cortisol levels and PTSR relationships. Notes: Gender-specific kernel density estimates of distributions of hair cortisol levels (pg of cortisol per mg of hair) for all respondents displayed in panel A, with overall means and gender-specific means. Gender-specific relationships between cortisol levels and PTSR scores 2 y after the tsunami for those respondents who were living in areas that were damaged by the tsunami displayed in panel B are locally weighted smoothed scatterplots estimated with 30% bandwidth and tricube weighting function.

Table 1 presents summary statistics and estimates of the relationship between ln(cortisol) and two measures of tsunami exposure. The regression coefficient estimates are multiplied by 100 and can be interpreted as indicating percentage change. Asymptotic t-statistics are based on variance-covariance estimates that take into account clustering at the baseline community level and arbitrary heteroskedasticity. All regression models control age (specified as linear splines allowing different slopes for <45 y, 45 to 55 y, and >55 y olds), years of completed education, and indicator variables for each of whether the respondent smokes, prays regularly, relaxes regularly, uses chemicals on their hair, and washed their hair within 24 h of sample collection.^†

Table 1.

Regression coefficients on ln(cortisol) and tsunami experiences, by gender

	Mean	A. Community exposure only			B. Add PTSR		C. Only PTSR
	(SE) [1]	All resp [2]	Females [3]	Males [4]	Females [5]	Males [6]	Females [7]	Males [8]
Community-level exposure
Indicator = 1 if area sustained
Heavy damage	0.44	−18.9	−29.1	−4.8	−30.8	−2.1
		[2.3]	[3.1]	[0.5]	[2.6]	[0.2]
Some damage	0.35	−7.9	−10.9	−4.9	−13.1	−4.9
		[0.8]	[1.0]	[0.4]	[1.2]	[0.4]
No damage (reference)	0.21
Individual-specific PTSR level
PTSR level
Maximum after tsunami	8.86				1.7	−0.9	1.2	−0.8
	(0.18)				[1.5]	[0.6]	[1.1]	[0.6]
1 y after tsunami	5.46				−0.3	0.9	−1.1	0.9
	(0.16)				[0.3]	[0.7]	[1.1]	[0.6]
2 y after tsunami	4.88				−2.0	−0.8	−2.3	−0.7
	(0.15)				[2.0]	[0.8]	[2.4]	[0.6]
Indicator = 1 if resp is male	0.47	−17.9
(reference is female resp)	0.53	[2.1]
Observations	615	615	327	288	327	288	327	288
R2		0.13	0.09	0.08	0.10	0.08	0.06	0.08
F test (P values)
All community exposures = 0		0.085	0.011	0.861	0.047	0.914
Effects same in heavy and some damage communities		0.237	0.083	0.989	0.132	0.783

Notes: Column 1 reports means (and SE) of covariates used in analyses. The remaining columns report coefficients from regression of ln(cortisol) on measures of tsunami exposures. Panel A includes only community-level exposures specified as indicator variables for the extent of damage where the respondent was living at the time of the tsunami. Panel B adds levels of individual-specific PTSR measured after the tsunami. Panel C excludes the damage indicators. All models include controls for age (specified as linear spline with knots at 45 y and 55 y), years completed education and separate indicator variables for whether respondent smokes, washed hair in last 24 h, uses chemicals on hair, prays regularly, regularly relaxes, PTSR missing. All coefficients multiplied by 100. Robust t statistics in parentheses take into account heteroscedasticity and clustering in the community of residence at the time of the tsunami.

Our first measure of tsunami exposure is an indicator of community-level damage constructed by comparing high-resolution satellite imagery immediately before and after the tsunami for each respondent’s residential location at the time of the tsunami. This is a plausibly exogenous measure of exposure to the stressors of the tsunami. We identify communities that sustained heavy damage, moderate damage, and no direct damage (the reference group). Forty-four percent of the study sample was living in areas that sustained heavy damage and 35% in areas that sustained moderate damage (Table 1, column 1). Regression estimates are displayed in panel A of the table, for all respondents in column 2, and stratified by gender in columns 3 and 4.

