8
Analysis Structure
OVERVIEW
The analysis plan for the Project SHAD (Shipboard Hazard and Defense) study was structured to check data validity, test hypotheses, and interactively explore data to follow leads arising from data analysis. The study was designed to address: (1) whether mortality (both cause-specific and overall) differed between Project SHAD participants and nonparticipants; (2) whether morbidity differed between Project SHAD participants and nonparticipants; and (3) whether mortality and morbidity differed among specific Project SHAD exposure groups.
The basic comparison involves the mortality and morbidity experiences of Project SHAD participants relative to that of referent cohort members. A number of measures from the study questionnaire were used to ascertain morbidity while fact of death and cause-specific mortality data were identified from the National Death Index (NDI), the Social Security Administration (SSA) Death Master File, and the Department of Veterans Affairs (VA) Beneficiary Identification and Records Locator Subsystem (BIRLS) file (see Chapter 4 regarding these sources).
AVAILABLE DATA
Data available for the analyses consist of measures or indicators of (1) presumed exposure; (2) demographic, lifestyle, and military service characteristics that might confound an association between exposure and outcome; (3) morbidity outcomes; and (4) mortality outcome. Table 8-1 presents the variables that were included in the analysis dataset. It should be noted that variables were not all of the same quality with regard to completeness and validity.
The variables included in the basic analyses are participant status, SHAD participant exposure group, age, race, branch of service, pay grade, smoking, drinking, body mass index (BMI; weight in kilograms divided by height squared, in meters), vital status, date of death, cause of death, and SF-36 score. Analyses also explore relationships using variables such as SF-36 subscale scores, Neuropsychological Impairment Scale (NIS) scores, Structured Clinical Interview for DSM-IV (SCID) somatization scores, and history of chronic medical conditions or symptoms. (The definitions and rationale for the use of these outcome variables are described in Chapter 7.)
As there were a large number of morbidity outcome variables collected in the questionnaire, the morbidity variables were categorized into primary, secondary, and tertiary outcomes. These categories were developed based on consultation with the advisory panel. Table 8-2 shows the list of primary, secondary, and tertiary outcome
TABLE 8-1 Variables Considered for Analysis and Their Sources
Variable |
Sample value |
Source |
Participant status |
Participant |
Military records |
Race |
White, nonwhite |
Military records |
Current marital status |
Single |
Questionnaire |
Education |
Bachelor degree |
Questionnaire |
Height |
5’7” |
Questionnaire |
Current weight |
175 lbs |
Questionnaire |
Date of birth |
1/2/1945 |
Military record/questionnaire |
General health status |
Excellent |
Questionnaire |
SF-36 score* |
|
Questionnaire |
SCID Somatization Scale score* |
|
Questionnaire |
Neuropsychological Impairment Scale score* |
|
Questionnaire |
History of 45 chronic medical conditions* |
Yes |
Questionnaire |
History of 12 general health problems* |
Yes |
Questionnaire |
History of 19 symptoms within past year* |
Yes |
Questionnaire |
Hospitalization while in Navy |
Yes |
Questionnaire |
Number of hospitalizations while in Navy |
3 |
Questionnaire |
Hospitalizations since discharge from active duty |
Yes |
Questionnaire |
Number of hospitalizations since discharge from active duty |
2 |
Questionnaire |
Length of time since last hospitalization |
More than 5 years ago |
Questionnaire |
Biological father of any pregnancy |
Yes |
Questionnaire |
Number of live birth pregnancies |
2 |
Questionnaire |
Number of children with birth defects |
0 |
Questionnaire |
Ever smoked |
Yes |
Questionnaire |
Current smoker |
Yes |
Questionnaire |
Age stopped smoking |
40 |
Questionnaire |
Years of smoking |
5 |
Questionnaire |
Cigarettes smoked/day |
7 |
Questionnaire |
Current drinker |
No |
Questionnaire |
Frequency of drinking |
3–4 times per week |
Questionnaire |
Problems with alcohol (series of 3 questions) |
Yes |
Questionnaire |
Ever drinker |
Yes |
Questionnaire |
Age stopped drinking |
35 |
Questionnaire |
Date of entry into military |
2/1965 |
Military record/questionnaire |
Date of discharge/separation |
10/1974 |
Military record/questionnaire |
Military handling of herbicides, insecticides, or hazardous chemicals |
Yes |
Questionnaire |
Perception of physical and mental risk of testing* |
