Associations Between Oil Spill Exposure Patterns and Acute Symptoms in United States Coast Guard Responders During the Deepwater Horizon Response

Abstract

Background:

Oil spill response workers encounter a unique mixture of hazardous exposures. Few studies have attempted to model these mixtures and evaluate the impact on worker health. The purpose of this investigation was to understand the association between clustered patterns of hazardous exposure and acute symptoms reported by United States Coast Guard (USCG) responders during the Deepwater Horizon oil spill (DWHOS).

Methods:

We conducted a cross-sectional study of USCG members who responded to the DWHOS and completed a post-deployment survey (n=4,855). The survey asked responders about a variety of hazardous exposures and acute symptoms experienced during their deployment. A previously conducted latent class analysis identified four unique latent classes (LC) which represent reported exposure patterns within this study population. We utilized the three higher LC levels to represent increasing probabilities of exposures to crude oil, exhaust, general outdoor environment, and experience of anxiety, with a “low overall exposure” group acting as the reference. Using multivariable log-binomial regression analyses, adjusted prevalence ratios (aPR) of acute symptoms in each LC compared to the reference LC were estimated.

Results:

There were significant positive associations between LCs and acute symptoms in all major organ systems with evidence of exposure-response relationships. Some of the strongest associations for individual acute symptoms representative of each organ system included chest pain, skin rash/itching, difficulty hearing, diarrhea, muscle pain, lightheadedness/dizziness, and shortness of breath.

Conclusions:

This study broadens the current understanding of oil spill response work hazards by focusing on a more holistic exposure assessment, modeled by LC-derived exposure patterns.

Introduction

On April 20, 2010, the oil drilling rig Deepwater Horizon (DWH) experienced structural failure of its ocean floor cement well resulting in hydrocarbon backflow to the rig’s surface and a subsequent onboard explosion. The United States Coast Guard (USCG) was immediately mobilized to lead a multiagency federal response to cap the leaking well and mitigate the environmental effects of the massive oil spill. The leaking oil well was finally capped on day 87 of the response. An estimated 200 million gallons of oil leaked from the well into the Gulf of Mexico over the course of the spill. By day 165, there were nearly 12 million gallons of oil burned in situ, 35 million gallons of oil recovered, and 1.8 million gallons of dispersant applied ¹.

Previous epidemiologic studies, including the DWH Oil Spill Coast Guard Cohort (DWH-CG) Study ² and the Gulf Long-Term Follow-Up (GuLF) Study ³, have identified a variety of acute symptoms and health conditions associated with widespread exposures (e.g., crude oil, oil dispersants, particulate matter) encountered during the DWHOS cleanup response. In cross-sectional studies, these exposures have been found to be positively associated with acute symptoms and conditions from various organ systems including cardiovascular ⁴, dermatologic ^2,5, eye/ear/nose/throat ⁶, gastrointestinal ^2,7, musculoskeletal ⁸, neurologic ^2,9–11, respiratory ^2,6,12, and urologic ² as well as mental health symptoms ¹³. Oil spill exposures have also been found to be positively associated with incident health conditions including myocardial infarction ^14–17, hypertension ¹⁸, eczema ⁵, tinnitus ¹⁹, peripheral nervous system dysfunction ^19,20, central nervous system dysfunction ²¹, migraine headaches ¹⁹, asthma ^2,22,23, chronic obstructive lung disease ²³, pulmonary function abnormalities ^24,25, hyperlipidemia ²⁶, obesity ²⁶, thyroid disease ²⁶, diabetes mellitus ^27,28, immunologic disorders ²⁹, and mental health conditions ^30,31.

Studies among workers and residents affected by other oil spills, including the Hebei Spirit ^32–36 and Prestige ^37–44, have also found positive associations between oil spill exposures and various dermatologic, neurologic, and respiratory symptoms in oil spill response workers.

These previous investigations have largely focused on individual exposures like crude oil and oil dispersants and the associated symptoms and health conditions. However, oil spill response workers, like most disaster response workers, can be exposed to a wide range of unpredictable hazards during their work that may have an additive or synergistic impact on their health. In the cleanup of the DWHOS, some of the key hazard types included chemical, noise, heat, vector-borne, ergonomic, and psychological ⁴⁵. Some studies, including the DWH-CG and GuLF Study, have attempted to capture this exposure mixture by creating exposure metrics that account for the combination of key exposures like the individual components of crude oil and oil dispersants ^{9,12,14,22,28,29,46}.

In this current study, to more holistically account for the unique mixture of exposures experienced by oil spill response workers, we utilized an exposure metric based on latent classes (LC) previously constructed using a latent class analysis (LCA) ⁴⁷. LCA is a probabilistic modeling algorithm that can identify unique exposure subgroups, also called LCs, within a study population based on several exposure input variables. The purpose of this investigation was to understand the association between LC-derived clustered patterns of hazardous exposure and acute symptoms reported by USCG responders during the DWHOS.

Methods

Data Sources

Data for this analysis are derived from two self-administered, computerized surveys available to USCG responders following their deployment. An earlier survey, “Survey 1,” was launched in June 2010. A later, more detailed survey, “Survey 2,” was launched in November 2010. Many of the variables (e.g., deployment duration and timing, general work tasks, exposures to crude oil, oil dispersants, engine exhaust, personal protective equipment use, and lifestyle factors) were similar between the two surveys, though Survey 1 evaluated exposures and symptoms on an ever/never scale, while Survey 2 evaluated them semi-quantitatively. Responders who completed Survey 1 were also asked to complete Survey 2 when it was launched, and the majority of responders (78%) completed both. The construction of the LC model, described later in more detail, was based on data from both Survey 1 and Survey 2.

