The field of hospital medicine, now the fastest growing specialty in medical history,[1] was born out of pressure to improve the efficiency and quality of clinical care in US hospitals.[2] Delivering safe and high‐value clinical care is a central goal of the field and has been an essential component of its growth and success.

The clinical demands on academic hospitalists have grown recently, fueled by the need to staff services previously covered by housestaff, whose hours are now restricted. Despite these new demands, expectations have grown in other arenas as well. Academic hospitalist groups (AHGs) are often expected to make significant contributions in quality improvement, patient safety, education, research, and administration. With broad expectations beyond clinical care, AHGs face unique challenges. Groups that focus mainly on providing coverage and improving clinical performance may find that they are unable to fully contribute in these other domains. To be successful, AHGs must develop strategies that balance their energies, resources, and performance.

The balanced scorecard (BSC) was introduced by Kaplan and Norton in 1992 to allow corporations to view their performance broadly, rather than narrowly focusing on financial measures. The BSC requires organizations to develop a balanced portfolio of performance metrics across 4 key perspectives: financial, customers, internal processes, and learning and growth. Metrics within these perspectives should help answer fundamental questions about the organization (Table 1).[3] Over time, the BSC evolved from a performance measurement tool to a strategic management system.[4] Successful organizations translate their mission and vision to specific strategic objectives in each of the 4 perspectives, delineate how these objectives will help the organization reach its vision with a strategy map,[5] and then utilize the BSC to track and monitor performance to ensure that the vision is achieved.[6]

Although originally conceived for businesses, the BSC has found its way into the healthcare industry, with reports of successful implementation in organizations ranging from individual departments to research collaboratives[7] to national healthcare systems.[8] However, there are few reports of BSC implementation in academic health centers.[9, 10] Because most academic centers are not‐for‐profit, Zelman suggests that the 4 BSC perspectives be modified to better fit their unique characteristics (Table 1).[11] To the best of our knowledge, there is no literature describing the development of a BSC in an academic hospitalist group. In this article, we describe the development of, and early experiences with, an academic hospital medicine BSC developed as part of a strategic planning initiative.

METHODS

The University of California, San Francisco (UCSF) Division of Hospital Medicine (DHM) was established in 2005. Currently, there are more than 50 faculty members, having doubled in the last 4 years. In addition to staffing several housestaff and nonhousestaff clinical services, faculty are involved in a wide variety of nonclinical endeavors at local and national levels. They participate and lead initiatives in education, faculty development, patient safety, care efficiency, quality improvement, information technology, and global health. There is an active research enterprise that generates nearly $5 million in grant funding annually.

RESULTS

There were 41 initial metrics considered by the division BSC task force (Table 3). Of these, 16 were chosen for the initial BSC through the modified Delphi method. Over the past 2 years, these initial metrics have been modified to reflect current strategic goals and objectives. Figure 1 illustrates the BSC for fiscal year (FY) 2012. An online version of this, complete with graphical representations of the data and metric definitions, can be found at http://hospitalmedicine.ucsf.edu/bsc/fy2012.html. Our strategy map (Figure 2) demonstrates how these metrics are interconnected across the 4 BSC perspectives and how they fit into our overall strategic plan.

Figure 1

Figure 2

DISCUSSION

Like many hospitalist groups, our division has experienced tremendous growth, both in our numbers and the breadth of roles that we fill. With this growth has come increasing expectations in multiple domains, competing priorities, and limited resources. We successfully developed a BSC as a tool to help our division reach its vision: balancing high quality clinical care, education, academics, and faculty development while maintaining a strong sense of community. We have found that the BSC has helped us meet several key goals.

The first goal was to allow for a broad view of our performance. This is the BSC's most basic function, and we saw immediate and tangible benefits. The scorecard provided a broad snapshot of our performance in a single place. For example, in the clinical domain, we saw that our direct cost per case was increasing despite our adjusted average length of stay remaining stable from FY2010‐FY2011. In academics and research, we saw that the number of abstracts accepted at national meetings increased by almost 30% in FY2011 (Figure 1).

The second goal was to create transparency and accountability. By measuring performance and displaying it on the division Web site, the BSC has promoted transparency. If performance does not meet our targets, the division as a whole becomes accountable. Leadership must understand why performance fell short and initiate changes to improve it. For instance, the rising direct cost per case has spurred the development of a high‐value care committee tasked with finding ways of reducing cost while providing high‐quality care.[12]

The third goal was to communicate goals and engage our faculty. As our division has grown, ensuring a shared vision among our entire faculty became an increasing challenge. The BSC functions as a communication platform between leadership and faculty, and yielded multiple benefits. As the metrics were born out of our mission and vision, the BSC has become a tangible representation of our core values. Moreover, individual faculty can see that they are part of a greater, high‐performing organization and realize they can impact the group's performance through their individual effort. For example, this has helped promote receptivity to carefully disseminated individual performance measures for billing and documentation, and patient satisfaction, in conjunction with faculty development in these areas.

The fourth goal was to ensure that we use data to guide strategic decisions. We felt that strategic decisions needed to be based on objective, rather than perceived or anecdotal, information. This meant translating our vision into measurable objectives that would drive performance improvement. For example, before the BSC, we were committed to the dissemination of our research and innovations. Yet, we quickly realized that we did not have a system to collect even basic data on academic performancea deficit we filled by leveraging information gathered from online databases and faculty curricula vitae. These data allowed us, for the first time, to objectively reflect on this as a strategic goal and to have an ongoing mechanism to monitor academic productivity.

Disclosure

Nothing to report.

The field of hospital medicine, now the fastest growing specialty in medical history,[1] was born out of pressure to improve the efficiency and quality of clinical care in US hospitals.[2] Delivering safe and high‐value clinical care is a central goal of the field and has been an essential component of its growth and success.

The clinical demands on academic hospitalists have grown recently, fueled by the need to staff services previously covered by housestaff, whose hours are now restricted. Despite these new demands, expectations have grown in other arenas as well. Academic hospitalist groups (AHGs) are often expected to make significant contributions in quality improvement, patient safety, education, research, and administration. With broad expectations beyond clinical care, AHGs face unique challenges. Groups that focus mainly on providing coverage and improving clinical performance may find that they are unable to fully contribute in these other domains. To be successful, AHGs must develop strategies that balance their energies, resources, and performance.

The balanced scorecard (BSC) was introduced by Kaplan and Norton in 1992 to allow corporations to view their performance broadly, rather than narrowly focusing on financial measures. The BSC requires organizations to develop a balanced portfolio of performance metrics across 4 key perspectives: financial, customers, internal processes, and learning and growth. Metrics within these perspectives should help answer fundamental questions about the organization (Table 1).[3] Over time, the BSC evolved from a performance measurement tool to a strategic management system.[4] Successful organizations translate their mission and vision to specific strategic objectives in each of the 4 perspectives, delineate how these objectives will help the organization reach its vision with a strategy map,[5] and then utilize the BSC to track and monitor performance to ensure that the vision is achieved.[6]

Although originally conceived for businesses, the BSC has found its way into the healthcare industry, with reports of successful implementation in organizations ranging from individual departments to research collaboratives[7] to national healthcare systems.[8] However, there are few reports of BSC implementation in academic health centers.[9, 10] Because most academic centers are not‐for‐profit, Zelman suggests that the 4 BSC perspectives be modified to better fit their unique characteristics (Table 1).[11] To the best of our knowledge, there is no literature describing the development of a BSC in an academic hospitalist group. In this article, we describe the development of, and early experiences with, an academic hospital medicine BSC developed as part of a strategic planning initiative.

METHODS

The University of California, San Francisco (UCSF) Division of Hospital Medicine (DHM) was established in 2005. Currently, there are more than 50 faculty members, having doubled in the last 4 years. In addition to staffing several housestaff and nonhousestaff clinical services, faculty are involved in a wide variety of nonclinical endeavors at local and national levels. They participate and lead initiatives in education, faculty development, patient safety, care efficiency, quality improvement, information technology, and global health. There is an active research enterprise that generates nearly $5 million in grant funding annually.

RESULTS

There were 41 initial metrics considered by the division BSC task force (Table 3). Of these, 16 were chosen for the initial BSC through the modified Delphi method. Over the past 2 years, these initial metrics have been modified to reflect current strategic goals and objectives. Figure 1 illustrates the BSC for fiscal year (FY) 2012. An online version of this, complete with graphical representations of the data and metric definitions, can be found at http://hospitalmedicine.ucsf.edu/bsc/fy2012.html. Our strategy map (Figure 2) demonstrates how these metrics are interconnected across the 4 BSC perspectives and how they fit into our overall strategic plan.

Figure 1

Figure 2

DISCUSSION

Like many hospitalist groups, our division has experienced tremendous growth, both in our numbers and the breadth of roles that we fill. With this growth has come increasing expectations in multiple domains, competing priorities, and limited resources. We successfully developed a BSC as a tool to help our division reach its vision: balancing high quality clinical care, education, academics, and faculty development while maintaining a strong sense of community. We have found that the BSC has helped us meet several key goals.

The first goal was to allow for a broad view of our performance. This is the BSC's most basic function, and we saw immediate and tangible benefits. The scorecard provided a broad snapshot of our performance in a single place. For example, in the clinical domain, we saw that our direct cost per case was increasing despite our adjusted average length of stay remaining stable from FY2010‐FY2011. In academics and research, we saw that the number of abstracts accepted at national meetings increased by almost 30% in FY2011 (Figure 1).

The second goal was to create transparency and accountability. By measuring performance and displaying it on the division Web site, the BSC has promoted transparency. If performance does not meet our targets, the division as a whole becomes accountable. Leadership must understand why performance fell short and initiate changes to improve it. For instance, the rising direct cost per case has spurred the development of a high‐value care committee tasked with finding ways of reducing cost while providing high‐quality care.[12]

The third goal was to communicate goals and engage our faculty. As our division has grown, ensuring a shared vision among our entire faculty became an increasing challenge. The BSC functions as a communication platform between leadership and faculty, and yielded multiple benefits. As the metrics were born out of our mission and vision, the BSC has become a tangible representation of our core values. Moreover, individual faculty can see that they are part of a greater, high‐performing organization and realize they can impact the group's performance through their individual effort. For example, this has helped promote receptivity to carefully disseminated individual performance measures for billing and documentation, and patient satisfaction, in conjunction with faculty development in these areas.

The fourth goal was to ensure that we use data to guide strategic decisions. We felt that strategic decisions needed to be based on objective, rather than perceived or anecdotal, information. This meant translating our vision into measurable objectives that would drive performance improvement. For example, before the BSC, we were committed to the dissemination of our research and innovations. Yet, we quickly realized that we did not have a system to collect even basic data on academic performancea deficit we filled by leveraging information gathered from online databases and faculty curricula vitae. These data allowed us, for the first time, to objectively reflect on this as a strategic goal and to have an ongoing mechanism to monitor academic productivity.

Disclosure

Nothing to report.

The field of hospital medicine, now the fastest growing specialty in medical history,[1] was born out of pressure to improve the efficiency and quality of clinical care in US hospitals.[2] Delivering safe and high‐value clinical care is a central goal of the field and has been an essential component of its growth and success.

