Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Mar 22.
Published in final edited form as: Suicide Life Threat Behav. 2010 Dec;40(6):620–627. doi: 10.1521/suli.2010.40.6.620

Likelihood of Suicidality at Varying Levels of Depression Severity: A Re-Analysis of NESARC Data

Lisa A Uebelacker 1, David Strong 1, Lauren M Weinstock 1, Ivan W Miller 1
PMCID: PMC13005707  NIHMSID: NIHMS2153501  PMID: 21198331

Abstract

Although it is clear that increasing depression severity is associated with more risk for suicidality, less is known about at what levels of depression severity the risk for different suicide symptoms increases. We used item response theory to estimate the likelihood of endorsing suicide symptoms across levels of depression severity in an epidemiological data set. Regardless of depression severity, suicide attempts were less frequently endorsed than ideation, which was less frequently endorsed than feeling like one wanted to die. All suicide symptoms were generally less likely to be endorsed than other depression symptoms. There was a low probability of suicidality at depression levels that likely would not merit a diagnosis of major depression.

Although suicidality occurs in the context of many different psychiatric disorders (Kessler, Borges, & Walters, 1999), it is one of nine DSM symptoms of major depressive disorder (MDD). It is well established that increasing depression symptom severity is a risk factor for different aspects of suicidality, including suicide ideation (Casey et al., 2006; Garlow et al., 2008), suicide attempts (Fordwood, Asarnow, Huizar, & Reise, 2007), and completed suicide (Blair-West, Mellsop, & Eyeson-Annan, 1997; Kessing, 2004). Previous analyses of the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) data set have documented that, among individuals who meet criteria for MDD, depression severity predicts likelihood of a suicide attempt (Bolton, Belik, Enns, Cox, & Sareen, 2008). In the U.S. National Comorbidity Survey (NCS), another epidemiological survey, having a diagnosis of MDD increased the risk of both suicide ideation (OR = 11.0) and suicide attempt (OR = 9.6; Kessler et al., 1999).

Although it is clear that increasing depression severity is associated with more suicidality, less is known about at what levels of depression severity the risk for various different aspects of suicidality increases, whether there is a depression severity level at which the risk increases rapidly, and by how much risk increases. Further, suicidality describes a range of thoughts and behaviors, and different aspects of suicidality are likely to occur at different levels of depression severity. Such information would give clinicians a better estimate of the risk for suicidality in patients with different levels of depression. The purpose of this study was to use the NESARC data set (Grant, Dawson, et al., 2004; Grant, Moore, Shepard, & Kaplan, 2003) to characterize the risk for different aspects of suicidality at various levels of depression severity. We examined four different aspects of suicidality corresponding to the four suicide questions asked of NESARC respondents: thoughts of one’s own death, feeling like one wanted to die, thoughts of suicide, and suicidal attempt. Specifically, we planned to: (1) test the hypothesis that, across all levels of depression severity, suicide attempt would be less frequently endorsed than thoughts of suicide, which in turn would be less frequently endorsed than both thoughts of one’s own death and feeling like one wanted to die; (2) compare rates of endorsement of the four aspects of suicidality to rates of endorsement of other depression symptoms across levels of depression severity; and (3) understand the likelihood of endorsing different aspects of suicidality at different levels of depression severity.

In order to meet these aims, we used methods based in item response theory (IRT, cf., Lord, 1980). IRT provides significant improvements on previous techniques (e.g., simply examining frequency counts of particular symptoms); IRT approaches allow one to examine the likelihood that a particular suicide symptom will be reported given a particular level of depression severity. Application of IRT methods is emerging in the evaluation of DSM-IV diagnostic criteria, including criteria for depression (Aggen, Neale, & Kendler, 2005; Simon & Von Korff, 2006), alcohol dependence (Kahler, Strong, Stuart, Moore, & Ramsey, 2003), and bulimia (Rowe, Pickles, Simonoff, Bulik, & Silberg, 2002). In a previous study, we used the NESARC data set to examine whether there were gender or race/ethnic differences in rates endorsement of various depression symptoms at different levels of depression severity (Uebelacker, Strong, Weinstock, & Miller, 2009).