Study Population

We carried out a cross-sectional study utilizing data from the DWH-CG Study, which has been described previously ². Briefly, there were 8,686 active duty or Selected Reserve USCG members identified by administrative databases who were deployed to the DWH response for at least one day from April 20, 2010 to December 17, 2010. The study population in the current investigation is the subset of USCG responders who completed Survey 2 (n=4,855).

Exposure Assessment

We utilized a previously developed LC model ⁴⁷ to identify unique patterns of hazardous exposure in the study population. The exposure data used for the LCA was originally collected from the two different post-deployment surveys. In order to harmonize data between Survey 1 and Survey 2 and to increase the power of the LCA, exposure data from Survey 2 were dichotomized to an ever/never scale, and we included responses from both surveys in the analysis (n=5,665). Figure 1 details the development of the LCA and its application to the current study population.

Figure 1:

Creation of latent class model and application to study population. The latent class analysis was conducted using responders who completed either Survey 1 or Survey 2 (n=5,665). In this present investigation, the study population consists of responders who completed Survey 2 (n=4,855). Each responder was then assigned to their most probable latent class.

The methodology for the construction of the LCA has been described previously ⁴⁷. Briefly, from a candidate list of 54 exposure variables, the LCA indicator variables were selected based on the following criteria: (1) high prevalence among the study population, (2) representation from distinct exposure domains (chemical, noise, heat, vector-borne, ergonomic, and psychological), and (3) plausibility of independence from each other within each LC. The following six indicator variables were selected to create the model: crude oil, exhaust fumes, hand sanitizer use, sunscreen use, mosquito bite, and experience of anxiety during deployment. Using these six indicator variables, models that consisted of two, three, four, and five LCs were considered. A four-class model was selected as it had best model fit by Bayesian Information Criterion compared to two-, three-, and five-class models, and had the same entropy as three-class and five-class models. Responders were then assigned to the LC in which they had the highest calculated probability.

Outcome Assessment

Acute symptoms experienced during deployment were queried in Survey 2 only, which is why the current investigation study population included only responders who completed Survey 2. Acute symptoms included in our analysis were from the following organ systems: cardiovascular, dermatologic, eye/ear/nose/throat, gastrointestinal, musculoskeletal, neurologic, and respiratory. Survey 2 elicited these outcomes on a three-level, frequency-based (never, sometimes, most of the time), scale that was then collapsed into binary outcomes on a never/ever scale, where “sometimes” and “most of the time” were combined into the “ever” category to create log-binomial regression models.

Statistical Analysis

We regarded the four LCs as surrogates for different histories of potentially toxic exposures and compared the prevalence of acute symptoms in responders whose most likely class was LC2, LC3, and LC4 to the symptoms of responders whose most likely class was the reference class, LC1, “low overall exposure.” We calculated adjusted prevalence ratios (aPRs) and 95% confidence intervals (CIs) using multivariable log-binomial regression analysis to determine the association between LC and acute symptoms. We selected this statistical method because odds ratios from logistic regression may overestimate the aPR for non-rare outcomes, while log-binomial regression provides a better estimate of the aPR ⁴⁸.

Covariates were included in models for each organ system if they were significantly associated (using log-binomial regression) with both LC and the acute symptoms of the organ system and determined not to be in the causal pathway. This determination was made using direct acyclic graphs as described in Figure 2.

Figure 2:

Sample directed acyclic graph (DAG) demonstrating identification of confounding covariates. Log binomial regression was conducted between each covariate and the latent class variable. If the covariate was significantly associated with the latent class variable (p≤0.05), it was labeled in green. If a covariate was also significantly associated with both the latent class variable and the outcome (acute symptom), it was considered a confounder and labeled pink. Confounding covariates were adjusted for in the final model for each organ system as indicated in Table 2 and Table 3. DAGs were created using the online resource: https://www.dagitty.net/.

We conducted a test for linear trend across LCs to determine if there was a statistically significant linear association by including the four-class LC variable as a pseudo-continuous variable in the model, adjusted for the same confounders as in the categorical analysis. We also conducted a test for interaction in stratified analyses by age (<32 years, ≥32 years, the median age of this cohort) and sex (male, female) to evaluate for effect modification. In stratified analysis, LCs were combined into two categories: low (LC1+LC2) and high (LC3+4). Statistical results were considered significant if test p-values were less than or equal to 0.05.

We conducted sensitivity analyses to determine if the relationship between LC and acute symptoms was impacted by pre-existing health conditions (Supplemental Table 1). Using clinician judgment (MH), we determined which ICD-9 codes (Supplemental Table 2) were likely to be associated with each acute symptom reported during the DWH deployment. The responders who had these diagnoses in their electronic medical record (as early as 2007) prior to their DWH deployment were identified and excluded in sensitivity analyses. Log-binomial regression analysis was then conducted as described above to determine the impact of their exclusion on the association between LC and acute symptoms.

Institutional and Ethics Approval

This study was approved by the Institutional Review Boards (IRB) of the Uniformed Services University (USU), the United States Coast Guard, and the University of North Carolina, Chapel Hill. A waiver for informed consent was approved by the USU IRB.

