 |
|
Variations in the Management of Primary Care
Effect on Cost in an HMO Network
Eugene C. Rich, MD;
John Kralewski, PhD;
Roger Feldman, PhD;
Bryan Dowd, PhD;
Terence S. Bernhardt
Arch Intern Med. 1998;158:2363-2371.
ABSTRACT
| |
Objective To determine the relation to cost of different aspects of the management of primary care among group practices within a health maintenance organization network.
Measures A cross-sectional survey study of medical practices conducted with Blue Cross Blue Shield of Minnesota, St Paul. The subjects were group practices accepting financial and administrative responsibility for primary care services in the managed care plans of Blue Cross Blue Shield of Minnesota. One hundred twelve primary care practices and 153,397 enrollees were included in this analysis. The principal resource use measure in this study was nonhospital cost per member per year estimated from payments to providers plus subscriber-eligible liability.
Results The medical directors' responses revealed considerable variability in the management of primary care in these 112 practices. Group practice characteristics consistently associated with lower nonhospital cost were patient identification of a primary care physician, cost of care profiling, more frequent physician profiling, more patients per hour in the clinic, a higher proportion of primary care physicians in the specialty of family or general practice, and a greater number of physicians in the group practice.
Conclusions Results of this study demonstrate substantial variation in the management of primary care among group practices participating in a health maintenance organization network. These differences are associated with significant variation in the nonhospital cost of care for enrollees.
INTRODUCTION
THERE IS substantial evidence of regional and community variation in the cost and outcomes of medical care. Although much of this research initially focused on the use of procedures and hospitalizations,1-3 there is now evidence of important variation in the process and cost of care across primary care practices.4-5
Although variations in consumer expectations, community characteristics, and access to health resources might contribute to these differences,6-10 results of several studies4, 11-12 suggest that the characteristics of medical practices are a major factor. Indeed, there is increasing evidence that a variety of features of primary care practices may be important to performance in the managed care environment, including the characteristics of the physicians,13-15 their workload,16-17 and the degree to which care is managed by the primary care providers.18-20 Of course, a variety of administrative interventions have been developed to help manage utilization by primary care physicians.21-26 There are also several attributes of primary care service that may be related to utilization,12, 27 including coordination of care, comprehensiveness of care, and continuity of provider.28-30
The purpose of this study was to determine the relation of these different aspects of the management of primary care to costs among group practices within a health maintenance organization (HMO) network. We chose the upper Midwest as the focus of our research because managed care and medical group practice have become well established there. Thus, we have an opportunity to observe the variety of responses of medical group practices to an advanced managed care environment.
SUBJECTS AND METHODS
SUBJECTS
The subjects of this study are group practices accepting financial and administrative responsibility for primary care services for enrollees in the managed care plans of Blue Cross Blue Shield of Minnesota (BCBSM), St Paul. To be eligible for study, the group practices must have had at least 200 members continuously enrolled in 1995. One hundred twenty-nine primary care practices were identified that met these criteria. These practices were in Minnesota, Wisconsin, North Dakota, and South Dakota.
MEASUREMENTS
The characteristics of primary care management in these practices were obtained by a mailed survey instrument. Topics relevant to managing primary care were identified from our previous studies, from a literature review, and from focus groups. These focus groups comprised clinic and health plan medical directors and administrators. A survey of medical directors was developed to investigate the management of primary care in these practices. Many items were adapted from previous research.31 The care management techniques identified were demand management, referral management, hospital care management, physician profiling, and use of guidelines.
Other aspects of primary care assessed in this survey included mechanisms for comprehensiveness of primary care, continuity of care, information support in the primary care office, and primary care physician workload. After the survey was developed, it was validated by pilot testing with physicians familiar with group practice management; responses were confirmed by group discussions with investigators. We also conducted a survey of administrators of these same group practices, addressing such issues as ownership, governance, sources of revenue, and physician incentive plans. The findings of the research with the administrators survey are discussed elsewhere.32
In designing the survey, we emphasized that medical directors were to identify programs that their group practice had in place in contrast to programs conducted centrally by health plans. During the study, BCBSM also supported an array of centrally administrated health plan strategies for utilization management, including a preservice approval program, high-cost case management, development and dissemination of clinical practice guidelines (eg, for preventive services), profiling of group practice on performance relative to utilization targets, and dissemination of performance data to group practice administrators.