METHOD

Data Collection

The NESARC is a survey of a nationally representative sample from the United States. Methods for obtaining the sample have been detailed elsewhere (Grant, Dawson, et al., 2004; Grant, Moore, et al., 2003). The NESARC involved interviews from 43,093 adults age 18 and older. Only those respondents who reported either depressed mood (n = 12,785) or anhedonia (n = 10,533) for at least 2 weeks in their lifetime completed the section of the NESARC survey assessing occurrence of all DSM-IV symptoms of major depressive disorder, including suicide-related symptoms during the time when depressed mood or anhedonia was at its worst. The present analysis consisted of those individuals (n = 13,753; or 32% of the total sample). This subsample had a mean age of 45.87 (SD = 17.16; range = 18–98) and included 65.7% women (n = 9,040). Individuals chose their racial group from the following categories: American Indian or Alaska Native (n = 468), Asian (n = 291), Black or African American (n = 2,396), Native Hawaiian or other Pacific Islander (n = 101), and White (n = 10,958). Individuals were also asked whether they were Latino or Hispanic; 2,258 responded affirmatively. A majority (83.2%) of the sample had completed high school or its equivalent, and 55.3% had completed college.

Symptoms of Major Depressive Disorder.

The Alcohol Use Disorders and Associated Disabilities Interview Schedule (AUDADIS; Grant, Dawson, & Hasin, 2001; Grant, Dawson, et al., 2003) was used to assess DSM-IV MDD criteria (AUDADIS-IV). Experienced interviewers received extensive training in this fully-structured interview and used computer-assisted software to decrease error in measurement (Grant, Stinson, et al., 2004). Developers of the AUDADIS-IV invested considerable effort to ensure that questions were comprehensible for lay persons (Grant, Dawson, et al., 2003). NESARC estimates of lifetime and 12-month prevalence of MDD were 13.2% and 5.3%, respectively (Hasin, Goodwin, Stinson, & Grant, 2005). These estimates are comparable to those found in the National Comorbidity Survey (Kessler, Berglund, Demler, Jin, & Walters, 2005; Kessler, Chiu, Demler, & Walters, 2005). Test-retest reliability for the MDD diagnosis was good in this sample (Grant, Dawson, et al., 2003).

The AUDADIS-IV includes four questions about suicidality. All refer to a time period during which the individual reported feeling depressed or anhedonic for at least 2 weeks. The stem for the four questions was: “During that time … did you …” The four items are: (1) “attempt suicide?” (suicide attempt); (2) “think about committing suicide?” (thoughts of suicide); (3) “feel like you wanted to die?;” and (4) “think a lot about your own death?”

Analyses

Item response modeling allowed us to estimate rates of endorsement of suicide symptoms while accounting for levels of depression severity. A 2-parameter model involved estimating the following for each symptom (or item): a severity parameter to describe the point on the latent continuum (i.e., depression severity) where a symptom becomes likely to be observed (e.g., likelihood exceeds 50%), and a discrimination parameter to describe how rapidly the probability of observing the symptom changes across increasing levels of the latent continuum (i.e., slope of the item response function).

Unidimensionality Assumption.

The primary assumption of unidimensional item response models is that responses to symptom queries are a function of individual variation along a single underlying dimension. We tested this assumption previously (Uebelacker et al., 2009) using confirmatory factor analyses of tetrachoric correlations with the robust weighted least squares method of parameter estimation. Fit statistics for the sample as a whole, X2 = 642.0; CFI = 0.978; TLI = 0.979; RMSEA = 0.059, indicated a reasonable fit to the data. We determined that these fit statistics were sufficient to proceed to fitting unidimensional IRT models.

Parametric Item Response Model.

Given previous support for unidimensionality and utility of fitting parametric models to the symptoms of MDD, we evaluated a 2-parameter parametric model. With four questions targeting suicidal ideation and behavior, responses within this set of questions share a relationship that extends beyond associations with increasing levels of depression. This relationship within a set of questions violates assumptions of standard unidimensional IRT models. However, the common variability within this set of questions about suicide can be accommodated statistically by expanding the standard IRT model to include a term to represent variability that is common to the group of questions or “testlets” (Thissen, 1993; Wainer, Bradlow, & Wang, 2007). This method, based in Testlet Response Theory (TRT) (Wainer et al., 2007), allows for the estimation of the relative discrimination and severity of each of the suicide questions using a metric representing overall depression severity.