Fourteen years after exposure, cortisol levels are lower among respondents who were living, at the time of the tsunami, in areas that were damaged relative to those who were living in areas that were not. The result is driven by a large impact of tsunami exposure on females in heavily damaged areas: Their cortisol levels are 30% lower than females in areas not-damaged (column 3). While females living in moderately damaged areas have 11% lower cortisol than those in not-damaged areas, that difference is not statistically significant. Combining females living in heavily and moderately damaged areas, cortisol levels are significantly lower relative to those who were living in areas that were not damaged (column 3, P-value = 0.01). (The gap between heavily and moderately damaged areas is not statistically significant). In contrast, location at the time of the tsunami is not related to variation in cortisol levels among males; the effects are small and not statistically significant (column 4).

Second, we investigate the links between individual-specific indicators of tsunami exposure using seven symptoms from the Post-Traumatic Stress Checklist–Civilian Version (PCL-C). The symptoms cover the three distinctive psychological domains of post-traumatic stress: hyperarousal, avoidance behaviors, and event reexperience. Respondent reports of the presence and severity of each symptom were scored (maximum = 3) and summed to construct post-traumatic stress reactivity (PTSR) which ranges from 0 to 21, with higher scores indicating greater reactivity. We use the full range of this scale rather than a dichotomous indicator (as used in post-traumatic stress disorder, PTSD, which is based on all 17 PCL-C items) (34, 35).

In addition to reflecting the individual-specific experience of each respondent, PTSR changes over time, providing an opportunity to investigate the dynamics underlying links between stress and cortisol. Measurement of the highest PTSR reported by each respondent at any point during the first 2 y after the tsunami as well as PTSR a year and 2 y after the tsunami enables us to separate the roles of peak PTSR and longevity of elevated PTSR. It is not clear that results based on self-reported PTSR can be given causal interpretation; we view them as complementary to the analyses in panel A that use community-level exposures.

Mean PTSR declines over time from a maximum of 8.86 to 5.46 1 y and 4.88 2 y after the tsunami (Table 1, column 1). Females report higher maximum PTSR, but this gender gap disappears over time (SI Appendix, Table S1) (35). Some respondents directly exposed to the tsunami presented with elevated PTSR levels immediately after the tsunami and with persistently elevated levels into the second year after the disaster. PTSR levels among those not directly exposed are relatively low and do not change substantially across the two survey waves.

The models in Table 1 panel B add the three measures of PTSR to the models in panel A. Estimates of community-level effects change little, indicating that, on average, females who were living in heavily damaged areas suffered from decreased cortisol irrespective of PTSR levels. Moreover, negative effects are largest in magnitude among females whose PTSR levels remain elevated 2 y after the tsunami and there are no significant independent effects of maximum or its level 1 y after the tsunami (columns 5 and 7). This suggests that multiyear sustained PTSR elevation drives lower cortisol over the long term.^‡ The models in panel C, which exclude controls for community-level exposure (columns 7 and 8). Comparing results for females, the negative effects of PTSR are slightly larger in magnitude in column 7 relative to column 5, suggesting that only a small portion of the PTSR effect is attributable to community of residence at the time of the tsunami. The results are summarized in Fig. 1B, which displays nonparametric estimates of the bivariate relationships between ln(cortisol) and PTSR 2 y after the tsunami for females and males. The shapes for both males and females are similar: Cortisol levels decline until PTSR reaches 6, and then, the relationships flatten out. The decline is more muted for males and indistinguishable from a horizontal line, consistent with the regression results.

To focus on the dose–response impact of PTSR, Table 2 restricts attention to respondents who were living, at the tsunami, in areas that sustained damage. The model in panel A repeats the model estimated in panel C of Table 1 with this restricted sample. The estimated impact of PTSR scores 2 y after the tsunami is about 50% larger in the restricted sample. Among females, cortisol levels are 3% lower for each additional unit increase in PTSR (column 1). For males, the effect is in the same direction but smaller in magnitude (1%) and not statistically significant.

