Original Article
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
POPULATION HEALTH MANAGEMENT Volume 00, Number 00, 2017 ª Mary Ann Liebert, Inc. DOI: 10.1089/pop.2017.0090
Health Care Resource Utilization for Outpatient Cardiovascular Disease and Diabetes Care Delivery Among Advanced Practice Providers and Physician Providers in Primary Care Salim S. Virani, MD, PhD,1–4 Julia M. Akeroyd, MPH,1 David J. Ramsey, PhD,1 Anita Deswal, MD, MPH,2,5 Khurram Nasir, MD,6,7 Suja S. Rajan, PhD, MS,8 Christie M. Ballantyne, MD,3,4 and Laura A. Petersen MD, MPH1
Abstract
Although effectiveness of diabetes or cardiovascular disease (CVD) care delivery between physicians and advanced practice providers (APPs) has been shown to be comparable, health care resource utilization between these 2 provider types in primary care is unknown. This study compared health care resource utilization between patients with diabetes or CVD receiving care from APPs or physicians. Diabetes (n = 1,022,588) or CVD (n = 1,187,035) patients with a primary care visit between October 2013 and September 2014 in 130 Veterans Affairs facilities were identified. Using hierarchical regression adjusting for covariates including patient illness burden, the authors compared number of primary or specialty care visits and number of lipid panels and hemoglobinA1c (HbA1c) tests among diabetes patients, and number of primary or specialty care visits and number of lipid panels and cardiac stress tests among CVD patients receiving care from physicians and APPs. Physicians had significantly larger patient panels compared with APPs. In adjusted analyses, diabetes patients receiving care from APPs received fewer primary and specialty care visits and a greater number of lipid panels and HbA1c tests compared with patients receiving care from physicians. CVD patients receiving care from APPs received more frequent lipid testing and fewer primary and specialty care visits compared with those receiving care from physicians, with no differences in the number of stress tests. Most of these differences, although statistically significant, were numerically small. Health care resource utilization among diabetes or CVD patients receiving care from APPs or physicians appears comparable, although physicians work with larger patient panels. Keywords: physicians, advanced practice providers, health care resource utilization, diabetes, cardiovascular disease debate, the system would be further strained if the 20 million Americans who obtained health insurance under the ACA retained some form of health care coverage. A greater use of advanced practice providers (APPs) (nurse practitioners [NPs] or physician assistants) has been proposed to address these shortages, especially for chronic disease care delivery.2
Introduction
T
he Association of American Medical Colleges estimates that there will be a shortage of 45,000 primary care physicians by 2020 and 65,000 by 2025.1 Although the future of the Affordable Care Act (ACA) is currently under
1 From the Health Policy, Quality & Informatics Program, Michael E. DeBakey Veterans Affairs Medical Center Health Services Research and Development Center for Innovations; and Section of Health Services Research, Department of Medicine, Baylor College of Medicine, Houston, Texas. 2 Section of Cardiology, Michael E. DeBakey Veterans Affairs Medical Center, Houston, Texas. 3 Section of Cardiovascular Research, Department of Medicine, Baylor College of Medicine, Houston, Texas. 4 Center for Cardiovascular Disease Prevention, Methodist DeBakey Heart and Vascular Center, Houston, Texas. 5 Section of Cardiology, Department of Medicine, Baylor College of Medicine, Houston, Texas. 6 Center for Healthcare Advancement & Outcomes, Baptist Health South Florida, Miami, Florida. 7 The Johns Hopkins Ciccarone Center for the Prevention of Heart Disease, Baltimore, Maryland. 8 Division of Management, Policy and Community Heath, School of Public Health, University of Texas—Health Science Center at Houston, Houston, Texas.
1
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
2
VIRANI ET AL.