Results

Baseline characteristics of the study population by the four LCs are presented in Table 1. In LC1, there was a higher frequency of older (35–50 years), educated (at least college degree), officers, short deployers (<30 days), and never smokers compared to other LCs. Frequencies of sex, race, home port, and duty status were comparable across LCs.

Table 1:

Baseline characteristics of the study population by latent class (n=4,855)

	Latent class 1		Latent class 2		Latent class 3		Latent class 4
	"Low overall exposure"		"Low crude oil/exhaust + moderate outdoor time/anxiety"		"High crude oil/exhaust + moderate outdoor time/anxiety"		"High overall exposure"
	N	%	N	%	N	%	N	%
Latent class	653	13.5	701	14.4	1301	26.8	2200	45.3
Age
< 25	97	14.9	123	17.6	223	17.1	384	17.5
25 – 34	252	38.6	310	44.2	560	43.0	1006	45.7
35 – 50	280	42.9	244	34.8	483	37.1	731	33.2
> 50	24	3.7	24	3.4	35	2.7	79	3.6
Sex
Female	115	17.6	148	21.1	156	12.0	309	14.1
Male	538	82.4	553	78.9	1145	88.0	1891	86.0
Race
White	491	75.2	547	78.0	966	74.3	1737	79.0
Black	41	6.3	27	3.9	71	5.5	64	2.9
Asian	26	4.0	28	4.0	59	4.5	69	3.1
Other	23	3.5	36	5.1	72	5.5	111	5.1
Unknown	72	11.0	63	9.0	133	10.2	219	10.0
Education
High school	315	48.2	383	54.6	753	57.9	1289	58.6
Some college	94	14.4	110	15.7	212	16.3	454	20.6
College degree	174	26.7	142	20.3	253	19.5	336	15.3
Masters	51	7.8	38	5.4	52	4.0	73	3.3
Other/unknown	19	2.9	28	4.0	31	2.4	48	2.2
Rank
Officer	281	43.0	213	30.4	387	29.8	430	19.6
Senior enlisted	130	19.9	162	23.1	351	27.0	668	30.4
Junior enlisted	242	37.1	326	46.5	563	43.3	1102	50.1
Duty status
Active duty	431	66.0	406	57.9	946	72.7	1319	60.0
Reserves	222	34.0	295	42.1	355	27.3	881	40.1
Home port
Non-Gulf Coast	445	68.2	558	79.6	881	67.7	1782	81.0
Gulf Coast	208	31.9	143	20.4	420	32.3	418	19.0
Days of deployment
< 30	234	35.8	191	27.3	410	31.5	478	21.7
30 – 59	327	50.1	387	55.2	627	48.2	1268	57.6
>= 60	92	14.1	123	17.6	264	20.3	454	20.6
Smoking status
Never	467	71.5	475	67.8	822	63.2	1395	63.4
Former	107	16.4	103	14.7	221	17.0	357	16.2
Current	79	12.1	123	17.6	258	19.8	448	20.4

Associations between LC and acute symptoms were estimated with aPRs (Table 2). There was a general pattern of increasing aPRs from LC1 to LC4 for acute cardiovascular, dermatologic, eye/ear/nose/throat, gastrointestinal, musculoskeletal, neurologic, and respiratory symptoms. In comparisons between LC3 vs. LC1 and LC4 vs. LC1 nearly all PRs were statistically significant. In all organ systems, there was evidence of an exposure-response relationship (p-trend≤0.05). Some of the strongest associations for individual acute symptoms representative of each organ system included chest pain (aPR_{LC4 vs LC1}=2.37, 95% CI=1.08–5.18; aPR_{LC3 vs LC1}=2.86, 95% CI=1.29–6.36), skin rash/itching (aPR_{LC4 vs LC1}=6.10, 95% CI=3.71–10.01), difficulty hearing (aPR_{LC4 vs LC1}=6.67, 95% CI =3.76–11.81), diarrhea (aPR_{LC4 vs LC1}=6.36, 95% CI=4.00–10.13), muscle pain (aPR_{LC4 vs LC1}=8.11, 95% CI=5.04–13.06), lightheadedness/dizziness (aPR_{LC4 vs LC1}=4.68, 95% CI=3.20–6.84), and shortness of breath (aPR_{LC4 vs LC1}=8.89, 95% CI=3.66–21.60).

Table 2:

Adjusted prevalence ratios and 95% confidence intervals of acute symptoms in each latent class compared to latent class 1