Yes (high risk) |
Questionnaire |
Days involved in Project SHAD |
5 |
Questionnaire |
Physical or mental problems during or after testing* |
Yes |
Questionnaire |
Perception of likelihood of long-term physical or mental effects* |
Somewhat unlikely |
Questionnaire |
Number of SHAD trials |
3 |
Military records |
Specific information about test and post-test activities |
Yes |
Questionnaire/military records |
Name of ship |
USS George Eastman |
DoD fact sheet |
Type of agent used in test |
Trioctyl phosphate |
DoD fact sheet |
Number of days or dates on ships |
35 or 7/1–8/5/1972 |
Military unit records |
Vital status |
Alive |
National Death Index/SSA/VA records |
Date of death |
8/15/2000 |
National Death Index/SSA/VA records |
Cause of death |
ICD-9 code |
National Death Index |
Branch of service |
Navy |
Military records |
Pay grade |
E1 |
Military records |
*See Appendix B for specific questionnaire items. |
TABLE 8-2 Primary, Secondary, and Tertiary Outcome Variables
Primary outcomes |
Description |
SF-36 summary score |
Physical and mental summary scores |
Vital status |
Alive/dead and date of death |
Cause of death |
Based on ICD groupings |
Secondary outcomes |
|
SF-36 subscale scores |
Physical functioning, role physical, bodily pain, general health perception, vitality, social functioning, role emotional, and mental health |
Neuropsychological Impairment Scale |
Memory and attention subscale |
SCID Somatization Scale |
Measure of somatization |
Medical condition groupings (created from 45 chronic medical conditions) |
Cardiovascular, visual, respiratory, renal, endocrine, liver, autoimmune, gastrointestinal, neurological, psychological, and cancer |
Tertiary outcomes |
|
History of 45 chronic medical conditions |
See questionnaire in Appendix B |
History of 19 symptoms within past year |
See questionnaire in Appendix B |
Number of children with birth defects |
See questionnaire in Appendix B |
Total number of postdischarge hospitalizations |
See questionnaire in Appendix B |
variables. Although there are a large number of health outcomes, we did not make adjustment for multiple statistical comparisons.
The primary exposure classification was defined as participant versus nonparticipant, but we also defined four exposure groups based on information in the Department of Defense (DoD) fact sheets and information on an individual’s test participation history (see below for details).
Data on the following potential confounders were also collected via questionnaire and from military records: smoking, drinking, age, general health status, perception of tests, branch of service, race, length of service, marital status, education, pay grade, and current BMI.
DEFINING EXPOSURE GROUPS
In addition to participant versus nonparticipant comparisons, it was desirable to define specific exposure groups within the Project SHAD participants to answer the question of whether outcomes differed by specific patterns of exposure. We also looked at whether health outcomes differed by individual ship. In defining the exposure groups, we took advantage of the fact that Project SHAD exposures fell into four natural groups. First, a large number of Project SHAD participants were exposed only to Bacillus globigii (BG) or methylacetoacetate (MAA), including those only in Autumn Gold, Eager Belle, Scarlet Sage, and Purple Sage. This exposure group we named group A and took additional advantage of the fact that there was a natural factorial design based on presence (+) or absence (–) of the two exposures. The four exposure groups were BG+/MAA+; BG+/MAA–;BG–/MAA+; and BG–/MAA–. Similarly, participants who were in only DTC test 69-10 were exposed only to trioctyl phosphate (TOF or TEHP) and were named group B. Removing the participants in groups A and B from further consideration, the remaining participants fell into two remaining groups: group C included participants who were at any test using active agents; and group D included participants who were at tests where no active agents were used. Thus, group D subjects might have been exposed to any one of the following agents or decontaminants: BG, betapropriolactone, calcofluor, DF-504, diethylphthlate with fluorescent dye, Echerichia coli, fluorescent particles, MAA, Serratia marcescens, trioctyl phosphate, uranine dye, or zinc cadmium sulfide. Table 8-3 shows the four exposure groups, numbers of participants, and number of controls.
Individual Exposure Data
During Project SHAD test DTC 69-10, Marine troops were subjected to a simulated chemical weapons assault with the purpose of determining the “operational effects of a persistent, toxic, chemical agent spray attack on U.S.