In prior studies performed in small practices3 and recent studies performed in large national cohorts of patients,4,5 it has been shown that the effectiveness of outpatient delivery of cardiovascular disease (CVD) and diabetes care is comparable among patients receiving care from physicians or APPs.5 Although effectiveness of routine outpatient CVD and diabetes care delivery has shown to be comparable between physicians and APPs,5 it is not clear whether this comes at the price of increased health care resource utilization by APPs compared with physicians. Prior studies on the topic were performed in small academic settings6–8 and did not compare these findings in large health care systems in a ‘‘real-world’’ primary care setting. Therefore, the aim of the present analyses was to compare health care resource utilization between patients receiving care from physicians or APPs. Using the Department of Veterans Affair’ (VA) robust administrative and clinical data set of national cohorts of patients with diabetes or CVD, this study assessed whether there were any clinically meaningful differences in health care resource utilization between patients receiving care from physicians or APPs. Diabetes and CVD were chosen as 2 representative chronic disease conditions, given their high prevalence, high cost, and their significant contribution to overall morbidity and mortality.9
FIG. 1.
Methods Cohort development and study population
Using VA’s administrative and clinical data sets, the study team created 2 separate national cohorts of patients with diabetes or CVD receiving care in the VA health care system from October 1, 2013, to September 30, 2014 (VA fiscal year [FY] 2014). These patients represent those with diabetes or CVD seeking care in 130 facilities and their associated community-based outpatient clinics within the VA health care system throughout the United States. Details of cohort development have been described previously.5,10,11 For each patient, the team first identified his/her primary care visit in VA FY 2014. If a patient had multiple primary care visits during VA FY 2014, then the most recent visit in the VA system was used as that patient’s index primary care visit. As described previously,5,10,11 patients with CVD were defined as those with a prior history of ischemic heart disease, peripheral artery disease (PAD), or ischemic cerebrovascular disease and were identified using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnoses and procedure, or current procedural terminology codes (see Supplementary Table S1;
Flowchart of the diabetes study cohort identification and exclusions. FY, fiscal year; PCP, primary care provider.
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
PROVIDER TYPE AND HEALTH CARE RESOURCE UTILIZATION
Supplementary Data are available online at www/liebertpub .com/pop) using well-validated methodology described in prior studies.5,11–17 The study team included patients with at least 2 outpatient diagnoses codes or 1 inpatient diagnosis code for unstable angina, or 1 code for myocardial infarction, percutaneous coronary intervention, or coronary artery bypass graft.18 On the basis of prior literature,17 the team also used exclusion criteria to improve specificity for the diagnosis of PAD (see Supplementary Table S2). Using this methodology, an overall positive predictive value of 88% was found for the identification of patients with CVD compared with manual chart review of 100 random CVD patients.5 The study team classified patients as having diabetes if any of the following were documented: 2 outpatient ICD-9-CM diagnoses codes or 1 inpatient ICD-9-CM diagnosis code indicating diabetes (250.xx, 357.2, 366.41), filled prescription for diabetes medications, any fasting glucose >126 mg/dL, hemoglobin A1c (HbA1c) >6.5%, or at least 2 outpatient blood glucose readings >200 mg/dL on 2 different days. 10,19,20 Using this methodology, an overall positive predictive value of 94% was found for identification of patients with diabetes compared with manual chart review of 100 random diabetes patients. As patients with metastatic cancers and those receiving hospice care are not
3
considered candidates for quality measurement,19 these patients were excluded from the study cohort. The study team created separate diabetes and CVD cohorts. Provider assignment for each diabetes or CVD patient was assessed using VA’s primary care management module files. For the diabetes cohort (Fig. 1), the team initially identified patients with diabetes and a primary care visit during VA FY 2014 (n = 1,483,164). The team excluded patients with metastatic cancer or those receiving hospice care (n = 23,539); patients with unknown sex or date of birth or those for whom medication records could not be reconciled (n = 12,343); and those with no assigned primary care provider, or inconsistent primary care provider type, or those assigned to a resident in training as a primary care provider (n = 81,471). Because most quality measures in patients with diabetes apply to those between 40–75 years of age, patients 75 (n = 343,223) also were excluded. The final diabetes cohort included 1,022,588 patients (811,872 receiving care from physicians and 210,716 receiving care from APPs).5,10,11 For the CVD cohort (Fig. 2), the study team identified CVD patients with a primary care visit during VA FY 2014 (n = 1,273,736). The team then excluded patients with metastatic cancer or those receiving hospice care (n = 24,675);
FIG. 2. Flowchart of the cardiovascular disease study cohort identification and exclusions. CVD, cardiovascular disease; FY, fiscal year; PCP, primary care provider.