To identify the appropriate recipients for these surveys, telephone calls were placed to each clinic organization to obtain the names of the chief administrator and the medical director. Surveys were then mailed to medical directors. We also surveyed a subset of group practices that operate multiple clinic sites; a description of the characteristics of these sites is reported elsewhere.33 Surveys were conducted between July 15 and October 15, 1996. We conducted telephone follow-up and multiple mailings for nonrespondents. One hundred twelve (86.8%) of 129 organizations surveyed provided analyzable responses (see the "Results" section for details).
Resource utilization data were abstracted from the Blue Cross Blue Shield financial database for the 180169 members eligible for this study in 1995. Enrollees were included in this study if they were continuously enrolled through 1995, were younger than 65 years, and were not receiving Medicare or Medicaid. We used 2 measures of resource utilization in this study. Our first analyses focused on overall cost per member per year, estimated from the allowed amount to providers plus subscriber liability. We also conducted analyses excluding hospital inpatient costs because in the BCBSM network hospital contracts vary among institutions and because most group practices have little effect on either hospital rate setting or the facility most convenient to their patients. Results of the 2 analyses are similar. We report the cost analyses excluding hospital inpatient costs.
Enrollee characteristics included age, sex, ambulatory care group (ACG), and patient insurance characteristics; these were also obtained from administrative data at BCBSM. We also observed enrollee insurance characteristics. Relevant categories of insurance were derived from the literature and from discussions with BCBSM underwriting and contracting staff. A group of categorical variables was developed to describe the individual patient's insurance characteristics as follows: underwriting pool (individual small group, rated group, and self-insured group), insurance products (point of service with comprehensive major medical, standard point of service, state government, and traditional HMO), coinsurance (copayment requirement and deductible amount), and pharmacy benefit coverage. Details of these categories are provided in Table 1. The predominance of point of service managed care products in our HMO network is consistent with national trends away from traditional gatekeeper HMOs. A point of service HMO meets all of the regulatory requirements of a traditional HMO but enrollees have the option of obtaining care from nonparticipating providers. Usually, the HMO charges enrollees (either through copayments, deductible differentials, or both) for care that was not managed by the designated primary care practice or that used the less restrictive list of participating physicians. Within the BCBSM products, the designated primary care clinic has all the costs for point of service enrollees attributed to it; furthermore, the group has a financial incentive (eg, partial withhold of payments and reconciliation based on a utilization target) to encourage patients to seek care through their group practice.
|
|
|
|
Table 1. Characteristics of 153397 Enrollees*
|
|
|
The characteristics of the physicians in the practices were obtained from credentialing data at BCBSM.
ANALYTIC APPROACH
We conducted a series of multivariate linear regression analyses to determine the relation to cost of clinic characteristics relevant to the management of primary care. The dependent variable in these analyses was the natural logarithm of cost per member per year; therefore, only members using some services were included in these analyses (ie, 153397 of 180169 members). We used the natural logarithm of dollar cost instead of raw dollars as the dependent variable because results of previous research34 indicate that health plan enrollee costs are not normally distributed.
In these multivariate models, we used the individual enrollee as our unit of analysis. Although we observed only 112 practices, we observed many patients with unique costs from each practice. Thus, the degrees of freedom for the practice-level variables is greater than 112. Nevertheless, there may be some correlation among the patients for each practice, even after controlling for observable characteristics (eg, practice size and specialty mix). Therefore, we also ran cost regression analyses using patient characteristics and a dummy variable for each practice. We inspected the estimated mean error ( ) for this regression and for the regressions from our full model. By this means, we were able to consider adjusted t statistics for our full model, correcting for interpractice correlation.
In this study, we used a sequential approach to building a model of primary care factors associated with cost. This approach consisted of 3 steps. Step 1 was the estimation of a "baseline" model that included only patient characteristics (demographics, ACG, and insurance plan).
