Are We Undercounting the True Burden of Mortality Related to Suicide, Alcohol Use, or Drug Use? An Analysis Using Death Certificate Data From Colorado Veterans

Abstract

Knowledge regarding deaths due to suicide or alcohol- or drug-related causes may be limited by inconsistent and/or restrictive case definitions, resulting in concerns regarding validity of findings and underestimates of burden. In this proof-of-concept study, we assessed varying case definitions (suicide, alcohol-related, and drug-related mortality using underlying-cause-of-death (UCOD) versus multiple-cause-of-death (MCOD) International Classification of Diseases, Tenth Revision (ICD-10) codes) on the basis of counts and rates among Colorado veterans who died (2009–2020). Suicide, alcohol-related, or drug-related ICD-10 codes were identified, and 2 case definitions were compared: UCOD (qualifying ICD-10 code listed as the UCOD) and MCOD (qualifying ICD-10 code in any cause-of-death field). Of 109,314 decedents, the number of deaths and the age-adjusted mortality rate (per 100,000 persons) significantly increased when MCOD codes were included: n = 4,930 (110.3 deaths/100,000 persons) for UCOD versus n = 6,954 (138.4 deaths/100,000 persons) for MCOD. While rates of suicide mortality did not change, rates of alcohol-related mortality doubled with the more inclusive case definition: 1,752 (27.3 deaths/100,000 persons) for UCOD versus 3,847 (59.8 deaths/100,000 persons) for MCOD. Alcohol-use disorder codes accounted for 71% of additional alcohol-related deaths captured with the MCOD definition. Studies that rely on UCOD codes may be underestimating the burden of deaths, especially alcohol-related deaths. Increased effort is required to reevaluate current classifications of deaths associated with suicide, alcohol use, or drug use.

This article is linked to 'Invited Commentary: Stop Analyzing Suicides, Drug-Related Deaths, and Alcohol Related Deaths Together' and 'Adams et al. Respond to "Stop Analyzing 'Despair' Deaths Together" (https://doi.org/10.1093/aje/kwad002 and https://doi.org/10.1093/aje/kwad004).

Abbreviations

 
• ACME

  Automated Classification of Medical Entities

 
• CI

  confidence interval

 
• ICD-10

  International Classification of Diseases, Tenth Revision

 
• MCOD

  multiple cause of death

 
• RR

  rate ratio

 
• UCOD

  underlying cause of death

Editor’s note: An invited commentary on this article appears on page 732, and the authors’ response appears on page 734.

Driven by increases in suicide, alcohol-related deaths, and drug-related deaths, US life expectancy has been declining since 2000 (1, 2). Over the past 2 decades, our nation has experienced 2 emerging public health crises: the suicide epidemic and the opioid epidemic (3, 4). Although each of these crises has been investigated, there has been less work regarding how these crises may be overlapping. Moreover, over the past 2 decades there has also been a doubling of alcohol-related deaths, which are often overlooked in terms of their contribution to suicide and drug-related mortality (5, 6). Though many researchers have used death certificate data to describe the burden of suicide, alcohol-related deaths, and drug-related deaths, as well as to increase understanding regarding population differences in risk, notable limitations remain (7–16).

First, instead of incorporating multiple-cause-of-death (MCOD) indicators, many studies use only the underlying cause of death (UCOD) to identify suicide, alcohol-related, and drug-related deaths (7–16). According to the World Health Organization, a UCOD is assigned as “the disease or injury which initiated the train of events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury” (17, p. 1). However, including MCOD fields (“the immediate cause of death and all other intermediate and contributory conditions listed on the death certificate” (17, p. 1)) can provide additional relevant information. Limiting data to the UCOD may result in an underestimate of the true burden of deaths (18, 19). For example, alcohol- and drug-use disorders can be implicated in both external (e.g., falls and overdoses) and internal (e.g., alcohol-related pancreatitis) causes of death, though only the former may be captured when limiting case definitions to UCOD indicators. Furthermore, researchers have found that when the cause of death is multidetermined (e.g., diabetes), use of only UCOD codes can result in an undercount of deaths from a given cause by nearly half (20–22).