4
patients with unknown sex or date of birth or those for whom medication records could not be reconciled (n = 7,046); and those with no assigned primary care provider, or inconsistent primary care provider type, or those assigned to a training resident as primary care provider (n = 54,980). The final CVD cohort included 1,187,035 patients (934,950 receiving care from physicians and 252,085 receiving care from APPs).5,9,10
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
Statistical analyses
For each cohort, the study team compared baseline characteristics between patients receiving care from physicians or APPs. These included age, sex, race, history of hypertension, receipt of care at a teaching facility, provider panel size, and the number of primary care visits in 12 months prior to the patient’s most recent primary care visit (index primary care visit) during the study interval. The team also calculated a diagnostic cost group (DCG) relative risk score (RRS) for each patient in the diabetes or CVD cohort. These are calculated using 1CD-9-CM codes available for each patient using structured VA data sets. The diagnoses, which are derived from more than 15 thousand ICD-9-CM codes, are then classified into clinically similar groups called condition categories that are ranked based on severity to form hierarchical condition categories (HCCs). The structured VA data sets currently provide HCCs on each patient receiving care in the VA health care system. HCCs are based on more than 100 disease condition categories. Using HCCs, the study team previously developed a custom regression model of VA costs, which produced the DCG RRS for each patient with a score of 1 being the mean of the population. These RRSs are calculated as the predicted VA cost divided by the mean VA cost for the population of veterans used to fit the regression model and have been used as a surrogate of a patient’s overall illness burden in prior studies.11,14,15 For example, a patient with an RRS of 5 would be expected to be 5 times as costly as an average patient in the population and to have an illness burden 5 times higher than an average patient. Conversely, a patient with an RRS of 0.5 would be expected to be half as costly as an average patient in the population and to have an illness burden half that of an average patient. The DCG RRS was compared among patients receiving care from physicians and APPs. The study team then compared health care resource utilization measures separately for diabetes or CVD patients receiving care from physicians or APPs. For the diabetes cohort, measures included mean number of primary care and specialty care (endocrinology) visits per patient with diabetes in the 12 months prior to the index primary care visit and mean number of lipid panels or HbA1c tests per patient in the past 12 months or within 2 weeks following his/her index primary care visit. The team included a 2-week window following a patient’s index primary care visit for the lipid panel or HbA1c tests to account for the fact that some patients may receive these tests following their primary care visit. Similarly, the study team compared health care resource utilization measures among CVD patients receiving care from physicians or APPs. These measures included mean number of primary care and specialty care (cardiology, vascular surgery) visits per CVD patient in the 12 months prior to the index primary care visit and mean number of lipid panels or stress tests per patient in the past 12 months
VIRANI ET AL.
or within 2 weeks following his/her index primary care visit. Stress test modalities assessed included treadmill stress tests, nuclear stress tests, and stress echocardiograms. The study team compared the mean for each measure for diabetes (or CVD) patients receiving care from physicians or APPs. Because the outcome measures included continuous variables, the team performed linear regression, adjusting for covariates to assess whether there were differences in the utilization of each measure among patients receiving care from physicians or APPs. The covariates among patients with diabetes included age, sex, race (white versus others), history of hypertension, DCG RRS (marker of patient’s illness burden as discussed above), receipt of care at a teaching versus nonteaching facility, and history of CVD. The covariates adjusted among CVD patients included age, sex, race (white versus others), history of hypertension, DCG RRS, receipt of care at a teaching versus nonteaching facility, and history of diabetes. The physician category was used as the referent category. Because the random variance in care could differ significantly secondary to clustering of patients between facilities, the study team further adjusted for clustering of patients at the