Frank Verloin Degruy III, MD, MSFM

For years we have heard arguments that splitting mental health from the biomedical sciences was harmful^1,2 and that repairing this split, at least in the primary care setting, would result in better and perhaps less expensive care.³ But contemporary primary medical care is structured so that even conscientious, sympathetic, psychologically minded primary care physicians (PCPs) have an extraordinarily difficult time expanding their scope of practice to include the full biopsychosocial range of their patients’ problems as part of normal business.

Many investigators have pursued this problem. They have studied the effects of patients’ disorders being diagnosed before the examination and providing the PCP with those diagnoses⁴; of providing simple diagnostic tools for mental health diagnoses⁵; of providing PCPs with diagnoses augmented with management suggestions or algorithms⁶; of introducing professionals into the primary care setting who can manage mental health disorders in conjunction with the PCP^7,8; and of removing these patients to special settings for individual or group treatment.⁹ The National Institute of Mental Health, the MacArthur Foundation, several pharmaceutical companies, and the Robert Wood Johnson Foundation are investing enormous amounts of money in research trying to figure out how to find and treat depression—the most common mental health disorder in primary care—as it occurs in this setting. We can anticipate similar efforts for other common mental health conditions.

What have we learned from all this? We know that PCPs generally agree that these are important health conditions that should be addressed, but the competing demands of a busy practice render simple physician education and physician-administered diagnostic instruments of transient and marginal benefit. Interventions that add resources to the practice, such as cognitive behavioral therapists, pharmacotherapists, psychiatric consultants, or group therapy sessions consistently show improved outcomes, but it is still unclear whether these benefits apply to the majority of affected primary care patients or selectively to only the subset of patients who meet research inclusion criteria. It would appear that the most effective interventions, involving the introduction of additional personnel into the system of care, do result in lower overall medical expenditures,^10,11 but these savings are less than the cost of the interventions themselves. There is no medical cost offset to justify the programs of integrated care that have been studied, and we are looking instead at cost-effectiveness equations.

Recently the usual care of mental disorders, particularly depression, in the primary care setting has come under closer scrutiny. Some patients who meet the criteria for major depression are treated, and some are not. Of those treated, some are treated adequately, and some are not. Some researchers are reporting that all these patients—untreated, inadequately treated, and fully treated—seem to experience similar clinical outcomes and incur similar medical costs.^12,13 This finding is hard to digest and suggests that we need to better understand what normally occurs between PCPs and their patients who have mental health disorders.

A Cost-Offset of Diagnoses?

In normal practices we encounter PCPs whose mental health care diagnostic behavior distributes along a continuum. Does the distribution of this behavior relate to a corresponding distribution in the cost of care? The answer, either way, sharpens subsequent research. It helps us know what kind of benefits we can expect from attention to mental conditions, and it gives us hints about where we might look for the underlying reasons. In this issue of the Journal Campbell and colleagues¹⁴ report the results of a large study designed to answer one aspect of this question. They report that PCPs who most frequently make mental health diagnoses are caring for the patients who incur the lowest medical expenditures. This initially suggests that in the usual care setting, there is a cost-offset effect associatedwith taking a biopsychosocial approach to primary care and that we are justified in encouraging primary care clinicians to make mental health diagnoses. But the situation is far more complicated than that, and the chain of inferences that lead from the study findings to this conclusion is long and tenuous. Alternative explanations must be entertained. We should look at this study a little more closely.

The patient and physician sample in the study by Campbell and coworkers is large and representative, and the utilization data is most likely complete and accurate. We have no reason to believe that these physicians practice a drastically different form of primary care than PCPs elsewhere in the country. Thus we can accept with reasonable confidence that the findings accurately represent primary care practice in Rochester, New York, and probably elsewhere in the United States. This requires corroboration, but provisional acceptance is justified.

Why These Disorders Are Not Diagnosed

As the authors point out, however, even the physicians in the highest quartile diagnose mental disorders in only 9% of their patients, which means that probably only one third of the patients meeting the criteria are identified by encounter diagnoses. Why are so many mental disorders not diagnosed? Is it just the particular diagnostic pattern we see here that is associated with low-cost behavior, or should we infer that diagnostic rates are generally associated with lower costs of care? There is no telling what would happen if a higher proportion of disorders were diagnosed or what would happen to expenditures in the low-diagnosis quartile if their diagnostic rate were somehow raised to 9%. We do not know if patients with those diagnoses are different from those who do not have such diagnosis, and if so, in what ways. I also question the physicians themselves. I am reminded of a practice that, compared with 2 similar practices, made fewer mental health diagnoses and scheduled far more diagnostic tests, medical consultations, and referrals. One observer said of this practice, “It’s easier to order tests than listen to the patient.” Thus, 2 threads of inquiry need to be followed up: (1) These patients may be unique in ways that affect the cost of their care—ways that may not apply to the larger pool of undiagnosed patients, and (2) the relationship between the diagnosis and expenditures may be confounded by a set of intervening physician-level behaviors that this study was not designed to capture. I agree with the authors’ conclusion that a fruitful next step would be to link diagnoses with expenditures at the patient level and to observe physician behavior for the presence of confounds.

