Original Research

Specific transesophageal echocardiography (TEE) findings associated with stroke include cardiac thrombi (particularly left atrial appendage [LAA]), left atrial spontaneous echo contrast, interatrial septal anomalies (particularly patent foramen ovale [PFO]), and atheromatous disease of the aorta. In younger patients (aged <50 years) with stroke of uncertain etiology, TEE is often recommended because of reported higher yield than transthoracic echocardiogram (TTE), particularly in detecting PFO or atrial septal aneurysm (ASA).[1]

Aside from oral anticoagulation in patients with an intracardiac thrombus, current guidelines and scientific evidence do not support specific therapeutic interventions for the other TEE findings. For example, the most effective therapy for stroke prevention with findings of aortic arch plaque remains uncertain. In addition, the very rare patient presenting with stroke from a cardiac tumor, which is generally visible on TTE, might benefit from surgical removal.[2]

We sought to examine the benefit of performing TEE after a normal TTE in patients over age 50 years admitted with a stroke of uncertain etiology. We hypothesized that there would be minimal change in management based on TEE findings after a normal TTE in older patients hospitalized with an unexplained stroke.

METHODS

Over a 4‐year period from 2009 to 2012, all patients over the age of 50 years admitted to our community‐based teaching hospital with a primary diagnosis of ischemic stroke were identified and retrospectively screened by review of our institutional echocardiography database during this time period. Stroke diagnosis had to be confirmed with acute or subacute ischemia on brain magnetic resonance imaging. Patients with an indication for anticoagulation or who had a known history of atrial fibrillation or flutter were excluded. Patients were monitored with continuous telemetry during hospital admission and were also excluded if they developed atrial fibrillation or flutter after admission. Additionally, patients were excluded if a neurologist‐directed evaluation revealed another etiology for the stroke.

A TTE acquired in all patients was performed according to Intersocietal Commission for the Accreditation of Echocardiography Laboratories standards and included 2‐dimensional, color Doppler, continuous wave, and pulse wave data. Images were obtained in the parasternal long and short axis, apical 4‐chamber, 2‐chamber, and long axis views. An abnormal TTE was defined as a study with a prosthetic valve, abnormal left ventricular (LV) systolic function, an intracardiac mass, intracardiac shunt, or severe valvular heart disease, as these significant findings may explain stroke.

Standardized TEE images were obtained with midesophageal 4‐chamber, mitral commissural, 2‐chamber, long axis, ascending aorta long axis, aortic valve short axis, right ventricular inflow‐outflow, and bicaval views. Detailed multiplanar evaluation of the LAA was performed. If no interatrial shunt was visualized with color flow Doppler in the bicaval view, agitated intravenous saline was administered for further evaluation. Additional standard images were obtained of the descending aorta and aortic arch in the short and long axis. Transgastric images were obtained when feasible or necessary.

The study was submitted to our institutional review board. As no patient identifiers were stored, and we used previously existing data from an institutional echocardiography database to conduct the study, it was determined to be exempt.

Statistical analysis was performed by recording the prevalence of each potential cardiac source of embolism.

RESULTS

Of the 853 consecutive patients screened, 456 were excluded because of atrial fibrillation, atrial flutter, or another etiology of stroke. An additional 134 patients were excluded with an abnormal TTE or if a TEE was not performed. The remaining 263 patients were analyzed based on TEE findings (Figure 1).

Figure 1

The mean age was 66.7 years (range, 5091 years), and 42.5% were female. A possible etiology of stroke (Table 1) discovered included complex plaque of the ascending aorta or arch 44/263 (16.7%), PFO 18/263 (6.8%), atrial septal aneurysm 25/263 (9.5%), and both ASA and PFO in 11/263 (4.2%), and spontaneous contrast was seen in the left atrium or LAA in 13/263 (4.9%) patients. One patient had a thrombus in the LAA for which anticoagulation was prescribed. No other intracardiac masses were identified.

Overall, 42.6% of patients had a TEE finding which could explain the etiology of stroke or transient ischemic attack (TIA), but only 1 patient (0.4%) had a finding that changed therapy. Follow‐up was available at 6 months for 85 patients, and 13 (15%) of these patients had been discovered to develop atrial fibrillation in the interim.

DISCUSSION

Our study retrospectively analyzed the utility of TEE in patients over age 50 years admitted with ischemic stroke without a clear etiology. We found that TEE provides significant incremental diagnostic benefit as compared to TTE in identifying a possible etiology of stroke in these patients. This is consistent with prior studies showing a high diagnostic yield of TEE in patient with ischemic stroke of uncertain etiology.[3] However, in our study, based on current guidelines, virtually none of these findings directly altered patient management.

The 2014 guidelines for secondary stroke prevention recommend antiplatelet and statin therapy (in addition to lifestyle modification, smoking cessation, and blood glucose and blood pressure control) as a standard medical regimen in patients with stroke or TIA of uncertain etiology. The finding of aortic arch atheroma does not warrant supplementary treatment in addition to an antiplatelet and statin according to current guidelines. Atherosclerosis of the aortic arch is an important source of cerebral embolism, particularly in cases where plaque is >4 mm in size.[4] A recent study by Amarenco et al., comparing efficacy of combined antiplatelet therapy (clopidogrel and aspirin) to warfarin in recurrent stroke prevention in patients with >4 mm aortic arch plaque, showed nonsignificant reduction in rate of recurrent stroke with dual antiplatelet therapy.[5] However, optimal therapy for these patients still remains uncertain beyond standard stroke‐prevention treatment. Although there are emerging data on therapeutic options in patients with complex atheroma, there is currently no specific guideline‐recommended therapy or consensus among stroke neurologists. Potentially, if an individual practitioner had a strong feeling on therapeutic modifications based on the presence of complex aortic arch atheroma, the TEE would have value to their patient. However, in our study, which had a prevalence of 16.8% of complex plaque of the ascending aorta or arch, there were no therapeutic changes based on this finding. This reinforces the limited value of this test that we observed in our study population.

Anticoagulation has not been shown to be superior to aspirin in patients with PFO (with or without ASA), and recent studies showed no benefit of procedural PFO closure compared to best medical management for stroke prevention (Randomized Evaluation of Recurrent Stroke Comparing PFO Closure to Established Current Standard of Care Treatment [RESPECT], Evaluation of the STARFlex Septal Closure System in Patients with a Stroke and/or Transient Ischemic Attack due to Presumed Paradoxical Embolism through a Patent Foramen Ovale [CLOSURE I]).[6, 7] However, a patient with a PFO and deep vein thrombosis would benefit from anticoagulation and consideration of PFO closure.[8] This rare entity could be excluded with a simple lower extremity duplex without the need for a TEE, which does come with a small risk of complications related to anesthesia and local oropharyngeal trauma as well as discomfort to the patient and increased cost. Spontaneous echo contrast is not an independent indication for anticoagulation. If spontaneous contrast were associated with mitral stenosis and an embolic event, then anticoagulation would be indicated.[9] Mitral stenosis is easily diagnosed with TTE.

LAA or left atrial thrombus is the predominant finding exclusive to TEE that would change management for secondary stroke prevention, specifically anticoagulation. Fifteen studies representing over 3000 patients in a 2014 meta‐analysis reported the prevalence of left atrial or LAA thrombus in patients aged 55 years with a cryptogenic stroke to be 4%, with a range in the studies of 0% to 21.2%.[3] The wide range of prevalence of this finding is likely related to the prevalence of known atrial arrhythmias or structural heart disease in the population of patients included in the analysis. Left atrial or LAA thrombus in the absence of systolic dysfunction, severe valve disease, or known atrial fibrillation is exceedingly uncommon (0.3%).[10] It is likely that the few patients with left atrial or LAA thrombus without 1 of these conditions probably has undiagnosed paroxysmal atrial fibrillation. In previous studies that showed a high prevalence of left atrial or LAA thrombus, there was no mention of the presence or absence of LV dysfunction or severe valve disease in patients with left atrial or LAA thrombus. Additionally, these studies only required a 12‐lead electrocardiogram or did not specify the presence or duration of continuous rhythm monitoring.[11, 12, 13, 14] Several of the studies with high incidence of left atrial or LAA thrombus specifically stated that some of these patients were known to have atrial fibrillation.[11, 13]

Approximately 8% of patients admitted with stroke are found to have atrial fibrillation only after admission with continuous electrocardiogram monitoring. The detection rate is nearly half if monitoring is limited to 24 hours instead of several days. Overall, detection rates of atrial fibrillation following stroke are relatively low during initial hospitalization.[15] More intense monitoring for atrial fibrillation in patients with a stroke of uncertain etiology with the use of a subcutaneous implantable cardiac monitor increases the detection rate to 12.4% at 1 year, and increases with longer monitoring time.[16] Therefore, identification of older stroke patients without significant stroke risk factors may be candidates for longer‐term cardiac monitoring to increase yield for detection of atrial fibrillation. Currently, continuous electrocardiographic monitoring of patients for the duration of their hospitalization and up to 30 days afterward is recommended.[8]

Our study differs from prior studies that showed a much higher prevalence of LAA or left atrial thrombus in 2 important ways. Patients with severe valve disease or LV dysfunction were excluded on the basis of TTE. Additionally, our patients underwent continuous electrocardiographic monitoring for the duration of their hospitalization and were excluded with a prior history or newly discovered atrial fibrillation or flutter. Our intention was to examine the value of adding TEE when no other etiology of stroke was identified. Value can be defined as healthcare outcomes achieved per dollar spent. Our study was not designed to look at long‐term outcomes; rather, we used immediate change in patient management as a surrogate.

There are several limitations to our study that must be noted. This was a single‐center study potentially creating a bias as less stringent selection of patients undergoing TEE may be the practice at other institutions. This analysis was retrospective; therefore, there may have been bias as to which patients were selected to undergo TEE. Additionally, stroke subtype was not specified, and the pretest probability of a cardioembolic source differs based on subtype. Last, we focused this study on immediate changes in clinical management prompted by TEE results, and did not assess patient perceptions of TEE value related to enhanced knowledge about the etiology of their stroke; this area represents an opportunity for further research.

CONCLUSIONS

TEE provides a substantial increase in possible explanation of stroke etiology in patients over age 50 years admitted with a stroke of uncertain cause and a normal TTE. However, there is minimal incremental value in regard to change in therapeutic management in these patients. In a time of increased focus on providing cost effective healthcare, our findings suggest that the need for TEE in this stroke population should be more closely examined.

Disclosure: Nothing to report.

Specific transesophageal echocardiography (TEE) findings associated with stroke include cardiac thrombi (particularly left atrial appendage [LAA]), left atrial spontaneous echo contrast, interatrial septal anomalies (particularly patent foramen ovale [PFO]), and atheromatous disease of the aorta. In younger patients (aged <50 years) with stroke of uncertain etiology, TEE is often recommended because of reported higher yield than transthoracic echocardiogram (TTE), particularly in detecting PFO or atrial septal aneurysm (ASA).[1]

Aside from oral anticoagulation in patients with an intracardiac thrombus, current guidelines and scientific evidence do not support specific therapeutic interventions for the other TEE findings. For example, the most effective therapy for stroke prevention with findings of aortic arch plaque remains uncertain. In addition, the very rare patient presenting with stroke from a cardiac tumor, which is generally visible on TTE, might benefit from surgical removal.[2]

We sought to examine the benefit of performing TEE after a normal TTE in patients over age 50 years admitted with a stroke of uncertain etiology. We hypothesized that there would be minimal change in management based on TEE findings after a normal TTE in older patients hospitalized with an unexplained stroke.

METHODS

Over a 4‐year period from 2009 to 2012, all patients over the age of 50 years admitted to our community‐based teaching hospital with a primary diagnosis of ischemic stroke were identified and retrospectively screened by review of our institutional echocardiography database during this time period. Stroke diagnosis had to be confirmed with acute or subacute ischemia on brain magnetic resonance imaging. Patients with an indication for anticoagulation or who had a known history of atrial fibrillation or flutter were excluded. Patients were monitored with continuous telemetry during hospital admission and were also excluded if they developed atrial fibrillation or flutter after admission. Additionally, patients were excluded if a neurologist‐directed evaluation revealed another etiology for the stroke.

A TTE acquired in all patients was performed according to Intersocietal Commission for the Accreditation of Echocardiography Laboratories standards and included 2‐dimensional, color Doppler, continuous wave, and pulse wave data. Images were obtained in the parasternal long and short axis, apical 4‐chamber, 2‐chamber, and long axis views. An abnormal TTE was defined as a study with a prosthetic valve, abnormal left ventricular (LV) systolic function, an intracardiac mass, intracardiac shunt, or severe valvular heart disease, as these significant findings may explain stroke.