facility level in the hierarchical regression models, using generalized linear latent and mixed models in Stata. The resultant beta coefficient in the adjusted models can be interpreted as the unit change (increase or decrease) in the dependent (outcome) variable associated with receipt of care from an APP compared with a physician provider. For example, a beta coefficient of -1.0 for the outcome of primary care visit indicates that patients receiving care from APPs in this sample had on average 1 less primary care visit per year than those receiving care from physicians, after adjustment for covariates. In contrast, a beta coefficient of 1.0 for the same outcome indicates that patients receiving care from APPs have on average 1 more primary care visit per year than those receiving care from physicians, after adjustment for covariates. To assess whether health care resource utilization is comparable among patients with high or low overall illness burden, the study team also performed stratified analyses comparing health care resource utilization among diabetes or CVD patients receiving care from physicians or APPs who fell above or below the median for DCG RRS. Finally, the team performed sensitivity analyses after excluding diabetes or CVD patients who switched provider categories in the 12 months prior to their index primary care visit. Analyses were conducted using SAS version 9.1.3 (SAS Institute Inc., Cary, NC) and STATA version 11 (StataCorp, LLC, College Station, TX). The protocol was approved by the Institutional Review Boards at Baylor College of Medicine and the Michael E. DeBakey VA Medical Center. Results
The final analyses included 1,022,588 patients with diabetes (811,872 receiving care from 5530 physicians and 210,716 receiving care from 2208 APPs) and 1,187,035 patients with CVD (934,950 receiving care from 5604 physicians and 252,085 receiving care from 2281 APPs). Tables 1 and 2 describe a comparison of baseline characteristics among diabetes or CVD patients receiving care from physicians or APPs. Patients with diabetes (Table 1) receiving care from physicians were more likely to be males and African Americans, to have a slightly higher overall
PROVIDER TYPE AND HEALTH CARE RESOURCE UTILIZATION
5
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
Table 1. Baseline Characteristics of Patients with Diabetes Assigned to Physicians and Advanced Practice Providers
Age, y, mean/SD Male sex, n (%) Race, n (%) White Black Other Unknown Diagnostic cost group RRS, mean/SD History of CVD, n (%) Hypertension, n(%) Receiving care at a teaching facility, n (%) Provider panel size, mean/SD
Patients assigned to physician providers (n = 811,872)
Patient assigned to advanced practice providers (n = 210,716)
P value
63.5/7.3 780,307 (96.1%)
63.6/7.4 199,754 (94.8%)
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
POPULATION HEALTH MANAGEMENT Volume 00, Number 00, 2017 ª Mary Ann Liebert, Inc. DOI: 10.1089/pop.2017.0090
Health Care Resource Utilization for Outpatient Cardiovascular Disease and Diabetes Care Delivery Among Advanced Practice Providers and Physician Providers in Primary Care Salim S. Virani, MD, PhD,1–4 Julia M. Akeroyd, MPH,1 David J. Ramsey, PhD,1 Anita Deswal, MD, MPH,2,5 Khurram Nasir, MD,6,7 Suja S. Rajan, PhD, MS,8 Christie M. Ballantyne, MD,3,4 and Laura A. Petersen MD, MPH1
Abstract
Although effectiveness of diabetes or cardiovascular disease (CVD) care delivery between physicians and advanced practice providers (APPs) has been shown to be comparable, health care resource utilization between these 2 provider types in primary care is unknown. This study compared health care resource utilization between patients with diabetes or CVD receiving care from APPs or physicians. Diabetes (n = 1,022,588) or CVD (n = 1,187,035) patients with a primary care visit between October 2013 and September 2014 in 130 Veterans Affairs facilities were identified. Using hierarchical regression adjusting for covariates including patient illness burden, the authors compared number of primary or specialty care visits and number of lipid panels and hemoglobinA1c (HbA1c) tests among diabetes patients, and number of primary or specialty care visits and number of lipid panels and cardiac stress tests among CVD patients receiving care from physicians and APPs. Physicians had significantly larger patient panels compared with APPs. In adjusted analyses, diabetes patients receiving care from APPs received fewer primary and specialty care visits and a greater number of lipid panels and HbA1c tests compared with patients receiving care from physicians. CVD patients receiving care from APPs received more frequent lipid testing and fewer primary and specialty care visits compared with those receiving care from physicians, with