Font Size Down Font Size Up Reset Font Size

Sign Up for Committee Updates

 

Witness Testimony of Stephen L. Rathbun, Ph.D., Department of Epidemiology and Biostatistics, Interim Head and Associate Professor of Biostatistics, University of Georgia, Athens, GA

Summary

In the fall of 2007, I was asked to perform a statistical analysis of veterans’ suicide data collected by CBS News for a storey that was broadcast on November 13, 2007. I agreed to do so, and was not compensated for my contributions. For 2004 and 2005, CBS provided data on numbers of veteran and non-veteran suicides and veteran and non-veteran population sizes, cross-tabulated by state, gender and age class. Suicide data were obtained from state death records and population sizes from the U.S. Census. These data are the property of CBS News and were erased from my computer following the broadcast on November 13, 2007. To avoid investigator biases, methods of statistical analyses were specified prior to receiving the data. Standard statistical procedures were used to analyze the data; estimated suicide rates adjusted for age and gender. Veteran suicide rates were estimated to be approximately double those of non-veterans. This pattern of higher estimated veteran suicide risk was observed for both genders and all age classes. The most striking pattern was an especially high risk ratio for 20-24 year old veterans; in this age class, veteran suicide rates were estimated to be about 3 times those of non-veterans.

Introduction

This statement concerns the analyses of veterans’ suicide data conducted at the request of CBS News during the fall of 2007, the results of which were broadcast during CBS Evening News on November 13, 2007. The following will describe my relationship with CBS News, the data that were analyzed, the method of statistical data analyses, and the results of those analyses.

Before proceeding, please note the following:

  1. I am an expert in biostatistics, but not an expert on veterans’ suicides. Although I have a 18-year record of teaching and research in biostatistics, I have not had any prior experience with suicide data. While I can comment on the methods of data analyses, estimated suicide rates and limits of statistical inference regarding the data analysis, I cannot make expert comments regarding what causes the observed patterns of suicide rates.
  2. The suicide data are a property of CBS News and at the request of CBS News, my copy was erased following the November 13, 2007 broadcast. This was done to comply with agreements made between CBS News and States to ensure the confidentiality of sensitive human-subjects data.
  3. On March 4, 2008 I testified on behalf of the plaintiffs in the case of Veterans for Common Sense and Veterans United for Truth, Inc. vs. Gordon H. Mansfield, Acting Secretary of Veterans Affairs (U.S. District Court, Northern District of California, San Francisco Division, Case No. C-07-3758-SC). No compensation was received from the plaintiffs for this testimony. I had no prior relation with either party in this action.

Relationship with CBS News

I had no prior relationship with CBS News before Pia Malbran, producer of the veterans’ suicide news storey, contacted me in August 2007 asking me to analyze veterans’ suicide data. Aside from the gift of a CBS News baseball cap (valued at less than $20), I was not paid by CBS News for the work that I have done. I understand from Pia Malbran, producer of the veterans’ suicide news storey, that I was contacted because I had no relationship with the Veterans Administration, veterans groups, or involvement with advocacy related to veterans’ issues.

Data Collection

During my initial consultations, Pia Malbran discussed the sources of the data, and I advised her regarding the format in which the data should be provided for data analysis. Ms. Malbran requested that suicide rates be adjusted for age, gender, and race. She obtained population totals from the U.S. Census, and population totals for veterans from the Veterans Administration. The states provided her information on the number of veteran suicides and total number of suicides. Veteran status was to be ascertained from state death records indicating that suicide was the cause of death, and including a check box indicating whether or not the subject was a veteran. In a few cases, veteran status was not available in the death records. Data were to be provided for 2004 and 2005. Death records after 2005 were not available at the time that the news storey was prepared.

In response to Pia Malbran’s description of the available data, I indicated that an excel spreadsheet should be prepared including columns for state, age class, gender, race, number of veteran suicides, number of suicides with unknown veteran status, total number of suicides, number of veterans, and population total. Suicides with unknown veteran status were allocated to veterans and non-veterans in numbers proportional to the respective sizes of veteran and non-veteran populations. For example, if 10% of all 25-29 year old males in a given state were veterans, then 10% of the suicides of unknown veteran status in that group would be allocated to veterans, while the remaining 90% of suicides in that group would be allocated to non-veterans. This proportional allocation results in conservative estimates, under-estimating the differences between veteran and non-veteran suicide rates.

Preliminary data analysis indicated that states and the Veterans Administration had different definitions of race with respect to the classification of black Hispanic-Americans. So, race was dropped from the data collection efforts. Thus, the data can be cross-classified by state, age class, gender, veteran status, and population size for data analysis. A total of 45 states provided data for 2004 and 2005.

For states with small populations of veterans, the cross classification of veterans by age, gender and suicide status may suffice to identify individual subjects. For that reason, CBS News had to agree that data be kept confidential before data were released to CBS News. To ensure this confidentiality, I was asked to erase the data immediately following the November 13, 2007 broadcast of the veterans’ suicides news storey. This was done as requested.