Table 2.

ln(cortisol) and PTSR of respondents who were directly exposed to the tsunami, by gender

	I. No fixed effects		II. Include fixed effects for community of residence at time of tsunami
	A. Base model		B. Base model		C. Minimal controls		D. Extended controls		E. Restrict age ≤ 55
	Females [1]	Males [2]	Females [3]	Males [4]	Females [5]	Males [6]	Females [7]	Males [8]	Females [9]	Males [10]
PTSR level
Maximum after tsunami	0.9	−0.8	1.5	−0.8	1.6	−1.0	1.4	−1.1	1.7	−2.4
	[0.8]	[0.5]	[1.1]	[0.6]	[1.2]	[0.7]	[1.1]	[0.8]	[1.2]	[1.5]
1 y after tsunami	−0.2	0.7	−0.5	0.3	−0.6	0.4	−0.7	1.1	−0.4	1.5
	[0.2]	[0.5]	[0.4]	[0.2]	[0.4]	[0.3]	[0.6]	[0.8]	[0.3]	[1.0]
2 y after tsunami	−3.0	−1.0	−2.7	−1.1	−2.8	−1.0	−2.4	−1.5	−2.6	−0.4
	[2.9]	[1.0]	[2.5]	[0.8]	[2.5]	[0.8]	[2.1]	[1.1]	[2.2]	[0.2]
Observations	249	235	249	235	249	235	249	235	209	179
R2	0.08	0.12	0.06	0.13	0.05	0.12	0.14	0.17	0.06	0.13

Notes: Coefficients reported from regression of ln(cortisol) on measures of individual-specific PTSR for respondents who were living at the time of the tsunami in areas damaged by the tsunami. Panel I repeats specification in Table 1, model C, but on restricted sample. All models in panel II include fixed effects for community of resident at the time of the tsunami. Model B controls are the same as model A, apart from inclusion of fixed effects. Model C specification is the same as B except drops prayer and relaxation controls. Model D specification extends controls in model B by also including characteristics measured at the time hair samples were collected. Extended controls are concurrent PTSR and nine indicator variables, one each for four psychosomatic symptoms (feelings of loneliness, sadness, having difficulty concentrating, or finding normal tasks an effort), difficulty sleeping, worrying excessively, feeling irritable, feeling bothered, and being quick to anger. Model E uses same specification as model B but restricts sample to those age 55 and younger at the time of assessment. All coefficients multiplied by 100. Robust t statistics in parentheses take into account heteroscedasticity and clustering in the community of residence at the time of the tsunami.

The extent to which PTSR effects are driven by other, correlated characteristics of the community of residence at the time of the tsunami is investigated in Table 2 panel B which includes a community-specific fixed effect. The dose–response effect of PTSR is identified by variation between respondents living in the same community at the time of the tsunami and takes into account all observed and unobserved community characteristics that are the same for respondents in each community. For females, the within-community estimates are slightly smaller in magnitude (−2.7%) but remain significant.

The models discussed thus far include respondent behaviors (measured 14 y after the tsunami) that potentially influence stress and cortisol production, including whether respondents pray or relax regularly. Since these behaviors may be correlated with other unobserved characteristics that affect PTSR, in panel C, those controls are excluded from the fixed effects models: Estimates change very little. In panel D, the model in B is extended to also include stressors measured concurrently with hair collection: The estimates change little.^§

Tsunami-related mortality was substantially higher among females relative to males because of differences in strength and swimming ability) (36, 37). To assess whether mortality selection drives the gendered cortisol results, the samples are restricted to younger respondents for whom tsunami-related mortality was lower: those younger than 42 y at the time of the tsunami (≤55 y at hair collection) (panel E). For females, the estimated effect of PTSR is little affected. For males, the coefficient is halved in magnitude and remains insignificant. Mortality selection does not explain the results.