	Latent class 1		Latent class 2		Latent class 3		Latent class 4
	"Low overall exposure"		"Low crude oil/exhaust + moderate outdoor time/anxiety"		"High crude oil/exhaust + moderate outdoor time/anxiety"		"High overall exposure"
	N cases/controls	aPR (95% CI)	N cases/controls	aPR (95% CI)	N cases/controls	aPR (95% CI)	N cases/controls	aPR (95% CI)	P-value for trend
Cardiovascular
Chest pain	7 / 646	Reference	5 / 696	0.66 (0.21 – 2.08)	40 / 1261	2.86 (1.29 – 6.36)	56 / 2144	2.37 (1.08 – 5.18)	0.003
Heartbeat changes	7 / 646	Reference	5 / 696	0.66 (0.21 – 2.08)	33 / 1268	2.36 (1.05 – 5.31)	51 / 2149	2.16 (0.98 – 4.74)	0.007
Dermatologic a
Skin rash/itching	16 / 637	Reference	27 / 674	1.47 (0.80 – 2.71)	133 / 1168	4.09 (2.45 – 6.82)	360 / 1840	6.10 (3.71 – 10.01)	<0.001
Sunburn	8 / 645	Reference	109 / 592	12.03 (5.91 – 24.5)	298 / 1003	17.97 (8.96 – 36.0)	1051 / 1149	35.46 (17.8 – 70.7)	<0.001
Eye/Ear/Nose/Throat b
Difficulty hearing	12 / 641	Reference	12 / 689	0.90 (0.41 – 2.00)	135 / 1166	5.26 (2.93 – 9.45)	289 / 1911	6.67 (3.76 – 11.81)	<0.001
Itchy eyes	30 / 623	Reference	56 / 645	1.72 (1.11 – 2.66)	251 / 1050	4.30 (2.96 – 6.24)	605 / 1595	5.87 (4.09 – 8.44)	<0.001
Ringing in the ears	22 / 631	Reference	24 / 677	0.99 (0.55 – 1.77)	172 / 1129	3.92 (2.51 – 6.12)	368 / 1832	4.80 (3.12 – 7.39)	<0.001
Runny nose	36 / 617	Reference	74 / 627	1.83 (1.24 – 2.70)	241 / 1060	3.39 (2.40 – 4.77)	556 / 1644	4.36 (3.13 – 6.07)	<0.001
Sinus problems	59 / 594	Reference	110 / 591	1.72 (1.27 – 2.32)	325 / 976	2.81 (2.15 – 3.66)	732 / 1468	3.57 (2.76 – 4.60)	<0.001
Sore throat	38 / 615	Reference	84 / 617	2.01 (1.38 – 2.92)	269 / 1032	3.58 (2.57 – 4.98)	657 / 1543	4.98 (3.62 – 6.86)	<0.001
Gastrointestinal c
Constipation	17 / 636	Reference	25 / 676	1.24 (0.67 – 2.30)	81 / 1220	2.52 (1.51 – 4.22)	202 / 1998	3.52 (2.16 – 5.74)	<0.001
Diarrhea	20 / 633	Reference	42 / 659	2.09 (1.21 – 3.61)	155 / 1146	4.11 (2.54 – 6.64)	405 / 1795	6.36 (4.00 – 10.13)	<0.001
Nausea/vomiting	14 / 639	Reference	18 / 683	1.18 (0.57 – 2.41)	80 / 1221	3.19 (1.79 – 5.69)	188 / 2012	4.29 (2.46 – 7.48)	<0.001
Stomach pain	20 / 633	Reference	27 / 674	1.22 (0.68 – 2.20)	108 / 1193	2.76 (1.71 – 4.47)	299 / 1901	4.50 (2.85 –7.11)	<0.001
Musculoskeletal d
Back pain	60 / 593	Reference	89 / 612	1.41 (1.04 – 1.93)	336 / 965	2.88 (2.23 – 3.73)	746 / 1454	3.80 (2.97 – 4.87)	<0.001
Difficulty walking	18 / 546	Reference	20 / 681	1.10 (0.58 – 2.06)	77 / 1224	2.28 (1.37 – 3.77)	151 / 2049	2.69 (1.66 – 4.35)	<0.001
Joint pain	28 / 625	Reference	58 / 643	2.00 (1.29 – 3.10)	244 / 1057	4.51 (3.09 – 6.60)	527 / 1673	5.83 (4.03 – 8.44)	<0.001
Joint stiffness	23 / 630	Reference	34 / 667	1.43 (0.85 – 2.40)	162 / 1139	3.71 (2.42 – 5.69)	344 / 1856	4.74 (3.13 – 7.16)	<0.001
Joint weakness	15 / 638	Reference	25 / 676	1.56 (0.83 – 2.94)	117 / 1184	3.99 (2.35 – 6.78)	269 / 1931	5.44 (3.26 – 9.10)	<0.001
Muscle pain	17 / 636	Reference	46 / 655	2.53 (1.46 – 4.37)	182 / 1119	5.52 (3.39 – 9.00)	453 / 1747	8.11 (5.04 – 13.06)	<0.001
Neurologic e
Blurred vision	17 / 636	Reference	22 / 679	1.18 (0.63 – 2.21)	61 / 1240	1.84 (1.08 – 3.12)	118 / 2082	2.09 (1.26 – 3.44)	<0.001
Difficulty concentrating	31 / 622	Reference	34 / 667	1.00 (0.62 – 1.62)	155 / 1146	2.57 (1.77 – 3.74)	248 / 1952	2.41 (1.67 – 3.46)	<0.001
Lightheadedness/ dizziness	27 / 626	Reference	29 / 672	0.97 (0.58 – 1.63)	167 / 1134	3.17 (2.13 – 4.71)	420 / 1780	4.68 (3.20 – 6.84)	<0.001
Headache	107 / 546	Reference	163 / 538	1.41 (1.13 – 1.76)	521 / 780	2.47 (2.05 – 2.97)	1100 / 1100	3.07 (2.57 – 3.67)	<0.001
Memory loss	9 / 644	Reference	6 / 695	0.62 (0.22 – 1.73)	36 / 1265	2.00 (0.97 – 4.14)	76 / 2124	2.50 (1.26 – 4.97)	<0.001
Numbness/tingling	10 / 643	Reference	14 / 687	1.29 (0.58 – 2.90)	55 / 1246	2.78 (1.42 – 5.42)	150 / 2050	4.47 (2.37 – 8.43)	<0.001
Respiratory f
Coughing	42 / 611	Reference	69 / 632	1.43 (0.99 – 2.07)	243 / 1058	2.85 (2.08 – 3.90)	588 / 1612	3.86 (2.85 – 5.22)	<0.001
Shortness of breath	5 / 648	Reference	18 / 683	3.07 (1.14 – 8.22)	73 / 1228	7.00 (2.84 – 17.26)	169 / 2031	8.89 (3.66 – 21.60)	<0.001
Wheezing	7 / 646	Reference	7 / 694	0.85 (0.30 – 2.42)	52 / 1249	3.55 (1.62 – 7.78)	107 / 2093	4.04 (1.88 – 8.67)	<0.001

