Click for next page ( 51


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 50
9 Analysis Structure OVERVIEW The analysis plan for the Five Series Study was structured to check data va- lidity, test hypotheses, and interactively explore data to follow leads arising from data analysis. The study was designed to address whether . participation in at least one of the five selected atmospheric nuclear weapons tests is associated with increased mortality hazard; and . participants who were more likely to have been highly exposed to radia- tion would have increased mortality hazard relative to participants who were less likely to have been highly exposed. The basic comparison involves the survival experience of participants rela- tive to that of referent cohort members. Because various diseases have different degrees of radiogenicity (Mettler and Upton, 1995), hazard ratios have been calculated for mortality from all causes, all malignancies, and leukemia (ex- cluding chronic lymphocytic leukemia). Other radiogenic cancers, along with a selection of presumed nonradiogenic diseases and conditions, are also examined. Disease categories were discussed and defined, using International Classifica- tiorl of Diseases, 9th Edition, codes, before analysis began (see below). AVAILABLE DATA Data available for the analysis of survival times consist of measures or indi- cators of (1) presumed radiation exposure, (2) individual and military service characteristics that might confound an association between exposure and out- come, and (3) mortality outcome. Table 9-1 presents the variables that were in- cluded in the analysis dataset. It should be noted that variables were not all of the same quality with regard to completeness and validity. 50

OCR for page 50
ANALYSIS STRUCTURE TABLE 9-1. Variables Considered for Analysis and Their Utility Variable Example Utility Participant status Sex Paygrade Branch of service Selection series Date of selection seriesb Location Shot Number of series Number of shots Unit category Unit of assignment Rank/rating Occupation Tasks during test Dose Device Date of birth Vital status Date of death Age at selection series Years since atomic test exposured Age at death or censured Calendar decade of deaths Decade since selection d series Underlying cause of death Associated causes of death Participant Male E3 Air Force, Army, Marines, Navy CASTLE April 1, 1953 Pacific BRAVO 3 12 Technical 9740 TSU Chemical Section Private, PVT2 E.g., pilot, navigator Cloud sampler pilot 2.4 rem Thermonuclear January 15, 1923 Dead March 7, 1972 Date of selection series minus date of birth Date of death or censoring minus date of selection series Date of death or censoring minus date of birth 1960s 20-30 years since shot +e +d,e +1 e l e +/_e +/_e Malignant neoplasm of the lung + Hypertensive heart disease + aPlus or minus assigned based on general consideration of validity and completeness; a plus indicates that data are available and of good quality; a minus indicates that data are either unavailable or of poor quality. bThe first day of the operational period of the selection series. CCategories created and assigned based on unit of assignment. Value calculated from specified dates in dataset. equality of fact and date data is discussed in this report. For known deaths, the quality is excellent; the Department of Veterans Affairs data system may not have ascertained all deaths of study cohort members. 51

OCR for page 50
52 THE FIVE SERIES SI UDY ANALYSIS Variables The variables included in the basic analyses are participant status, age at selection series, paygrade, branch of service, and selection series. Analyses also explore relationships using variables such as land versus sea series; age-calen- dar time; disease latency; series and post-series time periods; number of series; and associated causes of death. (The definitions and rationale for the use of these variables are described elsewhere in this report.) To appropriately test the second question whether a dose-response rela- tionship exists between radiation exposure and mortality hazard an at-least- ordinal variable that ranks an exposure surrogate measure would be needed. (Note: A working group of the committee overseeing the Five Series Study has reported kIOM, 1995; reprinted in Appendix A of this report] that the extensive dosimetry categorization and reconstruction data developed by the Nuclear Test Personnel Review Program of the Defense Threat Reduction Agency are not suitable for use in epidemiologic investigations of dose-response. This analysis plan does not, therefore, use the dosimetry data as exposure variables in the sta- tistical analysis.) These surrogates could incorporate information from military records, eyewitness accounts, and historical records known about groups more and less likely to have received higher radiation doses. Using the exposure sur- rogates, statistical models could test for these proxy dose-response relation- ships. (See Chapter 7.) Type of Analysis The research group defined two analytic approaches. The first uses standard- ized mortality ratios (SMRs), calculated for each cohort (participant and referent) separately using standard rates adjusted for age and time distributions (Marsh et al., 1998; Rothman and Greenland, 1998.~. The second involves proportional haz- ards modeling using the wider range of available covariates (Allison, 1995~. 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 "observed" deaths; one also calculates the "expected" number of deaths that would have occurred had the members of the cohort died at the same rate as the U.S. population with the same age, race, and sex distribution. The ratio of ob- served to expected deaths is an SMR, which is equal to 1.0 if the number of The organizational locus of the Nuclear Test Personnel Review Program within the Department of Defense has been the Defense Nuclear Agency (until June 1996), the De- fense Special Weapons Agency (until October 1998), and, currently, the Defense Threat Reduction Agency.

