Medical emergency teams (METs) were introduced more than a decade ago in Australia and the United Kingdom to rapidly identify and manage seriously ill patients at risk of cardiopulmonary arrest and other high‐risk conditions.1 METs, known in the United States as rapid response teams (RRTs), have been slow to be adopted thus far but are quickly gaining ground. Despite numerous studies indicating long‐term patient outcomes are poor following cardiac resuscitation in the hospital, the benefits of early intervention have sometimes been overlooked.25 Several observational studies and a retrospective analysis that included the Medical Emergency Response Improvement Team (MERIT) in Pittsburgh showed that introduction of a MET apparently has the potential to decrease the incidence of unanticipated intensive care unit (ICU) admissions and in‐hospital morbidity and mortality from unexpected cardiopulmonary arrest.69 Furthermore, the use of a MET as a quality improvement tool to detect medical errors and effect systemwide interventions is promising.10 Most recently, the Institute for Healthcare Improvement (IHI) and the American Hospital Association challenged health care organizations to redesign patient safety systems to prevent avoidable deaths in its 100K Lives Campaign. One of the 6 proposed core interventions was the deployment of rapid response teams at the first sign of patient decline.11

Despite these reports of success, a recent large cluster‐randomized controlled trial did not yield the same positive results. In this well‐designed study of 23 Australian hospitals, the Medical Early Response, Intervention and Therapy (MERIT) study investigators found the incidence of cardiac arrest, unplanned ICU admissions, and unexpected death essentially unchanged despite large increases in how often the emergency team was called.12 One possible explanation why these findings conflicted with previous favorable results is that the ultimate impact of a MET may depend on the effectiveness of implementation strategies. To derive the benefits of a MET/RRT, hospitals must increasingly focus on identifying barriers to implementation and address practical issues that may undermine their long‐term effectiveness.

In this article we describe in detail the process of establishing an RRT at our urban, academic hospital and the modifications that became necessary as we rolled out the intervention and encountered obstacles. This analysis was undertaken as a quality improvement (QI) activity. To our knowledge, this is one of the few recent published descriptions of the experiences of implementing an RRT in the United States since earlier work in Pittsburgh.9, 13

METHODS

Temple University Hospital is a tertiary care academic hospital in urban Philadelphia, Pennsylvania. Our RRT was first implemented July 1, 2004, and in the first 12 months of initiation, it was activated 307 times. The RRT at Temple University Hospital was designed to be accessible 24 hours a day, 7 days a week. The daytime team (8 am‐5 pm) is composed of an attending physician (a hospitalist trained as a general internist), a senior internal medicine resident, a critical care nurse, a nurse manager, a pharmacist, and a respiratory therapist. In addition, both a transporter and a member of the admissions office respond to all rapid response team calls but do not get clinically involved in patient care. For nighttime (5 pm‐8 am) and weekend coverage the hospitalist is replaced by an on‐site pulmonary critical care physician, but the remainder of the team is unchanged. All RRT members carry beepers synchronized to provide the location of an RRT activation. In addition, all RRT calls are simultaneously announced on the overhead paging system. No changes were made to the existing cardiac arrest team (code team) at the hospital, which remained a 24‐hour response team for patients found to be in true cardiopulmonary arrest and was comprised of on‐call internal medicine house staff (but no hospitalist attending physician), a respiratory therapist, a pharmacist, a critical care nurse, a nurse manager, and, most notably, an anesthesiologist for emergent intubation and airway management.

The RRT was intended for use within the physical confines of Temple University Hospital and its immediately adjacent grounds. Within the hospital the main locations defined were: inpatient areas, including patient rooms and hallways of the medical‐surgical units of the inpatient tower, as well as the burn, coronary, medical, neurological, neurosurgical, and surgical intensive care units; off‐unit/procedural areas, including diagnostic/emnterventional radiology, the gastroenterology endoscopy suite, the pulmonary procedure suite and pulmonary function lab, the cardiac catheterization/ECHO/stress Lab, the inpatient dialysis unit, and the physical therapy gym, all areas where inpatients are routinely transported during their hospital admission for workup/treatment and where outpatients go for scheduled procedures and therapies; and outpatient/common areas, including all the general medical and subspecialty outpatient clinics in 2 separate outpatient towers (Outpatient Building and Parkinson Pavilion) with direct access from the main hospital building, the outpatient pharmacy, the elevators, the hallways in the outpatient sections of the hospital, all lobbies, and the immediately adjacent outside grounds.

Prior to the launch date of the RRT, clinical criteria were established to help guide staff about when an RRT might be called (Fig. 1). These were based in part on early literature on the clinical markers that most often precede clinical deterioration.14, 15 In addition, 2 much broader categories for RRT activation were added (Inability to reach the patient's primary team of treating physicians for any of the above and Any potentially serious medical errors or adverse events) in order to minimize the need for a very specific physiologic definition to be met in order to activate the team. Physicians, nurses, and other staff with significant daily contact with inpatients and outpatients were in‐serviced about the purpose of the RRT and how to activate the system via the hospital paging operator. Laminated cards with RRT criteria were distributed to all hospital personnel, and educational posters were displayed prominently throughout the hospital.

Figure 1

Each RRT event was to be assessed by team members using a standardized evaluation form (Fig. 2), with primary responsibility going to the physician team leader. In the initial phases of implementation, these forms were kept in the offices of the Section of Hospital Medicine for the use of hospitalist attending physician team leaders. Later on in the year they were kept in the pharmacist's RRT medication bag. These forms were collected at the completion of each RRT event or faxed to a central location and then entered into a database maintained by the hospital's Department of Patient Safety Operations. Weekly debriefing meetings to review all RRT events from the preceding week were attended by representatives from patient safety, respiratory, nursing, hospital medicine, and the pharmacy. Attempts were made to identify the issues that led to selected RRT activations, to obtain patient follow‐up from the clinical event, and to evaluate the performance of the team. Throughout these weekly meetings, QI strategies for improving the effectiveness of the RRT were identified and implemented.

Figure 2

The core outcome measures that were used to assess RRT performance were: appropriateness of the RRT activation, percentage of patients who were stabilized, percentage of patients who were transferred to a higher level of care, and overall team performance.

In the weekly meeting of the RRT evaluation committee, at which each RRT was reviewed by the clinical team, each scenario and details of the event were reviewed to determine whether the RRT activation was appropriate, whether the intervention was successful, and whether there were any issues with the team performance. After a thorough discussion of each case and review of additional data from the chart if necessary, the RRT evaluation committee reached a consensus about each of these measures.

We also tracked the number of code team activations from the year preceding establishment of the RRT (2003‐2004) through the year during which the RRT was established (2004‐2005). Because all calls for both the RRT and the code team go first to the hospital operator, we reviewed the hospital paging operators' logs for the entire 12‐month period to track the rate of code team events to RRT events on a monthly basis.

RESULTS

In a 12‐month period, the RRT was activated 307 times, as recorded in the hospital operator logs. In the year preceding inception of the RRT, there were 272 code team activations. In the first 12 months concurrent with RRT implementation, the code team was activated 258 times. Overall, at their discretion the team leaders converted 13% of the 307 RRT activations to traditional code team activations.

There were 11 RRT activations in July, the first month of implementation, and 14 activations in the second month. At that point, the internal hospital newsletter released a feature on the new RRT, and our patient safety officer/director of patient safety operations made a concerted effort to educate hospital administration and the Graduate Medical Education Committee (GMEC); as a result, utilization picked up. From September onward through the remainder of the academic year, an average of 28 RRT activations occurred each month (range 20‐37), whereas an average of 22 codes took place each month (range 12‐27). The numbers of RRT versus code team activations are plotted in Figure 3. A trend line for the number of code team activations per month in 2003, the year prior to implementation of the RRT, was added for comparison; it conveys the slight overall decrease in the number of codes as the RRT took effect (average of 23 codes per month, range 15‐31).

Figure 3

Physician evaluation forms were returned for 170 of the 307 RRT events (55%). The main inpatient tower was the site of 42% of these RRT activations, followed by the outpatient/common areas, where 19% of the activations occurred, and off‐unit/procedural areas, the site of 18%. Table 2 provides information on specific location, reason for call, and disposition of a sample of the RRT activations in the non‐inpatient areas. Time of day was noted in 76.8% of events. Of these, 82.9% occurred during the traditional day shift (7 am‐7 pm) and 17.1% on night shift (7 pm‐7 am). Most RRT activations occurred between 8 am and 4 pm. Daytime events heavily outnumbered nighttime events regardless of location.

Physician team leaders largely believed a specific underlying clinical diagnosis was responsible for 59% of the RRT activations, followed by adverse drug reactions (3.5%), physician error (1.8%), and nursing error (0.6%). When an underlying clinical diagnosis or organ system was suspected, it was most frequently pulmonary (32%), followed by neurological (14%) and cardiac (11%). It was believed that 32% of events were for other reason not listed. Table 1 provides the breakdown of other underlying diagnoses in RRT events.

In the judgment of evaluators, the system was utilized appropriately in 98% of the evaluated events. Eighty‐five percent of RRT activations were believed to have prevented further clinical deterioration, though it was also thought that 3% of patients deteriorated despite the efforts of the team. Disposition of the patient following an RRT event was noted 87% of the time, and it was believed that 88% of the patients were stabilized. Of the formally evaluated RRT events, team members were largely satisfied with the response and the functioning of the team, stating for 68% of the events that the team performed without a problem.

DISCUSSION

We have found the RRT to be an effective but challenging‐to‐implement QI intervention to increase patient safety at our academic institution. The Australian MERIT investigators recently suggested that despite growing evidence of the benefits of MET/RRT systems, long‐term success may depend most on effective implementation strategies.12 We experienced firsthand these challenges in the first year of our new RRT system.

Large system changes in a hospital are especially fraught with danger because of the unique aspects of health care delivery systems. As Reid commented in an editorial about the emerging use of the MET system in the United Kingdom, Despite potential advantages to patients, ensuring appropriate utilization was difficult because of cultural barriers. Traditional hierarchical behaviors that dictate how doctors and nurses react and work got in the way of people calling these life saving teams.17

Our weekly multidisciplinary RRT debriefings were the most crucial component of our implementation strategy. Many latent systems issues were uncovered, as well as more subtle problems such as lack of coordination of care, communication errors, gaps in patient handoffs or sign‐out. Previous studies by the Pittsburgh MERIT team have validated such retrospective categorization of errors uncovered by MET responses.10

However, neither that group nor the Australian MERIT study investigators specifically addressed the importance of the feedback process in RRT implementation. A strength of our system is that modifications to the RRT are made prospectively and in real time based on feedback from active RRT members during debriefing. In fact, the success of our RRT underscores the importance of open communication among hospitalists, house staff, nurses, pharmacists, and ancillary staff in multidisciplinary patient safety and QI endeavors. Everything from the responsibilities of team members to equipment evolved over the 12‐month period in order to improve the function and effectiveness of the team and was almost entirely based on feedback from the RRT doctors and nurses on the front lines. Suggestions from the evaluation forms were given serious consideration at every RRT evaluation committee debriefing. By optimizing the efficient operation of the RRT, we hope to continue to improve outcomes.

We believe a key to the success of our debriefing process was the constant attendance of our patient safety officer/chief medical officer and director of patient safety operations, who both encouraged active participation. Early on in the process, comments were made principally by physician and critical care nurse RRT members, and the dynamic was a bit one‐sided. However, we quickly saw a noticeable and sustained increase in participation by pharmacists and respiratory therapists, and by year's end, they had offered some of the most valuable practical suggestions, which resulted in a more efficient response. As the year went on and real changes were made quickly, all groups were much more vocal and willing to bounce ideas around the room, and the team dynamic and spirit of the group effort improved substantially.

Previous studies have focused on the impact of METs/RRTs on the rate of inpatient cardiac arrests. However, we found that nearly as many RRT events occurred off the inpatient units, for instance, when admitted patients were transported to other areas such as radiology, procedural suites, physical therapy, or dialysis and when scheduled outpatients arrived for their appointments. In addition, a large number of RRT calls came from outpatient departments and common areas of the hospital such as lobbies, hallways, and waiting rooms, mostly involving outpatients and visitors, but not infrequently hospital employees were involved as well. This unexpected and, to our knowledge, previously unreported finding is mirrored in the distribution of RRT activations throughout the course of the day. Most events occurred during the traditional day shift of 7 am‐7 pm, and were heavily clustered between 8 am and 4 pm. In most American hospitals, these are the hours during which outpatients and visitors make up a significant proportion of the hospital population and during which most elective procedures on inpatients occur. Prior to the introduction of our RRT, no specific system was in place for emergent triage, assessment, and expedited treatment of off‐unit patients, outpatients, and visitors. Most often, the code team was mobilized, sometimes taking them to remote locations and making them unavailable for true inpatient cardiopulmonary arrests. Our RRT seems to have the potential to fill a much‐needed gap in patient safety, offering off‐unit patients, outpatients, and visitors a safety net while in our hospital. No prior descriptions of RRT or MET implementation have touched on this area. It would be interesting to see if other hospitals with RRTs have had a similar experience in order to determine whether having an RRT dedicated specifically to the outpatient and common areas of the hospital might provide even more targeted efforts and efficient response times. Thus, the benefits of our RRT seemed to extend beyond a simple reduction in the number of in‐hospital cardiopulmonary arrests and into an unanticipated patient safety black hole.

Implementation of the RRT specifically in academic medical centers has been limited to date. In our opinion, the academic environment is an ideal area for RRTs (because the most critically ill patients often are cared for on teaching services by junior house officers), but it is also a challenging arena in which to make change (because of the complex hierarchy of teaching hospitals). We chose to have an attending physician lead our RRT efforts for the most part. However, residents always participated, and not infrequently led, as key team members. As a commentator on the Australian RRT system pointed out, it is important that junior medical staff [feel empowered] to call for immediate assistance when they are concerned about their patient, but may not have the experience, knowledge, confidence or skills necessary to manage them appropriately.18 We believe that the RRT serves as a valuable educational forum for resident education. Academic centers that develop RRTs must work to integrate the teams into an educational context while simultaneously providing patients with the most experienced and knowledgeable clinical team to address their needs at a time when appropriate clinical decision making is critical. Therefore, the residents who participate in our RRT are formally evaluated by the hospitalists using a standard program evaluation form that encompasses the Accreditation Council for Graduate Medical Education (ACGME) core competencies.19

Through the first year of our RRT system and beyond, activation of the code team and RRT shifted as more RRT activations were recorded and fewer codes were called. Concerted educational efforts and reinforcement of the criteria for calling the RRT had a definite sustained impact of helping staff to become comfortable with using the system. At our institution, it has been difficult to definitively conclude whether RRT calls prevented codes or merely substituted for them at times, especially because 13% of all RRT activations were subsequently converted to code team calls. The Australian MERIT study investigators, despite an excellent study design of a large multicenter trial, also were unable to demonstrate a true decrease in the cardiac arrest rate.12 Much more significant to us, especially in the first year of implementation, was learning that the vast majority of physician RRT leaders perceived activation of the team to occur appropriately and to play a role in preventing clinical deterioration of patients. None of the other RRT or MET implementation studies that we reviewed commented specifically on these areas. It will be interesting to continue to follow these trends, as we expect the use of RRTs to become even more defined. Over time, we will no doubt be better able to determine whether RRTs have a true, sustained impact on preventing patient deterioration and inpatient cardiopulmonary arrests while maintaining a high rate of physician satisfaction that the team is being activated for legitimate reasons.

Our descriptive study had some limitations. The number of RRT evaluations received, while adequate for preliminary analysis, may not accurately represent the 307 activations of the system that occurred in the first 12 months. We suspect that this underreporting, especially in the first half of the year, was in large part a result of relying on team leaders to voluntarily return data forms at the conclusion of each RRT event. RRT evaluations in the second half of the year were more actively distributed at the point of care to the team leader directly by the pharmacist and were more diligently followed up on. Forms are now readily available in the team pharmacist's fanny pack, which was done because of quality improvement feedback from physicians at a debriefing meeting. Since those interventions, there has been a dramatic improvement in the capture of event data and the timely submission of forms. We expect and have demanded close to a 100% return of the forms in the second year of our RRT system, which will vastly improve our analysis. We were also surprised that despite the comprehensiveness of our RRT activation criteria, 32% of physicians were unable to find a match with a clinical indication on the list, indicating unanticipated reasons for calling an RRT. We will continually strive to improve the specificity of future data for planning purposes and training initiatives. However, in some way this confirms our belief that RRTs occur for such a wide variety of reasons that they cannot always be limited to the major clinical categories. On a similar note, we regret not adding a specific category under Outcomes on the evaluation form to include the possibility that RRT members might have offered palliative care or changes in code/do not resuscitate (DNR) status to patients or families. Given that our hospital has both a code team and an RRT begs the question of whether mortality rates might be affected if patients who prior to the RRT might have had a full resuscitation effort were made DNR. In the future, this would be an interesting issue to consider in analysis. Carefully categorizing RRT events is critical to continued success. Further work involving formal team skills training for RRT members, including use of the medical school's clinical simulators for mock RRT scenarios, is planned. These sessions are planned to review performance and clinical decision making for the most common scenarios that we have found to be involved in RRT activations. The 307 activations of the RRT in our first year have clearly set us on the path toward defining predictive rules and directed skills training for earlier identification of patient problems. Further outcome analyses of these efforts will be crucial.

CONCLUSIONS

An RRT was successfully introduced into an academic medical center. The team was heavily utilized in the first 12 months after the program was initiated, especially for off‐unit inpatients and those in outpatient/common areas, perhaps filling a gap in hospital patient safety. The keys to the early success of implementation of our RRT were multidisciplinary input and improvements made in real time. The long‐term effects of the RRT on the culture of patient safety in our institution and throughout the United States remain to be seen but are promising.

Medical emergency teams (METs) were introduced more than a decade ago in Australia and the United Kingdom to rapidly identify and manage seriously ill patients at risk of cardiopulmonary arrest and other high‐risk conditions.1 METs, known in the United States as rapid response teams (RRTs), have been slow to be adopted thus far but are quickly gaining ground. Despite numerous studies indicating long‐term patient outcomes are poor following cardiac resuscitation in the hospital, the benefits of early intervention have sometimes been overlooked.25 Several observational studies and a retrospective analysis that included the Medical Emergency Response Improvement Team (MERIT) in Pittsburgh showed that introduction of a MET apparently has the potential to decrease the incidence of unanticipated intensive care unit (ICU) admissions and in‐hospital morbidity and mortality from unexpected cardiopulmonary arrest.69 Furthermore, the use of a MET as a quality improvement tool to detect medical errors and effect systemwide interventions is promising.10 Most recently, the Institute for Healthcare Improvement (IHI) and the American Hospital Association challenged health care organizations to redesign patient safety systems to prevent avoidable deaths in its 100K Lives Campaign. One of the 6 proposed core interventions was the deployment of rapid response teams at the first sign of patient decline.11

Despite these reports of success, a recent large cluster‐randomized controlled trial did not yield the same positive results. In this well‐designed study of 23 Australian hospitals, the Medical Early Response, Intervention and Therapy (MERIT) study investigators found the incidence of cardiac arrest, unplanned ICU admissions, and unexpected death essentially unchanged despite large increases in how often the emergency team was called.12 One possible explanation why these findings conflicted with previous favorable results is that the ultimate impact of a MET may depend on the effectiveness of implementation strategies. To derive the benefits of a MET/RRT, hospitals must increasingly focus on identifying barriers to implementation and address practical issues that may undermine their long‐term effectiveness.

In this article we describe in detail the process of establishing an RRT at our urban, academic hospital and the modifications that became necessary as we rolled out the intervention and encountered obstacles. This analysis was undertaken as a quality improvement (QI) activity. To our knowledge, this is one of the few recent published descriptions of the experiences of implementing an RRT in the United States since earlier work in Pittsburgh.9, 13

METHODS

Temple University Hospital is a tertiary care academic hospital in urban Philadelphia, Pennsylvania. Our RRT was first implemented July 1, 2004, and in the first 12 months of initiation, it was activated 307 times. The RRT at Temple University Hospital was designed to be accessible 24 hours a day, 7 days a week. The daytime team (8 am‐5 pm) is composed of an attending physician (a hospitalist trained as a general internist), a senior internal medicine resident, a critical care nurse, a nurse manager, a pharmacist, and a respiratory therapist. In addition, both a transporter and a member of the admissions office respond to all rapid response team calls but do not get clinically involved in patient care. For nighttime (5 pm‐8 am) and weekend coverage the hospitalist is replaced by an on‐site pulmonary critical care physician, but the remainder of the team is unchanged. All RRT members carry beepers synchronized to provide the location of an RRT activation. In addition, all RRT calls are simultaneously announced on the overhead paging system. No changes were made to the existing cardiac arrest team (code team) at the hospital, which remained a 24‐hour response team for patients found to be in true cardiopulmonary arrest and was comprised of on‐call internal medicine house staff (but no hospitalist attending physician), a respiratory therapist, a pharmacist, a critical care nurse, a nurse manager, and, most notably, an anesthesiologist for emergent intubation and airway management.

The RRT was intended for use within the physical confines of Temple University Hospital and its immediately adjacent grounds. Within the hospital the main locations defined were: inpatient areas, including patient rooms and hallways of the medical‐surgical units of the inpatient tower, as well as the burn, coronary, medical, neurological, neurosurgical, and surgical intensive care units; off‐unit/procedural areas, including diagnostic/emnterventional radiology, the gastroenterology endoscopy suite, the pulmonary procedure suite and pulmonary function lab, the cardiac catheterization/ECHO/stress Lab, the inpatient dialysis unit, and the physical therapy gym, all areas where inpatients are routinely transported during their hospital admission for workup/treatment and where outpatients go for scheduled procedures and therapies; and outpatient/common areas, including all the general medical and subspecialty outpatient clinics in 2 separate outpatient towers (Outpatient Building and Parkinson Pavilion) with direct access from the main hospital building, the outpatient pharmacy, the elevators, the hallways in the outpatient sections of the hospital, all lobbies, and the immediately adjacent outside grounds.

Prior to the launch date of the RRT, clinical criteria were established to help guide staff about when an RRT might be called (Fig. 1). These were based in part on early literature on the clinical markers that most often precede clinical deterioration.14, 15 In addition, 2 much broader categories for RRT activation were added (Inability to reach the patient's primary team of treating physicians for any of the above and Any potentially serious medical errors or adverse events) in order to minimize the need for a very specific physiologic definition to be met in order to activate the team. Physicians, nurses, and other staff with significant daily contact with inpatients and outpatients were in‐serviced about the purpose of the RRT and how to activate the system via the hospital paging operator. Laminated cards with RRT criteria were distributed to all hospital personnel, and educational posters were displayed prominently throughout the hospital.