^a.Adjusted for rank and duty status

^b.Adjusted for sex, education, duty status, and smoking status

^c.Adjusted for sex, education, and smoking status

^d.Adjusted for age and sex

^e.Adjusted for sex

^f.Adjusted for sex, rank, duty status, and smoking status

N=number of responders; aPR=adjusted prevalence ratio; CI=confidence interval

We conducted sensitivity analyses to determine if the relationship between LC and acute symptoms was impacted by pre-existing health conditions (Supplemental Table 1). All comparisons remained statistically significant, and the magnitude of association was attenuated by greater than 10% in only 1 out of 56 comparisons. This comparison was for chest pain where the association was no longer significant when excluding pre-existing conditions (aPR_{LC4 vs LC1}=1.85, 95% CI=0.83–4.09). Sensitivity analyses were not conducted for sunburn since this acute symptom is unlikely to represent a chronic, pre-existing health condition that would require exclusion.

In analyses stratified by age at deployment (Table 3), there was a general pattern of more strongly elevated aPRs for musculoskeletal symptoms among responders <32 years of age than for responders ≥32 years of age. Statistically significant effect modification was found only for back pain (p=0.013) and difficulty walking (p=0.035). In sex-stratified analyses (Table 3), males had more highly elevated aPRs for most symptoms compared to females, with statistically significant effect modification found for runny nose (p=0.014), sore throat (p=0.012), diarrhea (p=0.035), stomach pain (p=0.012), back pain (p=0.027), difficulty concentrating (p=0.010), lightheadedness/dizziness (p=0.002), and headache (p<0.001).

Table 3:

Adjusted prevalence ratios and 95% confidence intervals of acute symptoms comparing high to low latent classes stratified by age and sex