Demographics were controlled through 8 categories of enrollee age and sex: males aged 0 to 2 years, 3 to 14 years, 15 to 44 years, and 45 to 64 years; and females aged 0 to 2 years, 3 to 14 years, 15 to 44 years, and 45 to 64 years. Females aged 3 to 14 years were the referenced age–sex category in the multivariate analysis. We controlled for diagnosis and morbidity through ACGs.35 Ambulatory care groups are a case-mix measure for application in ambulatory populations based primarily on categorization of diagnoses according to their likelihood of persistence.35 In developing our model, we tested 3 different approaches to case-mix control with ACG. The first was the morbidity index score; this is a score based on assigning patients to 1 of 5 morbidity groups of increasing severity based on their ACG. The morbidity index score algorithms are included as part of the Managed Care Information System licensed by Codman Research Group (Andover, Mass), which uses ACGs as the basis for its severity adjustment technique. The other 2 approaches evaluated were the use of a separate dummy variable for each of 5 morbidity groups and the use of a separate dummy variable for 50 of 52 ACGs. We chose the latter approach because it had the best explanatory power in our model. Ambulatory care groups 51 (ungroupable) and 52 (no diagnosis codes available) were combined as the reference category, selected because they were low-cost groups associated with a reasonable volume of enrollees (N=6877) in our sample.
We also controlled for whether the patient was from an urban or a rural county. Urban counties were defined as counties within a metropolitan statistical area; rural counties were defined as those outside a metropolitan statistical area. Most enrollees were from Minnesota; therefore, we did not use a "state of residence" variable in the model.
Step 2 of the model-building process was to estimate a series of "partial models" that tested survey items within each of the individual clusters of care management strategies while controlling for the patient characteristics already included in the baseline model (demographics, ACG, and insurance plan). A separate partial model was estimated for each of the previously described clusters of primary care management strategies (eg, demand management, referral management, hospital care management, physician profiling, physician guidelines, general aspects of primary care, and primary care physician workload).
Step 3 consisted of assembling a "combined model" that analyzed the effects of characteristics of the physicians in the group while controlling for patient characteristics and for all of the statistically significant (at P<.01) primary care management strategies selected from the partial models. The physician characteristics analyzed were the total number of physicians in the group, the proportion of physicians who practice primary care (defined as general internal medicine, general pediatrics, family practice, or general practice), the proportion of primary care physicians who specialize in family or general practice (defined as the number of family physicians and general practitioners divided by the total number of primary care physicians), the average number of years of experience of the physicians in the practice, and the proportion of male physicians in the practice. The combined model accounted for 53.4% of the variance in nonhospital cost among our enrollees (adjusted r2 =0.53)
In all of these analyses, for each survey variable for which data were missing, we created a dummy variable to indicate a missing response. We then scored the item that was missing as a 0, the lowest point on the scale of responses for that item. For most items used in our final analysis, fewer than 5% of responses were missing on any particular variable.
RESULTS
Of 129 group practice medical directors surveyed for this study, 110 provided data on the overall organization. Two medical directors of multisite organizations responded for the largest clinic within their organization but did not provide summary responses including the smaller satellite clinics. We included these 2 responses in our analysis. Thus, we had data on 112 of 129 organizations (86.8% response rate). Of enrollees of BCBSM managed care products assigned to these 112 clinics, 153397 submitted claims for services during 1995. As shown in Table 1, most enrollees were females aged 15 to 44 years. Most enrollees resided in urban areas. The most common insurance product was point of service; almost 45% of enrollees had a copayment requirement. The association with cost of each of these demographic characteristics is also shown in Table 1.
Table 2 shows common demand management techniques. The most common demand management technique was patient literature on after-hours care; 64 groups reported using this technique (not shown on Table 2 because it was not associated with cost in the partial models). Identification of high-cost patients, identification of patients with high emergency department use, distribution of patient literature on telephone advice, and distribution of literature on seeking care from the practice were all associated with cost in the partial models. In the combined multivariate cost model, only the use of strategies to identify patients with long-term illness approached our criterion for statistical significance in association with lower nonhospital cost.