Figure 1

Each RRT event was to be assessed by team members using a standardized evaluation form (Fig. 2), with primary responsibility going to the physician team leader. In the initial phases of implementation, these forms were kept in the offices of the Section of Hospital Medicine for the use of hospitalist attending physician team leaders. Later on in the year they were kept in the pharmacist's RRT medication bag. These forms were collected at the completion of each RRT event or faxed to a central location and then entered into a database maintained by the hospital's Department of Patient Safety Operations. Weekly debriefing meetings to review all RRT events from the preceding week were attended by representatives from patient safety, respiratory, nursing, hospital medicine, and the pharmacy. Attempts were made to identify the issues that led to selected RRT activations, to obtain patient follow‐up from the clinical event, and to evaluate the performance of the team. Throughout these weekly meetings, QI strategies for improving the effectiveness of the RRT were identified and implemented.

Figure 2

The core outcome measures that were used to assess RRT performance were: appropriateness of the RRT activation, percentage of patients who were stabilized, percentage of patients who were transferred to a higher level of care, and overall team performance.

In the weekly meeting of the RRT evaluation committee, at which each RRT was reviewed by the clinical team, each scenario and details of the event were reviewed to determine whether the RRT activation was appropriate, whether the intervention was successful, and whether there were any issues with the team performance. After a thorough discussion of each case and review of additional data from the chart if necessary, the RRT evaluation committee reached a consensus about each of these measures.

We also tracked the number of code team activations from the year preceding establishment of the RRT (2003‐2004) through the year during which the RRT was established (2004‐2005). Because all calls for both the RRT and the code team go first to the hospital operator, we reviewed the hospital paging operators' logs for the entire 12‐month period to track the rate of code team events to RRT events on a monthly basis.

RESULTS

In a 12‐month period, the RRT was activated 307 times, as recorded in the hospital operator logs. In the year preceding inception of the RRT, there were 272 code team activations. In the first 12 months concurrent with RRT implementation, the code team was activated 258 times. Overall, at their discretion the team leaders converted 13% of the 307 RRT activations to traditional code team activations.

There were 11 RRT activations in July, the first month of implementation, and 14 activations in the second month. At that point, the internal hospital newsletter released a feature on the new RRT, and our patient safety officer/director of patient safety operations made a concerted effort to educate hospital administration and the Graduate Medical Education Committee (GMEC); as a result, utilization picked up. From September onward through the remainder of the academic year, an average of 28 RRT activations occurred each month (range 20‐37), whereas an average of 22 codes took place each month (range 12‐27). The numbers of RRT versus code team activations are plotted in Figure 3. A trend line for the number of code team activations per month in 2003, the year prior to implementation of the RRT, was added for comparison; it conveys the slight overall decrease in the number of codes as the RRT took effect (average of 23 codes per month, range 15‐31).

Figure 3

Physician evaluation forms were returned for 170 of the 307 RRT events (55%). The main inpatient tower was the site of 42% of these RRT activations, followed by the outpatient/common areas, where 19% of the activations occurred, and off‐unit/procedural areas, the site of 18%. Table 2 provides information on specific location, reason for call, and disposition of a sample of the RRT activations in the non‐inpatient areas. Time of day was noted in 76.8% of events. Of these, 82.9% occurred during the traditional day shift (7 am‐7 pm) and 17.1% on night shift (7 pm‐7 am). Most RRT activations occurred between 8 am and 4 pm. Daytime events heavily outnumbered nighttime events regardless of location.

Physician team leaders largely believed a specific underlying clinical diagnosis was responsible for 59% of the RRT activations, followed by adverse drug reactions (3.5%), physician error (1.8%), and nursing error (0.6%). When an underlying clinical diagnosis or organ system was suspected, it was most frequently pulmonary (32%), followed by neurological (14%) and cardiac (11%). It was believed that 32% of events were for other reason not listed. Table 1 provides the breakdown of other underlying diagnoses in RRT events.

In the judgment of evaluators, the system was utilized appropriately in 98% of the evaluated events. Eighty‐five percent of RRT activations were believed to have prevented further clinical deterioration, though it was also thought that 3% of patients deteriorated despite the efforts of the team. Disposition of the patient following an RRT event was noted 87% of the time, and it was believed that 88% of the patients were stabilized. Of the formally evaluated RRT events, team members were largely satisfied with the response and the functioning of the team, stating for 68% of the events that the team performed without a problem.

DISCUSSION

We have found the RRT to be an effective but challenging‐to‐implement QI intervention to increase patient safety at our academic institution. The Australian MERIT investigators recently suggested that despite growing evidence of the benefits of MET/RRT systems, long‐term success may depend most on effective implementation strategies.12 We experienced firsthand these challenges in the first year of our new RRT system.

Large system changes in a hospital are especially fraught with danger because of the unique aspects of health care delivery systems. As Reid commented in an editorial about the emerging use of the MET system in the United Kingdom, Despite potential advantages to patients, ensuring appropriate utilization was difficult because of cultural barriers. Traditional hierarchical behaviors that dictate how doctors and nurses react and work got in the way of people calling these life saving teams.17

Our weekly multidisciplinary RRT debriefings were the most crucial component of our implementation strategy. Many latent systems issues were uncovered, as well as more subtle problems such as lack of coordination of care, communication errors, gaps in patient handoffs or sign‐out. Previous studies by the Pittsburgh MERIT team have validated such retrospective categorization of errors uncovered by MET responses.10

However, neither that group nor the Australian MERIT study investigators specifically addressed the importance of the feedback process in RRT implementation. A strength of our system is that modifications to the RRT are made prospectively and in real time based on feedback from active RRT members during debriefing. In fact, the success of our RRT underscores the importance of open communication among hospitalists, house staff, nurses, pharmacists, and ancillary staff in multidisciplinary patient safety and QI endeavors. Everything from the responsibilities of team members to equipment evolved over the 12‐month period in order to improve the function and effectiveness of the team and was almost entirely based on feedback from the RRT doctors and nurses on the front lines. Suggestions from the evaluation forms were given serious consideration at every RRT evaluation committee debriefing. By optimizing the efficient operation of the RRT, we hope to continue to improve outcomes.

We believe a key to the success of our debriefing process was the constant attendance of our patient safety officer/chief medical officer and director of patient safety operations, who both encouraged active participation. Early on in the process, comments were made principally by physician and critical care nurse RRT members, and the dynamic was a bit one‐sided. However, we quickly saw a noticeable and sustained increase in participation by pharmacists and respiratory therapists, and by year's end, they had offered some of the most valuable practical suggestions, which resulted in a more efficient response. As the year went on and real changes were made quickly, all groups were much more vocal and willing to bounce ideas around the room, and the team dynamic and spirit of the group effort improved substantially.

Previous studies have focused on the impact of METs/RRTs on the rate of inpatient cardiac arrests. However, we found that nearly as many RRT events occurred off the inpatient units, for instance, when admitted patients were transported to other areas such as radiology, procedural suites, physical therapy, or dialysis and when scheduled outpatients arrived for their appointments. In addition, a large number of RRT calls came from outpatient departments and common areas of the hospital such as lobbies, hallways, and waiting rooms, mostly involving outpatients and visitors, but not infrequently hospital employees were involved as well. This unexpected and, to our knowledge, previously unreported finding is mirrored in the distribution of RRT activations throughout the course of the day. Most events occurred during the traditional day shift of 7 am‐7 pm, and were heavily clustered between 8 am and 4 pm. In most American hospitals, these are the hours during which outpatients and visitors make up a significant proportion of the hospital population and during which most elective procedures on inpatients occur. Prior to the introduction of our RRT, no specific system was in place for emergent triage, assessment, and expedited treatment of off‐unit patients, outpatients, and visitors. Most often, the code team was mobilized, sometimes taking them to remote locations and making them unavailable for true inpatient cardiopulmonary arrests. Our RRT seems to have the potential to fill a much‐needed gap in patient safety, offering off‐unit patients, outpatients, and visitors a safety net while in our hospital. No prior descriptions of RRT or MET implementation have touched on this area. It would be interesting to see if other hospitals with RRTs have had a similar experience in order to determine whether having an RRT dedicated specifically to the outpatient and common areas of the hospital might provide even more targeted efforts and efficient response times. Thus, the benefits of our RRT seemed to extend beyond a simple reduction in the number of in‐hospital cardiopulmonary arrests and into an unanticipated patient safety black hole.

Implementation of the RRT specifically in academic medical centers has been limited to date. In our opinion, the academic environment is an ideal area for RRTs (because the most critically ill patients often are cared for on teaching services by junior house officers), but it is also a challenging arena in which to make change (because of the complex hierarchy of teaching hospitals). We chose to have an attending physician lead our RRT efforts for the most part. However, residents always participated, and not infrequently led, as key team members. As a commentator on the Australian RRT system pointed out, it is important that junior medical staff [feel empowered] to call for immediate assistance when they are concerned about their patient, but may not have the experience, knowledge, confidence or skills necessary to manage them appropriately.18 We believe that the RRT serves as a valuable educational forum for resident education. Academic centers that develop RRTs must work to integrate the teams into an educational context while simultaneously providing patients with the most experienced and knowledgeable clinical team to address their needs at a time when appropriate clinical decision making is critical. Therefore, the residents who participate in our RRT are formally evaluated by the hospitalists using a standard program evaluation form that encompasses the Accreditation Council for Graduate Medical Education (ACGME) core competencies.19

Through the first year of our RRT system and beyond, activation of the code team and RRT shifted as more RRT activations were recorded and fewer codes were called. Concerted educational efforts and reinforcement of the criteria for calling the RRT had a definite sustained impact of helping staff to become comfortable with using the system. At our institution, it has been difficult to definitively conclude whether RRT calls prevented codes or merely substituted for them at times, especially because 13% of all RRT activations were subsequently converted to code team calls. The Australian MERIT study investigators, despite an excellent study design of a large multicenter trial, also were unable to demonstrate a true decrease in the cardiac arrest rate.12 Much more significant to us, especially in the first year of implementation, was learning that the vast majority of physician RRT leaders perceived activation of the team to occur appropriately and to play a role in preventing clinical deterioration of patients. None of the other RRT or MET implementation studies that we reviewed commented specifically on these areas. It will be interesting to continue to follow these trends, as we expect the use of RRTs to become even more defined. Over time, we will no doubt be better able to determine whether RRTs have a true, sustained impact on preventing patient deterioration and inpatient cardiopulmonary arrests while maintaining a high rate of physician satisfaction that the team is being activated for legitimate reasons.

Our descriptive study had some limitations. The number of RRT evaluations received, while adequate for preliminary analysis, may not accurately represent the 307 activations of the system that occurred in the first 12 months. We suspect that this underreporting, especially in the first half of the year, was in large part a result of relying on team leaders to voluntarily return data forms at the conclusion of each RRT event. RRT evaluations in the second half of the year were more actively distributed at the point of care to the team leader directly by the pharmacist and were more diligently followed up on. Forms are now readily available in the team pharmacist's fanny pack, which was done because of quality improvement feedback from physicians at a debriefing meeting. Since those interventions, there has been a dramatic improvement in the capture of event data and the timely submission of forms. We expect and have demanded close to a 100% return of the forms in the second year of our RRT system, which will vastly improve our analysis. We were also surprised that despite the comprehensiveness of our RRT activation criteria, 32% of physicians were unable to find a match with a clinical indication on the list, indicating unanticipated reasons for calling an RRT. We will continually strive to improve the specificity of future data for planning purposes and training initiatives. However, in some way this confirms our belief that RRTs occur for such a wide variety of reasons that they cannot always be limited to the major clinical categories. On a similar note, we regret not adding a specific category under Outcomes on the evaluation form to include the possibility that RRT members might have offered palliative care or changes in code/do not resuscitate (DNR) status to patients or families. Given that our hospital has both a code team and an RRT begs the question of whether mortality rates might be affected if patients who prior to the RRT might have had a full resuscitation effort were made DNR. In the future, this would be an interesting issue to consider in analysis. Carefully categorizing RRT events is critical to continued success. Further work involving formal team skills training for RRT members, including use of the medical school's clinical simulators for mock RRT scenarios, is planned. These sessions are planned to review performance and clinical decision making for the most common scenarios that we have found to be involved in RRT activations. The 307 activations of the RRT in our first year have clearly set us on the path toward defining predictive rules and directed skills training for earlier identification of patient problems. Further outcome analyses of these efforts will be crucial.

CONCLUSIONS

An RRT was successfully introduced into an academic medical center. The team was heavily utilized in the first 12 months after the program was initiated, especially for off‐unit inpatients and those in outpatient/common areas, perhaps filling a gap in hospital patient safety. The keys to the early success of implementation of our RRT were multidisciplinary input and improvements made in real time. The long‐term effects of the RRT on the culture of patient safety in our institution and throughout the United States remain to be seen but are promising.

Medical emergency teams (METs) were introduced more than a decade ago in Australia and the United Kingdom to rapidly identify and manage seriously ill patients at risk of cardiopulmonary arrest and other high‐risk conditions.1 METs, known in the United States as rapid response teams (RRTs), have been slow to be adopted thus far but are quickly gaining ground. Despite numerous studies indicating long‐term patient outcomes are poor following cardiac resuscitation in the hospital, the benefits of early intervention have sometimes been overlooked.25 Several observational studies and a retrospective analysis that included the Medical Emergency Response Improvement Team (MERIT) in Pittsburgh showed that introduction of a MET apparently has the potential to decrease the incidence of unanticipated intensive care unit (ICU) admissions and in‐hospital morbidity and mortality from unexpected cardiopulmonary arrest.69 Furthermore, the use of a MET as a quality improvement tool to detect medical errors and effect systemwide interventions is promising.10 Most recently, the Institute for Healthcare Improvement (IHI) and the American Hospital Association challenged health care organizations to redesign patient safety systems to prevent avoidable deaths in its 100K Lives Campaign. One of the 6 proposed core interventions was the deployment of rapid response teams at the first sign of patient decline.11

Despite these reports of success, a recent large cluster‐randomized controlled trial did not yield the same positive results. In this well‐designed study of 23 Australian hospitals, the Medical Early Response, Intervention and Therapy (MERIT) study investigators found the incidence of cardiac arrest, unplanned ICU admissions, and unexpected death essentially unchanged despite large increases in how often the emergency team was called.12 One possible explanation why these findings conflicted with previous favorable results is that the ultimate impact of a MET may depend on the effectiveness of implementation strategies. To derive the benefits of a MET/RRT, hospitals must increasingly focus on identifying barriers to implementation and address practical issues that may undermine their long‐term effectiveness.

In this article we describe in detail the process of establishing an RRT at our urban, academic hospital and the modifications that became necessary as we rolled out the intervention and encountered obstacles. This analysis was undertaken as a quality improvement (QI) activity. To our knowledge, this is one of the few recent published descriptions of the experiences of implementing an RRT in the United States since earlier work in Pittsburgh.9, 13

METHODS

Temple University Hospital is a tertiary care academic hospital in urban Philadelphia, Pennsylvania. Our RRT was first implemented July 1, 2004, and in the first 12 months of initiation, it was activated 307 times. The RRT at Temple University Hospital was designed to be accessible 24 hours a day, 7 days a week. The daytime team (8 am‐5 pm) is composed of an attending physician (a hospitalist trained as a general internist), a senior internal medicine resident, a critical care nurse, a nurse manager, a pharmacist, and a respiratory therapist. In addition, both a transporter and a member of the admissions office respond to all rapid response team calls but do not get clinically involved in patient care. For nighttime (5 pm‐8 am) and weekend coverage the hospitalist is replaced by an on‐site pulmonary critical care physician, but the remainder of the team is unchanged. All RRT members carry beepers synchronized to provide the location of an RRT activation. In addition, all RRT calls are simultaneously announced on the overhead paging system. No changes were made to the existing cardiac arrest team (code team) at the hospital, which remained a 24‐hour response team for patients found to be in true cardiopulmonary arrest and was comprised of on‐call internal medicine house staff (but no hospitalist attending physician), a respiratory therapist, a pharmacist, a critical care nurse, a nurse manager, and, most notably, an anesthesiologist for emergent intubation and airway management.

The RRT was intended for use within the physical confines of Temple University Hospital and its immediately adjacent grounds. Within the hospital the main locations defined were: inpatient areas, including patient rooms and hallways of the medical‐surgical units of the inpatient tower, as well as the burn, coronary, medical, neurological, neurosurgical, and surgical intensive care units; off‐unit/procedural areas, including diagnostic/emnterventional radiology, the gastroenterology endoscopy suite, the pulmonary procedure suite and pulmonary function lab, the cardiac catheterization/ECHO/stress Lab, the inpatient dialysis unit, and the physical therapy gym, all areas where inpatients are routinely transported during their hospital admission for workup/treatment and where outpatients go for scheduled procedures and therapies; and outpatient/common areas, including all the general medical and subspecialty outpatient clinics in 2 separate outpatient towers (Outpatient Building and Parkinson Pavilion) with direct access from the main hospital building, the outpatient pharmacy, the elevators, the hallways in the outpatient sections of the hospital, all lobbies, and the immediately adjacent outside grounds.

Prior to the launch date of the RRT, clinical criteria were established to help guide staff about when an RRT might be called (Fig. 1). These were based in part on early literature on the clinical markers that most often precede clinical deterioration.14, 15 In addition, 2 much broader categories for RRT activation were added (Inability to reach the patient's primary team of treating physicians for any of the above and Any potentially serious medical errors or adverse events) in order to minimize the need for a very specific physiologic definition to be met in order to activate the team. Physicians, nurses, and other staff with significant daily contact with inpatients and outpatients were in‐serviced about the purpose of the RRT and how to activate the system via the hospital paging operator. Laminated cards with RRT criteria were distributed to all hospital personnel, and educational posters were displayed prominently throughout the hospital.

Figure 1

Each RRT event was to be assessed by team members using a standardized evaluation form (Fig. 2), with primary responsibility going to the physician team leader. In the initial phases of implementation, these forms were kept in the offices of the Section of Hospital Medicine for the use of hospitalist attending physician team leaders. Later on in the year they were kept in the pharmacist's RRT medication bag. These forms were collected at the completion of each RRT event or faxed to a central location and then entered into a database maintained by the hospital's Department of Patient Safety Operations. Weekly debriefing meetings to review all RRT events from the preceding week were attended by representatives from patient safety, respiratory, nursing, hospital medicine, and the pharmacy. Attempts were made to identify the issues that led to selected RRT activations, to obtain patient follow‐up from the clinical event, and to evaluate the performance of the team. Throughout these weekly meetings, QI strategies for improving the effectiveness of the RRT were identified and implemented.

Figure 2

The core outcome measures that were used to assess RRT performance were: appropriateness of the RRT activation, percentage of patients who were stabilized, percentage of patients who were transferred to a higher level of care, and overall team performance.

In the weekly meeting of the RRT evaluation committee, at which each RRT was reviewed by the clinical team, each scenario and details of the event were reviewed to determine whether the RRT activation was appropriate, whether the intervention was successful, and whether there were any issues with the team performance. After a thorough discussion of each case and review of additional data from the chart if necessary, the RRT evaluation committee reached a consensus about each of these measures.

We also tracked the number of code team activations from the year preceding establishment of the RRT (2003‐2004) through the year during which the RRT was established (2004‐2005). Because all calls for both the RRT and the code team go first to the hospital operator, we reviewed the hospital paging operators' logs for the entire 12‐month period to track the rate of code team events to RRT events on a monthly basis.

RESULTS

In a 12‐month period, the RRT was activated 307 times, as recorded in the hospital operator logs. In the year preceding inception of the RRT, there were 272 code team activations. In the first 12 months concurrent with RRT implementation, the code team was activated 258 times. Overall, at their discretion the team leaders converted 13% of the 307 RRT activations to traditional code team activations.

There were 11 RRT activations in July, the first month of implementation, and 14 activations in the second month. At that point, the internal hospital newsletter released a feature on the new RRT, and our patient safety officer/director of patient safety operations made a concerted effort to educate hospital administration and the Graduate Medical Education Committee (GMEC); as a result, utilization picked up. From September onward through the remainder of the academic year, an average of 28 RRT activations occurred each month (range 20‐37), whereas an average of 22 codes took place each month (range 12‐27). The numbers of RRT versus code team activations are plotted in Figure 3. A trend line for the number of code team activations per month in 2003, the year prior to implementation of the RRT, was added for comparison; it conveys the slight overall decrease in the number of codes as the RRT took effect (average of 23 codes per month, range 15‐31).

Figure 3

Physician evaluation forms were returned for 170 of the 307 RRT events (55%). The main inpatient tower was the site of 42% of these RRT activations, followed by the outpatient/common areas, where 19% of the activations occurred, and off‐unit/procedural areas, the site of 18%. Table 2 provides information on specific location, reason for call, and disposition of a sample of the RRT activations in the non‐inpatient areas. Time of day was noted in 76.8% of events. Of these, 82.9% occurred during the traditional day shift (7 am‐7 pm) and 17.1% on night shift (7 pm‐7 am). Most RRT activations occurred between 8 am and 4 pm. Daytime events heavily outnumbered nighttime events regardless of location.

Physician team leaders largely believed a specific underlying clinical diagnosis was responsible for 59% of the RRT activations, followed by adverse drug reactions (3.5%), physician error (1.8%), and nursing error (0.6%). When an underlying clinical diagnosis or organ system was suspected, it was most frequently pulmonary (32%), followed by neurological (14%) and cardiac (11%). It was believed that 32% of events were for other reason not listed. Table 1 provides the breakdown of other underlying diagnoses in RRT events.

In the judgment of evaluators, the system was utilized appropriately in 98% of the evaluated events. Eighty‐five percent of RRT activations were believed to have prevented further clinical deterioration, though it was also thought that 3% of patients deteriorated despite the efforts of the team. Disposition of the patient following an RRT event was noted 87% of the time, and it was believed that 88% of the patients were stabilized. Of the formally evaluated RRT events, team members were largely satisfied with the response and the functioning of the team, stating for 68% of the events that the team performed without a problem.

DISCUSSION

We have found the RRT to be an effective but challenging‐to‐implement QI intervention to increase patient safety at our academic institution. The Australian MERIT investigators recently suggested that despite growing evidence of the benefits of MET/RRT systems, long‐term success may depend most on effective implementation strategies.12 We experienced firsthand these challenges in the first year of our new RRT system.