We also do not know if the patients in the 4 groups are alike. The authors have undertaken a case-mix adjustment under the assumption that they are different. I accept their rationale for doing so, but this introduces a number of disturbing new factors into the interpretation of the results. The ambulatory diagnostic group methodology employed here contains mental health diagnoses, which means that the variable of interest is analyzed after its effects have been adjusted out. The logic for leaving the mental health diagnoses in the case-mix adjustment (apart from the reluctance to tamper with a validated instrument) are that the differences in mental health diagnoses are at least partly due to physician differences beyond any differences in actual rates of diagnosis, and adjusting will reveal, rather than obscure, those differences. This is a problem that should be pursued on the next iteration, with a design that does not require such an adjustment. Until our current ambulatory case-mix methodology has been sufficiently refined, such adjustments will introduce confusion into our research. It is critical that we clarify whether these physicians are practicing on panels of patients that are different or whether they are practicing differently on similar panels of patients.

The effect described in the study (lower expenditures with higher mental health diagnosis rates) would be most persuasive if it followed a linear pattern, but at least with respect to total costs, this effect is curvilinear: Expected costs fall from the lowest through the third quartile but then jump to a high in the fourth quartile. The actual costs were high in the first quartile, then fell and remained relatively flat through the other 3 quartiles. The difference after adjustment is significant. However, the pattern through the 4 quartiles does not describe a dose-response effect, and the pattern is not consistent across the outcomes examined in these analyses. The outcomes do not converge convincingly on a single pattern. This in no way invalidates the significance of the findings, but it does suggest that further refinements in design and analysis are desirable.

Thoughts for the Future

The study by Campbell and colleagues is extremely provocative and interesting and raises as many questions as it answers. It appears that physicians likely to place mental health diagnoses on an encounter form do so among patients who have lower health care costs. It is now up to us to learn whether this is actually true, and if so, why. Is it because these patients thereby become less expensive to care for? These patients are incidentally less expensive to care for? Physicians who record such diagnoses practice relatively less expensive medicine? These physicians attract and retain patients who are less expensive to care for? Or the case-mix adjustment has introduced an artifact? This is an interesting and important research agenda, and we are indebted to these researchers for putting it in such bold relief. If this finding is replicated, we will need to discover ways to magnify the result by supporting the particular behavior responsible for it, and to test whether it has the same effects in those patients who are affected but have not been given a mental diagnosis. Then we will be closer to our goal of practicing coherent integrated primary care in which mental and medical health are approached as 2 facets of the same stone.

For years we have heard arguments that splitting mental health from the biomedical sciences was harmful^1,2 and that repairing this split, at least in the primary care setting, would result in better and perhaps less expensive care.³ But contemporary primary medical care is structured so that even conscientious, sympathetic, psychologically minded primary care physicians (PCPs) have an extraordinarily difficult time expanding their scope of practice to include the full biopsychosocial range of their patients’ problems as part of normal business.

Many investigators have pursued this problem. They have studied the effects of patients’ disorders being diagnosed before the examination and providing the PCP with those diagnoses⁴; of providing simple diagnostic tools for mental health diagnoses⁵; of providing PCPs with diagnoses augmented with management suggestions or algorithms⁶; of introducing professionals into the primary care setting who can manage mental health disorders in conjunction with the PCP^7,8; and of removing these patients to special settings for individual or group treatment.⁹ The National Institute of Mental Health, the MacArthur Foundation, several pharmaceutical companies, and the Robert Wood Johnson Foundation are investing enormous amounts of money in research trying to figure out how to find and treat depression—the most common mental health disorder in primary care—as it occurs in this setting. We can anticipate similar efforts for other common mental health conditions.