Evidence in support of our interpretation that low cortisol reflects “burnout” is shown in Table 3, which examines the relationship between cortisol and self-reports of health collected at the same time as hair samples. The models include an indicator variable for whether self-reported general health is fair or poor (with good or very good health as the reference group) and the number of psychosomatic symptoms that the respondent reported feeling in the prior four weeks. We consider four symptoms: feeling lonely, sad, difficulty concentrating, and finding normal tasks an effort, that are not included in PTSR measures. The first dependent variable is an indicator for whether the respondent is in the bottom tercile of the cortisol distribution. For females, both contemporaneous health indicators are positively associated with low cortisol—estimates are individually significant and jointly significant. For males, only general health status is significantly related to low cortisol and the health indicators are not jointly significant. The second dependent variable is ln(cortisol), and the results for females mirror those in panel I of the table. For males, neither health indicator is significantly linked to ln(cortisol).

Table 3.

Relationship between cortisol and contemporaneous health indicators

	I. Bottom tercile of cortisol distribution		II. ln(Cortisol)
	Females	Males	Females	Males
	[1]	[2]	[2]	[3]
A. Indicator variable if self-reported general health status is fair or poor	17.0	20.0	−14.1	−8.2
	[2.8]	[2.4]	[2.1]	[0.9]
B. Number of psychosomatic symptoms experienced in previous 4 wk	6.7	−3.0	−8.5	4.9
	[3.0]	[1.0]	[3.0]	[1.2]
Observations	339	291	327	288
F test for joint significance health indicators (P-value)	0.000	0.076	0.000	0.398

Notes: Coefficients are reported from OLS regressions estimated separately by gender and adjusted for age. The dependent variable in panel I is indicator that takes value 1 if cortisol is in the bottom tercile of the distribution and zero otherwise. The dependent variable in panel II is ln(cortisol) and excludes respondents whose cortisol levels were above the assay maximum, the same sample used in Tables 1 and 2. All coefficients are multiplied by 100 and indicate percentage change. The reference category for fair or poor general health status in row A is respondents who reported good or very good general health. Four psychosomatic symptoms included in the index in row B are whether the respondent felt lonely, felt sad, had difficulty concentrating, or found normal tasks an effort. Psychosomatic symptoms and general health status were assessed at the same time as hair sampling. Robust t-statistics and F test statistics for joint significance take into account heteroscedasticity and clustering in the community of residence at the time of the tsunami.

Discussion

Exposure to a plausibly exogenous destructive natural disaster exerts a large and statistically significant negative impact on hair cortisol levels of females 14 y later. Cortisol levels of females living in heavily damaged communities at the time of the tsunami are 30% lower than for comparable females living in nearby undamaged communities. The direction of the effect is the same for males but smaller and statistically insignificant.

This pattern extends beyond our exogenous contextual measure of exposure to individual-specific markers of PTSR measured after the tsunami. Respondents with elevated PTSR 2 y after the tsunami present with lower cortisol levels 14 y later. In multivariable regression models, the effect is large in magnitude and statistically significant for females, but not for males. It is unlikely that sex differences reflect different exposures since the waves came ashore around 8:30 am on Sunday 26 December, when most were at home. Possibly the experience was processed differently by sex, reflecting different roles of social and economic resources and biological differences in sensitivity to stress and HPA axis responses from, for example, differences in estrogen (38).