Second, a lack of consensus remains regarding which causes of death should be included when looking at these 3 types of deaths together. While some studies focus only on suicides, drug overdoses, and all chronic liver disease deaths (7, 13, 16, 23), others include suicides and all alcohol- and drug-related causes (8, 9, 15). The first case definition may miss deaths which are indeed alcohol- or drug-related (e.g., alcohol use disorder) or wrongly include liver disease deaths that are not due to alcohol (e.g., hepatic fibrosis). Conversely, the latter may be overly inclusive (e.g., may include tobacco use).

Finally, studies have been conducted using trend analyses that combine mortality resulting from both acute alcohol/drug exposure (i.e., very high exposure in a short time; overdose) and chronic exposure (i.e., high exposure over a long period of time; alcohol-related cirrhosis) into 1 variable (8, 9, 14, 24). Because the natural histories of acute and chronic outcomes differ, mixing outcomes with heterogeneous latency periods could hinder the identification of causal factors (25). This may be especially salient when examining the association between secular trends (e.g., the Great Recession) and deaths, where the impact on acute mortality is generally observed over a shorter time frame while chronic alcohol- and/or drug-related mortality may take longer to observe.

Our purpose in this study was to assess these limitations and provide empirical findings to guide future efforts. Specifically, in this proof-of concept study, we analyzed data from Colorado veterans who died (2009–2020) to assess varying definitions (UCOD vs. MCOD) in terms of case counts, age-adjusted rates, and trends over time. We compared differences between case definitions overall and by type of death (suicide, alcohol-related, or drug-related (aim 1)). We posited that a more restrictive case definition, limited to UCOD, would result in an undercount of the burden of such deaths. Second, to understand how stratifying alcohol and drug case definitions could affect findings, we explored the extent to which alcohol-/drug-related deaths were the result of acute versus chronic substance exposure (aim 2). Finally, we evaluated which International Classification of Diseases, Tenth Revision (ICD-10) codes occurred most often in UCOD versus MCOD fields to understand which deaths are more likely to be missed with a restrictive definition using only UCOD (aim 3).

METHODS

Data sources

In the United States, death certificates provide specific information about factors that caused and contributed to the decedent’s death based on the judgment of an attending physician, coroner, medical examiner, or other clinician. Causes of death (i.e., MCOD) are provided sequentially starting with the immediate cause and ending with the underlying cause, or “disease or injury that initiated the events resulting in death” (17, p. 1). The Automated Classification of Medical Entities (ACME), developed by the National Center for Health Statistics, applies rules outlined by the World Health Organization to take MCOD ICD-10 codes (26) generated from the death certificate and assign a UCOD, returning the remaining MCOD codes (up to 20) in alphanumerical order (27). In this process, the last reported condition on the death certificate is used as the UCOD, unless it could not have given rise to the other conditions listed (28).

The Colorado Department of Public Health and Environment Vital Statistics Program provided death certificate data for all Colorado residents who died during the study period (2009–2020) and had served in the US Armed Forces. Data included the decedent’s year of death and age at death, sex, race, ethnicity, marital status, and census tract of residence, the UCOD, and up to 11 MCOD codes. Census tract of residence was combined with Rural-Urban Commuting Area Codes to determine rurality (1–3 = urban; 4–7 = town; 8–10 = rural) (29). Estimates for the Colorado veteran population were obtained from the American Community Survey (30).

Case definitions

Death counts were derived from ACME codes using suicide codes and 176 codes (see Web Table 1, available at https://doi.org/10.1093/aje/kwac194) from the Centers for Disease Control and Prevention’s Wide-Ranging Online Data for Epidemiologic Research (WONDER) database for alcohol-/drug-indicated deaths that included both physical and mental health mortality tied to the use of alcohol or drugs (31). Mortality related to drugs linked with illicit use and overdose was included. Codes indicating “mental and behavioral disorders due to use of tobacco” (ICD-10 code F17) were excluded.