|
|
|
|
Table 2. Effect of Strategies to Manage Patient Demand
|
|
|
We also investigated another aspect of demand management: strategies to provide first-contact care for short-term problems. We investigated this with respect to self-limited problems occurring during regular clinic hours and for such problems that arise after regular clinic hours. Physician clinic care was reported by essentially all groups as the predominant means of providing first-contact care during regular clinic hours. Nurse telephone care was used by 70.5% of groups for care during business hours, and clinic care by nurses was reported as a strategy by 51.8% of groups. Regarding first-contact care after regular clinic hours, physician telephone care was a strategy used by 97% of groups. Emergency department referral was used by 88.5% of groups for management of short-term, self-limited problems after regular business hours, whereas nurse telephone care was reported by only 36.6% of groups. Although several of these strategies were significantly associated with nonhospital cost in the partial models, none of these associations was confirmed in the combined multivariate model.
Table 3 demonstrates the association with cost of strategies to manage referrals and hospitalizations. Review of referrals was the most commonly used technique in these groups. All 4 referral management techniques were significantly associated with cost in the partial models. However, when tested in the combined model, none of these techniques was independently associated with nonhospital cost per member per year. Regarding the hospital care management techniques used in these 112 primary care practices, hospital-employed inpatient case managers and inpatient care paths are used relatively frequently. Hospital employment of inpatient case managers, a higher proportion of inpatients managed by case managers, a higher proportion of inpatients managed in care paths, a greater degree of involvement of physicians in care paths, and greater primary care physician involvement in inpatient care were associated with cost in the partial models. In the combined multivariate cost model excluding inpatient facility costs, 1 hospital care management strategy was still associated with lower expenditures: a greater degree of involvement of clinic physicians in developing care paths (P=.005).
|
|
|
|
Table 3. Effect of Strategies to Manage Referrals and Hospitalizations
|
|
|
Table 4 presents the findings on practice profiling. Several aspects of profiling were associated with cost in the partial models, including development of profiles by the clinics, more frequent preparation of profiles, and distribution of utilization profiles (on laboratory, radiology or imaging, pharmaceuticals, emergency department, and cost of care). Distribution of quality profiles (such as patient satisfaction, preventive services, and patient complaints) and a variety of applications of profiles (including individual feedback and distribution to peers) were also associated with cost. When these variables were included in the combined model with other care management strategies, use of physician profiles of hospital patient days was still associated with higher nonhospital cost, and the use of cost of care profiles was associated with lower nonhospital cost per member per year.
|
|
|
|
Table 4. Effect of Practice Profiling Strategies
|
|
|
Table 5 shows the use of guidelines within these primary care practices. Nearly half of the practices make some use of guidelines. The most common application of guidelines is distribution to physicians. Thirty-eight (22%) group practices use these guidelines in more sophisticated ways, eg, as protocols for telephone care. In the combined multivariate cost model, clinics that used formal educational programs for guidelines were associated with lower nonhospital cost per member. Those clinics that used a larger number of different guideline applications had higher nonhospital costs.
|
|
|
|
Table 5. Effect of Guideline Applications*
|
|
|
Table 6 describes the association of primary care features with nonhospital cost. Nearly half of the clinics request patients to identify a specific primary care physician. Almost 80% of patient problems are handled by their primary care physician, and more than 50% of urgent visits are attended to by the patient's primary care physician. In the partial models, all of these factors were associated with cost of care. In the combined model, clinics that requested patients to identify a primary care physician were associated with significantly lower cost of care.
|
|
|
|
Table 6. Effect of Primary Care Features*
|
|
|
In assessing the primary care features of the practices, we also obtained information regarding the quality of the medical record. There was a high degree of variability in the availability and completeness of the medical record presented to primary care physicians. Medical records were routinely available for urgent visits in 74% of practices, and emergency department reports and hospital discharge summaries were routinely available in only about 50% of groups. The use of such tools as problem lists and checklists for preventive services is also highly variable (problem lists were used in >85% of groups and checklists for preventive services were used in <70% of groups). In the partial models, the availability of medical records on routine and urgent visits, the availability of hospital discharge summaries, and the use of problem lists and checklists were significantly associated with nonhospital cost. In the combined model, these findings did not remain statistically significant.