Large system changes in a hospital are especially fraught with danger because of the unique aspects of health care delivery systems. As Reid commented in an editorial about the emerging use of the MET system in the United Kingdom, Despite potential advantages to patients, ensuring appropriate utilization was difficult because of cultural barriers. Traditional hierarchical behaviors that dictate how doctors and nurses react and work got in the way of people calling these life saving teams.17

Our weekly multidisciplinary RRT debriefings were the most crucial component of our implementation strategy. Many latent systems issues were uncovered, as well as more subtle problems such as lack of coordination of care, communication errors, gaps in patient handoffs or sign‐out. Previous studies by the Pittsburgh MERIT team have validated such retrospective categorization of errors uncovered by MET responses.10

However, neither that group nor the Australian MERIT study investigators specifically addressed the importance of the feedback process in RRT implementation. A strength of our system is that modifications to the RRT are made prospectively and in real time based on feedback from active RRT members during debriefing. In fact, the success of our RRT underscores the importance of open communication among hospitalists, house staff, nurses, pharmacists, and ancillary staff in multidisciplinary patient safety and QI endeavors. Everything from the responsibilities of team members to equipment evolved over the 12‐month period in order to improve the function and effectiveness of the team and was almost entirely based on feedback from the RRT doctors and nurses on the front lines. Suggestions from the evaluation forms were given serious consideration at every RRT evaluation committee debriefing. By optimizing the efficient operation of the RRT, we hope to continue to improve outcomes.

We believe a key to the success of our debriefing process was the constant attendance of our patient safety officer/chief medical officer and director of patient safety operations, who both encouraged active participation. Early on in the process, comments were made principally by physician and critical care nurse RRT members, and the dynamic was a bit one‐sided. However, we quickly saw a noticeable and sustained increase in participation by pharmacists and respiratory therapists, and by year's end, they had offered some of the most valuable practical suggestions, which resulted in a more efficient response. As the year went on and real changes were made quickly, all groups were much more vocal and willing to bounce ideas around the room, and the team dynamic and spirit of the group effort improved substantially.

Previous studies have focused on the impact of METs/RRTs on the rate of inpatient cardiac arrests. However, we found that nearly as many RRT events occurred off the inpatient units, for instance, when admitted patients were transported to other areas such as radiology, procedural suites, physical therapy, or dialysis and when scheduled outpatients arrived for their appointments. In addition, a large number of RRT calls came from outpatient departments and common areas of the hospital such as lobbies, hallways, and waiting rooms, mostly involving outpatients and visitors, but not infrequently hospital employees were involved as well. This unexpected and, to our knowledge, previously unreported finding is mirrored in the distribution of RRT activations throughout the course of the day. Most events occurred during the traditional day shift of 7 am‐7 pm, and were heavily clustered between 8 am and 4 pm. In most American hospitals, these are the hours during which outpatients and visitors make up a significant proportion of the hospital population and during which most elective procedures on inpatients occur. Prior to the introduction of our RRT, no specific system was in place for emergent triage, assessment, and expedited treatment of off‐unit patients, outpatients, and visitors. Most often, the code team was mobilized, sometimes taking them to remote locations and making them unavailable for true inpatient cardiopulmonary arrests. Our RRT seems to have the potential to fill a much‐needed gap in patient safety, offering off‐unit patients, outpatients, and visitors a safety net while in our hospital. No prior descriptions of RRT or MET implementation have touched on this area. It would be interesting to see if other hospitals with RRTs have had a similar experience in order to determine whether having an RRT dedicated specifically to the outpatient and common areas of the hospital might provide even more targeted efforts and efficient response times. Thus, the benefits of our RRT seemed to extend beyond a simple reduction in the number of in‐hospital cardiopulmonary arrests and into an unanticipated patient safety black hole.

Implementation of the RRT specifically in academic medical centers has been limited to date. In our opinion, the academic environment is an ideal area for RRTs (because the most critically ill patients often are cared for on teaching services by junior house officers), but it is also a challenging arena in which to make change (because of the complex hierarchy of teaching hospitals). We chose to have an attending physician lead our RRT efforts for the most part. However, residents always participated, and not infrequently led, as key team members. As a commentator on the Australian RRT system pointed out, it is important that junior medical staff [feel empowered] to call for immediate assistance when they are concerned about their patient, but may not have the experience, knowledge, confidence or skills necessary to manage them appropriately.18 We believe that the RRT serves as a valuable educational forum for resident education. Academic centers that develop RRTs must work to integrate the teams into an educational context while simultaneously providing patients with the most experienced and knowledgeable clinical team to address their needs at a time when appropriate clinical decision making is critical. Therefore, the residents who participate in our RRT are formally evaluated by the hospitalists using a standard program evaluation form that encompasses the Accreditation Council for Graduate Medical Education (ACGME) core competencies.19

Through the first year of our RRT system and beyond, activation of the code team and RRT shifted as more RRT activations were recorded and fewer codes were called. Concerted educational efforts and reinforcement of the criteria for calling the RRT had a definite sustained impact of helping staff to become comfortable with using the system. At our institution, it has been difficult to definitively conclude whether RRT calls prevented codes or merely substituted for them at times, especially because 13% of all RRT activations were subsequently converted to code team calls. The Australian MERIT study investigators, despite an excellent study design of a large multicenter trial, also were unable to demonstrate a true decrease in the cardiac arrest rate.12 Much more significant to us, especially in the first year of implementation, was learning that the vast majority of physician RRT leaders perceived activation of the team to occur appropriately and to play a role in preventing clinical deterioration of patients. None of the other RRT or MET implementation studies that we reviewed commented specifically on these areas. It will be interesting to continue to follow these trends, as we expect the use of RRTs to become even more defined. Over time, we will no doubt be better able to determine whether RRTs have a true, sustained impact on preventing patient deterioration and inpatient cardiopulmonary arrests while maintaining a high rate of physician satisfaction that the team is being activated for legitimate reasons.

Our descriptive study had some limitations. The number of RRT evaluations received, while adequate for preliminary analysis, may not accurately represent the 307 activations of the system that occurred in the first 12 months. We suspect that this underreporting, especially in the first half of the year, was in large part a result of relying on team leaders to voluntarily return data forms at the conclusion of each RRT event. RRT evaluations in the second half of the year were more actively distributed at the point of care to the team leader directly by the pharmacist and were more diligently followed up on. Forms are now readily available in the team pharmacist's fanny pack, which was done because of quality improvement feedback from physicians at a debriefing meeting. Since those interventions, there has been a dramatic improvement in the capture of event data and the timely submission of forms. We expect and have demanded close to a 100% return of the forms in the second year of our RRT system, which will vastly improve our analysis. We were also surprised that despite the comprehensiveness of our RRT activation criteria, 32% of physicians were unable to find a match with a clinical indication on the list, indicating unanticipated reasons for calling an RRT. We will continually strive to improve the specificity of future data for planning purposes and training initiatives. However, in some way this confirms our belief that RRTs occur for such a wide variety of reasons that they cannot always be limited to the major clinical categories. On a similar note, we regret not adding a specific category under Outcomes on the evaluation form to include the possibility that RRT members might have offered palliative care or changes in code/do not resuscitate (DNR) status to patients or families. Given that our hospital has both a code team and an RRT begs the question of whether mortality rates might be affected if patients who prior to the RRT might have had a full resuscitation effort were made DNR. In the future, this would be an interesting issue to consider in analysis. Carefully categorizing RRT events is critical to continued success. Further work involving formal team skills training for RRT members, including use of the medical school's clinical simulators for mock RRT scenarios, is planned. These sessions are planned to review performance and clinical decision making for the most common scenarios that we have found to be involved in RRT activations. The 307 activations of the RRT in our first year have clearly set us on the path toward defining predictive rules and directed skills training for earlier identification of patient problems. Further outcome analyses of these efforts will be crucial.

CONCLUSIONS

An RRT was successfully introduced into an academic medical center. The team was heavily utilized in the first 12 months after the program was initiated, especially for off‐unit inpatients and those in outpatient/common areas, perhaps filling a gap in hospital patient safety. The keys to the early success of implementation of our RRT were multidisciplinary input and improvements made in real time. The long‐term effects of the RRT on the culture of patient safety in our institution and throughout the United States remain to be seen but are promising.

Despite decades of intensive research and improvements in medical care, severe sepsis affects an estimated 751,000 patients in the United States every year, killing 215,000 of them at an annual cost of 16.7 billion dollars.1 Because the elderly experience a 100‐fold increase in incidence, as compared with children, and a nearly 4‐fold increase in mortality (38.4% of those more than 85 years old), this burden is expected to increase with the aging population.1 Patients with severe sepsis have prolonged ICU14 and hospital stays and incur substantially increased costs compared with other patients.36

New research continues to explore the complex pathophysiology of sepsis,7 and clinicians, who once relied primarily on clinical experience and expert opinion to guide therapy, now have an increasing array of evidenced‐based sepsis therapies to employ. Recent meta‐analyses have evaluated several major treatments for severe sepsis,810 and recommendations (the Surviving Sepsis Campaign guidelines) for the treatment of severe sepsis were recently endorsed by 11 international critical care and infectious disease organizations.11 This article summarizes the current definitions of sepsis syndromes, the trials supporting the specific therapies for sepsis that are currently recommended, ongoing controversies and research, and implications for hospitalists, with a focus on early, effective antibiotics, activated protein C, early goal‐directed therapy, stress dose steroids, and intensive insulin therapy. For space considerations, readers are directed elsewhere for data supporting prophylaxis for deep venous thrombosis (DVT)12 and stress ulcer bleeding13 and for therapies less often directed by hospitalists, such as lung protective ventilation.14

DEFINITIONS

Systemic inflammatory response syndrome (SIRS), sepsis, severe sepsis, and septic shock were defined in 1992 to standardize the terminology of sepsis.15 These definitions have recently been reviewed and supported by a variety of American and European intensive care societies.16

SIRS is defined by the presence of at least 2 of the following:

• Temperature > 38C or < 36C;

• Heart rate > 90 beats/min;

• Respiratory rate > 20 breaths/min or PaCO2 < 32 mm Hg;

• WBC >12,000 or < 4000 cells/mm³, or >10% immature (band) forms.

Sepsis is SIRS due to documented or strongly suspected infection.

Severe sepsis is sepsis with organ dysfunction (such as lactic acidosis, oliguria, thrombocytopenia, or delirium), hypoperfusion, or hypotension (< 90 mm Hg systolic or more than 40 mm Hg below baseline).

Septic shock is severe sepsis complicated by hypotension or pressor dependence despite adequate (20‐30 mL/kg; 1.5‐3 liters in most patients) fluid resuscitation.

Sepsis terminology must be applied carefully. Many hospitalized patients meet criteria for SIRS, yet it is inaccurate to say a patient who has acute leukemia with leukocytosis, anemia‐induced tachycardia, and thrombocytopenia has severe sepsis if those abnormalities are not a result of inflammation or infection. Accurate documentation of sepsis syndromes can improve professional and institutional reimbursement and provide prognostic information: the in‐hospital mortality rates for severe sepsis and septic shock are approximately 30% and 50%, respectively.17 More importantly, thoughtful application of these definitions can help a hospitalist identify septic patients who qualify for one of the proven therapies for severe sepsis.

EARLY, EFFECTIVE ANTIBIOTICS

For obvious ethical reasons, randomized, controlled trials to study the impact of inappropriate or delayed antibiotic therapy for serious infections are not possible. However, the evidence supporting early, effective antibiotic therapy is still compelling, and because many hospitalists often initiate treatment with antibiotics before transferring a patient to intensive care, this may represent the most important intervention hospitalists can provide to patients with serious infections. Several studies have estimated the impact of early, effective antibiotics on outcomes.

Houck et al. retrospectively reviewed 13,771 cases of community‐acquired pneumonia among elderly Medicare patients. They found that 39.1% of the patients waited more than 4 hours for antibiotics and 7.6% waited more than 12 hours; three quarters of these delays resulted from delayed ordering of antibiotics.18 Further, 21.2% received an antibiotic selection incompatible with recent professional guidelines. Receiving antibiotics within 4 hours reduced in‐hospital and 30‐day mortality by 15% and length of stay by 0.4 days.18 Similar conclusions were reported by 3 of 4 previous analyses.1922 Extending these findings to critically ill patients, Iregui et al. found that delayed treatment with appropriate antibiotics (odds ratio, 7.68) was a greater predictor of mortality for 107 patients with ventilator‐acquired pneumonia than were APACHE II scores and malignancy; 31% failed to receive appropriate antibiotics within 24 hours, and again, three quarters of these delays resulted from delays in writing antibiotic orders.23

Not surprisingly, antibiotic therapy must be effective as well as timely. MacArthur et al. studied the impact of adequate (ie, active against cultured organisms, if isolated) antibiotics on the outcomes of 2634 septic patients enrolled in a randomized trial of an anti‐TNF antibody. Nearly 91% received appropriate antibiotics; their mortality rate was 33%, 10% lower than that of the patients whose initial antibiotics were inadequate (P < .001).24 Leibovici et al. reported similar findings in a prospective study of patients with bacteremia. Only 63% of 3413 subjects received an antibiotic active against the infecting pathogen, and their mortality was 20%, 14% lower than that in the group that received ineffective antibiotics (P = .0001).25 Other authors have reported even worse outcomes with ineffective therapy: 62% mortality among inadequately treated bacteremic or fungemic ICU patients, compared with 28.4% among those who were adequately treated26 and an odds ratio of dying of 8.14 for the 46 of 270 septic ICU patients who received inadequate initial antibiotics,27 making inadequate antibiotic therapy the strongest risk factor for death. Finally, Kollef et al. reported that 26% of 655 infected ICU patients received inadequate antibiotics and suffered an infection‐related mortality rate of 40.2%, more than twice the 17.7% rate among adequately treated patients (P < .001). Inadequate antimicrobial therapy was a greater risk factor for death than early respiratory failure or sepsis‐related organ failure assessment scores.28

Guidelines for anti‐infective care now recommend obtaining appropriate cultures and administering broad‐spectrum antibiotics (appropriate for suspected infections, local susceptibility patterns, and any relevant prior culture data from individual patients) within 1 hour of presentation.11 In addition, any removable focus of infection must be identified and managed (eg, an abscess, infected catheter, tampon, or infection requiring surgery).

ACTIVATED PROTEIN C

Recombinant human activated protein C (APC) is a protein with anticoagulant and anti‐inflammatory properties that is relatively deficient in approximately 87% of septic patients.29 Although numerous trials of other anticoagulants (antithrombin III and tissue factor pathway inhibitor) and immunosuppressives (tumor necrosis factor inhibitors, high‐dose steroids, interleukin‐1 receptor antagonists, and others) have failed to show any benefit,7 in 2001 APC became the first proven therapy specifically for sepsis. The PROWESS trial, which established its efficacy, randomized 1690 patients who met 3 SIRS criteria and dysfunction of at least 1 organ system to APC (24 g/kg IV per hour for 96 hours interrupted for bleeding or urgent procedures) or placebo. APC reduced 28‐day mortality from 30.8% to 24.7%, yielding an absolute risk reduction of 6.1% and a corresponding number needed to treat (NNT) of 16.4. This benefit was seen across all subgroups including those with normal baseline APC levels.29

Not surprisingly, APC increases the risk of serious bleeding. Although this effect was of borderline significance in PROWESS (3.5% vs. 2% in the placebo group, P = .06),29 it was confirmed in subsequent trials (3.9% vs. 2.2%, P = .01)30 and may be larger still in open‐label use, at 6.5%.17, 31 Intracerebral hemorrhage (ICH), a particularly devastating complication, occurred in 0.2% of the PROWESS patients and 0.5% of patients in 2 subsequent studies30, 32; in both major trials, there was a single extra event in the APC arm.29, 30 Like serious bleeding in general, ICH was more common in open‐label use, occurring in 1.5% of patients.31, 33 Therefore, it is vital to have strict adherence to exclusion criteria and familiarity with the risk factors for serious bleeding. In the PROWESS trial, after randomization, risk factors for serious bleeding included procedures and injury to vascular organs, an activated partial‐thromboplastin time of more than 120 seconds, an international normalized ratio greater than 3, gastrointestinal ulceration, and development of severe thrombocytopenia (< 30,000/mm³)29; in a 2002 study of 2786 APC recipients, ICH was largely confined to patients with meningitis or a platelet count less than 30,000/mm³.32

APC therapy has several other limitations and drawbacks. Multiple contraindications, including predisposition to bleeding, a recent history of bleeding, anticoagulant use, immunosuppression, liver disease, dialysis dependence, and hypercoagulable states, restrict its use. APC appears to work best when administered early, within 24 hours of the onset of organ dysfunction.31 In addition, APC is indicated only in adults with Acute Physiology and Chronic Health Evaluation (APACHE II) scores greater than 24 and multiorgan failure. Post hoc analysis of the PROWESS data showed that although the relative risk (RR) of death for those with APACHE II scores of 25 or more was .71 and statistically significant, the RR for those with scores below 25 was a nonsignificant .99.34 A subsequent study, ADDRESS, confirmed there was no benefit to septic patients with a low risk of death.30 In the ADDRESS study 2613 patients with severe sepsis and either an APACHE II score less than 25 or single organ failure were randomized to APC or placebo. No differences were found in 28‐day and in‐hospital mortality; among patients who had undergone surgery in the previous 30 days, those receiving APC had a significantly increased risk of death (20.7% vs. 14.1%, P = .03).

An additional drawback of APC therapy is its cost, approximately $6800 per infusion, although the cost per year of life gained, $24,484, or $52,360 per life saved (NNT $6800), is reasonable for those with APACHE II scores greater than 24.34 Concerns have also been raised about the PROWESS trial itself: the production of the study drug and some exclusion criteria were changed midtrial, after which the effectiveness of APC improved. APACHE II scores had not been validated for selection of patients for therapies and may have varied with time or by observer. The original PROWESS study population may have been skewed away from chronically ill patients.35 Experts differ on the significance of these concerns and even whether APC therapy should be considered the standard of care pending further research.32, 35 The ADDRESS trial also failed to demonstrate a benefit in a subgroup of patients with APACHE II scores above 24, although it was underpowered to do so, and according to enrollment criteria, none of those patients had multiorgan failure.30 However, in the subgroup of PROWESS patients with APACHE II scores greater than 24, the absolute reduction in mortality was a full 13%,17 with a corresponding NNT of 7.7, and although the PROWESS findings have not been duplicated in a second randomized trial, a single‐arm, open‐label study of APC (ENHANCE) showed a nearly identical mortality rate.31 Pending confirmatory trials, APC remains a recommended therapy for selected patients sick enough to benefit and without excessive bleeding risk.11

EARLY GOAL‐DIRECTED THERAPY

Because physician‐directed resuscitation for sepsis may normalize vital signs, central venous pressures (CVP), and urine output without correcting hypoperfusion, Rivers et al. tested a resuscitation protocol that incorporated a central line that continuously monitored mixed‐venous oxygen saturation as a surrogate for cardiac output.36 They randomized 263 patients with septic shock (defined as hypotension < 90 mm Hg after a 20‐30 mL/kg bolus, or lactate > 4 mmol/L, which is associated with at least a 3‐fold increase in the mortality of emergency department patients with suspected infection37) to either standard care or early goal‐directed therapy (EGDT) for the initial 6 hours of hospital care. Patients with acute coronary ischemia, pulmonary edema, stroke, asthma, overdose, trauma, dysrhythmia, immunosuppression, uncontrolled cancer, or a need for urgent procedures were excluded. Standard care was directed by physiologic parameters such as vital signs, urine output, and CVP. EGDT used sequential therapies designed to support organ perfusion: 500 mL of normal saline was given every half hour until the CVP was at least 8‐12 mm Hg. Pressors were given until the mean arterial pressure was 65‐90 mm Hg (norepinephrine36 or dopamine were preferred agents, and vasopressin [0.01‐0.04 units/min] was an option for shock refractory to first‐line pressors)11, 38 Transfusion (to a hematocrit goal of 30) and dobutamine were given until mixed‐venous oxygenation saturation was 70% or better (Fig. 1). Lastly, patients who did not achieve this goal were sedated and mechanically ventilated.

Figure 1

Results were dramatic: mortality was reduced from 46.5% to 30.5%, with an ARR of 16% and an NNT of 6.25. Study patients received similar amounts of crystalloid, but received it earlier than the standard care patients and received more transfusions and inotropes. Substantially more patients in the EGDT group than the standard care group achieved a mixed venous oxygen saturation of 70%; 13.7% of the EGDT patients had occult hypoperfusion (low mixed‐venous oxygenation that responded to inotropes despite satisfactory vital signs). EGDT improved length of stay (4 days shorter among survivors) and duration of intubation, as well as APACHE scores and several physiologic parameters.36

Critics of this trial note the impossibility of adequate blinding and the high mortality in the placebo group. Further, because the trial tested the EGDT protocol as a whole, there was no way to know if each step was optimal. For example, a different CVP goal could have been used or adjustments made for mechanical ventilation, which can falsely elevate a low CVP into the desired range (the Surviving Sepsis Campaign guidelines recommend a CVP goal of 12‐15 mm Hg in mechanically ventilated patients11). Also, the selection of pressor, the use of inotropes, and the transfusion threshold were chosen on the basis of physiologic rationales, but all of these are arguable.39 This was also a single trial, and earlier goal‐directed protocols for ICU patients actually showed harm,40, 41 although those trials targeted supranormal physiologic goals in more established critical illness.42 Finally, on a practical level, hospitals and particularly emergency departments must commit resources to train physicians and staff, purchase the appropriate central venous catheters, and convince eligible patients to undergo an invasive procedure. In a survey of 30 attending physicians in academic referral hospitals, only 7% reported standard use of EGDT. Barriers included the requirement for specialty monitoring equipment and other resources, and central venous cannulation.43

Despite these concerns, the striking reduction in mortality associated with EGDT led to its endorsement by the Surviving Sepsis Campaign guidelines and underscores the principle of aggressive early resuscitation for patients who do not meet eligibility criteria but appear at risk for worsening sepsis. As yet, however, no strong evidence mandates a specific approach to the septic patient without shock.