In the testlet model (SCORIGHT; Wang, Bradlow, & Wainer, 2005), we employed Bayesian methods (Gelman, Carlin, Stern, & Rubin, 1995) for obtaining estimates of statistical parameters that reflect the level of depressive symptoms and discriminative power associated with each DSM-IV symptom. The model also generates a single parameter (Theta) reflecting each individual’s latent continuous level of depressive symptom severity. This method focuses on obtaining samples from the posterior distribution of each of the model parameters, accomplished by using a Markov chain that repeatedly samples for a parameter until it converges on a particular distribution (Markov chain Monte Carlo; MCMC). This posterior distribution then allows for posterior samples to be generated and used to make inferences. For example, in order to estimate the mean value for a parameter, one simply takes the mean value of the posterior sample for that parameter. A primary advantage of MCMC is that these methods do not rely on large-sample theories to estimate standard errors for a parameter (Wang et al., 2005).

RESULTS

First, we used IRT to test the hypothesis that, across levels of depression severity, suicide attempt would be less frequently endorsed than thoughts of suicide, which in turn would be less frequently endorsed than both thoughts of one’s own death and feeling like one wanted to die. As depicted in Figure 1, suicide attempt was less frequently endorsed than thoughts of suicide across all levels of depression severity, and thoughts of suicide was less frequently endorsed than feeling like one wanted to die. However, the item “thoughts of [one’s] own death” had a less steep slope than the other three items; that is, it was more likely to be endorsed than other suicide items at lower levels of depression, but less likely to be endorsed than two of the three other items at higher levels of depression.

Figure 1.

Figure 1.

Open in a new tab

Likelihood of endorsing suicide items across varying levels of depression severity

*Refers to mean expected number of symptoms endorsed of the total number of seven possible DSM-IV depressive symptoms, not including anhedonia or sad mood.

When we examine rates of endorsement of other individual depression symptoms compared to the suicide items across levels of depression (Table 1), we find that the four suicide items index a higher level of depression severity than all other measured depression symptoms. That is, the level of depression severity at which one would expect a 50% likelihood of endorsing each of the four suicide items (i.e., the severity parameter) is higher than the level of depression severity required for a 50% likelihood of endorsing any of the other depression symptoms. Inspection of ICCs from all MDD symptoms (see Figure 1) show that both suicide attempt and thoughts of one’s own death were less likely to be endorsed than all other symptoms of MDD across all levels of depression severity. The other two items, “thoughts of suicide” and “feeling like one wanted to die,” were also less likely to be endorsed than other depression items at most levels of depression severity.

TABLE 1.

Initial Symptom Parameters in Entire Subsample (n = 13,753)

DSM-IV MDD Symptoms Severity Parameter SE Discrimination Parameter SE
3. Appetite/weight disturbance −0.79 0.07 1.06 0.09
4. Sleep disturbance −0.99 0.08 1.83 0.15
5. Psychomotor symptoms 0.04 0.02 1.47 0.12
6. Fatigue −0.51 0.04 1.45 0.12
7. Worthlessness/guilt −0.19 0.02 1.64 0.13
8. Concentration −0.81 0.07 1.77 0.14
9. Suicide attempt 2.05 0.17 2.69 0.24
10. Thoughts of suicide 0.90 0.07 3.68 0.32
11. Felt like one wanted to die 0.62 0.05 4.57 0.45
12. Thoughts of one’s own death 1.19 0.10 1.41 0.11
Open in a new tab

Note. Symptoms 1 and 2 are sadness and anhedonia, respectively, and are not included in this table because participants were required to endorse one or both symptoms to be part of the sample.

