However, the data that support the veracity and magnitude of a cost-offset effect for the treatment of mental illness are equivocal [8, 19-22]. This ambiguity may arise from 1) the conceptual problems associated with approaching psychiatric illness as an entity separate from medical illness and 2) the assumption that the excess use of nonpsychiatric services is related to symptoms that have no organic basis. The alternate explanation is that patients with depression use more health services largely because they have more comorbid nonpsychiatric illness. Thus, even when somatic symptoms in these patients do not have an organic basis, diagnostic testing may be appropriate to rule out exacerbations of existing medical conditions. In addition, these medical conditions may complicate the recognition and treatment of depression, depression may complicate the treatment of medical conditions, symptoms of depression may arise from distress about the medical conditions, or any combination of these three situations may occur. Therefore, any attempt to address excess cost among patients with symptoms of depression must consider comorbid nonpsychiatric medical illness.

Older adults are particularly likely to have comorbid illness, to be functionally impaired, or to receive medication that can lead to symptoms of depression or shroud the presentation of depression, thereby complicating evaluation and treatment decisions. They are also more likely to present with psychiatric distress manifested as somatic symptoms, and primary care physicians are specifically trained to rule out organic illness as a cause of depression [16, 23-25]. Previous studies describing the excess use of health services among patients with symptoms of depression did not have access to detailed information on the content of diagnostic testing or more precise measures of the comorbid conditions that are likely to prompt more diagnostic testing. Our previous studies of depression among older patients in the primary care setting [26, 27] suggested that older adults with symptoms of depression have more medical conditions and greater functional impairment than do patients without symptoms of depression. Thus, we hypothesized that a greater burden of comorbid nonpsychiatric medical illness among older adults with symptoms of depression would largely explain the excess use of diagnostic testing in these patients.

Methods

Top Methods Results Discussion Author & Article Info References

Our study was done in a primary care group practice at an academic institution. In 1991, this practice delivered primary care to more than 12 000 patients, resulting in approximately 40 000 outpatient visits per year [28]. Beginning in January 1991, all patients 60 years of age and older were screened for depression, dementia, and alcoholism during their regularly scheduled visits. Detailed methods and results of the screening program for these three conditions have been reported elsewhere [26, 29, 30]. The screening instruments included the CAGE alcoholism screening questionnaire [31], the Short Portable Mental Status Questionnaire [32], and the Center for Epidemiologic Studies Depression Scale (CES-D) [33]. Professional research assistants administered the screening interview to patients during routine office visits. The CES-D is among the most widely used and evaluated instruments for identifying depression [34]. Its sensitivity (approximately 80%) and specificity (approximately 72%) are similar to those of other commonly used instruments [34]. We used the suggested cutoff score of 16 on the CES-D to indicate significant symptoms of depression [33]. Data that were obtained during the screening interview were merged with information that was routinely collected and stored in electronic medical records for the 2-year period after the screening interview [35].

One limitation of previous studies that have assessed the effect of depression on the use of health services has been reliance on a simple count of comorbid diagnoses as a measure of illness burden. We sought a mechanism that could summarize illness burden, particularly in relation to expected use of health services resources. The Ambulatory Care Group system [36, 37] provides a measure of health status and expected consumption of resources that is based on all diagnoses assigned to a patient by all providers during the course of 1 year. This system has been shown to predict concurrent and subsequent use of ambulatory care resources in several settings. An overview of this system, as described by its developers [37], follows.

"The crux of the Ambulatory Care Group case-mix approach is a scheme for clustering ambulatory diagnostic codes. Each of approximately 5000 common ICD-9-CM [International Classification of Diseases, version 9, clinical modification] codes has been placed into one of 34 clusters, based on its expected relationship to health care resource use. These clusters are termed ambulatory diagnostic groups (ADGs). After all of a patient's ICD-9-CM codes are assigned an appropriate ADG, a person is categorized into one of 51 distinct Ambulatory Care Groups based on their age, sex, and constellation of ADGs. The overall theoretical goal of the ICD to ADG assignment was to cluster together similar conditions based on their expected impact on health services resource consumption."