The clinical demands on academic hospitalists have grown recently, fueled by the need to staff services previously covered by housestaff, whose hours are now restricted. Despite these new demands, expectations have grown in other arenas as well. Academic hospitalist groups (AHGs) are often expected to make significant contributions in quality improvement, patient safety, education, research, and administration. With broad expectations beyond clinical care, AHGs face unique challenges. Groups that focus mainly on providing coverage and improving clinical performance may find that they are unable to fully contribute in these other domains. To be successful, AHGs must develop strategies that balance their energies, resources, and performance.

The balanced scorecard (BSC) was introduced by Kaplan and Norton in 1992 to allow corporations to view their performance broadly, rather than narrowly focusing on financial measures. The BSC requires organizations to develop a balanced portfolio of performance metrics across 4 key perspectives: financial, customers, internal processes, and learning and growth. Metrics within these perspectives should help answer fundamental questions about the organization (Table 1).[3] Over time, the BSC evolved from a performance measurement tool to a strategic management system.[4] Successful organizations translate their mission and vision to specific strategic objectives in each of the 4 perspectives, delineate how these objectives will help the organization reach its vision with a strategy map,[5] and then utilize the BSC to track and monitor performance to ensure that the vision is achieved.[6]

Although originally conceived for businesses, the BSC has found its way into the healthcare industry, with reports of successful implementation in organizations ranging from individual departments to research collaboratives[7] to national healthcare systems.[8] However, there are few reports of BSC implementation in academic health centers.[9, 10] Because most academic centers are not‐for‐profit, Zelman suggests that the 4 BSC perspectives be modified to better fit their unique characteristics (Table 1).[11] To the best of our knowledge, there is no literature describing the development of a BSC in an academic hospitalist group. In this article, we describe the development of, and early experiences with, an academic hospital medicine BSC developed as part of a strategic planning initiative.

METHODS

The University of California, San Francisco (UCSF) Division of Hospital Medicine (DHM) was established in 2005. Currently, there are more than 50 faculty members, having doubled in the last 4 years. In addition to staffing several housestaff and nonhousestaff clinical services, faculty are involved in a wide variety of nonclinical endeavors at local and national levels. They participate and lead initiatives in education, faculty development, patient safety, care efficiency, quality improvement, information technology, and global health. There is an active research enterprise that generates nearly $5 million in grant funding annually.

RESULTS

There were 41 initial metrics considered by the division BSC task force (Table 3). Of these, 16 were chosen for the initial BSC through the modified Delphi method. Over the past 2 years, these initial metrics have been modified to reflect current strategic goals and objectives. Figure 1 illustrates the BSC for fiscal year (FY) 2012. An online version of this, complete with graphical representations of the data and metric definitions, can be found at http://hospitalmedicine.ucsf.edu/bsc/fy2012.html. Our strategy map (Figure 2) demonstrates how these metrics are interconnected across the 4 BSC perspectives and how they fit into our overall strategic plan.

Figure 1

Figure 2

DISCUSSION

Like many hospitalist groups, our division has experienced tremendous growth, both in our numbers and the breadth of roles that we fill. With this growth has come increasing expectations in multiple domains, competing priorities, and limited resources. We successfully developed a BSC as a tool to help our division reach its vision: balancing high quality clinical care, education, academics, and faculty development while maintaining a strong sense of community. We have found that the BSC has helped us meet several key goals.

The first goal was to allow for a broad view of our performance. This is the BSC's most basic function, and we saw immediate and tangible benefits. The scorecard provided a broad snapshot of our performance in a single place. For example, in the clinical domain, we saw that our direct cost per case was increasing despite our adjusted average length of stay remaining stable from FY2010‐FY2011. In academics and research, we saw that the number of abstracts accepted at national meetings increased by almost 30% in FY2011 (Figure 1).

The second goal was to create transparency and accountability. By measuring performance and displaying it on the division Web site, the BSC has promoted transparency. If performance does not meet our targets, the division as a whole becomes accountable. Leadership must understand why performance fell short and initiate changes to improve it. For instance, the rising direct cost per case has spurred the development of a high‐value care committee tasked with finding ways of reducing cost while providing high‐quality care.[12]

The third goal was to communicate goals and engage our faculty. As our division has grown, ensuring a shared vision among our entire faculty became an increasing challenge. The BSC functions as a communication platform between leadership and faculty, and yielded multiple benefits. As the metrics were born out of our mission and vision, the BSC has become a tangible representation of our core values. Moreover, individual faculty can see that they are part of a greater, high‐performing organization and realize they can impact the group's performance through their individual effort. For example, this has helped promote receptivity to carefully disseminated individual performance measures for billing and documentation, and patient satisfaction, in conjunction with faculty development in these areas.

The fourth goal was to ensure that we use data to guide strategic decisions. We felt that strategic decisions needed to be based on objective, rather than perceived or anecdotal, information. This meant translating our vision into measurable objectives that would drive performance improvement. For example, before the BSC, we were committed to the dissemination of our research and innovations. Yet, we quickly realized that we did not have a system to collect even basic data on academic performancea deficit we filled by leveraging information gathered from online databases and faculty curricula vitae. These data allowed us, for the first time, to objectively reflect on this as a strategic goal and to have an ongoing mechanism to monitor academic productivity.

Disclosure

Nothing to report.

Clinical communication among healthcare providers to coordinate patient care is important, accounting for the majority of information exchanges in healthcare.[1, 2] Breakdowns in communication have therefore been identified as the major contributor to medical errors.[3, 4, 5]

There is a growing literature related to asynchronous clinical communication practices, or communication that does not occur at the same time in hospitals, and the limitations of using the traditional numeric pager. These include the inability to indicate the urgency of the message, frequent interruptions, contacting the wrong physician, and inefficiencies coordinating care across multiple disciplines and specialties.[6, 7, 8, 9, 10, 11, 12]

Hospitals have implemented a variety of health information technology (HIT) solutions to replace the numeric pager and address these clinical communication issues, including the use of alphanumeric pagers, smartphone devices, and Web‐based applications that allow clinicians to triage the urgency of issues.[13, 14, 15, 16] Although these solutions have resolved some of the deficiencies previously identified, issues relating to the impact on the interprofessional nature of healthcare remain unaddressed.[17] In some cases, the implementation of HIT has created unintended consequences that have an impact on effective communication.

One of the widely cited examples of HIT creating unintended consequences is the implementation of computerized physician order entry systems.[18, 19, 20] Other studies looking more broadly at patient information systems have identified problems caused by poor user interfaces that promoted errors in entry and retrieval of data, inflexible features forcing clinician workarounds, and technology designs that impeded clinical workflow.[21, 22, 23]

These observations suggest that although many of the issues with clinical communication stem from the reliance on numeric paging, simply replacing pagers with newer technology may not solve the problems and can in fact create other unintended consequences. These include unintended consequences resulting from the sociotechnical aspects of HIT, which is the interplay of technology with existing clinical workflow, culture, and social interactions.[21] Our institution recently implemented a Web‐based messaging system to replace the use of numeric pagers. We aimed to evaluate the unintended consequences resulting from the implementation of this system and to describe their impact on the delivery of clinical care on a general internal medicine (GIM) service.

METHODS

This was a pre‐post mixed‐methods study utilizing both quantitative and qualitative measures. We integrated these 2 data‐collection methods to improve the quality of the results. The study was conducted on the GIM service at the University Health Network Toronto Western Hospital site, a tertiary‐care academic teaching center fully affiliated with the University of Toronto (Toronto, Canada). The GIM service at Toronto Western Hospital consists of 4 clinical teams, each staffed by an attending physician, 3 to 4 residents, and 2 to 3 medical students.

Prior to this study, nonface‐to‐face communication on the wards was facilitated through numeric paging, where nurses, pharmacists, and social workers on the GIM wards would page residents to a hospital phone and wait for them to call back. Figures 1, 2, and 3 visualize the Web‐based messaging system we implemented at the University Health Network in May 2010. All residents on the service were provided smartphones that they used for communication. In addition to these, there was a dedicated team smartphone that acted as a central point of contact for the team 24 hours a day, 7 days a week and was carried by the physician covering the team at the time. The system allowed nurses, pharmacists, and social workers to triage the urgency of messages and include details providing context to the issue. Issues flagged as urgent were immediately sent to the team smartphone by e‐mail to alert the physician, who could respond from the smartphone. These messages could be forwarded to a team member, often the physician most familiar with the patient, to address. Issues flagged as nonurgent were posted to the system's message board, which the physicians accessed by logging into the Web‐based messaging system on a regular basis. The message board was designed to allow physicians to respond to multiple non‐urgent issues at once. To close the loop on communication, logic was developed so that if a physician did not respond to a non‐urgent message within the specified timeframe, the message was escalated and sent as an alerting e‐mail to the team smartphone every 15 minutes until it was addressed. The timeframes for responding varied from 1 hour to not needing a response until the next morning.

Figure 1

Figure 2

Figure 3

Ongoing training on the use of the system was built into the clinical orientation for the physician and nursing staff, as turnover in an academic teaching hospital is quite high. The orientation included instructions on how to use the features of the Web‐based messaging system but also provided guidelines on general etiquette with using the system to ensure sustainability of the communication process. For example, physicians were asked to check the system regularly, as the process worked only if they responded to messages. Nurses were asked not to send messages during scheduled educational sessions unless necessary and to limit messages they flagged as urgent to ones that were in fact urgent.

Our quantitative evaluation compared interruptions, which we define as communication that caused a medical resident to stop current activity to address, before and after the implementation of the Web‐based messaging system to assess the volume and time distribution of messages, and compared these results with our qualitative evaluation. For the pre‐implementation phase, interruptions were all numeric pages sent to all residents during the period of July 1427, 2008. For the post‐implementation phase, interruptions were the e‐mails sent directly to the team smartphones from the Web‐based messaging system to all residents during the period of October 1124, 2010. We excluded messages from the postimplementation phase if the same message was sent >10 times, typically indicating technical issues such as a malfunction of the smartphone causing the escalation process to continue.

Our qualitative evaluation consisted of semistructured interviews that were conducted after implementation. A research coordinator sent e‐mails to potential physician participants during a 1‐month rotation (n=16), and nurses (N=50), pharmacists (N=4), and social workers (N=4) from a representative ward inviting them to be part of this study. A set of open‐ended questions (see Supporting Information, Appendix A, in the online version of this article) developed based on informal feedback regarding the system provided by physicians, nurses, pharmacists, and social workers served as a guide to highlight key themes of interest. Based on the participants' responses, further questions were asked to drill down into more detail. Interviews were transcribed verbatim and anonymized.

The interview data were analyzed using thematic analysis to generate categories and overarching themes.[24, 25] Once the coding structure was developed, the transcripts were imported into qualitative analysis software (NVivo 9, QSR International) and then coded and analyzed, pulling the key themes that emerged from the text to be used in interpreting the data.

RESULTS

Our quantitative before‐after comparison of clinical messages sent to physicians revealed an increase in interruptions. We compared these results to the results of our interviews to understand why this might have occurred. Several key themes emerged from the analysis of the interviews, including increase in interruptions, accountability, and tactics to improve personal productivity. We interviewed 5 physicians, 8 nurses, 2 pharmacists, and 2 social workers.