What have we learned from all this? We know that PCPs generally agree that these are important health conditions that should be addressed, but the competing demands of a busy practice render simple physician education and physician-administered diagnostic instruments of transient and marginal benefit. Interventions that add resources to the practice, such as cognitive behavioral therapists, pharmacotherapists, psychiatric consultants, or group therapy sessions consistently show improved outcomes, but it is still unclear whether these benefits apply to the majority of affected primary care patients or selectively to only the subset of patients who meet research inclusion criteria. It would appear that the most effective interventions, involving the introduction of additional personnel into the system of care, do result in lower overall medical expenditures,^10,11 but these savings are less than the cost of the interventions themselves. There is no medical cost offset to justify the programs of integrated care that have been studied, and we are looking instead at cost-effectiveness equations.

Recently the usual care of mental disorders, particularly depression, in the primary care setting has come under closer scrutiny. Some patients who meet the criteria for major depression are treated, and some are not. Of those treated, some are treated adequately, and some are not. Some researchers are reporting that all these patients—untreated, inadequately treated, and fully treated—seem to experience similar clinical outcomes and incur similar medical costs.^12,13 This finding is hard to digest and suggests that we need to better understand what normally occurs between PCPs and their patients who have mental health disorders.

A Cost-Offset of Diagnoses?

In normal practices we encounter PCPs whose mental health care diagnostic behavior distributes along a continuum. Does the distribution of this behavior relate to a corresponding distribution in the cost of care? The answer, either way, sharpens subsequent research. It helps us know what kind of benefits we can expect from attention to mental conditions, and it gives us hints about where we might look for the underlying reasons. In this issue of the Journal Campbell and colleagues¹⁴ report the results of a large study designed to answer one aspect of this question. They report that PCPs who most frequently make mental health diagnoses are caring for the patients who incur the lowest medical expenditures. This initially suggests that in the usual care setting, there is a cost-offset effect associatedwith taking a biopsychosocial approach to primary care and that we are justified in encouraging primary care clinicians to make mental health diagnoses. But the situation is far more complicated than that, and the chain of inferences that lead from the study findings to this conclusion is long and tenuous. Alternative explanations must be entertained. We should look at this study a little more closely.

The patient and physician sample in the study by Campbell and coworkers is large and representative, and the utilization data is most likely complete and accurate. We have no reason to believe that these physicians practice a drastically different form of primary care than PCPs elsewhere in the country. Thus we can accept with reasonable confidence that the findings accurately represent primary care practice in Rochester, New York, and probably elsewhere in the United States. This requires corroboration, but provisional acceptance is justified.

Why These Disorders Are Not Diagnosed

As the authors point out, however, even the physicians in the highest quartile diagnose mental disorders in only 9% of their patients, which means that probably only one third of the patients meeting the criteria are identified by encounter diagnoses. Why are so many mental disorders not diagnosed? Is it just the particular diagnostic pattern we see here that is associated with low-cost behavior, or should we infer that diagnostic rates are generally associated with lower costs of care? There is no telling what would happen if a higher proportion of disorders were diagnosed or what would happen to expenditures in the low-diagnosis quartile if their diagnostic rate were somehow raised to 9%. We do not know if patients with those diagnoses are different from those who do not have such diagnosis, and if so, in what ways. I also question the physicians themselves. I am reminded of a practice that, compared with 2 similar practices, made fewer mental health diagnoses and scheduled far more diagnostic tests, medical consultations, and referrals. One observer said of this practice, “It’s easier to order tests than listen to the patient.” Thus, 2 threads of inquiry need to be followed up: (1) These patients may be unique in ways that affect the cost of their care—ways that may not apply to the larger pool of undiagnosed patients, and (2) the relationship between the diagnosis and expenditures may be confounded by a set of intervening physician-level behaviors that this study was not designed to capture. I agree with the authors’ conclusion that a fruitful next step would be to link diagnoses with expenditures at the patient level and to observe physician behavior for the presence of confounds.

We also do not know if the patients in the 4 groups are alike. The authors have undertaken a case-mix adjustment under the assumption that they are different. I accept their rationale for doing so, but this introduces a number of disturbing new factors into the interpretation of the results. The ambulatory diagnostic group methodology employed here contains mental health diagnoses, which means that the variable of interest is analyzed after its effects have been adjusted out. The logic for leaving the mental health diagnoses in the case-mix adjustment (apart from the reluctance to tamper with a validated instrument) are that the differences in mental health diagnoses are at least partly due to physician differences beyond any differences in actual rates of diagnosis, and adjusting will reveal, rather than obscure, those differences. This is a problem that should be pursued on the next iteration, with a design that does not require such an adjustment. Until our current ambulatory case-mix methodology has been sufficiently refined, such adjustments will introduce confusion into our research. It is critical that we clarify whether these physicians are practicing on panels of patients that are different or whether they are practicing differently on similar panels of patients.