We obtained ICD-9-CM codes [38] for all outpatient encounters for each patient-provider interaction during the year after the initial screening date. These codes are routinely stored in the electronic medical record for billing and are used to maintain lists for patients' problem-oriented medical records. The diagnoses are assigned by the patients' physicians and include specific diagnoses, poorly defined conditions, and patient-reported symptoms. Most of the poorly defined conditions and patient-reported symptoms have specific ICD-9-CM codes and would be included in the ambulatory diagnostic groups. However, we specifically excluded ICD-9-CM codes for affective disorders to avoid including them as comorbid conditions.

Our study was limited to patients 60 years of age and older; therefore, 17 of the 51 Ambulatory Care Groups are irrelevant because they apply only to patients younger than 60 years of age. Because of the illness burden of our patient population in general, the four Ambulatory Care Groups that are associated with the greatest future use of ambulatory resources accounted for 84% of our patient population. (In previous studies, these four Ambulatory Care Groups accounted for less than 10% of the sample population [37]). The remaining 16% of our patients were distributed among 30 other Ambulatory Care Groups. Thus, for the purpose of analysis, we constructed a dummy variable for each of the four Ambulatory Care Groups that represented most of our patient population and a fifth dummy variable to account for the other Ambulatory Care Groups. This fifth group was used as the reference group in the regression models.

The Regenstrief Medical Record System [35] contains data on frequency and charges for more than 850 diagnostic tests that are used in medical practice. It also includes diagnostic procedures, such as sigmoidoscopy, that are done by health care providers. These 850 or more tests and procedures can be grouped operationally into 22 revenue centers, such as radiology, general chemistry, hematology, pathology, or pulmonary physiology laboratory. For our purposes, we further combined these revenue centers into three groups: special procedures (such as exercise testing and colonoscopy), diagnostic imaging (such as radiography and ultrasonography), and clinical pathology (such as serum chemistry and microbiologic examination). Because diagnostic testing would typically be done in conjunction with a physician encounter, we describe the number of outpatient or emergency department visits and the number of hospitalizations and lengths of stay during each of the 2 years after the screening date.

To measure significant differences between patients with and without symptoms of depression, we used chi-square tests for categorical variables and t-tests for continuous variables. We compared charges between groups by using the nonparametric Wilcoxon rank-sum test. However, we repeated these analyses using parametric tests on the log-transformed charge data and found no substantive difference in the results. To explore the independent effects of symptoms of depression on the subsequent accrual of diagnostic test charges, we did a multiple linear regression on the log-transformed total diagnostic test charge using demographic variables, CES-D score, and Ambulatory Care Group assignment as independent variables.

The agencies that funded this project had no direct role in the gathering, analysis, or interpretation of our data and did not participate in the decision to publish the study.

Results

Top Methods Results Discussion Author & Article Info References

From January 1991 to May 1993, we contacted 4413 patients 60 years of age and older during their routine visits to the primary care practice; 3767 (86%) of these patients completed the screening examination. One hundred fifteen patients refused the screening examination, 284 were ineligible, and 247 could not complete the CES-D test. The mean age of the patients who were screened was 68 years (range, 60 to 102 years); 68.8% of the patients were women, 63.4% were black, 43.6% had 8 years of education or less, 15.7% had cognitive dysfunction (defined as 3 errors on the Short Portable Mental Status Questionnaire), 10.7% had evidence of alcohol dependency (defined as 2 positive responses on the CAGE test), and 16.2% had significant symptoms of depression (defined as a CES-D score 16). Figure 1 shows the distribution of CES-D scores for this cohort.

View larger version (12K): [in this window] [in a new window]   Figure 1. Results of testing on the Centers for Epidemiologic Studies Depression Scale (CES-D) for 3767 older adults who completed a screening examination.

Table 1 compares the clinical characteristics of patients who had significant symptoms of depression with those of patients who did not. Patients with significant symptoms of depression were younger and more likely to be white and female. Despite being younger, patients with symptoms of depression appeared to have a greater burden of nonpsychiatric illness as shown by the Ambulatory Care Group assignments. Most of our patients were classified in the four Ambulatory Care Groups with the highest expected consumption of resources. However, patients with symptoms of depression were more likely to have the Ambulatory Care Group assignments that were associated with the greatest subsequent use of ambulatory care resources (groups 49 and 50).

Total charges for diagnostic testing for 1 year in this cohort of 3767 older adults were $4 700 719. These charges were accrued in 30 372 visits to an ambulatory care center, 3270 emergency department visits, and 1063 hospitalizations. The mean charge for diagnostic testing for 1 year was $1248 (median, $420); 10.5% of the study population had no diagnostic test charges during the year after the screening date. Eliminating the cost outliers did not dramatically change mean and median charges for this group: The mean charge was $1099 (median, $410) when the 1% of patients with the greatest diagnostic test charges were not included.

