Regina Landis, BA

In 2014, there were more than 40,000 hospitalists in the United States, and approximately 20% were employed by academic medical centers.[1] Hospitalist physicians groups are committed to delivering excellent patient care. However, the published literature is limited with respect to defining optimal care in hospital medicine.

Patient satisfaction surveys, such as Press Ganey (PG)[2] and Hospital Consumer Assessment of Healthcare Providers and Systems,[3] are being used to assess patients' contentment with the quality of care they receive while hospitalized. The Society of Hospital Medicine, the largest professional medical society representing hospitalists, encourages the use of patient satisfaction surveys to measure hospitalist providers' quality of patient care.[4] There are, however, several problems with the current methods. First, the attribution to specific providers is questionable. Second, recall about the provider by the patients may be poor because surveys are sent to patients days after they return home. Third, the patients' recovery and health outcomes are likely to influence their assessment of the doctor. Finally, feedback is known to be most valuable and transformative when it is specific and given in real time. Thus, a tool that is able to provide feedback at the encounter level should be more helpful than a tool that offers assessment at the level of the admission, particularly when it can be also delivered immediately after the data are collected.

Comportment has been used to describe both the way a person behaves and also the way she carries herself (ie, her general manner).[5] Excellent comportment and communication can serve as the foundation for delivering patient‐centered care.[6, 7, 8] Patient centeredness has been shown to improve the patient experience and clinical outcomes, including compliance with therapeutic plans.[9, 10, 11] Respectful behavior, etiquette‐based medicine, and effective communication also lay the foundation upon which the therapeutic alliance between a doctor and patient can be built.

The goal of this study was to establish a metric that could comprehensively assess a hospitalist provider's comportment and communication skills during an encounter with a hospitalized patient.

METHODS

RESULTS

The average clinical experience of the 26 hospitalist physicians studied was 6 years (Table 1). Their mean age was 38 years, 13 (50%) were female, and 16 (62%) were of nonwhite race. Fourteen hospitalists (54%) worked at 1 of the nonacademic hospitals. In terms of clinical workload, most physicians (n = 17, 65%) devoted more than 70% of their time working in direct patient care. Mean time spent observing each physician was 280 minutes. During this time, the 26 physicians were observed for 181 separate clinical encounters; 54% of these patients were new encounters, patients who were not previously known to the physician. The average time each physician spent in a patient room was 10.8 minutes. Mean number of observed patient encounters per hospitalist was 7.

The distribution of HMCCOT scores was not statistically significantly different when analyzed by age, gender, race, amount of clinical experience, clinical workload of the hospitalist, hospital, time spent observing the hospitalist (all P > 0.05). The distribution of HMCCOT scores was statistically different in new patient encounters compared to follow‐ups (68.1% vs 39.7%, P 0.001). Encounters with patients that generated HMCCOT scores above versus below the mean were longer (13 minutes vs 8.7 minutes, P 0.001).

The mean HMCCOT score was 61 (standard deviation [SD] = 10.6), and it was normally distributed (Figure 1). Table 2 shows the data for the 23 behaviors that were objectively assessed as part of the HMCCOT for the 181 patient encounters. The most frequently observed behaviors were physicians washing hands after leaving the patient's room in 170 (94%) of the encounters and smiling (83%). The behaviors that were observed with the least regularity were using an empathic statement (26% of encounters), and employing teach‐back (13% of encounters). A common method of demonstrating interest in the patient as a person, seen in 41% of encounters, involved physicians asking about patients' personal histories and their interests.

jhm2647-fig-0001-m.png

The average composite PG scores for the physician sample was 38.95 (SD=39.64). A moderate correlation was found between the HMCCOT score and PG score (adjusted Pearson correlation: 0.45, P = 0.047).

DISCUSSION

In this study, we followed 26 hospitalist physicians during routine clinical care, and we focused intently on their communication and their comportment with patients at the bedside. Even among clinically respected hospitalists, the results reveal that there is wide variability in comportment and communication practices and behaviors at the bedside. The physicians' HMCCOT scores were associated with their PG scores. These findings suggest that improved bedside communication and comportment with patients might translate into enhanced patient satisfaction.

This is the first study that honed in on hospitalist communication and comportment. With validity evidence established for the HMCCOT, some may elect to more explicitly perform these behaviors themselves, and others may wish to watch other hospitalists to give them feedback that is tied to specific behaviors. Beginning with the basics, the hospitalists we studied introduced themselves to their patients at the initial encounter 78% of the time, less frequently than is done by primary care clinicians (89%) but more consistently than do emergency department providers (64%).[7] Other variables that stood out in the HMCCOT was that teach‐back was employed in only 13% of the encounters. Previous studies have shown that teach‐back corroborates patient comprehension and can be used to engage patients (and caregivers) in realistic goal setting and optimal health service utilization.[14] Further, patients who clearly understand their postdischarge plan are 30% less likely to be readmitted or visit the emergency department.[14] The data for our group have helped us to see areas of strengths, such as hand washing, where we are above compliance rates across hospitals in the United States,[15] as well as those matters that represent opportunities for improvement such as connecting more deeply with our patients.

Tackett et al. have looked at encounter length and its association with performance of etiquette‐based medicine behaviors.[7] Similar to their study, we found a positive correlation between spending more time with patients and higher HMCCOT scores. We also found that HMCCOT scores were higher when providers were caring for new patients. Patients' complaints about doctors often relate to feeling rushed, that their physicians did not listen to them, or that information was not conveyed in a clear manner.[16] Such challenges in physicianpatient communication are ubiquitous across clinical settings.[16] When successfully achieved, patient‐centered communication has been associated with improved clinical outcomes, including adherence to recommended treatment and better self‐management of chronic disease.[17, 18, 19, 20, 21, 22, 23, 24, 25, 26] Many of the components of the HMCCOT described in this article are at the heart of patient‐centered care.

Several limitations of the study should be considered. First, physicians may have behaved differently while they were being observed, which is known as the Hawthorne effect. We observed them for many hours and across multiple patient encounters, and the physicians were not aware of the specific types of data that we were collecting. These factors may have limited the biases along such lines. Second, there may be elements of optimal comportment and communication that were not captured by the HMCCOT. Hopefully, there are not big gaps, as we used multiple methods and an iterative process in the refinement of the HMCCOT metric. Third, one investigator did all of the observing, and it is possible that he might have missed certain behaviors. Through extensive pilot testing and comparisons with other raters, the observer became very skilled and facile with such data collection and the tool. Fourth, we did not survey the same patients that were cared for to compare their perspectives to the HMCCOT scores following the clinical encounters. For patient perspectives, we relied only on PG scores. Fifth, quality of care is a broad and multidimensional construct. The HMCCOT focuses exclusively on hospitalists' comportment and communication at the bedside; therefore, it does not comprehensively assess care quality. Sixth, with our goal to optimally validate the HMCCOT, we tested it on the top tier of hospitalists within each group. We may have observed different results had we randomly selected hospitalists from each hospital or had we conducted the study at hospitals in other geographic regions. Finally, all of the doctors observed worked at hospitals in the Mid‐Atlantic region. However, these five distinct hospitals each have their own cultures, and they are led by different administrators. We purposively chose to sample both academic as well as community settings.

In conclusion, this study reports on the development of a comportment and communication tool that was established and validated by following clinically excellent hospitalists at the bedside. Future studies are necessary to determine whether hospitalists of all levels of experience and clinical skill can improve when given data and feedback using the HMCCOT. Larger studies will then be needed to assess whether enhancing comportment and communication can truly improve patient satisfaction and clinical outcomes in the hospital.

Disclosures: Dr. Wright is a Miller‐Coulson Family Scholar and is supported through the Johns Hopkins Center for Innovative Medicine. Susrutha Kotwal, MD, and Waseem Khaliq, MD, contributed equally to this work. The authors report no conflicts of interest.

In 2014, there were more than 40,000 hospitalists in the United States, and approximately 20% were employed by academic medical centers.[1] Hospitalist physicians groups are committed to delivering excellent patient care. However, the published literature is limited with respect to defining optimal care in hospital medicine.

Patient satisfaction surveys, such as Press Ganey (PG)[2] and Hospital Consumer Assessment of Healthcare Providers and Systems,[3] are being used to assess patients' contentment with the quality of care they receive while hospitalized. The Society of Hospital Medicine, the largest professional medical society representing hospitalists, encourages the use of patient satisfaction surveys to measure hospitalist providers' quality of patient care.[4] There are, however, several problems with the current methods. First, the attribution to specific providers is questionable. Second, recall about the provider by the patients may be poor because surveys are sent to patients days after they return home. Third, the patients' recovery and health outcomes are likely to influence their assessment of the doctor. Finally, feedback is known to be most valuable and transformative when it is specific and given in real time. Thus, a tool that is able to provide feedback at the encounter level should be more helpful than a tool that offers assessment at the level of the admission, particularly when it can be also delivered immediately after the data are collected.

Comportment has been used to describe both the way a person behaves and also the way she carries herself (ie, her general manner).[5] Excellent comportment and communication can serve as the foundation for delivering patient‐centered care.[6, 7, 8] Patient centeredness has been shown to improve the patient experience and clinical outcomes, including compliance with therapeutic plans.[9, 10, 11] Respectful behavior, etiquette‐based medicine, and effective communication also lay the foundation upon which the therapeutic alliance between a doctor and patient can be built.

The goal of this study was to establish a metric that could comprehensively assess a hospitalist provider's comportment and communication skills during an encounter with a hospitalized patient.