DISCUSSION

The purpose of this study was to evaluate a Web‐based messaging system and identify the unintended consequences observed with implementing HIT to improve clinical communication. This is an important study because healthcare organizations are beginning to develop strategies for improving clinical communication but believe the solution involves simply replacing pager technology. Support for this approach is seen with larger vendors in the smartphone and communication industry, who promote their products as pager‐replacement solutions and even help customers develop pager‐replacement strategies.[26, 27, 28] Simply replacing pagers with smartphones and sending text messages will have only a limited impact on improving clinical communication and will likely result in unintended negative consequences, as seen in this study.

Whereas the Web‐based messaging system was designed to reduce interruptions from clinical messaging, interruptions actually increased, although the mental burden of each interruption was likely lower because responding to a text message is less interruptive than finding a telephone to answer a page. A key contributor to this effect was a culture of accountability among nurses, pharmacists, and social workers who felt it was their professional obligation to notify physicians about all issues of concern. This belief and related behavior is aligned with the standards promoted by professional regulatory bodies that identify accountability as a vital practice expectation.[29] Nurses and nursing staff take responsibility for the care they provide and answer for their own judgments and actions.[30] The system eliminated many of the previous barriers to paging and provided a less‐personal form of communication. The cumulative and unexpected outcome was an increase of interruptions for physicians and the adoption of workarounds by all healthcare professionals to improve personal productivity. Although the system was built in an iterative fashion with frontline clinicians, it is likely that oversights in the design of the system also contributed to these problems, which speaks to the complexity of clinical communication. Centralizing communication to the team smartphone could have overburdened the physicians covering it at the time, causing them to ignore messages because they were too busy to address them.

There were limitations to this study. One limitation was that this study examined only a cross‐section of messaging activity at a given point in time, and therefore it may not be representative of the behaviors of the physicians, nurses, pharmacists, and social workers over time as the culture of the environment evolves and they adjust to the new technology. The pre‐implementation data were collected 2 years prior to the post‐implementation data, but it was necessary to use data this old because other interventions were implemented prior to the Web‐based messaging system, so baseline paging data were no longer available. Whereas most clinical disciplines were represented in the interviews, the sample included only 17 participants from 1 clinical service, so generalizability of the results may be limited.

Although the reliance on numeric paging technology was previously identified as a primary source of problems with communication, the real issues are much more complex. This study highlighted that many of the underlying obstacles relate to existing social interactions and habits of multiple professions working together. Failures in collaboration among healthcare professionals have a negative impact on health outcomes and routinely stem from the lack of explicit definitions of roles, the absence of clear leadership, insufficient time for team‐building, the us‐and‐them effects created by professional socialization, and frustration created by power and status differentials of each discipline.[31, 32, 33] Therefore, it is critical that healthcare organizations focus on the people and clinical processes when implementing technology to solve issues with clinical communication. These observations are consistent with other studies examining the unintended consequences caused by the sociotechnical aspects of HIT implementation, where workarounds to game the system were also employed.[21]

In summary, improving clinical communication cannot be achieved simply by replacing pagers with newer technology; it requires a fundamental shift in how healthcare professionals interact, with a focus on the sociotechnical aspects of HIT. As patient volumes and the complexity of care continue to increase, more effective methods for facilitating interprofessional communication and collaboration must be developed.

Clinical communication among healthcare providers to coordinate patient care is important, accounting for the majority of information exchanges in healthcare.[1, 2] Breakdowns in communication have therefore been identified as the major contributor to medical errors.[3, 4, 5]

There is a growing literature related to asynchronous clinical communication practices, or communication that does not occur at the same time in hospitals, and the limitations of using the traditional numeric pager. These include the inability to indicate the urgency of the message, frequent interruptions, contacting the wrong physician, and inefficiencies coordinating care across multiple disciplines and specialties.[6, 7, 8, 9, 10, 11, 12]

Hospitals have implemented a variety of health information technology (HIT) solutions to replace the numeric pager and address these clinical communication issues, including the use of alphanumeric pagers, smartphone devices, and Web‐based applications that allow clinicians to triage the urgency of issues.[13, 14, 15, 16] Although these solutions have resolved some of the deficiencies previously identified, issues relating to the impact on the interprofessional nature of healthcare remain unaddressed.[17] In some cases, the implementation of HIT has created unintended consequences that have an impact on effective communication.

One of the widely cited examples of HIT creating unintended consequences is the implementation of computerized physician order entry systems.[18, 19, 20] Other studies looking more broadly at patient information systems have identified problems caused by poor user interfaces that promoted errors in entry and retrieval of data, inflexible features forcing clinician workarounds, and technology designs that impeded clinical workflow.[21, 22, 23]

These observations suggest that although many of the issues with clinical communication stem from the reliance on numeric paging, simply replacing pagers with newer technology may not solve the problems and can in fact create other unintended consequences. These include unintended consequences resulting from the sociotechnical aspects of HIT, which is the interplay of technology with existing clinical workflow, culture, and social interactions.[21] Our institution recently implemented a Web‐based messaging system to replace the use of numeric pagers. We aimed to evaluate the unintended consequences resulting from the implementation of this system and to describe their impact on the delivery of clinical care on a general internal medicine (GIM) service.

METHODS

This was a pre‐post mixed‐methods study utilizing both quantitative and qualitative measures. We integrated these 2 data‐collection methods to improve the quality of the results. The study was conducted on the GIM service at the University Health Network Toronto Western Hospital site, a tertiary‐care academic teaching center fully affiliated with the University of Toronto (Toronto, Canada). The GIM service at Toronto Western Hospital consists of 4 clinical teams, each staffed by an attending physician, 3 to 4 residents, and 2 to 3 medical students.

Prior to this study, nonface‐to‐face communication on the wards was facilitated through numeric paging, where nurses, pharmacists, and social workers on the GIM wards would page residents to a hospital phone and wait for them to call back. Figures 1, 2, and 3 visualize the Web‐based messaging system we implemented at the University Health Network in May 2010. All residents on the service were provided smartphones that they used for communication. In addition to these, there was a dedicated team smartphone that acted as a central point of contact for the team 24 hours a day, 7 days a week and was carried by the physician covering the team at the time. The system allowed nurses, pharmacists, and social workers to triage the urgency of messages and include details providing context to the issue. Issues flagged as urgent were immediately sent to the team smartphone by e‐mail to alert the physician, who could respond from the smartphone. These messages could be forwarded to a team member, often the physician most familiar with the patient, to address. Issues flagged as nonurgent were posted to the system's message board, which the physicians accessed by logging into the Web‐based messaging system on a regular basis. The message board was designed to allow physicians to respond to multiple non‐urgent issues at once. To close the loop on communication, logic was developed so that if a physician did not respond to a non‐urgent message within the specified timeframe, the message was escalated and sent as an alerting e‐mail to the team smartphone every 15 minutes until it was addressed. The timeframes for responding varied from 1 hour to not needing a response until the next morning.

Figure 1

Figure 2

Figure 3

Ongoing training on the use of the system was built into the clinical orientation for the physician and nursing staff, as turnover in an academic teaching hospital is quite high. The orientation included instructions on how to use the features of the Web‐based messaging system but also provided guidelines on general etiquette with using the system to ensure sustainability of the communication process. For example, physicians were asked to check the system regularly, as the process worked only if they responded to messages. Nurses were asked not to send messages during scheduled educational sessions unless necessary and to limit messages they flagged as urgent to ones that were in fact urgent.

Our quantitative evaluation compared interruptions, which we define as communication that caused a medical resident to stop current activity to address, before and after the implementation of the Web‐based messaging system to assess the volume and time distribution of messages, and compared these results with our qualitative evaluation. For the pre‐implementation phase, interruptions were all numeric pages sent to all residents during the period of July 1427, 2008. For the post‐implementation phase, interruptions were the e‐mails sent directly to the team smartphones from the Web‐based messaging system to all residents during the period of October 1124, 2010. We excluded messages from the postimplementation phase if the same message was sent >10 times, typically indicating technical issues such as a malfunction of the smartphone causing the escalation process to continue.

Our qualitative evaluation consisted of semistructured interviews that were conducted after implementation. A research coordinator sent e‐mails to potential physician participants during a 1‐month rotation (n=16), and nurses (N=50), pharmacists (N=4), and social workers (N=4) from a representative ward inviting them to be part of this study. A set of open‐ended questions (see Supporting Information, Appendix A, in the online version of this article) developed based on informal feedback regarding the system provided by physicians, nurses, pharmacists, and social workers served as a guide to highlight key themes of interest. Based on the participants' responses, further questions were asked to drill down into more detail. Interviews were transcribed verbatim and anonymized.

The interview data were analyzed using thematic analysis to generate categories and overarching themes.[24, 25] Once the coding structure was developed, the transcripts were imported into qualitative analysis software (NVivo 9, QSR International) and then coded and analyzed, pulling the key themes that emerged from the text to be used in interpreting the data.

RESULTS

Our quantitative before‐after comparison of clinical messages sent to physicians revealed an increase in interruptions. We compared these results to the results of our interviews to understand why this might have occurred. Several key themes emerged from the analysis of the interviews, including increase in interruptions, accountability, and tactics to improve personal productivity. We interviewed 5 physicians, 8 nurses, 2 pharmacists, and 2 social workers.

DISCUSSION

The purpose of this study was to evaluate a Web‐based messaging system and identify the unintended consequences observed with implementing HIT to improve clinical communication. This is an important study because healthcare organizations are beginning to develop strategies for improving clinical communication but believe the solution involves simply replacing pager technology. Support for this approach is seen with larger vendors in the smartphone and communication industry, who promote their products as pager‐replacement solutions and even help customers develop pager‐replacement strategies.[26, 27, 28] Simply replacing pagers with smartphones and sending text messages will have only a limited impact on improving clinical communication and will likely result in unintended negative consequences, as seen in this study.

Whereas the Web‐based messaging system was designed to reduce interruptions from clinical messaging, interruptions actually increased, although the mental burden of each interruption was likely lower because responding to a text message is less interruptive than finding a telephone to answer a page. A key contributor to this effect was a culture of accountability among nurses, pharmacists, and social workers who felt it was their professional obligation to notify physicians about all issues of concern. This belief and related behavior is aligned with the standards promoted by professional regulatory bodies that identify accountability as a vital practice expectation.[29] Nurses and nursing staff take responsibility for the care they provide and answer for their own judgments and actions.[30] The system eliminated many of the previous barriers to paging and provided a less‐personal form of communication. The cumulative and unexpected outcome was an increase of interruptions for physicians and the adoption of workarounds by all healthcare professionals to improve personal productivity. Although the system was built in an iterative fashion with frontline clinicians, it is likely that oversights in the design of the system also contributed to these problems, which speaks to the complexity of clinical communication. Centralizing communication to the team smartphone could have overburdened the physicians covering it at the time, causing them to ignore messages because they were too busy to address them.

There were limitations to this study. One limitation was that this study examined only a cross‐section of messaging activity at a given point in time, and therefore it may not be representative of the behaviors of the physicians, nurses, pharmacists, and social workers over time as the culture of the environment evolves and they adjust to the new technology. The pre‐implementation data were collected 2 years prior to the post‐implementation data, but it was necessary to use data this old because other interventions were implemented prior to the Web‐based messaging system, so baseline paging data were no longer available. Whereas most clinical disciplines were represented in the interviews, the sample included only 17 participants from 1 clinical service, so generalizability of the results may be limited.