Our final goal was to understand the likelihood of endorsing different aspects of suicidality at different levels of depression severity. In order to provide a more intuitive feel for our index of depression severity (θ), we calculated the mean expected number of symptoms (of the total number of DSM-IV depression symptoms endorsed; however, not including anhedonia or sad mood) across levels of depression severity (see top of Figure 1). Note that the required endorsement of either sad mood or anhedonia in order to be included in the data analysis means that the average number of depression symptoms (θ = 0) should be interpreted as the average number of depression symptoms in a group of people who endorse either sad mood or anhedonia for at least 2 weeks in their lifetime rather than the average number of depression symptoms in the population as a whole.

As depicted in Figure 1, we note that in this sample the likelihood of endorsing thoughts of suicide does not increase dramatically until above the average level of depression severity (θ = 0); the mean number of symptoms endorsed at the average level of depression severity is between 4 and 5 (not including endorsement of sad mood or anhedonia or both). Even at the average level of depressive severity, the probability of endorsing any of the four suicide symptoms is relatively low (ranging from <1% for suicide attempt, to 15% for “thoughts of own death”). We note that even a person with a depression severity 1 SD over the mean (i.e., θ = 1, which corresponds to an average of 7 of the additional MDD symptoms) has only a 59% likelihood of endorsing thoughts of suicide, and a 6% likelihood of reporting a suicide attempt during that particular episode.

DISCUSSION

Our purpose in this article was to provide a clinical picture of rates of endorsement of various aspects of suicidality during a given two-week time period at different levels of depression severity during that same two-week time period. As expected, across all levels of depression severity, suicide attempts were less frequently endorsed than thoughts of suicide, which in turn was less frequently endorsed than “felt like one wanted to die.” These three items all had relatively steep slopes as well (as indexed by the discrimination parameter), indicating that there is a threshold of depression severity at which risk increases sharply.

An important point is that all four aspects of suicide stand out as characteristic of more severe depression severity than all six of the other symptoms of depression included in this analysis (i.e., appetite/weight disturbance, sleep disturbance, psychomotor symptoms, fatigue, worthlessness/guilt, and concentration difficulties). That is, across most levels of depression severity, all four suicide items, including the least severe one—“felt like one wanted to die”—are less likely to be endorsed than the other six symptoms of depression. Because it is not until the average level of depression severity (i.e., θ = 0) that the rates of endorsement of all suicide symptoms increase in this sample, at lower levels of depression severity (where individuals likely would not have enough depression symptoms to meet criteria for MDD) individuals are unlikely to endorse any type of current suicidality. Clinicians may note that, based on these data, it is likely to be extremely unusual to have a suicide attempt when a patient does not have most or all of the other depression symptoms concurrently. Further, endorsement of aspects of suicidality among individuals who are subthreshold for MDD would be unexpected and should prompt exploration of problems not directly related to depression, which, in turn, may suggest additional treatment targets.

We note that the suicide item “thoughts of one’s own death” was not as strongly related to depression severity as the other three items. At lower levels of depression severity it was more likely to be endorsed than suicide ideation, but at higher levels of depression severity it was less likely to be endorsed than suicide ideation. This suggests that there are other factors besides depression that influence endorsement of “thoughts of one’s own death.” This may be because this question does not reference suicide at all (unlike the other three questions) and is more vague than the other questions. Clinically, this means that knowing that a patient thinks of his or her own death is not necessarily a good indicator of the severity of one’s current depression.

In evaluating study findings, it is important to consider limitations. One limitation is that it is necessary to endorse a certain level of depressed mood or anhedonia—namely, depressed mood most of the time for at least 2 weeks, or anhedonia for at least 2 weeks—in order to be included in the data analyses. Therefore, we may miss some participants with lower levels of depressed mood who should be considered to be part of the depressive spectrum. We acknowledge that results could be different in a sample that included individuals with lower levels of depression severity in the analyses. However, we note that fully 32% of the total NESARC sample endorsed depressed mood or anhedonia sometime in their life and were therefore included in our analyses. Furthermore, we were reassured by the similarity of our results to those of Aggen and colleagues (2005). In their community sample, all participants reported on all symptoms, regardless of whether they endorsed sad mood or anhedonia.