STRESS DOSE STEROIDS

Because of the importance of the inflammatory cascade in severe sepsis, a potential role for steroids in the management of sepsis has been repeatedly studied. More than 50 studies have been performed since the 1950s, generally with pharmacologic doses of steroid; a meta‐analysis showed that such a practice was ineffective.44, 45 However, data accumulated that relative adrenal insufficiency during severe sepsis was common and associated with an increased risk of death and that physiologic doses of steroids could reverse refractory hypotension.46 To define the role of a physiologic course of steroids in septic shock, Annane et al. randomized 299 critically ill adults to either 7 days of stress dose hydrocortisone (50 mg IV every 6 hours) and fludrocortisone (50 g NG every 24 hours) or matched placebos. Enrolled patients were severely ill; the placebo group had a 63% mortality, and patients had to have septic shock, oliguria or hypoxia, hypotension despite low‐dose dopamine, and lactate greater than 2 mmol/L and require mechanical ventilation. Pregnant women, those with myocardial infarction or pulmonary embolus, advanced malignancies, or immunodeficiency, and those with clear indications or contraindications to steroids were excluded.47 Enrollment criteria were modified midstudy; changes included the exclusion of patients who had recently received etomidate, which inhibits 11‐‐hydroxylase and has been identified as a risk factor for adrenal insufficiency in intensive care patients.48 All patients received a 250‐g cosyntropin stimulation test. The authors considered patients nonresponders to consyntropin if serum cortisol failed to increase to 9 g/dL or more.

Steroids reduced the duration that a vasopressor was required and reduced mortality from 63% to 53% among nonresponders, giving an NNT of only 10 to prevent 1 death at 28 days. Although the authors described no evidence of adverse effects, among the subset of 70 patients who responded appropriately to cosyntropin, there was a nonsignificant trend toward increased mortality, and rates of hyperglycemia were not provided.47 The authors concluded that physicians should test appropriate patients for adrenal reserve, give the studied steroid regimen while results are pending, and discontinue treatment if a patient retains adrenal reserve.

The literature on steroids and critical illness is complex, with more than 1300 articles on steroids and sepsis published since 1988, and several concerns were raised about the Annane study. For example, did much of the benefit for those patients enrolled before the protocol amendment come from reducing an adverse effect of etomidate?49 Does the high‐dose, 250‐g cosyntropin stimulation test overcome (and conceal) partial ACTH resistance that might benefit from treatment?50 Might not a robust baseline cortisol suggest sufficient adrenal function regardless of the incremental response to cosyntropin?51 Partial answers were provided by 2 subsequent meta‐analyses. Both found that more recent studies gave lower doses of steroids in longer, 5‐ to 7‐day courses to sicker patients and demonstrated improvement in mortality and shock reversal, with relative risk reductions of 14%‐22%; the NNT ranged from 8 to 11. One analysis found no difference in outcomes between adrenally sufficient and adrenally insufficient patients, and those authors advised considering treatment for all patients regardless of their adrenal function test results.8 The other analysis concluded that the data on steroids for those with adrenal reserve was too limited to recommend treating adrenally sufficient patients.9

Disputes about certain details, such as whether patients should be treated without regard to adrenal reserve, continue in the literature.45, 52 An ongoing randomized, controlled trial, CORTICUS, is expected to provide additional guidance on the use of low‐dose steroids in sepsis; in the meantime, the literature clearly supports a longer course of low‐dose steroid therapy for patients with pressor‐dependent septic shock with inadequate adrenal reserve by cosyntropin testing, and guidelines allow discretion about whether patients with adequate adrenal reserve should also be treated.11 Hospitalists may also want to treat septic shock with equivalent doses of dexamethasone (approximately 2 mg IV every 6 hours) if adrenal evaluation may be delayed, as this agent will not confound cosyntropin stimulation test results, and they may want to avoid etomidate in septic patients53, 54 for whom they perform or supervise intubations.

INTENSIVE INSULIN THERAPY

Mounting evidence supports the short‐term role of hyperglycemia in morbidity and mortality, especially in critical illness. Hyperglycemia impairs neutrophil and endothelial cell function as well as protective responses to cardiac and neuronal ischemia,55 whereas insulin has anti‐inflammatory and antiapoptotic effects,7, 56 suggesting that intensive insulin might improve the outcomes of critically ill patients. To test this theory, van den Berghe and colleagues randomized 1548 mostly surgical ICU patients to insulin infusions titrated for glucose goals of either 80‐110 or 180‐200 mg/dL, followed by subcutaneous insulin after ICU discharge. Although blinding was impossible, in both cases glucose management was performed by a separate research team. Multiple benefits were noted: ICU and total in‐hospital deaths were reduced, mostly among patients with an ICU stay of more than 5 days, whose risk of death fell from 20.2% to 10.6%. Intensive insulin also reduced septicemia, renal impairment, critical illness polyneuropathy, and duration of intensive care.57

Subsequently, a meta‐analysis of 35 trials suggested that insulin reduced the mortality of critically ill patients by 15%.10 Van den Berghe et al.'s results were also duplicated in a broad, medical‐surgical ICU population, although the reductions in morbidity and mortality were measured against historical controls.58 However, whether the results of the influential surgical ICU study could be applied to medical patients was not known until 2006, when the van den Berghe group reported the effects of similar insulin protocols on 1200 patients in the medical ICU who were expected to need intensive care for at least 3 days.59 In this study, intensive insulin failed to reduce overall mortality (40% and 37.3%, P = .33). However, intensive insulin did reduce mortality among the 64% of patients who stayed in the ICU 3 or more days from 52.5% to 43% (NNT 10.5, P = .009). This benefit was offset by an increased number of deaths in the intensive insulin group among patients with ICU stays of less than 3 days (P = .05‐.35 depending on the method used).59 Intensive insulin did reduce newly acquired kidney injury, duration of mechanical ventilation, and lengths of ICU and hospital stays, and the reduction in morbidity increased with the duration of intensive insulin therapy. Hypoglycemia (mean 32 mg/dL) occurred in 25% of patients with prolonged stays6.4 times as often as in the usual care group.60 Liver and renal failure were associated with hypoglycemia.59

Critics of the surgical ICU trial noted the high mortality among the usual care patients (5.1%), a robust 34% mortality reduction for a relatively small 50 mg/dL reduction in morning glucose levels, and the aggressive use of parenteral nutrition, raising the question of whether intensive insulin merely attenuated the side effects of intravenous glucose.61, 62 Also, the ideal blood glucose target is not known with certainty. Retrospective studies suggested the upper limit for target blood glucoses could be 145 mg/dL63 and found differing thresholds at which hyperglycemia increased mortality in nondiabetics (144 mg/dL) and diabetics (200 mg/dL).64 However, in the surgical ICU trial, there was no threshold below which there was no further reduction in risk; patients whose mean blood glucose was below 110 mg/dL had lower mortality than those whose levels were between 110 and 150 mg/dL (P = .026).65 Finally, the effects of hyperglycemia and intensive insulin may vary by population: retrospective studies found that ICU hyperglycemia was more strongly associated with mortality among nondiabetics,64, 66 and van den Berghe et al. noted no benefit from intensive insulin in a small subgroup of diabetics.59

In summary, large, well‐designed trials have demonstrated that intensive insulin reduced mortality in critically ill patients after a delay of 3‐5 days, but this benefit did not extend to all patients in the medical ICU.57, 59 Some authors have suggested deferring intensive insulin for 3 days,67 but because early therapy probably contributes to the delayed mortality benefit, this approach may deprive patients of the observed benefits. Ongoing clinical trials (NICE‐SUGAR) are likely to provide useful information about how hyperglycemia should be managed in different populations, including septic ICU patients.61 In the meantime, institutions can select the intensity of their insulin therapy by weighing morbidity and long‐term mortality benefits against possible short‐term harms and ensuring that hospital staff members are sufficiently trained to control hyperglycemia safely. For example, in critical illness, intravenous insulin is preferable to subcutaneous insulin, and the frequent measurement of whole‐blood glucose instead of finger‐stick glucose helps to avoid errors.55, 68 And although researchers were unable to prospectively identify patients with long ICU stays,59 severely septic patients have long ICU stays (generally 7.5‐16.6 days),14 and individual ICUs might observe enough stays of more than 2 days in their patient population to justify intensive insulin for this subgroup. And finally, although no conclusive evidence mandates a specific approach to hyperglycemia outside the ICU, the ICU data provide a physiologic rationale for cautious but tight control of glucose in more moderately ill patients. Guidelines for the management of inpatient hyperglycemia were published previously.55

SEPSIS AND THE HOSPITALIST

Hospitalists who provide critical care may make frequent decisions about the inclusion and exclusion criteria for the major trials of sepsis, weigh their relative benefits against risks and costs, contemplate gray areas such as adrenal testing in shock, and employ evidence‐based therapies for severe sepsis. However, hospitalists may also see patients who qualify for these therapies when they are called to see septic patients in the emergency department, when severe sepsis develops in patients on the medicine ward, or when they provide consultation services in an ICU. Sepsis care must be implemented urgently; patients in the pivotal trial of steroids had to be randomized within 3 hours of shock onset,47 data suggest that the window for optimal antibiotic therapy may be no greater than 4 hours from diagnosis,18 whereas guidelines suggest therapy within 1 hour,11 and early goal‐directed therapy was studied only for the first 6 or more hours of hospitalization.36 Thus, hospitalists who do not provide ICU care should be able to identify patients with severe sepsis and either deliver initial care or recognize the need for immediate consultation. Specifically, hospitalists can:

• Recognize that both absolute (< 90 mm Hg) and relative hypotension (> 40 mm Hg below baseline) indicate septic shock;

• Identify normotensive candidates for EGDT (severe sepsis with serum lactate > 4 mmol/L) by requesting a serum lactate in addition to prompt appropriate cultures for severe acute infection69;

• Recognize atypical presentations of sepsis (tachypnea, tachycardia, confusion, etc.) and maintain a high suspicion for sepsis in patients who may be predisposed to infection and to atypical presentation because of age, immunosuppression, neutropenia, diabetes, or other conditions;

• Initiate effective antibiotics and EGDT promptly for individual patients or by coordinating efforts to improve sepsis care at an institutional level, for example, as a component of medical emergency team services70, 71;

• Rapidly identify and manage removable foci of infection such as abscesses, empyemas, necrotizing fasciitis, or infected vascular catheters; and

• Competently educate hospital staff, residents, and medical students about sepsis care.

Hospitalists are busy physicians, and the task of reviewing sepsis literature and implementing recommendations is daunting. However, hospitalists can turn to resources such as the Surviving Sepsis Campaign Guidelines, a series of recommendations for managing severe sepsis that were endorsed by 11 international critical care and infectious disease societies and published in Critical Care Medicine in 2004.11 The Institute for Healthcare Improvement has also published a series of online severe sepsis bundles, or groups of proven interventions, complete with implementation tips and supporting literature, available at http://www.survivingsepsis.org or http://www.ihi.org/IHI/Topics/CriticalCare/Sepsis/,72 and individual institutions have described their implementation of evidence‐based, protocol‐directed sepsis care (http://www.mustprotocol.com).69

CONCLUSION: DEADLY YET TREATABLE

The death toll from severe sepsis in the United States exceeds that of lung, breast, and colon cancer combined and equals that of myocardial infarction (MI),1 a condition that appropriately triggers a series of emergency interventions. Physicians now have an arsenal of therapies for severe sepsis analogous to those employed for MI, and a comparison between the 2 conditions underscores the high mortality rate of severe sepsis and the enormous impact on patient outcomes provided by evidence‐based sepsis therapy. Figure 2 compares the 9.5%‐16% ARR for death associated with APC in patients with APACHE 2 scores greater than 24 and multiorgan failure,29 EGDT,36 stress dose steroids in shock complicated by adrenal insufficiency,47 and intensive insulin in patients with medical ICU stays longer than 3 days,59 with the benefits of thrombolysis for ST‐elevation MI (2%‐3%)73 or antiplatelet therapy for acute MI (2.3%).74 Figure 3 compares the corresponding NNT values to save 1 life; according to the available data, a hospitalist is 5‐8 times more likely to save a life with EGDT than with fibrinolysis.

Figure 2

Figure 3

Because the literature supporting several major sepsis therapies have been limited to retrospective studies1828 and single randomized, controlled trials29, 36 and because key trials are still underway (CORTICUS, NICE‐SUGAR), the benefits of sepsis therapies are less certain than are those for the treatment of MI. This was underscored by the finding that the benefit in reduced mortality of intensive insulin in the surgical ICU57 did not extend to all patients in the medical ICU.59 However, the potentially marked survival benefit of early effective antibiotics, APC, EGDT, stress dose steroids, and intensive insulin and the urgency with which they must be applied demand that all hospitalists become or remain familiar with the evolving sepsis literature.

Despite decades of intensive research and improvements in medical care, severe sepsis affects an estimated 751,000 patients in the United States every year, killing 215,000 of them at an annual cost of 16.7 billion dollars.1 Because the elderly experience a 100‐fold increase in incidence, as compared with children, and a nearly 4‐fold increase in mortality (38.4% of those more than 85 years old), this burden is expected to increase with the aging population.1 Patients with severe sepsis have prolonged ICU14 and hospital stays and incur substantially increased costs compared with other patients.36

New research continues to explore the complex pathophysiology of sepsis,7 and clinicians, who once relied primarily on clinical experience and expert opinion to guide therapy, now have an increasing array of evidenced‐based sepsis therapies to employ. Recent meta‐analyses have evaluated several major treatments for severe sepsis,810 and recommendations (the Surviving Sepsis Campaign guidelines) for the treatment of severe sepsis were recently endorsed by 11 international critical care and infectious disease organizations.11 This article summarizes the current definitions of sepsis syndromes, the trials supporting the specific therapies for sepsis that are currently recommended, ongoing controversies and research, and implications for hospitalists, with a focus on early, effective antibiotics, activated protein C, early goal‐directed therapy, stress dose steroids, and intensive insulin therapy. For space considerations, readers are directed elsewhere for data supporting prophylaxis for deep venous thrombosis (DVT)12 and stress ulcer bleeding13 and for therapies less often directed by hospitalists, such as lung protective ventilation.14

DEFINITIONS

Systemic inflammatory response syndrome (SIRS), sepsis, severe sepsis, and septic shock were defined in 1992 to standardize the terminology of sepsis.15 These definitions have recently been reviewed and supported by a variety of American and European intensive care societies.16

SIRS is defined by the presence of at least 2 of the following:

• Temperature > 38C or < 36C;

• Heart rate > 90 beats/min;

• Respiratory rate > 20 breaths/min or PaCO2 < 32 mm Hg;

• WBC >12,000 or < 4000 cells/mm³, or >10% immature (band) forms.

Sepsis is SIRS due to documented or strongly suspected infection.

Severe sepsis is sepsis with organ dysfunction (such as lactic acidosis, oliguria, thrombocytopenia, or delirium), hypoperfusion, or hypotension (< 90 mm Hg systolic or more than 40 mm Hg below baseline).

Septic shock is severe sepsis complicated by hypotension or pressor dependence despite adequate (20‐30 mL/kg; 1.5‐3 liters in most patients) fluid resuscitation.

Sepsis terminology must be applied carefully. Many hospitalized patients meet criteria for SIRS, yet it is inaccurate to say a patient who has acute leukemia with leukocytosis, anemia‐induced tachycardia, and thrombocytopenia has severe sepsis if those abnormalities are not a result of inflammation or infection. Accurate documentation of sepsis syndromes can improve professional and institutional reimbursement and provide prognostic information: the in‐hospital mortality rates for severe sepsis and septic shock are approximately 30% and 50%, respectively.17 More importantly, thoughtful application of these definitions can help a hospitalist identify septic patients who qualify for one of the proven therapies for severe sepsis.

EARLY, EFFECTIVE ANTIBIOTICS

For obvious ethical reasons, randomized, controlled trials to study the impact of inappropriate or delayed antibiotic therapy for serious infections are not possible. However, the evidence supporting early, effective antibiotic therapy is still compelling, and because many hospitalists often initiate treatment with antibiotics before transferring a patient to intensive care, this may represent the most important intervention hospitalists can provide to patients with serious infections. Several studies have estimated the impact of early, effective antibiotics on outcomes.

Houck et al. retrospectively reviewed 13,771 cases of community‐acquired pneumonia among elderly Medicare patients. They found that 39.1% of the patients waited more than 4 hours for antibiotics and 7.6% waited more than 12 hours; three quarters of these delays resulted from delayed ordering of antibiotics.18 Further, 21.2% received an antibiotic selection incompatible with recent professional guidelines. Receiving antibiotics within 4 hours reduced in‐hospital and 30‐day mortality by 15% and length of stay by 0.4 days.18 Similar conclusions were reported by 3 of 4 previous analyses.1922 Extending these findings to critically ill patients, Iregui et al. found that delayed treatment with appropriate antibiotics (odds ratio, 7.68) was a greater predictor of mortality for 107 patients with ventilator‐acquired pneumonia than were APACHE II scores and malignancy; 31% failed to receive appropriate antibiotics within 24 hours, and again, three quarters of these delays resulted from delays in writing antibiotic orders.23

Not surprisingly, antibiotic therapy must be effective as well as timely. MacArthur et al. studied the impact of adequate (ie, active against cultured organisms, if isolated) antibiotics on the outcomes of 2634 septic patients enrolled in a randomized trial of an anti‐TNF antibody. Nearly 91% received appropriate antibiotics; their mortality rate was 33%, 10% lower than that of the patients whose initial antibiotics were inadequate (P < .001).24 Leibovici et al. reported similar findings in a prospective study of patients with bacteremia. Only 63% of 3413 subjects received an antibiotic active against the infecting pathogen, and their mortality was 20%, 14% lower than that in the group that received ineffective antibiotics (P = .0001).25 Other authors have reported even worse outcomes with ineffective therapy: 62% mortality among inadequately treated bacteremic or fungemic ICU patients, compared with 28.4% among those who were adequately treated26 and an odds ratio of dying of 8.14 for the 46 of 270 septic ICU patients who received inadequate initial antibiotics,27 making inadequate antibiotic therapy the strongest risk factor for death. Finally, Kollef et al. reported that 26% of 655 infected ICU patients received inadequate antibiotics and suffered an infection‐related mortality rate of 40.2%, more than twice the 17.7% rate among adequately treated patients (P < .001). Inadequate antimicrobial therapy was a greater risk factor for death than early respiratory failure or sepsis‐related organ failure assessment scores.28

Guidelines for anti‐infective care now recommend obtaining appropriate cultures and administering broad‐spectrum antibiotics (appropriate for suspected infections, local susceptibility patterns, and any relevant prior culture data from individual patients) within 1 hour of presentation.11 In addition, any removable focus of infection must be identified and managed (eg, an abscess, infected catheter, tampon, or infection requiring surgery).

ACTIVATED PROTEIN C

Recombinant human activated protein C (APC) is a protein with anticoagulant and anti‐inflammatory properties that is relatively deficient in approximately 87% of septic patients.29 Although numerous trials of other anticoagulants (antithrombin III and tissue factor pathway inhibitor) and immunosuppressives (tumor necrosis factor inhibitors, high‐dose steroids, interleukin‐1 receptor antagonists, and others) have failed to show any benefit,7 in 2001 APC became the first proven therapy specifically for sepsis. The PROWESS trial, which established its efficacy, randomized 1690 patients who met 3 SIRS criteria and dysfunction of at least 1 organ system to APC (24 g/kg IV per hour for 96 hours interrupted for bleeding or urgent procedures) or placebo. APC reduced 28‐day mortality from 30.8% to 24.7%, yielding an absolute risk reduction of 6.1% and a corresponding number needed to treat (NNT) of 16.4. This benefit was seen across all subgroups including those with normal baseline APC levels.29

Not surprisingly, APC increases the risk of serious bleeding. Although this effect was of borderline significance in PROWESS (3.5% vs. 2% in the placebo group, P = .06),29 it was confirmed in subsequent trials (3.9% vs. 2.2%, P = .01)30 and may be larger still in open‐label use, at 6.5%.17, 31 Intracerebral hemorrhage (ICH), a particularly devastating complication, occurred in 0.2% of the PROWESS patients and 0.5% of patients in 2 subsequent studies30, 32; in both major trials, there was a single extra event in the APC arm.29, 30 Like serious bleeding in general, ICH was more common in open‐label use, occurring in 1.5% of patients.31, 33 Therefore, it is vital to have strict adherence to exclusion criteria and familiarity with the risk factors for serious bleeding. In the PROWESS trial, after randomization, risk factors for serious bleeding included procedures and injury to vascular organs, an activated partial‐thromboplastin time of more than 120 seconds, an international normalized ratio greater than 3, gastrointestinal ulceration, and development of severe thrombocytopenia (< 30,000/mm³)29; in a 2002 study of 2786 APC recipients, ICH was largely confined to patients with meningitis or a platelet count less than 30,000/mm³.32

APC therapy has several other limitations and drawbacks. Multiple contraindications, including predisposition to bleeding, a recent history of bleeding, anticoagulant use, immunosuppression, liver disease, dialysis dependence, and hypercoagulable states, restrict its use. APC appears to work best when administered early, within 24 hours of the onset of organ dysfunction.31 In addition, APC is indicated only in adults with Acute Physiology and Chronic Health Evaluation (APACHE II) scores greater than 24 and multiorgan failure. Post hoc analysis of the PROWESS data showed that although the relative risk (RR) of death for those with APACHE II scores of 25 or more was .71 and statistically significant, the RR for those with scores below 25 was a nonsignificant .99.34 A subsequent study, ADDRESS, confirmed there was no benefit to septic patients with a low risk of death.30 In the ADDRESS study 2613 patients with severe sepsis and either an APACHE II score less than 25 or single organ failure were randomized to APC or placebo. No differences were found in 28‐day and in‐hospital mortality; among patients who had undergone surgery in the previous 30 days, those receiving APC had a significantly increased risk of death (20.7% vs. 14.1%, P = .03).