TABLE 8-3 SHAD Exposure Groups
Group Name |
Type of Exposure* |
Number of Participants |
Number of Controls |
Group A |
Only BG or MAA |
3,392 |
3,615 |
Group B |
Only TOF |
856 |
870 |
Group C |
Nerve agent or biological agent (with or without possible simulant exposure) |
749 |
1,093 |
Group D |
No active agents |
870 |
1,212 |
Total |
— |
5,867 |
6,790 |
NOTE: In Project SHAD, test Magic Sword uninfected mosquitoes were released from a ship to see if they would make it to a nearby island. These participants were not exposed to any agents. *BG = Bacillus globigii; MAA = methylacetoacetate; TOF = trioctyl phosphate; nerve agents = sarin or VX; biological agents = Coxiella burnetti, Pasteurella tularensis, staphyloccocal enterotoxin B; no active agents = remainder of participants after Groups A, B, and C have been removed that were exposed to some other type of agent. |
amphibious forces” (DoD, 2006). During this test, sampling was conducted on exposed personnel and their clothing to determine the extent of exposure to the simulant agent TOF. DTC test 69-10 was conducted at Vieques Island, east of Puerto Rico, on May 3, 4, 5, and 7, 1969.
We received a redacted version of the DTC test 69-10 final report from the DoD. Tables 12 through 15 of that report showed estimates of contamination on landing force personnel for trials on the days May 3, 4, 5, and 7, respectively. Each table showed the military unit (down to platoon level) and listed individuals, along with their estimated magnitude of contamination, on an ordinal scale: VH (very heavy), H (heavy), M (medium), L (light), VL (very light), T (trace), and N (negligible). In these tables, individuals were identified by last name or last name and initial or initials. Presumably, initials were shown when there were duplicates of last names.
Using data from the Marine unit roster, we attempted to identify all the individuals with DTC test 69-10 exposure data, determine their military service number, and link their exposure data with their responses on the health survey. There were 706 daily exposure records (including multiple records per individual), of which 672 (95 percent) were successfully linked to an individual on our study roster. When multiple exposures were taken into account, there were 428 individuals who had ordinal contamination data from one or more trials. Because the DoD was unable to provide quantitative data regarding the contamination levels, we analyzed the TOF exposure data by arbitrarily assigning the following exposure values: T (trace) and N (negligible) = 0.5 ; VL (very light) = 1.0; L (light) = 2.0; M (medium) = 3.0; H (heavy) = 4.0; and VH (very heavy) = 5.0. We further assigned a dose of zero to Marine controls in DTC test 69-10.
METHODS OF ANALYSIS
Mortality Analyses
The research group defined two analytic approaches for the mortality outcome. The first uses standardized mortality ratios (SMRs), calculated for each cohort (participant and referent) separately using standard rates adjusted for age, race, sex, and calendar year of death. The second involves proportional hazards modeling using a wider range of available covariates.
SMRs are a commonly used tool to compare death rates among a cohort of interest to those in a larger, reference population, customarily the U.S. general population. The deaths that actually occur in the cohort of interest are labeled as “observed” deaths; one also calculates the “expected” number of deaths that would have occurred had the numbers of the cohort died at the same rate as the U.S. population with the same age, race, and sex distribution. The ratio of observed to expected deaths is an SMR, which is equal to 1.0 if the number of deaths observed in the cohort of interest is the same as the number of deaths expected to have occurred if the cohort members had died at the same rate as the rest of the U.S. population.
SMRs show whether the mortality of the cohort of interest is higher or lower than that of the U.S. population. One typically sees SMRs for veterans’ cohorts that are less than 1.0. Reasons given for this refer to the requirement that military servicemen pass an entrance physical and also pass periodic physical fitness exams while in military service, both effectively screening in favor of healthier individuals versus their general civilian counterparts. Not only is this healthiness thought to produce lower death rates among active duty military personnel, but lower mortality rates apparently persist even after discharge from active duty (Seltzer and Jablon, 1974, 1977). Such effects seen among occupational groups have been labeled as the “healthy worker effect,” and by analogy, lower SMRs among military veterans can be attributed to a “healthy soldier effect.” Despite this limitation, SMRs provide a way to compare the mortality of the cohort of interest to that of the general population. Also, because SMRs are based on standard distributions of deaths, they can be compared across studies. We used OCMAP Plus software to compute SMRs (Marsh et al., 1998). SMR results were also stratified by exposure group, ship/unit, officer or enlisted, and branch as sample size permitted. All-cause and cause-specific mortality were investigated.