OCR for page 50
ANALYSIS STRUCTURE 53 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 U.S. population death rates. Sex and race information was not included in the datasets for this study. For the 61.7 percent of deaths for which we were able to acquire death certificates, race and sex information is available. Less than half a percent (0.4%) of the death certificates were coded as female; between 8 and 9 percent as black. These proportions may not accurately reflect the unknown percentages of male and black members of the participant and referent cohorts. Because both race and sex are associated with mortality (both survival time and cause of death), they do not provide valid estimates of the full at-risk cohort. We used white male population rates for SMR calculations. SMRs thus 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 focus on the requirement that mili- tary 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 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~. Cox proportional hazard ratio analysis (Cox, 1972) is used for the core analyses in this report. We implemented these analyses using the SAS program PHREG (SAS Institute, 1996~. 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 in this case nuclear test series participation. The coefficient associated with a factor represents a hazard ratio, 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 test series participation, age at time of first participation, and age at time of first participation squared and cubed. Haz- ard ratios are considered statistically significant if their associated 95 percent confidence interval excludes the value 1.0. The time scale for these models was attained age, which is thought to be the most appropriate scale for the kinds of analyses we undertook (Korn et al., 19974. Rather than fashion regression models that included risk estimates for other factors such as test series, branch of service, and paygrade, we chose to include these as stratification variables. Although this choice does not permit the esti- mation of risks associated with the stratification variables, it does allow more

OCR for page 50
54 THE FIVE SERIES SILLY complete control of the effects of these variables (see Allison t1995] for further detail). Finally, if we did not have definitive evidence of death for an individual, he was considered to be not known dead (alive); if the individual was thought to be dead, but there was no date of death or date of birth (there were only 38 of these), this record was excluded from the analyses. Diagnosis Groups Based on tables from other studies of atomic veterans (e.g., Pearce et al., 1997) yet expanded, the staff and committee chose which categories of diagno- sis codes to examine: (1) the broad categories of noncancer causes of death; (2) all malignant neoplasms; and (3) focused groups of malignancies, including leu- kemias and other putatively radiogenic malignancies. Tables 9-2 and 9-3 present the cause-of-death categories considered in this report. TABLE 9-2. Broad Categories of Noncancer Causes of Death as Grouped by ICD-9 Codes Broad Category ICD-9 Code Infectious and parasitic diseases Benign neoplasms Endocrine, nutritional, and metabolic diseases and immunity disorders Diseases of the blood and blood-forming organs Mental disorders Diseases of the nervous system and sense organs Circulatory disease Respiratory disease Digestive disease Diseases of the genitourinary system Diseases of the skin and subcutaneous tissue Diseases of the musculoskeletal system and connective tissue Congenital anomalies Symptoms, signs, and ill-defined conditions All external causes * Total 001-139 210-239 240 279 280-289 290-319 320-389 390~59 460-519 520-579 580~29 680-709 710-739 740-759 780-799 800-999 001-139, 210-999 Not included are complications of pregnancy, childbirth, and the puer- perium, and certain conditions originating in the perinatal period. SOURCE: International Classification of Diseases, 9th Edition (ICD-9) (USDHHS, 1991~.

OCR for page 50
ANALYSIS STRUCTURE TABLE 9-3. Cause-of-Death Categories Within Broad Category of Malignant Neoplasms Site ICD-9 Code Lip, oral cavity, and pharynx Digestive organs and peritoneum Esophagus Stomach Small intestine Colon Rectum Liver and intrahepatic bile ducts Gallbladder Pancreas Respiratory and intrathoracic organs Nasal Larynx Lung Bone, connective tissue, skin, and breast Bone Connective tissue Skin Skin nonmelanoma Breast Genitourinary organs Prostate Testis Bladder Kidney Other Brain and nervous system Thyroid Other solid cancer Total hematological Non-Hodgkin's lymphoma Hodgkin's disease Multiple myeloma Leukemia Leukemia, excluding chronic lymphoid leukemia Total 140 149 150 151 152 153 154 155 156 157 160 161 162 170 171 172 173 185 186 188 189 191, 192 193 14~199 20~208 200, 202 201 203 20~208 204.0, 204.2-208.9 14~208 Not listed separately are malignant neoplasms of the retroperitoneum and peritoneum; other digestive organs; pleura; thymus, heart, and mediasti- num; other respiratory and intrathoracic organs; female breast; female genital organs; penis and other male genital organs; eye; other (than thy- roid) endocrine glands; other, ill-defined, secondary, and unspecified sites. SOURCE: International Classification of Diseases, 9th Edition (ICD-9) (USDHHS, 1991). 55