METHODS

RESULTS

The average clinical experience of the 26 hospitalist physicians studied was 6 years (Table 1). Their mean age was 38 years, 13 (50%) were female, and 16 (62%) were of nonwhite race. Fourteen hospitalists (54%) worked at 1 of the nonacademic hospitals. In terms of clinical workload, most physicians (n = 17, 65%) devoted more than 70% of their time working in direct patient care. Mean time spent observing each physician was 280 minutes. During this time, the 26 physicians were observed for 181 separate clinical encounters; 54% of these patients were new encounters, patients who were not previously known to the physician. The average time each physician spent in a patient room was 10.8 minutes. Mean number of observed patient encounters per hospitalist was 7.

The distribution of HMCCOT scores was not statistically significantly different when analyzed by age, gender, race, amount of clinical experience, clinical workload of the hospitalist, hospital, time spent observing the hospitalist (all P > 0.05). The distribution of HMCCOT scores was statistically different in new patient encounters compared to follow‐ups (68.1% vs 39.7%, P 0.001). Encounters with patients that generated HMCCOT scores above versus below the mean were longer (13 minutes vs 8.7 minutes, P 0.001).

The mean HMCCOT score was 61 (standard deviation [SD] = 10.6), and it was normally distributed (Figure 1). Table 2 shows the data for the 23 behaviors that were objectively assessed as part of the HMCCOT for the 181 patient encounters. The most frequently observed behaviors were physicians washing hands after leaving the patient's room in 170 (94%) of the encounters and smiling (83%). The behaviors that were observed with the least regularity were using an empathic statement (26% of encounters), and employing teach‐back (13% of encounters). A common method of demonstrating interest in the patient as a person, seen in 41% of encounters, involved physicians asking about patients' personal histories and their interests.

jhm2647-fig-0001-m.png

The average composite PG scores for the physician sample was 38.95 (SD=39.64). A moderate correlation was found between the HMCCOT score and PG score (adjusted Pearson correlation: 0.45, P = 0.047).

DISCUSSION

In this study, we followed 26 hospitalist physicians during routine clinical care, and we focused intently on their communication and their comportment with patients at the bedside. Even among clinically respected hospitalists, the results reveal that there is wide variability in comportment and communication practices and behaviors at the bedside. The physicians' HMCCOT scores were associated with their PG scores. These findings suggest that improved bedside communication and comportment with patients might translate into enhanced patient satisfaction.

This is the first study that honed in on hospitalist communication and comportment. With validity evidence established for the HMCCOT, some may elect to more explicitly perform these behaviors themselves, and others may wish to watch other hospitalists to give them feedback that is tied to specific behaviors. Beginning with the basics, the hospitalists we studied introduced themselves to their patients at the initial encounter 78% of the time, less frequently than is done by primary care clinicians (89%) but more consistently than do emergency department providers (64%).[7] Other variables that stood out in the HMCCOT was that teach‐back was employed in only 13% of the encounters. Previous studies have shown that teach‐back corroborates patient comprehension and can be used to engage patients (and caregivers) in realistic goal setting and optimal health service utilization.[14] Further, patients who clearly understand their postdischarge plan are 30% less likely to be readmitted or visit the emergency department.[14] The data for our group have helped us to see areas of strengths, such as hand washing, where we are above compliance rates across hospitals in the United States,[15] as well as those matters that represent opportunities for improvement such as connecting more deeply with our patients.

Tackett et al. have looked at encounter length and its association with performance of etiquette‐based medicine behaviors.[7] Similar to their study, we found a positive correlation between spending more time with patients and higher HMCCOT scores. We also found that HMCCOT scores were higher when providers were caring for new patients. Patients' complaints about doctors often relate to feeling rushed, that their physicians did not listen to them, or that information was not conveyed in a clear manner.[16] Such challenges in physicianpatient communication are ubiquitous across clinical settings.[16] When successfully achieved, patient‐centered communication has been associated with improved clinical outcomes, including adherence to recommended treatment and better self‐management of chronic disease.[17, 18, 19, 20, 21, 22, 23, 24, 25, 26] Many of the components of the HMCCOT described in this article are at the heart of patient‐centered care.

Several limitations of the study should be considered. First, physicians may have behaved differently while they were being observed, which is known as the Hawthorne effect. We observed them for many hours and across multiple patient encounters, and the physicians were not aware of the specific types of data that we were collecting. These factors may have limited the biases along such lines. Second, there may be elements of optimal comportment and communication that were not captured by the HMCCOT. Hopefully, there are not big gaps, as we used multiple methods and an iterative process in the refinement of the HMCCOT metric. Third, one investigator did all of the observing, and it is possible that he might have missed certain behaviors. Through extensive pilot testing and comparisons with other raters, the observer became very skilled and facile with such data collection and the tool. Fourth, we did not survey the same patients that were cared for to compare their perspectives to the HMCCOT scores following the clinical encounters. For patient perspectives, we relied only on PG scores. Fifth, quality of care is a broad and multidimensional construct. The HMCCOT focuses exclusively on hospitalists' comportment and communication at the bedside; therefore, it does not comprehensively assess care quality. Sixth, with our goal to optimally validate the HMCCOT, we tested it on the top tier of hospitalists within each group. We may have observed different results had we randomly selected hospitalists from each hospital or had we conducted the study at hospitals in other geographic regions. Finally, all of the doctors observed worked at hospitals in the Mid‐Atlantic region. However, these five distinct hospitals each have their own cultures, and they are led by different administrators. We purposively chose to sample both academic as well as community settings.

In conclusion, this study reports on the development of a comportment and communication tool that was established and validated by following clinically excellent hospitalists at the bedside. Future studies are necessary to determine whether hospitalists of all levels of experience and clinical skill can improve when given data and feedback using the HMCCOT. Larger studies will then be needed to assess whether enhancing comportment and communication can truly improve patient satisfaction and clinical outcomes in the hospital.

Disclosures: Dr. Wright is a Miller‐Coulson Family Scholar and is supported through the Johns Hopkins Center for Innovative Medicine. Susrutha Kotwal, MD, and Waseem Khaliq, MD, contributed equally to this work. The authors report no conflicts of interest.

In 2014, there were more than 40,000 hospitalists in the United States, and approximately 20% were employed by academic medical centers.[1] Hospitalist physicians groups are committed to delivering excellent patient care. However, the published literature is limited with respect to defining optimal care in hospital medicine.

Patient satisfaction surveys, such as Press Ganey (PG)[2] and Hospital Consumer Assessment of Healthcare Providers and Systems,[3] are being used to assess patients' contentment with the quality of care they receive while hospitalized. The Society of Hospital Medicine, the largest professional medical society representing hospitalists, encourages the use of patient satisfaction surveys to measure hospitalist providers' quality of patient care.[4] There are, however, several problems with the current methods. First, the attribution to specific providers is questionable. Second, recall about the provider by the patients may be poor because surveys are sent to patients days after they return home. Third, the patients' recovery and health outcomes are likely to influence their assessment of the doctor. Finally, feedback is known to be most valuable and transformative when it is specific and given in real time. Thus, a tool that is able to provide feedback at the encounter level should be more helpful than a tool that offers assessment at the level of the admission, particularly when it can be also delivered immediately after the data are collected.

Comportment has been used to describe both the way a person behaves and also the way she carries herself (ie, her general manner).[5] Excellent comportment and communication can serve as the foundation for delivering patient‐centered care.[6, 7, 8] Patient centeredness has been shown to improve the patient experience and clinical outcomes, including compliance with therapeutic plans.[9, 10, 11] Respectful behavior, etiquette‐based medicine, and effective communication also lay the foundation upon which the therapeutic alliance between a doctor and patient can be built.

The goal of this study was to establish a metric that could comprehensively assess a hospitalist provider's comportment and communication skills during an encounter with a hospitalized patient.

METHODS

RESULTS

The average clinical experience of the 26 hospitalist physicians studied was 6 years (Table 1). Their mean age was 38 years, 13 (50%) were female, and 16 (62%) were of nonwhite race. Fourteen hospitalists (54%) worked at 1 of the nonacademic hospitals. In terms of clinical workload, most physicians (n = 17, 65%) devoted more than 70% of their time working in direct patient care. Mean time spent observing each physician was 280 minutes. During this time, the 26 physicians were observed for 181 separate clinical encounters; 54% of these patients were new encounters, patients who were not previously known to the physician. The average time each physician spent in a patient room was 10.8 minutes. Mean number of observed patient encounters per hospitalist was 7.

The distribution of HMCCOT scores was not statistically significantly different when analyzed by age, gender, race, amount of clinical experience, clinical workload of the hospitalist, hospital, time spent observing the hospitalist (all P > 0.05). The distribution of HMCCOT scores was statistically different in new patient encounters compared to follow‐ups (68.1% vs 39.7%, P 0.001). Encounters with patients that generated HMCCOT scores above versus below the mean were longer (13 minutes vs 8.7 minutes, P 0.001).

The mean HMCCOT score was 61 (standard deviation [SD] = 10.6), and it was normally distributed (Figure 1). Table 2 shows the data for the 23 behaviors that were objectively assessed as part of the HMCCOT for the 181 patient encounters. The most frequently observed behaviors were physicians washing hands after leaving the patient's room in 170 (94%) of the encounters and smiling (83%). The behaviors that were observed with the least regularity were using an empathic statement (26% of encounters), and employing teach‐back (13% of encounters). A common method of demonstrating interest in the patient as a person, seen in 41% of encounters, involved physicians asking about patients' personal histories and their interests.

jhm2647-fig-0001-m.png

The average composite PG scores for the physician sample was 38.95 (SD=39.64). A moderate correlation was found between the HMCCOT score and PG score (adjusted Pearson correlation: 0.45, P = 0.047).

DISCUSSION

In this study, we followed 26 hospitalist physicians during routine clinical care, and we focused intently on their communication and their comportment with patients at the bedside. Even among clinically respected hospitalists, the results reveal that there is wide variability in comportment and communication practices and behaviors at the bedside. The physicians' HMCCOT scores were associated with their PG scores. These findings suggest that improved bedside communication and comportment with patients might translate into enhanced patient satisfaction.