Medical directors described the work schedule for the primary care physicians in their practices. The average number of hours in the clinic per week for these primary care physicians was 30.1, with an average of 3.94 patients scheduled per hour. All of these features were associated with cost in the partial models. In the combined multivariate cost model, clinics with higher average numbers of patients per hour for primary care physicians were associated with lower nonhospital cost.
As shown in Table 7, several physician characteristics were strongly associated with nonhospital cost per member in the combined multivariate cost model. The proportion of physicians who are in a primary care specialty (general internal medicine, general pediatrics, and family practice) approached statistical significance (P=.02). A greater proportion of family practice and general practice physicians among the primary care specialties was independently associated with lower nonhospital cost (P<.001). Larger clinics with greater total numbers of physicians were also associated with lower nonhospital costs (P<.001). Neither the average number of years of experience nor the sex of the physicians of the clinic was associated with cost.
|
|
|
|
Table 7. Effect of Physician Characteristics
|
|
|
COMMENT
The medical directors' responses revealed considerable variability in the management of primary care in these group practices. Although many of these care management strategies were associated with cost in the model-building process, only a few characteristics were consistently associated with cost in the full multivariate model (which combined all potentially significant primary care management characteristics).
Variables consistently associated with lower nonhospital costs at the P<.01 level were physician involvement in care path development, profiling cost of care, more frequent physician profiling, more patients scheduled per hour, having patients identify a specific primary care physician, a higher proportion of primary care physicians in family and general practice, and a greater number of physicians in the group. Group practice characteristics consistently associated with higher nonhospital cost per member were use of profiles of hospital days and use of a greater number of applications of guidelines.
The relative importance of these effects on nonhospital costs varies somewhat, depending not only on the actual coefficient observed but also on the specification of the independent variable. Because our dependent variable is the natural logarithm of nonhospital cost, the coefficient can be interpreted as the proportion of change in the raw nonhospital cost for a 1-unit change in a continuous independent variable. For example, with a coefficient of -0.22 for the proportion of primary care physicians in the practice, an increase by 10% in the proportion of primary care physicians would be associated with a 2% decrease in nonhospital cost per member per year (0.10 x -0.22). Similarly, with a coefficient of -0.001 for size of group, an increase in 10 for the size of group would be associated with a 1% decrease in nonhospital cost per member per year (ie, 10 x -0.001). Using the same approach, with a coefficient of -0.07 for the average number of patients per hour, adding 1 additional patient per hour to the primary care physician's schedule would be associated with a 7% decrease in nonhospital cost per member per year. For a categorical variable such as "requesting patient to identify primary care physician," the association with nonhospital cost may be calculated36 using the following formula: e(coefficient - SE / 2 x coefficient) - 1. Thus, with a coefficient of -0.22 and an SE of 0.08, clinics that request patients to identify a specific primary care physician are associated with a 19% lower nonhospital cost per member per year in this study. Similarly, using cost of care profiles, with a coefficient of -0.19 and an SE of 0.05, clinics that use cost of care profiles are associated with a 17% lower nonhospital cost per member per year in this study.
The utilization management strategy of practice profiling proved to be consistently associated with lower nonhospital cost of care in our various models. Profiling physicians on cost of care and increased frequency of profiling were each independently associated with lower cost in our final models, controlling for patient and physician characteristics. Results of several other studies23, 26, 37 demonstrate the power of profiling and feedback to change physician practice. Our data suggest that some additional utilization management strategies may also be important to cost performance in the managed care environment. The demand management strategy of identification of patients with long-term illness approached our stringent P<.01 standard for significance. Similarly, several hospital care management strategies (proportion of inpatient case managers and substantial involvement of physicians in care paths) approached or met significant association with lower cost in the combined model.
