Relationship Between Primary Care Physician Financial Risk and Member Emergency Department Use in a Commercial HMO Population

Published on: 
The American Journal of Managed Care, June 2006, Volume 12, Issue 6

Objective: To demonstrate whether, despite factors beyond the control of primary care physicians (PCPs), health maintenance organization (HMO) members under the care of PCPs with greater financial risk for the cost of emergency care have lower rates of emergency department use.

Study Design: Cohort study using calendar year 2000 administrative data on 217 298 commercial members enrolled in a nonprofit statewide HMO (Blue Care Network of Michigan) under the care of non-staff-model PCPs with varying levels of financial risk for emergency care.

Methods: Ordinary least squares (OLS) and binary logistic regression models were developed to assess the influence of PCP financial risk, net effects of member and PCP demographics, and emergency care accessibility on use of emergency treat-andrelease services by members.



Results: OLS results indicated emergency use was lower by 33 visits per 1000 (< .001) and 51 visits per 1000 (< .001) for members with PCPs who had medium and high financial risk, respectively, compared with members whose PCPs had low financial risk for emergency care. Emergency care availability, member copayment and demographics, and the number of all Blue Care Network members assigned to the PCP also were significant predictors of emergency use.

Conclusions: PCPs do have the ability to influence their patients' emergency department use if financially motivated to do so. Ensuring that the PCP has a large enough number of patients under a specific contractual arrangement also is an important component of the success of an associated financial incentive.

(Am J Manag Care. 2006;12:329-340)

Not all care delivered in the emergency setting is truly emergent. According to national data from the Centers for Disease Control and Prevention, the percentage of emergency visits classified as "nonurgent" varied from 9% to 12.8% between 1999 and 2003.1-5 In addition, an increasing proportion of emergency care utilization is driven by the discretionary demand of the insured, as opposed to an actual or perceived clinical need for immediate care in the emergency setting.6-8 This type of utilization actually has 3 possible drivers: patients, primary care physicians (PCPs), and, to a degree, hospital-affiliated emergency departments themselves.

Patients may demand discretionary emergency care because of its immediate availability and convenience.9 Patients often want to access care on their schedule and want immediate diagnostic testing with results.

PCPs may drive their patients' demand for discretionary emergency care from a different perspective. A PCP may want a place for his or her patients to receive immediate care and evaluation without disrupting daytime office schedules or impinging on the PCP's afterhours time. A PCP may not wish to assign nonphysician staff or other resources to address unscheduled patient clinical issues. In addition, PCPs in a capitated health maintenance organization (HMO) environment have another potential reason to add to the overall demand for emergency care. Such PCPs may wish to devote office time to non-HMO patients who will provide positive incremental (marginal) revenue, rather than to a capitated HMO member who offers no incremental financial gain.10 Although such PCP-driven demand would be manifest as patient demand, the cause of the patient demand would be limited (ie, difficult) access to the PCP as determined by the PCP.11,12

Hospital emergency departments also have worked to drive patient demand for their services with "fast-track" or "urgent care" areas located either in or adjacent to the emergency department itself, along with marketing campaigns to attract greater numbers of patients.13-17 These efforts to drive greater demand are occurring despite contemporaneous concerns about emergency department overcrowding.18-21 Hospital-located urgent care areas initially were intended as a low-cost infrastructure solution to decompress emergency departments during peak times due to growing numbers of patients. However, these urgent care areas have themselves become primary care destinations. Hospital-located urgent care now can be viewed as competing both with PCP offices and with efforts by health plans to redirect members to their PCP for PCP-treatable clinical events.22

From the perspective of a health plan with prepaid, capitated PCP office visits, a hospital emergency department is not a financially desirable point of care for PCP-treatable conditions. However, low-intensity urgent care visits have low marginal operating costs for emergency departments and may be billed and paid at emergency care rates.23 In the study by Williams, 32% of 24 010 emergency visits were categorized as nonurgent for reasons such as acute pharyngitis, otitis media, or upper respiratory tract infection.23

Blue Care Network of Michigan (BCN) is a nonprofit, mixed-model statewide HMO and wholly owned subsidiary of Blue Cross Blue Shield of Michigan. BCN of Michigan had approximately 565 000 commercial members and more than 3000 PCPs in 2000. Many BCN PCPs have stated that they have little or no ability to influence emergency department utilization by their patients. Anecdotally, PCPs have cited factors including no or low member emergency copayments, prudent layperson laws, aggressive hospital advertising campaigns, and illness burden of their patients as explanations for why they cannot manage emergency use by their patients. Although these factors may contribute to emergency utilization patterns, PCP-related factors such as managed care financial risk arrangements also may influence emergency care utilization.