An additional drawback of APC therapy is its cost, approximately $6800 per infusion, although the cost per year of life gained, $24,484, or $52,360 per life saved (NNT $6800), is reasonable for those with APACHE II scores greater than 24.34 Concerns have also been raised about the PROWESS trial itself: the production of the study drug and some exclusion criteria were changed midtrial, after which the effectiveness of APC improved. APACHE II scores had not been validated for selection of patients for therapies and may have varied with time or by observer. The original PROWESS study population may have been skewed away from chronically ill patients.35 Experts differ on the significance of these concerns and even whether APC therapy should be considered the standard of care pending further research.32, 35 The ADDRESS trial also failed to demonstrate a benefit in a subgroup of patients with APACHE II scores above 24, although it was underpowered to do so, and according to enrollment criteria, none of those patients had multiorgan failure.30 However, in the subgroup of PROWESS patients with APACHE II scores greater than 24, the absolute reduction in mortality was a full 13%,17 with a corresponding NNT of 7.7, and although the PROWESS findings have not been duplicated in a second randomized trial, a single‐arm, open‐label study of APC (ENHANCE) showed a nearly identical mortality rate.31 Pending confirmatory trials, APC remains a recommended therapy for selected patients sick enough to benefit and without excessive bleeding risk.11

EARLY GOAL‐DIRECTED THERAPY

Because physician‐directed resuscitation for sepsis may normalize vital signs, central venous pressures (CVP), and urine output without correcting hypoperfusion, Rivers et al. tested a resuscitation protocol that incorporated a central line that continuously monitored mixed‐venous oxygen saturation as a surrogate for cardiac output.36 They randomized 263 patients with septic shock (defined as hypotension < 90 mm Hg after a 20‐30 mL/kg bolus, or lactate > 4 mmol/L, which is associated with at least a 3‐fold increase in the mortality of emergency department patients with suspected infection37) to either standard care or early goal‐directed therapy (EGDT) for the initial 6 hours of hospital care. Patients with acute coronary ischemia, pulmonary edema, stroke, asthma, overdose, trauma, dysrhythmia, immunosuppression, uncontrolled cancer, or a need for urgent procedures were excluded. Standard care was directed by physiologic parameters such as vital signs, urine output, and CVP. EGDT used sequential therapies designed to support organ perfusion: 500 mL of normal saline was given every half hour until the CVP was at least 8‐12 mm Hg. Pressors were given until the mean arterial pressure was 65‐90 mm Hg (norepinephrine36 or dopamine were preferred agents, and vasopressin [0.01‐0.04 units/min] was an option for shock refractory to first‐line pressors)11, 38 Transfusion (to a hematocrit goal of 30) and dobutamine were given until mixed‐venous oxygenation saturation was 70% or better (Fig. 1). Lastly, patients who did not achieve this goal were sedated and mechanically ventilated.

Figure 1

Results were dramatic: mortality was reduced from 46.5% to 30.5%, with an ARR of 16% and an NNT of 6.25. Study patients received similar amounts of crystalloid, but received it earlier than the standard care patients and received more transfusions and inotropes. Substantially more patients in the EGDT group than the standard care group achieved a mixed venous oxygen saturation of 70%; 13.7% of the EGDT patients had occult hypoperfusion (low mixed‐venous oxygenation that responded to inotropes despite satisfactory vital signs). EGDT improved length of stay (4 days shorter among survivors) and duration of intubation, as well as APACHE scores and several physiologic parameters.36

Critics of this trial note the impossibility of adequate blinding and the high mortality in the placebo group. Further, because the trial tested the EGDT protocol as a whole, there was no way to know if each step was optimal. For example, a different CVP goal could have been used or adjustments made for mechanical ventilation, which can falsely elevate a low CVP into the desired range (the Surviving Sepsis Campaign guidelines recommend a CVP goal of 12‐15 mm Hg in mechanically ventilated patients11). Also, the selection of pressor, the use of inotropes, and the transfusion threshold were chosen on the basis of physiologic rationales, but all of these are arguable.39 This was also a single trial, and earlier goal‐directed protocols for ICU patients actually showed harm,40, 41 although those trials targeted supranormal physiologic goals in more established critical illness.42 Finally, on a practical level, hospitals and particularly emergency departments must commit resources to train physicians and staff, purchase the appropriate central venous catheters, and convince eligible patients to undergo an invasive procedure. In a survey of 30 attending physicians in academic referral hospitals, only 7% reported standard use of EGDT. Barriers included the requirement for specialty monitoring equipment and other resources, and central venous cannulation.43

Despite these concerns, the striking reduction in mortality associated with EGDT led to its endorsement by the Surviving Sepsis Campaign guidelines and underscores the principle of aggressive early resuscitation for patients who do not meet eligibility criteria but appear at risk for worsening sepsis. As yet, however, no strong evidence mandates a specific approach to the septic patient without shock.

STRESS DOSE STEROIDS

Because of the importance of the inflammatory cascade in severe sepsis, a potential role for steroids in the management of sepsis has been repeatedly studied. More than 50 studies have been performed since the 1950s, generally with pharmacologic doses of steroid; a meta‐analysis showed that such a practice was ineffective.44, 45 However, data accumulated that relative adrenal insufficiency during severe sepsis was common and associated with an increased risk of death and that physiologic doses of steroids could reverse refractory hypotension.46 To define the role of a physiologic course of steroids in septic shock, Annane et al. randomized 299 critically ill adults to either 7 days of stress dose hydrocortisone (50 mg IV every 6 hours) and fludrocortisone (50 g NG every 24 hours) or matched placebos. Enrolled patients were severely ill; the placebo group had a 63% mortality, and patients had to have septic shock, oliguria or hypoxia, hypotension despite low‐dose dopamine, and lactate greater than 2 mmol/L and require mechanical ventilation. Pregnant women, those with myocardial infarction or pulmonary embolus, advanced malignancies, or immunodeficiency, and those with clear indications or contraindications to steroids were excluded.47 Enrollment criteria were modified midstudy; changes included the exclusion of patients who had recently received etomidate, which inhibits 11‐‐hydroxylase and has been identified as a risk factor for adrenal insufficiency in intensive care patients.48 All patients received a 250‐g cosyntropin stimulation test. The authors considered patients nonresponders to consyntropin if serum cortisol failed to increase to 9 g/dL or more.

Steroids reduced the duration that a vasopressor was required and reduced mortality from 63% to 53% among nonresponders, giving an NNT of only 10 to prevent 1 death at 28 days. Although the authors described no evidence of adverse effects, among the subset of 70 patients who responded appropriately to cosyntropin, there was a nonsignificant trend toward increased mortality, and rates of hyperglycemia were not provided.47 The authors concluded that physicians should test appropriate patients for adrenal reserve, give the studied steroid regimen while results are pending, and discontinue treatment if a patient retains adrenal reserve.

The literature on steroids and critical illness is complex, with more than 1300 articles on steroids and sepsis published since 1988, and several concerns were raised about the Annane study. For example, did much of the benefit for those patients enrolled before the protocol amendment come from reducing an adverse effect of etomidate?49 Does the high‐dose, 250‐g cosyntropin stimulation test overcome (and conceal) partial ACTH resistance that might benefit from treatment?50 Might not a robust baseline cortisol suggest sufficient adrenal function regardless of the incremental response to cosyntropin?51 Partial answers were provided by 2 subsequent meta‐analyses. Both found that more recent studies gave lower doses of steroids in longer, 5‐ to 7‐day courses to sicker patients and demonstrated improvement in mortality and shock reversal, with relative risk reductions of 14%‐22%; the NNT ranged from 8 to 11. One analysis found no difference in outcomes between adrenally sufficient and adrenally insufficient patients, and those authors advised considering treatment for all patients regardless of their adrenal function test results.8 The other analysis concluded that the data on steroids for those with adrenal reserve was too limited to recommend treating adrenally sufficient patients.9

Disputes about certain details, such as whether patients should be treated without regard to adrenal reserve, continue in the literature.45, 52 An ongoing randomized, controlled trial, CORTICUS, is expected to provide additional guidance on the use of low‐dose steroids in sepsis; in the meantime, the literature clearly supports a longer course of low‐dose steroid therapy for patients with pressor‐dependent septic shock with inadequate adrenal reserve by cosyntropin testing, and guidelines allow discretion about whether patients with adequate adrenal reserve should also be treated.11 Hospitalists may also want to treat septic shock with equivalent doses of dexamethasone (approximately 2 mg IV every 6 hours) if adrenal evaluation may be delayed, as this agent will not confound cosyntropin stimulation test results, and they may want to avoid etomidate in septic patients53, 54 for whom they perform or supervise intubations.

INTENSIVE INSULIN THERAPY

Mounting evidence supports the short‐term role of hyperglycemia in morbidity and mortality, especially in critical illness. Hyperglycemia impairs neutrophil and endothelial cell function as well as protective responses to cardiac and neuronal ischemia,55 whereas insulin has anti‐inflammatory and antiapoptotic effects,7, 56 suggesting that intensive insulin might improve the outcomes of critically ill patients. To test this theory, van den Berghe and colleagues randomized 1548 mostly surgical ICU patients to insulin infusions titrated for glucose goals of either 80‐110 or 180‐200 mg/dL, followed by subcutaneous insulin after ICU discharge. Although blinding was impossible, in both cases glucose management was performed by a separate research team. Multiple benefits were noted: ICU and total in‐hospital deaths were reduced, mostly among patients with an ICU stay of more than 5 days, whose risk of death fell from 20.2% to 10.6%. Intensive insulin also reduced septicemia, renal impairment, critical illness polyneuropathy, and duration of intensive care.57

Subsequently, a meta‐analysis of 35 trials suggested that insulin reduced the mortality of critically ill patients by 15%.10 Van den Berghe et al.'s results were also duplicated in a broad, medical‐surgical ICU population, although the reductions in morbidity and mortality were measured against historical controls.58 However, whether the results of the influential surgical ICU study could be applied to medical patients was not known until 2006, when the van den Berghe group reported the effects of similar insulin protocols on 1200 patients in the medical ICU who were expected to need intensive care for at least 3 days.59 In this study, intensive insulin failed to reduce overall mortality (40% and 37.3%, P = .33). However, intensive insulin did reduce mortality among the 64% of patients who stayed in the ICU 3 or more days from 52.5% to 43% (NNT 10.5, P = .009). This benefit was offset by an increased number of deaths in the intensive insulin group among patients with ICU stays of less than 3 days (P = .05‐.35 depending on the method used).59 Intensive insulin did reduce newly acquired kidney injury, duration of mechanical ventilation, and lengths of ICU and hospital stays, and the reduction in morbidity increased with the duration of intensive insulin therapy. Hypoglycemia (mean 32 mg/dL) occurred in 25% of patients with prolonged stays6.4 times as often as in the usual care group.60 Liver and renal failure were associated with hypoglycemia.59

Critics of the surgical ICU trial noted the high mortality among the usual care patients (5.1%), a robust 34% mortality reduction for a relatively small 50 mg/dL reduction in morning glucose levels, and the aggressive use of parenteral nutrition, raising the question of whether intensive insulin merely attenuated the side effects of intravenous glucose.61, 62 Also, the ideal blood glucose target is not known with certainty. Retrospective studies suggested the upper limit for target blood glucoses could be 145 mg/dL63 and found differing thresholds at which hyperglycemia increased mortality in nondiabetics (144 mg/dL) and diabetics (200 mg/dL).64 However, in the surgical ICU trial, there was no threshold below which there was no further reduction in risk; patients whose mean blood glucose was below 110 mg/dL had lower mortality than those whose levels were between 110 and 150 mg/dL (P = .026).65 Finally, the effects of hyperglycemia and intensive insulin may vary by population: retrospective studies found that ICU hyperglycemia was more strongly associated with mortality among nondiabetics,64, 66 and van den Berghe et al. noted no benefit from intensive insulin in a small subgroup of diabetics.59

In summary, large, well‐designed trials have demonstrated that intensive insulin reduced mortality in critically ill patients after a delay of 3‐5 days, but this benefit did not extend to all patients in the medical ICU.57, 59 Some authors have suggested deferring intensive insulin for 3 days,67 but because early therapy probably contributes to the delayed mortality benefit, this approach may deprive patients of the observed benefits. Ongoing clinical trials (NICE‐SUGAR) are likely to provide useful information about how hyperglycemia should be managed in different populations, including septic ICU patients.61 In the meantime, institutions can select the intensity of their insulin therapy by weighing morbidity and long‐term mortality benefits against possible short‐term harms and ensuring that hospital staff members are sufficiently trained to control hyperglycemia safely. For example, in critical illness, intravenous insulin is preferable to subcutaneous insulin, and the frequent measurement of whole‐blood glucose instead of finger‐stick glucose helps to avoid errors.55, 68 And although researchers were unable to prospectively identify patients with long ICU stays,59 severely septic patients have long ICU stays (generally 7.5‐16.6 days),14 and individual ICUs might observe enough stays of more than 2 days in their patient population to justify intensive insulin for this subgroup. And finally, although no conclusive evidence mandates a specific approach to hyperglycemia outside the ICU, the ICU data provide a physiologic rationale for cautious but tight control of glucose in more moderately ill patients. Guidelines for the management of inpatient hyperglycemia were published previously.55

SEPSIS AND THE HOSPITALIST

Hospitalists who provide critical care may make frequent decisions about the inclusion and exclusion criteria for the major trials of sepsis, weigh their relative benefits against risks and costs, contemplate gray areas such as adrenal testing in shock, and employ evidence‐based therapies for severe sepsis. However, hospitalists may also see patients who qualify for these therapies when they are called to see septic patients in the emergency department, when severe sepsis develops in patients on the medicine ward, or when they provide consultation services in an ICU. Sepsis care must be implemented urgently; patients in the pivotal trial of steroids had to be randomized within 3 hours of shock onset,47 data suggest that the window for optimal antibiotic therapy may be no greater than 4 hours from diagnosis,18 whereas guidelines suggest therapy within 1 hour,11 and early goal‐directed therapy was studied only for the first 6 or more hours of hospitalization.36 Thus, hospitalists who do not provide ICU care should be able to identify patients with severe sepsis and either deliver initial care or recognize the need for immediate consultation. Specifically, hospitalists can:

• Recognize that both absolute (< 90 mm Hg) and relative hypotension (> 40 mm Hg below baseline) indicate septic shock;

• Identify normotensive candidates for EGDT (severe sepsis with serum lactate > 4 mmol/L) by requesting a serum lactate in addition to prompt appropriate cultures for severe acute infection69;

• Recognize atypical presentations of sepsis (tachypnea, tachycardia, confusion, etc.) and maintain a high suspicion for sepsis in patients who may be predisposed to infection and to atypical presentation because of age, immunosuppression, neutropenia, diabetes, or other conditions;

• Initiate effective antibiotics and EGDT promptly for individual patients or by coordinating efforts to improve sepsis care at an institutional level, for example, as a component of medical emergency team services70, 71;

• Rapidly identify and manage removable foci of infection such as abscesses, empyemas, necrotizing fasciitis, or infected vascular catheters; and

• Competently educate hospital staff, residents, and medical students about sepsis care.

Hospitalists are busy physicians, and the task of reviewing sepsis literature and implementing recommendations is daunting. However, hospitalists can turn to resources such as the Surviving Sepsis Campaign Guidelines, a series of recommendations for managing severe sepsis that were endorsed by 11 international critical care and infectious disease societies and published in Critical Care Medicine in 2004.11 The Institute for Healthcare Improvement has also published a series of online severe sepsis bundles, or groups of proven interventions, complete with implementation tips and supporting literature, available at http://www.survivingsepsis.org or http://www.ihi.org/IHI/Topics/CriticalCare/Sepsis/,72 and individual institutions have described their implementation of evidence‐based, protocol‐directed sepsis care (http://www.mustprotocol.com).69

CONCLUSION: DEADLY YET TREATABLE

The death toll from severe sepsis in the United States exceeds that of lung, breast, and colon cancer combined and equals that of myocardial infarction (MI),1 a condition that appropriately triggers a series of emergency interventions. Physicians now have an arsenal of therapies for severe sepsis analogous to those employed for MI, and a comparison between the 2 conditions underscores the high mortality rate of severe sepsis and the enormous impact on patient outcomes provided by evidence‐based sepsis therapy. Figure 2 compares the 9.5%‐16% ARR for death associated with APC in patients with APACHE 2 scores greater than 24 and multiorgan failure,29 EGDT,36 stress dose steroids in shock complicated by adrenal insufficiency,47 and intensive insulin in patients with medical ICU stays longer than 3 days,59 with the benefits of thrombolysis for ST‐elevation MI (2%‐3%)73 or antiplatelet therapy for acute MI (2.3%).74 Figure 3 compares the corresponding NNT values to save 1 life; according to the available data, a hospitalist is 5‐8 times more likely to save a life with EGDT than with fibrinolysis.

Figure 2

Figure 3

Because the literature supporting several major sepsis therapies have been limited to retrospective studies1828 and single randomized, controlled trials29, 36 and because key trials are still underway (CORTICUS, NICE‐SUGAR), the benefits of sepsis therapies are less certain than are those for the treatment of MI. This was underscored by the finding that the benefit in reduced mortality of intensive insulin in the surgical ICU57 did not extend to all patients in the medical ICU.59 However, the potentially marked survival benefit of early effective antibiotics, APC, EGDT, stress dose steroids, and intensive insulin and the urgency with which they must be applied demand that all hospitalists become or remain familiar with the evolving sepsis literature.

Despite decades of intensive research and improvements in medical care, severe sepsis affects an estimated 751,000 patients in the United States every year, killing 215,000 of them at an annual cost of 16.7 billion dollars.1 Because the elderly experience a 100‐fold increase in incidence, as compared with children, and a nearly 4‐fold increase in mortality (38.4% of those more than 85 years old), this burden is expected to increase with the aging population.1 Patients with severe sepsis have prolonged ICU14 and hospital stays and incur substantially increased costs compared with other patients.36

New research continues to explore the complex pathophysiology of sepsis,7 and clinicians, who once relied primarily on clinical experience and expert opinion to guide therapy, now have an increasing array of evidenced‐based sepsis therapies to employ. Recent meta‐analyses have evaluated several major treatments for severe sepsis,810 and recommendations (the Surviving Sepsis Campaign guidelines) for the treatment of severe sepsis were recently endorsed by 11 international critical care and infectious disease organizations.11 This article summarizes the current definitions of sepsis syndromes, the trials supporting the specific therapies for sepsis that are currently recommended, ongoing controversies and research, and implications for hospitalists, with a focus on early, effective antibiotics, activated protein C, early goal‐directed therapy, stress dose steroids, and intensive insulin therapy. For space considerations, readers are directed elsewhere for data supporting prophylaxis for deep venous thrombosis (DVT)12 and stress ulcer bleeding13 and for therapies less often directed by hospitalists, such as lung protective ventilation.14

DEFINITIONS

Systemic inflammatory response syndrome (SIRS), sepsis, severe sepsis, and septic shock were defined in 1992 to standardize the terminology of sepsis.15 These definitions have recently been reviewed and supported by a variety of American and European intensive care societies.16

SIRS is defined by the presence of at least 2 of the following:

• Temperature > 38C or < 36C;

• Heart rate > 90 beats/min;

• Respiratory rate > 20 breaths/min or PaCO2 < 32 mm Hg;

• WBC >12,000 or < 4000 cells/mm³, or >10% immature (band) forms.

Sepsis is SIRS due to documented or strongly suspected infection.

Severe sepsis is sepsis with organ dysfunction (such as lactic acidosis, oliguria, thrombocytopenia, or delirium), hypoperfusion, or hypotension (< 90 mm Hg systolic or more than 40 mm Hg below baseline).

Septic shock is severe sepsis complicated by hypotension or pressor dependence despite adequate (20‐30 mL/kg; 1.5‐3 liters in most patients) fluid resuscitation.

Sepsis terminology must be applied carefully. Many hospitalized patients meet criteria for SIRS, yet it is inaccurate to say a patient who has acute leukemia with leukocytosis, anemia‐induced tachycardia, and thrombocytopenia has severe sepsis if those abnormalities are not a result of inflammation or infection. Accurate documentation of sepsis syndromes can improve professional and institutional reimbursement and provide prognostic information: the in‐hospital mortality rates for severe sepsis and septic shock are approximately 30% and 50%, respectively.17 More importantly, thoughtful application of these definitions can help a hospitalist identify septic patients who qualify for one of the proven therapies for severe sepsis.

EARLY, EFFECTIVE ANTIBIOTICS

For obvious ethical reasons, randomized, controlled trials to study the impact of inappropriate or delayed antibiotic therapy for serious infections are not possible. However, the evidence supporting early, effective antibiotic therapy is still compelling, and because many hospitalists often initiate treatment with antibiotics before transferring a patient to intensive care, this may represent the most important intervention hospitalists can provide to patients with serious infections. Several studies have estimated the impact of early, effective antibiotics on outcomes.

Houck et al. retrospectively reviewed 13,771 cases of community‐acquired pneumonia among elderly Medicare patients. They found that 39.1% of the patients waited more than 4 hours for antibiotics and 7.6% waited more than 12 hours; three quarters of these delays resulted from delayed ordering of antibiotics.18 Further, 21.2% received an antibiotic selection incompatible with recent professional guidelines. Receiving antibiotics within 4 hours reduced in‐hospital and 30‐day mortality by 15% and length of stay by 0.4 days.18 Similar conclusions were reported by 3 of 4 previous analyses.1922 Extending these findings to critically ill patients, Iregui et al. found that delayed treatment with appropriate antibiotics (odds ratio, 7.68) was a greater predictor of mortality for 107 patients with ventilator‐acquired pneumonia than were APACHE II scores and malignancy; 31% failed to receive appropriate antibiotics within 24 hours, and again, three quarters of these delays resulted from delays in writing antibiotic orders.23

Not surprisingly, antibiotic therapy must be effective as well as timely. MacArthur et al. studied the impact of adequate (ie, active against cultured organisms, if isolated) antibiotics on the outcomes of 2634 septic patients enrolled in a randomized trial of an anti‐TNF antibody. Nearly 91% received appropriate antibiotics; their mortality rate was 33%, 10% lower than that of the patients whose initial antibiotics were inadequate (P < .001).24 Leibovici et al. reported similar findings in a prospective study of patients with bacteremia. Only 63% of 3413 subjects received an antibiotic active against the infecting pathogen, and their mortality was 20%, 14% lower than that in the group that received ineffective antibiotics (P = .0001).25 Other authors have reported even worse outcomes with ineffective therapy: 62% mortality among inadequately treated bacteremic or fungemic ICU patients, compared with 28.4% among those who were adequately treated26 and an odds ratio of dying of 8.14 for the 46 of 270 septic ICU patients who received inadequate initial antibiotics,27 making inadequate antibiotic therapy the strongest risk factor for death. Finally, Kollef et al. reported that 26% of 655 infected ICU patients received inadequate antibiotics and suffered an infection‐related mortality rate of 40.2%, more than twice the 17.7% rate among adequately treated patients (P < .001). Inadequate antimicrobial therapy was a greater risk factor for death than early respiratory failure or sepsis‐related organ failure assessment scores.28

Guidelines for anti‐infective care now recommend obtaining appropriate cultures and administering broad‐spectrum antibiotics (appropriate for suspected infections, local susceptibility patterns, and any relevant prior culture data from individual patients) within 1 hour of presentation.11 In addition, any removable focus of infection must be identified and managed (eg, an abscess, infected catheter, tampon, or infection requiring surgery).