This is the first study that honed in on hospitalist communication and comportment. With validity evidence established for the HMCCOT, some may elect to more explicitly perform these behaviors themselves, and others may wish to watch other hospitalists to give them feedback that is tied to specific behaviors. Beginning with the basics, the hospitalists we studied introduced themselves to their patients at the initial encounter 78% of the time, less frequently than is done by primary care clinicians (89%) but more consistently than do emergency department providers (64%).[7] Other variables that stood out in the HMCCOT was that teach‐back was employed in only 13% of the encounters. Previous studies have shown that teach‐back corroborates patient comprehension and can be used to engage patients (and caregivers) in realistic goal setting and optimal health service utilization.[14] Further, patients who clearly understand their postdischarge plan are 30% less likely to be readmitted or visit the emergency department.[14] The data for our group have helped us to see areas of strengths, such as hand washing, where we are above compliance rates across hospitals in the United States,[15] as well as those matters that represent opportunities for improvement such as connecting more deeply with our patients.

Tackett et al. have looked at encounter length and its association with performance of etiquette‐based medicine behaviors.[7] Similar to their study, we found a positive correlation between spending more time with patients and higher HMCCOT scores. We also found that HMCCOT scores were higher when providers were caring for new patients. Patients' complaints about doctors often relate to feeling rushed, that their physicians did not listen to them, or that information was not conveyed in a clear manner.[16] Such challenges in physicianpatient communication are ubiquitous across clinical settings.[16] When successfully achieved, patient‐centered communication has been associated with improved clinical outcomes, including adherence to recommended treatment and better self‐management of chronic disease.[17, 18, 19, 20, 21, 22, 23, 24, 25, 26] Many of the components of the HMCCOT described in this article are at the heart of patient‐centered care.

Several limitations of the study should be considered. First, physicians may have behaved differently while they were being observed, which is known as the Hawthorne effect. We observed them for many hours and across multiple patient encounters, and the physicians were not aware of the specific types of data that we were collecting. These factors may have limited the biases along such lines. Second, there may be elements of optimal comportment and communication that were not captured by the HMCCOT. Hopefully, there are not big gaps, as we used multiple methods and an iterative process in the refinement of the HMCCOT metric. Third, one investigator did all of the observing, and it is possible that he might have missed certain behaviors. Through extensive pilot testing and comparisons with other raters, the observer became very skilled and facile with such data collection and the tool. Fourth, we did not survey the same patients that were cared for to compare their perspectives to the HMCCOT scores following the clinical encounters. For patient perspectives, we relied only on PG scores. Fifth, quality of care is a broad and multidimensional construct. The HMCCOT focuses exclusively on hospitalists' comportment and communication at the bedside; therefore, it does not comprehensively assess care quality. Sixth, with our goal to optimally validate the HMCCOT, we tested it on the top tier of hospitalists within each group. We may have observed different results had we randomly selected hospitalists from each hospital or had we conducted the study at hospitals in other geographic regions. Finally, all of the doctors observed worked at hospitals in the Mid‐Atlantic region. However, these five distinct hospitals each have their own cultures, and they are led by different administrators. We purposively chose to sample both academic as well as community settings.

In conclusion, this study reports on the development of a comportment and communication tool that was established and validated by following clinically excellent hospitalists at the bedside. Future studies are necessary to determine whether hospitalists of all levels of experience and clinical skill can improve when given data and feedback using the HMCCOT. Larger studies will then be needed to assess whether enhancing comportment and communication can truly improve patient satisfaction and clinical outcomes in the hospital.

Disclosures: Dr. Wright is a Miller‐Coulson Family Scholar and is supported through the Johns Hopkins Center for Innovative Medicine. Susrutha Kotwal, MD, and Waseem Khaliq, MD, contributed equally to this work. The authors report no conflicts of interest.

Approximately 70 million adults within the United States have sleep disorders,[1] and up to 30% of adults report sleeping less than 6 hours per night.[2] Poor sleep has been associated with undesirable health outcomes.[1] Suboptimal sleep duration and sleep quality has been associated with a higher prevalence of chronic health conditions including hypertension, type 2 diabetes, coronary artery disease, stroke, and obesity, as well as increased overall mortality.[3, 4, 5, 6, 7]

Sleep plays an important role in restoration of wellness. Poor sleep is associated with physiological disturbances that may result in poor healing.[8, 9, 10] In the literature, prevalence of insomnia among elderly hospitalized patients was 36.7%,[11] whereas in younger hospitalized patients it was 50%.[12] Hospitalized patients frequently cite their acute illness, hospital‐related environmental factors, and disruptions that are part of routine care as causes for poor sleep during hospitalization.[13, 14, 15] Although the pervasiveness of poor sleep among hospitalized patients is high, interventions that prioritize sleep optimization as routine care, are uncommon. Few studies have reviewed the effect of sleep‐promoting measures on both sleep quality and sleep duration among patients hospitalized on general medicine units.

In this study, we aimed to assess the feasibility of incorporating sleep‐promoting interventions on a general medicine unit. We sought to identify differences in sleep measures between intervention and control groups. The primary outcome that we hoped to influence and lengthen in the intervention group was sleep duration. This outcome was measured both by sleep diary and with actigraphy. Secondary outcomes that we hypothesized should improve in the intervention group included feeling more refreshed in the mornings, sleep efficiency, and fewer sleep disruptions. As a feasibility pilot, we also wanted to explore the ease or difficulty with which sleep‐promoting interventions could be incorporated to the team's workflow.

METHODS

RESULTS

Of the 112 study patients, 48 were in the intervention unit and 64 in the control unit. Eighty‐five percent of study participants endorsed poor sleep prior to hospital admission on the PSQI sleep quality measure, which was similar in both groups (Table 1).

Participants completed 1 to 8 sleep diary entries (mean = 2.5, standard deviation = 1.1). Because only 6 participants completed 5 or more diaries, we constrained the number of diaries included in the inferential analysis to 4 to avoid influential outliers identified by scatterplots. Fifty‐seven percent of participants had 1 night of valid actigraphy data (n = 64); 29%, 2 nights (n = 32), 8% had 3 or 4 nights, and 9 participants did not have any usable actigraphy data. The extent to which the intervention was accepted by patients in the intervention group was highly variable. Unit‐wide patient adherence with the 10 _pm lights off, telephone off, and TV off policy was 87%, 67%, and 64% of intervention patients, respectively. Uptake of sleep menu items was also highly variable, and not a single element was used by more than half of patients (acceptance rates ranged from 11% to 44%). Eye masks (44%) and ear plugs (32%) were the most commonly utilized items.

A greater proportion of patients in the control arm (33%) had been taking sleep medications prior to hospitalization compared to the intervention arm (15%; ² = 4.6, P < 0.05). However, hypnotic medication use in the hospital was similar across the both groups (intervention unit patients: 25% and controls: 21%, P = 0.49).

Intraclass correlations for the 7 sleep outcomes ranged from 0.59 to 0.76 on sleep diary outcomes, and from 0.61 to 0.85 on actigraphy. Dependency of sleep measures within patients accounted for 59% to 85% of variance in sleep outcomes. The best‐fit mixed models included random intercepts only. The results of mixed models testing the main effect of intervention versus comparison arm on sleep outcome measures, adjusted for covariates, are presented in Table 2. Total sleep time was the only outcome that was significantly different between groups; the average total sleep time, calculated from sleep diary data, was longer in the intervention group by 49 minutes.

Table 3 lists slopes representing average change in sleep measures over hospital days in both groups. The P values represent z tests of interaction terms in mixed models, after adjustment for covariates, testing whether slopes significantly differed between groups. Of the 7 outcomes, 3 sleep diary measures had significant interaction terms. For ratings of sleep quality, refreshing sleep, and sleep disruptions, slopes in the control group were flat, whereas slopes in the intervention group demonstrated improvements in ratings of sleep quality and refreshed sleep, and a decrease in the impact of sleep disruptions over the course of subsequent nights in the hospital. Figure 1 illustrates a plot of the adjusted average slopes for the refreshed sleep score across hospital days in intervention and control groups.

jhm2578-fig-0001-m.png

DISCUSSION

Poor sleep is common among hospitalized adults, both at home prior to the admission and especially when in the hospital. This pilot study demonstrated the feasibility of rolling out a sleep‐promoting intervention on a hospital's general medicine unit. Although participants on the intervention unit reported improved sleep quality and feeling more refreshed, this was not supported by actigraphy data (such as sleep time or sleep efficiency). Although care team engagement and implementation of unit‐wide interventions were high, patient use of individual components was imperfect. Of particular interest, however, the intervention group actually began to have improved sleep quality and fewer disruptions with subsequent nights sleeping in the hospital.

Our findings of the high prevalence of poor sleep among hospitalized patients is congruent with prior studies and supports the great need to screen for and address poor sleep within the hospital setting.[24, 25, 26] Attempts to promote sleep among hospitalized patients may be effective. Prior literature on sleep‐promoting intervention studies demonstrated relaxation techniques improved sleep quality by almost 38%,[27] and ear plugs and eye masks showed some benefit in promoting sleep within the hospital.[28] Our study's multicomponent intervention that attempted to minimize disruptions led to improvement in sleep quality, more restorative sleep, and decreased report of sleep disruptions, especially among patients who had a longer length of stay. As suggested by Thomas et al.[29] and seen in our data, this temporal relationship with improvement across subsequent nights suggests there may be an adaptation to the new environment and that it may take time for the sleep intervention to work.

Hospitalized patients often fail to reclaim the much‐needed restorative sleep at the time when they are most vulnerable. Patients cite routine care as the primary cause of sleep disruption, and often recognize the way that the hospital environment interferes with their ability to sleep.[30, 31, 32] The sleep‐promoting interventions used in our study would be characterized by most as low effort[33] and a potential for high yield, even though our patients only appreciated modest improvements in sleep outcomes.