In a review article, Hellinger concluded that past studies consistently have shown financial risk to be an important element in explaining lower levels of healthcare utilization.24 Hellinger states that individually these studies have suffered from bias due to patient selection, physician selection, and/or other unmeasured characteristics. A Council on Medical Service report discussed how other prior studies have measured the influence of managed care in its entirety on utilization, assessing the combined impact of noneconomic methods (eg, prior authorization) and coexisting financial incentives.25 These investigations, which compared fee-for-service (FFS) utilization with managed care as a whole, did not truly isolate the influence of the financial incentives.

The purpose of this study is to explore the effect of physician financial arrangements with a health plan on 1 specific aspect of healthcare utilization, while attempting to address some of the gaps in the literature. The hypothesis is that patient demand for discretionary emergency care is significantly lower when PCPs have greater financial risk sharing for emergency use.


This study used BCN administrative data (ie, claims, member and physician demographics, provider contractual characteristics) from calendar year 2000. Publication of this article is with the permission of BCN.

The use of administrative claims data does not allow a definitive differentiation of discretionary (in the mind of the patient) emergency utilization from nondiscretionary utilization. One option to identify potentially discretionary visits is to separate out claims by final disposition. Claims were segregated as to whether the emergency visit resulted in either an inpatient admission or overnight observation stay versus those where the patient was treated and released. Only the treat-and-release cases were used in the remainder of this analysis. Certainly not every treat-and-release emergency case is discretionary; however, the assumption is that the prevalence of discretionary utilization would be higher in treat-and-release patients than in patients whose condition required an actual hospitalization. The member's overall illness burden was incorporated into the analysis in an effort to account for the incidence of treat-and-release emergency use that may not have been deemed discretionary by the member or the PCP.

BCN service areas and membership are divided among 4 geographic regions for administrative purposes (Figure). There was no coexisting prior authorization requirement for emergency care during the study period.


All PCPs included in the study were affiliated with a primary care group (PCG). PCGs (and thus their PCPs) are affiliated with only 1 BCN region. The PCGs are not physician groups. Each PCG is a business entity comprised of 1 or more physician practices, and functions essentially like an independent practice association or physician organization. The PCP practices belonging to a PCG may consist of physicians who are salaried, independent, or a mixture of both. Generally, each PCP (as opposed to the practice the PCP may belong to) has an individual contract with the PCG with which he or she is affiliated. The PCGs vary as to the comprehensiveness and sophistication of their administrative structures in support of the financial goals of the PCG, and in their risk-sharing arrangements both with BCN and with their PCPs.


Because financial arrangements are largely between the PCG and the PCP (without the involvement of BCN), they were not always known with certainty for purposes of this study. However, PCG self-reported information has indicated that the greater the financial risk the PCG agreed to in their contract with BCN, the greater the financial risk their PCPs are exposed to. The PCG self-reported information has face validity because the PCG would have no business incentive to shield their PCPs from the same or similar magnitude of financial risk. Thus, the PCG risk arrangement was used in this study as a surrogate measure for the financial risk experienced by the PCP.

Although BCN operates staff-model locations where the PCPs are BCN employees, the vast majority (>90%) of overall BCN membership had non-staff-model PCPs in 2000. Non-staff-model individual base PCP reimbursement was either by FFS or capitation in 2000. The PCG risk sharing varied from 0% to 100% of funded risk pools for defined categories of utilization (inpatient, outpatient, referral, and others). The PCGs that had 0% risk for all risk pools were comprised of PCPs who were paid solely on an FFS basis for primary care. Essentially, all (>99%) of the remaining PCPs received capitated payments for primary care services.

Each PCG had complex financial arrangements with BCN. Depending on the PCG, these arrangements could include maximum overall annual gain/loss limits, capitation withholds, and individual patient stop-loss thresholds used in calculating PCG settlements based on overall risk-pool performance to targets set by BCN. These other arrangements were not included in the final analysis because they are assessed after a year has concluded and include all areas of utilization covered by the different risk pools, not emergency utilization alone. At the start of each year, all that is known is the level of stated risk and funding for each utilization category, not the final overall outcome. Therefore, the relevant metric to assess provider behavior during the course of a year is the level of stated prospective risk on January 1.