ACTIVATED PROTEIN C

Recombinant human activated protein C (APC) is a protein with anticoagulant and anti‐inflammatory properties that is relatively deficient in approximately 87% of septic patients.29 Although numerous trials of other anticoagulants (antithrombin III and tissue factor pathway inhibitor) and immunosuppressives (tumor necrosis factor inhibitors, high‐dose steroids, interleukin‐1 receptor antagonists, and others) have failed to show any benefit,7 in 2001 APC became the first proven therapy specifically for sepsis. The PROWESS trial, which established its efficacy, randomized 1690 patients who met 3 SIRS criteria and dysfunction of at least 1 organ system to APC (24 g/kg IV per hour for 96 hours interrupted for bleeding or urgent procedures) or placebo. APC reduced 28‐day mortality from 30.8% to 24.7%, yielding an absolute risk reduction of 6.1% and a corresponding number needed to treat (NNT) of 16.4. This benefit was seen across all subgroups including those with normal baseline APC levels.29

Not surprisingly, APC increases the risk of serious bleeding. Although this effect was of borderline significance in PROWESS (3.5% vs. 2% in the placebo group, P = .06),29 it was confirmed in subsequent trials (3.9% vs. 2.2%, P = .01)30 and may be larger still in open‐label use, at 6.5%.17, 31 Intracerebral hemorrhage (ICH), a particularly devastating complication, occurred in 0.2% of the PROWESS patients and 0.5% of patients in 2 subsequent studies30, 32; in both major trials, there was a single extra event in the APC arm.29, 30 Like serious bleeding in general, ICH was more common in open‐label use, occurring in 1.5% of patients.31, 33 Therefore, it is vital to have strict adherence to exclusion criteria and familiarity with the risk factors for serious bleeding. In the PROWESS trial, after randomization, risk factors for serious bleeding included procedures and injury to vascular organs, an activated partial‐thromboplastin time of more than 120 seconds, an international normalized ratio greater than 3, gastrointestinal ulceration, and development of severe thrombocytopenia (< 30,000/mm³)29; in a 2002 study of 2786 APC recipients, ICH was largely confined to patients with meningitis or a platelet count less than 30,000/mm³.32

APC therapy has several other limitations and drawbacks. Multiple contraindications, including predisposition to bleeding, a recent history of bleeding, anticoagulant use, immunosuppression, liver disease, dialysis dependence, and hypercoagulable states, restrict its use. APC appears to work best when administered early, within 24 hours of the onset of organ dysfunction.31 In addition, APC is indicated only in adults with Acute Physiology and Chronic Health Evaluation (APACHE II) scores greater than 24 and multiorgan failure. Post hoc analysis of the PROWESS data showed that although the relative risk (RR) of death for those with APACHE II scores of 25 or more was .71 and statistically significant, the RR for those with scores below 25 was a nonsignificant .99.34 A subsequent study, ADDRESS, confirmed there was no benefit to septic patients with a low risk of death.30 In the ADDRESS study 2613 patients with severe sepsis and either an APACHE II score less than 25 or single organ failure were randomized to APC or placebo. No differences were found in 28‐day and in‐hospital mortality; among patients who had undergone surgery in the previous 30 days, those receiving APC had a significantly increased risk of death (20.7% vs. 14.1%, P = .03).

An additional drawback of APC therapy is its cost, approximately $6800 per infusion, although the cost per year of life gained, $24,484, or $52,360 per life saved (NNT $6800), is reasonable for those with APACHE II scores greater than 24.34 Concerns have also been raised about the PROWESS trial itself: the production of the study drug and some exclusion criteria were changed midtrial, after which the effectiveness of APC improved. APACHE II scores had not been validated for selection of patients for therapies and may have varied with time or by observer. The original PROWESS study population may have been skewed away from chronically ill patients.35 Experts differ on the significance of these concerns and even whether APC therapy should be considered the standard of care pending further research.32, 35 The ADDRESS trial also failed to demonstrate a benefit in a subgroup of patients with APACHE II scores above 24, although it was underpowered to do so, and according to enrollment criteria, none of those patients had multiorgan failure.30 However, in the subgroup of PROWESS patients with APACHE II scores greater than 24, the absolute reduction in mortality was a full 13%,17 with a corresponding NNT of 7.7, and although the PROWESS findings have not been duplicated in a second randomized trial, a single‐arm, open‐label study of APC (ENHANCE) showed a nearly identical mortality rate.31 Pending confirmatory trials, APC remains a recommended therapy for selected patients sick enough to benefit and without excessive bleeding risk.11

EARLY GOAL‐DIRECTED THERAPY

Because physician‐directed resuscitation for sepsis may normalize vital signs, central venous pressures (CVP), and urine output without correcting hypoperfusion, Rivers et al. tested a resuscitation protocol that incorporated a central line that continuously monitored mixed‐venous oxygen saturation as a surrogate for cardiac output.36 They randomized 263 patients with septic shock (defined as hypotension < 90 mm Hg after a 20‐30 mL/kg bolus, or lactate > 4 mmol/L, which is associated with at least a 3‐fold increase in the mortality of emergency department patients with suspected infection37) to either standard care or early goal‐directed therapy (EGDT) for the initial 6 hours of hospital care. Patients with acute coronary ischemia, pulmonary edema, stroke, asthma, overdose, trauma, dysrhythmia, immunosuppression, uncontrolled cancer, or a need for urgent procedures were excluded. Standard care was directed by physiologic parameters such as vital signs, urine output, and CVP. EGDT used sequential therapies designed to support organ perfusion: 500 mL of normal saline was given every half hour until the CVP was at least 8‐12 mm Hg. Pressors were given until the mean arterial pressure was 65‐90 mm Hg (norepinephrine36 or dopamine were preferred agents, and vasopressin [0.01‐0.04 units/min] was an option for shock refractory to first‐line pressors)11, 38 Transfusion (to a hematocrit goal of 30) and dobutamine were given until mixed‐venous oxygenation saturation was 70% or better (Fig. 1). Lastly, patients who did not achieve this goal were sedated and mechanically ventilated.

Figure 1

Results were dramatic: mortality was reduced from 46.5% to 30.5%, with an ARR of 16% and an NNT of 6.25. Study patients received similar amounts of crystalloid, but received it earlier than the standard care patients and received more transfusions and inotropes. Substantially more patients in the EGDT group than the standard care group achieved a mixed venous oxygen saturation of 70%; 13.7% of the EGDT patients had occult hypoperfusion (low mixed‐venous oxygenation that responded to inotropes despite satisfactory vital signs). EGDT improved length of stay (4 days shorter among survivors) and duration of intubation, as well as APACHE scores and several physiologic parameters.36

Critics of this trial note the impossibility of adequate blinding and the high mortality in the placebo group. Further, because the trial tested the EGDT protocol as a whole, there was no way to know if each step was optimal. For example, a different CVP goal could have been used or adjustments made for mechanical ventilation, which can falsely elevate a low CVP into the desired range (the Surviving Sepsis Campaign guidelines recommend a CVP goal of 12‐15 mm Hg in mechanically ventilated patients11). Also, the selection of pressor, the use of inotropes, and the transfusion threshold were chosen on the basis of physiologic rationales, but all of these are arguable.39 This was also a single trial, and earlier goal‐directed protocols for ICU patients actually showed harm,40, 41 although those trials targeted supranormal physiologic goals in more established critical illness.42 Finally, on a practical level, hospitals and particularly emergency departments must commit resources to train physicians and staff, purchase the appropriate central venous catheters, and convince eligible patients to undergo an invasive procedure. In a survey of 30 attending physicians in academic referral hospitals, only 7% reported standard use of EGDT. Barriers included the requirement for specialty monitoring equipment and other resources, and central venous cannulation.43

Despite these concerns, the striking reduction in mortality associated with EGDT led to its endorsement by the Surviving Sepsis Campaign guidelines and underscores the principle of aggressive early resuscitation for patients who do not meet eligibility criteria but appear at risk for worsening sepsis. As yet, however, no strong evidence mandates a specific approach to the septic patient without shock.

STRESS DOSE STEROIDS

Because of the importance of the inflammatory cascade in severe sepsis, a potential role for steroids in the management of sepsis has been repeatedly studied. More than 50 studies have been performed since the 1950s, generally with pharmacologic doses of steroid; a meta‐analysis showed that such a practice was ineffective.44, 45 However, data accumulated that relative adrenal insufficiency during severe sepsis was common and associated with an increased risk of death and that physiologic doses of steroids could reverse refractory hypotension.46 To define the role of a physiologic course of steroids in septic shock, Annane et al. randomized 299 critically ill adults to either 7 days of stress dose hydrocortisone (50 mg IV every 6 hours) and fludrocortisone (50 g NG every 24 hours) or matched placebos. Enrolled patients were severely ill; the placebo group had a 63% mortality, and patients had to have septic shock, oliguria or hypoxia, hypotension despite low‐dose dopamine, and lactate greater than 2 mmol/L and require mechanical ventilation. Pregnant women, those with myocardial infarction or pulmonary embolus, advanced malignancies, or immunodeficiency, and those with clear indications or contraindications to steroids were excluded.47 Enrollment criteria were modified midstudy; changes included the exclusion of patients who had recently received etomidate, which inhibits 11‐‐hydroxylase and has been identified as a risk factor for adrenal insufficiency in intensive care patients.48 All patients received a 250‐g cosyntropin stimulation test. The authors considered patients nonresponders to consyntropin if serum cortisol failed to increase to 9 g/dL or more.

Steroids reduced the duration that a vasopressor was required and reduced mortality from 63% to 53% among nonresponders, giving an NNT of only 10 to prevent 1 death at 28 days. Although the authors described no evidence of adverse effects, among the subset of 70 patients who responded appropriately to cosyntropin, there was a nonsignificant trend toward increased mortality, and rates of hyperglycemia were not provided.47 The authors concluded that physicians should test appropriate patients for adrenal reserve, give the studied steroid regimen while results are pending, and discontinue treatment if a patient retains adrenal reserve.

The literature on steroids and critical illness is complex, with more than 1300 articles on steroids and sepsis published since 1988, and several concerns were raised about the Annane study. For example, did much of the benefit for those patients enrolled before the protocol amendment come from reducing an adverse effect of etomidate?49 Does the high‐dose, 250‐g cosyntropin stimulation test overcome (and conceal) partial ACTH resistance that might benefit from treatment?50 Might not a robust baseline cortisol suggest sufficient adrenal function regardless of the incremental response to cosyntropin?51 Partial answers were provided by 2 subsequent meta‐analyses. Both found that more recent studies gave lower doses of steroids in longer, 5‐ to 7‐day courses to sicker patients and demonstrated improvement in mortality and shock reversal, with relative risk reductions of 14%‐22%; the NNT ranged from 8 to 11. One analysis found no difference in outcomes between adrenally sufficient and adrenally insufficient patients, and those authors advised considering treatment for all patients regardless of their adrenal function test results.8 The other analysis concluded that the data on steroids for those with adrenal reserve was too limited to recommend treating adrenally sufficient patients.9

Disputes about certain details, such as whether patients should be treated without regard to adrenal reserve, continue in the literature.45, 52 An ongoing randomized, controlled trial, CORTICUS, is expected to provide additional guidance on the use of low‐dose steroids in sepsis; in the meantime, the literature clearly supports a longer course of low‐dose steroid therapy for patients with pressor‐dependent septic shock with inadequate adrenal reserve by cosyntropin testing, and guidelines allow discretion about whether patients with adequate adrenal reserve should also be treated.11 Hospitalists may also want to treat septic shock with equivalent doses of dexamethasone (approximately 2 mg IV every 6 hours) if adrenal evaluation may be delayed, as this agent will not confound cosyntropin stimulation test results, and they may want to avoid etomidate in septic patients53, 54 for whom they perform or supervise intubations.

INTENSIVE INSULIN THERAPY

Mounting evidence supports the short‐term role of hyperglycemia in morbidity and mortality, especially in critical illness. Hyperglycemia impairs neutrophil and endothelial cell function as well as protective responses to cardiac and neuronal ischemia,55 whereas insulin has anti‐inflammatory and antiapoptotic effects,7, 56 suggesting that intensive insulin might improve the outcomes of critically ill patients. To test this theory, van den Berghe and colleagues randomized 1548 mostly surgical ICU patients to insulin infusions titrated for glucose goals of either 80‐110 or 180‐200 mg/dL, followed by subcutaneous insulin after ICU discharge. Although blinding was impossible, in both cases glucose management was performed by a separate research team. Multiple benefits were noted: ICU and total in‐hospital deaths were reduced, mostly among patients with an ICU stay of more than 5 days, whose risk of death fell from 20.2% to 10.6%. Intensive insulin also reduced septicemia, renal impairment, critical illness polyneuropathy, and duration of intensive care.57

Subsequently, a meta‐analysis of 35 trials suggested that insulin reduced the mortality of critically ill patients by 15%.10 Van den Berghe et al.'s results were also duplicated in a broad, medical‐surgical ICU population, although the reductions in morbidity and mortality were measured against historical controls.58 However, whether the results of the influential surgical ICU study could be applied to medical patients was not known until 2006, when the van den Berghe group reported the effects of similar insulin protocols on 1200 patients in the medical ICU who were expected to need intensive care for at least 3 days.59 In this study, intensive insulin failed to reduce overall mortality (40% and 37.3%, P = .33). However, intensive insulin did reduce mortality among the 64% of patients who stayed in the ICU 3 or more days from 52.5% to 43% (NNT 10.5, P = .009). This benefit was offset by an increased number of deaths in the intensive insulin group among patients with ICU stays of less than 3 days (P = .05‐.35 depending on the method used).59 Intensive insulin did reduce newly acquired kidney injury, duration of mechanical ventilation, and lengths of ICU and hospital stays, and the reduction in morbidity increased with the duration of intensive insulin therapy. Hypoglycemia (mean 32 mg/dL) occurred in 25% of patients with prolonged stays6.4 times as often as in the usual care group.60 Liver and renal failure were associated with hypoglycemia.59

Critics of the surgical ICU trial noted the high mortality among the usual care patients (5.1%), a robust 34% mortality reduction for a relatively small 50 mg/dL reduction in morning glucose levels, and the aggressive use of parenteral nutrition, raising the question of whether intensive insulin merely attenuated the side effects of intravenous glucose.61, 62 Also, the ideal blood glucose target is not known with certainty. Retrospective studies suggested the upper limit for target blood glucoses could be 145 mg/dL63 and found differing thresholds at which hyperglycemia increased mortality in nondiabetics (144 mg/dL) and diabetics (200 mg/dL).64 However, in the surgical ICU trial, there was no threshold below which there was no further reduction in risk; patients whose mean blood glucose was below 110 mg/dL had lower mortality than those whose levels were between 110 and 150 mg/dL (P = .026).65 Finally, the effects of hyperglycemia and intensive insulin may vary by population: retrospective studies found that ICU hyperglycemia was more strongly associated with mortality among nondiabetics,64, 66 and van den Berghe et al. noted no benefit from intensive insulin in a small subgroup of diabetics.59

In summary, large, well‐designed trials have demonstrated that intensive insulin reduced mortality in critically ill patients after a delay of 3‐5 days, but this benefit did not extend to all patients in the medical ICU.57, 59 Some authors have suggested deferring intensive insulin for 3 days,67 but because early therapy probably contributes to the delayed mortality benefit, this approach may deprive patients of the observed benefits. Ongoing clinical trials (NICE‐SUGAR) are likely to provide useful information about how hyperglycemia should be managed in different populations, including septic ICU patients.61 In the meantime, institutions can select the intensity of their insulin therapy by weighing morbidity and long‐term mortality benefits against possible short‐term harms and ensuring that hospital staff members are sufficiently trained to control hyperglycemia safely. For example, in critical illness, intravenous insulin is preferable to subcutaneous insulin, and the frequent measurement of whole‐blood glucose instead of finger‐stick glucose helps to avoid errors.55, 68 And although researchers were unable to prospectively identify patients with long ICU stays,59 severely septic patients have long ICU stays (generally 7.5‐16.6 days),14 and individual ICUs might observe enough stays of more than 2 days in their patient population to justify intensive insulin for this subgroup. And finally, although no conclusive evidence mandates a specific approach to hyperglycemia outside the ICU, the ICU data provide a physiologic rationale for cautious but tight control of glucose in more moderately ill patients. Guidelines for the management of inpatient hyperglycemia were published previously.55

SEPSIS AND THE HOSPITALIST

Hospitalists who provide critical care may make frequent decisions about the inclusion and exclusion criteria for the major trials of sepsis, weigh their relative benefits against risks and costs, contemplate gray areas such as adrenal testing in shock, and employ evidence‐based therapies for severe sepsis. However, hospitalists may also see patients who qualify for these therapies when they are called to see septic patients in the emergency department, when severe sepsis develops in patients on the medicine ward, or when they provide consultation services in an ICU. Sepsis care must be implemented urgently; patients in the pivotal trial of steroids had to be randomized within 3 hours of shock onset,47 data suggest that the window for optimal antibiotic therapy may be no greater than 4 hours from diagnosis,18 whereas guidelines suggest therapy within 1 hour,11 and early goal‐directed therapy was studied only for the first 6 or more hours of hospitalization.36 Thus, hospitalists who do not provide ICU care should be able to identify patients with severe sepsis and either deliver initial care or recognize the need for immediate consultation. Specifically, hospitalists can:

• Recognize that both absolute (< 90 mm Hg) and relative hypotension (> 40 mm Hg below baseline) indicate septic shock;

• Identify normotensive candidates for EGDT (severe sepsis with serum lactate > 4 mmol/L) by requesting a serum lactate in addition to prompt appropriate cultures for severe acute infection69;

• Recognize atypical presentations of sepsis (tachypnea, tachycardia, confusion, etc.) and maintain a high suspicion for sepsis in patients who may be predisposed to infection and to atypical presentation because of age, immunosuppression, neutropenia, diabetes, or other conditions;

• Initiate effective antibiotics and EGDT promptly for individual patients or by coordinating efforts to improve sepsis care at an institutional level, for example, as a component of medical emergency team services70, 71;

• Rapidly identify and manage removable foci of infection such as abscesses, empyemas, necrotizing fasciitis, or infected vascular catheters; and

• Competently educate hospital staff, residents, and medical students about sepsis care.

Hospitalists are busy physicians, and the task of reviewing sepsis literature and implementing recommendations is daunting. However, hospitalists can turn to resources such as the Surviving Sepsis Campaign Guidelines, a series of recommendations for managing severe sepsis that were endorsed by 11 international critical care and infectious disease societies and published in Critical Care Medicine in 2004.11 The Institute for Healthcare Improvement has also published a series of online severe sepsis bundles, or groups of proven interventions, complete with implementation tips and supporting literature, available at http://www.survivingsepsis.org or http://www.ihi.org/IHI/Topics/CriticalCare/Sepsis/,72 and individual institutions have described their implementation of evidence‐based, protocol‐directed sepsis care (http://www.mustprotocol.com).69

CONCLUSION: DEADLY YET TREATABLE

The death toll from severe sepsis in the United States exceeds that of lung, breast, and colon cancer combined and equals that of myocardial infarction (MI),1 a condition that appropriately triggers a series of emergency interventions. Physicians now have an arsenal of therapies for severe sepsis analogous to those employed for MI, and a comparison between the 2 conditions underscores the high mortality rate of severe sepsis and the enormous impact on patient outcomes provided by evidence‐based sepsis therapy. Figure 2 compares the 9.5%‐16% ARR for death associated with APC in patients with APACHE 2 scores greater than 24 and multiorgan failure,29 EGDT,36 stress dose steroids in shock complicated by adrenal insufficiency,47 and intensive insulin in patients with medical ICU stays longer than 3 days,59 with the benefits of thrombolysis for ST‐elevation MI (2%‐3%)73 or antiplatelet therapy for acute MI (2.3%).74 Figure 3 compares the corresponding NNT values to save 1 life; according to the available data, a hospitalist is 5‐8 times more likely to save a life with EGDT than with fibrinolysis.

Figure 2

Figure 3

Because the literature supporting several major sepsis therapies have been limited to retrospective studies1828 and single randomized, controlled trials29, 36 and because key trials are still underway (CORTICUS, NICE‐SUGAR), the benefits of sepsis therapies are less certain than are those for the treatment of MI. This was underscored by the finding that the benefit in reduced mortality of intensive insulin in the surgical ICU57 did not extend to all patients in the medical ICU.59 However, the potentially marked survival benefit of early effective antibiotics, APC, EGDT, stress dose steroids, and intensive insulin and the urgency with which they must be applied demand that all hospitalists become or remain familiar with the evolving sepsis literature.

A 35‐year‐old man presented to the emergency department of a community hospital with 3 days of nausea, vomiting, abdominal pain, and diarrhea. He had developed a sore throat, nasal congestion, and green sputum on a business trip to Las Vegas 10 days prior. He then traveled to Shanghai, China, where he developed frequent diarrhea with mucous and urgency. The stool was mustard colored without blood or melena. He became nauseated and unable to keep down food or fluids. He noted a 4‐kg weight loss since the beginning of his symptoms. He had not traveled outside Shanghai or eaten exotic foods. His travel companions remained unaffected. The patient had had type I diabetes mellitus for 23 years with known retinopathy and microalbuminuria. He had hypertension and hyperlipidemia but was otherwise healthy. He was a married traveling salesman with 3 healthy children. He did not smoke or drink and reported no drug use or extramarital relations. He had no known allergies. His medications included insulin, lisinopril, atorvastatin, and valsartan. He was afebrile, his vital signs were stable and the physical examination was unremarkable.

The patient presents with gastrointestinal symptoms following a trip to China. He may have an infection that began in the respiratory system and now is causing some gastrointestinal symptoms, like Legionnella. If he had received antibiotics, the diarrheal illness could be a complication. With his recent travel to China, typical enteric pathogens would have to be considered: enterotoxigenic E. coli, Shigella, Salmonella, Campylobacter, or perhaps Giardia.

He was admitted and treated with intravenous fluids and ciprofloxacin for presumed gastroenteritis. He was discharged the next day but returned 2 days later because of continued nausea and vomiting and limited oral intake. He was febrile, to 38.9C, with chills. The results of an abdominal exam were normal. Stool studies for Salmonella, Shigella, Campylobacter, Yersinia, enterotoxigenic E. coli, Giardia, and Cryptosporidium were negative. No blood parasites were seen. The white‐cell count was 3900/mm³ with a normal differential count. Hemoglobin level, platelet count, serum electrolytes and creatinine were normal. He was readmitted for intravenous fluids.