The literature suggests two possible patterns linking major stressors and downstream cortisol levels. On the one hand, consistently elevated stress may result in higher cortisol levels at a later point, suggesting chronic stress. On the other hand, cumulative effects of extended stress may lead to HPA axis dysregulation and reduced cortisol levels long term (39, 40). Our findings are consistent with the latter hypothesis both in models of the exogenous effect of community-level exposure and in models isolating a dose-response of PTSR levels. We establish that the latter are driven by those whose PTSR levels are persistently high, consistent with extended hyperactivation of the HPA axis resulting in eventual “burnout.” A limitation of this study is the lack of multiple waves of hair cortisol, so we cannot document the trajectory of cortisol in the aftermath of the tsunami. Nonetheless, the fact the tsunami was a large-scale, unanticipated, and plausibly exogenous shock suggests that both PTSR-related symptoms and hypocortisolism among those exposed to the disaster result from exposure-related stressors rather than other, unobserved characteristics (17).

Beyond the deleterious impacts of tsunami exposure on health and well-being, hypocortisolism may have other important impacts, potentially including the “hypocortisolemic symptom triad” of high stress sensitivity, fatigue, and chronic pain (2, 10, 41). Among females, low cortisol is significantly associated with fair or poor self-rated general health status and with more psychosomatic symptoms reported by the respondent at the time we collected hair samples. These results are consistent with the interpretation that, among females, low cortisol linked to tsunami exposures reflects “burnout.”

We conclude that, in these data, high levels of tsunami-related stress result in reduced hair cortisol measured 14 y after the tsunami reflecting HPA axis dysregulation which likely has deleterious implications for future health and well-being. Interventions designed to mitigate these effects of disaster exposure will yield large positive returns to individuals and society.

Methods and Data

Exposure to the Tsunami and Study Setting.

On December 26, 2004, the Sumatra–Andaman earthquake spawned a tsunami with waves up to 30 m that devastated some communities along Aceh’s western coast, on the northern end of Sumatra, Indonesia. Other, nearby communities were untouched (32, 42). Community-level exposure to tsunami-associated stressors is plausibly exogenous with respect to future health, including cortisol levels, for two reasons.

First, the tsunami was sudden and unexpected. Tsunami risk was thought to be low, no warning sensors were in the Indian Ocean, and no tsunami had hit mainland Sumatra for over 600 y (43).

Second, the height, force, and inland reach of water from the tsunami depended on topographical features of the ocean floor and shoreline (33, 44). In some areas, communities were destroyed, whereas other, nearby communities protected by a promontory or facing away from the waves, were not. We leverage this variation in combination with STAR survey data to isolate a causal impact of exposure to the stressors on cortisol levels.

The baseline was conducted before the tsunami in February and March 2004, as part of Statistics Indonesia’s annual national socioeconomic survey, SUSENAS. We conducted follow-ups annually for 5 y and at 10 and 13 y after the tsunami. Duke University Campus IRB approved the project (Protocol 2126). After the study goals, questions and sample procedures were explained, each respondent provided informed consent verbally. Using baseline data, we have established that the age, gender, education, and household per capita expenditure (a measure of socioeconomic status) of those who were living in tsunami-damaged coastal communities are neither substantively nor statistically different from those in neighboring coastal communities (36). Further, in the baseline conducted before the tsunami, respondents were asked about disaster risks. Only 15 respondents mentioned a tsunami as most likely. 90% reported no risks and the reported likelihood is no different between those living in communities with subsequent tsunami damage and those in places not directly affected (36).

Our primary measure of stress exposure is based on the degree of destruction the tsunami caused at the community level. Based on their community of residence at the time of the tsunami, individuals are stratified into three damage categories: heavily, moderately, and not directly damaged. These categories are defined by triangulating information from satellite imagery immediately before and after the tsunami, community informant reports, and interviewer observations after the tsunami (36). Community damage classifications have been cross-validated with community-specific mortality data, administrative and census data collected independently by Statistics Indonesia.

We treat this community-level exposure measure as exogenous since the tsunami was unanticipated and residential location at the time of the tsunami is unlikely to be related to factors that affect stress responses. These analyses are complemented with individual-specific measures of tsunami trauma that reflect experiences at the time and responses to those experiences. They are summarized in a scale capturing post-traumatic stress reactivity (PTSR), discussed at length in Frankenberg et al. (35). In surveys conducted after the tsunami, we use seven items from the 17-item PTSD Checklist Civilian Version (PCL-C) that in combination reflect three key symptoms of post-traumatic stress disorder (34). The 17-item PCL-C has been validated with veterans, victims of accidents and sexual assault, and survivors of bone marrow transplants (45, 46). We used a shortened version to reduce respondent burden.