Although the reliance on numeric paging technology was previously identified as a primary source of problems with communication, the real issues are much more complex. This study highlighted that many of the underlying obstacles relate to existing social interactions and habits of multiple professions working together. Failures in collaboration among healthcare professionals have a negative impact on health outcomes and routinely stem from the lack of explicit definitions of roles, the absence of clear leadership, insufficient time for team‐building, the us‐and‐them effects created by professional socialization, and frustration created by power and status differentials of each discipline.[31, 32, 33] Therefore, it is critical that healthcare organizations focus on the people and clinical processes when implementing technology to solve issues with clinical communication. These observations are consistent with other studies examining the unintended consequences caused by the sociotechnical aspects of HIT implementation, where workarounds to game the system were also employed.[21]

In summary, improving clinical communication cannot be achieved simply by replacing pagers with newer technology; it requires a fundamental shift in how healthcare professionals interact, with a focus on the sociotechnical aspects of HIT. As patient volumes and the complexity of care continue to increase, more effective methods for facilitating interprofessional communication and collaboration must be developed.

Clinical communication among healthcare providers to coordinate patient care is important, accounting for the majority of information exchanges in healthcare.[1, 2] Breakdowns in communication have therefore been identified as the major contributor to medical errors.[3, 4, 5]

There is a growing literature related to asynchronous clinical communication practices, or communication that does not occur at the same time in hospitals, and the limitations of using the traditional numeric pager. These include the inability to indicate the urgency of the message, frequent interruptions, contacting the wrong physician, and inefficiencies coordinating care across multiple disciplines and specialties.[6, 7, 8, 9, 10, 11, 12]

Hospitals have implemented a variety of health information technology (HIT) solutions to replace the numeric pager and address these clinical communication issues, including the use of alphanumeric pagers, smartphone devices, and Web‐based applications that allow clinicians to triage the urgency of issues.[13, 14, 15, 16] Although these solutions have resolved some of the deficiencies previously identified, issues relating to the impact on the interprofessional nature of healthcare remain unaddressed.[17] In some cases, the implementation of HIT has created unintended consequences that have an impact on effective communication.

One of the widely cited examples of HIT creating unintended consequences is the implementation of computerized physician order entry systems.[18, 19, 20] Other studies looking more broadly at patient information systems have identified problems caused by poor user interfaces that promoted errors in entry and retrieval of data, inflexible features forcing clinician workarounds, and technology designs that impeded clinical workflow.[21, 22, 23]

These observations suggest that although many of the issues with clinical communication stem from the reliance on numeric paging, simply replacing pagers with newer technology may not solve the problems and can in fact create other unintended consequences. These include unintended consequences resulting from the sociotechnical aspects of HIT, which is the interplay of technology with existing clinical workflow, culture, and social interactions.[21] Our institution recently implemented a Web‐based messaging system to replace the use of numeric pagers. We aimed to evaluate the unintended consequences resulting from the implementation of this system and to describe their impact on the delivery of clinical care on a general internal medicine (GIM) service.

METHODS

This was a pre‐post mixed‐methods study utilizing both quantitative and qualitative measures. We integrated these 2 data‐collection methods to improve the quality of the results. The study was conducted on the GIM service at the University Health Network Toronto Western Hospital site, a tertiary‐care academic teaching center fully affiliated with the University of Toronto (Toronto, Canada). The GIM service at Toronto Western Hospital consists of 4 clinical teams, each staffed by an attending physician, 3 to 4 residents, and 2 to 3 medical students.

Prior to this study, nonface‐to‐face communication on the wards was facilitated through numeric paging, where nurses, pharmacists, and social workers on the GIM wards would page residents to a hospital phone and wait for them to call back. Figures 1, 2, and 3 visualize the Web‐based messaging system we implemented at the University Health Network in May 2010. All residents on the service were provided smartphones that they used for communication. In addition to these, there was a dedicated team smartphone that acted as a central point of contact for the team 24 hours a day, 7 days a week and was carried by the physician covering the team at the time. The system allowed nurses, pharmacists, and social workers to triage the urgency of messages and include details providing context to the issue. Issues flagged as urgent were immediately sent to the team smartphone by e‐mail to alert the physician, who could respond from the smartphone. These messages could be forwarded to a team member, often the physician most familiar with the patient, to address. Issues flagged as nonurgent were posted to the system's message board, which the physicians accessed by logging into the Web‐based messaging system on a regular basis. The message board was designed to allow physicians to respond to multiple non‐urgent issues at once. To close the loop on communication, logic was developed so that if a physician did not respond to a non‐urgent message within the specified timeframe, the message was escalated and sent as an alerting e‐mail to the team smartphone every 15 minutes until it was addressed. The timeframes for responding varied from 1 hour to not needing a response until the next morning.

Figure 1

Figure 2

Figure 3

Ongoing training on the use of the system was built into the clinical orientation for the physician and nursing staff, as turnover in an academic teaching hospital is quite high. The orientation included instructions on how to use the features of the Web‐based messaging system but also provided guidelines on general etiquette with using the system to ensure sustainability of the communication process. For example, physicians were asked to check the system regularly, as the process worked only if they responded to messages. Nurses were asked not to send messages during scheduled educational sessions unless necessary and to limit messages they flagged as urgent to ones that were in fact urgent.

Our quantitative evaluation compared interruptions, which we define as communication that caused a medical resident to stop current activity to address, before and after the implementation of the Web‐based messaging system to assess the volume and time distribution of messages, and compared these results with our qualitative evaluation. For the pre‐implementation phase, interruptions were all numeric pages sent to all residents during the period of July 1427, 2008. For the post‐implementation phase, interruptions were the e‐mails sent directly to the team smartphones from the Web‐based messaging system to all residents during the period of October 1124, 2010. We excluded messages from the postimplementation phase if the same message was sent >10 times, typically indicating technical issues such as a malfunction of the smartphone causing the escalation process to continue.

Our qualitative evaluation consisted of semistructured interviews that were conducted after implementation. A research coordinator sent e‐mails to potential physician participants during a 1‐month rotation (n=16), and nurses (N=50), pharmacists (N=4), and social workers (N=4) from a representative ward inviting them to be part of this study. A set of open‐ended questions (see Supporting Information, Appendix A, in the online version of this article) developed based on informal feedback regarding the system provided by physicians, nurses, pharmacists, and social workers served as a guide to highlight key themes of interest. Based on the participants' responses, further questions were asked to drill down into more detail. Interviews were transcribed verbatim and anonymized.

The interview data were analyzed using thematic analysis to generate categories and overarching themes.[24, 25] Once the coding structure was developed, the transcripts were imported into qualitative analysis software (NVivo 9, QSR International) and then coded and analyzed, pulling the key themes that emerged from the text to be used in interpreting the data.

RESULTS

Our quantitative before‐after comparison of clinical messages sent to physicians revealed an increase in interruptions. We compared these results to the results of our interviews to understand why this might have occurred. Several key themes emerged from the analysis of the interviews, including increase in interruptions, accountability, and tactics to improve personal productivity. We interviewed 5 physicians, 8 nurses, 2 pharmacists, and 2 social workers.

DISCUSSION

The purpose of this study was to evaluate a Web‐based messaging system and identify the unintended consequences observed with implementing HIT to improve clinical communication. This is an important study because healthcare organizations are beginning to develop strategies for improving clinical communication but believe the solution involves simply replacing pager technology. Support for this approach is seen with larger vendors in the smartphone and communication industry, who promote their products as pager‐replacement solutions and even help customers develop pager‐replacement strategies.[26, 27, 28] Simply replacing pagers with smartphones and sending text messages will have only a limited impact on improving clinical communication and will likely result in unintended negative consequences, as seen in this study.

Whereas the Web‐based messaging system was designed to reduce interruptions from clinical messaging, interruptions actually increased, although the mental burden of each interruption was likely lower because responding to a text message is less interruptive than finding a telephone to answer a page. A key contributor to this effect was a culture of accountability among nurses, pharmacists, and social workers who felt it was their professional obligation to notify physicians about all issues of concern. This belief and related behavior is aligned with the standards promoted by professional regulatory bodies that identify accountability as a vital practice expectation.[29] Nurses and nursing staff take responsibility for the care they provide and answer for their own judgments and actions.[30] The system eliminated many of the previous barriers to paging and provided a less‐personal form of communication. The cumulative and unexpected outcome was an increase of interruptions for physicians and the adoption of workarounds by all healthcare professionals to improve personal productivity. Although the system was built in an iterative fashion with frontline clinicians, it is likely that oversights in the design of the system also contributed to these problems, which speaks to the complexity of clinical communication. Centralizing communication to the team smartphone could have overburdened the physicians covering it at the time, causing them to ignore messages because they were too busy to address them.

There were limitations to this study. One limitation was that this study examined only a cross‐section of messaging activity at a given point in time, and therefore it may not be representative of the behaviors of the physicians, nurses, pharmacists, and social workers over time as the culture of the environment evolves and they adjust to the new technology. The pre‐implementation data were collected 2 years prior to the post‐implementation data, but it was necessary to use data this old because other interventions were implemented prior to the Web‐based messaging system, so baseline paging data were no longer available. Whereas most clinical disciplines were represented in the interviews, the sample included only 17 participants from 1 clinical service, so generalizability of the results may be limited.

Although the reliance on numeric paging technology was previously identified as a primary source of problems with communication, the real issues are much more complex. This study highlighted that many of the underlying obstacles relate to existing social interactions and habits of multiple professions working together. Failures in collaboration among healthcare professionals have a negative impact on health outcomes and routinely stem from the lack of explicit definitions of roles, the absence of clear leadership, insufficient time for team‐building, the us‐and‐them effects created by professional socialization, and frustration created by power and status differentials of each discipline.[31, 32, 33] Therefore, it is critical that healthcare organizations focus on the people and clinical processes when implementing technology to solve issues with clinical communication. These observations are consistent with other studies examining the unintended consequences caused by the sociotechnical aspects of HIT implementation, where workarounds to game the system were also employed.[21]

In summary, improving clinical communication cannot be achieved simply by replacing pagers with newer technology; it requires a fundamental shift in how healthcare professionals interact, with a focus on the sociotechnical aspects of HIT. As patient volumes and the complexity of care continue to increase, more effective methods for facilitating interprofessional communication and collaboration must be developed.

Acute respiratory failure (ARF), a common and serious complication in hospitalized patients, may be caused by several conditions including pneumonia, chronic obstructive pulmonary disease (COPD), adult respiratory distress syndrome (ARDS), and congestive heart failure (CHF). Although ARF is conventionally defined by an arterial oxygen tension of <60 mm Hg, an arterial carbon dioxide tension of >45 mm Hg, or both, these thresholds serve as a guide to be used in combination with history and clinical assessment of the patient.[1, 2] Supplemental oxygen and treatment of the underlying cause is the mainstay of therapy for ARF, but in severe cases patients are treated with invasive mechanical ventilation (IMV) or noninvasive ventilation (NIV). ARF is the most frequent reason for admission to the intensive care unit (ICU)[3, 4] and has an in‐hospital mortality rate of 33% to 37% among those who require IMV.[5, 6] The majority of epidemiologic studies of ARF have been limited to patients requiring mechanical ventilation or those admitted to the ICU, and information about the characteristics and outcomes of patients across the full spectrum of severity is much more limited.[5, 7, 8, 9, 10, 11] General improvements in the management of underlying conditions, implementation of more effective ventilation strategies,[12, 13] and increasing use of NIV[14, 15] may have led to better outcomes for patients with ARF, yet empirical evidence of a change in the adjusted mortality rate over time is lacking.

