The current study applied 3 previously used methodological approaches for identifying psychiatric comorbidity and examined the impact of these different methods on hospital mortality for 2 of the most common causes of hospitalization—CHF and pneumonia. We emphasize the following 3 findings. First, rates of psychiatric comorbidity varied according to the method used to identify such comorbidity. Rates were relatively similar using prior outpatient diagnoses or mental health clinic visits but were substantially lower using inpatient secondary diagnosis codes. Second, the agreement between the different approaches was only fair to moderate; agreement was highest between prior outpatient diagnosis codes and mental health visits. Third and perhaps most important, the risk of hospital death varied depending on the method of identifying psychiatric comorbidity. Thus, the method of identifying psychiatric comorbidity substantially affected the prognostic impact. These findings suggest that different diagnostic approaches likely identify individuals with different types of psychiatric comorbidity and different illness constructs.
In interpreting our findings, several explanations are possible. First, the findings of lower mortality for patients with secondary diagnoses of psychiatric comorbidity may indicate that these codes are a marker for lower unmeasured severity. Because the number of secondary diagnoses in discharge abstracts is limited, medical record coding may be more likely to include a psychiatric comorbidity in patients with less complicated hospital courses. This explanation may also underlie the lower mortality in some studies for patients with secondary diagnoses of hypertension and diabetes.16,17
Indeed, in the current analysis, patients with psychiatric comorbidity from inpatient codes had the lowest predicted mortality (Table )—supporting the hypothesis that secondary inpatient codes may serve as a marker for lower severity. Moreover, some prior studies have also demonstrated that patients with psychiatric comorbidity identified by secondary inpatient codes had lower mean predicted risk of mortality.7,9
Second, it is possible that the lower mortality in patients with psychiatric secondary diagnoses may be due to greater vigilance by providers or greater likelihood to admit lower severity patients with psychiatric comorbidities. Third, the lower mortality may reflect lower utilization of invasive diagnostic or therapeutic modalities. For example, 2 prior studies of acute myocardial infarction (AMI) found that patients with psychiatric secondary diagnoses were less likely to undergo coronary revascularization,9,20
which may be associated with worse hospital outcomes but better long-term outcomes.18
In contrast, a study of general medical and surgical inpatients4
found that patients with an inpatient diagnosis code of schizophrenia (0.2% of the sample) had a 2- to 2.5-fold higher risk of several adverse events (e.g., postoperative sepsis, iatrogenic infections).
Other studies of patients with ischemic heart disease provide further evidence of the potential impact of different diagnostic methods. In contrast to the findings noted above9,20
of lower use of coronary revascularization in patients with inpatient psychiatric codes, Jones and Carney19
examined privately insured patients with AMI in a single state and found that patients with psychiatric diagnoses captured before or during hospitalization or within 30 days after discharge had similar rates of coronary revascularization. However, in an analysis of VA patients with AMI during 1994–1995, Petersen et al.20
found that patients identified by prior psychiatric admissions, secondary inpatient codes, or mental health provider visits had lower rates of coronary angiography and revascularization but had similar 30-day mortality and were equally likely to receive indicated medications. Nonetheless, the latter finding contrasts with the study by Druss et al.,9
who found that psychiatric patients identified by inpatient secondary diagnosis inpatient codes were less likely to receive indicated medications after AMI. A final analysis by Young and Foster8
of patients with acute coronary syndromes found that in-hospital mortality was 21% lower in patients with psychiatric secondary diagnosis codes in among patients 65 years and older but was higher for patients with schizophrenia and substance abuse diagnoses who were less than 65.
Whereas the inconsistent findings across prior studies may reflect differences in patient characteristics (e.g., age, type of health insurance), it is also likely that the inconsistencies may reflect the different methods used to identify psychiatric comorbidity. Consistent with our findings, rates of identifying psychiatric comorbidity varied widely in these earlier studies depending on the methods employed, varying from roughly 5% using inpatient codes7–9
to 40% using outpatient codes.19
Thus, our findings build on prior studies and represent the first analysis to directly compare the differences in rates of identification and in prognostic impact of psychiatric comorbidity in the same patient population.
Several limitations should be discussed. First, mental illnesses often go unrecognized21
and inpatient and outpatient diagnoses from claims data may underestimate the prevalence of such conditions. Second, the sensitivity and specificity of ICD-9-CM codes in administrative databases may vary across individual diagnosis.22
Nevertheless, numerous prior studies have used administrative data to measure quality23,24
including several VA studies.25–27
Moreover, in a comparison of statistical models based on administrative data and clinical data from medical records, Krakauer et al.28
found that administrative data are satisfactory for characterizing variations in hospital mortality rates, whereas Ash et al.29
concluded that prediction models based on claims data can be accurate. Additionally, our inclusion of a laboratory-based measure addresses some of the limitations of administrative data regarding unmeasured severity of illness.30,31
Third, it is important to acknowledge that methods we used to identify psychiatric comorbidity may identify constructs of psychiatric disease that vary with respect to disease severity or spectrum (e.g., acute, chronic, newly diagnosed illness). Such constructs may, in turn, have different associations with hospital mortality either directly or indirectly through their influence on health care delivery. Differences in findings across prior studies may, in fact, reflect the differences in disease constructs identified by these studies.
Fourth, our use of mental health visits may identify a heterogeneous group of patients and may not capture patients with ongoing illness. These concerns may also be true for diagnosis codes captured on encounter data.
Despite these limitations, this study has several implications for research and practice. First, the study highlights the limitations of using secondary psychiatric diagnoses from inpatient codes to identify patients. Second, the study highlights that assessing the care delivered to patients with psychiatric illnesses using claims data may require the triangulation of multiple methodological approaches, given that different approaches may identify different disease constructs. Lastly, the findings are reassuring that patients with psychiatric illnesses do not appear to be at increased risk of adverse hospital outcomes for 2 common causes of hospitalization within the VA system. However, given the robust resources and novel clinical programs for managing mental health problems32
and the VA’s integrated electronic medical record, our findings should be replicated in other settings.
In conclusion, as the recognition of psychiatric illnesses improves through better screening approaches and public awareness, the monitoring of outcomes in such populations will become increasingly important. This is particularly true for the increasing number of veterans and nonveterans being diagnosed with serious mental illnesses, such as PTSD. The current findings suggest that additional research in determining sensitive and specific methods of identifying such patients, using existing data sources, is warranted.