Data from each STAR follow-up respondent are used to construct PTSR scales consistent with empirical evidence of a stress-response continuum (47, 48). Beginning five months after the tsunami, we reinterviewed surviving baseline respondents and elicited whether they had ever experienced each symptom since the tsunami, scoring responses from 0 (no occurrence) to 3 (occurred often when experienced most intensely) and summed across the symptoms. The resulting scale, ranging from 0 to 21, is our first PTSR measure: the respondent’s highest level experienced at any point. We repeated the symptom questions at the first after the tsunami interview to create our second measure, PTSR about a year after the tsunami. Approximately 2 y after the tsunami, we assessed the same symptoms in the past 12 mo for our third measure, PTSR 2 y after the tsunami. Cronbach’s alpha ranges between 0.69 and 0.72 for the measures. We include all three PTSR measures in the analytical models to distinguish the effects of elevated PTSR immediately after the tsunami from elevated PTSR that persists over time.

Whereas spatially based measures of exposure are plausibly exogenous, PTSR measures reflect both exogenous exposures and individual-specific processing of those exposures, which are potentially related to other, unobserved factors in HPA activation (49). They have the advantage, however, of reflecting the perceived level of stress as recalled by the respondent.

PTSR has been linked to individual-specific stressful experiences at the time of the tsunami (35), including loss of family and friends, being swept into the water, watching others struggle and disappear, losing homes, and seeing victims’ bodies. As these stressors are correlated, we do not distinguish the roles of specific stressors; our exposure measures capture the influence of all stressors taken together.

Sample and Cortisol Measurement.

Fourteen years after the tsunami, we randomly sampled 35 communities where respondents had resided at the time of the tsunami, oversampling heavily damaged communities. All respondents from those communities, regardless of residence at the time of hair collection, were eligible for the study as long as they were age 21 or older at the tsunami: We successfully measured cortisol in hair for 91% of the 695 target respondents. Loss to follow-up is unrelated to tsunami exposure, age, gender, or socioeconomic status, indicating that attrition is unlikely to be an important source of bias. We could not locate 1% of target individuals, 6.8% refused to provide a hair sample, and 1.6% were bald. Cortisol was quantified in 615 of 630 hair samples. Interviewers collected over 30 mg of hair by cutting from the posterior vertex, as close to the scalp as possible. Hair was stored in foil and refrigerated until analysis. At collection, participants were asked about medication use (including corticosteroids), hair coloring, hair treatment, hair washing, physical activity, meditation, prayer behavior, and relaxation.

Hair samples were ground in a laboratory in Yogyakarta, Indonesia, following methods derived from Davenport et al. (50). The proximal 3 cm was weighed and washed with isopropanol and then dried. After extensive testing, the protocol was modified by washing the sample four instead of two times to remove hair surface contamination. Samples were ground using a bead-beater and processed under methanol for 18 h. The supernatant was extracted and dried under a stream of air. The resulting dried samples were reconstituted in assay buffer, filtered, and frozen until assayed. Assays were run using a salivary cortisol kit (EIA, Arbor Assays). Fifteen out of 630 samples were above the assay maximum even after 50% dilution and were excluded from analysis.^¶ Hair cortisol concentrations were calculated in pg of cortisol per mg of hair assayed.

We rigorously validated lab procedures using a series of paired hair samples from volunteers in the United States that were tested in the reference Meyer laboratory at the University of Massachusetts–Amherst prior to study initiation and retested in our lab in Indonesia. The paired samples correlate well: ρ = 0.982.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Digital data have been deposited in stardata.org (See Site) (51).