Primary care physician workload also proved to be significantly associated with nonhospital cost. Our analyses revealed a consistent association of increased patient visits per hour with lower nonhospital cost of care. Of course, the rate at which physicians see patients in their practice is related to a variety of factors, including the complexity of patients and the specialty of the physicians. These factors were accounted for in our analysis. Increased numbers of patients seen per hour by primary care physicians could reflect improved productivity in the organization (ie, greater numbers of services provided per employee). Improved productivity certainly has been identified as a key feature in improving cost performance in health care delivery.31 Nonetheless, increasing patient output and reducing time available per patient also have been identified as potential problems exacerbated by the rise of managed care. These have been suggested to have adverse effects on physician satisfaction,38 patient satisfaction,17, 39-40 and even risk of malpractice.41 Results of a few studies42-43 suggest circumstances in which increased physician workload may be associated with lower costs by means of truncated care. Clearly, further research is required to determine if improved cost performance associated with greater physician productivity is associated with reduced appropriateness of care.
NEITHER CONTINUITY nor comprehensiveness of care was associated with nonhospital cost in our analyses, although continuity has been associated with improved cost or satisfaction in several previous studies.29-30,44 In this study, our measures of continuity and comprehensiveness were crude (eg, medical directors' estimate of the proportion of visits attended to by the primary care physician), in contrast to much more precise methods used in other studies.12, 45 Also, the patients in this study were a relatively young and healthy, commercially insured population. Results of other studies46 suggest that older patients with more severe illnesses may particularly benefit from continuity of provider.
Even if variations across group practices in degree of continuity or comprehensiveness of care were not associated with cost in our analyses, we confirmed that practices that had patients identify a specific primary care physician are associated with significantly lower nonhospital cost of care. This result is consistent with the finding by Forrest and Starfield20 that episodes of illness that begin with a visit to the individual's primary care physician (as opposed to other sources) were associated with lower health care expenditures. Group practice efforts to have patients identify a primary care physician may be more important in larger, multispecialty groups than in the smaller primary care only practices. Results of our previous research47 show that efforts to have patients identify a specific primary care physician are less common in smaller group practices.
The use of guidelines also has been thought to be critical to improving cost and appropriateness of care.22, 48 Results of our study demonstrate that the use of guidelines varies across practice organizations in a managed care environment. The most prominent finding regarding guidelines was that groups that made greater use of guidelines were associated with higher nonhospital cost of care. Many guidelines used in these practices involved enhancing the provision of services (eg, periodic health examination, preventive screening, immunizations, diabetes testing, and asthma care). It may not be surprising that these interventions, although improving appropriateness of care and possibly averting long-term sequelae, would not be associated with lower cost per member per year.49 Analyses that focused on specific subgroups of severely ill patients (ie, diabetes and asthma) might demonstrate beneficial effects from guideline interventions that cannot be detected in the larger population.
The associations of physician characteristics with nonhospital cost are perhaps our most provocative findings. We found that the proportion of physicians who were in primary care (ie, general internal medicine, general pediatrics, family practice, and general practice) approached the P<.01 level of significance in association with lower nonhospital cost. Results of several recent studies15 suggest that primary care physicians may be less expensive than specialty physicians in caring for common problems. Perhaps more controversial is the additional finding that practices with a greater proportion of family physicians were associated with lower nonhospital cost. There is certainly evidence that internal medicine physicians may receive more difficult cases.4, 50 Although in this analysis we performed substantial adjustment of patient characteristics, there could be patients who have higher expectations for service that were not adequately measured in this study. However, there is also considerable literature suggesting that general internists are more expensive than family physicians for patients who have common problems.51-53 There is also some evidence suggesting that internists may be less well prepared than family physicians for practice in a managed care environment.54 Additional research is required to address this controversy.
Group practices with larger numbers of physicians are associated with lower nonhospital cost. Although considerable research has indicated economies of scale associated with group vs solo practice, these efficiencies may be accomplished by group size of less than 10,55 a relatively small group in our study. Other research has suggested that larger group practices are associated with higher cost.56 Of course, results of several studies15, 57 demonstrate lower cost of care for staff model HMOs; however, these studies compared 1 or a few large clinics to physicians practicing in smaller fee-for-service–oriented groups. There has been a continuing wave of consolidation of physician practices into larger organizations.58 Similar findings have been noted by 1 of us (J.K., unpublished data, January 1995). It has been postulated that these integration activities are motivated, in part, by efforts to improve efficiency.59
There are several limitations that should be considered when interpreting these results. A fundamental limitation is the fact that the cost data for this study came from 1995, whereas the surveys reflected clinic characteristics in early 1996. Thus, the interpretation of clinic characteristics associated with an increased cost is fraught with hazard. The most obvious difficulty is the possibility that a clinic that recognized a cost problem during 1995 might have subsequently introduced an intervention (which would have been reported in our survey). Given the rapid dissemination of care management interventions in clinics and hospitals, it is certainly plausible that some of the associations with higher cost could be confounded by this problem.