no differences in the number of stress tests. Most of these differences, although statistically significant, were numerically small. Health care resource utilization among diabetes or CVD patients receiving care from APPs or physicians appears comparable, although physicians work with larger patient panels. Keywords: physicians, advanced practice providers, health care resource utilization, diabetes, cardiovascular disease debate, the system would be further strained if the 20 million Americans who obtained health insurance under the ACA retained some form of health care coverage. A greater use of advanced practice providers (APPs) (nurse practitioners [NPs] or physician assistants) has been proposed to address these shortages, especially for chronic disease care delivery.2
Introduction
T
he Association of American Medical Colleges estimates that there will be a shortage of 45,000 primary care physicians by 2020 and 65,000 by 2025.1 Although the future of the Affordable Care Act (ACA) is currently under
1 From the Health Policy, Quality & Informatics Program, Michael E. DeBakey Veterans Affairs Medical Center Health Services Research and Development Center for Innovations; and Section of Health Services Research, Department of Medicine, Baylor College of Medicine, Houston, Texas. 2 Section of Cardiology, Michael E. DeBakey Veterans Affairs Medical Center, Houston, Texas. 3 Section of Cardiovascular Research, Department of Medicine, Baylor College of Medicine, Houston, Texas. 4 Center for Cardiovascular Disease Prevention, Methodist DeBakey Heart and Vascular Center, Houston, Texas. 5 Section of Cardiology, Department of Medicine, Baylor College of Medicine, Houston, Texas. 6 Center for Healthcare Advancement & Outcomes, Baptist Health South Florida, Miami, Florida. 7 The Johns Hopkins Ciccarone Center for the Prevention of Heart Disease, Baltimore, Maryland. 8 Division of Management, Policy and Community Heath, School of Public Health, University of Texas—Health Science Center at Houston, Houston, Texas.
1
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
2
VIRANI ET AL.
In prior studies performed in small practices3 and recent studies performed in large national cohorts of patients,4,5 it has been shown that the effectiveness of outpatient delivery of cardiovascular disease (CVD) and diabetes care is comparable among patients receiving care from physicians or APPs.5 Although effectiveness of routine outpatient CVD and diabetes care delivery has shown to be comparable between physicians and APPs,5 it is not clear whether this comes at the price of increased health care resource utilization by APPs compared with physicians. Prior studies on the topic were performed in small academic settings6–8 and did not compare these findings in large health care systems in a ‘‘real-world’’ primary care setting. Therefore, the aim of the present analyses was to compare health care resource utilization between patients receiving care from physicians or APPs. Using the Department of Veterans Affair’ (VA) robust administrative and clinical data set of national cohorts of patients with diabetes or CVD, this study assessed whether there were any clinically meaningful differences in health care resource utilization between patients receiving care from physicians or APPs. Diabetes and CVD were chosen as 2 representative chronic disease conditions, given their high prevalence, high cost, and their significant contribution to overall morbidity and mortality.9
FIG. 1.
Methods Cohort development and study population
Using VA’s administrative and clinical data sets, the study team created 2 separate national cohorts of patients with diabetes or CVD receiving care in the VA health care system from October 1, 2013, to September 30, 2014 (VA fiscal year [FY] 2014). These patients represent those with diabetes or CVD seeking care in 130 facilities and their associated community-based outpatient clinics within the VA health care system throughout the United States. Details of cohort development have been described previously.5,10,11 For each patient, the team first identified his/her primary care visit in VA FY 2014. If a patient had multiple primary care visits during VA FY 2014, then the most recent visit in the VA system was used as that patient’s index primary care visit. As described previously,5,10,11 patients with CVD were defined as those with a prior history of ischemic heart disease, peripheral artery disease (PAD), or ischemic cerebrovascular disease and were identified using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnoses and procedure, or current procedural terminology codes (see Supplementary Table S1;
Flowchart of the diabetes study cohort identification and exclusions. FY, fiscal year; PCP, primary care provider.