A second limitation of this study is the use of retrospective reporting. Participants were asked about a lifetime mood episode; for many participants, this was not a current mood episode. This means that individuals who were successful in their suicide attempt are not included in the sample. Although this is likely a small number of individuals relative to those who made attempts (Moscicki, 2001, based on epidemiologic data, estimated the attempt:death ratio as 18:1), we cannot rule out the possibility that the group who was successful in their suicide attempt exhibited different symptom patterns than what we found in this study.

Finally, as is true with any assessment instrument, the way in which the questions were asked—including the exact wording—may influence rates of affirmative responses. Had a different assessment instrument been used to assess similar constructs, there may have been similar response rates. This points to the need for replication of our findings. We do note that one advantage of the NESARC data set is that, because interviewers were trained carefully, we can be relatively confident that questions were asked in the same way across participants.

Notable strengths of this study include the use of a large epidemiological data set and statistical techniques that allow us to examine suicidality at different levels of depression severity. We found that all aspects of suicidality, including the less severe ones, are typically found only at relatively high levels of depression severity. It is encouraging to note that there is a low probability of endorsement of any type of suicidality at depression levels below those that would merit a diagnosis of MDD.

Acknowledgments

The National Epidemiologic Survey on Alcohol and Related Conditions was funded by the National Institute on Alcohol Abuse and Alcoholism with supplemental funding from the National Institute on Drug Abuse. Work on this manuscript was supported by grant MH067779 from the National Institute of Mental Health to Dr. Uebelacker. The authors have no conflicts of interest to report.