Standardized TEE images were obtained with midesophageal 4‐chamber, mitral commissural, 2‐chamber, long axis, ascending aorta long axis, aortic valve short axis, right ventricular inflow‐outflow, and bicaval views. Detailed multiplanar evaluation of the LAA was performed. If no interatrial shunt was visualized with color flow Doppler in the bicaval view, agitated intravenous saline was administered for further evaluation. Additional standard images were obtained of the descending aorta and aortic arch in the short and long axis. Transgastric images were obtained when feasible or necessary.

The study was submitted to our institutional review board. As no patient identifiers were stored, and we used previously existing data from an institutional echocardiography database to conduct the study, it was determined to be exempt.

Statistical analysis was performed by recording the prevalence of each potential cardiac source of embolism.

RESULTS

Of the 853 consecutive patients screened, 456 were excluded because of atrial fibrillation, atrial flutter, or another etiology of stroke. An additional 134 patients were excluded with an abnormal TTE or if a TEE was not performed. The remaining 263 patients were analyzed based on TEE findings (Figure 1).

Figure 1

The mean age was 66.7 years (range, 5091 years), and 42.5% were female. A possible etiology of stroke (Table 1) discovered included complex plaque of the ascending aorta or arch 44/263 (16.7%), PFO 18/263 (6.8%), atrial septal aneurysm 25/263 (9.5%), and both ASA and PFO in 11/263 (4.2%), and spontaneous contrast was seen in the left atrium or LAA in 13/263 (4.9%) patients. One patient had a thrombus in the LAA for which anticoagulation was prescribed. No other intracardiac masses were identified.

Overall, 42.6% of patients had a TEE finding which could explain the etiology of stroke or transient ischemic attack (TIA), but only 1 patient (0.4%) had a finding that changed therapy. Follow‐up was available at 6 months for 85 patients, and 13 (15%) of these patients had been discovered to develop atrial fibrillation in the interim.

DISCUSSION

Our study retrospectively analyzed the utility of TEE in patients over age 50 years admitted with ischemic stroke without a clear etiology. We found that TEE provides significant incremental diagnostic benefit as compared to TTE in identifying a possible etiology of stroke in these patients. This is consistent with prior studies showing a high diagnostic yield of TEE in patient with ischemic stroke of uncertain etiology.[3] However, in our study, based on current guidelines, virtually none of these findings directly altered patient management.

The 2014 guidelines for secondary stroke prevention recommend antiplatelet and statin therapy (in addition to lifestyle modification, smoking cessation, and blood glucose and blood pressure control) as a standard medical regimen in patients with stroke or TIA of uncertain etiology. The finding of aortic arch atheroma does not warrant supplementary treatment in addition to an antiplatelet and statin according to current guidelines. Atherosclerosis of the aortic arch is an important source of cerebral embolism, particularly in cases where plaque is >4 mm in size.[4] A recent study by Amarenco et al., comparing efficacy of combined antiplatelet therapy (clopidogrel and aspirin) to warfarin in recurrent stroke prevention in patients with >4 mm aortic arch plaque, showed nonsignificant reduction in rate of recurrent stroke with dual antiplatelet therapy.[5] However, optimal therapy for these patients still remains uncertain beyond standard stroke‐prevention treatment. Although there are emerging data on therapeutic options in patients with complex atheroma, there is currently no specific guideline‐recommended therapy or consensus among stroke neurologists. Potentially, if an individual practitioner had a strong feeling on therapeutic modifications based on the presence of complex aortic arch atheroma, the TEE would have value to their patient. However, in our study, which had a prevalence of 16.8% of complex plaque of the ascending aorta or arch, there were no therapeutic changes based on this finding. This reinforces the limited value of this test that we observed in our study population.

Anticoagulation has not been shown to be superior to aspirin in patients with PFO (with or without ASA), and recent studies showed no benefit of procedural PFO closure compared to best medical management for stroke prevention (Randomized Evaluation of Recurrent Stroke Comparing PFO Closure to Established Current Standard of Care Treatment [RESPECT], Evaluation of the STARFlex Septal Closure System in Patients with a Stroke and/or Transient Ischemic Attack due to Presumed Paradoxical Embolism through a Patent Foramen Ovale [CLOSURE I]).[6, 7] However, a patient with a PFO and deep vein thrombosis would benefit from anticoagulation and consideration of PFO closure.[8] This rare entity could be excluded with a simple lower extremity duplex without the need for a TEE, which does come with a small risk of complications related to anesthesia and local oropharyngeal trauma as well as discomfort to the patient and increased cost. Spontaneous echo contrast is not an independent indication for anticoagulation. If spontaneous contrast were associated with mitral stenosis and an embolic event, then anticoagulation would be indicated.[9] Mitral stenosis is easily diagnosed with TTE.

LAA or left atrial thrombus is the predominant finding exclusive to TEE that would change management for secondary stroke prevention, specifically anticoagulation. Fifteen studies representing over 3000 patients in a 2014 meta‐analysis reported the prevalence of left atrial or LAA thrombus in patients aged 55 years with a cryptogenic stroke to be 4%, with a range in the studies of 0% to 21.2%.[3] The wide range of prevalence of this finding is likely related to the prevalence of known atrial arrhythmias or structural heart disease in the population of patients included in the analysis. Left atrial or LAA thrombus in the absence of systolic dysfunction, severe valve disease, or known atrial fibrillation is exceedingly uncommon (0.3%).[10] It is likely that the few patients with left atrial or LAA thrombus without 1 of these conditions probably has undiagnosed paroxysmal atrial fibrillation. In previous studies that showed a high prevalence of left atrial or LAA thrombus, there was no mention of the presence or absence of LV dysfunction or severe valve disease in patients with left atrial or LAA thrombus. Additionally, these studies only required a 12‐lead electrocardiogram or did not specify the presence or duration of continuous rhythm monitoring.[11, 12, 13, 14] Several of the studies with high incidence of left atrial or LAA thrombus specifically stated that some of these patients were known to have atrial fibrillation.[11, 13]

Approximately 8% of patients admitted with stroke are found to have atrial fibrillation only after admission with continuous electrocardiogram monitoring. The detection rate is nearly half if monitoring is limited to 24 hours instead of several days. Overall, detection rates of atrial fibrillation following stroke are relatively low during initial hospitalization.[15] More intense monitoring for atrial fibrillation in patients with a stroke of uncertain etiology with the use of a subcutaneous implantable cardiac monitor increases the detection rate to 12.4% at 1 year, and increases with longer monitoring time.[16] Therefore, identification of older stroke patients without significant stroke risk factors may be candidates for longer‐term cardiac monitoring to increase yield for detection of atrial fibrillation. Currently, continuous electrocardiographic monitoring of patients for the duration of their hospitalization and up to 30 days afterward is recommended.[8]

Our study differs from prior studies that showed a much higher prevalence of LAA or left atrial thrombus in 2 important ways. Patients with severe valve disease or LV dysfunction were excluded on the basis of TTE. Additionally, our patients underwent continuous electrocardiographic monitoring for the duration of their hospitalization and were excluded with a prior history or newly discovered atrial fibrillation or flutter. Our intention was to examine the value of adding TEE when no other etiology of stroke was identified. Value can be defined as healthcare outcomes achieved per dollar spent. Our study was not designed to look at long‐term outcomes; rather, we used immediate change in patient management as a surrogate.

There are several limitations to our study that must be noted. This was a single‐center study potentially creating a bias as less stringent selection of patients undergoing TEE may be the practice at other institutions. This analysis was retrospective; therefore, there may have been bias as to which patients were selected to undergo TEE. Additionally, stroke subtype was not specified, and the pretest probability of a cardioembolic source differs based on subtype. Last, we focused this study on immediate changes in clinical management prompted by TEE results, and did not assess patient perceptions of TEE value related to enhanced knowledge about the etiology of their stroke; this area represents an opportunity for further research.

CONCLUSIONS

TEE provides a substantial increase in possible explanation of stroke etiology in patients over age 50 years admitted with a stroke of uncertain cause and a normal TTE. However, there is minimal incremental value in regard to change in therapeutic management in these patients. In a time of increased focus on providing cost effective healthcare, our findings suggest that the need for TEE in this stroke population should be more closely examined.

Disclosure: Nothing to report.

Specific transesophageal echocardiography (TEE) findings associated with stroke include cardiac thrombi (particularly left atrial appendage [LAA]), left atrial spontaneous echo contrast, interatrial septal anomalies (particularly patent foramen ovale [PFO]), and atheromatous disease of the aorta. In younger patients (aged <50 years) with stroke of uncertain etiology, TEE is often recommended because of reported higher yield than transthoracic echocardiogram (TTE), particularly in detecting PFO or atrial septal aneurysm (ASA).[1]

Aside from oral anticoagulation in patients with an intracardiac thrombus, current guidelines and scientific evidence do not support specific therapeutic interventions for the other TEE findings. For example, the most effective therapy for stroke prevention with findings of aortic arch plaque remains uncertain. In addition, the very rare patient presenting with stroke from a cardiac tumor, which is generally visible on TTE, might benefit from surgical removal.[2]

We sought to examine the benefit of performing TEE after a normal TTE in patients over age 50 years admitted with a stroke of uncertain etiology. We hypothesized that there would be minimal change in management based on TEE findings after a normal TTE in older patients hospitalized with an unexplained stroke.

METHODS

Over a 4‐year period from 2009 to 2012, all patients over the age of 50 years admitted to our community‐based teaching hospital with a primary diagnosis of ischemic stroke were identified and retrospectively screened by review of our institutional echocardiography database during this time period. Stroke diagnosis had to be confirmed with acute or subacute ischemia on brain magnetic resonance imaging. Patients with an indication for anticoagulation or who had a known history of atrial fibrillation or flutter were excluded. Patients were monitored with continuous telemetry during hospital admission and were also excluded if they developed atrial fibrillation or flutter after admission. Additionally, patients were excluded if a neurologist‐directed evaluation revealed another etiology for the stroke.

A TTE acquired in all patients was performed according to Intersocietal Commission for the Accreditation of Echocardiography Laboratories standards and included 2‐dimensional, color Doppler, continuous wave, and pulse wave data. Images were obtained in the parasternal long and short axis, apical 4‐chamber, 2‐chamber, and long axis views. An abnormal TTE was defined as a study with a prosthetic valve, abnormal left ventricular (LV) systolic function, an intracardiac mass, intracardiac shunt, or severe valvular heart disease, as these significant findings may explain stroke.

Standardized TEE images were obtained with midesophageal 4‐chamber, mitral commissural, 2‐chamber, long axis, ascending aorta long axis, aortic valve short axis, right ventricular inflow‐outflow, and bicaval views. Detailed multiplanar evaluation of the LAA was performed. If no interatrial shunt was visualized with color flow Doppler in the bicaval view, agitated intravenous saline was administered for further evaluation. Additional standard images were obtained of the descending aorta and aortic arch in the short and long axis. Transgastric images were obtained when feasible or necessary.

The study was submitted to our institutional review board. As no patient identifiers were stored, and we used previously existing data from an institutional echocardiography database to conduct the study, it was determined to be exempt.

Statistical analysis was performed by recording the prevalence of each potential cardiac source of embolism.

RESULTS

Of the 853 consecutive patients screened, 456 were excluded because of atrial fibrillation, atrial flutter, or another etiology of stroke. An additional 134 patients were excluded with an abnormal TTE or if a TEE was not performed. The remaining 263 patients were analyzed based on TEE findings (Figure 1).

Figure 1

The mean age was 66.7 years (range, 5091 years), and 42.5% were female. A possible etiology of stroke (Table 1) discovered included complex plaque of the ascending aorta or arch 44/263 (16.7%), PFO 18/263 (6.8%), atrial septal aneurysm 25/263 (9.5%), and both ASA and PFO in 11/263 (4.2%), and spontaneous contrast was seen in the left atrium or LAA in 13/263 (4.9%) patients. One patient had a thrombus in the LAA for which anticoagulation was prescribed. No other intracardiac masses were identified.

Overall, 42.6% of patients had a TEE finding which could explain the etiology of stroke or transient ischemic attack (TIA), but only 1 patient (0.4%) had a finding that changed therapy. Follow‐up was available at 6 months for 85 patients, and 13 (15%) of these patients had been discovered to develop atrial fibrillation in the interim.

DISCUSSION

Our study retrospectively analyzed the utility of TEE in patients over age 50 years admitted with ischemic stroke without a clear etiology. We found that TEE provides significant incremental diagnostic benefit as compared to TTE in identifying a possible etiology of stroke in these patients. This is consistent with prior studies showing a high diagnostic yield of TEE in patient with ischemic stroke of uncertain etiology.[3] However, in our study, based on current guidelines, virtually none of these findings directly altered patient management.

