A Call to Link Data to Answer Pressing Questions About Suicide Risk Among Veterans

OUR PAPER IN THE CURRENT issue of Journal1 makes the case that male veterans of conflicts prior to Operation Iraqi Freedom and Operation Enduring Freedom (OIF/OEF) have an age- and race-adjusted suicide risk that is modestly, but not significantly, higher than risk among male nonveterans. Our estimate is consistent with findings from prior military cohort studies2–6 but not with findings from Kaplan et al. 2007,7 who, using the same underlying data set, found a statistically significant and greater than 2-fold increase in suicide risk among male veterans. Because both our study and Kaplan's use the National Health Interview Survey-National Death Index (NHIS-NDI) linked database (and we attempted, unsuccessfully, to replicate Kaplan's findings), our study unavoidably challenges the validity of Kaplan's. Unfortunately, Robert Gibbons, PhD, declined the opportunity to adjudicate between our findings8 and concludes that “there is no way to comment on the accuracy” of the 2 sets of results, despite his and our inability to replicate the higher risk observed by Kaplan et al. Clearly, had Gibbons requested the datasets and analytic code from both published studies, it would have been possible to ascertain whether differences in the analytic datasets, analytic strategy, or simple human error explained these differences. Because Gibbons chose not to undertake this effort, the most transparent way to resolve the issue is for the NHIS to conduct a reanalysis and report the findings. We have done so and urge Kaplan to do the same. Instead of offering the specific findings of his own reanalysis of the current NHIS-NDI data set, Gibbons uses his editorial to explore rates of suicide among men identified by the National Violent Death Reporting System (NVDRS)9 as current or former members of the US military. We feel obliged to respond to this new analysis, both because our long involvement with the development of NVDRS (C. B. and D. A. codirected its pilot and are familiar with its strengths and limitations) leads us to question whether the NVDRS can currently identify veterans with accuracy and because Gibbons conflates 3 important but distinct issues in his analysis: suicide risk among veterans of OIF/OEF, risk among younger veterans (irrespective of conflict), and risk among those recently separated from active military duty (irrespective of conflict and age). NVDRS is a rich dataset, but without linking NVDRS records to Department of Defense (DOD) data to determine which decedents were truly veterans (according to the same definition used in the US veteran population estimates), findings from NVDRS are problematic when used to answer questions about suicide risk among veterans. In short, suicide decedents categorized by NVDRS as having served in the military will correctly include current active duty personnel (which Gibbons attempts to adjust for, as we demonstrate next) but may also incorrectly include persons who never served as active duty personnel, such as decedents who, at the time of death, were current or former National Guard or Reserves but were never activated, civilians serving in the military, or people in training for the military. Because this overestimation occurs only among decedents, it leads to overestimates of risk among veterans, particularly among the youngest cohort, for whom it is also problematic to separate out recency of service from age–group related effects. To his credit, Gibbons tries to subtract out known active duty suicides from suicides identified as ever having served. However, as exemplified by the ways in which he distributes active duty suicides across NVDRS versus non-NVDRS states (and across age strata), Gibbons makes assumptions that further bias his estimates. For example, Gibbons distributes active duty suicides evenly across NVDRS and non-NVDRS states, despite the fact that several of the states with the largest active duty populations are non-NVRDS states. The estimate of the risk of veteran suicide among the youngest age group may be further exaggerated because Gibbons distributes the military suicides according to the age distribution of military enrollment rather than according to the observed distribution of military suicides by age group (which would attribute a greater proportion of suicides, approximately half, to the youngest age stratum).10 Mitigating this bias, to some extent, is the fact that Gibbons subtracts out all active duty suicides, including those that occured abroad, when, in fact, deaths abroad are not counted in NVDRS statistics. Understanding the risk of suicide among recently separated veterans, especially veterans of the Iraq and Afghanistan conflicts, is of pressing concern given the unprecedented increase in rates of suicide in the active US armed forces since 2005. This issue, however, is not relevant to the discrepancies between our study and Kaplan's, both of which used (NHIS) data based on interviews that not only do not specify period of service but also took place before these wars began. Moreover, the issue cannot be resolved using unlinked NVDRS data. It can be answered, however, if data routinely collected by the DOD are linked to NVDRS data or to the NDI. Indeed, Kang and Bullman2,4–6 have published first-rate analyses along these lines on select populations of veterans. Following their lead, data from the DOD, the NDI, and the NVDRS could readily be linked not only retrospectively (DOD, NDI), allowing resolution of the historical issue of suicide risk among veterans remote from military service but also prospectively (DOD, NDI, NVDRS, and possibly VHA), thus providing an ongoing surveillance system that would enable policymakers and health care providers to make decisions aimed at saving lives based on unbiased risk assessment.