7
The Comparison Cohort
A significant feature of the design of this study of Operation CROSSROADS mortality is the inclusion of a military comparison cohort. One of the criticisms of MFUA's earlier study of nuclear weapons test participants was its lack of a control group (GAO 1992). While the CROSSROADS study does compare participant mortality with that of the U.S. general population, the study has assembled, at significant cost, a separate control group from military unit and individual records. The comparison of participant death rates with rates for comparable controls should eliminate the risk of bias introduced by the ''healthy soldier effect." 14
Comparison Cohort Selection
Navy and Shipboard Marines
The Defense Nuclear Agency (DNA) provided lists of potential control units15 to the Medical Follow-up Agency (MFUA). For Navy controls, initial plans called for selecting a control ship to pair with each participant ship.
14 |
See discussion in Chapter 3. |
15 |
In the Navy, a "unit" is usually a ship (so that all personnel assigned to a particular ship would be in one unit associated with the ship's name). Exceptions occur, especially during special non-combat procedures. Examples of unit assignment could then include "RADSAFE" or "JTF" (joint task force). Army units are more frequently based on function, such as artillery or infantry. |
Pairing criteria included type of ship (i.e., destroyer, cruiser, tender), Pacific location, and service near to the time of CROSSROADS.
The initial selection plan was amended twice. First, because CROSSROADS occurred in the summer of 1946, well after demobilization efforts had gotten under way following World War II, ships in the Atlantic were chosen when it was impossible to find sufficient ships of a given type in the Pacific. Second, because selected control ships did not always supply sufficient numbers of personnel for the comparison cohort to match the desired rating distribution, additional control ships were selected to provide supplements. The additional ships maintain the type of ship choice initially derived from the to-be-matched participant ship.
As discussed above, DNA supplied MFUA with lists of potential control ships; MFUA selected actual control ships from those lists. To identify the enlisted men on these ships, microfilmed muster rolls were obtained from the National Archives. Each muster roll of a Navy ship from the CROSSROADS era provides, for a given date, the complete list in alphabetical order of enlisted personnel on board that particular ship, showing name, service number, and rank/rating (for example, "MMI" is machinist mate, first class).
We selected Navy enlisted controls in such a way that they would have the same rating distribution as participants and would have served on the same types of ships. Before keying individual control names, MFUA developed a distribution of ranks and ratings (based on distribution within participant units) required from each control ship to create a balanced comparison cohort. A muster roll was selected for a control ship, and once the ship identifying information was entered into the computer, the program randomly selected a letter of the alphabet at which keying of the muster roll data for that ship was begun. As each individual rank and rating was entered, it was compared to the distribution of participant ranks and ratings for a specific ship or ship type. If a particular category had not already been filled, the keyer was allowed to enter the corresponding name and military service number. As more names were entered, more rank-rating categories were filled up, and the program accepted fewer names. Although the original software was programmed to match each individual's participant ship to a single control ship, it was later modified to match the type of participant ship to the type of control ship (e.g., submarine to submarine, destroyer to destroyer, etc.). The identification of control officers relied on a different source—deck logs—because Navy officers are not routinely listed on muster rolls.
Marine controls were selected, by and large, in the same way as Navy controls, starting from ship muster rolls. Because 52 percent had missing military service numbers (MSN), the Marine Corps Stat list was searched for MSN; the majority of missing numbers were found. Also, there was no random alphabetical starting point for keying and no range checks on keyed data because, for each selected control unit, the entire list was keyed.
Army and Army Air Corps
The assembly of the Army and Army Air Corps control rosters was somewhat more complicated than that for Navy and Marines. First, although data on rank were available, there was no information on individual military occupational specialties readily at hand. In addition, the choice of Army and Army Air Corps units was less straightforward than for Navy units, because there were many more types of Army units than there were types of ships. Most important, however, was the part played by a fire at the National Personnel Records Center (NPRC) in St. Louis.
In 1973, a fire at NPRC (Stender 1974) destroyed some 80 percent of Army and Air Force personnel records from World War I through the Korean Conflict. After the fire, NPRC created a computerized registry, in part to index which military personnel records were reconstructed from fragments of burned records and which only suffered water damage and subsequently were refiled. The postfire protocol calls for all newly accessioned records to be indexed in the registry. The registry file is thus an index to the location of military medical and personnel records at NPRC. It contains information on approximately 23 million veterans. Because all selected controls were to have their military personnel records abstracted, the NPRC registry was used to obtain unburned records for controls. The 80 percent loss of records meant that there was a need to oversample from Army and Army Air Corps unit records—by roughly a factor of five—to obtain sufficient potential controls to ensure finding a control with an unburned record. Because of this oversampling strategy, the data entry software for Army and Army Air Corps records did not keep a count of each rank category; for each control unit selected, the entire unit roster was keyed.
The processing of Navy muster rolls and Army unit rosters began in October 1988 and continued through January 1990. A total of 57,734 control records were keyed, allowing for the oversampling of Army controls.
Military Personnel Records
It was necessary to locate (using military service number) and abstract the military personnel record for each selected control individual to obtain the necessary information for a mortality analysis. Data abstracted included full name; branch of service; additional military service numbers; date, place and state of birth; place of entry onto active duty; race; rank at separation; date of separation; and whether records had been sent to the VA.
Due to blurred or unreadable microfilm, many of the service numbers keyed from the muster rolls did not correspond to records available on known military personnel. Thus, it was necessary to go to an independent source, the VA Master Index, to verify identification data. The VA Master Index (VAMI),
now a large microfilm file, originally consisted of some tens of millions of 3 × 5 cards, one card for each veteran beneficiary. Each 3 × 5 card in VAMI, which was maintained in active use through the early 1970s, contains identifying and VA benefits data for an individually identified veteran. When a muster roll entry could not be matched to an available record folder, an attempt was made to locate the corresponding 3 × 5 card in VAMI and to verify correct identifying data, such as name and MSN. At the same time, additional data, if available, were also recorded: for example, VA claim number or date of death. Using these new identification data, another attempt was then made to find a record folder. Finally, despite the deliberate oversampling of Army and Army Air Corps controls, it was not always possible to find enough unburned records for men of a particular rank, and partially burned records were occasionally ordered. A total of 40,354 military personnel records were abstracted.
Additional Controls
In 1989, DNA notified MFUA that the DNA list of CROSSROADS participants reflected substantial numbers of individuals who should not be on the list and omitted many who should be on the list.l6 The final participant roster used in this study is about 2,500 individuals larger than the 1986 list on which control selection was based. The balance of ranks and ratings in ship types in the existing control pool and the participant cohort were compared and noted to be similar. Since the analysis did not involve pairwise comparisons, we determined that additional controls would not be required.
16 |
See Chapter 2 of this report. |