The 2014 guidelines for secondary stroke prevention recommend antiplatelet and statin therapy (in addition to lifestyle modification, smoking cessation, and blood glucose and blood pressure control) as a standard medical regimen in patients with stroke or TIA of uncertain etiology. The finding of aortic arch atheroma does not warrant supplementary treatment in addition to an antiplatelet and statin according to current guidelines. Atherosclerosis of the aortic arch is an important source of cerebral embolism, particularly in cases where plaque is >4 mm in size.[4] A recent study by Amarenco et al., comparing efficacy of combined antiplatelet therapy (clopidogrel and aspirin) to warfarin in recurrent stroke prevention in patients with >4 mm aortic arch plaque, showed nonsignificant reduction in rate of recurrent stroke with dual antiplatelet therapy.[5] However, optimal therapy for these patients still remains uncertain beyond standard stroke‐prevention treatment. Although there are emerging data on therapeutic options in patients with complex atheroma, there is currently no specific guideline‐recommended therapy or consensus among stroke neurologists. Potentially, if an individual practitioner had a strong feeling on therapeutic modifications based on the presence of complex aortic arch atheroma, the TEE would have value to their patient. However, in our study, which had a prevalence of 16.8% of complex plaque of the ascending aorta or arch, there were no therapeutic changes based on this finding. This reinforces the limited value of this test that we observed in our study population.

Anticoagulation has not been shown to be superior to aspirin in patients with PFO (with or without ASA), and recent studies showed no benefit of procedural PFO closure compared to best medical management for stroke prevention (Randomized Evaluation of Recurrent Stroke Comparing PFO Closure to Established Current Standard of Care Treatment [RESPECT], Evaluation of the STARFlex Septal Closure System in Patients with a Stroke and/or Transient Ischemic Attack due to Presumed Paradoxical Embolism through a Patent Foramen Ovale [CLOSURE I]).[6, 7] However, a patient with a PFO and deep vein thrombosis would benefit from anticoagulation and consideration of PFO closure.[8] This rare entity could be excluded with a simple lower extremity duplex without the need for a TEE, which does come with a small risk of complications related to anesthesia and local oropharyngeal trauma as well as discomfort to the patient and increased cost. Spontaneous echo contrast is not an independent indication for anticoagulation. If spontaneous contrast were associated with mitral stenosis and an embolic event, then anticoagulation would be indicated.[9] Mitral stenosis is easily diagnosed with TTE.

LAA or left atrial thrombus is the predominant finding exclusive to TEE that would change management for secondary stroke prevention, specifically anticoagulation. Fifteen studies representing over 3000 patients in a 2014 meta‐analysis reported the prevalence of left atrial or LAA thrombus in patients aged 55 years with a cryptogenic stroke to be 4%, with a range in the studies of 0% to 21.2%.[3] The wide range of prevalence of this finding is likely related to the prevalence of known atrial arrhythmias or structural heart disease in the population of patients included in the analysis. Left atrial or LAA thrombus in the absence of systolic dysfunction, severe valve disease, or known atrial fibrillation is exceedingly uncommon (0.3%).[10] It is likely that the few patients with left atrial or LAA thrombus without 1 of these conditions probably has undiagnosed paroxysmal atrial fibrillation. In previous studies that showed a high prevalence of left atrial or LAA thrombus, there was no mention of the presence or absence of LV dysfunction or severe valve disease in patients with left atrial or LAA thrombus. Additionally, these studies only required a 12‐lead electrocardiogram or did not specify the presence or duration of continuous rhythm monitoring.[11, 12, 13, 14] Several of the studies with high incidence of left atrial or LAA thrombus specifically stated that some of these patients were known to have atrial fibrillation.[11, 13]

Approximately 8% of patients admitted with stroke are found to have atrial fibrillation only after admission with continuous electrocardiogram monitoring. The detection rate is nearly half if monitoring is limited to 24 hours instead of several days. Overall, detection rates of atrial fibrillation following stroke are relatively low during initial hospitalization.[15] More intense monitoring for atrial fibrillation in patients with a stroke of uncertain etiology with the use of a subcutaneous implantable cardiac monitor increases the detection rate to 12.4% at 1 year, and increases with longer monitoring time.[16] Therefore, identification of older stroke patients without significant stroke risk factors may be candidates for longer‐term cardiac monitoring to increase yield for detection of atrial fibrillation. Currently, continuous electrocardiographic monitoring of patients for the duration of their hospitalization and up to 30 days afterward is recommended.[8]

Our study differs from prior studies that showed a much higher prevalence of LAA or left atrial thrombus in 2 important ways. Patients with severe valve disease or LV dysfunction were excluded on the basis of TTE. Additionally, our patients underwent continuous electrocardiographic monitoring for the duration of their hospitalization and were excluded with a prior history or newly discovered atrial fibrillation or flutter. Our intention was to examine the value of adding TEE when no other etiology of stroke was identified. Value can be defined as healthcare outcomes achieved per dollar spent. Our study was not designed to look at long‐term outcomes; rather, we used immediate change in patient management as a surrogate.

There are several limitations to our study that must be noted. This was a single‐center study potentially creating a bias as less stringent selection of patients undergoing TEE may be the practice at other institutions. This analysis was retrospective; therefore, there may have been bias as to which patients were selected to undergo TEE. Additionally, stroke subtype was not specified, and the pretest probability of a cardioembolic source differs based on subtype. Last, we focused this study on immediate changes in clinical management prompted by TEE results, and did not assess patient perceptions of TEE value related to enhanced knowledge about the etiology of their stroke; this area represents an opportunity for further research.

CONCLUSIONS

TEE provides a substantial increase in possible explanation of stroke etiology in patients over age 50 years admitted with a stroke of uncertain cause and a normal TTE. However, there is minimal incremental value in regard to change in therapeutic management in these patients. In a time of increased focus on providing cost effective healthcare, our findings suggest that the need for TEE in this stroke population should be more closely examined.

Disclosure: Nothing to report.

Staphylococcus aureus is one the most common pathogens isolated in nosocomial and community‐onset bloodstream infections (BSI) in the United States.[1, 2] S aureus bacteremia (SAB) has been reported in the literature to have substantial morbidity and mortality, with rates ranging between 15% and 60% worldwide.[3, 4, 5, 6] In the United States, patients with infections due to S aureus have on average 3 times the length of hospital stay than inpatients without these infections (14.3 days vs 4.5 days; P<0.01).[7] Healthcare costs are negatively impacted by these infections. In a recent meta‐analysis, Zimlichman et al.[8] reported that central‐line BSI (CLABSI) and surgical‐site infection (SSI) caused by methicillin‐resistant S aureus (MRSA) resulted in the highest estimated costs associated with hospital‐acquired infections in the United States ($58,614 [95% CI: $16,760‐$174,755] for CLABSI and $42,300 [95% CI: $4,005‐$82,670] for SSIs).

Appropriate management of SAB includes not only selecting the correct antimicrobial based on susceptibilities but also timely control of the source of infection, appropriate use of ancillary studies when indicated, and pharmacokinetic and pharmacodynamic therapeutic monitoring of antimicrobial therapy when vancomycin is used.[9] Consultation with an infectious diseases (ID) specialist has been associated with increased compliance with evidence‐based strategies in the management of SAB,[10, 11, 12, 13, 14] such as appropriate antibiotic choice, optimized duration of treatment, removal of the source of infection, and better use of cardiac echocardiography, resulting in improved outcomes.[13, 14]

Some, but not all, institutions have adopted bundles,[14] mandatory ID consultation[10] or daily prospective audit and feedback review[15] as part of antimicrobial stewardship program (ASP) interventions aiming to optimize the management of SABs. As part of our ASP quality improvement activities we performed the present study to determine our institutional rate of clinical failure in the treatment of SAB, to identify current practice patterns in the delivery of processes of care, and evaluate their association with clinical outcomes of hospitalized patients with SAB to identify future areas of improvement.

METHODS

A retrospective cohort study was performed at a 1558 licensed‐bed tertiary teaching hospital in Miami, Florida. All hospitalized patients 18 years of age or older with at least 1 positive blood culture with MRSA or methicillin‐susceptible S aureus (MSSA) between January 1, 2012 and April 30, 2013 were included. Patients were identified from the electronic microbiology laboratory database. For the purposes of this study, only the first episode of SAB was included in the analysis. Patients were excluded if aged younger than 18 years or if SAB was detected in an outpatient setting. The primary outcome was clinical failure, defined as a composite endpoint of in‐hospital mortality or persistent bacteremia; persistent bacteremia was defined as bacteremia for 7 or more days after the first positive blood culture. S aureus isolates were identified by standard methods.[16] Species identification was performed by latex agglutination. Antimicrobial susceptibility testing was performed using an automated system (Vitek 2; bioMerieux, Durham, NC) according to standard guidelines.

Data collected included baseline demographics, comorbidities, and treating healthcare provider's service; provider's service was categorized into 1 of 5 groups: internal medicine (academic), internal medicine (hospitalist), surgery, trauma, or neurosurgery. Duration of bacteremia was recorded and defined as the time between first positive and first negative blood culture. The time of first positive culture was defined as the date in which the culture was obtained. Patients who failed to have at least 1 follow‐up blood culture were not counted toward the main outcome. Additionally, presence of a foreign body (cardiac device, orthopedic prosthesis, tunneled catheter, nontunneled catheter) and presumed source of infection as documented in the electronic medical record by the treating service was also collected. Infections were considered community associated when onset of bacteremia occurred within the first 72 hours of admission, and hospital associated if onset of bacteremia occurred after 72 hours of admission.

Based on current practice guidelines,[9] the variables considered processes of care were the time to obtain the first follow‐up blood culture, time from first positive blood culture to initiation of appropriate antibiotic therapy (defined as a loading dose of vancomycin of 15 mg/kg, or a ‐lactam if the organism was susceptible), time to obtain the first vancomycin trough (when indicated), time from first positive blood culture to consultation with ID specialist, appropriate antibiotic de‐escalation (vancomycin to ‐lactam antibiotic if the organism was susceptible and the patient had no allergies or contraindications), and obtaining an echocardiographic study (transthoracic echocardiogram or transesophageal echocardiogram).

Statistical analyses were performed using SAS 9.2 (SAS Institute, Cary, NC). Differences in proportions were analyzed with ² or Fisher exact test, accordingly. Differences in means among continuous variables were evaluated using independent samples of paired samples t tests as appropriate for the analysis. Continuous variables were dichotomized using a clinically established cutoff to determine relative risk (RR). A univariate analysis of risk factors associated with clinical failure was performed. Multivariable analyses were performed using logistic regression. Models were created using the backward stepwise approach and included all variables found to be statistically significant at less than 0.05 level in the univariate model and those of clinical significance. The study was reviewed and approved by the institutional review boards at the University of Miami and Jackson Memorial Hospital.

RESULTS

During the study period, 241 patients with a first episode of SAB were identified. MRSA and MSSA were isolated in 124 (51.4%) and 117 (48.5%) patients, respectively. Demographic and clinical characteristics of the study population based on isolate are summarized in Table 1. One hundred seventy‐nine (74.3%) patients were under the care of internal medicine services. There was no association between treating service (medical vs surgical) and clinical failure.

The onset of infection occurred in the community in 77 (62.1%) patients with MRSA and in 77 (65.8%) patients with MSSA. The documented source of bacteremia was unknown in 30% of patients with MRSA and 44% of those with MSSA BSI. When ID specialists were consulted, patients were more likely to have a source of infection identified (RR: 1.5; 95% confidence interval [CI]: 1.2‐1.8; P<0.0001). The most commonly documented sources of infection were CLABSI, which occurred in 32 (25.8%) patients with MRSA and 21 (17.9%) patients with MSSA, followed by skin and soft tissue infections in 24 (19.3%) patients with MRSA BSI and 20 (17.1%) patients with MSSA BSI. All patients with CLABSI had documentation of catheter removal.

Clinical failure (defined as in‐hospital mortality or persistent bacteremia) occurred in 78 (32.4%) patients. Of these, 50 (20.7%) represented in‐hospital mortality, and 31 (12.9%) had persistent bacteremia. Table 2 summarizes the demographic and clinical characteristics associated with clinical failure. In the univariate analysis, the variables statistically significantly associated with clinical failure were: age greater than 60 years (RR: 1.4; 95% CI: 1.1‐1.8; P=0.001), bacteremia due to MRSA (RR: 1.7; 95% CI: 1.1‐2.5; P=0.008), white race (RR: 0.7; 95% CI: 0.6‐1; P=0.03), acute kidney injury during admission (RR: 2.2; 95% CI: 1.3‐3.7; P=0.004), presence of nontunneled central venous catheters at the onset of bacteremia (RR: 1.9; 95% CI: 1.3‐2.7; P=0.004), and endocarditis (RR: 2.9; 95% CI: 2.1‐3.9; P<0.0001). In the multivariable analysis, age greater than 60 years and endocarditis were found to be independent risk factors for the development of clinical failure.

DISCUSSION

Our study showed a significant rate of morbidity associated with S aureus bacteremia and identified processes of care in the management of SAB that impact patient outcomes.