Crude mortality was also examined using Kaplan-Meier survival curves to assess mortality differences between analysis groups. Cox proportional hazard regression analysis was used to assess mortality differences while adjusting for potential confounders. We implemented these analyses using the SAS PHREG procedure (SAS Institute, Inc., 1999). In this approach, the risk of death—in statistical terms, the hazard—is modeled in a regression that includes a baseline hazard as well as coefficients that represent the additional hazards associated with various factors such as participation in Project SHAD. The coefficient associated with a factor represents a hazard ratio (HR), which can be interpreted as a relative risk of death that remains constant over the follow-up period. In our analyses, coefficients were included for participation, age at time of first participation, race (white versus nonwhite), service branch (Navy versus Marines), and pay grade. Hazard ratios are considered statistically significant if their associated 95 percent confidence interval (CI) excludes the value 1.0. For those participants who were missing a date of birth (roughly 7%) the following procedure was used to impute a date of birth. The day of birth was randomly assigned as 1 to 28 with each day having a uniform chance of being assigned. The month of birth was randomly assigned as 1 to 12 with each month having a uniform chance of being assigned. The year of birth was randomly assigned based on the following probabilities: 1939 = 4 percent; 1940 = 5 percent; 1941 = 8 percent; 1942 = 14 percent; 1943 = 15 percent; 1944 = 15 percent; 1945 = 12 percent; 1946 = 10 percent; 1947 = 8 percent; 1948 = 5 percent; and 1949 = 4 percent. For those participants who were missing a value for race (roughly 29%), the value was set to white.
The International Classification of Diseases, Tenth Revision (ICD-10) and the International Classification of Diseases, Ninth Revision (ICD-9) were used to identify deaths due to malignant neoplasm (ICD-10 codes C00–C97 and ICD-9 codes 140–208), cardiovascular disease (ICD-10 codes I00–I99 and ICD-9 codes 390–459), respiratory disease (ICD-10 codes J00–J99 and ICD-9 codes 460–519), endocrine and metabolic diseases (ICD-10 codes E00–E90 and ICD-9 codes 240–279), infectious diseases (ICD-10 codes A00–B99 and ICD-9 codes 001–139), and injury/external causes (ICD-10 codes S00–T90 and V01–X85 and ICD-9 codes 800–959 and E codes 800–999). For all-cause survival analysis, persons who were not matched to a death record were considered alive through the follow-up period and administratively censored as of the end of the study period. For cause-specific analyses, follow-up for those who died from other causes was censored at the age of death. Although the main mortality analysis included an overall comparison of total and cause-specific mortality for Project SHAD participants versus nonparticipants, similar analyses were done for each of the four exposure groups as defined above.
Morbidity Analyses
The main morbidity analysis focused on differences between Project SHAD participants and nonparticipant controls for the primary outcome of the SF-36 score, physical and mental summary scores. Differences in secondary and tertiary outcomes as described above were also examined. With regard to morbidity outcomes, crude comparison of differences in mean scale measurements were made using analysis of variance and Student’s t-test as appropriate to compare the outlined exposure and control groups. Comparison of differences in mean scale measurements with adjustment for potential confounders of age, race, branch, pay grade, smoking, drinking, and BMI was accomplished using a general linear models analysis. SF-36 scales were also analyzed to examine differ-
ences in dose groupings of BG and MAA. In addition, a subgroup analysis of SF-36 scores among the DTC test 69-10 Marines was conducted to look for exposure-response relationships. The mean NIS and SCID Somatization Scale scores were analyzed as outlined above for the SF-36 scales and subscales. The NIS scores were also used to create a dichotomous outcome for memory and attention problems. Crude comparisons of prevalence of these outcomes, as well as comparisons of the prevalence of medical conditions and symptoms, were conducted using odds ratios and 95 percent CI. Comparison of prevalence rates adjusted for the potential confounders of age, race, pay grade, smoking, drinking, and BMI was done using logistic regression analysis. Medical conditions were analyzed as individual items and also in the following 11 major groupings: cardiovascular, visual, respiratory, renal, endocrine, liver, autoimmune, gastrointestinal, neurological, psychological, and cancer.
REFERENCES
DoD (Department of Defense). 2006. Project 112. http://deploymentlink.osd.mil/current_issues/shad/shad_intro.shtml (accessed November 28, 2006).
Marsh, G.M., A.O. Youk, R.A. Stone, S. Sefcik, and C. Alcorn. 1998. OCMAP-PLUS: A program for the comprehensive analysis of occupational cohort data. Journal of Occupational and Environmental Medicine 40:351-362.
SAS Institute, Inc. 1999. SAS/STAT Software version 8. Cary, NC.
Seltzer, C.C., and S. Jablon. 1974. Effects of selection on mortality. American Journal of Epidemiology 100:367-372.
Seltzer, C.C., and S. Jablon. 1977. Army rank and subsequent mortality by cause: 23-year follow-up. American Journal of Epidemiology 105:559-566.