Data and results aggregation was accomplished using Microsoft Excel and Access (Microsoft Corp, Redmond, Wash). Statistical analysis was done by using SPSS software (SPSS, Chicago, Ill). The BCN staff provided certain data elements using GeoAccess (Ingenix, Lenexa, Kan) and DxCG (DxCG, Boston, Mass) products.

To initially control for known areas of potential bias as much as possible, data were assembled for a subset of BCN members with stable member and PCP characteristics throughout the study period. These characteristics were plan benefits, plan enrollment, PCP affiliation, and PCP reimbursement methodology (ie, PCG affiliation). Members with a staff-model PCP were excluded because staff-model PCPs operate under a different reimbursement model than non-staff-model PCPs. The analysis was done on member-level data rather than aggregate information.

A total of 217 298 members were identified, representing approximately 38% of the total BCN commercial population in 2000. They had the following characteristics:

  • Commercial plan benefits.
  • The same PCP for all of 2000.
  • A PCP who was affiliated with the same PCG for all of 2000.

Additionally, the following data elements were identified for each of the 217 298 members:

  • Number of treat-and-release emergency visits (visits that did not result in an inpatient admission or observation stay) in 2000.
  • The total number of BCN members assigned to the member's PCP (panel size), regardless of whether the additional members were not in the study population.
  • The member's office visit copayment ($0, $5, $10 or $15).
  • The financial arrangement between BCN and the PCG (matched to the member's PCP) for emergency visits and PCP-provided primary care.

Bivariate correlation was performed for member emergency copayment and office visit copayment. The Pearson's correlation coefficient was 0.808. Therefore, office visit copayment was excluded from further analysis due to the high correlation and collinearity concerns. The direct cost of a treat-and-release emergency visit impacted 2 different risk pool funds for a PCG: the outpatient fund and the referral fund. The facility claim from the hospital emergency department was assigned to the outpatient fund, as were other specified types of hospital outpatient services. Emergency physician professional claims impacted the referral fund, as did any professional claim paid to a specialist who receives a referral from a PCP. Thus, although a PCP who was capitated for primary care would be better off financially to refer as much care as possible, such behavior was tempered by any coexisting financial risk for the costs of such referrals. In the case of emergency care, the cost of a member seeking medical attention in that venue impacted a physician who was financially at risk in both the outpatient and the referral funds. Although the risk percentage was frequently the same for both funds, there were occasional differences. For the purposes of this study, the emergency visit risk percentage was calculated as the simple average of these 2 funds for each PCG.

The average BCN combined facility and professional treat-and-release emergency visit payment was $362 for all visits that occurred for all BCN commercial members in 2000. Hospital-located urgent care centers were billed and paid as emergency visits. It is unknown which encounters were handled in an adjacent urgent care area (if present) versus the emergency department proper. For the study set of 217 298 members who incurred a total of 42 888 treat-and-release emergency visits in 2000, the average BCN total cost per visit was $371. The total expense for the year was nearly $16 million, or $6.10 per member per month. All of the above-stated amounts were the net of copayments, coordination of benefits, and any other member or provider contractual adjustments.

Study of the BCN cost of these visits was problematic because reimbursement was governed by a diverse mix of reimbursement schedules and other factors that varied across emergency departments. These variations represented a major threat to the validity of any in-depth analysis of the relationship between emergency care expense and PCP financial risk. Thus, this study focused on use as opposed to cost.

The overall BCN membership assigned to the PCP (regardless of whether members were part of the study population) was included in an attempt to determine (1) how much the PCP might have cared about BCN financial programs specifically (ie, PCP engagement with BCN members) and (2) possibly the general managed care emphasis of the practice. No information was available on the overall makeup of each PCP practice (ie, what percentage were in HMOs vs traditional FFS insurance, what percentage belonged to BCN).


