Though rare, suicide is not rare enough in VA settings. Early identification of suicidality and its risk factors is crucial for effective intervention to prevent this disastrous outcome. Luoma et al. have found that 45% of those who committed suicide had contact with a primary care provider (PCP) in the month prior to suicide,(Luoma, Martin et al. 2002) thus highlighting the importance of PCPs in identifying and intervening with the 2-3% of primary care patients who endorse suicidal ideation.(Gaynes, West et al. 2004)
Based on a large, nationally representative sample (N=320,890), a recent publication reported that military veterans are twice as likely to commit suicide as non-veterans.3 Chronic pain and depression - two conditions not uncommon in VA medical settings -- are leading contributing factors in deaths by suicide. A systematic literature review of pain and suicide, suicide attempts, and suicidal ideation revealed that patients with chronic pain had a two-fold risk of death by suicide, a 14% prevalence of suicide attempts (compared to 5% without chronic pain), and a 20% prevalence of suicidal ideation.4 The relationship of depression with suicidality is well known,5 and we have demonstrated in our own work that VA patients with post-traumatic stress disorder (PTSD) have a significantly greater suicide risk than those without PTSD.6 Older age and male gender are also highly related to suicide - both being predominant characteristics of VA patients.(Gaynes, West et al. 2004)
In our previous work we have studied a number of domains related to the present proposal. For example, we have demonstrated a significant relationship between pain (SF-36 pain score) and provider recognition of depression7 and between diagnoses involving musculoskeletal pain and provider recognition of PTSD8; however, suicidality was not included in these analyses.
The overall objective of this secondary data analysis project was to determine meaningful correlates of suicidality in the context of VA primary care that would prove useful to clinicians for effective detection and intervention. The first aim was to determine the relationship of patient level characteristics and patterns of service use to suicidality in VA primary care patients. In addition to socio-demographic characteristics, we examined psychiatric diagnoses, unrecognized psychiatric diagnoses, medical diagnoses associated with pain, self-reported experience of pain, service use (primary care, mental health, and other specialty), service intensity, and medication management. The second aim of this project was to consider the relationship of suicidality to health status and burden of disease, as measured by preference-weighted health status (PWHS)9 and the Charlson Comorbidity Index (CCI)10, respectively. These measures complement each other in that PWHS is based on patients' subjective reports reflecting desirability (or lack thereof) for specific health states while the Charlson comorbidity index is based on physician recorded ICD-9 diagnoses and reflects mortality risk. Preference weights (derived using standard gamble approach) are applied to specific health states (SF-36 profiles) for each patient, yielding a preference weighted health status coefficient ranging from 0 (death) to 1 (perfect health). The CCI applies a coding algorithm to each patient's ICD-9 data to provide a global assessment of a patient's 1-year mortality risk attributable to comorbidity.
Overall, 8.71% of patients (77/884) endorsed items indicating suicidal ideation, meaning that at least one of the six module C items was coded 'yes'; these patients were classified, by the MINI, as "suicide risk current" (see Appendix A). The scoring algorithm for module C of the MINI also provides a level of suicide risk based on the number and pattern of affirmative responses. In our data only one patient responded in a manner that indicated a high level of suicide risk. The remaining patients were classified as either moderate (n=16) or low (n=53) risk. Even though the presence of suicidality was determined, suicide risk level was not available for 7 patients.
Due to the small number of patients in the high and moderate risk categories, and because any level of suicide risk is clinically important, we compared patients with no suicide risk (n=884) to those with any suicide risk (n=77).
Table 1 shows the socio-demographic characteristics of our sample by suicidality status with simple chi-square results. Table 2 shows unadjusted and adjusted (age, gender, race, education) odds ratios for the same characteristics. Middle-aged patients (50 to 64 years old) were the most likely to endorse suicidality (11.8%) followed by younger patients (<50 years old) (8.5%), with older patients (65 and older) (5.9%) the least likely to endorse suicidality items. The unadjusted odds ratio comparing middle aged with older patients was significant (OR=0.47; CIs 0.27, 0.83). Education was also significant, with those with a college degree or greater least likely to endorse suicidality, and significantly less likely than those with less than a high school diploma (OR = 0.35; CIs 0.13, 0.98). Veterans working had lower odds of suicidality than those not working due to disability (OR=0.23; CIs 0.12, 0.45) as did those not working due to retirement (OR=.38; CIs 0.22, 0.67). Those who served in a warzone were more likely to be suicidal than those without warzone service (OR = 2.16; CIs 1.33, 3.51). Gender, race, marital status, and war era were not significantly related to suicidality.
We next conducted multi-variable analyses controlling for age, gender, race, and education. All univariate findings were confirmed; additionally the relationship between younger versus older veterans became statistically significant with the oldest group having lower odds of endorsing suicidality items (OR = 0.43; CIs 0.18, 0.98).
We examined functional status using SF-36 scores. For each of the subscales investigated (general health, mental health, vitality, physical functioning, social functioning, role physical, role emotional, pain), suicidal patients had significantly worse functioning (See Table 3). Because some of the subscales on the SF-36 have unusual distributions, we also ran analyses using the ranksum test; the results were unchanged. These relationships persisted after adjustments for age, gender, race, and education (Table 4). The composite mental health score combining the mental health, social functioning, and role emotional domains was also significant, as was the composite physical health score which combines the general health, vitality, physical functioning, role physical, and pain domains.