Several limitations of this study should be considered. First, although we had hoped to collect substantial amounts of objective data, the average time of actigraphy observation was less than 48 hours. This may have constrained the group by time interaction analysis with actigraphy data, as studies have shown increased accuracy in actigraphy measures with longer wear.[34] By contrast, the sleep diary survey collected throughout hospitalization yielded significant improvements in consecutive daily measurements. Second, the proximity of the study units raised concern for study contamination, which could have reduced the differences in the outcome measures that may have been observed. Although the physicians work on both units, the nursing and support care teams are distinct and unit dependent. Finally, this was not a randomized trial. Patient assignment to the treatment arms was haphazard and occurred within the hospital's admitting strategy. Allocation of patients to either the intervention or the control group was based on bed availability at the time of admission. Although both groups were similar in most characteristics, more of the control participants reported taking more sleep medications prior to admission as compared to the intervention participants. Fortunately, hypnotic use was not different between groups during the admission, the time when sleep data were being captured.

Overall, this pilot study suggests that patients admitted to general medical ward fail to realize sufficient restorative sleep when they are in the hospital. Sleep disruption is rather frequent. This study demonstrates the opportunity for and feasibility of sleep‐promoting interventions where facilitating sleep is considered to be a top priority and vital component of the healthcare delivery. When trying to improve patients' sleep in the hospital, it may take several consecutive nights to realize a return on investment.

Approximately 70 million adults within the United States have sleep disorders,[1] and up to 30% of adults report sleeping less than 6 hours per night.[2] Poor sleep has been associated with undesirable health outcomes.[1] Suboptimal sleep duration and sleep quality has been associated with a higher prevalence of chronic health conditions including hypertension, type 2 diabetes, coronary artery disease, stroke, and obesity, as well as increased overall mortality.[3, 4, 5, 6, 7]

Sleep plays an important role in restoration of wellness. Poor sleep is associated with physiological disturbances that may result in poor healing.[8, 9, 10] In the literature, prevalence of insomnia among elderly hospitalized patients was 36.7%,[11] whereas in younger hospitalized patients it was 50%.[12] Hospitalized patients frequently cite their acute illness, hospital‐related environmental factors, and disruptions that are part of routine care as causes for poor sleep during hospitalization.[13, 14, 15] Although the pervasiveness of poor sleep among hospitalized patients is high, interventions that prioritize sleep optimization as routine care, are uncommon. Few studies have reviewed the effect of sleep‐promoting measures on both sleep quality and sleep duration among patients hospitalized on general medicine units.

In this study, we aimed to assess the feasibility of incorporating sleep‐promoting interventions on a general medicine unit. We sought to identify differences in sleep measures between intervention and control groups. The primary outcome that we hoped to influence and lengthen in the intervention group was sleep duration. This outcome was measured both by sleep diary and with actigraphy. Secondary outcomes that we hypothesized should improve in the intervention group included feeling more refreshed in the mornings, sleep efficiency, and fewer sleep disruptions. As a feasibility pilot, we also wanted to explore the ease or difficulty with which sleep‐promoting interventions could be incorporated to the team's workflow.

METHODS

RESULTS

Of the 112 study patients, 48 were in the intervention unit and 64 in the control unit. Eighty‐five percent of study participants endorsed poor sleep prior to hospital admission on the PSQI sleep quality measure, which was similar in both groups (Table 1).

Participants completed 1 to 8 sleep diary entries (mean = 2.5, standard deviation = 1.1). Because only 6 participants completed 5 or more diaries, we constrained the number of diaries included in the inferential analysis to 4 to avoid influential outliers identified by scatterplots. Fifty‐seven percent of participants had 1 night of valid actigraphy data (n = 64); 29%, 2 nights (n = 32), 8% had 3 or 4 nights, and 9 participants did not have any usable actigraphy data. The extent to which the intervention was accepted by patients in the intervention group was highly variable. Unit‐wide patient adherence with the 10 _pm lights off, telephone off, and TV off policy was 87%, 67%, and 64% of intervention patients, respectively. Uptake of sleep menu items was also highly variable, and not a single element was used by more than half of patients (acceptance rates ranged from 11% to 44%). Eye masks (44%) and ear plugs (32%) were the most commonly utilized items.

A greater proportion of patients in the control arm (33%) had been taking sleep medications prior to hospitalization compared to the intervention arm (15%; ² = 4.6, P < 0.05). However, hypnotic medication use in the hospital was similar across the both groups (intervention unit patients: 25% and controls: 21%, P = 0.49).

Intraclass correlations for the 7 sleep outcomes ranged from 0.59 to 0.76 on sleep diary outcomes, and from 0.61 to 0.85 on actigraphy. Dependency of sleep measures within patients accounted for 59% to 85% of variance in sleep outcomes. The best‐fit mixed models included random intercepts only. The results of mixed models testing the main effect of intervention versus comparison arm on sleep outcome measures, adjusted for covariates, are presented in Table 2. Total sleep time was the only outcome that was significantly different between groups; the average total sleep time, calculated from sleep diary data, was longer in the intervention group by 49 minutes.

Table 3 lists slopes representing average change in sleep measures over hospital days in both groups. The P values represent z tests of interaction terms in mixed models, after adjustment for covariates, testing whether slopes significantly differed between groups. Of the 7 outcomes, 3 sleep diary measures had significant interaction terms. For ratings of sleep quality, refreshing sleep, and sleep disruptions, slopes in the control group were flat, whereas slopes in the intervention group demonstrated improvements in ratings of sleep quality and refreshed sleep, and a decrease in the impact of sleep disruptions over the course of subsequent nights in the hospital. Figure 1 illustrates a plot of the adjusted average slopes for the refreshed sleep score across hospital days in intervention and control groups.

jhm2578-fig-0001-m.png

DISCUSSION

Poor sleep is common among hospitalized adults, both at home prior to the admission and especially when in the hospital. This pilot study demonstrated the feasibility of rolling out a sleep‐promoting intervention on a hospital's general medicine unit. Although participants on the intervention unit reported improved sleep quality and feeling more refreshed, this was not supported by actigraphy data (such as sleep time or sleep efficiency). Although care team engagement and implementation of unit‐wide interventions were high, patient use of individual components was imperfect. Of particular interest, however, the intervention group actually began to have improved sleep quality and fewer disruptions with subsequent nights sleeping in the hospital.

Our findings of the high prevalence of poor sleep among hospitalized patients is congruent with prior studies and supports the great need to screen for and address poor sleep within the hospital setting.[24, 25, 26] Attempts to promote sleep among hospitalized patients may be effective. Prior literature on sleep‐promoting intervention studies demonstrated relaxation techniques improved sleep quality by almost 38%,[27] and ear plugs and eye masks showed some benefit in promoting sleep within the hospital.[28] Our study's multicomponent intervention that attempted to minimize disruptions led to improvement in sleep quality, more restorative sleep, and decreased report of sleep disruptions, especially among patients who had a longer length of stay. As suggested by Thomas et al.[29] and seen in our data, this temporal relationship with improvement across subsequent nights suggests there may be an adaptation to the new environment and that it may take time for the sleep intervention to work.

Hospitalized patients often fail to reclaim the much‐needed restorative sleep at the time when they are most vulnerable. Patients cite routine care as the primary cause of sleep disruption, and often recognize the way that the hospital environment interferes with their ability to sleep.[30, 31, 32] The sleep‐promoting interventions used in our study would be characterized by most as low effort[33] and a potential for high yield, even though our patients only appreciated modest improvements in sleep outcomes.

Several limitations of this study should be considered. First, although we had hoped to collect substantial amounts of objective data, the average time of actigraphy observation was less than 48 hours. This may have constrained the group by time interaction analysis with actigraphy data, as studies have shown increased accuracy in actigraphy measures with longer wear.[34] By contrast, the sleep diary survey collected throughout hospitalization yielded significant improvements in consecutive daily measurements. Second, the proximity of the study units raised concern for study contamination, which could have reduced the differences in the outcome measures that may have been observed. Although the physicians work on both units, the nursing and support care teams are distinct and unit dependent. Finally, this was not a randomized trial. Patient assignment to the treatment arms was haphazard and occurred within the hospital's admitting strategy. Allocation of patients to either the intervention or the control group was based on bed availability at the time of admission. Although both groups were similar in most characteristics, more of the control participants reported taking more sleep medications prior to admission as compared to the intervention participants. Fortunately, hypnotic use was not different between groups during the admission, the time when sleep data were being captured.

Overall, this pilot study suggests that patients admitted to general medical ward fail to realize sufficient restorative sleep when they are in the hospital. Sleep disruption is rather frequent. This study demonstrates the opportunity for and feasibility of sleep‐promoting interventions where facilitating sleep is considered to be a top priority and vital component of the healthcare delivery. When trying to improve patients' sleep in the hospital, it may take several consecutive nights to realize a return on investment.

Approximately 70 million adults within the United States have sleep disorders,[1] and up to 30% of adults report sleeping less than 6 hours per night.[2] Poor sleep has been associated with undesirable health outcomes.[1] Suboptimal sleep duration and sleep quality has been associated with a higher prevalence of chronic health conditions including hypertension, type 2 diabetes, coronary artery disease, stroke, and obesity, as well as increased overall mortality.[3, 4, 5, 6, 7]

Sleep plays an important role in restoration of wellness. Poor sleep is associated with physiological disturbances that may result in poor healing.[8, 9, 10] In the literature, prevalence of insomnia among elderly hospitalized patients was 36.7%,[11] whereas in younger hospitalized patients it was 50%.[12] Hospitalized patients frequently cite their acute illness, hospital‐related environmental factors, and disruptions that are part of routine care as causes for poor sleep during hospitalization.[13, 14, 15] Although the pervasiveness of poor sleep among hospitalized patients is high, interventions that prioritize sleep optimization as routine care, are uncommon. Few studies have reviewed the effect of sleep‐promoting measures on both sleep quality and sleep duration among patients hospitalized on general medicine units.

