Study population and study design
The DWH-CG Cohort has been described in detail elsewhere . Briefly, this cohort was established with an aim of examining associations between the DWH oil spill response and both acute health symptoms and longer-term health conditions. The DWH-CG Cohort consists of 8696 USCG members who responded to the DWH oil spill for at least one day (i.e., responders) and 44823 USCG personnel who were either on active duty or in the Selected Reserve for at least one day during the main operational phase of the cleanup (April 20, 2010 – December 17, 2010 ), but who did not deploy to the spill (i.e., non-responders). For the current prospective study, we were able to include only active duty responders (n=5964, 68.6%) and non-responders (n=39260, 87.6%) because only active duty military personnel, and not the Select Reservists, have comprehensive medical coverage through the Military Health System (MHS) and, thus, ongoing health encounter data available for querying neurological diagnoses. Detailed information about the benefits of MHS, a healthcare system designed for equal access, was previously published .
This study was approved by the Institutional Review Boards (IRB) of the Uniformed Services University (USU), the U.S. Coast Guard, and the University of North Carolina, Chapel Hill. A waiver for informed consent was approved by the USU IRB.
Our first exposure metric included the spill response work comparing all active duty responders (n=5964) to non-responders (n=39260). The rest of the oil spill related exposures were applicable to responders only. For the within-responder comparisons, we relied on self-reported exposures to crude oil/oily water (henceforth referred to as “crude oil”) and oil dispersants (henceforth referred to as “dispersants”) that were ascertained from two, previously described , post-deployment surveys.
Briefly, Survey 1 was launched on June 25, 2010 and Survey 2 on November 1, 2010. The questions on the two surveys were similar, however, Survey 1 assessed self-reported exposures to crude oil via different routes (i.e., inhalation, direct skin contact, ingestion, and submersion) on a binary scale (never, ever), while Survey 2 evaluated these crude oil exposures on a 5-point Likert scale (never, rarely, sometimes, most of the time, and all of the time). Self-reported exposure of coming into contact with dispersants, also on a 5-point Likert scale, was ascertained only on Survey 2. There were 3492 active duty responders who completed at least one of the two surveys, however, 390 active duty responders completed only Survey 1, and therefore had no information on dispersants.
For the within-responder analyses, we used several survey-based exposure metrics: 1) crude oil exposure via any route (i.e., inhalation, direct skin contact, ingestion, or submersion), 2) crude oil exposure via inhalation, 3) crude oil exposure via direct skin contact, 4) crude oil exposure via submersion, and 5) combined crude oil/dispersants exposure. For the crude oil exposure via any route and via submersion (n=3492), we combined self-reported data from both post-deployment surveys and created the following exposure metrics: ever (combining "ever" from Survey 1 and “rarely,” "sometimes," "most of the time,” or "all of the time" from Survey 2) vs. never (combining "never" from Survey 1 and "never" from Survey 2). For analyses relating to crude oil exposure via inhalation and via direct skin contact (n=3492), we combined self-reported data from both surveys and created the following exposure metrics: ever (combining "ever" from Survey 1 and "sometimes," "most of the time,” or "all of the time" from Survey 2) vs. never (combining "never" from Survey 1 and "never" or "rarely" from Survey 2). We combined the responses of “rarely” and “never” for the crude oil exposures via inhalation and via direct skin contact metrics because these two exposures were the most commonly reported.
For the combined crude oil/dispersants exposure metric, we created the following exposure groups: “Neither” (i.e., reported “never” exposure to crude oil via any route and “never” exposure to dispersants); “Oil only” (i.e., reported “ever” exposure to crude oil via any route and “never” exposure to dispersants); and “Both” (i.e., reported “ever” exposure to crude oil via any route and exposure to dispersants of “rarely”, “sometimes,” “most of the time,” or “all of the time”). For this comparison, the “Neither” group was the referent category. Because there were only 20 responders who reported any exposure to dispersants (i.e., “rarely” or greater) but no exposure to crude oil (i.e., “never” exposure to crude oil via any route), we did not create a “Dispersants only” exposure category.
We ascertained neurological outcomes from the Military Health System Data Repository (MDR), a medical health encounter data repository maintained by the military [40, 45, 46]. The MDR contains data from inpatient and outpatient health encounters obtained in both military treatment facilities and clinics (“direct care”) and in civilian treatment facilities for which care is billed to the military (“purchased care”). For all active duty cohort members, we obtained full medical health encounter coverage for a period between October 1, 2007 (i.e., ~ two and a half years before the DWH spill) and September 30, 2015 (i.e., five and a half years post-spill) by combining MDR data from four major sources: 1) inpatient direct/military care, 2) outpatient direct/military care, 3) inpatient purchased/civilian care, and 4) outpatient purchased/civilian care. For the time period we queried, health encounter MDR diagnoses were coded using the International Classification of Diseases, 9th Revision (ICD-9) codes. We focused on chronic neurological diseases and symptoms classified by three-, four-, or five-digit ICD-9 codes. We considered individual and grouped ICD-9 codes for various neurological conditions including migraines and headaches, conditions impairing memory, signs and symptoms involving cognition, peripheral nerve disorders, visual, hearing, and disturbance of skin sensation, as well as conditions affecting balance/gait. A full listing of individual and grouped diseases and symptoms that we evaluated, along with the corresponding ICD-9 codes, is provided in Supplemental Table 1.