The patient was treated for traveler's diarrhea, although this is not a typical case and is now becoming a protracted illness. Amebiasis would be a consideration, as well as typhoid or perhaps an abdominal abscess. The normal platelet count reduces the likelihood of hemolytic uremic syndrome, as does the absence of bloody diarrhea. I would evaluate for a systemic illness: check for adenopathy, do a thorough abdominal exam, and get a chest radiogram, blood cultures, and liver function tests. I am also concerned about metabolic abnormalities that could occur as a consequence of the diarrhea.

Over 3 days, oliguria developed along with urinary hesitancy, a 9‐kg weight gain, and the development of marked edema. Blood pressure and heart rate remained normal, and a chest radiograph was clear. Liver function tests were normal. A urinary catheter was inserted. Urinalysis revealed a specific gravity of 1.031, protein of 100 mg/dL, and trace glucose but was otherwise negative; no casts or cells were seen in the sediment. Chemistries included sodium of 133 mmol/L, potassium of 3.9 mmol/L, and serum bicarbonate of 18.4 mmol/L. Blood and urine cultures were sterile. The creatinine increased from 1.0 mg/dL (88.4 mol/L) to 1.3 mg/dL (115 mol/L). He was transferred to a tertiary care hospital for renal consultation because of concerns of impending renal failure and for consideration of a kidney biopsy.

In a typical case of a malabsorptive diarrhea, the patient could be volume depleted, but in this case he has gained 9‐kg and is grossly edematous. The chest radiograph and liver tests point to renal rather than cardiac or hepatologic causes for the edema. A glomerulonephritis may be driving the salt and water retention.

The proteinuria could be related to hemodynamics, or it could be from a glomerular lesion secondary to immune complexes. The specific gravity of 1.031 indicates the kidney is able to concentrate, and we are not seeing acute tubular necrosis. There is only minimal elevation in creatinine at this point. Quantitation or estimation of the degree of proteinuria by a protein‐to‐creatinine ratio would be helpful.

A further workup should include additional blood cultures and a CT scan of the lungs and abdomen to look for occult infection. Is he unfortunate enough to have developed a malignancy? Is this a connective tissue disease? Reexamination of the urine sediment is important to evaluate for glomerulonephritis.

The patient reported ongoing nausea and vomiting, but his diarrhea resolved. He was tachypneic, with a respiratory rate of 26 breaths/minute and an oxygen saturation of 98% breathing ambient air. His temperature was 37.1C, heart rate 84 beats/minute, and blood pressure 120/65 mm Hg. His mucous membranes were moist, and his jugular venous pressure was 6 cm. No lymphadenopathy was present. The heart and lungs were normal. The abdomen was soft, nontender, and without organomegaly, masses, or shifting dullness; bowel sounds were hypoactive. Severe edema of his legs, sacrum, hands, arms, and orbits was noted. His right hand and wrist were painful with limited mobility and small joint effusions of the wrist and metacarpophalangeal joints, but without erythema or warmth. Small petechiae were noted on his eyelids; skin examination was otherwise unremarkable. He had been given ciprofloxacin, phenergan, calcium carbonate, and pantoprazole prior to transfer. Stool was negative for occult blood.

His lungs are clear, but it is possible to have early pulmonary congestion with normal breath sounds. As he is normotensive and has a normal JVP, I would not give further intravenous fluid. Unless he has evidence of symptomatic pulmonary edema, I would not give diuretics but would simply observe his course. At this point I would ultrasound his kidneys to make sure there is no obstruction. The proteinuria could be a result of underlying diabetic glomerulopathy that may predispose to fluid retention.

He has significant oliguria but only a mild rise in creatinine. As the referring physicians requested a biopsy, we should consider it. A decision to biopsy the kidney would rest on the degree of proteinuria and the activity of the sediment. For example, if red blood cell casts or dysmorphic red cells were present, postinfectious glomerulonephritis or IgA nephropathy would be more likely. However, if the proteinuria is in the non‐nephrotic range and the sediment is nonreactive, the yield of a biopsy would be low.

His right hand and wrist are painful with limited mobility. This could be a sequela of endocarditis, although the absence of a murmur and negative blood cultures make it unlikely. He could have an infectious arthritis, although this usually presents more dramatically. He could have gout or pseudogout, which could be determined by joint aspiration. Finally, could this be iatrogenic? A drug reaction could explain some of the features, including rash, fever, joint symptoms, and renal abnormalities.

Repeat urinalysis revealed protein of more than 300 mg/dL, hematuria (1+), mucous, renal tubular epithelial cells, renal tubular epithelial cell casts, and granular casts. Eosinophiluria was absent. Laboratory evaluation revealed a hemoglobin level of 10.7 g/dL (decreased from 13 g/dL on initial presentation), white‐cell count of 7300/mm³, platelet count of 337,000/mm³, serum creatinine of 1.2 mg/dL (106 mol/L), blood urea nitrogen of 23 mg/dL (8.2 mol.L), total serum protein of 6.3 g/dL (normal range, 6.3‐8.7), and albumin of 2.7 g/dL (normal range, 3.2‐5.2). Other liver tests and serum electrolytes were normal.

This degree of proteinuria is significant, but it is unclear whether this is related to the underlying disease process or his advancing diabetes. He has some hematuria, but that could be from the urinary catheter. It would be helpful to know if the red blood cells are dysmorphic, which would point to a glomerulonephritis. He has renal cells, renal cell casts, and granular casts, which are nonspecific. He has a mild anemia, which is unexplained, but could relate to phlebotomy or overhydration. The hypoalbuminemia may be a result of renal losses or a catabolic state.

A renal ultrasonogram was normal, apart from evidence of bilateral pleural effusions. Antinuclear antibody and rheumatoid factor test results were negative, as were those for hepatitis A, hepatitis B surface antigen, and hepatitis C antibodies. Antistreptolysin O and antideoxyribonuclease B titers were normal. Total complement Ch50 was low at 29 U/mL (normal range, 30‐75) as was complement factor C3 at 67 mg/dL (normal range, 90‐180). Complement factor C4 was normal. Serum and urine electrophoresis revealed no monoclonal protein spike. Vitamin B12 and serum folate were normal, serum ferritin was 584 ng/mL (normal range, 30‐400), iron serum was 29 g/dL (normal range, 45‐160), transferrin saturation was 16%, and total iron‐binding capacity was 164 g/dL (normal range, 250‐450). The reticulocyte count was 2.9% with an absolute reticulocyte count of 102/cm³ and a reticulocyte production index of 0.96 (normal range, 1.0‐2.0).

It is reassuring that his urinary tract ultrasound is normal. In addition to edema, he has bilateral effusions, which are probably transudative, related to fluid overload. The urinalysis does not suggest a rapidly progressive glomerulonephritis, but autoimmune disease is still in the differential.

He has a mild complement C3 deficiency. In nephrology we think of lupus, infective endocarditis, cryoglobulinemia, and specific glomerular lesions such as membranoproliferative glomerulonephritis and postinfectious glomerulonephritis as being associated with the development of circulating immune complexes that may lead to low complement levels. There is no evidence of a paraprotein, but testing for cryoglobulins should be considered. Cryoglobulins are associated with hepatitis C but may be induced by a variety of infections. Acting like immune complexes, they can lead to low complement levels and could cause some of this patient's symptoms. However, this whole illness seems most likely to be secondary to infection. The normal antistreptolysin O and antideoxyribonuclease B titers make streptococcal disease unlikely, but another bacterial infection could cause postinfectious glomerulonephritis.

Over the course of his 5‐day hospital stay, the patient received furosemide with increased urine output and normalization of his serum creatinine to its baseline level of 1.0 mg.dL (88.4 mol/L). Proteinuria resolved to 44 mg/dL. A kidney biopsy was not performed. The parvovirus IgG index, checked because of anemia and oligoarthralgias, was 3.67 (normal 0‐1.10), and the IgM index was 8.13 (normal 0‐1.10), suggesting recent infection. The patient was discharged after 5 days. His edema had resolved on discharge; he continued to be nauseated but was able to eat and drink normally. Six months after his hospitalization, his symptoms had completely resolved.

Parvovirus! It could cause the pulmonary infection and the gastroenteric symptoms. Parvovirus usually causes more anemia than nausea and vomiting. We see it occasionally in our transplant patients. The underlying diabetic nephropathy may have made him more symptomatic with a superimposed glomerulonephritis. The most important pedagogic point is that he did well with a very conservative approach, and the possible iatrogenic consequences of a kidney biopsy, had it been performed, were avoided.

COMMENTARY

Parvovirus B19 is endemic, with as many as 80% of adults showing serologic evidence of past infection. Although most adults with detectable B19‐specific IgG do not recall having had specific symptoms, a number of syndromes have been associated with acute infection.1, 2 Parvovirus B19 should be included in the differential for postinfectious glomerulonephritis, especially if a patient presents with marked edema with preserved renal function.

Human parvovirus B19, a member of the erythrovirus genus, is a nonenveloped single‐stranded DNA virus that propagates in erythroid progenitor cells, arresting erythropoiesis.3 The cellular receptor for the virus is globoside (erythrocyte P antigen), a neutral glycosphingolipid densely present on erythroid cells and also found on hepatocytes, nephrons, and bowel mucosa.3, 4

The most common clinical presentation of parvovirus B19 in children is erythema infectiosum, or fifth disease.3 In adults, the infection is known to cause symmetric polyarthropathy, rash, malaise, coryza, headache, and gastrointestinal symptoms (nausea, abdominal pain) and may mimic systemic lupus erythematosus.1, 3 In patients with sickle cell anemia or other chronic hemolytic disorders, parvovirus B19 can cause a transient aplastic crisis.3 Immunosuppressed patients (eg, organ transplant recipients, patients with certain cancers or advanced AIDS) may develop chronic infection and anemia because of an inability to mount an immune response to clear viremia. Mild anemia or pancytopenia is frequently observed in normal infected hosts.

The syndrome of renal involvement in parvovirus B19 includes the typical features of fever, a maculopapular or reticular erythematous rash on the face or extremities, and polyarthritis, accompanied by oliguria that leads to systemic edema. Mild pancytopenia, proteinuria, hematuria, and hypocomplementemia are often present. Creatinine is usually normal or near normal. These symptoms typically appear 1‐2 weeks after the initial viral syndrome.5, 6 With supportive care, most recover spontaneously, although chronic kidney disease has been reported.7, 8

Published kidney biopsy findings of parvovirus B19 show endocapillary or mesangial proliferative glomerulonephritis with subendothelial electron‐dense deposits and granular deposition of C3, IgG, or IgM along the capillary walls and mesangium. These lesions suggest immune complex deposition and are consistent with postinfectious glomerulonephritis.5, 9, 10 Indeed, increased levels of circulating immune complexes have been seen during acute parvovirus B19 infection.6, 9 It is likely that the protracted symptoms our patient experienced resulted from the formation, circulation, and deposition of immune complexes. The presence of globoside in the kidneys and bowel also raises the possibility of direct infection of these organs.

Postinfectious glomerulonephritis is often thought to be synonymous with poststreptococcal glomerulonephritis. However, viruses, including hepatitis B and C viruses, human immunodeficiency virus, cytomegalovirus, hantavirus, and parvovirus B19 may cause postinfectious glomerulonephritis. As with poststreptococcal glomerulonephritis, glomerular disease associated with viral infection appears to be mediated by the immune complexes. The pathogenic series of events leading to glomerular injury includes formation of circulating immune complexes with subsequent deposition in the glomerulus, or formation of in situ antigen‐antibody reactions.11 Immune complexes in the glomerulus lead to activation of the complement cascade, which in turn leads to hypocomplementemia, as the complement cascade is activated faster than the synthesis of new complement proteins.12 Histologically, a number of different renal lesions may be seen in postviral glomerulonephritis, including membranous, membranoproliferative, and mesangial glomerulonephritis, as well as focal segmental glomerulosclerosis.

Our patient presented with symptoms compatible with but not specific for parvovirus B19. Using a pattern recognition approach to diagnosis, our discussant correctly identified the disease pattern as a postinfectious glomerulonephritis but was unable to identify the correct pathogen, as bacterial infections were the main focus of concern, and viruses, parvovirus B19 in particular, were not considered. The clinical pattern of arthralgia, gastrointestinal symptoms, fever combined with anemia or pancytopenia, and hypocomplementemia is typical of the clues for parvovirus B19. Although renal involvement is unusual, the presence of oliguria, hematuria, and edema with minimal creatinine elevation is typical of parvovirus renal disease.

An essential part of clinical judgment is carefully determining which of a patient's often myriad complaints must be considered part of the disease process. Common and nonspecific signs and symptoms often fall off the clinician's radar screen. In this instance, several of the hallmark features of parvovirus B19 disease were dismissed by our discussant as due to the patient's previous medical conditions or hospital‐related interventions. Anemia (due to interruption of erythropoiesis by parvovirus B19 replication) was attributed to hydration or phlebotomy, fluid retention was attributed to advancing diabetes, and hematuria was attributed to a urinary catheter. It is important to evaluate the entire clinical picture prior to excluding potential clues to the diagnosis. Another reasonable approach would have been to choose a less general sign or symptom to narrow the possible diagnoses. For example, had the wrist arthralgia been more central in the discussant's thoughts, parvovirus B19 might have appeared on the differential.

Finally, the discussant wrestled with the decision to perform a renal biopsy for a definitive diagnosis versus the potential complications of the procedure. In this case, it was possible to achieve a clinical diagnosis, support it with serologic evidence, and thus avoid the need for biopsy. The current medical climate emphasizes the importance of reaching a definitive diagnosis as rapidly as possible. There are pressures to act quickly and utilize technology that may add both cost and risk. This case emphasizes the value of clinical reasoning and patience, which led to a correct diagnosis and a favorable outcome without the need for invasive procedures. Clinical acumen must occasionally include avoiding the temptation to perform the next test and merely standing at the patient's bedside instead.

A 35‐year‐old man presented to the emergency department of a community hospital with 3 days of nausea, vomiting, abdominal pain, and diarrhea. He had developed a sore throat, nasal congestion, and green sputum on a business trip to Las Vegas 10 days prior. He then traveled to Shanghai, China, where he developed frequent diarrhea with mucous and urgency. The stool was mustard colored without blood or melena. He became nauseated and unable to keep down food or fluids. He noted a 4‐kg weight loss since the beginning of his symptoms. He had not traveled outside Shanghai or eaten exotic foods. His travel companions remained unaffected. The patient had had type I diabetes mellitus for 23 years with known retinopathy and microalbuminuria. He had hypertension and hyperlipidemia but was otherwise healthy. He was a married traveling salesman with 3 healthy children. He did not smoke or drink and reported no drug use or extramarital relations. He had no known allergies. His medications included insulin, lisinopril, atorvastatin, and valsartan. He was afebrile, his vital signs were stable and the physical examination was unremarkable.

The patient presents with gastrointestinal symptoms following a trip to China. He may have an infection that began in the respiratory system and now is causing some gastrointestinal symptoms, like Legionnella. If he had received antibiotics, the diarrheal illness could be a complication. With his recent travel to China, typical enteric pathogens would have to be considered: enterotoxigenic E. coli, Shigella, Salmonella, Campylobacter, or perhaps Giardia.

He was admitted and treated with intravenous fluids and ciprofloxacin for presumed gastroenteritis. He was discharged the next day but returned 2 days later because of continued nausea and vomiting and limited oral intake. He was febrile, to 38.9C, with chills. The results of an abdominal exam were normal. Stool studies for Salmonella, Shigella, Campylobacter, Yersinia, enterotoxigenic E. coli, Giardia, and Cryptosporidium were negative. No blood parasites were seen. The white‐cell count was 3900/mm³ with a normal differential count. Hemoglobin level, platelet count, serum electrolytes and creatinine were normal. He was readmitted for intravenous fluids.

The patient was treated for traveler's diarrhea, although this is not a typical case and is now becoming a protracted illness. Amebiasis would be a consideration, as well as typhoid or perhaps an abdominal abscess. The normal platelet count reduces the likelihood of hemolytic uremic syndrome, as does the absence of bloody diarrhea. I would evaluate for a systemic illness: check for adenopathy, do a thorough abdominal exam, and get a chest radiogram, blood cultures, and liver function tests. I am also concerned about metabolic abnormalities that could occur as a consequence of the diarrhea.

Over 3 days, oliguria developed along with urinary hesitancy, a 9‐kg weight gain, and the development of marked edema. Blood pressure and heart rate remained normal, and a chest radiograph was clear. Liver function tests were normal. A urinary catheter was inserted. Urinalysis revealed a specific gravity of 1.031, protein of 100 mg/dL, and trace glucose but was otherwise negative; no casts or cells were seen in the sediment. Chemistries included sodium of 133 mmol/L, potassium of 3.9 mmol/L, and serum bicarbonate of 18.4 mmol/L. Blood and urine cultures were sterile. The creatinine increased from 1.0 mg/dL (88.4 mol/L) to 1.3 mg/dL (115 mol/L). He was transferred to a tertiary care hospital for renal consultation because of concerns of impending renal failure and for consideration of a kidney biopsy.

In a typical case of a malabsorptive diarrhea, the patient could be volume depleted, but in this case he has gained 9‐kg and is grossly edematous. The chest radiograph and liver tests point to renal rather than cardiac or hepatologic causes for the edema. A glomerulonephritis may be driving the salt and water retention.

The proteinuria could be related to hemodynamics, or it could be from a glomerular lesion secondary to immune complexes. The specific gravity of 1.031 indicates the kidney is able to concentrate, and we are not seeing acute tubular necrosis. There is only minimal elevation in creatinine at this point. Quantitation or estimation of the degree of proteinuria by a protein‐to‐creatinine ratio would be helpful.

A further workup should include additional blood cultures and a CT scan of the lungs and abdomen to look for occult infection. Is he unfortunate enough to have developed a malignancy? Is this a connective tissue disease? Reexamination of the urine sediment is important to evaluate for glomerulonephritis.

The patient reported ongoing nausea and vomiting, but his diarrhea resolved. He was tachypneic, with a respiratory rate of 26 breaths/minute and an oxygen saturation of 98% breathing ambient air. His temperature was 37.1C, heart rate 84 beats/minute, and blood pressure 120/65 mm Hg. His mucous membranes were moist, and his jugular venous pressure was 6 cm. No lymphadenopathy was present. The heart and lungs were normal. The abdomen was soft, nontender, and without organomegaly, masses, or shifting dullness; bowel sounds were hypoactive. Severe edema of his legs, sacrum, hands, arms, and orbits was noted. His right hand and wrist were painful with limited mobility and small joint effusions of the wrist and metacarpophalangeal joints, but without erythema or warmth. Small petechiae were noted on his eyelids; skin examination was otherwise unremarkable. He had been given ciprofloxacin, phenergan, calcium carbonate, and pantoprazole prior to transfer. Stool was negative for occult blood.

His lungs are clear, but it is possible to have early pulmonary congestion with normal breath sounds. As he is normotensive and has a normal JVP, I would not give further intravenous fluid. Unless he has evidence of symptomatic pulmonary edema, I would not give diuretics but would simply observe his course. At this point I would ultrasound his kidneys to make sure there is no obstruction. The proteinuria could be a result of underlying diabetic glomerulopathy that may predispose to fluid retention.

He has significant oliguria but only a mild rise in creatinine. As the referring physicians requested a biopsy, we should consider it. A decision to biopsy the kidney would rest on the degree of proteinuria and the activity of the sediment. For example, if red blood cell casts or dysmorphic red cells were present, postinfectious glomerulonephritis or IgA nephropathy would be more likely. However, if the proteinuria is in the non‐nephrotic range and the sediment is nonreactive, the yield of a biopsy would be low.

His right hand and wrist are painful with limited mobility. This could be a sequela of endocarditis, although the absence of a murmur and negative blood cultures make it unlikely. He could have an infectious arthritis, although this usually presents more dramatically. He could have gout or pseudogout, which could be determined by joint aspiration. Finally, could this be iatrogenic? A drug reaction could explain some of the features, including rash, fever, joint symptoms, and renal abnormalities.

Repeat urinalysis revealed protein of more than 300 mg/dL, hematuria (1+), mucous, renal tubular epithelial cells, renal tubular epithelial cell casts, and granular casts. Eosinophiluria was absent. Laboratory evaluation revealed a hemoglobin level of 10.7 g/dL (decreased from 13 g/dL on initial presentation), white‐cell count of 7300/mm³, platelet count of 337,000/mm³, serum creatinine of 1.2 mg/dL (106 mol/L), blood urea nitrogen of 23 mg/dL (8.2 mol.L), total serum protein of 6.3 g/dL (normal range, 6.3‐8.7), and albumin of 2.7 g/dL (normal range, 3.2‐5.2). Other liver tests and serum electrolytes were normal.

This degree of proteinuria is significant, but it is unclear whether this is related to the underlying disease process or his advancing diabetes. He has some hematuria, but that could be from the urinary catheter. It would be helpful to know if the red blood cells are dysmorphic, which would point to a glomerulonephritis. He has renal cells, renal cell casts, and granular casts, which are nonspecific. He has a mild anemia, which is unexplained, but could relate to phlebotomy or overhydration. The hypoalbuminemia may be a result of renal losses or a catabolic state.

A renal ultrasonogram was normal, apart from evidence of bilateral pleural effusions. Antinuclear antibody and rheumatoid factor test results were negative, as were those for hepatitis A, hepatitis B surface antigen, and hepatitis C antibodies. Antistreptolysin O and antideoxyribonuclease B titers were normal. Total complement Ch50 was low at 29 U/mL (normal range, 30‐75) as was complement factor C3 at 67 mg/dL (normal range, 90‐180). Complement factor C4 was normal. Serum and urine electrophoresis revealed no monoclonal protein spike. Vitamin B12 and serum folate were normal, serum ferritin was 584 ng/mL (normal range, 30‐400), iron serum was 29 g/dL (normal range, 45‐160), transferrin saturation was 16%, and total iron‐binding capacity was 164 g/dL (normal range, 250‐450). The reticulocyte count was 2.9% with an absolute reticulocyte count of 102/cm³ and a reticulocyte production index of 0.96 (normal range, 1.0‐2.0).

It is reassuring that his urinary tract ultrasound is normal. In addition to edema, he has bilateral effusions, which are probably transudative, related to fluid overload. The urinalysis does not suggest a rapidly progressive glomerulonephritis, but autoimmune disease is still in the differential.