	Age < 32 (n=2,551)		Age >=32 (n=2,304)			Female (n=728)		Male (n=4,127)
	Low LC	High LC	Low LC	High LC	Int. p-value	Low LC	High LC	Low LC	High LC	Int. p-value
Cardiovascular					0.338
Chest pain	Ref.	4.46 (1.62 – 12.31)	Ref.	2.41 (1.14 – 5.09)	0.130	Ref.	1.58 (0.57 – 4.34)	Ref.	4.20 (1.95 – 9.07)	0.131
Heartbeat changes	Ref.	6.65 (1.61 – 27.52)	Ref.	1.97 (1.00 – 3.89)		Ref.	1.83 (0.60 – 5.57)	Ref.	3.18 (1.54 – 6.60)	0.416
Dermatologic a					0.229
Skin rash/itching	Ref.	5.34 (3.19 – 8.93)	Ref.	3.71 (2.53 – 5.44)	0.145	Ref.	3.90 (2.17 – 7.00)	Ref.	4.50 (3.14 – 6.45)	0.629
Sunburn	Ref.	3.72 (2.97 – 4.67)	Ref.	4.98 (3.73 – 6.64)		Ref.	3.42 (2.31 – 5.07)	Ref.	4.39 (3.59 – 5.36)	0.308
Eye/Ear/Nose/Throat b					0.776
Difficulty hearing	Ref.	7.07 (3.64 – 13.72)	Ref.	6.24 (3.72 – 10.45)	0.309	Ref.	7.41 (2.71 – 20.25)	Ref.	6.29 (4.03 – 9.82)	0.773
Itchy eyes	Ref.	4.43 (3.10 – 6.31)	Ref.	3.57 (2.72 – 4.67)	0.786	Ref.	2.74 (1.87 – 4.03)	Ref.	4.34 (3.34 – 5.63)	0.054
Ringing in the ears	Ref.	4.84 (2.94 – 7.98)	Ref.	4.39 (3.01 – 6.41)	0.15	Ref.	3.60 (1.81 – 7.15)	Ref.	4.71 (3.36 – 6.59)	0.494
Runny nose	Ref.	2.40 (1.84 – 3.12)	Ref.	3.22 (2.44 – 4.25)	0.989	Ref.	1.90 (1.35 – 2.67)	Ref.	3.20 (2.54 – 4.04)	0.014
Sinus problems	Ref.	2.37 (1.89 – 2.98)	Ref.	2.41 (1.97 – 2.96)	0.848	Ref.	2.04 (1.53 – 2.71)	Ref.	2.50 (2.09 – 2.99)	0.24
Sore throat	Ref.	2.83 (2.18 – 3.68)	Ref.	2.99 (2.34 – 3.84)		Ref.	1.99 (1.43 – 2.76)	Ref.	3.32 (2.68 – 4.13)	0.012
Gastrointestinal c					0.645
Constipation	Ref.	3.14 (1.82 – 5.43)	Ref.	2.70 (1.80 – 4.05)	0.201	Ref.	2.36 (1.42 – 3.91)	Ref.	3.12 (2.03 – 4.78)	0.423
Diarrhea	Ref.	2.93 (2.01 – 4.26)	Ref.	4.23 (2.91 – 6.13)	0.629	Ref.	2.09 (1.23 – 3.55)	Ref.	4.05 (2.98 – 5.51)	0.035
Nausea/vomiting	Ref.	3.19 (1.94 – 5.23)	Ref.	3.99 (2.27 – 7.04)	0.434	Ref.	2.72 (1.52 – 4.85)	Ref.	4.15 (2.54 – 6.78)	0.271
Stomach pain	Ref.	3.05 (1.98 – 4.71)	Ref.	3.92 (2.54 – 6.03)		Ref.	1.91 (1.14 – 3.21)	Ref.	4.40 (2.99 – 6.47)	0.012
Musculoskeletal d Back pain					0.013
Difficulty walking	Ref.	3.77 (2.84 – 5.00)	Ref.	2.43 (2.00 – 2.94)	0.035	Ref.	2.07 (1.51 – 2.84)	Ref.	3.13 (2.60 – 3.77)	0.027
Joint pain	Ref.	5.08 (2.23 – 11.59)	Ref.	1.91 (1.31 – 2.78)	0.351	Ref.	3.10 (1.31 – 7.31)	Ref.	2.29 (1.58 – 3.30)	0.052
Joint stiffness	Ref.	4.07 (2.81 – 5.88)	Ref.	3.27 (2.52 – 4.25)	0.13	Ref.	2.75 (1.74 – 4.34)	Ref.	3.76 (2.95 – 4.80)	0.234
Joint weakness	Ref.	4.90 (2.92 – 8.21)	Ref.	3.10 (2.27 – 4.23)	0.29	Ref.	2.91 (1.71 – 4.95)	Ref.	3.79 (2.79 – 5.17)	0.392
Muscle pain	Ref.	4.63 (2.76 – 7.78)	Ref.	3.29 (2.18 – 4.94)	0.296	Ref.	2.67 (1.45 – 4.88)	Ref.	4.26 (2.91 – 6.23)	0.192
Neurologic e	Ref.	4.68 (3.15 – 6.97)	Ref.	3.53 (2.55 – 4.89)		Ref.	2.75 (1.76 – 4.29)	Ref.	4.60 (3.38 – 6.25)	0.064
Blurred vision					0.517
Difficulty concentrating	Ref.	2.03 (1.20 – 3.43)	Ref.	1.69 (1.08 – 2.66)	0.489	Ref.	1.46 (0.76 – 2.79)	Ref.	1.96 (1.31 – 2.95)	0.444
Lightheadedness/ dizziness	Ref.	2.78 (1.83 – 4.22)	Ref.	2.34 (1.70 – 3.23)	0.947	Ref.	1.52 (0.99 – 2.32)	Ref.	3.06 (2.21 – 4.24)	0.01
Headache	Ref.	4.16 (2.89 – 5.98)	Ref.	4.11 (2.77 – 6.09)	0.795	Ref.	2.39 (1.60 – 3.56)	Ref.	5.71 (3.97 – 8.20)	0.002
Memory loss	Ref.	2.38 (2.03 – 2.81)	Ref.	2.31 (1.98 – 2.70)	0.263	Ref.	1.67 (1.39 – 2.02)	Ref.	2.69 (2.34 – 3.09)	<0.001
Numbness/tingling	Ref.	5.09 (1.58 – 16.37)	Ref.	2.46 (1.34 – 4.52)	0.98	Ref.	1.47 (0.53 – 4.07)	Ref.	3.55 (1.86 – 6.79)	0.152
Respiratory f	Ref.	3.39 (1.64 – 6.99)	Ref.	3.50 (2.10 – 5.85)		Ref.	4.52 (1.61 – 12.65)	Ref.	3.10 (1.96 – 4.91)	0.513
Coughing					0.597
Shortness of breath	Ref.	2.93 (2.24 – 3.85)	Ref.	2.71 (2.08 – 3.53)	0.385	Ref.	2.25 (1.57 – 3.23)	Ref.	3.04 (2.44 – 3.80)	0.176
Wheezing	Ref.	3.13 (1.81 – 5.39)	Ref.	5.11 (2.59 – 10.06)	0.798	Ref.	3.08 (1.39 – 6.79)	Ref.	4.17 (2.51 – 6.91)	0.476
	Ref.	3.74 (1.63 – 8.58)	Ref.	4.56 (2.21 – 9.38)		Ref.	2.40 (0.92 – 6.25)	Ref.	5.16 (2.63 – 10.12)	0.208

^a.Adjusted for rank and duty status

^b.Adjusted for sex, education, duty status, and smoking status

^c.Adjusted for sex, education, and smoking status

^d.Adjusted for age and sex

^e.Adjusted for sex

^f.Adjusted for sex, rank, duty status, and smoking status

Ref=reference group; Low LC=latent classes 1+2; High LC=latent classes 3+4; Int. p-value=interaction p- value

Discussion

In this cross-sectional study, we found that oil spill exposure patterns, as represented by LCs of increasing hazardous exposure, were positively associated with a variety of acute symptoms in DWHOS responders. Some of the strongest associations by organ system included chest pain, skin rash/itching, difficulty hearing, diarrhea, muscle pain, lightheadedness/dizziness, and shortness of breath. Stratified analyses showed that younger responders were more likely to report acute musculoskeletal symptoms compared to older responders. In male responders, the strength of association between LC and acute symptom was generally stronger across all organ systems compared to the association in female responders. Sensitivity analyses for pre-existing health conditions demonstrated the same patterns.