Our results show that early consultation with an ID specialist was associated with a decreased risk of developing clinical failure, increased likelihood of identification of a source of infection, and positively impacted administration of appropriate antibiotic therapy, especially in cases of MSSA BSI, with overall improvement in patient outcomes. However, consultation with an ID specialist was only obtained in 40.2% of our cases, which is consistent with published data.[10, 11, 12, 13] Consultation with an ID specialist itself did not impact clinical failure, but rather timeliness in obtaining expert guidance was associated with better outcomes. As shown in previous studies,[10, 11, 12, 13, 14] compliance with the standards of care and patient prognosis are improved when ID specialists are involved in the management of SAB. Our study reiterates that early consultation with an ID specialist has a positive outcome in patient care, as opposed to delaying consultation once the patient has persistent bacteremia for more than 7 days. This association could be explained by considering that the majority of the standards of care are time sensitive, which include: obtaining surveillance blood cultures 48 to 96 hours after initial detection[10] or initiating therapy,[11, 14] removal of foci of infection,[10, 11, 12, 14] use of parenteral ‐lactams for the treatment of MSSA,[10, 11, 13, 14] performing echocardiography when clinically indicated,[10, 11, 13, 14] and appropriate duration of therapy.[10, 13, 14] Importantly, studies have shown that when ID specialists' recommendations are followed, patients are more likely to be cured,[10, 11, 13] and are less likely to relapse.[10, 11, 12] Given the complexities of treating patients with SAB and high rates of clinical failures, routine guidance could be beneficial to healthcare providers as part of a multidisciplinary structured strategy that is set in motion the moment a patient with SAB is identified by the microbiology laboratory. The processes of care outlined in this this study can serve as quality of care indicators and be integrated into a structured strategy to optimize the management of SAB.

Regarding optimal timing for follow‐up blood cultures, our results show that delays in obtaining follow‐up blood cultures (more than 4 days from onset of bacteremia) was independently associated with increased risk of clinical failure. Timely follow‐up blood cultures have been previously identified as quality of care indicators.[10, 11, 13, 14] Compliance with obtaining follow‐up blood cultures improves when this step is integrated into a bundle of care.[14]

Antimicrobial therapy was promptly initiated in the majority of the patients in our study. However, areas for improvement were identified. Vancomycin was the empirical therapy of choice in most of the cases, but an appropriate dose was only received by 65% of the patients, and vancomycin levels after the fourth dose were obtained in 85.9% of instances when indicated. Although in our cohort these results were not significantly associated with clinical failure, previous studies have described attainment of a target therapeutic vancomycin trough (1520 mg/dL) as a factor for treatment success.[17, 18] This problem could be addressed through physician education on therapeutic drug monitoring,[19] as well as through an ASP intervention, which have successfully led efforts to improve vancomycin utilization and dosing.[20] Among patients with MSSA BSI, therapy with ‐lactams was associated with improved outcomes, and was more likely to be administered when an ID specialist was consulted. This is in accordance with previous studies that have shown that higher rates of appropriate antimicrobial therapy are achieved when ID specialists are involved in management of SAB.[10, 11, 13, 14] The use of ‐lactams for treatment of MSSA BSI has been consistently associated with lower SAB‐related mortality and relapse.[21, 22, 23, 24, 25, 26]

Echocardiographic studies were obtained in only half of the patients in our cohort, and they were twice more likely to be obtained when an ID specialist was consulted. Although we did not evaluate the appropriateness of the echocardiographic study, the increased proportion of studies performed when ID specialists were consulted could indicate a more in‐depth evaluation of the case. Moreover, in our cohort, when ID specialists where involved in direct patient care, a source of infection was more likely to be identified. This is in accordance with previous studies proposing that because evaluation by ID specialists are more detailed, they lead to increased use in ancillary studies and recognition of complicated cases.[10, 12]

Limitations of this study include its retrospective design and the fact that it was performed in a single institution. The source of infection was defined as documented by treating providers and not by independent diagnostic criteria. Antibiotic use was collected throughout duration of admission, and was not followed after patients were discharged, as these data were not available on the electronic medical record for all patients. Deaths that may have occurred after hospital discharge were not included. We did not account for elevated vancomycin minimum inhibitory concentration as a risk factor for the main outcome, and adjustment of vancomycin based on serum levels was not factored in. Acute kidney injury was accounted for anytime during hospitalization, but not in relation to antimicrobial administration. Despite the limitations, our study has strengths that make our results generalizable. Although our institution is a single medical center, it serves a large and diverse population as reflected in our cases. Even though this is a retrospective cohort study, the use of a centralized electronic medical record allowed us to identify each aspect of the management of SAB, as implemented by different treating services (medical and surgical), as continuous variables (days) rather than only in a dichotomous fashion. Moreover, by being a community teaching hospital, we were able to explore aspects of the practice of physicians in training versus practicing clinicians. These results could be extrapolated to other healthcare facilities aiming to improve the management of SAB.

CONCLUSIONS

Our results suggest that obtaining timely follow‐up blood cultures, use of ‐lactams in patients with MSSA BSI, and early consultation with infectious diseases are the processes of care that could serve as quality and patient‐safety indicators for the management of SAB. These results contribute to a growing body of evidence supporting the implementation of structured processes of care to optimize the management and clinical outcomes of hospitalized patients with SAB.

Disclosure: Nothing to report.

Staphylococcus aureus is one the most common pathogens isolated in nosocomial and community‐onset bloodstream infections (BSI) in the United States.[1, 2] S aureus bacteremia (SAB) has been reported in the literature to have substantial morbidity and mortality, with rates ranging between 15% and 60% worldwide.[3, 4, 5, 6] In the United States, patients with infections due to S aureus have on average 3 times the length of hospital stay than inpatients without these infections (14.3 days vs 4.5 days; P<0.01).[7] Healthcare costs are negatively impacted by these infections. In a recent meta‐analysis, Zimlichman et al.[8] reported that central‐line BSI (CLABSI) and surgical‐site infection (SSI) caused by methicillin‐resistant S aureus (MRSA) resulted in the highest estimated costs associated with hospital‐acquired infections in the United States ($58,614 [95% CI: $16,760‐$174,755] for CLABSI and $42,300 [95% CI: $4,005‐$82,670] for SSIs).

Appropriate management of SAB includes not only selecting the correct antimicrobial based on susceptibilities but also timely control of the source of infection, appropriate use of ancillary studies when indicated, and pharmacokinetic and pharmacodynamic therapeutic monitoring of antimicrobial therapy when vancomycin is used.[9] Consultation with an infectious diseases (ID) specialist has been associated with increased compliance with evidence‐based strategies in the management of SAB,[10, 11, 12, 13, 14] such as appropriate antibiotic choice, optimized duration of treatment, removal of the source of infection, and better use of cardiac echocardiography, resulting in improved outcomes.[13, 14]

Some, but not all, institutions have adopted bundles,[14] mandatory ID consultation[10] or daily prospective audit and feedback review[15] as part of antimicrobial stewardship program (ASP) interventions aiming to optimize the management of SABs. As part of our ASP quality improvement activities we performed the present study to determine our institutional rate of clinical failure in the treatment of SAB, to identify current practice patterns in the delivery of processes of care, and evaluate their association with clinical outcomes of hospitalized patients with SAB to identify future areas of improvement.

METHODS

A retrospective cohort study was performed at a 1558 licensed‐bed tertiary teaching hospital in Miami, Florida. All hospitalized patients 18 years of age or older with at least 1 positive blood culture with MRSA or methicillin‐susceptible S aureus (MSSA) between January 1, 2012 and April 30, 2013 were included. Patients were identified from the electronic microbiology laboratory database. For the purposes of this study, only the first episode of SAB was included in the analysis. Patients were excluded if aged younger than 18 years or if SAB was detected in an outpatient setting. The primary outcome was clinical failure, defined as a composite endpoint of in‐hospital mortality or persistent bacteremia; persistent bacteremia was defined as bacteremia for 7 or more days after the first positive blood culture. S aureus isolates were identified by standard methods.[16] Species identification was performed by latex agglutination. Antimicrobial susceptibility testing was performed using an automated system (Vitek 2; bioMerieux, Durham, NC) according to standard guidelines.

Data collected included baseline demographics, comorbidities, and treating healthcare provider's service; provider's service was categorized into 1 of 5 groups: internal medicine (academic), internal medicine (hospitalist), surgery, trauma, or neurosurgery. Duration of bacteremia was recorded and defined as the time between first positive and first negative blood culture. The time of first positive culture was defined as the date in which the culture was obtained. Patients who failed to have at least 1 follow‐up blood culture were not counted toward the main outcome. Additionally, presence of a foreign body (cardiac device, orthopedic prosthesis, tunneled catheter, nontunneled catheter) and presumed source of infection as documented in the electronic medical record by the treating service was also collected. Infections were considered community associated when onset of bacteremia occurred within the first 72 hours of admission, and hospital associated if onset of bacteremia occurred after 72 hours of admission.

Based on current practice guidelines,[9] the variables considered processes of care were the time to obtain the first follow‐up blood culture, time from first positive blood culture to initiation of appropriate antibiotic therapy (defined as a loading dose of vancomycin of 15 mg/kg, or a ‐lactam if the organism was susceptible), time to obtain the first vancomycin trough (when indicated), time from first positive blood culture to consultation with ID specialist, appropriate antibiotic de‐escalation (vancomycin to ‐lactam antibiotic if the organism was susceptible and the patient had no allergies or contraindications), and obtaining an echocardiographic study (transthoracic echocardiogram or transesophageal echocardiogram).

Statistical analyses were performed using SAS 9.2 (SAS Institute, Cary, NC). Differences in proportions were analyzed with ² or Fisher exact test, accordingly. Differences in means among continuous variables were evaluated using independent samples of paired samples t tests as appropriate for the analysis. Continuous variables were dichotomized using a clinically established cutoff to determine relative risk (RR). A univariate analysis of risk factors associated with clinical failure was performed. Multivariable analyses were performed using logistic regression. Models were created using the backward stepwise approach and included all variables found to be statistically significant at less than 0.05 level in the univariate model and those of clinical significance. The study was reviewed and approved by the institutional review boards at the University of Miami and Jackson Memorial Hospital.

RESULTS

During the study period, 241 patients with a first episode of SAB were identified. MRSA and MSSA were isolated in 124 (51.4%) and 117 (48.5%) patients, respectively. Demographic and clinical characteristics of the study population based on isolate are summarized in Table 1. One hundred seventy‐nine (74.3%) patients were under the care of internal medicine services. There was no association between treating service (medical vs surgical) and clinical failure.

The onset of infection occurred in the community in 77 (62.1%) patients with MRSA and in 77 (65.8%) patients with MSSA. The documented source of bacteremia was unknown in 30% of patients with MRSA and 44% of those with MSSA BSI. When ID specialists were consulted, patients were more likely to have a source of infection identified (RR: 1.5; 95% confidence interval [CI]: 1.2‐1.8; P<0.0001). The most commonly documented sources of infection were CLABSI, which occurred in 32 (25.8%) patients with MRSA and 21 (17.9%) patients with MSSA, followed by skin and soft tissue infections in 24 (19.3%) patients with MRSA BSI and 20 (17.1%) patients with MSSA BSI. All patients with CLABSI had documentation of catheter removal.

Clinical failure (defined as in‐hospital mortality or persistent bacteremia) occurred in 78 (32.4%) patients. Of these, 50 (20.7%) represented in‐hospital mortality, and 31 (12.9%) had persistent bacteremia. Table 2 summarizes the demographic and clinical characteristics associated with clinical failure. In the univariate analysis, the variables statistically significantly associated with clinical failure were: age greater than 60 years (RR: 1.4; 95% CI: 1.1‐1.8; P=0.001), bacteremia due to MRSA (RR: 1.7; 95% CI: 1.1‐2.5; P=0.008), white race (RR: 0.7; 95% CI: 0.6‐1; P=0.03), acute kidney injury during admission (RR: 2.2; 95% CI: 1.3‐3.7; P=0.004), presence of nontunneled central venous catheters at the onset of bacteremia (RR: 1.9; 95% CI: 1.3‐2.7; P=0.004), and endocarditis (RR: 2.9; 95% CI: 2.1‐3.9; P<0.0001). In the multivariable analysis, age greater than 60 years and endocarditis were found to be independent risk factors for the development of clinical failure.

DISCUSSION

Our study showed a significant rate of morbidity associated with S aureus bacteremia and identified processes of care in the management of SAB that impact patient outcomes.