The effect described in the study (lower expenditures with higher mental health diagnosis rates) would be most persuasive if it followed a linear pattern, but at least with respect to total costs, this effect is curvilinear: Expected costs fall from the lowest through the third quartile but then jump to a high in the fourth quartile. The actual costs were high in the first quartile, then fell and remained relatively flat through the other 3 quartiles. The difference after adjustment is significant. However, the pattern through the 4 quartiles does not describe a dose-response effect, and the pattern is not consistent across the outcomes examined in these analyses. The outcomes do not converge convincingly on a single pattern. This in no way invalidates the significance of the findings, but it does suggest that further refinements in design and analysis are desirable.

Thoughts for the Future

The study by Campbell and colleagues is extremely provocative and interesting and raises as many questions as it answers. It appears that physicians likely to place mental health diagnoses on an encounter form do so among patients who have lower health care costs. It is now up to us to learn whether this is actually true, and if so, why. Is it because these patients thereby become less expensive to care for? These patients are incidentally less expensive to care for? Physicians who record such diagnoses practice relatively less expensive medicine? These physicians attract and retain patients who are less expensive to care for? Or the case-mix adjustment has introduced an artifact? This is an interesting and important research agenda, and we are indebted to these researchers for putting it in such bold relief. If this finding is replicated, we will need to discover ways to magnify the result by supporting the particular behavior responsible for it, and to test whether it has the same effects in those patients who are affected but have not been given a mental diagnosis. Then we will be closer to our goal of practicing coherent integrated primary care in which mental and medical health are approached as 2 facets of the same stone.

For years we have heard arguments that splitting mental health from the biomedical sciences was harmful^1,2 and that repairing this split, at least in the primary care setting, would result in better and perhaps less expensive care.³ But contemporary primary medical care is structured so that even conscientious, sympathetic, psychologically minded primary care physicians (PCPs) have an extraordinarily difficult time expanding their scope of practice to include the full biopsychosocial range of their patients’ problems as part of normal business.

Many investigators have pursued this problem. They have studied the effects of patients’ disorders being diagnosed before the examination and providing the PCP with those diagnoses⁴; of providing simple diagnostic tools for mental health diagnoses⁵; of providing PCPs with diagnoses augmented with management suggestions or algorithms⁶; of introducing professionals into the primary care setting who can manage mental health disorders in conjunction with the PCP^7,8; and of removing these patients to special settings for individual or group treatment.⁹ The National Institute of Mental Health, the MacArthur Foundation, several pharmaceutical companies, and the Robert Wood Johnson Foundation are investing enormous amounts of money in research trying to figure out how to find and treat depression—the most common mental health disorder in primary care—as it occurs in this setting. We can anticipate similar efforts for other common mental health conditions.

What have we learned from all this? We know that PCPs generally agree that these are important health conditions that should be addressed, but the competing demands of a busy practice render simple physician education and physician-administered diagnostic instruments of transient and marginal benefit. Interventions that add resources to the practice, such as cognitive behavioral therapists, pharmacotherapists, psychiatric consultants, or group therapy sessions consistently show improved outcomes, but it is still unclear whether these benefits apply to the majority of affected primary care patients or selectively to only the subset of patients who meet research inclusion criteria. It would appear that the most effective interventions, involving the introduction of additional personnel into the system of care, do result in lower overall medical expenditures,^10,11 but these savings are less than the cost of the interventions themselves. There is no medical cost offset to justify the programs of integrated care that have been studied, and we are looking instead at cost-effectiveness equations.

Recently the usual care of mental disorders, particularly depression, in the primary care setting has come under closer scrutiny. Some patients who meet the criteria for major depression are treated, and some are not. Of those treated, some are treated adequately, and some are not. Some researchers are reporting that all these patients—untreated, inadequately treated, and fully treated—seem to experience similar clinical outcomes and incur similar medical costs.^12,13 This finding is hard to digest and suggests that we need to better understand what normally occurs between PCPs and their patients who have mental health disorders.

A Cost-Offset of Diagnoses?

In normal practices we encounter PCPs whose mental health care diagnostic behavior distributes along a continuum. Does the distribution of this behavior relate to a corresponding distribution in the cost of care? The answer, either way, sharpens subsequent research. It helps us know what kind of benefits we can expect from attention to mental conditions, and it gives us hints about where we might look for the underlying reasons. In this issue of the Journal Campbell and colleagues¹⁴ report the results of a large study designed to answer one aspect of this question. They report that PCPs who most frequently make mental health diagnoses are caring for the patients who incur the lowest medical expenditures. This initially suggests that in the usual care setting, there is a cost-offset effect associatedwith taking a biopsychosocial approach to primary care and that we are justified in encouraging primary care clinicians to make mental health diagnoses. But the situation is far more complicated than that, and the chain of inferences that lead from the study findings to this conclusion is long and tenuous. Alternative explanations must be entertained. We should look at this study a little more closely.