Total charges for diagnostic testing in the 2 years after the screening date differed significantly between the five Ambulatory Care Group categories listed in Table 1. Mean charges for each Ambulatory Care Group category in the year after screening were $700 (median, $282) for group 41, $1066 (median, $374) for group 44, $1691 (median, $620) for group 49, $2016 (median, $882) for group 50, and $670 (median, $206) for all other groups. In addition, patients who were hospitalized at least once during the year after the screening date had significantly greater diagnostic test charges (mean, $4222; median, $2961) than did those who were not hospitalized (mean, $600; median, $303) (P < 0.001). Patients who visited the emergency department at least once in the year after the screening date had significantly greater diagnostic test charges (mean, $2204; median, $1054) than did patients who did not visit the emergency department (mean, $556; median, $248) (P < 0.001).

Charges for clinical pathology (such as those for routine blood and urine tests) accounted for 54.2% of diagnostic test charges, diagnostic imaging accounted for 16%, and special procedures accounted for 29.7%. As Table 2 shows, patients with symptoms of depression had substantially greater use of health services during the subsequent year in terms of ambulatory care center visits, emergency department visits, and hospitalizations; they also had greater diagnostic testing charges in all three categories of tests. Patients with symptoms of depression had greater total diagnostic test charges in the year after the screening date than did patients without symptoms of depression; these differences persisted in the second year after the screening date. Differences in diagnostic charges between groups remained significant when patients with CES-D scores of 24 or higher (mean charges, $1567; median, $612) were compared with patients with CES-D scores of 16 to 23 (mean charges, $1413; median $535) and patients with CES-D scores less than 16 (mean charges, $1203; median, $387) (P < 0.001).

Although the CES-D summary score and the total log-transformed charge for diagnostic testing are significantly correlated, the CES-D summary score explains very little of the variance in total diagnostic test charges (Pearson correlation coefficient, 0.06; P < 0.001). Furthermore, the CES-D summary score remains significantly correlated with the log-transformed charge for diagnostic testing when demographic variables in multivariable regression are controlled for, but the effects of the CES-D summary score on diagnostic testing charges are no longer significant if the Ambulatory Care Group is added to the model. Thus, controlling for comorbid nonpsychiatric illness appears to eliminate the independent effects of CES-D score on diagnostic charges.

Discussion

Top Methods Results Discussion Author & Article Info References

Our objectives were 1) to provide a detailed prospective account of the volume and content of diagnostic testing consumed by older adult patients in the primary care setting who did and did not have significant symptoms of depression and 2) to determine whether a greater burden of nonpsychiatric medical illness contributes to the excess diagnostic test charges of patients with symptoms of depression. The mean diagnostic test charge per year for our cohort of primary care patients was $1248 (median, $420). Low-cost items, such as routine blood tests, accounted for most diagnostic test charges [39]; Ambulatory Care Group assignment, emergency department use, and hospitalization were highly correlated with diagnostic charges. Patients with baseline CES-D scores of 16 or more accrued approximately $200 in excess diagnostic test charges during the first year after screening compared with patients with lower CES-D scores; this difference persisted into the second year after screening. Although these excess charges are modest on an individual or yearly basis, the total could quickly reach millions of dollars if these charges are incurred by an entire population of older adults for a period of several years.

At least three lines of evidence in these data suggest that patients with symptoms of depression have a greater burden of comorbid nonpsychiatric illness. This issue is important because it addresses the question of whether symptoms of depression (especially if left untreated) cause excess diagnostic test charges or whether they are simply more common among patients with a greater burden of comorbid nonpsychiatric illness. First, patients with symptoms of depression were more likely to be categorized in the Ambulatory Care Groups associated with the greatest subsequent use of ambulatory care resources. Second, patients with symptoms of depression were more likely to be hospitalized and to use the emergency department during the year after the screening date. Third, in bivariate analyses, a patient's CES-D score as a continuous or categorical variable accounted for less than 0.5% of the variance in total diagnostic test charges, and this association was no longer significant when Ambulatory Care Group assignment was controlled for. Also, in data that were reported elsewhere [26], older adults with symptoms of depression had nearly twice the functional impairment on the physical disability subscale of the Sickness Impact Profile [40] compared with a random sample of older adults in the primary care setting who had low CES-D scores (23.2 compared with 12.8; P < 0.001). However, the presence of excess medical illness does not preclude the possibility that treatment of the symptoms of depression might reduce the overall costs of health care.