Our results show that early consultation with an ID specialist was associated with a decreased risk of developing clinical failure, increased likelihood of identification of a source of infection, and positively impacted administration of appropriate antibiotic therapy, especially in cases of MSSA BSI, with overall improvement in patient outcomes. However, consultation with an ID specialist was only obtained in 40.2% of our cases, which is consistent with published data.[10, 11, 12, 13] Consultation with an ID specialist itself did not impact clinical failure, but rather timeliness in obtaining expert guidance was associated with better outcomes. As shown in previous studies,[10, 11, 12, 13, 14] compliance with the standards of care and patient prognosis are improved when ID specialists are involved in the management of SAB. Our study reiterates that early consultation with an ID specialist has a positive outcome in patient care, as opposed to delaying consultation once the patient has persistent bacteremia for more than 7 days. This association could be explained by considering that the majority of the standards of care are time sensitive, which include: obtaining surveillance blood cultures 48 to 96 hours after initial detection[10] or initiating therapy,[11, 14] removal of foci of infection,[10, 11, 12, 14] use of parenteral ‐lactams for the treatment of MSSA,[10, 11, 13, 14] performing echocardiography when clinically indicated,[10, 11, 13, 14] and appropriate duration of therapy.[10, 13, 14] Importantly, studies have shown that when ID specialists' recommendations are followed, patients are more likely to be cured,[10, 11, 13] and are less likely to relapse.[10, 11, 12] Given the complexities of treating patients with SAB and high rates of clinical failures, routine guidance could be beneficial to healthcare providers as part of a multidisciplinary structured strategy that is set in motion the moment a patient with SAB is identified by the microbiology laboratory. The processes of care outlined in this this study can serve as quality of care indicators and be integrated into a structured strategy to optimize the management of SAB.

Regarding optimal timing for follow‐up blood cultures, our results show that delays in obtaining follow‐up blood cultures (more than 4 days from onset of bacteremia) was independently associated with increased risk of clinical failure. Timely follow‐up blood cultures have been previously identified as quality of care indicators.[10, 11, 13, 14] Compliance with obtaining follow‐up blood cultures improves when this step is integrated into a bundle of care.[14]

Antimicrobial therapy was promptly initiated in the majority of the patients in our study. However, areas for improvement were identified. Vancomycin was the empirical therapy of choice in most of the cases, but an appropriate dose was only received by 65% of the patients, and vancomycin levels after the fourth dose were obtained in 85.9% of instances when indicated. Although in our cohort these results were not significantly associated with clinical failure, previous studies have described attainment of a target therapeutic vancomycin trough (1520 mg/dL) as a factor for treatment success.[17, 18] This problem could be addressed through physician education on therapeutic drug monitoring,[19] as well as through an ASP intervention, which have successfully led efforts to improve vancomycin utilization and dosing.[20] Among patients with MSSA BSI, therapy with ‐lactams was associated with improved outcomes, and was more likely to be administered when an ID specialist was consulted. This is in accordance with previous studies that have shown that higher rates of appropriate antimicrobial therapy are achieved when ID specialists are involved in management of SAB.[10, 11, 13, 14] The use of ‐lactams for treatment of MSSA BSI has been consistently associated with lower SAB‐related mortality and relapse.[21, 22, 23, 24, 25, 26]

Echocardiographic studies were obtained in only half of the patients in our cohort, and they were twice more likely to be obtained when an ID specialist was consulted. Although we did not evaluate the appropriateness of the echocardiographic study, the increased proportion of studies performed when ID specialists were consulted could indicate a more in‐depth evaluation of the case. Moreover, in our cohort, when ID specialists where involved in direct patient care, a source of infection was more likely to be identified. This is in accordance with previous studies proposing that because evaluation by ID specialists are more detailed, they lead to increased use in ancillary studies and recognition of complicated cases.[10, 12]

Limitations of this study include its retrospective design and the fact that it was performed in a single institution. The source of infection was defined as documented by treating providers and not by independent diagnostic criteria. Antibiotic use was collected throughout duration of admission, and was not followed after patients were discharged, as these data were not available on the electronic medical record for all patients. Deaths that may have occurred after hospital discharge were not included. We did not account for elevated vancomycin minimum inhibitory concentration as a risk factor for the main outcome, and adjustment of vancomycin based on serum levels was not factored in. Acute kidney injury was accounted for anytime during hospitalization, but not in relation to antimicrobial administration. Despite the limitations, our study has strengths that make our results generalizable. Although our institution is a single medical center, it serves a large and diverse population as reflected in our cases. Even though this is a retrospective cohort study, the use of a centralized electronic medical record allowed us to identify each aspect of the management of SAB, as implemented by different treating services (medical and surgical), as continuous variables (days) rather than only in a dichotomous fashion. Moreover, by being a community teaching hospital, we were able to explore aspects of the practice of physicians in training versus practicing clinicians. These results could be extrapolated to other healthcare facilities aiming to improve the management of SAB.

CONCLUSIONS

Our results suggest that obtaining timely follow‐up blood cultures, use of ‐lactams in patients with MSSA BSI, and early consultation with infectious diseases are the processes of care that could serve as quality and patient‐safety indicators for the management of SAB. These results contribute to a growing body of evidence supporting the implementation of structured processes of care to optimize the management and clinical outcomes of hospitalized patients with SAB.

Disclosure: Nothing to report.

Staphylococcus aureus is one the most common pathogens isolated in nosocomial and community‐onset bloodstream infections (BSI) in the United States.[1, 2] S aureus bacteremia (SAB) has been reported in the literature to have substantial morbidity and mortality, with rates ranging between 15% and 60% worldwide.[3, 4, 5, 6] In the United States, patients with infections due to S aureus have on average 3 times the length of hospital stay than inpatients without these infections (14.3 days vs 4.5 days; P<0.01).[7] Healthcare costs are negatively impacted by these infections. In a recent meta‐analysis, Zimlichman et al.[8] reported that central‐line BSI (CLABSI) and surgical‐site infection (SSI) caused by methicillin‐resistant S aureus (MRSA) resulted in the highest estimated costs associated with hospital‐acquired infections in the United States ($58,614 [95% CI: $16,760‐$174,755] for CLABSI and $42,300 [95% CI: $4,005‐$82,670] for SSIs).

Appropriate management of SAB includes not only selecting the correct antimicrobial based on susceptibilities but also timely control of the source of infection, appropriate use of ancillary studies when indicated, and pharmacokinetic and pharmacodynamic therapeutic monitoring of antimicrobial therapy when vancomycin is used.[9] Consultation with an infectious diseases (ID) specialist has been associated with increased compliance with evidence‐based strategies in the management of SAB,[10, 11, 12, 13, 14] such as appropriate antibiotic choice, optimized duration of treatment, removal of the source of infection, and better use of cardiac echocardiography, resulting in improved outcomes.[13, 14]

Some, but not all, institutions have adopted bundles,[14] mandatory ID consultation[10] or daily prospective audit and feedback review[15] as part of antimicrobial stewardship program (ASP) interventions aiming to optimize the management of SABs. As part of our ASP quality improvement activities we performed the present study to determine our institutional rate of clinical failure in the treatment of SAB, to identify current practice patterns in the delivery of processes of care, and evaluate their association with clinical outcomes of hospitalized patients with SAB to identify future areas of improvement.

METHODS

A retrospective cohort study was performed at a 1558 licensed‐bed tertiary teaching hospital in Miami, Florida. All hospitalized patients 18 years of age or older with at least 1 positive blood culture with MRSA or methicillin‐susceptible S aureus (MSSA) between January 1, 2012 and April 30, 2013 were included. Patients were identified from the electronic microbiology laboratory database. For the purposes of this study, only the first episode of SAB was included in the analysis. Patients were excluded if aged younger than 18 years or if SAB was detected in an outpatient setting. The primary outcome was clinical failure, defined as a composite endpoint of in‐hospital mortality or persistent bacteremia; persistent bacteremia was defined as bacteremia for 7 or more days after the first positive blood culture. S aureus isolates were identified by standard methods.[16] Species identification was performed by latex agglutination. Antimicrobial susceptibility testing was performed using an automated system (Vitek 2; bioMerieux, Durham, NC) according to standard guidelines.

Data collected included baseline demographics, comorbidities, and treating healthcare provider's service; provider's service was categorized into 1 of 5 groups: internal medicine (academic), internal medicine (hospitalist), surgery, trauma, or neurosurgery. Duration of bacteremia was recorded and defined as the time between first positive and first negative blood culture. The time of first positive culture was defined as the date in which the culture was obtained. Patients who failed to have at least 1 follow‐up blood culture were not counted toward the main outcome. Additionally, presence of a foreign body (cardiac device, orthopedic prosthesis, tunneled catheter, nontunneled catheter) and presumed source of infection as documented in the electronic medical record by the treating service was also collected. Infections were considered community associated when onset of bacteremia occurred within the first 72 hours of admission, and hospital associated if onset of bacteremia occurred after 72 hours of admission.

Based on current practice guidelines,[9] the variables considered processes of care were the time to obtain the first follow‐up blood culture, time from first positive blood culture to initiation of appropriate antibiotic therapy (defined as a loading dose of vancomycin of 15 mg/kg, or a ‐lactam if the organism was susceptible), time to obtain the first vancomycin trough (when indicated), time from first positive blood culture to consultation with ID specialist, appropriate antibiotic de‐escalation (vancomycin to ‐lactam antibiotic if the organism was susceptible and the patient had no allergies or contraindications), and obtaining an echocardiographic study (transthoracic echocardiogram or transesophageal echocardiogram).

Statistical analyses were performed using SAS 9.2 (SAS Institute, Cary, NC). Differences in proportions were analyzed with ² or Fisher exact test, accordingly. Differences in means among continuous variables were evaluated using independent samples of paired samples t tests as appropriate for the analysis. Continuous variables were dichotomized using a clinically established cutoff to determine relative risk (RR). A univariate analysis of risk factors associated with clinical failure was performed. Multivariable analyses were performed using logistic regression. Models were created using the backward stepwise approach and included all variables found to be statistically significant at less than 0.05 level in the univariate model and those of clinical significance. The study was reviewed and approved by the institutional review boards at the University of Miami and Jackson Memorial Hospital.

RESULTS

During the study period, 241 patients with a first episode of SAB were identified. MRSA and MSSA were isolated in 124 (51.4%) and 117 (48.5%) patients, respectively. Demographic and clinical characteristics of the study population based on isolate are summarized in Table 1. One hundred seventy‐nine (74.3%) patients were under the care of internal medicine services. There was no association between treating service (medical vs surgical) and clinical failure.

The onset of infection occurred in the community in 77 (62.1%) patients with MRSA and in 77 (65.8%) patients with MSSA. The documented source of bacteremia was unknown in 30% of patients with MRSA and 44% of those with MSSA BSI. When ID specialists were consulted, patients were more likely to have a source of infection identified (RR: 1.5; 95% confidence interval [CI]: 1.2‐1.8; P<0.0001). The most commonly documented sources of infection were CLABSI, which occurred in 32 (25.8%) patients with MRSA and 21 (17.9%) patients with MSSA, followed by skin and soft tissue infections in 24 (19.3%) patients with MRSA BSI and 20 (17.1%) patients with MSSA BSI. All patients with CLABSI had documentation of catheter removal.

Clinical failure (defined as in‐hospital mortality or persistent bacteremia) occurred in 78 (32.4%) patients. Of these, 50 (20.7%) represented in‐hospital mortality, and 31 (12.9%) had persistent bacteremia. Table 2 summarizes the demographic and clinical characteristics associated with clinical failure. In the univariate analysis, the variables statistically significantly associated with clinical failure were: age greater than 60 years (RR: 1.4; 95% CI: 1.1‐1.8; P=0.001), bacteremia due to MRSA (RR: 1.7; 95% CI: 1.1‐2.5; P=0.008), white race (RR: 0.7; 95% CI: 0.6‐1; P=0.03), acute kidney injury during admission (RR: 2.2; 95% CI: 1.3‐3.7; P=0.004), presence of nontunneled central venous catheters at the onset of bacteremia (RR: 1.9; 95% CI: 1.3‐2.7; P=0.004), and endocarditis (RR: 2.9; 95% CI: 2.1‐3.9; P<0.0001). In the multivariable analysis, age greater than 60 years and endocarditis were found to be independent risk factors for the development of clinical failure.

DISCUSSION

Our study showed a significant rate of morbidity associated with S aureus bacteremia and identified processes of care in the management of SAB that impact patient outcomes.

Our results show that early consultation with an ID specialist was associated with a decreased risk of developing clinical failure, increased likelihood of identification of a source of infection, and positively impacted administration of appropriate antibiotic therapy, especially in cases of MSSA BSI, with overall improvement in patient outcomes. However, consultation with an ID specialist was only obtained in 40.2% of our cases, which is consistent with published data.[10, 11, 12, 13] Consultation with an ID specialist itself did not impact clinical failure, but rather timeliness in obtaining expert guidance was associated with better outcomes. As shown in previous studies,[10, 11, 12, 13, 14] compliance with the standards of care and patient prognosis are improved when ID specialists are involved in the management of SAB. Our study reiterates that early consultation with an ID specialist has a positive outcome in patient care, as opposed to delaying consultation once the patient has persistent bacteremia for more than 7 days. This association could be explained by considering that the majority of the standards of care are time sensitive, which include: obtaining surveillance blood cultures 48 to 96 hours after initial detection[10] or initiating therapy,[11, 14] removal of foci of infection,[10, 11, 12, 14] use of parenteral ‐lactams for the treatment of MSSA,[10, 11, 13, 14] performing echocardiography when clinically indicated,[10, 11, 13, 14] and appropriate duration of therapy.[10, 13, 14] Importantly, studies have shown that when ID specialists' recommendations are followed, patients are more likely to be cured,[10, 11, 13] and are less likely to relapse.[10, 11, 12] Given the complexities of treating patients with SAB and high rates of clinical failures, routine guidance could be beneficial to healthcare providers as part of a multidisciplinary structured strategy that is set in motion the moment a patient with SAB is identified by the microbiology laboratory. The processes of care outlined in this this study can serve as quality of care indicators and be integrated into a structured strategy to optimize the management of SAB.