The objective of this study was to provide a broad characterization of the epidemiology of ARF among adults hospitalized in the United States using a large nationally representative database. We sought to evaluate whether incidence, mortality, cost, or ventilation practice associated with ARF in the United States changed over the period of 2001 to 2009.

METHODS

RESULTS

Hospitalization Trends

The number of hospitalizations with an ARF diagnosis code increased at an average annual rate of 11.3% from 1,007,549 (standard deviation [SD] = 19,268) in 2001 to 1,917,910 (SD = 47,558) in 2009. More than two‐thirds of ARF admissions were associated with medical, rather than surgical, conditions (69.5% in 2001 and 71.2% in 2009). The median age, racial make‐up, and gender did not change significantly. Over the study period we observed an increase in ARF‐related hospitalizations in large, urban, teaching hospitals and in hospitals located in the Midwest (Table 1).

Table 1.Hospitalizations With Acute Respiratory Failure in the United States, 2001 to 2009, by Patient and Hospital Characteristics

	2001	2003	2005	2007	2009
NOTE: Abbreviations: ARF, acute respiratory failure; IMV, invasive mechanical ventilation; LOS, length of stay; NIV, noninvasive ventilation; SD, standard deviation; SE standard error. P value for trend <0.01, including all years 20012009, *P value for trend <0.0001, including all years 20012009. Adjusted for sex, age, race, hospital characteristics, and comorbidities. Geometric mean reported, standard errors from Taylor series expansion.
Patient characteristics
All, N (SD)	1,007,549 (19,268)	1,184,928 (25,542)	1,288,594 (30,493)	1,480,270 (32,002)	1,917,910 (47,558)
Age, mean (SE), y	66.6 (0.2)	66.0 (0.2)	66.1 (0.2)	65.8 (0.2)	65.8 (0.2)
Age group, %
1844	11.5	12.0	11.5	11.6	10.9
4564*	26.7	28.9	29.6	30.7	31.7
6584*	50.2	47.8	47.0	45.7	45.3
85+	11.5	11.4	11.9	12.0	12.1
Male*	48.1	48.2	48.6	49.3	49.2
Race
White	75.8	71.9	76.5	71.8	73.4
Black	12.7	13.6	11.2	14.2	12.5
Hispanic	7.2	9.8	7.7	8.5	7.8
Other	4.2	4.7	4.7	5.5	6.3
Primary ARF	20.7	20.9	25.9	26.1	19.9
Secondary ARF	79.3	79.1	74.1	73.9	80.1
Medical*	69.5	69.1	69.9	70.2	71.2
Surgical*	30.5	30.8	30.1	29.8	28.8
Hospital characteristics, %
Number of beds
Small	10.0	10.1	10.5	10.8	11.3
Medium	25.2	25.3	24.6	24.0	22.7
Large	64.7	64.6	64.9	65.2	66.0
Region
South*	18.5	18.5	17.6	17.0	16.3
Midwest	21.4	22.0	23.6	23.2	23.5
Northeast	42.6	41.7	41.4	42.2	42.1
West*	17.5	17.8	17.3	17.6	18.1
Hospital type
Rural	13.6	13.0	11.8	11.0	10.8
Urban nonteaching	45.5	44.5	50.1	45.3	45.7
Urban teaching	40.9	42.5	38.1	43.7	43.6
Patient outcomes
Ventilation strategy
IMV*	48.5	48.4	47.5	46.5	42.1
NIV*	3.8	5.3	6.9	9.4	10.1
IMV or NIV	50.9	51.7	52.1	52.9	49.7
Disposition
Home/home healthcare*	42.1	43.8	42.8	43.4	45.7
Transfer to acute care	5.2	4.7	4.6	4.6	4.4
Nursing facility*	24.4	24.9	27.4	28.6	29.0
Other	0.7	0.8	0.9	0.9	1.0
Adjusted mortality, % (SE)*	27.6 (0.3)	26.4 (0.4)	24.9 (0.4)	22.7 (0.4)	20.6 (0.3)
Adjusted mean, LOS/case, d (SE)*	7.8 (0.1)	7.9 (0.1)	7.7 (0.1)	7.5 (0.1)	7.1 (0.1)
Adjusted mean cost/case, 2009 US$, (SE)	15,818 (251)	16,981 (419)	17,236 (411)	16,941 (436)	15,987 (402)

After adjusting for age and sex, the population incidence of ARF increased from 502 (standard error [SE] = 10) cases per 100,000 in 2001 to 784 (SE = 19) cases per 100,000 in 2009 (a 56% increase, P < 0.0001). Hispanics had the lowest rates of ARF, with both black and white groups having similar rates (Table 2).

Table 2.Cases of Acute Respiratory Failure per 100,000 Population

	2001	2003	2005	2007	2009
NOTE: Data are presented as number per 100,000 population (standard error), standardized to 2000 US Census population. *P value for trend < 0.0001, including all years 2001 to 2009.
All*	502 (10)	569 (12)	595 (14)	627 (14)	784 (19)
Age group
1844*	107 (3)	130 (4)	137 (4)	153 (5)	189 (6)
4564*	422 (9)	500 (12)	521 (13)	580 (14)	739 (19)
6584*	1697 (35)	1863 (42)	1950 (50)	2066 (46)	2578 (69)
85+	3449 (86)	3792 (106)	3981 (120)	3429 (97)	4163 (123)
Sex
Male*	491 (10)	553 (13)	582 (14)	629 (14)	782 (20)
Female*	512 (10)	583 (12)	607 (15)	625 (13)	786 (19)
Race/ethnicity
White*	398 (11)	427 (12)	466 (16)	450 (13)	699 (21)
Black*	423 (27)	513 (33)	432 (26)	574 (38)	738 (37)
Hispanic*	247 (24)	381 (42)	307 (27)	353 (34)	478 (42)
Other*	268 (20)	342 (29)	347 (26)	424 (29)	713 (77)
In‐hospital mortality	140 (3)	148 (3)	146 (3)	140 (3)	154 (4)

The most common etiologies of ARF among medical patients were pneumonia, CHF, ARDS, COPD exacerbation, and sepsis. Over the 9‐year study, the proportion of cases secondary to pneumonia and sepsis rose significantly: from 39% to 46% and 13% to 21%, respectively (Figure 1).

Figure 1

2009 Data Analysis

In 2009 the 1,917,910 hospitalizations with ARF resulted in 381,155 (SD = 8965) deaths and a total inpatient cost of $54 billion. The most common etiologies in patients over 65 years old were pneumonia, CHF, COPD, ARDS, and sepsis. In patients younger than 45 years the most frequent diagnoses were pneumonia, ARDS, sepsis, asthma, drug ingestion, and trauma. Stratified analysis by gender and by age groups showed that mortality rates among men were higher than for women and were highest in patients older than 85 years (Table 3).

Table 3.Characteristics of Hospitalizations by Etiology (Medical, Surgical, Comorbidities, Procedures) in 2009

Disease	Total	Age <45 Years	4565 Years	6584 Years	85+ Years	Male	Female
NOTE: One patient can have more than 1 diagnosis. Abbreviations: AMI, acute myocardial infarction; ARDS, adult respiratory distress syndrome; CHF, congestive heart failure; CI, confidence interval; COPD, chronic obstructive pulmonary disease; IMV, invasive mechanical ventilation; NIV, noninvasive ventilation. *P < 0.0001 for age group. P < 0.0001 for gender. The P values are not from Rao‐Scott 2 test.
Medical
Total, N (%)	1,364,624 (71.2)	144,715 (10.6)	416,922 (30.6)	615,009 (45.1)	187,977 (13.8)	647,894 (47.5)	716,635 (52.5)
Pneumonia, %*	46.1	41.7	42.8	46.9	54.3	48.8	43.7
CHF, %*	36.6	10.4	27.3	43.6	54.8	35.0	38.1
ARDS, %*	16.1	22.9	16.2	14.5	15.9	15.5	16.7
Sepsis, %*	21.2	18.1	21.3	21.3	23.1	22.8	19.8
COPD, %*	25.4	4.2	25.6	32.3	18.3	25.0	25.7
AMI, %*	9.0	2.6	7.1	10.5	13.3	9.3	8.8
Asthma, %*	9.2	18.1	11.6	6.7	5.4	6.2	12.0
Stroke, %*	4.8	2.3	4.1	5.5	6.0	5.0	4.7
Trauma or burns, %*	3.4	5.4	2.9	3.0	4.1	4.3	2.5
Cardiorespiratory arrest, %*	4.1	3.9	4.4	4.1	3.8	4.6	3.7
Drug, %*	3.7	16.6	5.1	0.8	0.3	3.8	3.6
IMV, %*	37.7	54.6	43.7	33.5	24.8	41.1	34.5
NIV, %*	11.9	7.1	11.5	13.0	12.7	11.4	12.3
In‐hospital mortality (CI)	22 (21.322.7)	12.9 (11.913.9)	18.5 (17.619.4)	23.9 (23.024.9)	31.8 (30.633.1)	24.2 (23.325.1)	20.9 (20.121.7)
Surgical
Total, N (%)	552971 (28.8)	64983 (11.8)	190225 (34.4)	254336 (46)	43426 (7.9)	295660 (53.5)	257287 (46.5)
Pneumonia, %*	34.9	33.0	34.0	35.0	40.5	37.1	32.2
CHF, %*	27.2	8.9	21.7	33.3	42.6	26.7	27.7
ARDS, %*	45.5	51.5	45.2	44.7	42.7	45.0	46.1
Sepsis, %*	25.1	22.8	25.4	25.2	26.1	25.4	24.7
COPD, %*	8.2	1.1	7.4	10.8	7.5	8.3	8.1
AMI, %*	16.9	4.9	17.0	19.8	17.9	19.1	14.4
Asthma, %*	6.1	7.6	7.2	5.4	3.6	4.1	8.5
Stroke, %*	8.9	6.6	9.2	9.4	7.2	8.9	8.8
Trauma or burns, %*	12.2	26.5	9.6	9.2	20.3	13.8	10.4
Cardiorespiratory arrest, %*	5.5	4.4	6.0	5.4	5.2	6.1	4.7
Drug, %*	0.5	1.3	0.7	0.2	0.2	0.4	0.6
IMV, %*	52.9	57.1	54.3	51.3	50.0	54.5	51.0
NIV, %*	5.8	3.5	5.5	6.4	6.4	5.6	6.0
In‐hospital mortality, % (CI)	18.6 (17.819.5)	10.7 (9.312.0)	15.5 (14.216.8)	20.8 (19.821.9)	29.4 (27.831.1)	19.0 (18.219.8)	18.3 (17.319.2)

When we examined ventilation practices among medical patients we found that patients older than 85 years, when compared to patients younger than 45 years, were less likely to be treated with IMV (25% vs 55%) and more likely to be treated with NIV (12.7% vs 7%). At the same time, the average cost per case was lowest among patients 85 years and older, and hospital costs per case fell sharply after age 70 years. Costs were considerably higher for those who did not survive during hospitalization, particularly for patients younger than 45 years (Figure 2).