Deaths were considered overall and by type, categorized as suicide, alcohol-related, or drug-related, with suicides due to overdose of alcohol or drugs included in multiple categories. Alcohol and drug-related ICD-10 codes were further categorized into mutually exclusive categories of acute, chronic, and undetermined. Two physicians reviewed ICD-10 codes and descriptions and came to consensus on categorizations. Acute deaths were defined as those that can result from high use in a short period of time, with overdoses being the largest contributor to this category. Chronic alcohol- or drug-use deaths were defined as those requiring long-term use of a substance prior to development of the condition and listed as contributing to death (e.g., alcohol-related cirrhosis of the liver). Undetermined deaths were those that could result from either chronic or acute substance use (e.g., alcoholic liver disease, unspecified). Mental health codes indicating substance use as a cause of death (e.g., mental and behavioral disorders due to use of alcohol) were categorized on the basis of likely length of use given the code description. Specific codes within each category are presented in Web Table 1.

Two case definitions for deaths were identified: UCOD (i.e., the more restrictive case definition, including only the UCOD field to identify qualifying deaths) and MCOD (i.e., the more inclusive case definition, including all available ACME codes to identify qualifying deaths). We pooled numbers of deaths into four 3-year periods (2009–2011, 2012–2014, 2015–2017, and 2018–2020), by case definition and type of death (suicide, alcohol-related, and drug-related), to explore trends over time.

Analysis

We used multiple descriptive approaches to compare the numbers, rates, and differences in deaths by type and specific ICD-10 code in UCOD versus MCOD case definitions.

First, we compared the demographic characteristics (i.e., year of death, sex, age, race, ethnicity, marital status, and rurality) of suicide, alcohol-related, or drug-related decedents with those of all other decedents, using the restrictive UCOD case definition. We also compared demographic characteristics for suicide, alcohol-related, or drug-related deaths identified with the UCOD definition to those with qualifying suicide, alcohol-related, or drug-related codes present only in the MCOD definition. Chi-squared tests were used to identify statistically significant differences.

Next, we compared the numbers and annual age-adjusted suicide, alcohol-related, or drug-related mortality rates, by case definition, overall and for each type of death, with alcohol/drug death types considered overall and stratified by chronicity of use. Direct age adjustment, using the 2000 US Census population for standardization, was used to improve comparison of rates over time, given that age is associated with suicide, alcohol-related, or drug-related death and the age distribution for veterans changed over the study time period (25, 32). Mortality counts and rates were calculated for the full 12-year period and for 4 consecutive 3-year periods (2009–2011, 2012–2014, 2015–2017, and 2018–2020) to explore trends over time. Age-specific counts for the time period of interest were summed, and the midpoint (e.g., 2014 for 2009–2020 and 2010 for 2009–2011) American Community Survey 5-year veteran population estimates were used as the denominators to calculate rates for the whole time period. These rates were then divided by the number of years in the time period (i.e., 12 or 3) to obtain an average annual rate.

The “epitools” package (33) in R, version 4.0.5 (R Foundation for Statistical Computing, Vienna, Austria; www.r-project.org/), was used to calculate crude and age-adjusted mortality rates and 95% confidence intervals (CIs). To compare rates between case definitions, we calculated rate ratios (RRs) and 95% CIs using the “popEpi” package (34). To test for trends over time in age-adjusted rates between case definitions, we used Joinpoint, version 4.9.0.0 (35), also testing for parallelism (36). The best-fit trendlines by case definition were graphed. Significant slopes indicated change in the age-adjusted rate over time; a significant parallelism test indicated that the slopes differed from each other by case definition.

Finally, we identified most common ICD-10 codes by case definition. Results for the count of each code were grouped by type using the mutually exclusive categories and whether they were present in the UCOD or MCOD fields. To identify which suicide, alcohol-related, or drug-related ICD-10 codes are most often missed with the UCOD case definition, we limited results to those with qualifying MCOD codes only, identified the top 20 codes by frequency, and explored patterns by ICD-10 chapter.

Numbers of deaths and rates calculated on the basis of counts less than 10 were suppressed to maintain confidentiality (37). Figures were created using the “ggplot” package (38). P values less than 0.05 from 2-sided tests and 95% CIs excluding the null value (i.e., RR = 1.0) were considered statistically significant. This study was approved by the Department of Veterans Affairs Office of Research and Development, the Colorado Multiple Institutional Review Board, and the Colorado Department of Public Health and Environment Institutional Review Board.