Regarding optimal timing for follow‐up blood cultures, our results show that delays in obtaining follow‐up blood cultures (more than 4 days from onset of bacteremia) was independently associated with increased risk of clinical failure. Timely follow‐up blood cultures have been previously identified as quality of care indicators.[10, 11, 13, 14] Compliance with obtaining follow‐up blood cultures improves when this step is integrated into a bundle of care.[14]

Antimicrobial therapy was promptly initiated in the majority of the patients in our study. However, areas for improvement were identified. Vancomycin was the empirical therapy of choice in most of the cases, but an appropriate dose was only received by 65% of the patients, and vancomycin levels after the fourth dose were obtained in 85.9% of instances when indicated. Although in our cohort these results were not significantly associated with clinical failure, previous studies have described attainment of a target therapeutic vancomycin trough (1520 mg/dL) as a factor for treatment success.[17, 18] This problem could be addressed through physician education on therapeutic drug monitoring,[19] as well as through an ASP intervention, which have successfully led efforts to improve vancomycin utilization and dosing.[20] Among patients with MSSA BSI, therapy with ‐lactams was associated with improved outcomes, and was more likely to be administered when an ID specialist was consulted. This is in accordance with previous studies that have shown that higher rates of appropriate antimicrobial therapy are achieved when ID specialists are involved in management of SAB.[10, 11, 13, 14] The use of ‐lactams for treatment of MSSA BSI has been consistently associated with lower SAB‐related mortality and relapse.[21, 22, 23, 24, 25, 26]

Echocardiographic studies were obtained in only half of the patients in our cohort, and they were twice more likely to be obtained when an ID specialist was consulted. Although we did not evaluate the appropriateness of the echocardiographic study, the increased proportion of studies performed when ID specialists were consulted could indicate a more in‐depth evaluation of the case. Moreover, in our cohort, when ID specialists where involved in direct patient care, a source of infection was more likely to be identified. This is in accordance with previous studies proposing that because evaluation by ID specialists are more detailed, they lead to increased use in ancillary studies and recognition of complicated cases.[10, 12]

Limitations of this study include its retrospective design and the fact that it was performed in a single institution. The source of infection was defined as documented by treating providers and not by independent diagnostic criteria. Antibiotic use was collected throughout duration of admission, and was not followed after patients were discharged, as these data were not available on the electronic medical record for all patients. Deaths that may have occurred after hospital discharge were not included. We did not account for elevated vancomycin minimum inhibitory concentration as a risk factor for the main outcome, and adjustment of vancomycin based on serum levels was not factored in. Acute kidney injury was accounted for anytime during hospitalization, but not in relation to antimicrobial administration. Despite the limitations, our study has strengths that make our results generalizable. Although our institution is a single medical center, it serves a large and diverse population as reflected in our cases. Even though this is a retrospective cohort study, the use of a centralized electronic medical record allowed us to identify each aspect of the management of SAB, as implemented by different treating services (medical and surgical), as continuous variables (days) rather than only in a dichotomous fashion. Moreover, by being a community teaching hospital, we were able to explore aspects of the practice of physicians in training versus practicing clinicians. These results could be extrapolated to other healthcare facilities aiming to improve the management of SAB.

CONCLUSIONS

Our results suggest that obtaining timely follow‐up blood cultures, use of ‐lactams in patients with MSSA BSI, and early consultation with infectious diseases are the processes of care that could serve as quality and patient‐safety indicators for the management of SAB. These results contribute to a growing body of evidence supporting the implementation of structured processes of care to optimize the management and clinical outcomes of hospitalized patients with SAB.

Disclosure: Nothing to report.

The hospitalized patient experience has become an area of increased focus for hospitals given the recent coupling of patient satisfaction to reimbursement rates for Medicare patients.[1] Although patient experiences are multifactorial, 1 component is the relationship that hospitalized patients develop with their inpatient physicians. In recognition of the importance of this relationship, several organizations including the Society of Hospital Medicine, Society of General Internal Medicine, American College of Physicians, the American College of Emergency Physicians, and the Accreditation Council for Graduate Medical Education have recommended that patients know and understand who is guiding their care at all times during their hospitalization.[2, 3] Unfortunately, previous studies have shown that hospitalized patients often lack the ability to identify[4, 5] and understand their course of care.[6, 7] This may be due to numerous clinical factors including lack of a prior relationship, rapid pace of clinical care, and the frequent transitions of care found in both hospitalists and general medicine teaching services.[5, 8, 9] Regardless of the cause, one could hypothesize that patients who are unable to identify or understand the role of their physician may be less informed about their hospitalization, which may lead to further confusion, dissatisfaction, and ultimately a poor experience.

Given the proliferation of nonteaching hospitalist services in teaching hospitals, it is important to understand if patient experiences differ between general medicine teaching and hospitalist services. Several reasons could explain why patient experiences may vary on these services. For example, patients on a hospitalist service will likely interact with a single physician caretaker, which may give a feeling of more personalized care. In contrast, patients on general medicine teaching services are cared for by larger teams of residents under the supervision of an attending physician. Residents are also subjected to duty‐hour restrictions, clinic responsibilities, and other educational requirements that may impede the continuity of care for hospitalized patients.[10, 11, 12] Although 1 study has shown that hospitalist‐intensive hospitals perform better on patient satisfaction measures,[13] no study to date has compared patient‐reported experiences on general medicine teaching and nonteaching hospitalist services. This study aimed to evaluate the hospitalized patient experience on both teaching and nonteaching hospitalist services by assessing several patient‐reported measures of their experience, namely their confidence in their ability to identify their physician(s), understand their roles, and their rating of both the coordination and overall care.

METHODS

RESULTS

In total, 14,855 patients were enrolled during their hospitalization with 57% and 61% completing the 30‐day follow‐up survey on the hospitalist and general medicine teaching service, respectively. In total, 4131 (69%) and 4322 (48%) of the hospitalist and general medicine services, respectively, either did not answer all survey questions, or were missing basic demographic data, and thus were excluded. Data from 4591 patients on the general medicine teaching (52% of those enrolled at hospitalization) and 1811 on the hospitalist service (31% of those enrolled at hospitalization) were used for final analysis (Figure 1). Respondents were predominantly female (61% and 56%), African American (75% and 63%), with a mean age of 56.2 (19.4) and 57.1 (16.1) years, for the general medicine teaching and hospitalist services, respectively. A majority of patients (71% and 66%) had a CCI of 0 to 3 on both services. There were differences in self‐reported comorbidities between the 2 groups, with hospitalist services having a higher prevalence of cancer (20% vs 7%), renal disease (25% vs 18%), and liver disease (23% vs 7%). Patients on the hospitalist service had a longer mean LOS (5.5 vs 4.8 days), a greater percentage of a hospitalization within 1 year (58% vs 52%), and a larger proportion who were admitted in 2011 to 2013 compared to 2007 to 2010 (75% vs 39%), when compared to the general medicine teaching services. Median LOS and interquartile ranges were similar between both groups. Although most baseline demographics were statistically different between the 2 groups (Table 1), these differences were likely clinically insignificant. Compared to those who responded to the follow‐up survey, nonresponders were more likely to be African American (73% and 64%, P < 0.001) and female (60% and 56%, P < 0.01). The nonresponders were more likely to be hospitalized in the past 1 year (62% and 53%, P < 0.001) and have a lower CCI (CCI 03 [75% and 80%, P < 0.001]) compared to responders. Demographics between responders and nonresponders were also statistically different from one another.

Figure 1

Unadjusted results revealed that patients on the hospitalist service were more confident in their abilities to identify their physician(s) (50% vs 45%, P < 0.001), perceived greater ability in understanding the role of their physician(s) (54% vs 50%, P < 0.001), and reported greater satisfaction with coordination and teamwork (68% vs 64%, P = 0.006) and with overall care (73% vs 67%, P < 0.001) (Figure 2).

Figure 2

From the mixed‐effects regression models it was discovered that admission to the hospitalist service was associated with a higher odds ratio (OR) of reporting overall care as excellent or very good (OR: 1.33; 95% confidence interval [CI]: 1.15‐1.47). There was no difference between services in patients' ability to identify their physician(s) (OR: 0.89; 95% CI: 0.61‐1.11), in patients reporting a better understanding of the role of their physician(s) (OR: 1.09; 95% CI: 0.94‐1.23), or in their rating of overall coordination and teamwork (OR: 0.71; 95% CI: 0.42‐1.89).

A subgroup analysis was performed on the 25% of hospitalist attendings in the general medicine teaching service comparing this cohort to the hospitalist services, and it was found that patients perceived better overall care on the hospitalist service (OR: 1.17; 95% CI: 1.01‐ 1.31) than on the general medicine service (Table 2). All other domains in the subgroup analysis were not statistically significant. Finally, an ordinal logistic regression was performed for each of these outcomes, but it did not show any major differences compared to the logistic regression of dichotomous outcomes.

DISCUSSION

This study is the first to directly compare measures of patient experience on hospitalist and general medicine teaching services in a large, multiyear comparison across multiple domains. In adjusted analysis, we found that patients on nonteaching hospitalist services rated their overall care better than those on general medicine teaching services, whereas no differences in patients' ability to identify their physician(s), understand their role in their care, or rating of coordination of care were found. Although the magnitude of the differences in rating of overall care may appear small, it remains noteworthy because of the recent focus on patient experience at the reimbursement level, where small differences in performance can lead to large changes in payment. Because of the observational design of this study, it is important to consider mechanisms that could account for our findings.

The first are the structural differences between the 2 services. Our subgroup analysis comparing patients rating of overall care on a general medicine service with a hospitalist attending to a pure hospitalist cohort found a significant difference between the groups, indicating that the structural differences between the 2 groups may be a significant contributor to patient satisfaction ratings. Under the care of a hospitalist service, a patient would only interact with a single physician on a daily basis, possibly leading to a more meaningful relationship and improved communication between patient and provider. Alternatively, while on a general medicine teaching service, patients would likely interact with multiple physicians, as a result making their confidence in their ability to identify and perception at understanding physicians' roles more challenging.[18] This dilemma is further compounded by duty hour restrictions, which have subsequently led to increased fragmentation in housestaff scheduling. The patient experience on the general medicine teaching service may be further complicated by recent data that show residents spend a minority of time in direct patient care,[19, 20] which could additionally contribute to patients' inability to understand who their physicians are and to the decreased satisfaction with their care. This combination of structural complexity, duty hour reform, and reduced direct patient interaction would likely decrease the chance a patient will interact with the same resident on a consistent basis,[5, 21] thus making the ability to truly understand who their caretakers are, and the role they play, more difficult.

Another contributing factor could be the use of NPAs on our hospitalist service. Given that these providers often see the patient on a more continual basis, hospitalized patients' exposure to a single, continuous caretaker may be a factor in our findings.[22] Furthermore, with studies showing that hospitalists also spend a small fraction of their day in direct patient care,[23, 24, 25] the use of NPAs may allow our hospitalists to spend greater amounts of time with their patients, thus improving patients' rating of their overall care and influencing their perceived ability to understand their physician's role.

Although there was no difference between general medicine teaching and hospitalist services with respect to patient understanding of their roles, our data suggest that both groups would benefit from interventions to target this area. Focused attempts at improving patient's ability to identify and explain the roles of their inpatient physician(s) have been performed. For example, previous studies have attempted to improve a patient's ability to identify their physician through physician facecards[8, 9] or the use of other simple interventions (ie, bedside whiteboards).[4, 26] Results from such interventions are mixed, as they have demonstrated the capacity to improve patients' ability to identify who their physician is, whereas few have shown any appreciable improvement in patient satisfaction.[26]

Although our findings suggest that structural differences in team composition may be a possible explanation, it is also important to consider how the quality of care a patient receives affects their experience. For instance, hospitalists have been shown to produce moderate improvements in patient‐centered outcomes such as 30‐day readmission[27] and hospital length of stay[14, 28, 29, 30, 31] when compared to other care providers, which in turn could be reflected in the patient's perception of their overall care. In a large national study of acute care hospitals using the Hospital Consumer Assessment of Healthcare Providers and Systems survey, Chen and colleagues found that for most measures of patient satisfaction, hospitals with greater use of hospitalist care were associated with better patient‐centered care.[13] These outcomes were in part driven by patient‐centered domains such as discharge planning, pain control, and medication management. It is possible that patients are sensitive to the improved outcomes that are associated with hospitalist services, and reflect this in their measures of patient satisfaction.