There are several additional limitations related to our measure of cost that also must be considered when reviewing the results. First, our analyses only related to users of medical resources. Individuals in the managed care plans at BCBSM who did not use services during 1995 were excluded from this analysis. Second, our estimate of costs excludes mental health services. Third, the analyses exclude inpatient expenditures because the group practices have little effect on hospital rate setting; in many cases (particularly for rural hospitals), groups have little choice of the hospital facility most convenient for their patients. We also conducted cost analyses, including hospital expenditures, that produced similar results to those reported herein.
In addition, our cost variables are derived from the estimated actual dollar expenditures by BCBSM for services provided by the various medical group practices in this study. The typical basis for payment to these practices is derived from a standard fee schedule and a withhold pool (with annual settlement based on group practice performance relative to utilization targets). Nonetheless, the nature of these agreements varies among clinics based on annual negotiations between the health plan and group practice leaders. Thus, what 1 group practice receives in payment from BCBSM for performing an individual physical examination, endoscopy, or complete blood cell count may differ from the specific payments received by another clinic. We used our best available estimate of the actual dollars expended by BCBSM for nonhospital clinical care on these enrollees during 1995.
This study was conducted in the relatively unique environment of the upper Midwest. Health maintenance organizations have long been an important aspect of financing care in Minnesota, and this area remains one of the most advanced managed care markets in North America. The utilization management and care management techniques used by Minnesota health plans could, of course, suppress the benefits of such strategies conducted within group practices; however, such health plan–directed strategies are now common to managed care products in metropolitan areas throughout the United States. More than 30% of our subject enrollees reside in rural areas and obtain their primary care from local medical practices. In these areas, intensively managed health plans are a relatively recent method of medical care financing.
Minnesota is also characterized by group-style medical practice; this is evidenced not only by the long-standing prominence of several large multispecialty group practices (which have been joined by numerous others in recent years) but also by the relatively early disappearance of solo-style primary care practice. Thus, both physicians and administrators are familiar with the problems and opportunities presented by group medical practice. Less well-established group practices, or newer forms of organizing physicians (eg, group practices without walls), may not be as effective in implementing these strategies. Furthermore, the practice characteristics we are studying were not assigned randomly to physicians' practices. Group practices decided whether they wanted to adopt specific characteristics. Thus, the results we report are only measures of association between those characteristics and cost per member per month for practices that chose to adopt the characteristic. We cannot conclude that applying a specific utilization management technique, for example, to a group practice that had not chosen to adopt it would produce the same result that we observe for group practices that chose voluntarily to adopt it.
Despite these limitations, the results of our study demonstrate substantial variation in the management of primary care in group practices participating in an HMO network. These differences are associated with significant variation in the nonhospital cost of care for enrollees. Group practice features most strongly associated with lower nonhospital cost include the use of physician profiling, patient identification of primary care physicians, the workload of the primary care physicians, and the characteristics of the physicians practicing in the group.
AUTHOR INFORMATION
Accepted for publication March 23, 1998.
Funded by Blue Cross Blue Shield of Minnesota and the Blue Cross Foundation of Minnesota, St Paul.
Reprints: Eugene C. Rich, MD, Department of Internal Medicine, Creighton University, 601 N 30th St, Suite 5850, Omaha, NE 68131-2197 (e-mail: richec{at}creighton.edu).
From the Department of Internal Medicine, Creighton University, Omaha, Neb (Dr Rich); Institute for Health Services Research and Policy, University of Minnesota, Minneapolis (Drs Kralewski, Feldman, and Dowd); and Network Management Division, Blue Cross Blue Shield of Minnesota, St Paul (Mr Bernhardt).
REFERENCES
| |
1. McMahon L, Wolfe R, Tedeschi P. Variation in hospital admissions among small areas: a comparison of Maine and Michigan. Med Care. 1989;27:623-631.