The incident case definition for classifying neurological outcomes included at least one inpatient encounter or two outpatient encounters for a specific individual neurological disease/symptom or a group of neurological diseases/symptoms. We only retained outcomes with at least 10 cases among all responders and 10 cases among all non-responders, in order to avoid data sparsity issues. Prevalent cases among the responders and non-responders who had a pre-existing neurological condition documented in MDR before the spill (October 1, 2007 - April 20, 2010), ascertained via the same case definition as a post-DWH incident case, were excluded from all analyses of that particular neurological outcome.
For our main analyses, we included ICD-9 codes in any diagnostic position. We conducted multivariable Cox Proportional Hazards regression analyses to model associations between the oil spill exposures described above and risk of neurological diseases/symptoms by calculating adjusted hazard ratios (aHRs) and 95% confidence intervals (95% CI). The within-responder comparisons (ever vs. never crude oil exposure metrics and combined crude oil/dispersants comparisons) were adjusted for age at baseline (years), sex (male, female), race (white, Black, other/unknown), and smoking status (never, former, current, unknown) based on prior literature [40, 44]. The responder vs. non-responder models were adjusted only for age, sex, and race because smoking information was not available for non-responders or for responders who did not complete a post-deployment survey (n=2472). We calculated p-values for linear trend (p-trend) for the combined oil/dispersants exposure treating it as a continuous variable in the Cox regression models.
For the responder vs. non-responder comparison, the start of follow-up time for all cohort members was the later of April 20, 2010 or the USCG entry date. Responders contributed events and person-time as non-responders until the first day of their spill deployment. Because responders may have sought care outside of the MHS during the DWH spill deployment, their health encounters may have not been recorded in a systematic way. To account for this potential issue, we excluded responder events and person-time during deployment from the study observational period. From the day after their deployment ended, responders contributed events and person-time to the responder group. For the within-responder comparisons, the start of follow-up time for all responders was the day after the last day of their spill deployment. The end of follow-up time for all study comparisons was the earliest of 1) the date of becoming a case of a particular neurological condition, 2) the end of follow-up period, i.e., September 30, 2015, or 3) the USCG exit date.
We tested the assumption of proportionality of hazards across the entire follow-up period (April 20, 2010/end of deployment through September 30, 2015) by evaluating Pearson correlations between Schoenfeld residuals and follow-up time. A p-value < 0.05 for the corresponding Pearson correlation coefficient suggested non-proportionality of hazards. In the cases where the proportionality assumption was violated, we calculated aHRs and 95% CIs for two approximately equal-length time periods: April 20, 2010/end of deployment –December 31, 2012 (i.e., the earlier period) and January 1, 2013 –September 30, 2015 (i.e., the later period).
To evaluate the robustness of the main associations, we performed four sensitivity analyses. First, we refined incident case definitions by restricting the relevant ICD-9 codes to either the first or the second diagnostic position instead of to any diagnostic position. For the second sensitivity analysis, we excluded cohort members who were potentially exposed to occupational hazards and, therefore, under more intensive periodic medical surveillance through enrollment in the Coast Guard’s Occupational Medical Surveillance and Evaluation Program (OMSEP) from before the DWH oil spill through the end of the study follow-up period. USCG members with occupations with a high likelihood for occupational exposure to known or suspected toxins (e.g., benzene) require enrollment in OMSEP and are followed more closely through baseline and periodic physical examinations in accordance with the Occupational Safety and Health Administration (OSHA) requirements . We excluded OMSEP enrollees who were in the program due to occupational exposures to benzene, chromium compounds, lead, pesticides, and/or solvents because these cohort members could be a higher risk group for developing neurological conditions due to their usual occupational exposures and could, therefore, bias our risk estimates for neurological conditions. Because tobacco smoke contains some of the same constituents as crude oil (i.e., benzene, PAHs, heavy metals) , for our third sensitivity analysis, we restricted the within-responder comparisons for the ever/never crude oil exposure via inhalation to those responders who reported never smoking. This restriction allowed us to rule out any potential residual confounding by smoking. Lastly, to assess exposure-response relationships between increasing levels of self-reported crude oil inhalation exposure and neurological outcomes, we performed a sensitivity analysis restricted to spill responders with Survey 2 data (n=3102).
All analyses were performed in SAS Version 9.4 (SAS Institute, Cary, NC, USA).