RESULTS

Between 2009 and 2020, a total of 109,314 Colorado veterans died; 25 decedents were excluded (no ICD-10 codes listed). Eighteen percent of decedents had only 1 UCOD code present, 44% had up to 3 additional MCOD codes, and 38% had more than 3 MCOD codes. Five percent (n = 4,930) of decedents had a qualifying death code listed as their UCOD (Table 1). Most decedents (96%) were male. Compared with veterans dying from non–suicide, alcohol-related, or drug-related causes, those identified using the UCOD code were significantly younger, more often of non-White race, more often of Hispanic ethnicity, and more often divorced or never married, and more often lived in urban census tracts.

Case definition comparisons—aim 1

The number of decedents classified as having a suicide, alcohol-related, or drug-related death increased by 41% (n = 2,024) using the MCOD definition (n = 6,954) instead of the UCOD definition (n = 4,930) (Table 2). There were 618 (8.9%) decedents identified using either UCOD or MCOD codes that were included in more than 1 type of death (e.g., suicide and alcohol-related). Of those, 229 suicide decedents had a self-poisoning with alcohol (i.e., X65) or drug (i.e., X60–X62, X64) code, and 389 decedents had more than 1 qualifying ICD-10 code, which placed them into multiple type categories.

Decedents identified only with the MCOD case definition (i.e., those with a nonqualifying UCOD) were significantly older and more likely to be non-White and married (Table 1). Comparing the capture of each definition by time period, we noted that the proportion of decedents captured via the MCOD definition was higher in more recent years. Furthermore, most suicide, alcohol-related, or drug-related deaths identified only via the MCOD definition were alcohol- or drug-related. The number of alcohol- or drug-related deaths increased by 77% overall, while only 3 additional suicide deaths were identified with the MCOD definition.

More specifically, alcohol-related deaths contributed to most of the observed increase, with an additional 2,095 deaths identified using the MCOD case definition. As a result, age-adjusted rates of alcohol-related deaths more than doubled from the restrictive definition to the inclusive definition. Fifty-four percent of all deaths with alcohol mortality codes present were identified only when using the MCOD case definition (Figure 1). Drug-related mortality rates also increased significantly from the UCOD case definition to the MCOD case definition, but to a lesser degree (n = 361 deaths added; 25% of total).

No significant differences in trendlines (i.e., test for parallelism) in age-adjusted rates between case definitions were identified for suicide, alcohol-related, or drug-related deaths overall or by mortality type (Figure 2A–D; crude and age-adjusted rates are provided in Web Table 2). Testing for individual linear trends between age-adjusted rates from 2009–2011 to 2018–2020 identified only suicide rates as significantly increasing (P = 0.01). However, computing RRs and 95% CIs between 2009–2011 and 2018–2020, we found that age-adjusted mortality rates for overall suicide, alcohol-related, or drug-related deaths increased significantly for both the UCOD (RR = 1.33, 95% CI: 1.31, 1.34) and MCOD (RR = 1.32, 95% CI: 1.31, 1.33) case definitions (Figure 2A). Suicide rates also increased significantly (RR = 1.45, 95% CI: 1.42, 1.49), though no difference was observed by case definition, given how few additional suicides were added by using MCOD codes (Figure 2B). Additionally, alcohol-related mortality rates increased significantly for both the UCOD (RR = 1.36, 95% CI: 1.32, 1.40) and MCOD (RR = 1.27, 95% CI: 1.25, 1.29) definitions (Figure 2C). Meanwhile, drug mortality rates increased significantly only for the MCOD case definition (RR = 1.15, 95% CI: 1.11, 1.20) and not the UCOD definition (RR = 1.01, 95% CI: 0.95, 1.07) (Figure 2D).