FULL TEXT
|
ISI
| PUBMED
2. Chassin M, Brook R, Park R, et al. Variations in the use of medical and surgical services by the Medicare population. N Engl J Med. 1986;314:285-290.
ABSTRACT
3. McPherson K, Wennberg J, Hovind O, et al. Small-area variations in the use of common surgical procedures: an international comparison of New England, England, and Norway. N Engl J Med. 1982;307:1310-1314.
ABSTRACT
4. Weiner JP, Starfield BH, Powe NR, Stuart ME, Steinwachs DM. Ambulatory care practice variation within a Medicaid program. Health Serv Res. 1996;30:751-770.
ISI
| PUBMED
5. Starfield B, Powe NR, Weiner JR, et al. Costs vs quality in different types of primary care settings. JAMA. 1994;272:1903-1908.
ABSTRACT
6. Piette JD, Moos RH. The influence of distance on ambulatory care use, death, and readmission following a myocardial infarction. Health Serv Res. 1996;31:573-591.
ISI
| PUBMED
7. Komaromy M, Lurie N, Osmond D, Vranizan K, Keane D, Bindman AB. Physician practice style and rates of hospitalization for chronic medical conditions. Med Care. 1996;34:594-609.
FULL TEXT
|
ISI
| PUBMED
8. Bindman AB, Grumbach K, Osmond D, et al. Preventable hospitalization and access to health care. JAMA. 1995;274:305-311.
ABSTRACT
9. Laine CL, Davidoff F, Lewis CE, et al. Important elements of outpatient care: a comparison of patients' and physicians' opinions. Ann Intern Med. 1996;125:640-645.
FREE FULL TEXT
10. Bartman BA, Clancy CM, Moy E, Langenberg P. Cost differences among women's primary care physicians. Health Aff (Millwood). 1996;15:177-182.
ABSTRACT
11. Shortell SM, Gillies RR, Anderson DA. The new world of managed care: creating organized delivery systems. Health Aff (Millwood). 1994;13:46-64.
ABSTRACT
12. Starfield B. Organization, financing, and access to services. In: Primary Care: Concept, Evaluation, and Policy. New York, NY: Oxford University Press; 1992:chap 8.
13. Hargraves JL, Palmer RH, Orav EJ, Wright EA. Practice characteristics and performance of primary care practitioners. Med Care. 1996;34(suppl 9):SS67-SS76.
14. Lurie N, Margolis KL, McGovern PG, Mink PJ, Slater JS. Why do patients of female physicians have higher rates of breast and cervical cancer screening? J Gen Intern Med. 1997;12:34-43.
FULL TEXT
|
ISI
| PUBMED
15. Greenfield S, Nelson EC, Zubkoff M, et al. Variations in resource utilization among medical specialties and systems of care: results from the Medical Outcomes Study. JAMA. 1992;267:1624-1630.
ABSTRACT
16. Kaplan SH, Sullivan LM. Maximizing the quality of the physician-patient encounter. Ann Intern Med. 1996;121:187-188.
17. Hall JA, Roter DL, Katz NR. Meta-analysis of correlates of provider behavior in medical encounter. Med Care. 1988;26:657-675.
ISI
| PUBMED
18. Eisenberg JM. The internist as gatekeeper: preparing the general internist for a new role. Ann Intern Med. 1985;102:537-543.
19. Reagan MD. Physicians as gatekeepers: a complex challenge. N Engl J Med. 1987;317:1731-1734.
ISI
| PUBMED
20. Forrest CB, Starfield B. The effect of first-contact care with primary care clinicians on ambulatory health care expenditures. J Fam Pract. 1996;43:40-48.
ISI
| PUBMED
21. Powe NR, Weiner JP, Starfield B, Stuart M, Baker A, Steinwachs DM. Systemwide provider performance in a Medicaid program: profiling the care of patients with chronic illnesses. Med Care. 1996;34:798-810.
FULL TEXT
|
ISI
| PUBMED
22. Greco PJ, Eisenberg JM. Changing physicians' practices. N Engl J Med. 1993;329:1271-1274.
|