Figure 2

DISCUSSION

In this large population‐based study, we found that the number of hospitalizations associated with a diagnosis of ARF almost doubled over a 9‐year period. In 2009 there were nearly 2 million hospitalizations with ARF in the United States, resulting in approximately 380,000 deaths and inpatient costs of over $54 billion. The population‐adjusted ARF hospitalization rates increased in all age groups, and patients 85 years and older had the highest age‐specific hospitalization rate. Although overall rates of mechanical ventilation (NIV or IMV) remained stable over the 9‐year period, there was an important shift away from IMV (which decreased from 48% in 2001 to 42% in 2009) toward NIV (which increased from 4% in 2001 to 10% in 2009). Overall, there was a significant increase in the number of total deaths despite a decline in adjusted in‐hospital mortality rates. In‐hospital mortality rates decreased for all cases of ARF regardless of ventilation choice.

The findings of this study mirror results of others that have shown that although the incidence of critical care illnesses like sepsis[17, 20, 21, 26] and acute renal failure[27] has increased over the last decade, in‐hospital mortality rates have decreased.[20, 21, 28] Our results also compliment the results of a recent study that looked at hospitalizations for noncardiogenic ARF, which observed a 3.7‐fold increase in the number of cases and a steady decline in case fatality.[11]

Most prior studies addressing the incidence of ARF have included only patients receiving mechanical ventilation. In 1994, the estimated number of cases of ARF requiring IMV was 329,766,[9] which increased to 790,257 in 2005.[10] In our study we found that in 2009, the number of patients with ARF hospitalizations with IMV increased to 806,538. The increase in the overall number of cases with ARF was mainly driven by a surge in cases of sepsis and pneumonia. Our findings are consistent with national trends over time in noncardiogenic ARF[11] and in conditions that predispose patients to ARF such as sepsis[17, 20, 28] and acute renal failure.[27] As the number of claims for ARF doubled and the number of deaths increased, we found that adjusted hospital mortality improved from 27.6% in 2001 to 20.6% in 2009. This decline in hospital mortality was observed among all patients groups, regardless of ventilation choice. The decline in overall case fatality is consistent with prior findings in noncardiogenic ARF,[11] sepsis,[17, 28] and CHF.[29]

There are a number of potential explanations for the reduction in mortality observed over the study period, including improvements in hospital management of the underlying conditions leading to ARF, an increase in the proportion of patients being treated with NIV,[30] and advances in the care of critically ill patients such as the use of low‐tidal volume ventilation.[31, 32] Another contributor may be an increase in the proportion of discharges to nursing facilities, although this change in discharge disposition cannot fully explain our findings. For example, from 2007 to 2009, mortality decreased by 2 percentage points, and nursing home discharges increased by only 0.4 percentage points. Growth and aging of the US population only partially explain the increase we observed in the incidence of ARF, as age‐ and sex‐adjusted population rates increased by 56% from 2001 to 2009. In addition, the NIS captures data on hospital discharges and not individual patients; thus, a patient may have had multiple admissions. Over the last decade adoption of a more intensive practice style has been associated with improved in‐hospital mortality,[33, 34] and although these patients may be living longer they may have multiple readmissions.[35, 36]

We also observed that older patients were less likely to be treated with IMV, had a higher mortality rate, and less expensive care. These results are consistent with other studies and suggest that the intensity of treatment decreases with increasing age, and decisions to withhold or withdraw life‐supporting treatments are more frequent in the elderly.[26, 37] Prior research has shown that severity of illness is more important than age on patients' prognosis,[38, 39] and aggressive treatment strategies are not less cost‐effective when provided to older patients.[40]

Another important finding of this study is the marked increase in the use of NIV paired with a modest reduction in the use of IMV in the treatment of patients with ARF. This finding adds to evidence from other studies, which have similarly reported a dramatic increase in the use of NIV and a decrease in the use of IMV in patients with COPD as well as in ARF of other etiologies.[30, 41]

Our work has several limitations. First, we identified ARF based on ICD‐9‐CM codes and therefore cannot exclude disease misclassification. We did not find any studies in the literature addressing the accuracy and the completeness of ARF coding. However, we employed the same codes used to define ARF as has been used to define organ dysfunction in studies of severe sepsis,[17, 20] and the ICD‐9‐CM codes that we used to identify cases of ARDS have been used in prior studies.[11, 22, 23] Another limitation is that it is not clear to what extent the trends we observed may be due to changes over time in documentation and coding practices. Although this should be considered given the additional reimbursement associated with the diagnosis of ARF, our observation that rates of assisted ventilation have remained almost flat over the 9‐year period of the study suggest that would not wholly account for the rise in ARF. Second, because we did not have access to physiological data such as results of blood gas testing, we could not determine whether the threshold for applying the diagnosis of ARF or for delivering ventilatory support has changed over time. Third, for the purpose of this study we employed a broad definition of ARF, not limiting cases to those requiring mechanical ventilation, and this led to a more heterogeneous cohort including less severe cases of ARF. However, this is not dissimilar to the heterogeneity in disease severity observed among patients who receive a diagnosis of heart failure or acute renal failure. Fourth, survivors of ARF remain at high risk of death in the months after hospitalization,[42] but we assessed only in‐hospital mortality. It is possible that although in‐hospital mortality has improved, 30‐day mortality remained stable. Finally, as the NIS contains only discharge‐level data, we could not distinguish between patients admitted for ARF from those who developed ARF (potentially iatrogenic) after admission.

In summary, over the period of 2001 to 2009, there was a large increase in the number of patients given a diagnosis of ARF and a concomitant reduction in inpatient mortality. Although rates of mechanical ventilation remained relatively constant, there was a significant shift toward greater use of NIV at the expense of IMV.

Acute respiratory failure (ARF), a common and serious complication in hospitalized patients, may be caused by several conditions including pneumonia, chronic obstructive pulmonary disease (COPD), adult respiratory distress syndrome (ARDS), and congestive heart failure (CHF). Although ARF is conventionally defined by an arterial oxygen tension of <60 mm Hg, an arterial carbon dioxide tension of >45 mm Hg, or both, these thresholds serve as a guide to be used in combination with history and clinical assessment of the patient.[1, 2] Supplemental oxygen and treatment of the underlying cause is the mainstay of therapy for ARF, but in severe cases patients are treated with invasive mechanical ventilation (IMV) or noninvasive ventilation (NIV). ARF is the most frequent reason for admission to the intensive care unit (ICU)[3, 4] and has an in‐hospital mortality rate of 33% to 37% among those who require IMV.[5, 6] The majority of epidemiologic studies of ARF have been limited to patients requiring mechanical ventilation or those admitted to the ICU, and information about the characteristics and outcomes of patients across the full spectrum of severity is much more limited.[5, 7, 8, 9, 10, 11] General improvements in the management of underlying conditions, implementation of more effective ventilation strategies,[12, 13] and increasing use of NIV[14, 15] may have led to better outcomes for patients with ARF, yet empirical evidence of a change in the adjusted mortality rate over time is lacking.

The objective of this study was to provide a broad characterization of the epidemiology of ARF among adults hospitalized in the United States using a large nationally representative database. We sought to evaluate whether incidence, mortality, cost, or ventilation practice associated with ARF in the United States changed over the period of 2001 to 2009.

METHODS

RESULTS

Hospitalization Trends

The number of hospitalizations with an ARF diagnosis code increased at an average annual rate of 11.3% from 1,007,549 (standard deviation [SD] = 19,268) in 2001 to 1,917,910 (SD = 47,558) in 2009. More than two‐thirds of ARF admissions were associated with medical, rather than surgical, conditions (69.5% in 2001 and 71.2% in 2009). The median age, racial make‐up, and gender did not change significantly. Over the study period we observed an increase in ARF‐related hospitalizations in large, urban, teaching hospitals and in hospitals located in the Midwest (Table 1).

Table 1.Hospitalizations With Acute Respiratory Failure in the United States, 2001 to 2009, by Patient and Hospital Characteristics

	2001	2003	2005	2007	2009
NOTE: Abbreviations: ARF, acute respiratory failure; IMV, invasive mechanical ventilation; LOS, length of stay; NIV, noninvasive ventilation; SD, standard deviation; SE standard error. P value for trend <0.01, including all years 20012009, *P value for trend <0.0001, including all years 20012009. Adjusted for sex, age, race, hospital characteristics, and comorbidities. Geometric mean reported, standard errors from Taylor series expansion.
Patient characteristics
All, N (SD)	1,007,549 (19,268)	1,184,928 (25,542)	1,288,594 (30,493)	1,480,270 (32,002)	1,917,910 (47,558)
Age, mean (SE), y	66.6 (0.2)	66.0 (0.2)	66.1 (0.2)	65.8 (0.2)	65.8 (0.2)
Age group, %
1844	11.5	12.0	11.5	11.6	10.9
4564*	26.7	28.9	29.6	30.7	31.7
6584*	50.2	47.8	47.0	45.7	45.3
85+	11.5	11.4	11.9	12.0	12.1
Male*	48.1	48.2	48.6	49.3	49.2
Race
White	75.8	71.9	76.5	71.8	73.4
Black	12.7	13.6	11.2	14.2	12.5
Hispanic	7.2	9.8	7.7	8.5	7.8
Other	4.2	4.7	4.7	5.5	6.3
Primary ARF	20.7	20.9	25.9	26.1	19.9
Secondary ARF	79.3	79.1	74.1	73.9	80.1
Medical*	69.5	69.1	69.9	70.2	71.2
Surgical*	30.5	30.8	30.1	29.8	28.8
Hospital characteristics, %
Number of beds
Small	10.0	10.1	10.5	10.8	11.3
Medium	25.2	25.3	24.6	24.0	22.7
Large	64.7	64.6	64.9	65.2	66.0
Region
South*	18.5	18.5	17.6	17.0	16.3
Midwest	21.4	22.0	23.6	23.2	23.5
Northeast	42.6	41.7	41.4	42.2	42.1
West*	17.5	17.8	17.3	17.6	18.1
Hospital type
Rural	13.6	13.0	11.8	11.0	10.8
Urban nonteaching	45.5	44.5	50.1	45.3	45.7
Urban teaching	40.9	42.5	38.1	43.7	43.6
Patient outcomes
Ventilation strategy
IMV*	48.5	48.4	47.5	46.5	42.1
NIV*	3.8	5.3	6.9	9.4	10.1
IMV or NIV	50.9	51.7	52.1	52.9	49.7
Disposition
Home/home healthcare*	42.1	43.8	42.8	43.4	45.7
Transfer to acute care	5.2	4.7	4.6	4.6	4.4
Nursing facility*	24.4	24.9	27.4	28.6	29.0
Other	0.7	0.8	0.9	0.9	1.0
Adjusted mortality, % (SE)*	27.6 (0.3)	26.4 (0.4)	24.9 (0.4)	22.7 (0.4)	20.6 (0.3)
Adjusted mean, LOS/case, d (SE)*	7.8 (0.1)	7.9 (0.1)	7.7 (0.1)	7.5 (0.1)	7.1 (0.1)
Adjusted mean cost/case, 2009 US$, (SE)	15,818 (251)	16,981 (419)	17,236 (411)	16,941 (436)	15,987 (402)

After adjusting for age and sex, the population incidence of ARF increased from 502 (standard error [SE] = 10) cases per 100,000 in 2001 to 784 (SE = 19) cases per 100,000 in 2009 (a 56% increase, P < 0.0001). Hispanics had the lowest rates of ARF, with both black and white groups having similar rates (Table 2).