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
PROVIDER TYPE AND HEALTH CARE RESOURCE UTILIZATION
Supplementary Data are available online at www/liebertpub .com/pop) using well-validated methodology described in prior studies.5,11–17 The study team included patients with at least 2 outpatient diagnoses codes or 1 inpatient diagnosis code for unstable angina, or 1 code for myocardial infarction, percutaneous coronary intervention, or coronary artery bypass graft.18 On the basis of prior literature,17 the team also used exclusion criteria to improve specificity for the diagnosis of PAD (see Supplementary Table S2). Using this methodology, an overall positive predictive value of 88% was found for the identification of patients with CVD compared with manual chart review of 100 random CVD patients.5 The study team classified patients as having diabetes if any of the following were documented: 2 outpatient ICD-9-CM diagnoses codes or 1 inpatient ICD-9-CM diagnosis code indicating diabetes (250.xx, 357.2, 366.41), filled prescription for diabetes medications, any fasting glucose >126 mg/dL, hemoglobin A1c (HbA1c) >6.5%, or at least 2 outpatient blood glucose readings >200 mg/dL on 2 different days. 10,19,20 Using this methodology, an overall positive predictive value of 94% was found for identification of patients with diabetes compared with manual chart review of 100 random diabetes patients. As patients with metastatic cancers and those receiving hospice care are not
3
considered candidates for quality measurement,19 these patients were excluded from the study cohort. The study team created separate diabetes and CVD cohorts. Provider assignment for each diabetes or CVD patient was assessed using VA’s primary care management module files. For the diabetes cohort (Fig. 1), the team initially identified patients with diabetes and a primary care visit during VA FY 2014 (n = 1,483,164). The team excluded patients with metastatic cancer or those receiving hospice care (n = 23,539); patients with unknown sex or date of birth or those for whom medication records could not be reconciled (n = 12,343); and those with no assigned primary care provider, or inconsistent primary care provider type, or those assigned to a resident in training as a primary care provider (n = 81,471). Because most quality measures in patients with diabetes apply to those between 40–75 years of age, patients 75 (n = 343,223) also were excluded. The final diabetes cohort included 1,022,588 patients (811,872 receiving care from physicians and 210,716 receiving care from APPs).5,10,11 For the CVD cohort (Fig. 2), the study team identified CVD patients with a primary care visit during VA FY 2014 (n = 1,273,736). The team then excluded patients with metastatic cancer or those receiving hospice care (n = 24,675);
FIG. 2. Flowchart of the cardiovascular disease study cohort identification and exclusions. CVD, cardiovascular disease; FY, fiscal year; PCP, primary care provider.
4
patients with unknown sex or date of birth or those for whom medication records could not be reconciled (n = 7,046); and those with no assigned primary care provider, or inconsistent primary care provider type, or those assigned to a training resident as primary care provider (n = 54,980). The final CVD cohort included 1,187,035 patients (934,950 receiving care from physicians and 252,085 receiving care from APPs).5,9,10
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
Statistical analyses
For each cohort, the study team compared baseline characteristics between patients receiving care from physicians or APPs. These included age, sex, race, history of hypertension, receipt of care at a teaching facility, provider panel size, and the number of primary care visits in 12 months prior to the patient’s most recent primary care visit (index primary care visit) during the study interval. The team also calculated a diagnostic cost group (DCG) relative risk score (RRS) for each patient in the diabetes or CVD cohort. These are calculated using 1CD-9-CM codes available for each patient using structured VA data sets. The diagnoses, which are derived from more than 15 thousand ICD-9-CM codes, are then classified into clinically similar groups called condition categories that are ranked based on severity to form hierarchical condition categories (HCCs). The structured VA data sets currently provide HCCs on each patient receiving care in the VA health care system. HCCs are based on more than 100 disease condition categories. Using HCCs, the study team previously developed a custom regression model of VA costs, which produced the DCG RRS for each patient with a score of 1 being the mean of the population. These RRSs are calculated as the predicted VA cost divided by the mean VA cost for the population of veterans used to fit the regression model and have been used as a surrogate of a patient’s overall illness burden in prior studies.11,14,15 For example, a patient with an RRS of 5 would be expected to be 5 times as costly as an average patient in the population and to have an illness burden 5 times higher than an average patient. Conversely, a patient with an RRS of 0.5 would be expected to be half as costly as an average patient in the population and to have an illness burden half that of an average patient. The DCG RRS was compared among patients receiving care from physicians and APPs. The study team then compared health care resource utilization measures separately for diabetes or CVD patients receiving care from physicians or APPs. For the diabetes cohort, measures included mean number of primary care and specialty care (endocrinology) visits per patient with diabetes in the 12 months prior to the index primary care visit and mean number of lipid panels or HbA1c tests per patient in the past 12 months or within 2 weeks following his/her index primary care visit. The team included a 2-week window following a patient’s index primary care visit for the lipid panel or HbA1c tests to account for the fact that some patients may receive these tests following their primary care visit. Similarly, the study team compared health care resource utilization measures among CVD patients receiving care from physicians or APPs. These measures included mean number of primary care and specialty care (cardiology, vascular surgery) visits per CVD patient in the 12 months prior to the index primary care visit and mean number of lipid panels or stress tests per patient in the past 12 months
VIRANI ET AL.