Much literature addresses the possibility that excess use of nonpsychiatric health services by patients with mental disorders may be amenable to improved treatment of mental illness. However, this literature is heterogeneous in terms of study design, patient population, study setting, intervention, and outcome measure. In most studies, the investigators could not randomly assign patients to treatment, control for comorbid illness, or examine the precise reasons for ambulatory visits or hospitalizations [8]. We found no previous studies that showed reductions in the subsequent use of ambulatory care resources by patients in the primary care setting who had late-life depression and were randomly assigned to psychiatric treatment. In one of the few randomized trials in primary care to examine the effects of psychiatric consultation on subsequent use of health care resources, Katon and colleagues [41] reported no significant differences between controls and patients receiving intervention in distress, disability, or use of health care resources among patients identified as distressed high-utilizers. They noted that

"the physician's ability to attend to psychiatric illness was complicated by coexisting problems in the management of physical disease. Additionally, any improvement in psychological status may be overshadowed by the multiplicity of distress from chronic physical disorders."

More than a decade ago, Mumford and colleagues [8] did a meta-analysis of 58 controlled studies that investigated the cost-offset effects of outpatient treatment of mental illness. Among randomized trials, these authors found that reductions in the use of health care services were more commonly associated with inpatient than with outpatient services, but reductions tended to be greater for older patients. The presence and magnitude of the cost-offset effect that was reported in these studies largely depended on the study methods mentioned above. Variation in study methods may also explain the mixed results of reducing health care costs through psychiatric interventions in more recent randomized trials [22, 41-46]. Notably, nearly all of the successful interventions in these studies involved bringing psychiatric services to the medical or surgical ward or primary care clinic rather than sending patients to a psychiatric referral setting.

Our study has several limitations. First, we identified patients on the basis of symptoms of depression rather than criteria-based syndromes of depression. Older patients who meet the criteria for major depression might generate a very different pattern of diagnostic test use. However, patients with subclinical, minor, or chronic depression account for most cases of affective disorders seen in primary care [47]. Also, symptoms of minor and subclinical depression are associated with excess illness and excess use of health services [12, 48-51]. Second, our sample was drawn from an urban primary care group practice at an academic institution; therefore, our data may not apply to other populations. Third, our measure of comorbid conditions could be confounded with the health service use patterns of patients with symptoms of depression. The Ambulatory Care Group system, however, is specifically designed to predict the subsequent use of ambulatory care resources; other measures of comorbid illness predict mortality or use of hospital services. Finally, we could not measure the diagnostic testing charges that were accumulated by patients in facilities of health systems other than our own.

Diagnostic testing is a costly and discretionary maneuver that is largely driven by the judgments of the physician but is frequently influenced by the perceived or overt demands of the patient. Although diagnostic testing is usually not considered a treatment per se, it is frequently therapeutic and some patients equate quality with the intensity and novelty of diagnostic testing. The vagaries of medical decision making for diagnostic testing provide a foundation for the hypothesis that the use of diagnostic testing might be inappropriately high among older adults with depression who present with ill-defined symptoms. However, our results suggest that the effects of psychiatric symptoms on diagnostic testing cannot easily be separated from the effects of comorbid medical illness. The literature contains abundant evidence showing that the patients who accrue more charges and have greater psychiatric distress are also the patients who have more extensive medical illness. Little evidence supports the existence of a large group of "worried well" among older adults [52, 53]. Primary care physicians and their psychiatric colleagues are not likely to be successful in reducing health care costs and improving patient outcomes for depression unless organic and psychiatric illnesses are addressed simultaneously. Carving out psychiatric services as a separate entity would appear to decrease the likelihood for success. Older adults generally prefer to receive psychiatric treatment in the context of primary care. However, primary care physicians need better treatment models for these complex patients [54], and older adults need better access to psychiatric services [21]. Future research on intervention that can decrease the morbidity and cost of late-life depression must improve the care of psychiatric and medical illnesses concurrently.

From the Regenstrief Institute for Health Care, the Richard L. Roudebush Veterans Affairs Medical Center, and Indiana University School of Medicine, Indianapolis, Indiana.