In this study, we aimed to assess the feasibility of incorporating sleep‐promoting interventions on a general medicine unit. We sought to identify differences in sleep measures between intervention and control groups. The primary outcome that we hoped to influence and lengthen in the intervention group was sleep duration. This outcome was measured both by sleep diary and with actigraphy. Secondary outcomes that we hypothesized should improve in the intervention group included feeling more refreshed in the mornings, sleep efficiency, and fewer sleep disruptions. As a feasibility pilot, we also wanted to explore the ease or difficulty with which sleep‐promoting interventions could be incorporated to the team's workflow.

METHODS

RESULTS

Of the 112 study patients, 48 were in the intervention unit and 64 in the control unit. Eighty‐five percent of study participants endorsed poor sleep prior to hospital admission on the PSQI sleep quality measure, which was similar in both groups (Table 1).

Participants completed 1 to 8 sleep diary entries (mean = 2.5, standard deviation = 1.1). Because only 6 participants completed 5 or more diaries, we constrained the number of diaries included in the inferential analysis to 4 to avoid influential outliers identified by scatterplots. Fifty‐seven percent of participants had 1 night of valid actigraphy data (n = 64); 29%, 2 nights (n = 32), 8% had 3 or 4 nights, and 9 participants did not have any usable actigraphy data. The extent to which the intervention was accepted by patients in the intervention group was highly variable. Unit‐wide patient adherence with the 10 _pm lights off, telephone off, and TV off policy was 87%, 67%, and 64% of intervention patients, respectively. Uptake of sleep menu items was also highly variable, and not a single element was used by more than half of patients (acceptance rates ranged from 11% to 44%). Eye masks (44%) and ear plugs (32%) were the most commonly utilized items.

A greater proportion of patients in the control arm (33%) had been taking sleep medications prior to hospitalization compared to the intervention arm (15%; ² = 4.6, P < 0.05). However, hypnotic medication use in the hospital was similar across the both groups (intervention unit patients: 25% and controls: 21%, P = 0.49).

Intraclass correlations for the 7 sleep outcomes ranged from 0.59 to 0.76 on sleep diary outcomes, and from 0.61 to 0.85 on actigraphy. Dependency of sleep measures within patients accounted for 59% to 85% of variance in sleep outcomes. The best‐fit mixed models included random intercepts only. The results of mixed models testing the main effect of intervention versus comparison arm on sleep outcome measures, adjusted for covariates, are presented in Table 2. Total sleep time was the only outcome that was significantly different between groups; the average total sleep time, calculated from sleep diary data, was longer in the intervention group by 49 minutes.

Table 3 lists slopes representing average change in sleep measures over hospital days in both groups. The P values represent z tests of interaction terms in mixed models, after adjustment for covariates, testing whether slopes significantly differed between groups. Of the 7 outcomes, 3 sleep diary measures had significant interaction terms. For ratings of sleep quality, refreshing sleep, and sleep disruptions, slopes in the control group were flat, whereas slopes in the intervention group demonstrated improvements in ratings of sleep quality and refreshed sleep, and a decrease in the impact of sleep disruptions over the course of subsequent nights in the hospital. Figure 1 illustrates a plot of the adjusted average slopes for the refreshed sleep score across hospital days in intervention and control groups.

jhm2578-fig-0001-m.png

DISCUSSION

Poor sleep is common among hospitalized adults, both at home prior to the admission and especially when in the hospital. This pilot study demonstrated the feasibility of rolling out a sleep‐promoting intervention on a hospital's general medicine unit. Although participants on the intervention unit reported improved sleep quality and feeling more refreshed, this was not supported by actigraphy data (such as sleep time or sleep efficiency). Although care team engagement and implementation of unit‐wide interventions were high, patient use of individual components was imperfect. Of particular interest, however, the intervention group actually began to have improved sleep quality and fewer disruptions with subsequent nights sleeping in the hospital.

Our findings of the high prevalence of poor sleep among hospitalized patients is congruent with prior studies and supports the great need to screen for and address poor sleep within the hospital setting.[24, 25, 26] Attempts to promote sleep among hospitalized patients may be effective. Prior literature on sleep‐promoting intervention studies demonstrated relaxation techniques improved sleep quality by almost 38%,[27] and ear plugs and eye masks showed some benefit in promoting sleep within the hospital.[28] Our study's multicomponent intervention that attempted to minimize disruptions led to improvement in sleep quality, more restorative sleep, and decreased report of sleep disruptions, especially among patients who had a longer length of stay. As suggested by Thomas et al.[29] and seen in our data, this temporal relationship with improvement across subsequent nights suggests there may be an adaptation to the new environment and that it may take time for the sleep intervention to work.

Hospitalized patients often fail to reclaim the much‐needed restorative sleep at the time when they are most vulnerable. Patients cite routine care as the primary cause of sleep disruption, and often recognize the way that the hospital environment interferes with their ability to sleep.[30, 31, 32] The sleep‐promoting interventions used in our study would be characterized by most as low effort[33] and a potential for high yield, even though our patients only appreciated modest improvements in sleep outcomes.

Several limitations of this study should be considered. First, although we had hoped to collect substantial amounts of objective data, the average time of actigraphy observation was less than 48 hours. This may have constrained the group by time interaction analysis with actigraphy data, as studies have shown increased accuracy in actigraphy measures with longer wear.[34] By contrast, the sleep diary survey collected throughout hospitalization yielded significant improvements in consecutive daily measurements. Second, the proximity of the study units raised concern for study contamination, which could have reduced the differences in the outcome measures that may have been observed. Although the physicians work on both units, the nursing and support care teams are distinct and unit dependent. Finally, this was not a randomized trial. Patient assignment to the treatment arms was haphazard and occurred within the hospital's admitting strategy. Allocation of patients to either the intervention or the control group was based on bed availability at the time of admission. Although both groups were similar in most characteristics, more of the control participants reported taking more sleep medications prior to admission as compared to the intervention participants. Fortunately, hypnotic use was not different between groups during the admission, the time when sleep data were being captured.

Overall, this pilot study suggests that patients admitted to general medical ward fail to realize sufficient restorative sleep when they are in the hospital. Sleep disruption is rather frequent. This study demonstrates the opportunity for and feasibility of sleep‐promoting interventions where facilitating sleep is considered to be a top priority and vital component of the healthcare delivery. When trying to improve patients' sleep in the hospital, it may take several consecutive nights to realize a return on investment.

Patient satisfaction scores are being reported publicly and will affect hospital reimbursement rates under Hospital Value Based Purchasing.[1] Patient satisfaction scores are currently obtained through metrics such as Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS)[2] and Press Ganey (PG)[3] surveys. Such surveys are mailed to a variable proportion of patients following their discharge from the hospital, and ask patients about the quality of care they received during their admission. Domains assessed regarding the patients' inpatient experiences range from room cleanliness to the amount of time the physician spent with them.

The Society of Hospital Medicine (SHM), the largest professional medical society representing hospitalists, encourages the use of patient satisfaction surveys to measure hospitalist providers' quality of patient care.[4] Ideally, accurate information would be delivered as feedback to individual providers in a timely manner in hopes of improving performance; however, the current methodology has shortcomings that limit its usefulness. First, several hospitalists and consultants may be involved in the care of 1 patient during the hospital stay, but the score can only be tied to a single physician. Current survey methods attribute all responses to that particular doctor, usually the attending of record, although patients may very well be thinking of other physicians when responding to questions. Second, only a few questions on the surveys ask about doctors' performance. Aforementioned surveys have 3 to 8 questions about doctors' care, which limits the ability to assess physician performance comprehensively. Finally, the surveys are mailed approximately 1 week after the patient's discharge, usually without a name or photograph of the physician to facilitate patient/caregiver recall. This time lag and lack of information to prompt patient recall likely lead to impreciseness in assessment. In addition, the response rates to these surveys are typically low, around 25% (personal oral communication with our division's service excellence stakeholder Dr. L.P. in September 2013). These deficiencies limit the usefulness of such data in coaching individual providers about their performance because they cannot be delivered in a timely fashion, and the reliability of the attribution is suspect.

With these considerations in mind, we developed and validated a new survey metric, the Tool to Assess Inpatient Satisfaction with Care from Hospitalists (TAISCH). We hypothesized that the results would be different from those collected using conventional methodologies.

PATIENTS AND METHODS

RESULTS

A total of 330 patients were considered to be eligible through medical record screening. Of those patients, 73 (22%) were already discharged by the time the research assistant attempted to enroll them after 2 days of care by a single physician. Of 257 inpatients approached, 30 patients (12%) refused to participate. Among the 227 consented patients, 24 (9%) were excluded as they were unable to correctly identify their hospitalist provider. A total of 203 patients were enrolled, and each patient rated a single hospitalist; a total of 29 unique hospitalists were assessed by these patients. The patients' mean age was 60 years, 114 (56%) were female, and 61 (30%) were of nonwhite race (Table 1). The hospitalist physicians' demographic information is also shown in Table 1. Two hospitalists with fewer than 4 surveys collected were excluded from the analysis. Thus, final analysis included 200 unique patients assessing 1 of the 27 hospitalists (mean=7.4 surveys per hospitalist).

DISCUSSION

Our new metric, TAISCH, was found to be a reliable and valid measurement tool to assess patient satisfaction with the hospitalist physician's care. Because we only surveyed patients who could correctly identify their hospitalist physicians after interacting for at least 2 consecutive days, the attribution of the data to the individual hospitalist is almost certainly correct. The high participation rate indicates that the patients were not hesitant about rating their hospitalist provider's quality of care, even when asked while they were still in the hospital.