Preference weighted health status
Based on SF-36 responses, we calculated preference weighted health status for each patient according to the method of Brazier et al. (2002). Suicidal patients had significantly worse preference weighted health status (0.652) than non-suicidal patients (0.546) (p=0.000) (see Table 5).
We examined the presence of current psychiatric disorders as measured by the MINI and CAPS (for PTSD) (see Table 6). In every category measured (depression including dysthymia, generalized anxiety disorder, PTSD, and substance use disorders), suicidal patients were much more likely to be suicidal. Not surprisingly, 80.5% of those who were suicidal were depressed compared with 18.2% of those who were not suicidal.
ICD9 pain related diagnoses
We grouped patients according to whether they had an ICD9 diagnosis related to back, chest, neurological, musculoskeletal, or "other" pain (see Appendix B for groupings by diagnosis). Suicidal patients were more likely to have chest pain than non-suicidal patients (21.6% vs. 9.8%; p=0.002) and musculoskeletal pain (59.6% vs. 79.7%; p=0.001); findings were not significant for back, neurological, or "other" pain. Overall, 87.8% of patients endorsing suicidality received one or more pain diagnosis compared to 68.9% of patients with no suicidal ideation (p=001). (See Table 7)
Charlson Comorbidity Index
As a measure of overall medical morbidity, we used ICD9 codes to calculate a CCI for every patient. This index ranges from 0 to 2, with 0 the best and 2 the worst. The CCI distribution for suicidal and non-suicidal patients was not statistically significantly (see Table 8).
Health services use
For the two year observation period, we collected each patient's use of VA ambulatory health services. We focused on use of primary care, psychiatric services, urgent care, and pain clinics. As shown in Table 9, suicidal patients were significantly more likely to have had at least one visit to mental health clinics, PTSD clinics, substance abuse clinics, and urgent care. In addition, as shown in Table 10, they had significantly more visits in every category. In particular, they had more primary care visits (13.7 vs. 10.4; p=0.003) and mental health visits (11.3 vs. 3.2; p=0.003). Of note, although suicidal patients were more likely to have mental health visits, 29.9% did not have any mental health visit of any kind.
We examined prescription medication fills in four categories: opioid, non-opioid, sedative/hypnotic, non-tricyclic antidepressants (see Table 11). Based on fill patterns during the two-year observation period, for each medication category we classified patients into those who had no fills, 1-2 fills, 3-5 fills, or >5 fills. For pain medications, suicidal patients were more likely to have higher opioid fill rates (p=0.008), but not non-opioid fill rates (p=0.522). They were more likely to have higher sedative/hypnotic fill rates (p=0.000) and non-tricyclic antidepressant fill rates (p=0.000). We did not examine tricyclics, as dosing was not at therapeutic levels for depression.
Given the known relationship of depression and suicidality and the relationships suggested by our analyses of pain variables and suicidality, we conducted multivariable analyses to understand better the contribution of both pain and depression to suicidality. We first conducted logistic regressions with suicidality as the dependent variable and depression from the MINI (including dysthymia) and the SF-36 pain score as the independent variables. We also included our three age categories. Both depression and SF-36 pain were significant. We ran the same analyses, only replaced SF-36 pain with all of our ICD9 pain categories (back, chest, neurological, musculoskeletal, other) as independent variables. Depression was again highly significant, but only chest pain and musculoskeletal pain were significant as pain indicators. We made one index for chest pain and musculoskeletal pain, so that patients could have neither, one, or both (see Table 12). In a similar model with depression and both SF-36 pain and our chest/musculoskeletal pain index, SF-36 pain drops to non-significance. We then tested for interaction between chest/musculoskeletal pain and depression, and there was none. Thus, in the final model, only the odds ratios for depression (OR=16.29; CIs 8.16, 32.51) and pain (either musculoskeletal or chest OR=2.39; CIs 1.11, 5.17; or both musculoskeletal and chest OR=3.24; CIs 1.12, 9.40) were statistically significant.
In summary, VA patients reporting symptoms of suicidality have significantly different socio-demographic characteristics, clinical characteristics, and service use patterns than non-suicidal patients. While many of our findings are not unexpected (e.g., high prevalence of depression among suicidal patients), some are new. In contrast to other published reports, we found that middle-aged veterans had the highest percentage of suicidal patients and elderly patients the lowest. This may be a cohort related phenomenon related to service era, or it may reflect variations in how VA accepts veterans for care over time. We expected to find higher rates of depression and pain among suicidal patients; however, the two have rarely been measured simultaneously. Our MINI generated diagnoses for depression, combined with two ways of measuring pain (SF-36 and diagnostic classification) provide strong evidence for this association.
Our multivariable analyses simultaneously including pain (musculoskeletal and chest) and depression provide strong evidence of the separate contribution of each to suicidality. The presence of physician diagnosed pain conditions was a more robust predictor than self-reported pain.
We examined health services use patterns and medication patterns for additional relationships. While higher use of primary care, mental health clinics, pain clinics, and urgent care was expected among suicidal patients, we were surprised that 30% of suicidal patients had not had a single mental health visit of any kind in the one year prior or one year post assessment for suicide. Though not all suicidal patients met criteria for a mental disorder, it is worrisome that 30% of suicidal patients were not being more closely monitored by a mental health professional during this two year period.
External Links for this Project
- Magruder KM. Prevention and Management of Suicidality. Paper presented at: Institute for Defense and Government Advancement Military Healthcare Convention and Conference; 2010 Jun 24; San Antonio, TX. [view]