Table 2.Cases of Acute Respiratory Failure per 100,000 Population

	2001	2003	2005	2007	2009
NOTE: Data are presented as number per 100,000 population (standard error), standardized to 2000 US Census population. *P value for trend < 0.0001, including all years 2001 to 2009.
All*	502 (10)	569 (12)	595 (14)	627 (14)	784 (19)
Age group
1844*	107 (3)	130 (4)	137 (4)	153 (5)	189 (6)
4564*	422 (9)	500 (12)	521 (13)	580 (14)	739 (19)
6584*	1697 (35)	1863 (42)	1950 (50)	2066 (46)	2578 (69)
85+	3449 (86)	3792 (106)	3981 (120)	3429 (97)	4163 (123)
Sex
Male*	491 (10)	553 (13)	582 (14)	629 (14)	782 (20)
Female*	512 (10)	583 (12)	607 (15)	625 (13)	786 (19)
Race/ethnicity
White*	398 (11)	427 (12)	466 (16)	450 (13)	699 (21)
Black*	423 (27)	513 (33)	432 (26)	574 (38)	738 (37)
Hispanic*	247 (24)	381 (42)	307 (27)	353 (34)	478 (42)
Other*	268 (20)	342 (29)	347 (26)	424 (29)	713 (77)
In‐hospital mortality	140 (3)	148 (3)	146 (3)	140 (3)	154 (4)

The most common etiologies of ARF among medical patients were pneumonia, CHF, ARDS, COPD exacerbation, and sepsis. Over the 9‐year study, the proportion of cases secondary to pneumonia and sepsis rose significantly: from 39% to 46% and 13% to 21%, respectively (Figure 1).

Figure 1

2009 Data Analysis

In 2009 the 1,917,910 hospitalizations with ARF resulted in 381,155 (SD = 8965) deaths and a total inpatient cost of $54 billion. The most common etiologies in patients over 65 years old were pneumonia, CHF, COPD, ARDS, and sepsis. In patients younger than 45 years the most frequent diagnoses were pneumonia, ARDS, sepsis, asthma, drug ingestion, and trauma. Stratified analysis by gender and by age groups showed that mortality rates among men were higher than for women and were highest in patients older than 85 years (Table 3).

Table 3.Characteristics of Hospitalizations by Etiology (Medical, Surgical, Comorbidities, Procedures) in 2009

Disease	Total	Age <45 Years	4565 Years	6584 Years	85+ Years	Male	Female
NOTE: One patient can have more than 1 diagnosis. Abbreviations: AMI, acute myocardial infarction; ARDS, adult respiratory distress syndrome; CHF, congestive heart failure; CI, confidence interval; COPD, chronic obstructive pulmonary disease; IMV, invasive mechanical ventilation; NIV, noninvasive ventilation. *P < 0.0001 for age group. P < 0.0001 for gender. The P values are not from Rao‐Scott 2 test.
Medical
Total, N (%)	1,364,624 (71.2)	144,715 (10.6)	416,922 (30.6)	615,009 (45.1)	187,977 (13.8)	647,894 (47.5)	716,635 (52.5)
Pneumonia, %*	46.1	41.7	42.8	46.9	54.3	48.8	43.7
CHF, %*	36.6	10.4	27.3	43.6	54.8	35.0	38.1
ARDS, %*	16.1	22.9	16.2	14.5	15.9	15.5	16.7
Sepsis, %*	21.2	18.1	21.3	21.3	23.1	22.8	19.8
COPD, %*	25.4	4.2	25.6	32.3	18.3	25.0	25.7
AMI, %*	9.0	2.6	7.1	10.5	13.3	9.3	8.8
Asthma, %*	9.2	18.1	11.6	6.7	5.4	6.2	12.0
Stroke, %*	4.8	2.3	4.1	5.5	6.0	5.0	4.7
Trauma or burns, %*	3.4	5.4	2.9	3.0	4.1	4.3	2.5
Cardiorespiratory arrest, %*	4.1	3.9	4.4	4.1	3.8	4.6	3.7
Drug, %*	3.7	16.6	5.1	0.8	0.3	3.8	3.6
IMV, %*	37.7	54.6	43.7	33.5	24.8	41.1	34.5
NIV, %*	11.9	7.1	11.5	13.0	12.7	11.4	12.3
In‐hospital mortality (CI)	22 (21.322.7)	12.9 (11.913.9)	18.5 (17.619.4)	23.9 (23.024.9)	31.8 (30.633.1)	24.2 (23.325.1)	20.9 (20.121.7)
Surgical
Total, N (%)	552971 (28.8)	64983 (11.8)	190225 (34.4)	254336 (46)	43426 (7.9)	295660 (53.5)	257287 (46.5)
Pneumonia, %*	34.9	33.0	34.0	35.0	40.5	37.1	32.2
CHF, %*	27.2	8.9	21.7	33.3	42.6	26.7	27.7
ARDS, %*	45.5	51.5	45.2	44.7	42.7	45.0	46.1
Sepsis, %*	25.1	22.8	25.4	25.2	26.1	25.4	24.7
COPD, %*	8.2	1.1	7.4	10.8	7.5	8.3	8.1
AMI, %*	16.9	4.9	17.0	19.8	17.9	19.1	14.4
Asthma, %*	6.1	7.6	7.2	5.4	3.6	4.1	8.5
Stroke, %*	8.9	6.6	9.2	9.4	7.2	8.9	8.8
Trauma or burns, %*	12.2	26.5	9.6	9.2	20.3	13.8	10.4
Cardiorespiratory arrest, %*	5.5	4.4	6.0	5.4	5.2	6.1	4.7
Drug, %*	0.5	1.3	0.7	0.2	0.2	0.4	0.6
IMV, %*	52.9	57.1	54.3	51.3	50.0	54.5	51.0
NIV, %*	5.8	3.5	5.5	6.4	6.4	5.6	6.0
In‐hospital mortality, % (CI)	18.6 (17.819.5)	10.7 (9.312.0)	15.5 (14.216.8)	20.8 (19.821.9)	29.4 (27.831.1)	19.0 (18.219.8)	18.3 (17.319.2)

When we examined ventilation practices among medical patients we found that patients older than 85 years, when compared to patients younger than 45 years, were less likely to be treated with IMV (25% vs 55%) and more likely to be treated with NIV (12.7% vs 7%). At the same time, the average cost per case was lowest among patients 85 years and older, and hospital costs per case fell sharply after age 70 years. Costs were considerably higher for those who did not survive during hospitalization, particularly for patients younger than 45 years (Figure 2).

Figure 2

DISCUSSION

In this large population‐based study, we found that the number of hospitalizations associated with a diagnosis of ARF almost doubled over a 9‐year period. In 2009 there were nearly 2 million hospitalizations with ARF in the United States, resulting in approximately 380,000 deaths and inpatient costs of over $54 billion. The population‐adjusted ARF hospitalization rates increased in all age groups, and patients 85 years and older had the highest age‐specific hospitalization rate. Although overall rates of mechanical ventilation (NIV or IMV) remained stable over the 9‐year period, there was an important shift away from IMV (which decreased from 48% in 2001 to 42% in 2009) toward NIV (which increased from 4% in 2001 to 10% in 2009). Overall, there was a significant increase in the number of total deaths despite a decline in adjusted in‐hospital mortality rates. In‐hospital mortality rates decreased for all cases of ARF regardless of ventilation choice.

The findings of this study mirror results of others that have shown that although the incidence of critical care illnesses like sepsis[17, 20, 21, 26] and acute renal failure[27] has increased over the last decade, in‐hospital mortality rates have decreased.[20, 21, 28] Our results also compliment the results of a recent study that looked at hospitalizations for noncardiogenic ARF, which observed a 3.7‐fold increase in the number of cases and a steady decline in case fatality.[11]

Most prior studies addressing the incidence of ARF have included only patients receiving mechanical ventilation. In 1994, the estimated number of cases of ARF requiring IMV was 329,766,[9] which increased to 790,257 in 2005.[10] In our study we found that in 2009, the number of patients with ARF hospitalizations with IMV increased to 806,538. The increase in the overall number of cases with ARF was mainly driven by a surge in cases of sepsis and pneumonia. Our findings are consistent with national trends over time in noncardiogenic ARF[11] and in conditions that predispose patients to ARF such as sepsis[17, 20, 28] and acute renal failure.[27] As the number of claims for ARF doubled and the number of deaths increased, we found that adjusted hospital mortality improved from 27.6% in 2001 to 20.6% in 2009. This decline in hospital mortality was observed among all patients groups, regardless of ventilation choice. The decline in overall case fatality is consistent with prior findings in noncardiogenic ARF,[11] sepsis,[17, 28] and CHF.[29]

There are a number of potential explanations for the reduction in mortality observed over the study period, including improvements in hospital management of the underlying conditions leading to ARF, an increase in the proportion of patients being treated with NIV,[30] and advances in the care of critically ill patients such as the use of low‐tidal volume ventilation.[31, 32] Another contributor may be an increase in the proportion of discharges to nursing facilities, although this change in discharge disposition cannot fully explain our findings. For example, from 2007 to 2009, mortality decreased by 2 percentage points, and nursing home discharges increased by only 0.4 percentage points. Growth and aging of the US population only partially explain the increase we observed in the incidence of ARF, as age‐ and sex‐adjusted population rates increased by 56% from 2001 to 2009. In addition, the NIS captures data on hospital discharges and not individual patients; thus, a patient may have had multiple admissions. Over the last decade adoption of a more intensive practice style has been associated with improved in‐hospital mortality,[33, 34] and although these patients may be living longer they may have multiple readmissions.[35, 36]

We also observed that older patients were less likely to be treated with IMV, had a higher mortality rate, and less expensive care. These results are consistent with other studies and suggest that the intensity of treatment decreases with increasing age, and decisions to withhold or withdraw life‐supporting treatments are more frequent in the elderly.[26, 37] Prior research has shown that severity of illness is more important than age on patients' prognosis,[38, 39] and aggressive treatment strategies are not less cost‐effective when provided to older patients.[40]

Another important finding of this study is the marked increase in the use of NIV paired with a modest reduction in the use of IMV in the treatment of patients with ARF. This finding adds to evidence from other studies, which have similarly reported a dramatic increase in the use of NIV and a decrease in the use of IMV in patients with COPD as well as in ARF of other etiologies.[30, 41]

Our work has several limitations. First, we identified ARF based on ICD‐9‐CM codes and therefore cannot exclude disease misclassification. We did not find any studies in the literature addressing the accuracy and the completeness of ARF coding. However, we employed the same codes used to define ARF as has been used to define organ dysfunction in studies of severe sepsis,[17, 20] and the ICD‐9‐CM codes that we used to identify cases of ARDS have been used in prior studies.[11, 22, 23] Another limitation is that it is not clear to what extent the trends we observed may be due to changes over time in documentation and coding practices. Although this should be considered given the additional reimbursement associated with the diagnosis of ARF, our observation that rates of assisted ventilation have remained almost flat over the 9‐year period of the study suggest that would not wholly account for the rise in ARF. Second, because we did not have access to physiological data such as results of blood gas testing, we could not determine whether the threshold for applying the diagnosis of ARF or for delivering ventilatory support has changed over time. Third, for the purpose of this study we employed a broad definition of ARF, not limiting cases to those requiring mechanical ventilation, and this led to a more heterogeneous cohort including less severe cases of ARF. However, this is not dissimilar to the heterogeneity in disease severity observed among patients who receive a diagnosis of heart failure or acute renal failure. Fourth, survivors of ARF remain at high risk of death in the months after hospitalization,[42] but we assessed only in‐hospital mortality. It is possible that although in‐hospital mortality has improved, 30‐day mortality remained stable. Finally, as the NIS contains only discharge‐level data, we could not distinguish between patients admitted for ARF from those who developed ARF (potentially iatrogenic) after admission.