Last, because this is an observational study and not a randomized trial, it is possible that the clinical differences in the patients cared for by these services could have led to our findings. Although the clinical significance of the differences in patient demographics were small, patients seen on the hospitalist service were more likely to be older white males, with a slightly longer LOS, greater comorbidities, and more hospitalizations in the previous year than those seen on the general medicine teaching service. Additionally, our hospitalist service frequently cares for highly specific subpopulations (ie, liver and renal transplant patients, and oncology patients), which could have influenced our results. For example, transplant patients who may be very grateful for their second chance, are preferentially admitted to the hospitalist service, which could have biased our results in favor of hospitalists.[32] Unfortunately, we were unable to control for all such factors.

Although we hope that multivariable analysis can adjust for many of these differences, we are not able to account for possible unmeasured confounders such as time of day of admission, health literacy, personality differences, physician turnover, or nursing and other ancillary care that could contribute to these findings. In addition to its observational study design, our study has several other limitations. First, our study was performed at a single institution, thus limiting its generalizability. Second, as a retrospective study based on observational data, no definitive conclusions regarding causality can be made. Third, although our response rate was low, it is comparable to other studies that have examined underserved populations.[33, 34] Fourth, because our survey was performed 30 days after hospitalization, this may impart imprecision on our outcomes measures. Finally, we were not able to mitigate selection bias through imputation for missing data .

All together, given the small absolute differences between the groups in patients' ratings of their overall care compared to large differences in possible confounders, these findings call for further exploration into the significance and possible mechanisms of these outcomes. Our study raises the potential possibility that the structural component of a care team may play a role in overall patient satisfaction. If this is the case, future studies of team structure could help inform how best to optimize this component for the patient experience. On the other hand, if process differences are to explain our findings, it is important to distill the types of processes hospitalists are using to improve the patient experience and potentially export this to resident services.

Finally, if similar results were found in other institutions, these findings could have implications on how hospitals respond to new payment models that are linked to patient‐experience measures. For example, the Hospital Value‐Based Purchasing Program currently links the Centers for Medicare and Medicaid Services payments to a set of quality measures that consist of (1) clinical processes of care (70%) and (2) the patient experience (30%).[1] Given this linkage, any small changes in the domain of patient satisfaction could have large payment implications on a national level.

CONCLUSION

In summary, in this large‐scale multiyear study, patients cared for by a nonteaching hospitalist service reported greater satisfaction with their overall care than patients cared for by a general medicine teaching service. This difference could be mediated by the structural differences between these 2 services. As hospitals seek to optimize patient experiences in an era where reimbursement models are now being linked to patient‐experience measures, future work should focus on further understanding the mechanisms for these findings.

The hospitalized patient experience has become an area of increased focus for hospitals given the recent coupling of patient satisfaction to reimbursement rates for Medicare patients.[1] Although patient experiences are multifactorial, 1 component is the relationship that hospitalized patients develop with their inpatient physicians. In recognition of the importance of this relationship, several organizations including the Society of Hospital Medicine, Society of General Internal Medicine, American College of Physicians, the American College of Emergency Physicians, and the Accreditation Council for Graduate Medical Education have recommended that patients know and understand who is guiding their care at all times during their hospitalization.[2, 3] Unfortunately, previous studies have shown that hospitalized patients often lack the ability to identify[4, 5] and understand their course of care.[6, 7] This may be due to numerous clinical factors including lack of a prior relationship, rapid pace of clinical care, and the frequent transitions of care found in both hospitalists and general medicine teaching services.[5, 8, 9] Regardless of the cause, one could hypothesize that patients who are unable to identify or understand the role of their physician may be less informed about their hospitalization, which may lead to further confusion, dissatisfaction, and ultimately a poor experience.

Given the proliferation of nonteaching hospitalist services in teaching hospitals, it is important to understand if patient experiences differ between general medicine teaching and hospitalist services. Several reasons could explain why patient experiences may vary on these services. For example, patients on a hospitalist service will likely interact with a single physician caretaker, which may give a feeling of more personalized care. In contrast, patients on general medicine teaching services are cared for by larger teams of residents under the supervision of an attending physician. Residents are also subjected to duty‐hour restrictions, clinic responsibilities, and other educational requirements that may impede the continuity of care for hospitalized patients.[10, 11, 12] Although 1 study has shown that hospitalist‐intensive hospitals perform better on patient satisfaction measures,[13] no study to date has compared patient‐reported experiences on general medicine teaching and nonteaching hospitalist services. This study aimed to evaluate the hospitalized patient experience on both teaching and nonteaching hospitalist services by assessing several patient‐reported measures of their experience, namely their confidence in their ability to identify their physician(s), understand their roles, and their rating of both the coordination and overall care.

METHODS

RESULTS

In total, 14,855 patients were enrolled during their hospitalization with 57% and 61% completing the 30‐day follow‐up survey on the hospitalist and general medicine teaching service, respectively. In total, 4131 (69%) and 4322 (48%) of the hospitalist and general medicine services, respectively, either did not answer all survey questions, or were missing basic demographic data, and thus were excluded. Data from 4591 patients on the general medicine teaching (52% of those enrolled at hospitalization) and 1811 on the hospitalist service (31% of those enrolled at hospitalization) were used for final analysis (Figure 1). Respondents were predominantly female (61% and 56%), African American (75% and 63%), with a mean age of 56.2 (19.4) and 57.1 (16.1) years, for the general medicine teaching and hospitalist services, respectively. A majority of patients (71% and 66%) had a CCI of 0 to 3 on both services. There were differences in self‐reported comorbidities between the 2 groups, with hospitalist services having a higher prevalence of cancer (20% vs 7%), renal disease (25% vs 18%), and liver disease (23% vs 7%). Patients on the hospitalist service had a longer mean LOS (5.5 vs 4.8 days), a greater percentage of a hospitalization within 1 year (58% vs 52%), and a larger proportion who were admitted in 2011 to 2013 compared to 2007 to 2010 (75% vs 39%), when compared to the general medicine teaching services. Median LOS and interquartile ranges were similar between both groups. Although most baseline demographics were statistically different between the 2 groups (Table 1), these differences were likely clinically insignificant. Compared to those who responded to the follow‐up survey, nonresponders were more likely to be African American (73% and 64%, P < 0.001) and female (60% and 56%, P < 0.01). The nonresponders were more likely to be hospitalized in the past 1 year (62% and 53%, P < 0.001) and have a lower CCI (CCI 03 [75% and 80%, P < 0.001]) compared to responders. Demographics between responders and nonresponders were also statistically different from one another.

Figure 1

Unadjusted results revealed that patients on the hospitalist service were more confident in their abilities to identify their physician(s) (50% vs 45%, P < 0.001), perceived greater ability in understanding the role of their physician(s) (54% vs 50%, P < 0.001), and reported greater satisfaction with coordination and teamwork (68% vs 64%, P = 0.006) and with overall care (73% vs 67%, P < 0.001) (Figure 2).

Figure 2

From the mixed‐effects regression models it was discovered that admission to the hospitalist service was associated with a higher odds ratio (OR) of reporting overall care as excellent or very good (OR: 1.33; 95% confidence interval [CI]: 1.15‐1.47). There was no difference between services in patients' ability to identify their physician(s) (OR: 0.89; 95% CI: 0.61‐1.11), in patients reporting a better understanding of the role of their physician(s) (OR: 1.09; 95% CI: 0.94‐1.23), or in their rating of overall coordination and teamwork (OR: 0.71; 95% CI: 0.42‐1.89).

A subgroup analysis was performed on the 25% of hospitalist attendings in the general medicine teaching service comparing this cohort to the hospitalist services, and it was found that patients perceived better overall care on the hospitalist service (OR: 1.17; 95% CI: 1.01‐ 1.31) than on the general medicine service (Table 2). All other domains in the subgroup analysis were not statistically significant. Finally, an ordinal logistic regression was performed for each of these outcomes, but it did not show any major differences compared to the logistic regression of dichotomous outcomes.

DISCUSSION

This study is the first to directly compare measures of patient experience on hospitalist and general medicine teaching services in a large, multiyear comparison across multiple domains. In adjusted analysis, we found that patients on nonteaching hospitalist services rated their overall care better than those on general medicine teaching services, whereas no differences in patients' ability to identify their physician(s), understand their role in their care, or rating of coordination of care were found. Although the magnitude of the differences in rating of overall care may appear small, it remains noteworthy because of the recent focus on patient experience at the reimbursement level, where small differences in performance can lead to large changes in payment. Because of the observational design of this study, it is important to consider mechanisms that could account for our findings.

The first are the structural differences between the 2 services. Our subgroup analysis comparing patients rating of overall care on a general medicine service with a hospitalist attending to a pure hospitalist cohort found a significant difference between the groups, indicating that the structural differences between the 2 groups may be a significant contributor to patient satisfaction ratings. Under the care of a hospitalist service, a patient would only interact with a single physician on a daily basis, possibly leading to a more meaningful relationship and improved communication between patient and provider. Alternatively, while on a general medicine teaching service, patients would likely interact with multiple physicians, as a result making their confidence in their ability to identify and perception at understanding physicians' roles more challenging.[18] This dilemma is further compounded by duty hour restrictions, which have subsequently led to increased fragmentation in housestaff scheduling. The patient experience on the general medicine teaching service may be further complicated by recent data that show residents spend a minority of time in direct patient care,[19, 20] which could additionally contribute to patients' inability to understand who their physicians are and to the decreased satisfaction with their care. This combination of structural complexity, duty hour reform, and reduced direct patient interaction would likely decrease the chance a patient will interact with the same resident on a consistent basis,[5, 21] thus making the ability to truly understand who their caretakers are, and the role they play, more difficult.

Another contributing factor could be the use of NPAs on our hospitalist service. Given that these providers often see the patient on a more continual basis, hospitalized patients' exposure to a single, continuous caretaker may be a factor in our findings.[22] Furthermore, with studies showing that hospitalists also spend a small fraction of their day in direct patient care,[23, 24, 25] the use of NPAs may allow our hospitalists to spend greater amounts of time with their patients, thus improving patients' rating of their overall care and influencing their perceived ability to understand their physician's role.

Although there was no difference between general medicine teaching and hospitalist services with respect to patient understanding of their roles, our data suggest that both groups would benefit from interventions to target this area. Focused attempts at improving patient's ability to identify and explain the roles of their inpatient physician(s) have been performed. For example, previous studies have attempted to improve a patient's ability to identify their physician through physician facecards[8, 9] or the use of other simple interventions (ie, bedside whiteboards).[4, 26] Results from such interventions are mixed, as they have demonstrated the capacity to improve patients' ability to identify who their physician is, whereas few have shown any appreciable improvement in patient satisfaction.[26]

Although our findings suggest that structural differences in team composition may be a possible explanation, it is also important to consider how the quality of care a patient receives affects their experience. For instance, hospitalists have been shown to produce moderate improvements in patient‐centered outcomes such as 30‐day readmission[27] and hospital length of stay[14, 28, 29, 30, 31] when compared to other care providers, which in turn could be reflected in the patient's perception of their overall care. In a large national study of acute care hospitals using the Hospital Consumer Assessment of Healthcare Providers and Systems survey, Chen and colleagues found that for most measures of patient satisfaction, hospitals with greater use of hospitalist care were associated with better patient‐centered care.[13] These outcomes were in part driven by patient‐centered domains such as discharge planning, pain control, and medication management. It is possible that patients are sensitive to the improved outcomes that are associated with hospitalist services, and reflect this in their measures of patient satisfaction.

Last, because this is an observational study and not a randomized trial, it is possible that the clinical differences in the patients cared for by these services could have led to our findings. Although the clinical significance of the differences in patient demographics were small, patients seen on the hospitalist service were more likely to be older white males, with a slightly longer LOS, greater comorbidities, and more hospitalizations in the previous year than those seen on the general medicine teaching service. Additionally, our hospitalist service frequently cares for highly specific subpopulations (ie, liver and renal transplant patients, and oncology patients), which could have influenced our results. For example, transplant patients who may be very grateful for their second chance, are preferentially admitted to the hospitalist service, which could have biased our results in favor of hospitalists.[32] Unfortunately, we were unable to control for all such factors.

Although we hope that multivariable analysis can adjust for many of these differences, we are not able to account for possible unmeasured confounders such as time of day of admission, health literacy, personality differences, physician turnover, or nursing and other ancillary care that could contribute to these findings. In addition to its observational study design, our study has several other limitations. First, our study was performed at a single institution, thus limiting its generalizability. Second, as a retrospective study based on observational data, no definitive conclusions regarding causality can be made. Third, although our response rate was low, it is comparable to other studies that have examined underserved populations.[33, 34] Fourth, because our survey was performed 30 days after hospitalization, this may impart imprecision on our outcomes measures. Finally, we were not able to mitigate selection bias through imputation for missing data .