The majority of the patients approached were able to correctly identify their hospitalist provider. This rate (91%) was much higher than the rate previously reported in the literature where a picture card was used to improve provider recognition.[22] It is also likely that 1 physician, rather than a team of physicians, taking care of patients make it easier for patients to recall the name and recognize the face of their inpatient provider.

The CFA of TAISCH showed good fit but suggests that 2 variables, both from Kahn's etiquette‐based medicine (EtBM) checklist,[9] may not load in the same way as the other items. Tackett and colleagues reported that hospitalists who performed more EtBM behaviors scored higher on PG evaluations.[23] Such results, along with the comparable explanation of variance and reliability, convinced us to retain these 2 items in the final 15‐item TAISCH as dictated by the CFA. Although the literature supports the fact that physician etiquette is related to perception of high‐quality care, it is possible that these 2 questions were answered differently (and thereby failed to load the same way), because environmental limitations may be preventing physicians' ability to perform them consistently. We prefer the 15‐item version of TAISCH and future studies may provide additional information about its performance as compared to the 13‐item adaptation.

The significantly negative association between the Wong‐Baker pain scale and TAISCH stresses the importance of adequately addressing and treating the patient's pain. Hanna et al. showed that the patients' perceptions of pain control was associated with their overall satisfaction score measured by HCAHPS.[24] The association seen in our study was not unexpected, because TAISCH is administered while the patients are acutely ill in the hospital, when pain is likely more prevalent and severe than it is during the postdischarge settings (when the HCAHPS or PG surveys are administered). Interestingly, Hanna et al. discovered that the team's attention to controlling pain was more strongly correlated with overall satisfaction than was the actual pain control.[24] These data, now confirmed by our study, should serve to remind us that a hospitalist's concern and effort to relieve pain may augment patient satisfaction with the quality of care, even when eliminating the pain may be difficult or impossible.

TAISCH was found not to be correlated with PG scores. Several explanations for this deserve consideration. First, the postdischarge PG survey that is used for our institution does not list the name of the specific hospitalist providers for the patients to evaluate. Because patients encounter multiple physicians during their hospital stay (eg, emergency department physicians, hospitalist providers, consultants), it is possible that patients are not reflecting on the named doctor when assessing the the attending of record on the PG mailed questionnaire. Second, the representation of patients who responded to TAISCH and PG were different; almost all patients completed TAISCH as opposed to a small minority who decide to respond to the PG survey. Third, TAISCH measures the physicians' performance more comprehensively with a larger number of variables. Last, it is possible that we were underpowered to detect significant correlation, because there were only 24 providers who had data from both TAISCH and PG. However, our results endorse using caution in interpreting PG scores for individual hospitalist's performance, particularly for high‐stakes consequences (including the provision of incentives to high performer and the insistence on remediation for low performers).

Several limitations of this study should be considered. First, only hospitalist providers from a single division were assessed. This may limit the generalizability of our findings. Second, although patients were assured about confidentiality of their responses, they might have provided more favorable answers, because they may have felt uncomfortable rating their physician poorly. One review article of the measurement of healthcare satisfaction indicated that impersonal (mailed) methods result in more criticism and lower satisfaction than assessments made in person using interviews. As the trade‐off, the mailed surveys yield lower response rates that may introduce other forms of bias.[25] Even on the HCHAPS survey report for the same period from our institution, 78% of patients gave top box ratings for our doctors' communication skills, which is at the state average.[26] Similarly, a study that used postdischarge telephone interviews to collect patients' satisfaction with hospitalists' care quality reported an average score of 4.20 out of 5.[27] These findings confirm that highly skewed ratings are common for these types of surveys, irrespective of how or when the data are collected.

Despite the aforementioned limitations, TAISCH use need not be limited to hospitalist physicians. It may also be used to assess allied health professionals or trainees performance, which cannot be assessed by HCHAPS or PG. Applying TAISCH in different hospital settings (eg, emergency department or critical care units), assessing hospitalists' reactions to TAISCH, learning whether TAISCH leads to hospitalists' behavior changes or appraising whether performance can improve in response to coaching interventions for those performing poorly are all research questions that merit additional consideration.

CONCLUSION

TAISCH allows for obtaining patient satisfaction data that are highly attributable to specific hospitalist providers. The data collection method also permits high response rates so that input comes from almost all patients. The timeliness of the TAISCH assessments also makes it possible for real‐time service recovery, which is impossible with other commonly used metrics assessing patient satisfaction. Our next step will include testing the most effective way to provide feedback to providers and to coach these individuals so as to improve performance.

Patient satisfaction scores are being reported publicly and will affect hospital reimbursement rates under Hospital Value Based Purchasing.[1] Patient satisfaction scores are currently obtained through metrics such as Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS)[2] and Press Ganey (PG)[3] surveys. Such surveys are mailed to a variable proportion of patients following their discharge from the hospital, and ask patients about the quality of care they received during their admission. Domains assessed regarding the patients' inpatient experiences range from room cleanliness to the amount of time the physician spent with them.

The Society of Hospital Medicine (SHM), the largest professional medical society representing hospitalists, encourages the use of patient satisfaction surveys to measure hospitalist providers' quality of patient care.[4] Ideally, accurate information would be delivered as feedback to individual providers in a timely manner in hopes of improving performance; however, the current methodology has shortcomings that limit its usefulness. First, several hospitalists and consultants may be involved in the care of 1 patient during the hospital stay, but the score can only be tied to a single physician. Current survey methods attribute all responses to that particular doctor, usually the attending of record, although patients may very well be thinking of other physicians when responding to questions. Second, only a few questions on the surveys ask about doctors' performance. Aforementioned surveys have 3 to 8 questions about doctors' care, which limits the ability to assess physician performance comprehensively. Finally, the surveys are mailed approximately 1 week after the patient's discharge, usually without a name or photograph of the physician to facilitate patient/caregiver recall. This time lag and lack of information to prompt patient recall likely lead to impreciseness in assessment. In addition, the response rates to these surveys are typically low, around 25% (personal oral communication with our division's service excellence stakeholder Dr. L.P. in September 2013). These deficiencies limit the usefulness of such data in coaching individual providers about their performance because they cannot be delivered in a timely fashion, and the reliability of the attribution is suspect.

With these considerations in mind, we developed and validated a new survey metric, the Tool to Assess Inpatient Satisfaction with Care from Hospitalists (TAISCH). We hypothesized that the results would be different from those collected using conventional methodologies.

PATIENTS AND METHODS

RESULTS

A total of 330 patients were considered to be eligible through medical record screening. Of those patients, 73 (22%) were already discharged by the time the research assistant attempted to enroll them after 2 days of care by a single physician. Of 257 inpatients approached, 30 patients (12%) refused to participate. Among the 227 consented patients, 24 (9%) were excluded as they were unable to correctly identify their hospitalist provider. A total of 203 patients were enrolled, and each patient rated a single hospitalist; a total of 29 unique hospitalists were assessed by these patients. The patients' mean age was 60 years, 114 (56%) were female, and 61 (30%) were of nonwhite race (Table 1). The hospitalist physicians' demographic information is also shown in Table 1. Two hospitalists with fewer than 4 surveys collected were excluded from the analysis. Thus, final analysis included 200 unique patients assessing 1 of the 27 hospitalists (mean=7.4 surveys per hospitalist).

DISCUSSION

Our new metric, TAISCH, was found to be a reliable and valid measurement tool to assess patient satisfaction with the hospitalist physician's care. Because we only surveyed patients who could correctly identify their hospitalist physicians after interacting for at least 2 consecutive days, the attribution of the data to the individual hospitalist is almost certainly correct. The high participation rate indicates that the patients were not hesitant about rating their hospitalist provider's quality of care, even when asked while they were still in the hospital.

The majority of the patients approached were able to correctly identify their hospitalist provider. This rate (91%) was much higher than the rate previously reported in the literature where a picture card was used to improve provider recognition.[22] It is also likely that 1 physician, rather than a team of physicians, taking care of patients make it easier for patients to recall the name and recognize the face of their inpatient provider.

The CFA of TAISCH showed good fit but suggests that 2 variables, both from Kahn's etiquette‐based medicine (EtBM) checklist,[9] may not load in the same way as the other items. Tackett and colleagues reported that hospitalists who performed more EtBM behaviors scored higher on PG evaluations.[23] Such results, along with the comparable explanation of variance and reliability, convinced us to retain these 2 items in the final 15‐item TAISCH as dictated by the CFA. Although the literature supports the fact that physician etiquette is related to perception of high‐quality care, it is possible that these 2 questions were answered differently (and thereby failed to load the same way), because environmental limitations may be preventing physicians' ability to perform them consistently. We prefer the 15‐item version of TAISCH and future studies may provide additional information about its performance as compared to the 13‐item adaptation.

The significantly negative association between the Wong‐Baker pain scale and TAISCH stresses the importance of adequately addressing and treating the patient's pain. Hanna et al. showed that the patients' perceptions of pain control was associated with their overall satisfaction score measured by HCAHPS.[24] The association seen in our study was not unexpected, because TAISCH is administered while the patients are acutely ill in the hospital, when pain is likely more prevalent and severe than it is during the postdischarge settings (when the HCAHPS or PG surveys are administered). Interestingly, Hanna et al. discovered that the team's attention to controlling pain was more strongly correlated with overall satisfaction than was the actual pain control.[24] These data, now confirmed by our study, should serve to remind us that a hospitalist's concern and effort to relieve pain may augment patient satisfaction with the quality of care, even when eliminating the pain may be difficult or impossible.