He has a mild complement C3 deficiency. In nephrology we think of lupus, infective endocarditis, cryoglobulinemia, and specific glomerular lesions such as membranoproliferative glomerulonephritis and postinfectious glomerulonephritis as being associated with the development of circulating immune complexes that may lead to low complement levels. There is no evidence of a paraprotein, but testing for cryoglobulins should be considered. Cryoglobulins are associated with hepatitis C but may be induced by a variety of infections. Acting like immune complexes, they can lead to low complement levels and could cause some of this patient's symptoms. However, this whole illness seems most likely to be secondary to infection. The normal antistreptolysin O and antideoxyribonuclease B titers make streptococcal disease unlikely, but another bacterial infection could cause postinfectious glomerulonephritis.

Over the course of his 5‐day hospital stay, the patient received furosemide with increased urine output and normalization of his serum creatinine to its baseline level of 1.0 mg.dL (88.4 mol/L). Proteinuria resolved to 44 mg/dL. A kidney biopsy was not performed. The parvovirus IgG index, checked because of anemia and oligoarthralgias, was 3.67 (normal 0‐1.10), and the IgM index was 8.13 (normal 0‐1.10), suggesting recent infection. The patient was discharged after 5 days. His edema had resolved on discharge; he continued to be nauseated but was able to eat and drink normally. Six months after his hospitalization, his symptoms had completely resolved.

Parvovirus! It could cause the pulmonary infection and the gastroenteric symptoms. Parvovirus usually causes more anemia than nausea and vomiting. We see it occasionally in our transplant patients. The underlying diabetic nephropathy may have made him more symptomatic with a superimposed glomerulonephritis. The most important pedagogic point is that he did well with a very conservative approach, and the possible iatrogenic consequences of a kidney biopsy, had it been performed, were avoided.

COMMENTARY

Parvovirus B19 is endemic, with as many as 80% of adults showing serologic evidence of past infection. Although most adults with detectable B19‐specific IgG do not recall having had specific symptoms, a number of syndromes have been associated with acute infection.1, 2 Parvovirus B19 should be included in the differential for postinfectious glomerulonephritis, especially if a patient presents with marked edema with preserved renal function.

Human parvovirus B19, a member of the erythrovirus genus, is a nonenveloped single‐stranded DNA virus that propagates in erythroid progenitor cells, arresting erythropoiesis.3 The cellular receptor for the virus is globoside (erythrocyte P antigen), a neutral glycosphingolipid densely present on erythroid cells and also found on hepatocytes, nephrons, and bowel mucosa.3, 4

The most common clinical presentation of parvovirus B19 in children is erythema infectiosum, or fifth disease.3 In adults, the infection is known to cause symmetric polyarthropathy, rash, malaise, coryza, headache, and gastrointestinal symptoms (nausea, abdominal pain) and may mimic systemic lupus erythematosus.1, 3 In patients with sickle cell anemia or other chronic hemolytic disorders, parvovirus B19 can cause a transient aplastic crisis.3 Immunosuppressed patients (eg, organ transplant recipients, patients with certain cancers or advanced AIDS) may develop chronic infection and anemia because of an inability to mount an immune response to clear viremia. Mild anemia or pancytopenia is frequently observed in normal infected hosts.

The syndrome of renal involvement in parvovirus B19 includes the typical features of fever, a maculopapular or reticular erythematous rash on the face or extremities, and polyarthritis, accompanied by oliguria that leads to systemic edema. Mild pancytopenia, proteinuria, hematuria, and hypocomplementemia are often present. Creatinine is usually normal or near normal. These symptoms typically appear 1‐2 weeks after the initial viral syndrome.5, 6 With supportive care, most recover spontaneously, although chronic kidney disease has been reported.7, 8

Published kidney biopsy findings of parvovirus B19 show endocapillary or mesangial proliferative glomerulonephritis with subendothelial electron‐dense deposits and granular deposition of C3, IgG, or IgM along the capillary walls and mesangium. These lesions suggest immune complex deposition and are consistent with postinfectious glomerulonephritis.5, 9, 10 Indeed, increased levels of circulating immune complexes have been seen during acute parvovirus B19 infection.6, 9 It is likely that the protracted symptoms our patient experienced resulted from the formation, circulation, and deposition of immune complexes. The presence of globoside in the kidneys and bowel also raises the possibility of direct infection of these organs.

Postinfectious glomerulonephritis is often thought to be synonymous with poststreptococcal glomerulonephritis. However, viruses, including hepatitis B and C viruses, human immunodeficiency virus, cytomegalovirus, hantavirus, and parvovirus B19 may cause postinfectious glomerulonephritis. As with poststreptococcal glomerulonephritis, glomerular disease associated with viral infection appears to be mediated by the immune complexes. The pathogenic series of events leading to glomerular injury includes formation of circulating immune complexes with subsequent deposition in the glomerulus, or formation of in situ antigen‐antibody reactions.11 Immune complexes in the glomerulus lead to activation of the complement cascade, which in turn leads to hypocomplementemia, as the complement cascade is activated faster than the synthesis of new complement proteins.12 Histologically, a number of different renal lesions may be seen in postviral glomerulonephritis, including membranous, membranoproliferative, and mesangial glomerulonephritis, as well as focal segmental glomerulosclerosis.

Our patient presented with symptoms compatible with but not specific for parvovirus B19. Using a pattern recognition approach to diagnosis, our discussant correctly identified the disease pattern as a postinfectious glomerulonephritis but was unable to identify the correct pathogen, as bacterial infections were the main focus of concern, and viruses, parvovirus B19 in particular, were not considered. The clinical pattern of arthralgia, gastrointestinal symptoms, fever combined with anemia or pancytopenia, and hypocomplementemia is typical of the clues for parvovirus B19. Although renal involvement is unusual, the presence of oliguria, hematuria, and edema with minimal creatinine elevation is typical of parvovirus renal disease.

An essential part of clinical judgment is carefully determining which of a patient's often myriad complaints must be considered part of the disease process. Common and nonspecific signs and symptoms often fall off the clinician's radar screen. In this instance, several of the hallmark features of parvovirus B19 disease were dismissed by our discussant as due to the patient's previous medical conditions or hospital‐related interventions. Anemia (due to interruption of erythropoiesis by parvovirus B19 replication) was attributed to hydration or phlebotomy, fluid retention was attributed to advancing diabetes, and hematuria was attributed to a urinary catheter. It is important to evaluate the entire clinical picture prior to excluding potential clues to the diagnosis. Another reasonable approach would have been to choose a less general sign or symptom to narrow the possible diagnoses. For example, had the wrist arthralgia been more central in the discussant's thoughts, parvovirus B19 might have appeared on the differential.

Finally, the discussant wrestled with the decision to perform a renal biopsy for a definitive diagnosis versus the potential complications of the procedure. In this case, it was possible to achieve a clinical diagnosis, support it with serologic evidence, and thus avoid the need for biopsy. The current medical climate emphasizes the importance of reaching a definitive diagnosis as rapidly as possible. There are pressures to act quickly and utilize technology that may add both cost and risk. This case emphasizes the value of clinical reasoning and patience, which led to a correct diagnosis and a favorable outcome without the need for invasive procedures. Clinical acumen must occasionally include avoiding the temptation to perform the next test and merely standing at the patient's bedside instead.

A 35‐year‐old man presented to the emergency department of a community hospital with 3 days of nausea, vomiting, abdominal pain, and diarrhea. He had developed a sore throat, nasal congestion, and green sputum on a business trip to Las Vegas 10 days prior. He then traveled to Shanghai, China, where he developed frequent diarrhea with mucous and urgency. The stool was mustard colored without blood or melena. He became nauseated and unable to keep down food or fluids. He noted a 4‐kg weight loss since the beginning of his symptoms. He had not traveled outside Shanghai or eaten exotic foods. His travel companions remained unaffected. The patient had had type I diabetes mellitus for 23 years with known retinopathy and microalbuminuria. He had hypertension and hyperlipidemia but was otherwise healthy. He was a married traveling salesman with 3 healthy children. He did not smoke or drink and reported no drug use or extramarital relations. He had no known allergies. His medications included insulin, lisinopril, atorvastatin, and valsartan. He was afebrile, his vital signs were stable and the physical examination was unremarkable.

The patient presents with gastrointestinal symptoms following a trip to China. He may have an infection that began in the respiratory system and now is causing some gastrointestinal symptoms, like Legionnella. If he had received antibiotics, the diarrheal illness could be a complication. With his recent travel to China, typical enteric pathogens would have to be considered: enterotoxigenic E. coli, Shigella, Salmonella, Campylobacter, or perhaps Giardia.

He was admitted and treated with intravenous fluids and ciprofloxacin for presumed gastroenteritis. He was discharged the next day but returned 2 days later because of continued nausea and vomiting and limited oral intake. He was febrile, to 38.9C, with chills. The results of an abdominal exam were normal. Stool studies for Salmonella, Shigella, Campylobacter, Yersinia, enterotoxigenic E. coli, Giardia, and Cryptosporidium were negative. No blood parasites were seen. The white‐cell count was 3900/mm³ with a normal differential count. Hemoglobin level, platelet count, serum electrolytes and creatinine were normal. He was readmitted for intravenous fluids.

The patient was treated for traveler's diarrhea, although this is not a typical case and is now becoming a protracted illness. Amebiasis would be a consideration, as well as typhoid or perhaps an abdominal abscess. The normal platelet count reduces the likelihood of hemolytic uremic syndrome, as does the absence of bloody diarrhea. I would evaluate for a systemic illness: check for adenopathy, do a thorough abdominal exam, and get a chest radiogram, blood cultures, and liver function tests. I am also concerned about metabolic abnormalities that could occur as a consequence of the diarrhea.

Over 3 days, oliguria developed along with urinary hesitancy, a 9‐kg weight gain, and the development of marked edema. Blood pressure and heart rate remained normal, and a chest radiograph was clear. Liver function tests were normal. A urinary catheter was inserted. Urinalysis revealed a specific gravity of 1.031, protein of 100 mg/dL, and trace glucose but was otherwise negative; no casts or cells were seen in the sediment. Chemistries included sodium of 133 mmol/L, potassium of 3.9 mmol/L, and serum bicarbonate of 18.4 mmol/L. Blood and urine cultures were sterile. The creatinine increased from 1.0 mg/dL (88.4 mol/L) to 1.3 mg/dL (115 mol/L). He was transferred to a tertiary care hospital for renal consultation because of concerns of impending renal failure and for consideration of a kidney biopsy.

In a typical case of a malabsorptive diarrhea, the patient could be volume depleted, but in this case he has gained 9‐kg and is grossly edematous. The chest radiograph and liver tests point to renal rather than cardiac or hepatologic causes for the edema. A glomerulonephritis may be driving the salt and water retention.

The proteinuria could be related to hemodynamics, or it could be from a glomerular lesion secondary to immune complexes. The specific gravity of 1.031 indicates the kidney is able to concentrate, and we are not seeing acute tubular necrosis. There is only minimal elevation in creatinine at this point. Quantitation or estimation of the degree of proteinuria by a protein‐to‐creatinine ratio would be helpful.

A further workup should include additional blood cultures and a CT scan of the lungs and abdomen to look for occult infection. Is he unfortunate enough to have developed a malignancy? Is this a connective tissue disease? Reexamination of the urine sediment is important to evaluate for glomerulonephritis.

The patient reported ongoing nausea and vomiting, but his diarrhea resolved. He was tachypneic, with a respiratory rate of 26 breaths/minute and an oxygen saturation of 98% breathing ambient air. His temperature was 37.1C, heart rate 84 beats/minute, and blood pressure 120/65 mm Hg. His mucous membranes were moist, and his jugular venous pressure was 6 cm. No lymphadenopathy was present. The heart and lungs were normal. The abdomen was soft, nontender, and without organomegaly, masses, or shifting dullness; bowel sounds were hypoactive. Severe edema of his legs, sacrum, hands, arms, and orbits was noted. His right hand and wrist were painful with limited mobility and small joint effusions of the wrist and metacarpophalangeal joints, but without erythema or warmth. Small petechiae were noted on his eyelids; skin examination was otherwise unremarkable. He had been given ciprofloxacin, phenergan, calcium carbonate, and pantoprazole prior to transfer. Stool was negative for occult blood.

His lungs are clear, but it is possible to have early pulmonary congestion with normal breath sounds. As he is normotensive and has a normal JVP, I would not give further intravenous fluid. Unless he has evidence of symptomatic pulmonary edema, I would not give diuretics but would simply observe his course. At this point I would ultrasound his kidneys to make sure there is no obstruction. The proteinuria could be a result of underlying diabetic glomerulopathy that may predispose to fluid retention.

He has significant oliguria but only a mild rise in creatinine. As the referring physicians requested a biopsy, we should consider it. A decision to biopsy the kidney would rest on the degree of proteinuria and the activity of the sediment. For example, if red blood cell casts or dysmorphic red cells were present, postinfectious glomerulonephritis or IgA nephropathy would be more likely. However, if the proteinuria is in the non‐nephrotic range and the sediment is nonreactive, the yield of a biopsy would be low.

His right hand and wrist are painful with limited mobility. This could be a sequela of endocarditis, although the absence of a murmur and negative blood cultures make it unlikely. He could have an infectious arthritis, although this usually presents more dramatically. He could have gout or pseudogout, which could be determined by joint aspiration. Finally, could this be iatrogenic? A drug reaction could explain some of the features, including rash, fever, joint symptoms, and renal abnormalities.

Repeat urinalysis revealed protein of more than 300 mg/dL, hematuria (1+), mucous, renal tubular epithelial cells, renal tubular epithelial cell casts, and granular casts. Eosinophiluria was absent. Laboratory evaluation revealed a hemoglobin level of 10.7 g/dL (decreased from 13 g/dL on initial presentation), white‐cell count of 7300/mm³, platelet count of 337,000/mm³, serum creatinine of 1.2 mg/dL (106 mol/L), blood urea nitrogen of 23 mg/dL (8.2 mol.L), total serum protein of 6.3 g/dL (normal range, 6.3‐8.7), and albumin of 2.7 g/dL (normal range, 3.2‐5.2). Other liver tests and serum electrolytes were normal.

This degree of proteinuria is significant, but it is unclear whether this is related to the underlying disease process or his advancing diabetes. He has some hematuria, but that could be from the urinary catheter. It would be helpful to know if the red blood cells are dysmorphic, which would point to a glomerulonephritis. He has renal cells, renal cell casts, and granular casts, which are nonspecific. He has a mild anemia, which is unexplained, but could relate to phlebotomy or overhydration. The hypoalbuminemia may be a result of renal losses or a catabolic state.

A renal ultrasonogram was normal, apart from evidence of bilateral pleural effusions. Antinuclear antibody and rheumatoid factor test results were negative, as were those for hepatitis A, hepatitis B surface antigen, and hepatitis C antibodies. Antistreptolysin O and antideoxyribonuclease B titers were normal. Total complement Ch50 was low at 29 U/mL (normal range, 30‐75) as was complement factor C3 at 67 mg/dL (normal range, 90‐180). Complement factor C4 was normal. Serum and urine electrophoresis revealed no monoclonal protein spike. Vitamin B12 and serum folate were normal, serum ferritin was 584 ng/mL (normal range, 30‐400), iron serum was 29 g/dL (normal range, 45‐160), transferrin saturation was 16%, and total iron‐binding capacity was 164 g/dL (normal range, 250‐450). The reticulocyte count was 2.9% with an absolute reticulocyte count of 102/cm³ and a reticulocyte production index of 0.96 (normal range, 1.0‐2.0).

It is reassuring that his urinary tract ultrasound is normal. In addition to edema, he has bilateral effusions, which are probably transudative, related to fluid overload. The urinalysis does not suggest a rapidly progressive glomerulonephritis, but autoimmune disease is still in the differential.

He has a mild complement C3 deficiency. In nephrology we think of lupus, infective endocarditis, cryoglobulinemia, and specific glomerular lesions such as membranoproliferative glomerulonephritis and postinfectious glomerulonephritis as being associated with the development of circulating immune complexes that may lead to low complement levels. There is no evidence of a paraprotein, but testing for cryoglobulins should be considered. Cryoglobulins are associated with hepatitis C but may be induced by a variety of infections. Acting like immune complexes, they can lead to low complement levels and could cause some of this patient's symptoms. However, this whole illness seems most likely to be secondary to infection. The normal antistreptolysin O and antideoxyribonuclease B titers make streptococcal disease unlikely, but another bacterial infection could cause postinfectious glomerulonephritis.

Over the course of his 5‐day hospital stay, the patient received furosemide with increased urine output and normalization of his serum creatinine to its baseline level of 1.0 mg.dL (88.4 mol/L). Proteinuria resolved to 44 mg/dL. A kidney biopsy was not performed. The parvovirus IgG index, checked because of anemia and oligoarthralgias, was 3.67 (normal 0‐1.10), and the IgM index was 8.13 (normal 0‐1.10), suggesting recent infection. The patient was discharged after 5 days. His edema had resolved on discharge; he continued to be nauseated but was able to eat and drink normally. Six months after his hospitalization, his symptoms had completely resolved.

Parvovirus! It could cause the pulmonary infection and the gastroenteric symptoms. Parvovirus usually causes more anemia than nausea and vomiting. We see it occasionally in our transplant patients. The underlying diabetic nephropathy may have made him more symptomatic with a superimposed glomerulonephritis. The most important pedagogic point is that he did well with a very conservative approach, and the possible iatrogenic consequences of a kidney biopsy, had it been performed, were avoided.

COMMENTARY

Parvovirus B19 is endemic, with as many as 80% of adults showing serologic evidence of past infection. Although most adults with detectable B19‐specific IgG do not recall having had specific symptoms, a number of syndromes have been associated with acute infection.1, 2 Parvovirus B19 should be included in the differential for postinfectious glomerulonephritis, especially if a patient presents with marked edema with preserved renal function.

Human parvovirus B19, a member of the erythrovirus genus, is a nonenveloped single‐stranded DNA virus that propagates in erythroid progenitor cells, arresting erythropoiesis.3 The cellular receptor for the virus is globoside (erythrocyte P antigen), a neutral glycosphingolipid densely present on erythroid cells and also found on hepatocytes, nephrons, and bowel mucosa.3, 4

The most common clinical presentation of parvovirus B19 in children is erythema infectiosum, or fifth disease.3 In adults, the infection is known to cause symmetric polyarthropathy, rash, malaise, coryza, headache, and gastrointestinal symptoms (nausea, abdominal pain) and may mimic systemic lupus erythematosus.1, 3 In patients with sickle cell anemia or other chronic hemolytic disorders, parvovirus B19 can cause a transient aplastic crisis.3 Immunosuppressed patients (eg, organ transplant recipients, patients with certain cancers or advanced AIDS) may develop chronic infection and anemia because of an inability to mount an immune response to clear viremia. Mild anemia or pancytopenia is frequently observed in normal infected hosts.

The syndrome of renal involvement in parvovirus B19 includes the typical features of fever, a maculopapular or reticular erythematous rash on the face or extremities, and polyarthritis, accompanied by oliguria that leads to systemic edema. Mild pancytopenia, proteinuria, hematuria, and hypocomplementemia are often present. Creatinine is usually normal or near normal. These symptoms typically appear 1‐2 weeks after the initial viral syndrome.5, 6 With supportive care, most recover spontaneously, although chronic kidney disease has been reported.7, 8

Published kidney biopsy findings of parvovirus B19 show endocapillary or mesangial proliferative glomerulonephritis with subendothelial electron‐dense deposits and granular deposition of C3, IgG, or IgM along the capillary walls and mesangium. These lesions suggest immune complex deposition and are consistent with postinfectious glomerulonephritis.5, 9, 10 Indeed, increased levels of circulating immune complexes have been seen during acute parvovirus B19 infection.6, 9 It is likely that the protracted symptoms our patient experienced resulted from the formation, circulation, and deposition of immune complexes. The presence of globoside in the kidneys and bowel also raises the possibility of direct infection of these organs.

Postinfectious glomerulonephritis is often thought to be synonymous with poststreptococcal glomerulonephritis. However, viruses, including hepatitis B and C viruses, human immunodeficiency virus, cytomegalovirus, hantavirus, and parvovirus B19 may cause postinfectious glomerulonephritis. As with poststreptococcal glomerulonephritis, glomerular disease associated with viral infection appears to be mediated by the immune complexes. The pathogenic series of events leading to glomerular injury includes formation of circulating immune complexes with subsequent deposition in the glomerulus, or formation of in situ antigen‐antibody reactions.11 Immune complexes in the glomerulus lead to activation of the complement cascade, which in turn leads to hypocomplementemia, as the complement cascade is activated faster than the synthesis of new complement proteins.12 Histologically, a number of different renal lesions may be seen in postviral glomerulonephritis, including membranous, membranoproliferative, and mesangial glomerulonephritis, as well as focal segmental glomerulosclerosis.

Our patient presented with symptoms compatible with but not specific for parvovirus B19. Using a pattern recognition approach to diagnosis, our discussant correctly identified the disease pattern as a postinfectious glomerulonephritis but was unable to identify the correct pathogen, as bacterial infections were the main focus of concern, and viruses, parvovirus B19 in particular, were not considered. The clinical pattern of arthralgia, gastrointestinal symptoms, fever combined with anemia or pancytopenia, and hypocomplementemia is typical of the clues for parvovirus B19. Although renal involvement is unusual, the presence of oliguria, hematuria, and edema with minimal creatinine elevation is typical of parvovirus renal disease.

An essential part of clinical judgment is carefully determining which of a patient's often myriad complaints must be considered part of the disease process. Common and nonspecific signs and symptoms often fall off the clinician's radar screen. In this instance, several of the hallmark features of parvovirus B19 disease were dismissed by our discussant as due to the patient's previous medical conditions or hospital‐related interventions. Anemia (due to interruption of erythropoiesis by parvovirus B19 replication) was attributed to hydration or phlebotomy, fluid retention was attributed to advancing diabetes, and hematuria was attributed to a urinary catheter. It is important to evaluate the entire clinical picture prior to excluding potential clues to the diagnosis. Another reasonable approach would have been to choose a less general sign or symptom to narrow the possible diagnoses. For example, had the wrist arthralgia been more central in the discussant's thoughts, parvovirus B19 might have appeared on the differential.

Finally, the discussant wrestled with the decision to perform a renal biopsy for a definitive diagnosis versus the potential complications of the procedure. In this case, it was possible to achieve a clinical diagnosis, support it with serologic evidence, and thus avoid the need for biopsy. The current medical climate emphasizes the importance of reaching a definitive diagnosis as rapidly as possible. There are pressures to act quickly and utilize technology that may add both cost and risk. This case emphasizes the value of clinical reasoning and patience, which led to a correct diagnosis and a favorable outcome without the need for invasive procedures. Clinical acumen must occasionally include avoiding the temptation to perform the next test and merely standing at the patient's bedside instead.