Acute versus chronic alcohol- and drug-related deaths—aim 2

Significant increases in age-adjusted mortality rates using the MCOD definition (vs. UCOD) were observed for acute, chronic, and undetermined categories of alcohol-related deaths, with increases of 80%, 186%, and 249%, respectively (Table 2). For drug-related deaths, a significant increase across case definitions was observed for acute (12%) and undetermined (1,952%) subcategories. Most (81% for the MCOD definition) drug-related deaths were acute, while numbers of deaths related to chronic drug use were low regardless of case definition. This precluded meaningful comparisons across case definitions for the chronic drug category. Eighty-nine percent of acute drug-related deaths were identified using UCOD codes, while 95% of undetermined drug deaths were identified using MCOD codes.

When comparing age-adjusted rates over time for alcohol and drug mortality according to chronicity categories, alcohol mortality rates were consistently higher than drug mortality rates for chronic and undetermined mortality but not for acute mortality (Figure 2E–J). Trends in age-adjusted mortality rates for chronic alcohol deaths did not significantly change over time for either case definition, though trendlines differed significantly from each other (P < 0.001; Figure 2G). Similarly, trends in acute alcohol, acute drug, chronic drug, undetermined alcohol, and undetermined drug mortality did not significantly increase over time.

Suicide, alcohol-related, or drug-related ICD-10 codes—aim 3

Of 190 suicide, alcohol-related, or drug-related death codes considered for this analysis, 92 were present in one of the 11 MCOD fields in this population. The breakdown of the presence of each code by case definition is available in Web Table 3. Most (n = 5,437; 78.2%) decedents identified under either the UCOD or MCOD case definition had only 1 qualifying code on the death certificate; 13% had 2 qualifying codes, and the rest had 3–7 qualifying codes. “Mental and behavioral disorders due to use of alcohol, harmful use” (ICD-10 code F10.1; n = 1,171) was the most common qualifying code in this population, followed by “intentional self-harm by other and unspecified firearm discharge” (code X74; n = 993) and “alcoholic cirrhosis of the liver” (code K70.3; n = 934).

Of the 2,023 individuals with only qualifying MCOD codes and no UCOD codes, alcohol-related codes were the most commonly present, with “mental and behavioral disorders due to use of alcohol” (ICD-10 codes F10.1, F10.2, and F10.9), “alcoholic cirrhosis of the liver” (code K70.3), and “accidental poisoning by exposure to alcohol” (code X45) being included most frequently (Table 3). The most common UCOD codes in this group were cardiovascular disease (ICD-10 chapter I), external causes (chapters V, W, X, or Y), and cancer or blood disease (chapters C and D) (Web Table 4). Alcohol and drug use are known risk factors for many of these outcomes (e.g., heart disease, falls, traffic-related injuries, liver cancer, chronic viral hepatitis, etc.) (Web Table 5).

DISCUSSION

While an increase in suicide, alcohol-related, or drug-related deaths would be expected when expanding inclusion criteria, findings from this proof-of concept study highlight how including MCOD codes meaningfully increased the estimated burden of such deaths. Overall, 2,000 additional deaths were identified—a 41% increase. Most notably, when all available codes were used versus only the UCOD code, the alcohol-related mortality rate increased over 100%. This increase was driven by codes for alcohol use disorder, alcohol-related cirrhosis, and alcohol poisoning in UCOD fields. These findings suggest that use of a restrictive case definition based only on the UCOD probably results in deaths of interest being systematically underreported (7–16). Another notable finding was differences in UCOD versus MCOD case definitions by sex, race, and age. This finding requires further exploration in a larger population to increase understanding regarding how a narrow case definition may bias reporting by race, ethnicity, age, sex, and/or region (39).

More inclusive case definition

Relying solely on the UCOD is standard practice for mortality surveillance. While we found that adding MCOD codes might not necessarily change observed trends in mortality, research studies relying only on the UCOD may be underestimating the full burden of alcohol- and drug-related deaths. Low sensitivity of an outcome measure can be consequential if the purpose of research is to explore, understand, and eventually prevent negative consequences of such deaths. Similar issues have been identified in research on diabetes-related mortality, where half of deaths with diabetes indicated as a major contributor to mortality have other downstream causes (e.g., kidney failure or heart disease) listed as the UCOD (20–22).