REFERENCES

  1. Aggen SH, Neale MC, & Kendler KS (2005). DSM criteria for major depression: Evaluating symptom patterns using latent trait item response models. Psychological Medicine, 35, 475–487. [DOI] [PubMed] [Google Scholar]
  2. Blair-West GW, Mellsop GW, & Eyeson-Annan ML (1997). Down-rating lifetime suicide risk in major depression. Acta Psychiatrica Scandinavica, 95, 259–263. [DOI] [PubMed] [Google Scholar]
  3. Bolton JM, Belik SL, Enns MW, Cox BJ, & Sareen J (2008). Exploring the correlates of suicide attempts among individuals with major depressive disorder: Findings from the National Epidemiologic Survey on Alcohol and Related Conditions. Journal of Clinical Psychiatry, 69, 1139–1149. [DOI] [PubMed] [Google Scholar]
  4. Casey PR, Dunn G, Kelly BD, Birkbeck G, Dalgard OS, Lehtinen V, et al. (2006). Factors associated with suicidal ideation in the general population: Five-centre analysis from the ODIN study. British Journal of Psychiatry, 189, 410–415. [DOI] [PubMed] [Google Scholar]
  5. Fordwood SR, Asarnow JR, Huizar DP, & Reise SP (2007). Suicide attempts among depressed adolescents in primary care. Journal of Clinical Child and Adolescent Psychology, 36, 392–404. [DOI] [PubMed] [Google Scholar]
  6. Garlow SJ, Rosenberg J, Moore JD, Haas AP, Koestner B, Hendin H, et al. (2008). Depression, desperation, and suicidal ideation in college students: Results from the American Foundation for Suicide Prevention College Screening Project at Emory University. Depression and Anxiety, 25, 482–488. [DOI] [PubMed] [Google Scholar]
  7. Gelman A, Carlin J, Stern H, & Rubin D (1995). Bayesian data analysis. London: Chapman and Hall. [Google Scholar]
  8. Grant BF, Dawson DA, & Hasin DS (2001). The Alcohol Use Disorders and Associated Disabilities Interview Schedule—DSM-IV Version (AUDADIS-IV). Bethesda, MD: National Institute on Alcohol Abuse & Alcoholism. [Google Scholar]
  9. Grant BF, Dawson DA, Stinson FS, Chou PS, Kay W, & Pickering R (2003). The Alcohol Use Disorder and Associated Disabilities Interview Schedule-IV (AUDADIS-IV): Reliability of alcohol consumption, tobacco use, family history of depression and psychiatric diagnostic modules in a general population sample. Drug and Alcohol Dependence, 71, 7–16. [DOI] [PubMed] [Google Scholar]
  10. Grant BF, Dawson DA, Stinson FS, Chou SP, Dufour MC, & Pickering RP (2004). The 12-month prevalence and trends in DSM-IV alcohol abuse and dependence: United States, 1991–1992 and 2001–2002. Drug and Alcohol Dependence, 74, 223–234. [DOI] [PubMed] [Google Scholar]
  11. Grant BF, Moore TC, Shepard J, & Kaplan K (2003). Source and accuracy statement: Wave 1 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). Bethesda, MD: National Institute on Alcohol Abuse and Alcoholism. [Google Scholar]
  12. Grant BF, Stinson FS, Dawson DA, Chou SP, Dufour MC, Compton W, et al. (2004). Prevalence and co-occurrence of substance use disorders and independent mood and anxiety disorders: Results from the National Epidemiologic Survey on Alcohol and Related Conditions. Archives of General Psychiatry, 61, 807–816. [DOI] [PubMed] [Google Scholar]
  13. Hasin DS, Goodwin RD, Stinson FS, & Grant BF (2005). Epidemiology of major depressive disorder: Results from the National Epidemiologic Survey on Alcoholism and Related Conditions. Archives of General Psychiatry, 62, 1097–1106. [DOI] [PubMed] [Google Scholar]
  14. Kahler CW, Strong DR, Stuart GL, Moore TM, & Ramsey SE (2003). Item functioning of the alcohol dependence scale in a high-risk sample. Drug and Alcohol Dependence, 72, 183–192. [DOI] [PubMed] [Google Scholar]
  15. Kessing LV (2004). Severity of depressive episodes according to ICD-10: Prediction of risk of relapse and suicide. British Journal of Psychiatry, 184, 153–156. [DOI] [PubMed] [Google Scholar]
  16. Kessler RC, Berglund P, Demler O, Jin R, & Walters EE (2005). Lifetime prevalence and age-of-onset distribution of DSM-IV disorders in the National Comorbidity Survey replication. Archives of General Psychiatry, 62, 593–602. [DOI] [PubMed] [Google Scholar]
  17. Kessler RC, Borges G, & Walters EE (1999). Prevalence of and risk factors for lifetime suicide attemps in the National Comorbidity Survey. Archives of General Psychiatry, 56, 617–626. [DOI] [PubMed] [Google Scholar]
  18. Kessler RC, Chiu WT, Demler O, & Walters EE (2005). Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the National Comorbidity Survey replication. Archives of General Psychiatry, 62, 617–627. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Lord FM (1980). Applications of Item Response Theory to practial testing problems. Hillsdale, NJ: Erlbaum. [Google Scholar]
  20. Moscicki EK (2001). Epidemiology of completed and attempted suicide: Toward a framework for prevention. Clinical Neuroscience Research, 1, 310–323. [Google Scholar]
  21. Rowe R, Pickles A, Simonoff E, Bulik CM, & Silberg JL (2002). Bulimic symptoms in the Virginia Twin Study of Adolescent Behavioral Development: Correlates, comorbidity, and genetics. Biological Psychiatry, 51, 172–182. [DOI] [PubMed] [Google Scholar]
  22. Simon GE, & Von Korff M (2006). Medical comorbidity and validity of DSM-IV depression criteria. Psychological Medicine, 36, 27–36. [DOI] [PubMed] [Google Scholar]
  23. Thissen D (1993). Repeating rules that no longer apply to psychological measurement. In Frederikson N, Mislevy RJ, & Bejar I (Eds.), Test theory for a new generation of test. Hillsdale, NJ: Erlbaum. [Google Scholar]
  24. Uebelacker LA, Strong D, Weinstock LM, & Miller IW (2009). Use of item response theory to understand differential functioning of DSM-IV major depression symptoms by race, ethnicity and gender. Psychological Medicine, 39, 591–601. [DOI] [PubMed] [Google Scholar]
  25. Wainer H, Bradlow ET, & Wang X (2007). Testlet Response Theory and its applications. New York: Cambridge University Press. [Google Scholar]
  26. Wang X, Bradlow ET, & Wainer H (2005). User’s guide for SCORIGHT (Version 3.0). A computer program for scoring tests built of testlets including a module for covariate analysis (Research Report 04–49). Princeton, NJ: Educational Testing Services. [Google Scholar]

RESOURCES