TAISCH was found not to be correlated with PG scores. Several explanations for this deserve consideration. First, the postdischarge PG survey that is used for our institution does not list the name of the specific hospitalist providers for the patients to evaluate. Because patients encounter multiple physicians during their hospital stay (eg, emergency department physicians, hospitalist providers, consultants), it is possible that patients are not reflecting on the named doctor when assessing the the attending of record on the PG mailed questionnaire. Second, the representation of patients who responded to TAISCH and PG were different; almost all patients completed TAISCH as opposed to a small minority who decide to respond to the PG survey. Third, TAISCH measures the physicians' performance more comprehensively with a larger number of variables. Last, it is possible that we were underpowered to detect significant correlation, because there were only 24 providers who had data from both TAISCH and PG. However, our results endorse using caution in interpreting PG scores for individual hospitalist's performance, particularly for high‐stakes consequences (including the provision of incentives to high performer and the insistence on remediation for low performers).

Several limitations of this study should be considered. First, only hospitalist providers from a single division were assessed. This may limit the generalizability of our findings. Second, although patients were assured about confidentiality of their responses, they might have provided more favorable answers, because they may have felt uncomfortable rating their physician poorly. One review article of the measurement of healthcare satisfaction indicated that impersonal (mailed) methods result in more criticism and lower satisfaction than assessments made in person using interviews. As the trade‐off, the mailed surveys yield lower response rates that may introduce other forms of bias.[25] Even on the HCHAPS survey report for the same period from our institution, 78% of patients gave top box ratings for our doctors' communication skills, which is at the state average.[26] Similarly, a study that used postdischarge telephone interviews to collect patients' satisfaction with hospitalists' care quality reported an average score of 4.20 out of 5.[27] These findings confirm that highly skewed ratings are common for these types of surveys, irrespective of how or when the data are collected.

Despite the aforementioned limitations, TAISCH use need not be limited to hospitalist physicians. It may also be used to assess allied health professionals or trainees performance, which cannot be assessed by HCHAPS or PG. Applying TAISCH in different hospital settings (eg, emergency department or critical care units), assessing hospitalists' reactions to TAISCH, learning whether TAISCH leads to hospitalists' behavior changes or appraising whether performance can improve in response to coaching interventions for those performing poorly are all research questions that merit additional consideration.

CONCLUSION

TAISCH allows for obtaining patient satisfaction data that are highly attributable to specific hospitalist providers. The data collection method also permits high response rates so that input comes from almost all patients. The timeliness of the TAISCH assessments also makes it possible for real‐time service recovery, which is impossible with other commonly used metrics assessing patient satisfaction. Our next step will include testing the most effective way to provide feedback to providers and to coach these individuals so as to improve performance.

Patient satisfaction scores are being reported publicly and will affect hospital reimbursement rates under Hospital Value Based Purchasing.[1] Patient satisfaction scores are currently obtained through metrics such as Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS)[2] and Press Ganey (PG)[3] surveys. Such surveys are mailed to a variable proportion of patients following their discharge from the hospital, and ask patients about the quality of care they received during their admission. Domains assessed regarding the patients' inpatient experiences range from room cleanliness to the amount of time the physician spent with them.

The Society of Hospital Medicine (SHM), the largest professional medical society representing hospitalists, encourages the use of patient satisfaction surveys to measure hospitalist providers' quality of patient care.[4] Ideally, accurate information would be delivered as feedback to individual providers in a timely manner in hopes of improving performance; however, the current methodology has shortcomings that limit its usefulness. First, several hospitalists and consultants may be involved in the care of 1 patient during the hospital stay, but the score can only be tied to a single physician. Current survey methods attribute all responses to that particular doctor, usually the attending of record, although patients may very well be thinking of other physicians when responding to questions. Second, only a few questions on the surveys ask about doctors' performance. Aforementioned surveys have 3 to 8 questions about doctors' care, which limits the ability to assess physician performance comprehensively. Finally, the surveys are mailed approximately 1 week after the patient's discharge, usually without a name or photograph of the physician to facilitate patient/caregiver recall. This time lag and lack of information to prompt patient recall likely lead to impreciseness in assessment. In addition, the response rates to these surveys are typically low, around 25% (personal oral communication with our division's service excellence stakeholder Dr. L.P. in September 2013). These deficiencies limit the usefulness of such data in coaching individual providers about their performance because they cannot be delivered in a timely fashion, and the reliability of the attribution is suspect.

With these considerations in mind, we developed and validated a new survey metric, the Tool to Assess Inpatient Satisfaction with Care from Hospitalists (TAISCH). We hypothesized that the results would be different from those collected using conventional methodologies.

PATIENTS AND METHODS

RESULTS

A total of 330 patients were considered to be eligible through medical record screening. Of those patients, 73 (22%) were already discharged by the time the research assistant attempted to enroll them after 2 days of care by a single physician. Of 257 inpatients approached, 30 patients (12%) refused to participate. Among the 227 consented patients, 24 (9%) were excluded as they were unable to correctly identify their hospitalist provider. A total of 203 patients were enrolled, and each patient rated a single hospitalist; a total of 29 unique hospitalists were assessed by these patients. The patients' mean age was 60 years, 114 (56%) were female, and 61 (30%) were of nonwhite race (Table 1). The hospitalist physicians' demographic information is also shown in Table 1. Two hospitalists with fewer than 4 surveys collected were excluded from the analysis. Thus, final analysis included 200 unique patients assessing 1 of the 27 hospitalists (mean=7.4 surveys per hospitalist).

DISCUSSION

Our new metric, TAISCH, was found to be a reliable and valid measurement tool to assess patient satisfaction with the hospitalist physician's care. Because we only surveyed patients who could correctly identify their hospitalist physicians after interacting for at least 2 consecutive days, the attribution of the data to the individual hospitalist is almost certainly correct. The high participation rate indicates that the patients were not hesitant about rating their hospitalist provider's quality of care, even when asked while they were still in the hospital.

The majority of the patients approached were able to correctly identify their hospitalist provider. This rate (91%) was much higher than the rate previously reported in the literature where a picture card was used to improve provider recognition.[22] It is also likely that 1 physician, rather than a team of physicians, taking care of patients make it easier for patients to recall the name and recognize the face of their inpatient provider.

The CFA of TAISCH showed good fit but suggests that 2 variables, both from Kahn's etiquette‐based medicine (EtBM) checklist,[9] may not load in the same way as the other items. Tackett and colleagues reported that hospitalists who performed more EtBM behaviors scored higher on PG evaluations.[23] Such results, along with the comparable explanation of variance and reliability, convinced us to retain these 2 items in the final 15‐item TAISCH as dictated by the CFA. Although the literature supports the fact that physician etiquette is related to perception of high‐quality care, it is possible that these 2 questions were answered differently (and thereby failed to load the same way), because environmental limitations may be preventing physicians' ability to perform them consistently. We prefer the 15‐item version of TAISCH and future studies may provide additional information about its performance as compared to the 13‐item adaptation.

The significantly negative association between the Wong‐Baker pain scale and TAISCH stresses the importance of adequately addressing and treating the patient's pain. Hanna et al. showed that the patients' perceptions of pain control was associated with their overall satisfaction score measured by HCAHPS.[24] The association seen in our study was not unexpected, because TAISCH is administered while the patients are acutely ill in the hospital, when pain is likely more prevalent and severe than it is during the postdischarge settings (when the HCAHPS or PG surveys are administered). Interestingly, Hanna et al. discovered that the team's attention to controlling pain was more strongly correlated with overall satisfaction than was the actual pain control.[24] These data, now confirmed by our study, should serve to remind us that a hospitalist's concern and effort to relieve pain may augment patient satisfaction with the quality of care, even when eliminating the pain may be difficult or impossible.

TAISCH was found not to be correlated with PG scores. Several explanations for this deserve consideration. First, the postdischarge PG survey that is used for our institution does not list the name of the specific hospitalist providers for the patients to evaluate. Because patients encounter multiple physicians during their hospital stay (eg, emergency department physicians, hospitalist providers, consultants), it is possible that patients are not reflecting on the named doctor when assessing the the attending of record on the PG mailed questionnaire. Second, the representation of patients who responded to TAISCH and PG were different; almost all patients completed TAISCH as opposed to a small minority who decide to respond to the PG survey. Third, TAISCH measures the physicians' performance more comprehensively with a larger number of variables. Last, it is possible that we were underpowered to detect significant correlation, because there were only 24 providers who had data from both TAISCH and PG. However, our results endorse using caution in interpreting PG scores for individual hospitalist's performance, particularly for high‐stakes consequences (including the provision of incentives to high performer and the insistence on remediation for low performers).

Several limitations of this study should be considered. First, only hospitalist providers from a single division were assessed. This may limit the generalizability of our findings. Second, although patients were assured about confidentiality of their responses, they might have provided more favorable answers, because they may have felt uncomfortable rating their physician poorly. One review article of the measurement of healthcare satisfaction indicated that impersonal (mailed) methods result in more criticism and lower satisfaction than assessments made in person using interviews. As the trade‐off, the mailed surveys yield lower response rates that may introduce other forms of bias.[25] Even on the HCHAPS survey report for the same period from our institution, 78% of patients gave top box ratings for our doctors' communication skills, which is at the state average.[26] Similarly, a study that used postdischarge telephone interviews to collect patients' satisfaction with hospitalists' care quality reported an average score of 4.20 out of 5.[27] These findings confirm that highly skewed ratings are common for these types of surveys, irrespective of how or when the data are collected.

Despite the aforementioned limitations, TAISCH use need not be limited to hospitalist physicians. It may also be used to assess allied health professionals or trainees performance, which cannot be assessed by HCHAPS or PG. Applying TAISCH in different hospital settings (eg, emergency department or critical care units), assessing hospitalists' reactions to TAISCH, learning whether TAISCH leads to hospitalists' behavior changes or appraising whether performance can improve in response to coaching interventions for those performing poorly are all research questions that merit additional consideration.