Multidetermined causes connected to alcohol and drug use (e.g., alcohol-related cirrhosis or car accidents caused by intoxication) were undercounted by half with the more restrictive case definitions. Another study found that 31% of US decedents with alcohol-related liver disease indicated on the death certificate had a different UCOD (40). Inclusion of MCOD fields in case definitions provides a more robust picture of mortality and factors contributing to increases over time (10, 18, 19, 41). While some investigators used the MCOD field to specify opioid mortality (7, 23), we found no other studies that used the MCOD fields when examining suicide, alcohol-related, or drug-related deaths in the same study. Given our findings, in future studies researchers should consider use of the MCOD mortality files and fields to improve outcome sensitivity.

Relevant inclusion criteria are study-specific

Inclusion criteria for ICD-10 codes in the literature are not necessarily built around an agreed-upon conceptual model. This contributes to variability. Some studies are overly inclusive of codes—for example, including tobacco use disorder or outcomes that are more likely due to an adverse reaction to medication rather than an illicit substance (e.g., “drug-induced thyroiditis”) (9, 15). Other studies include liver disease codes that are not specific to alcohol (e.g., “chronic hepatitis,” “fibrosis,” “cirrhosis of the liver”) (1, 8, 12, 13, 16, 23). Conversely, some studies use a narrower definition, focusing specifically on deaths due to suicide, overdose, and alcoholic liver disease (2, 7, 10, 11). This definition excludes other alcohol- and drug-related deaths, such as alcohol- or substance-use disorder.

We found that substance-use disorder codes were common in the MCOD fields, especially for alcohol. Excluding these codes could result in missing important deaths. Ultimately, more work is needed to establish a valid and relevant case definition with inclusion criteria consistent with a conceptual theory linking mechanisms to mortality. Until consensus is achieved, we propose that investigators consider which causes of death are most pertinent to their research question.

Stratify acute and chronic conditions

Difficulties remain in defining subcategories of suicide, alcohol-related, or drug-related deaths to minimize differences in the natural history between exposure and mortality. For example, alcohol poisoning results from extreme exposure within a day, while alcohol-related liver disease typically develops after long-term excessive alcohol consumption (42). In causal research identifying the association between an exposure and an outcome, having heterogeneous outcomes composed of acute and chronic conditions can increase type II error (25). As such, research investigating causal models for such deaths will require thoughtful stratification. Some researchers have addressed this using the categories suicide, overdose, and liver disease (7, 11, 13, 23).

Specifically, we proposed an approach to stratifying deaths into chronic substance exposure versus acute substance exposure, recognizing that some codes are not specific enough and should be categorized as undetermined. The benefit to this approach is that it provides some logical categorization of the deaths into potential proximal and distal causes of these deaths. Notably, this approach also has limitations. For example, for the current analysis, we coded opioid overdoses as acute, though a known risk factor for opioid overdose is long-term opioid use (43), so the time frame between initiation of opioid use and opioid overdose could be longer. Additionally, large proportions of the alcohol and drug deaths were marked as undetermined, complicating easy stratification into meaningful groups. To improve the validity of causal research, efforts to identify best approaches to stratifying deaths or improved ways of dealing with heterogenous outcomes are warranted.

Study limitations

This study had several limitations. First, our analysis was limited to veterans. Findings may not be generalizable to nonveteran civilians. Likewise, our sample was limited to the state of Colorado. Colorado has a coroner system (i.e., elected officials who may or may not have formal medical or forensic training oversee death investigations), and thus these findings may not be generalizable to states with a medical examiner system or to the United States overall (44).

In addition to potentially limited generalizability, we note possible information bias in important death certificate variables. First, to define the study sample, we used a death certificate variable indicating whether decedents had ever served in the US Armed Forces. Thus, deaths could include current active-duty service members or exclude veterans for whom this information was not correctly specified on the death certificate. That said, in one study in Colorado, Bahraini et al. (45) found that accuracy of correctly specifying veteran status is fairly high. In addition, suicide, alcohol-related, or drug-related deaths are broadly known to be underreported on death certificates. For example, suicides could be misclassified as unintentional or undetermined, so the overall burden is probably higher than what we report (46–48). Finally, we did not find significant linear trends in most types of death by case definition, which could be because we did not have enough statistical power with only 4 time points or that trends over time were not necessarily linear.