The patient and physician sample in the study by Campbell and coworkers is large and representative, and the utilization data is most likely complete and accurate. We have no reason to believe that these physicians practice a drastically different form of primary care than PCPs elsewhere in the country. Thus we can accept with reasonable confidence that the findings accurately represent primary care practice in Rochester, New York, and probably elsewhere in the United States. This requires corroboration, but provisional acceptance is justified.

Why These Disorders Are Not Diagnosed

As the authors point out, however, even the physicians in the highest quartile diagnose mental disorders in only 9% of their patients, which means that probably only one third of the patients meeting the criteria are identified by encounter diagnoses. Why are so many mental disorders not diagnosed? Is it just the particular diagnostic pattern we see here that is associated with low-cost behavior, or should we infer that diagnostic rates are generally associated with lower costs of care? There is no telling what would happen if a higher proportion of disorders were diagnosed or what would happen to expenditures in the low-diagnosis quartile if their diagnostic rate were somehow raised to 9%. We do not know if patients with those diagnoses are different from those who do not have such diagnosis, and if so, in what ways. I also question the physicians themselves. I am reminded of a practice that, compared with 2 similar practices, made fewer mental health diagnoses and scheduled far more diagnostic tests, medical consultations, and referrals. One observer said of this practice, “It’s easier to order tests than listen to the patient.” Thus, 2 threads of inquiry need to be followed up: (1) These patients may be unique in ways that affect the cost of their care—ways that may not apply to the larger pool of undiagnosed patients, and (2) the relationship between the diagnosis and expenditures may be confounded by a set of intervening physician-level behaviors that this study was not designed to capture. I agree with the authors’ conclusion that a fruitful next step would be to link diagnoses with expenditures at the patient level and to observe physician behavior for the presence of confounds.

We also do not know if the patients in the 4 groups are alike. The authors have undertaken a case-mix adjustment under the assumption that they are different. I accept their rationale for doing so, but this introduces a number of disturbing new factors into the interpretation of the results. The ambulatory diagnostic group methodology employed here contains mental health diagnoses, which means that the variable of interest is analyzed after its effects have been adjusted out. The logic for leaving the mental health diagnoses in the case-mix adjustment (apart from the reluctance to tamper with a validated instrument) are that the differences in mental health diagnoses are at least partly due to physician differences beyond any differences in actual rates of diagnosis, and adjusting will reveal, rather than obscure, those differences. This is a problem that should be pursued on the next iteration, with a design that does not require such an adjustment. Until our current ambulatory case-mix methodology has been sufficiently refined, such adjustments will introduce confusion into our research. It is critical that we clarify whether these physicians are practicing on panels of patients that are different or whether they are practicing differently on similar panels of patients.

The effect described in the study (lower expenditures with higher mental health diagnosis rates) would be most persuasive if it followed a linear pattern, but at least with respect to total costs, this effect is curvilinear: Expected costs fall from the lowest through the third quartile but then jump to a high in the fourth quartile. The actual costs were high in the first quartile, then fell and remained relatively flat through the other 3 quartiles. The difference after adjustment is significant. However, the pattern through the 4 quartiles does not describe a dose-response effect, and the pattern is not consistent across the outcomes examined in these analyses. The outcomes do not converge convincingly on a single pattern. This in no way invalidates the significance of the findings, but it does suggest that further refinements in design and analysis are desirable.

Thoughts for the Future

The study by Campbell and colleagues is extremely provocative and interesting and raises as many questions as it answers. It appears that physicians likely to place mental health diagnoses on an encounter form do so among patients who have lower health care costs. It is now up to us to learn whether this is actually true, and if so, why. Is it because these patients thereby become less expensive to care for? These patients are incidentally less expensive to care for? Physicians who record such diagnoses practice relatively less expensive medicine? These physicians attract and retain patients who are less expensive to care for? Or the case-mix adjustment has introduced an artifact? This is an interesting and important research agenda, and we are indebted to these researchers for putting it in such bold relief. If this finding is replicated, we will need to discover ways to magnify the result by supporting the particular behavior responsible for it, and to test whether it has the same effects in those patients who are affected but have not been given a mental diagnosis. Then we will be closer to our goal of practicing coherent integrated primary care in which mental and medical health are approached as 2 facets of the same stone.