Previous epidemiologic studies have found that common oil spill exposures such as crude oil and oil dispersants were associated with acute symptoms across a variety of organ systems. However, the magnitude of the aPRs reported here are generally greater compared to those in prior studies from the DWH-CG that focused on individual exposures. For example, a cross-sectional study identified positive associations between crude oil exposure by any route and skin rash/itching (aPR_{high oil vs low oil}=1.87, 95% CI=1.45–2.40), numbness/tingling (aPR_{high oil vs low oil}=1.26, 95% CI=0.88–1.82), and shortness of breath (aPR_{high oil vs low oil}=2.30, 95% CI=1.51–3.56) ². In contrast, this investigation identified stronger associations with skin rash/itching (aPR_{LC4 vs LC1}=6.10, 95% CI=3.71–10.01), numbness/tingling (aPR_{LC4 vs LC1}=4.47, 95% CI=2.37–8.43), and shortness of breath (aPR_{LC4 vs LC1}=8.89, 95% CI=3.66–21.60) using LC-derived exposure patterns. The greater strength of association reported here may be due to the holistic exposure assessment conducted in this investigation where the impact of simultaneous exposures was considered.

Cardiovascular Symptoms

Chest pain (aPR_{LC3 vs LC1}=2.86, 95% CI=1.29–6.36; aPR_{LC4 vs LC1}=2.37, 95% CI=1.08–5.18) may be attributable to a variety of factors. The increased physical demands of the LC3 and LC4 missions compared to the LC1 mission ⁴⁷ could result in chest wall pain, the most common cause of clinical chest pain ⁴⁹. Cardiac chest pain, though less likely in a relatively young and healthy population, is a potential medical emergency. There are several studies now linking oil spill-related exposures to acute coronary events ^14–17. The strength of association between LC and acute chest pain was attenuated somewhat after accounting for pre-existing cardiovascular conditions (Supplemental Table 1). This could indicate that some responders’ reported acute chest pain represents exacerbation of a pre-existing condition.

Dermatologic Symptoms

Skin rash/itching (aPR_{LC4 vs LC1}=6.10, 95% CI=3.71–10.01) could be caused by dermal contact with crude oil and other oil spill-related chemicals given that LC4 responders were more likely to participate in oil-based missions compared to LC1 responders ⁴⁷. Polycyclic aromatic hydrocarbons, a component of crude oil, have been shown to act as a skin irritant and sensitizer that can trigger or exacerbate skin diseases ⁵⁰. Chemical dispersants were used extensively in the DWHOS cleanup response and have been shown to act as an irritant and sensitizer in animal models ⁵¹. A variety of other chemical exposures with soaps, detergents, and cleaners may cause acute skin symptoms in oil spill response workers ⁵².

Eye/Ear/Nose/Throat Symptoms

Difficulty hearing (aPR_{LC4 vs LC1}=6.67, 95% CI =3.76–11.81) could be explained by exposure to environmental noise and to ototoxic chemicals. LC4 responders were more likely to encounter increased noise from diesel engines and other equipment due to the nature of their mission compared to LC1 responders ⁴⁷. The causal relationship between occupational noise exposure and hearing loss is well established ⁵³. LC4 responders were also more likely to encounter crude oil and its ototoxic components including toluene, ethylbenzene, and xylenes ⁵⁴. Exposure to ototoxic chemicals may act synergistically with environmental noise to cause hearing loss ⁵⁵.

Gastrointestinal Symptoms

Diarrhea (aPR_{LC4 vs LC1}=6.36, 95% CI=4.00–10.13) and other acute gastrointestinal symptoms may occur with inhalational exposure to components of crude oil, particularly naphthalene ⁵⁶ and xylene ⁵⁷. A recent study in the DWH-CG found a positive exposure-response relationship for crude oil exposure via dermal and inhalation pathways and acute gastrointestinal symptoms, including diarrhea ⁷. A possible mechanism is the disruption of the human gut microbiome. Human gut microbiome samples have demonstrated increased levels of pathogenic bacteria and decreased levels of beneficial bacteria when exposed to crude oil ⁵⁸. Anxiety may be another contributing factor to the prevalence of diarrhea and acute gastrointestinal symptoms given LC3 and LC4 responders were three- and four-times more likely, respectively, to report anxiety compared to LC1 responders ⁴⁷, and emotional and cognitive centers in the brain can influence peripheral gastrointestinal functions through the bidirectional neural, endocrine, and immune pathways known as the gut-brain axis ⁵⁹.

Musculoskeletal Symptoms

Muscle pain (aPR_{LC4 vs LC1}=8.11, 95% CI=5.04–13.06) and other musculoskeletal symptoms could be related to both acute and chronic physical stressors experienced during deployment. A recent cross-sectional study using the same LC-derived exposure patterns utilized in this investigation showed that LC4 responders were significantly more likely to experience slips, trips, and falls compared to LC1 responders, which could explain the increase in acute musculoskeletal symptoms such as back pain or muscle pain ⁸. Furthermore, the mission requirements of LC4 responders were more likely to involve heavy lifting and greater ergonomic stress compared to LC1 responders ⁴⁷.