Classification of Diseases, Ninth Revision


Relative risk scores were obtained for each of the studied members based on BCN claims data. The relative risk score was calculated using DxCG software, which uses Diagnostic Cost Group/Hierarchal Condition Category models. The basic components of the Diagnostic Cost Group/Hierarchal Condition Category model are Diagnostic Groups, which are used to combine related conditions into Condition Categories, which are then organized into hierarchies. Risk is assessed with only the highest cost category, with cost based on a large national representative commercially insured population sample. Age and sex are also components of the disease groupings and risk score calculation.26 This same methodology has been used by the Centers for Medicare and Medicaid Services for Medicare HMO capitation payment calculations since 2004.27 For BCN, diagnosis codes from all member inpatient and outpatient claims (excluding lab, radiology, and pharmacy) during all of 2000 were used in the DxCG concurrent model. Concurrent models (year 1 data to explain year 1 illness burden), as opposed to predictive models (year 1 data to predict year 2 illness burden), are primarily useful for retrospective analyses, because knowing all the medical conditions treated during a time period is especially germane when assessing the resources required.28 Cumming and Cameron compared different claims-based methods of health-risk assessment and found that the DxCG concurrent methodology had the highest 2 value compared with 6 other concurrent-model methodologies (0.564 with claims truncated at $50 000).26

The benchmark population used for this study was the entire BCN commercial population for 2000, which was given a value of 1. The scores represent the overall retrospective relative illness burden of these members during all of 2000, which was then used to adjust the total treat-and-release emergency utilization of each member for the same year.


For BCN, the Southeast region has generally had higher emergency utilization rates than the other BCN regions. Because immediate availability and access to care are thought to be component drivers of emergency utilization, the density of emergency departments in each region was assessed. GeoAccess software was used to generate a report on inpatient facilities within the BCN regions. Almost all had a 24-hour emergency department in 2000. The output included a count of how many facilities were within 10 miles of each listed facility. In other words, facility density was used as a surrogate for ready access to an emergency department. The Southeast average facility density was about 10 times the average density of the other geographic regions. (The averages for the other 3 regions were nearly equivalent.) Thus, a dichotomy was created between the Southeast region and the rest of the state for emergency department density.

Age and sex of study group members were accounted for in the relative risk score. To obtain other demographic information, US Census Bureau data were reviewed as a proxy measure for unknown BCN member socioeconomic metrics.29,30 The assumption was that the characteristics of those who reside within the same geographic area as the member (as determined by zip code) could serve as an indirect indicator for unavailable member-specific demographics. Of the 217 298 BCN members included in the study, 96.7% had valid Michigan zip codes on file with BCN for analysis. These member mailing address zip codes were mapped to the 4 BCN regions. Subsequently, the member's PCP region was used in this analysis to be consistent with later analyses, as the member's mailing address region and member's PCP region matched 95.8% of the time.

Analysis of the 4 BCN regions was done. Review of per capita income, percent nonwhite race, and percent urban revealed substantive differences between the Southeast and the other 3 BCN regions (Table 1). The Southeast region had the highest per capita income and the highest percentage nonwhite race, and was the most urban. The percent urban differential is consistent with the hospital density difference noted previously.

The basic model specification was the number of treat-and-release emergency visits for each member as the dependent variable. PCP region (Southeast or not), member relative risk score, member emergency copayment, PCP (PCG) risk pool financial risk percentage, and the total number of BCN members assigned to the member's PCP (panel size) were the independent variables.

Preliminary and final analyses were done using ordinary least squares (OLS) regression. An OLS model allowed for calculation of adjusted treat-and-release utilization rates, which have practical utility from a business perspective. As the dependent variable represented positively skewed count data with a high proportion of zeros, confirmation analysis of the OLS results was done with binary logistic regression. For the logistic model, the dependent variable was defined as emergency use during the year studied (either zero visits or at least 1 visit). The 2 models were compared to ascertain whether the relationships remained consistent to support the use of the OLS methodology. Regardless of the violation of basic assumptions, such as linearity of relationships and normally distributed residuals, which would be expected to result in erroneous output if not mitigated by the large study population in a social science analysis.31(p6)

The data were grouped into categories as noted in Table 2. Copayments were already categorized into equal discrete intervals. PCP financial risk was discrete with irregular intervals, and therefore was grouped to produce uniform category intervals so that model coefficients would have a clear interpretation. Relative risk score and PCP panel size were continuous, but were categorized for various practical reasons. After calculation of the relative risk scores, the DxCG software can produce aggregated diagnostic cost groups for either the concurrent or predictive models at the member level. The value range is 1 through 5, with 5 representing the most resourceintensive (ie, most ill) category of members, from either a backward (concurrent) or forward (predictive) perspective depending on the DxCG model used. The relative risk score distribution for the study population was positively skewed, with a mean of 1.14, median of 0.27, and a maximum value of 140. Because of BCN technical issues, the actual aggregated diagnostic cost group scores for this set of data were not available at the time of this analysis. Therefore, the study population's relative risk scores were categorized by quintiles, as these were functionally equivalent to the categorization scheme produced for the aggregated diagnostic cost groups; the quintile groupings also reduced the effect of extreme high outliers without having to trim cases from the data.