Conclusion

In summary, a consensus approach to developing a case definition for suicide, alcohol-related, and drug-related deaths is required to improve the quality and validity of emerging research. We provide proof-of-concept data which suggest that case definitions frequently employed (i.e., UCOD only) are probably too narrow. Current approaches may be underestimating the true burden of such deaths, especially those related to alcohol. This study should be repeated in a national data set with special consideration of subpopulations (e.g., race/ethnicity, sex, region) to understand the possible impact of discrimination or coroner–versus–medical examiner systems on UCOD versus MCOD classification. Given our findings, existing research could be limiting epidemiologists’ ability to call attention to this public health crisis and is minimizing alcohol as a major contributor. Additionally, depending on the study, meaningful stratification of deaths should be considered. Investigators conducting causal research should consider the possible natural course of disease to a mortality outcome and conduct analyses accordingly. Improved case definitions can improve research and thus help facilitate public health action to address troubling trends in mortality.

ACKNOWLEDGMENTS

Author affiliations: Rocky Mountain Mental Illness Research Education and Clinical Center (MIRECC) for Veteran Suicide Prevention, Rocky Mountain Regional VA Medical Center, Aurora, Colorado (Talia L. Spark, Rachel Sayko Adams, Claire A. Hoffmire, Jeri E. Foster, Lisa A. Brenner); Department of Physical Medicine and Rehabilitation, Anschutz School of Medicine, University of Colorado, Aurora, Colorado (Talia L. Spark, Claire A. Hoffmire, Jeri E. Foster, Lisa A. Brenner); Injury and Violence Prevention Center, Colorado School of Public Health, University of Colorado, Aurora, Colorado (Talia L. Spark, Lisa A. Brenner); Institute for Behavioral Health, Heller School for Social Policy and Management, Brandeis University, Boston, Massachusetts (Rachel Sayko Adams); Department of Psychiatry, Anschutz School of Medicine, University of Colorado, Aurora, Colorado (Lisa A. Brenner); and Department of Neurology, Anschutz School of Medicine, University of Colorado, Aurora, Colorado (Lisa A. Brenner).

This work was supported by the US Department of Veterans Affairs, Rocky Mountain MIRECC for Veteran Suicide Prevention.

The Colorado mortality data analyzed during the current study are not publicly available due to Colorado state guidelines regarding confidentiality. Data can be requested through the Colorado Department of Public Health and Environment request system (https://www.datarequest.dphe.state.co.us/) and may require institutional review board approval.

We thank Drs. Joseph Simonetti and Hal Wortzel for reviewing ICD-10 codes to assign “acute,” “chronic,” or “undetermined” categorizations, Alexandra Schneider and Carlee Kreisel for help with data management and regulatory approvals, the Colorado Department of Public Health and Environment for providing the data, and Kirk Bol for answering questions related to the data and the national vital statistics surveillance process.

The views expressed in this article are those of the authors and do not necessarily represent the views or policy of the Department of Veterans Affairs or the US government.

T.A.S. reports employment with Westat, Inc. (Rockville, Maryland). R.S.A. reports receiving grants from the National Institutes of Health and the Henry M. Jackson Foundation for the Advancement of Military Medicine (Bethesda, Maryland) on behalf of the Uniformed Services University of the Health Sciences. She consults for The Informatics Applications Group (TIAG; Reston, Virginia) in support of the National Intrepid Center of Excellence at Walter Reed National Military Medical Center. C.A.H., J.E.F., and L.A.B. report receiving grants from the Department of Veterans Affairs, the US Department of Defense, and the National Institutes of Health. J.E.F. and L.A.B. additionally report receiving grants from the state of Colorado. L.A.B. reports receiving editorial renumeration from Wolters Kluwer N.V. (Alphen aan den Rijn, the Netherlands) and the RAND Corporation (Santa Monica, California) and royalties from the American Psychological Association (Washington, DC) and Oxford University Press (Oxford, United Kingdom). In addition, she consults with sports leagues via her university affiliation.

REFERENCES