Data Analysis

To avoid investigator bias, the methods of statistical data analyses were specified before data were received from CBS News. The specific choice of methods was based on the type of data collected, and the specific estimates that Pia Malbran requested in her memo of October 1, 2007:

  1. What is the overall rate of suicide (per 100,000 - age and gender adjusted) for veterans verses non-veterans nationwide?
  2. What is rate (per 100,000 - age and gender adjusted) of suicide for veterans verses non-veterans state by state ranked highest to lowest?
  3. What is the overall rate of suicide (per 100,000) for veteran males (all ages) verses non-veteran males nationwide?   
  4. What is the rate of suicide (per 100,000) among male veterans, 65 or older?  And, how does that compare with male non-veterans, 65 or older?
  5. What is the rate of suicide (per 100,000) among veterans (both genders) aged 20 to 34

A logistic regression model was fit to the data from each year (2004 and 2005), including main effects for age-class, gender, and veteran status, as well as two-way interactions among pairs of these explanatory variables. Logistic regression is the standard statistical method for modeling binary responses such as suicide status; a person has either committed suicide or not committed suicide. The inclusion of interactions allows the effect of veteran status on suicide risk to depend on gender and age. As per the pre-specified protocol for data analysis, the three-way interaction among age, gender and veteran status was dropped from the model since it was not statistically significant.

All estimated suicide rates were adjusted for the impact of age and gender. This was done because suicide rates depend on age and gender. For example, suicide rates are higher among males and than among females. Moreover, males are over-represented among veterans since males are more likely to serve in the military than females. Given the higher suicide rates among males, and the over-representation of males in veteran populations, failure to adjust for gender will result in over-estimates of veteran suicide rates. Similar arguments can be made for the impact of age.

Initial data analysis was completed before my portion of the storey was taped by CBS News on October 3, 2007. On October 16, Pia Malbran contacted me with data on the numbers of active-duty soldiers committing suicide in 2004 and 2005, cross-classified by age and gender. She expressed concern that some of these soldiers may have been mistakenly classified as veterans in the state death records, and asked me to re-analyze the data subtracting these cases from the veteran suicide counts. Thus, two estimates of veteran suicide rates will be presented. The higher estimate is based on the original analysis, while the lower estimate was obtained after these active-duty suicides were subtracted.

All analyses were carried out using SAS. Analyses using this statistical software involve written code documenting exactly how the analyses are carried out. This is as opposed to other statistical software packages that rely on point-and-click menu-driving procedures for data analysis that leave no record documenting the method of analysis. SAS code used for my data analysis was provided to Ira Katz following the November 13 CBS News broadcast.

Results

The estimated suicide risk is higher among veterans than non-veterans. Table 1 presents the age- and gender-adjusted estimates of suicide rates by veteran status for each of the two years. The suicide risk among veterans was estimated to be 1.86-2.32 times the risk among non-veterans in 2004, and 2.10-2.34 times the risk among non-veterans in 2005. These risk ratios were computed by dividing the suicide rates among veterans by the suicide rates among non-veterans.

Table 1. Estimates of overall suicide rates adjusted for age and gender. All rates are expressed as numbers of suicides per 100,000 people.

Veteran Status

2004

2005

Veterans

17.5-21.8

18.7-20.8

Non-Veterans

9.4

8.9

Table 2 presents the age-adjusted estimates of suicide rates for males and females by veteran status for each of the two years. For both veterans and non-veterans, suicide rates were higher among males than among females. Among males, the suicide risk for veterans was estimated to be 1.67-2.09 times the risk among non-veterans in 2004, and 1.79-2.01 times the risk among non-veterans in 2005. Among females, the suicide risk for veterans was estimated to be 2.08-2.60 times the risk among non-veterans, and 2.47-2.73 times the risk among non-veterans in 2005.

Table 2. Estimates of suicide rates by gender adjusted for age. All rates are expressed as numbers of suicides per 100,000 people.

Gender

Veteran Status

2004

2005

Males

Veterans

30.6-38.3

31.5-35.3

 

Non-Veterans

18.3

17.6

Females

Veterans

10.0-12.5

11.1-12.3

 

Non-Veterans

4.8

4.5

Table 3 presents the gender-adjusted estimates of suicide rates for the various age-classes by veteran status. The most striking result is the high relative risk of suicide among 20-24 year old veterans when compared to non-veterans. For this group of young veterans, the suicide risk is estimated to be 2.81-4.31 times the risk among non-veterans in 2004, and 2.75-3.84 times the risk among non-veterans in 2005. Veterans in their 40s also had high estimated suicide rates, but the risk of suicide among non-veterans in their 40s was also estimated to be high. Consequently, the risk ratio did not exceed 1.73 in this age group.

Table 3. Estimates of suicide rates by age adjusted for gender. All rates are expressed as numbers of suicides per 100,000 people.

Age Class

Veteran Status

2004

2005

20-24

Veteran

23.3-35.8

22.9-31.9

 

Non-Veteran

8.3

8.3

25-29

Veteran

12.8-15.5

13.1-16.1

 

Non-Veteran

9.0

8.3

30-34

Veteran

14.1-14.7

16.1-17.7

 

Non-Veteran

10.2

9.9

35-39

Veteran

16.4-17.8

16.1-16.5

 

Non-Veteran

11.4

10.5

40-44

Veteran

20.4-20.6

19.4-19.7

 

Non-Veteran

13.4

12.6

45-49

Veteran

22.4-22.5

23.4

 

Non-Veteran

14.4

13.5

50-54

Veteran

20.3

21.0

 

Non-Veteran

13.4

12.9

55-59

Veteran

15.3-15.4

16.0-16.1

 

Non-Veteran

11.4

11.0

60-64

Veteran

14.3-14.4

13.6

 

Non-Veteran

10.3

10.5

65+

Veteran

11.8

14.9-15.0

 

Non-Veteran

9.6

9.7