PCP BCN panel size also was positively skewed, with a mean of 184.5, a median of 153, and a maximum value of 707. Therefore, PCP BCN panel size was categorized to mitigate the nonnormal distribution of the continuous data. In addition, physician behavior would more likely vary in a stepped fashion as thresholds of assigned members are attained. The panel size categories, while arbitrary, represent generally how BCN internally views this measure, as well as resulting in a reasonably symmetric distribution.

For the OLS model, the original continuous-count data values were used for the dependent variable. All other independent variables were specified as binary variables for each variable's category grouping (Table 2), with the lowest value category serving as the reference group for each independent variable.


The distributions of member and PCP characteristics are summarized in Table 3 and Table 4. The vast majority of members never used the emergency department for care. Only a small percentage (3.3%) could be labeled as frequent users (>1 visit in 12 months). The lowest category of PCP financial risk (=25%) was comprised exclusively of PCPs (n = 10) from 2 solely FFS (ie, 0% financial risk) PCGs in different regions. Additionally, the highest category of PCP financial risk (>75%) was comprised of PCPs from a single PCG in the Southeast region. This single PCG had BCN members whose indirectly measured demographic characteristic values were greater on 2 of the 3 measures from the characteristic values of members of the Southeast region in its entirety, possibly resulting in an underestimation of the effect of the high financial-risk category (Table 5).

Most (96%) of the study members were assigned to PCPs in either the low financial risk or medium financial risk categories. Thus, FFS could not be the reference group, because the low financial risk and medium financial risk dummy variables were highly correlated (Pearson's correlation coefficient of -0.892), resulting in an ill-conditioned model due to collinearity. Therefore, the first exploratory OLS model was altered to make the PCP low financial-risk category the reference group for this variable.

Results of a second exploratory OLS model indicated that neither the FFS category for the PCP financial-risk variable nor the $75 copayment category for the copayment variable was statistically significant. Observed power for both of these dummy variables (0.104 for FFS and 0.061 for $75 copayment) indicated insufficient sample size for each of these 2 groups. Therefore, the decision was made to eliminate members (n = 547) with a $75 emergency copayment per year or PCPs paid solely FFS from further analysis.

For the final OLS model (Table 6) all coefficients showed statistical significance. Review of SPSS model diagnostics revealed no issues with collinearity. Examination of the residuals revealed a nonnormal distribution. Although the large sample sizes resulted in observed power of more than 99% for all independent variables, the issue is the substantive (practical) significance of the coefficients.31(p46-51)




The binary logistic regression model (Table 7), which also excluded members with PCPs paid FFS or a $75 copayment, had independent dummy variables specified exactly as in the final OLS model. The model converged and had a Nagelkerke 2 value of 0.228. The Nagelkerke 2 is a measure of the strength of association and ranges from 0 to 1. This value is an approximation of the OLS-adjusted 2 statistic, but does not represent the actual percent variance explained.

For members who visited the emergency department at least once, increasing relative risk score (overall illness burden) was the strongest predictor, followed by the PCP being in the Southeast region, consistent with the standardized coefficients in the OLS model. Member copayments, higher PCP financial risk, and a PCP having a larger overall BCN panel membership all decreased the odds of visiting the emergency department. All odds ratios were statistically significant.

The OLS coefficients were used to calculate average adjusted effects on rates of emergency treat-and-release utilization per 1000 members by model dimension categories (Table 8). Reference group utilization was inclusive of all study members identified using only the reference group characteristic noted for that model dimension.


This study supports the view that the financial risk of a PCP can influence his or her behavior, and thus health plan member behavior, as it relates to emergency department utilization. Although capitation is a form of financial risk, as a stand-alone reimbursement mechanism it offers only a negative incentive to the PCP to maintain adequate member access to primary care and to avoid increased discretionary emergency utilization. (Primary care and PCP access are defined for this study to include PCP-provided primary care along with any PCP-controlled nonphysician staff and services that may be available, whether in person or by other means of communication.)