Neurologic Symptoms

Lightheadedness/dizziness (aPR_{LC4 vs LC1}=4.68, 95% CI=3.20–6.84) may be attributed to a variety of factors including heat stress and the neurotoxic effects of chemical exposures. DWH responders experienced environmental heat and symptoms of heat stress during deployment ⁶⁰. LC4 responders likely experienced more environmental heat as they were more likely to work outdoors compared to LC1 responders ⁴⁷. The crude oil components xylene and toluene may cause lightheadedness/dizziness and other neurologic symptoms via inhalation ⁶¹. A recent cross-sectional study identified inhalational and dermal exposure to crude oil to be significantly associated with lightheadedness/dizziness and other acute neurologic symptoms in DWH responders ⁹.

Respiratory Symptoms

Shortness of breath (aPR_{LC4 vs LC1}=8.89, 95% CI=3.66–21.60) can occur with inhalation of volatile organic compounds that cause pathologic changes to the respiratory system. A cross-sectional study in the DWH-CG found an exposure-response relationship with crude oil inhalational exposure and shortness of breath ¹². A follow-up study identified crude oil via the inhalation pathway as a risk factor for chronic obstructive pulmonary disease and possible reactive airway disease including asthma ²³. These health conditions may be the result of pathological changes in airway epithelial cells that have been shown to occur after in vitro treatment with crude oil and dispersants ⁶².

Effect Modification

In stratified analyses, there was evidence of effect modification by age in associations between LC and acute musculoskeletal symptoms. Younger responders showed a stronger positive association between LC and back pain than older responders (p-int=0.013). This is likely not due to acute injuries as older responders in this study population were more likely to experience slips/trips/falls compared to younger responders ⁸. An occupational case-crossover study demonstrated that younger workers working with heavy loads are more likely to develop acute low back pain compared to older workers ⁶³. It is possible that older workers have more experience lifting heavy loads and the important safety techniques required to prevent back injury or that older responders were assigned to take on less physically arduous tasks compared to younger responders during their deployment.

In analyses stratified by sex, the association between LC and acute symptoms was stronger in males for runny nose (p-int=0.014), sore throat (p-int=0.012), diarrhea (p-int=0.035), stomach pain (p-int=0.012), back pain (p-int=0.027), difficulty concentrating (p-int=0.010), lightheadedness/dizziness (p-int=0.002), and headache (p-int<0.001) than in females. That these differences occur across nearly all organ systems may indicate a systematic difference in the exposure patterns of male and female responders instead of a distinct biological mechanism that modulates the association between LC and acute symptoms.

Strengths and Limitations

This investigation has several strengths. LCA is a relatively novel approach to develop a holistic exposure assessment that captures the mixture of exposures experienced by oil spill response workers. The positive associations found in this analysis across many organ systems indicate the utility of LCA in capturing the impact of complex exposure mixtures on the human body as a whole. The relatively large sample size (n=4,855) provided adequate power for most comparisons including those in stratified analyses. This work includes acute symptoms that have been understudied among oil spill response workers such as eye/ear/nose/throat and musculoskeletal symptoms. The study population was relatively young and healthy with few comorbidities, making confounding by pre-existing health conditions less likely. Additionally, we expect minimal risk of selection and recall bias, as responders who completed the survey versus did not complete the survey were generally similar ². The DWH-CG Study contains data linkage to electronic medical records which enabled sensitivity analyses to exclude pre-existing health conditions and can allow for follow-up studies on incident health conditions.

This investigation should be interpreted considering several potential limitations. Due to the cross-sectional study design, there is limited understanding of the temporal relationship between LC exposure patterns and acute symptoms. It is possible that pre-existing health conditions, rather than exposure patterns experienced during deployment, contributed to the reported acute symptoms. However, our sensitivity analyses showed a generally minor impact on effect sizes when responders with pre-existing conditions were excluded. Although the post-deployment survey was administered immediately after deployment, recall bias is possible since responders were queried on exposures and acute symptoms that occurred in the preceding several months. The median completion date for Survey 2 was 185 days post-deployment. Some acute symptoms had small counts (n<10) such as chest pain, heartbeat changes, sunburn, memory loss, shortness of breath, and wheezing, which increases the uncertainty of these comparisons. Because of the large number of comparisons we performed, some of the associations we found to be statistically significant may have occurred by chance. However, we attempt to mitigate this problem by focusing on and describing patterns of results, rather than individual, statistically significant findings. Finally, the rank order of the latent classes we used as the exposure of interest in this study is not the rank order for every exposure input variable included in the latent class analysis. For example, while the probabilities of exposure to crude oil, exhaust/carbon monoxide, mosquito bites, and experience of anxiety increase across the four LCs, they do not for sunscreen use or hand sanitizer use.

Conclusions

This study broadens the current understanding of oil spill response work hazards by focusing on a more holistic exposure assessment, based on LC-derived exposure patterns, rather than individual exposures. This novel approach represents a more realistic scenario of exposure and could be applied in other study populations and future investigations of oil spill-related health. Future studies may consider longitudinal follow-up and correlation between self-reported symptoms and incident, objectively ascertained, health conditions.

Data availability

Since these data are from military data resources, the process for obtaining assurances for public use requires authorizations by the Department of Defense and the United States Coast Guard. The corresponding author can provide additional information on the request process.