or within 2 weeks following his/her index primary care visit. Stress test modalities assessed included treadmill stress tests, nuclear stress tests, and stress echocardiograms. The study team compared the mean for each measure for diabetes (or CVD) patients receiving care from physicians or APPs. Because the outcome measures included continuous variables, the team performed linear regression, adjusting for covariates to assess whether there were differences in the utilization of each measure among patients receiving care from physicians or APPs. The covariates among patients with diabetes included age, sex, race (white versus others), history of hypertension, DCG RRS (marker of patient’s illness burden as discussed above), receipt of care at a teaching versus nonteaching facility, and history of CVD. The covariates adjusted among CVD patients included age, sex, race (white versus others), history of hypertension, DCG RRS, receipt of care at a teaching versus nonteaching facility, and history of diabetes. The physician category was used as the referent category. Because the random variance in care could differ significantly secondary to clustering of patients between facilities, the study team further adjusted for clustering of patients at the facility level in the hierarchical regression models, using generalized linear latent and mixed models in Stata. The resultant beta coefficient in the adjusted models can be interpreted as the unit change (increase or decrease) in the dependent (outcome) variable associated with receipt of care from an APP compared with a physician provider. For example, a beta coefficient of -1.0 for the outcome of primary care visit indicates that patients receiving care from APPs in this sample had on average 1 less primary care visit per year than those receiving care from physicians, after adjustment for covariates. In contrast, a beta coefficient of 1.0 for the same outcome indicates that patients receiving care from APPs have on average 1 more primary care visit per year than those receiving care from physicians, after adjustment for covariates. To assess whether health care resource utilization is comparable among patients with high or low overall illness burden, the study team also performed stratified analyses comparing health care resource utilization among diabetes or CVD patients receiving care from physicians or APPs who fell above or below the median for DCG RRS. Finally, the team performed sensitivity analyses after excluding diabetes or CVD patients who switched provider categories in the 12 months prior to their index primary care visit. Analyses were conducted using SAS version 9.1.3 (SAS Institute Inc., Cary, NC) and STATA version 11 (StataCorp, LLC, College Station, TX). The protocol was approved by the Institutional Review Boards at Baylor College of Medicine and the Michael E. DeBakey VA Medical Center. Results
The final analyses included 1,022,588 patients with diabetes (811,872 receiving care from 5530 physicians and 210,716 receiving care from 2208 APPs) and 1,187,035 patients with CVD (934,950 receiving care from 5604 physicians and 252,085 receiving care from 2281 APPs). Tables 1 and 2 describe a comparison of baseline characteristics among diabetes or CVD patients receiving care from physicians or APPs. Patients with diabetes (Table 1) receiving care from physicians were more likely to be males and African Americans, to have a slightly higher overall
PROVIDER TYPE AND HEALTH CARE RESOURCE UTILIZATION
5
Downloaded by Gothenburg University Library from online.liebertpub.com at 10/18/17. For personal use only.
Table 1. Baseline Characteristics of Patients with Diabetes Assigned to Physicians and Advanced Practice Providers
Age, y, mean/SD Male sex, n (%) Race, n (%) White Black Other Unknown Diagnostic cost group RRS, mean/SD History of CVD, n (%) Hypertension, n(%) Receiving care at a teaching facility, n (%) Provider panel size, mean/SD
Patients assigned to physician providers (n = 811,872)
Patient assigned to advanced practice providers (n = 210,716)
P value
63.5/7.3 780,307 (96.1%)
63.6/7.4 199,754 (94.8%)