CONCLUSION

TAISCH allows for obtaining patient satisfaction data that are highly attributable to specific hospitalist providers. The data collection method also permits high response rates so that input comes from almost all patients. The timeliness of the TAISCH assessments also makes it possible for real‐time service recovery, which is impossible with other commonly used metrics assessing patient satisfaction. Our next step will include testing the most effective way to provide feedback to providers and to coach these individuals so as to improve performance.

Sepsis is a major cause of death in hospitalized patients.13 It is recommended that patients with sepsis be treated with early appropriate antibiotics, as well as early goal‐directed therapy including fluid and vasopressor support according to evidence‐based guidelines.46 Following such evidence‐based protocols and process‐of‐care interventions has been shown to be associated with better patient outcomes, including decreased mortality.7, 8

Most patients with severe sepsis are cared for in intensive care units (ICUs). At times, there are no beds available in the primary ICU and patients presenting to the hospital with sepsis are cared for in other units. Patients admitted to a non‐preferred clinical inpatient setting are sometimes referred to as overflow.9 ICUs can differ significantly in staffing patterns, equipment, and training.10 It is not known if overflow sepsis patients receive similar care when admitted to non‐primary ICUs.

At our hospital, we have an active bed management system led by the hospitalist division.11 This system includes protocols to place sepsis patients in the overflow ICU if the primary ICU is full. We hypothesized that process‐of‐care interventions would be more strictly adhered to when sepsis patients were in the primary ICU rather than in the overflow unit at our institution.

METHODS

RESULTS

There were 100 patients admitted to the MICU and 41 patients admitted to the CICU during the study period (Table 1). The majority of the patients were admitted to the ICUs directly from the emergency department (ED) (n = 129), with a small number of patients who were transferred from the Medicine floors (n = 12).

There were no significant differences between the 2 study groups in terms of age, sex, primary site of infection, mean APACHE II score, SOFA scores on day 1, chronic organ insufficiency, immune suppression, or need for mechanical ventilation (Table 1). The most common site of infection was lung. There were significantly more patients with severe sepsis in the MICU (88% vs 51%, P <0.001).

DISCUSSION

Since sepsis is more commonly treated in the medical ICU and some data suggests that specialty ICUs may be better at providing desired care,18, 19 we believed that patients treated in the MICU would be more likely to receive guideline‐concordant care. The study refutes our a priori hypothesis and reveals that evidence‐based processes‐of‐care associated with improved outcomes for sepsis are similarly implemented at our institution in the primary and overflow ICU. These findings are important, as ICU bed availability is a frequent problem and many hospitals overflow patients to non‐primary ICUs.9, 20

The observed equivalence in the care delivered may be a function of the relatively high number of patients with sepsis treated in the overflow unit, thereby giving the delivery teams enough experience to provide the desired care. An alternative explanation could be that the residents in CICU brought with them the experience from having previously trained in the MICU. Although, some of the care processes for sepsis patients are influenced by the CPOE (with embedded order sets and protocols), it is unlikely that CPOE can fully account for similarity in care because many processes and therapies (like use of steroids, amount of fluid delivered in first 24 hours, packed red blood cells [PRBC] transfusion, and spontaneous breathing trials) are not embedded within order sets.

The significant difference noted in the areas of deep vein thrombosis (DVT) and gastrointestinal (GI) prophylaxis within 24 hours of ICU admission was unexpected. These preventive therapies are included in initial order sets in the CPOE, which prompt physicians to order them as standard‐of‐care. With respect to DVT prophylaxis, we suspect that some of the difference might be attributable to specific contraindications to its use, which could have been more common in one of the units. There were more patients in MICU on mechanical ventilation (although not statistically significant) and with severe sepsis (statistically significant) at time of admission, which might have contributed to the difference noted in use of GI prophylaxis. It is also plausible that these differences might have disappeared if they were reassessed beyond 24 hours into the ICU admission. We cannot rule out the presence of unit‐ and physician‐level differences that contributed to this. Likewise, there was an unexpected trend towards significance, wherein more patients in CICU had spontaneous breathing trials within 24 hours of admission. This might also be explained by the higher number of patients with severe sepsis in the MICU (preempting any weaning attempts). These caveats aside, it is reassuring that, at our institution, admitting septic patients to the first available ICU bed does not adversely affect important processes‐of‐care.

One might ask whether this study's data should reassure other sites who are boarding septic patients in non‐primary ICUs. Irrespective of the number of patients studied or the degree of statistical significance of the associations, an observational study design cannot prove that boarding septic patients in non‐primary ICUs is either safe or unsafe. However, we hope that readers reflect on, and take inventory of, systems issues that may be different between unitswith an eye towards eliminating variation such that all units managing septic patients are primed to deliver guideline‐concordant care. Other hospitals that use CPOE with sepsis order sets, have protocols for sepsis care, and who train nursing and respiratory therapists to meet high standards might be pleased to see that the patients in our study received comparable, high‐quality care across the 2 units. While our data suggests that boarding patients in overflow units may be safe, these findings would need to be replicated at other sites using prospective designs to prove safety.

Length of emergency room stay prior to admission is associated with higher mortality rates.2123 At many hospitals, critical care beds are a scarce resource such that most hospitals have a policy for the triage of patients to critical care beds.24, 25 Lundberg and colleagues' study demonstrated that patients who developed septic shock on the medical wards experienced delays in receipt of intravenous fluids, inotropic agents and transfer to a critical care setting.26 Thus, rather than waiting in the ED or on the medical service for an MICU bed to become available, it may be most wise to admit a critically sick septic patient to the first available ICU bed, even to an overflow ICU. In a recent study by Sidlow and Aggarwal, 1104 patients discharged from the coronary care unit (CCU) with a non‐cardiac primary diagnosis were compared to patients admitted to the MICU in the same hospital.27 The study found no differences in patient mortality, 30‐day readmission rate, hospital LOS, ICU LOS, and safety outcomes of ventilator‐associated pneumonia and catheter‐associated bloodstream infections between ICUs. However, their study did not examine processes‐of‐care delivered between the primary ICU and the overflow unit, and did not validate the primary diagnoses of patients admitted to the ICU.

Several limitations of this study should be considered. First, this study was conducted at a single center. Second, we used a retrospective study design; however, a prospective study randomizing patients to 1 of the 2 units would likely never be possible. Third, the relatively small number of patients limited the power of the study to detect mortality differences between the units. However, this was a pilot study focused on processes of care as opposed to clinical outcomes. Fourth, it is possible that we did not capture every single patient with sepsis with our keyword search. Our use of a previously validated screening process should have limited the number of missed cases.15, 16 Fifth, although the 2 ICUs have exclusive nursing staff and attending physicians, the housestaff and respiratory therapists do rotate between the 2 ICUs and place orders in the common CPOE. The rotating housestaff may certainly represent a source for confounding, but the large numbers (>30) of evenly spread housestaff over the study period minimizes the potential for any trainee to be responsible for a large proportion of observed practice. Sixth, ICU attendings are the physicians of record and could influence the results. Because no attending physician was on service for more than 4 weeks during the study period, and patients were equally spread over this same time, concerns about clustering and biases this may have created should be minimal but cannot be ruled out. Seventh, some interventions and processes, such as antibiotic administration and measurement of lactate, may have been initiated in the ED, thereby decreasing the potential for differences between the groups. Additionally, we cannot rule out the possibility that factors other than bed availability drove the admission process (we found that the relative proportion of patients admitted to overflow ICU during hours of ambulance diversion was similar to the overflow ICU admissions during non‐ambulance diversion hours). It is possible that some selection bias by the hospitalist assigning patients to specific ICUs influenced their triage decisionsalthough all triaging doctors go through the same process of training in active bed management.11 While more patients admitted to the MICU had severe sepsis, there were no differences between groups in APACHE II or SOFA scores. However, we cannot rule out that there were other residual confounders. Finally, in a small number of cases (4/41, 10%), the CICU team consulted the MICU attending for assistance. This input had the potential to reduce disparities in care between the units.

Overflowing patients to non‐primary ICUs occurs in many hospitals. Our study demonstrates that sepsis treatment for overflow patients may be similar to that received in the primary ICU. While a large multicentered and randomized trial could determine whether significant management and outcome differences exist between primary and overflow ICUs, feasibility concerns make it unlikely that such a study will ever be conducted.

Sepsis is a major cause of death in hospitalized patients.13 It is recommended that patients with sepsis be treated with early appropriate antibiotics, as well as early goal‐directed therapy including fluid and vasopressor support according to evidence‐based guidelines.46 Following such evidence‐based protocols and process‐of‐care interventions has been shown to be associated with better patient outcomes, including decreased mortality.7, 8

Most patients with severe sepsis are cared for in intensive care units (ICUs). At times, there are no beds available in the primary ICU and patients presenting to the hospital with sepsis are cared for in other units. Patients admitted to a non‐preferred clinical inpatient setting are sometimes referred to as overflow.9 ICUs can differ significantly in staffing patterns, equipment, and training.10 It is not known if overflow sepsis patients receive similar care when admitted to non‐primary ICUs.

At our hospital, we have an active bed management system led by the hospitalist division.11 This system includes protocols to place sepsis patients in the overflow ICU if the primary ICU is full. We hypothesized that process‐of‐care interventions would be more strictly adhered to when sepsis patients were in the primary ICU rather than in the overflow unit at our institution.

METHODS

RESULTS

There were 100 patients admitted to the MICU and 41 patients admitted to the CICU during the study period (Table 1). The majority of the patients were admitted to the ICUs directly from the emergency department (ED) (n = 129), with a small number of patients who were transferred from the Medicine floors (n = 12).

There were no significant differences between the 2 study groups in terms of age, sex, primary site of infection, mean APACHE II score, SOFA scores on day 1, chronic organ insufficiency, immune suppression, or need for mechanical ventilation (Table 1). The most common site of infection was lung. There were significantly more patients with severe sepsis in the MICU (88% vs 51%, P <0.001).