All together, given the small absolute differences between the groups in patients' ratings of their overall care compared to large differences in possible confounders, these findings call for further exploration into the significance and possible mechanisms of these outcomes. Our study raises the potential possibility that the structural component of a care team may play a role in overall patient satisfaction. If this is the case, future studies of team structure could help inform how best to optimize this component for the patient experience. On the other hand, if process differences are to explain our findings, it is important to distill the types of processes hospitalists are using to improve the patient experience and potentially export this to resident services.

Finally, if similar results were found in other institutions, these findings could have implications on how hospitals respond to new payment models that are linked to patient‐experience measures. For example, the Hospital Value‐Based Purchasing Program currently links the Centers for Medicare and Medicaid Services payments to a set of quality measures that consist of (1) clinical processes of care (70%) and (2) the patient experience (30%).[1] Given this linkage, any small changes in the domain of patient satisfaction could have large payment implications on a national level.

CONCLUSION

In summary, in this large‐scale multiyear study, patients cared for by a nonteaching hospitalist service reported greater satisfaction with their overall care than patients cared for by a general medicine teaching service. This difference could be mediated by the structural differences between these 2 services. As hospitals seek to optimize patient experiences in an era where reimbursement models are now being linked to patient‐experience measures, future work should focus on further understanding the mechanisms for these findings.

The hospitalized patient experience has become an area of increased focus for hospitals given the recent coupling of patient satisfaction to reimbursement rates for Medicare patients.[1] Although patient experiences are multifactorial, 1 component is the relationship that hospitalized patients develop with their inpatient physicians. In recognition of the importance of this relationship, several organizations including the Society of Hospital Medicine, Society of General Internal Medicine, American College of Physicians, the American College of Emergency Physicians, and the Accreditation Council for Graduate Medical Education have recommended that patients know and understand who is guiding their care at all times during their hospitalization.[2, 3] Unfortunately, previous studies have shown that hospitalized patients often lack the ability to identify[4, 5] and understand their course of care.[6, 7] This may be due to numerous clinical factors including lack of a prior relationship, rapid pace of clinical care, and the frequent transitions of care found in both hospitalists and general medicine teaching services.[5, 8, 9] Regardless of the cause, one could hypothesize that patients who are unable to identify or understand the role of their physician may be less informed about their hospitalization, which may lead to further confusion, dissatisfaction, and ultimately a poor experience.

Given the proliferation of nonteaching hospitalist services in teaching hospitals, it is important to understand if patient experiences differ between general medicine teaching and hospitalist services. Several reasons could explain why patient experiences may vary on these services. For example, patients on a hospitalist service will likely interact with a single physician caretaker, which may give a feeling of more personalized care. In contrast, patients on general medicine teaching services are cared for by larger teams of residents under the supervision of an attending physician. Residents are also subjected to duty‐hour restrictions, clinic responsibilities, and other educational requirements that may impede the continuity of care for hospitalized patients.[10, 11, 12] Although 1 study has shown that hospitalist‐intensive hospitals perform better on patient satisfaction measures,[13] no study to date has compared patient‐reported experiences on general medicine teaching and nonteaching hospitalist services. This study aimed to evaluate the hospitalized patient experience on both teaching and nonteaching hospitalist services by assessing several patient‐reported measures of their experience, namely their confidence in their ability to identify their physician(s), understand their roles, and their rating of both the coordination and overall care.

METHODS

RESULTS

In total, 14,855 patients were enrolled during their hospitalization with 57% and 61% completing the 30‐day follow‐up survey on the hospitalist and general medicine teaching service, respectively. In total, 4131 (69%) and 4322 (48%) of the hospitalist and general medicine services, respectively, either did not answer all survey questions, or were missing basic demographic data, and thus were excluded. Data from 4591 patients on the general medicine teaching (52% of those enrolled at hospitalization) and 1811 on the hospitalist service (31% of those enrolled at hospitalization) were used for final analysis (Figure 1). Respondents were predominantly female (61% and 56%), African American (75% and 63%), with a mean age of 56.2 (19.4) and 57.1 (16.1) years, for the general medicine teaching and hospitalist services, respectively. A majority of patients (71% and 66%) had a CCI of 0 to 3 on both services. There were differences in self‐reported comorbidities between the 2 groups, with hospitalist services having a higher prevalence of cancer (20% vs 7%), renal disease (25% vs 18%), and liver disease (23% vs 7%). Patients on the hospitalist service had a longer mean LOS (5.5 vs 4.8 days), a greater percentage of a hospitalization within 1 year (58% vs 52%), and a larger proportion who were admitted in 2011 to 2013 compared to 2007 to 2010 (75% vs 39%), when compared to the general medicine teaching services. Median LOS and interquartile ranges were similar between both groups. Although most baseline demographics were statistically different between the 2 groups (Table 1), these differences were likely clinically insignificant. Compared to those who responded to the follow‐up survey, nonresponders were more likely to be African American (73% and 64%, P < 0.001) and female (60% and 56%, P < 0.01). The nonresponders were more likely to be hospitalized in the past 1 year (62% and 53%, P < 0.001) and have a lower CCI (CCI 03 [75% and 80%, P < 0.001]) compared to responders. Demographics between responders and nonresponders were also statistically different from one another.

Figure 1

Unadjusted results revealed that patients on the hospitalist service were more confident in their abilities to identify their physician(s) (50% vs 45%, P < 0.001), perceived greater ability in understanding the role of their physician(s) (54% vs 50%, P < 0.001), and reported greater satisfaction with coordination and teamwork (68% vs 64%, P = 0.006) and with overall care (73% vs 67%, P < 0.001) (Figure 2).

Figure 2

From the mixed‐effects regression models it was discovered that admission to the hospitalist service was associated with a higher odds ratio (OR) of reporting overall care as excellent or very good (OR: 1.33; 95% confidence interval [CI]: 1.15‐1.47). There was no difference between services in patients' ability to identify their physician(s) (OR: 0.89; 95% CI: 0.61‐1.11), in patients reporting a better understanding of the role of their physician(s) (OR: 1.09; 95% CI: 0.94‐1.23), or in their rating of overall coordination and teamwork (OR: 0.71; 95% CI: 0.42‐1.89).

A subgroup analysis was performed on the 25% of hospitalist attendings in the general medicine teaching service comparing this cohort to the hospitalist services, and it was found that patients perceived better overall care on the hospitalist service (OR: 1.17; 95% CI: 1.01‐ 1.31) than on the general medicine service (Table 2). All other domains in the subgroup analysis were not statistically significant. Finally, an ordinal logistic regression was performed for each of these outcomes, but it did not show any major differences compared to the logistic regression of dichotomous outcomes.

DISCUSSION

This study is the first to directly compare measures of patient experience on hospitalist and general medicine teaching services in a large, multiyear comparison across multiple domains. In adjusted analysis, we found that patients on nonteaching hospitalist services rated their overall care better than those on general medicine teaching services, whereas no differences in patients' ability to identify their physician(s), understand their role in their care, or rating of coordination of care were found. Although the magnitude of the differences in rating of overall care may appear small, it remains noteworthy because of the recent focus on patient experience at the reimbursement level, where small differences in performance can lead to large changes in payment. Because of the observational design of this study, it is important to consider mechanisms that could account for our findings.

The first are the structural differences between the 2 services. Our subgroup analysis comparing patients rating of overall care on a general medicine service with a hospitalist attending to a pure hospitalist cohort found a significant difference between the groups, indicating that the structural differences between the 2 groups may be a significant contributor to patient satisfaction ratings. Under the care of a hospitalist service, a patient would only interact with a single physician on a daily basis, possibly leading to a more meaningful relationship and improved communication between patient and provider. Alternatively, while on a general medicine teaching service, patients would likely interact with multiple physicians, as a result making their confidence in their ability to identify and perception at understanding physicians' roles more challenging.[18] This dilemma is further compounded by duty hour restrictions, which have subsequently led to increased fragmentation in housestaff scheduling. The patient experience on the general medicine teaching service may be further complicated by recent data that show residents spend a minority of time in direct patient care,[19, 20] which could additionally contribute to patients' inability to understand who their physicians are and to the decreased satisfaction with their care. This combination of structural complexity, duty hour reform, and reduced direct patient interaction would likely decrease the chance a patient will interact with the same resident on a consistent basis,[5, 21] thus making the ability to truly understand who their caretakers are, and the role they play, more difficult.

Another contributing factor could be the use of NPAs on our hospitalist service. Given that these providers often see the patient on a more continual basis, hospitalized patients' exposure to a single, continuous caretaker may be a factor in our findings.[22] Furthermore, with studies showing that hospitalists also spend a small fraction of their day in direct patient care,[23, 24, 25] the use of NPAs may allow our hospitalists to spend greater amounts of time with their patients, thus improving patients' rating of their overall care and influencing their perceived ability to understand their physician's role.

Although there was no difference between general medicine teaching and hospitalist services with respect to patient understanding of their roles, our data suggest that both groups would benefit from interventions to target this area. Focused attempts at improving patient's ability to identify and explain the roles of their inpatient physician(s) have been performed. For example, previous studies have attempted to improve a patient's ability to identify their physician through physician facecards[8, 9] or the use of other simple interventions (ie, bedside whiteboards).[4, 26] Results from such interventions are mixed, as they have demonstrated the capacity to improve patients' ability to identify who their physician is, whereas few have shown any appreciable improvement in patient satisfaction.[26]

Although our findings suggest that structural differences in team composition may be a possible explanation, it is also important to consider how the quality of care a patient receives affects their experience. For instance, hospitalists have been shown to produce moderate improvements in patient‐centered outcomes such as 30‐day readmission[27] and hospital length of stay[14, 28, 29, 30, 31] when compared to other care providers, which in turn could be reflected in the patient's perception of their overall care. In a large national study of acute care hospitals using the Hospital Consumer Assessment of Healthcare Providers and Systems survey, Chen and colleagues found that for most measures of patient satisfaction, hospitals with greater use of hospitalist care were associated with better patient‐centered care.[13] These outcomes were in part driven by patient‐centered domains such as discharge planning, pain control, and medication management. It is possible that patients are sensitive to the improved outcomes that are associated with hospitalist services, and reflect this in their measures of patient satisfaction.

Last, because this is an observational study and not a randomized trial, it is possible that the clinical differences in the patients cared for by these services could have led to our findings. Although the clinical significance of the differences in patient demographics were small, patients seen on the hospitalist service were more likely to be older white males, with a slightly longer LOS, greater comorbidities, and more hospitalizations in the previous year than those seen on the general medicine teaching service. Additionally, our hospitalist service frequently cares for highly specific subpopulations (ie, liver and renal transplant patients, and oncology patients), which could have influenced our results. For example, transplant patients who may be very grateful for their second chance, are preferentially admitted to the hospitalist service, which could have biased our results in favor of hospitalists.[32] Unfortunately, we were unable to control for all such factors.

Although we hope that multivariable analysis can adjust for many of these differences, we are not able to account for possible unmeasured confounders such as time of day of admission, health literacy, personality differences, physician turnover, or nursing and other ancillary care that could contribute to these findings. In addition to its observational study design, our study has several other limitations. First, our study was performed at a single institution, thus limiting its generalizability. Second, as a retrospective study based on observational data, no definitive conclusions regarding causality can be made. Third, although our response rate was low, it is comparable to other studies that have examined underserved populations.[33, 34] Fourth, because our survey was performed 30 days after hospitalization, this may impart imprecision on our outcomes measures. Finally, we were not able to mitigate selection bias through imputation for missing data .

All together, given the small absolute differences between the groups in patients' ratings of their overall care compared to large differences in possible confounders, these findings call for further exploration into the significance and possible mechanisms of these outcomes. Our study raises the potential possibility that the structural component of a care team may play a role in overall patient satisfaction. If this is the case, future studies of team structure could help inform how best to optimize this component for the patient experience. On the other hand, if process differences are to explain our findings, it is important to distill the types of processes hospitalists are using to improve the patient experience and potentially export this to resident services.

Finally, if similar results were found in other institutions, these findings could have implications on how hospitals respond to new payment models that are linked to patient‐experience measures. For example, the Hospital Value‐Based Purchasing Program currently links the Centers for Medicare and Medicaid Services payments to a set of quality measures that consist of (1) clinical processes of care (70%) and (2) the patient experience (30%).[1] Given this linkage, any small changes in the domain of patient satisfaction could have large payment implications on a national level.

CONCLUSION

In summary, in this large‐scale multiyear study, patients cared for by a nonteaching hospitalist service reported greater satisfaction with their overall care than patients cared for by a general medicine teaching service. This difference could be mediated by the structural differences between these 2 services. As hospitals seek to optimize patient experiences in an era where reimbursement models are now being linked to patient‐experience measures, future work should focus on further understanding the mechanisms for these findings.