Additional risk arrangements working in concert with capitation are needed to offset this negative incentive (eg, secondary risk pools for services such as emergency care). The odds ratios in the logistic regression output reveal that BCN members whose PCPs had medium and high financial risk were less likely than members whose PCPs had low financial risk to have had 1 or more treat-and-release emergency visits in 2000, after adjusting for PCP BCN panel size, copayment, region, and member relative risk score. Compared with members whose PCPs were at low financial risk, members whose PCPs were at medium and high financial risk had 15.9% and 24.4% lower rates of emergency department utilization, respectively, as determined by the OLS model (controlling for the other independent variables).

The RAND Health Insurance Experiment supports, conceptually, that member copayments affect emergency department use.32 It also is substantiated more specifically by Selby et al, who reported an overall 14.6% decrease in emergency visits after the introduction of a new $25 or $35 copayment for emergency care.33 The greatest effect was reported with cases classified as "often not an emergency," which comprised 25% of emergency visits before the introduction of the copayment. The reduction in emergency use noted by Selby et al compares with the 7.8% reduction when the copayment changed from $0 to $25 in the BCN study population. In the study by Selby et al, only age, sex, and socioeconomic status were controlled for.

The percent reduction in emergency department use when the copayment changed from $0 to $50 and when the PCP financial risk changed from low to high was less than twice the percent reduction when the copayment changed from $0 to $25 and when PCP financial risk changed from low to medium, despite uniform category intervals. This suggests that the effect of cost sharing on emergency utilization lessens, for both the member and the PCP, as cost sharing increases to higher levels. This conclusion is supported by the smaller OLS standardized coefficients for the $50 copayment and high PCP financial risk compared with those for the $25 copayment and medium PCP financial risk. The emergency visits included in the study represent an unknown mixture of actual discretionary and actual nondiscretionary events. It makes intuitive sense that as cost sharing increases to higher levels, the residual amount of discretionary use to deter will decrease.

Review of the effect of PCP panel size from the OLS model shows a plateau in the relationship between panel size and emergency use. The percent change in emergency use was essentially the same for the mediumpanel and large-panel categories versus the small-panel category. When PCPs had a total BCN panel of 101 to 300 members, emergency use decreased by 12.5% compared with use when PCPs were responsible for 100 or fewer BCN members. Panel sizes greater than 300 members appeared to not be associated with additional reductions in use.

Lower emergency department use likely is a reflection of greater member access to their PCPs' practice for medical direction and/or the provision of medical care. Although it is postulated that members assigned to PCPs with greater financial risk received the necessary access to their PCP as a substitute for the emergency department, it is not definitively known from this study. As PCP access can be viewed as a quality outcome in and of itself, the assumption is that these PCPs and their PCGs had better structures and processes in place to achieve that outcome.34

The temporal relationship between physician assumption of financial risk and the reduction in emergency utilization is not clear. It is not known whether only PCGs (and their affiliated PCPs) with superior internal structures and processes were willing to accept risk, or if they developed superior infrastructure in response to a past decision to accept risk. Complete details of each PCP's financial arrangement with its PCG, along with actual member demographics beyond age and sex, were not available, nor were overall PCP practice demographics.

PCPs can influence their patients' emergency department use if financially motivated to do so. The exact structures and processes that are used to achieve these results are beyond the scope of this paper. The relationship between panel size and emergency use suggests that making sure that the PCP has a large enough number of patients under a specific contractual arrangement (ie, health plan engagement) is an important component for the success of an associated financial incentive.

Due to variations in benefit designs, contractual relationships, and unmeasured member characteristics, these results may not be generalizable to other insurance products or populations, such as Medicare or Medicaid, or to other areas of utilization.


I acknowledge Douglas R. Woll, MD, BCN Chief Medical Officer, for his review of the manuscript; Janet Jennings, MS, BCN Director of Medical Informatics, for her statistical review and review of the manuscript; and the BCN Medical Informatics, Membership, and Experience Reporting areas for their support in providing various necessary data extracts.

From Blue Care Network of Michigan, Southfield, Mich. Dr Goodman performed this study while a full-time employee of Blue Care Network of Michigan.

Address correspondence to: Robert M. Goodman, DO, MHSA, Associate Medical Director, Blue Care Network of Michigan, Mail Code C336, 20500 Civic Center Dr, Southfield, MI 48076. E-mail: