Avoidable Hospitalizations
The ability to measure appropriateness of care and outcomes in claims data is limited. One risk of a focus on mental health disorders might be inadequate attention to biomedical problems. In turn, a consequence of this inattention might be less timely or effective care that increases the risk of avoidable hospitalization for biomedical conditions.17 We used the approach of Weisman and Epstein18 to classify whether hospitalizations were avoidable. On the basis of previous research,18-20 6 medical conditions met the criteria for avoidable hospitalizations: angina, congestive heart failure, hypertension, asthma, chronic obstructive pulmonary disease, and diabetes mellitus. Patients were classified according to whether they were admitted during the year for an avoidable hospitalization condition.
Analyses
We explored the relationships between physicians’ coding of mental health diagnoses and both their observed expenditures and the risk of their patients being admitted for an avoidable hospitalization condition using regression analyses. To avoid confounding by differences in case mix, we developed measures of expected expenditures and predicted risk of avoidable hospitalizations according to case mix. The primary independent variable of interest—the physicians’ coding of mental health diagnoses—was defined as the proportion of all diagnoses coded by the physician during the year within the mental health category of the International Classification of Diseases, 9th edition. Because the proportion of mental health diagnoses variable exhibited marked skewing and because we wanted to provide summary measures of the effect of recording more mental health diagnoses, this variable was categorized into quartiles. The derivation of the measures of observed expenditures per panel member, expected expenditures per panel member (case-mix adjustment), avoidable hospitalizations, and predicted risk of avoidable hospitalization (case-mix adjustment) are described in detail in the Appendix.* For methodologic reasons described in the Appendix, we conducted the expenditure analyses at the physician level and the avoidable hospitalization analyses at the patient level.
We used physician-level ordinary least squares regression analyses to examine the relationship between the proportion of mental health diagnoses recorded by the physician and observed log expenditures per panel member. This analysis adjusted for case-mix–predicted expenditures. We used patient-level logistic regression analyses to examine the relationship between the proportion of mental health diagnoses recorded by the patient’s physician and the dependent dichotomous variable. This analysis for the patient’s case mix predicted risk of avoidable hospitalization and the nesting of patients with a physician. A detailed description of these analyses is also included in the Appendix.
Results
The characteristics of physicians by their quartile of proportion of mental health diagnoses and their per panel member expenditures are shown in [Table 1]. In general, there were few statistically significant differences among the groups. Physicians recording a greater proportion of mental health diagnoses had greater expected expenditures per panel member (r = .14; P = .0017) and higher proportions of panel members who used health care services (r = .11; P = .017). Of the 243,150 patients in the database in 1995, 859 (0.35%) had at least one avoidable hospitalization.
After adjustment for expected per panel member expenditures, physicians in the higher quartiles of proportion of mental health diagnoses recorded had lower expenditures than physicians in the lowest quartile; expenditures were 9% lower for physicians in the highest quartile [Table 2]. These results were little changed after further adjustment for physician specialty, age, and sex (results not shown). The cost reductions were observed for total inpatient services, total outpatient services, and diagnostic services, but not for ambulatory visit services [Table 3].
After adjustment for predicted risk of avoidable hospitalization (case-mix adjustment), there was a trend (P = .051) for patients whose physicians were in the highest quartile of recorded mental health diagnoses to be less likely to be admitted for an avoidable condition. Patients of physicians in the highest quartile were at lower risk for an avoidable hospitalization than patients of physicians in the lowest quartile (adjusted odds ratio = 0.73; 95% confidence interval, 0.54 - 1.00). These results were little changed after further adjustment for physician specialty, age, and sex (results not shown).
Discussion
Our study demonstrates that primary care physicians who record higher proportions of mental health diagnoses generate lower overall health care costs. The lower costs can be attributed to reduced inpatient diagnostic testing and total outpatient costs, but not to physician visit costs. In addition, the patients of physicians with the highest proportion of recorded mental health diagnoses may be less likely to be admitted for an avoidable hospitalization.
In our study the largest reduction in health care expenditures occurred with inpatient services, where there was a 20% difference between physicians in the highest and lowest quartiles of recorded mental health diagnoses. This result is consistent with previous research showing reduced total and inpatient costs after treatment of mental health disorders.21 Katon and colleagues16 found that patients with depression who received recommended levels of antidepressant treatment had dramatically lower non–mental health-related inpatient costs. These findings suggest that the detection and treatment of mental health disorders may have its main impact on health costs by reducing hospital admissions or lengths of stay.