In summary, over the period of 2001 to 2009, there was a large increase in the number of patients given a diagnosis of ARF and a concomitant reduction in inpatient mortality. Although rates of mechanical ventilation remained relatively constant, there was a significant shift toward greater use of NIV at the expense of IMV.

Acute respiratory failure (ARF), a common and serious complication in hospitalized patients, may be caused by several conditions including pneumonia, chronic obstructive pulmonary disease (COPD), adult respiratory distress syndrome (ARDS), and congestive heart failure (CHF). Although ARF is conventionally defined by an arterial oxygen tension of <60 mm Hg, an arterial carbon dioxide tension of >45 mm Hg, or both, these thresholds serve as a guide to be used in combination with history and clinical assessment of the patient.[1, 2] Supplemental oxygen and treatment of the underlying cause is the mainstay of therapy for ARF, but in severe cases patients are treated with invasive mechanical ventilation (IMV) or noninvasive ventilation (NIV). ARF is the most frequent reason for admission to the intensive care unit (ICU)[3, 4] and has an in‐hospital mortality rate of 33% to 37% among those who require IMV.[5, 6] The majority of epidemiologic studies of ARF have been limited to patients requiring mechanical ventilation or those admitted to the ICU, and information about the characteristics and outcomes of patients across the full spectrum of severity is much more limited.[5, 7, 8, 9, 10, 11] General improvements in the management of underlying conditions, implementation of more effective ventilation strategies,[12, 13] and increasing use of NIV[14, 15] may have led to better outcomes for patients with ARF, yet empirical evidence of a change in the adjusted mortality rate over time is lacking.

The objective of this study was to provide a broad characterization of the epidemiology of ARF among adults hospitalized in the United States using a large nationally representative database. We sought to evaluate whether incidence, mortality, cost, or ventilation practice associated with ARF in the United States changed over the period of 2001 to 2009.

METHODS

RESULTS

Hospitalization Trends

The number of hospitalizations with an ARF diagnosis code increased at an average annual rate of 11.3% from 1,007,549 (standard deviation [SD] = 19,268) in 2001 to 1,917,910 (SD = 47,558) in 2009. More than two‐thirds of ARF admissions were associated with medical, rather than surgical, conditions (69.5% in 2001 and 71.2% in 2009). The median age, racial make‐up, and gender did not change significantly. Over the study period we observed an increase in ARF‐related hospitalizations in large, urban, teaching hospitals and in hospitals located in the Midwest (Table 1).

Table 1.Hospitalizations With Acute Respiratory Failure in the United States, 2001 to 2009, by Patient and Hospital Characteristics

	2001	2003	2005	2007	2009
NOTE: Abbreviations: ARF, acute respiratory failure; IMV, invasive mechanical ventilation; LOS, length of stay; NIV, noninvasive ventilation; SD, standard deviation; SE standard error. P value for trend <0.01, including all years 20012009, *P value for trend <0.0001, including all years 20012009. Adjusted for sex, age, race, hospital characteristics, and comorbidities. Geometric mean reported, standard errors from Taylor series expansion.
Patient characteristics
All, N (SD)	1,007,549 (19,268)	1,184,928 (25,542)	1,288,594 (30,493)	1,480,270 (32,002)	1,917,910 (47,558)
Age, mean (SE), y	66.6 (0.2)	66.0 (0.2)	66.1 (0.2)	65.8 (0.2)	65.8 (0.2)
Age group, %
1844	11.5	12.0	11.5	11.6	10.9
4564*	26.7	28.9	29.6	30.7	31.7
6584*	50.2	47.8	47.0	45.7	45.3
85+	11.5	11.4	11.9	12.0	12.1
Male*	48.1	48.2	48.6	49.3	49.2
Race
White	75.8	71.9	76.5	71.8	73.4
Black	12.7	13.6	11.2	14.2	12.5
Hispanic	7.2	9.8	7.7	8.5	7.8
Other	4.2	4.7	4.7	5.5	6.3
Primary ARF	20.7	20.9	25.9	26.1	19.9
Secondary ARF	79.3	79.1	74.1	73.9	80.1
Medical*	69.5	69.1	69.9	70.2	71.2
Surgical*	30.5	30.8	30.1	29.8	28.8
Hospital characteristics, %
Number of beds
Small	10.0	10.1	10.5	10.8	11.3
Medium	25.2	25.3	24.6	24.0	22.7
Large	64.7	64.6	64.9	65.2	66.0
Region
South*	18.5	18.5	17.6	17.0	16.3
Midwest	21.4	22.0	23.6	23.2	23.5
Northeast	42.6	41.7	41.4	42.2	42.1
West*	17.5	17.8	17.3	17.6	18.1
Hospital type
Rural	13.6	13.0	11.8	11.0	10.8
Urban nonteaching	45.5	44.5	50.1	45.3	45.7
Urban teaching	40.9	42.5	38.1	43.7	43.6
Patient outcomes
Ventilation strategy
IMV*	48.5	48.4	47.5	46.5	42.1
NIV*	3.8	5.3	6.9	9.4	10.1
IMV or NIV	50.9	51.7	52.1	52.9	49.7
Disposition
Home/home healthcare*	42.1	43.8	42.8	43.4	45.7
Transfer to acute care	5.2	4.7	4.6	4.6	4.4
Nursing facility*	24.4	24.9	27.4	28.6	29.0
Other	0.7	0.8	0.9	0.9	1.0
Adjusted mortality, % (SE)*	27.6 (0.3)	26.4 (0.4)	24.9 (0.4)	22.7 (0.4)	20.6 (0.3)
Adjusted mean, LOS/case, d (SE)*	7.8 (0.1)	7.9 (0.1)	7.7 (0.1)	7.5 (0.1)	7.1 (0.1)
Adjusted mean cost/case, 2009 US$, (SE)	15,818 (251)	16,981 (419)	17,236 (411)	16,941 (436)	15,987 (402)

After adjusting for age and sex, the population incidence of ARF increased from 502 (standard error [SE] = 10) cases per 100,000 in 2001 to 784 (SE = 19) cases per 100,000 in 2009 (a 56% increase, P < 0.0001). Hispanics had the lowest rates of ARF, with both black and white groups having similar rates (Table 2).

Table 2.Cases of Acute Respiratory Failure per 100,000 Population

	2001	2003	2005	2007	2009
NOTE: Data are presented as number per 100,000 population (standard error), standardized to 2000 US Census population. *P value for trend < 0.0001, including all years 2001 to 2009.
All*	502 (10)	569 (12)	595 (14)	627 (14)	784 (19)
Age group
1844*	107 (3)	130 (4)	137 (4)	153 (5)	189 (6)
4564*	422 (9)	500 (12)	521 (13)	580 (14)	739 (19)
6584*	1697 (35)	1863 (42)	1950 (50)	2066 (46)	2578 (69)
85+	3449 (86)	3792 (106)	3981 (120)	3429 (97)	4163 (123)
Sex
Male*	491 (10)	553 (13)	582 (14)	629 (14)	782 (20)
Female*	512 (10)	583 (12)	607 (15)	625 (13)	786 (19)
Race/ethnicity
White*	398 (11)	427 (12)	466 (16)	450 (13)	699 (21)
Black*	423 (27)	513 (33)	432 (26)	574 (38)	738 (37)
Hispanic*	247 (24)	381 (42)	307 (27)	353 (34)	478 (42)
Other*	268 (20)	342 (29)	347 (26)	424 (29)	713 (77)
In‐hospital mortality	140 (3)	148 (3)	146 (3)	140 (3)	154 (4)

The most common etiologies of ARF among medical patients were pneumonia, CHF, ARDS, COPD exacerbation, and sepsis. Over the 9‐year study, the proportion of cases secondary to pneumonia and sepsis rose significantly: from 39% to 46% and 13% to 21%, respectively (Figure 1).

Figure 1

2009 Data Analysis

In 2009 the 1,917,910 hospitalizations with ARF resulted in 381,155 (SD = 8965) deaths and a total inpatient cost of $54 billion. The most common etiologies in patients over 65 years old were pneumonia, CHF, COPD, ARDS, and sepsis. In patients younger than 45 years the most frequent diagnoses were pneumonia, ARDS, sepsis, asthma, drug ingestion, and trauma. Stratified analysis by gender and by age groups showed that mortality rates among men were higher than for women and were highest in patients older than 85 years (Table 3).

Table 3.Characteristics of Hospitalizations by Etiology (Medical, Surgical, Comorbidities, Procedures) in 2009

Disease	Total	Age <45 Years	4565 Years	6584 Years	85+ Years	Male	Female
NOTE: One patient can have more than 1 diagnosis. Abbreviations: AMI, acute myocardial infarction; ARDS, adult respiratory distress syndrome; CHF, congestive heart failure; CI, confidence interval; COPD, chronic obstructive pulmonary disease; IMV, invasive mechanical ventilation; NIV, noninvasive ventilation. *P < 0.0001 for age group. P < 0.0001 for gender. The P values are not from Rao‐Scott 2 test.
Medical
Total, N (%)	1,364,624 (71.2)	144,715 (10.6)	416,922 (30.6)	615,009 (45.1)	187,977 (13.8)	647,894 (47.5)	716,635 (52.5)
Pneumonia, %*	46.1	41.7	42.8	46.9	54.3	48.8	43.7
CHF, %*	36.6	10.4	27.3	43.6	54.8	35.0	38.1
ARDS, %*	16.1	22.9	16.2	14.5	15.9	15.5	16.7
Sepsis, %*	21.2	18.1	21.3	21.3	23.1	22.8	19.8
COPD, %*	25.4	4.2	25.6	32.3	18.3	25.0	25.7
AMI, %*	9.0	2.6	7.1	10.5	13.3	9.3	8.8
Asthma, %*	9.2	18.1	11.6	6.7	5.4	6.2	12.0
Stroke, %*	4.8	2.3	4.1	5.5	6.0	5.0	4.7
Trauma or burns, %*	3.4	5.4	2.9	3.0	4.1	4.3	2.5
Cardiorespiratory arrest, %*	4.1	3.9	4.4	4.1	3.8	4.6	3.7
Drug, %*	3.7	16.6	5.1	0.8	0.3	3.8	3.6
IMV, %*	37.7	54.6	43.7	33.5	24.8	41.1	34.5
NIV, %*	11.9	7.1	11.5	13.0	12.7	11.4	12.3
In‐hospital mortality (CI)	22 (21.322.7)	12.9 (11.913.9)	18.5 (17.619.4)	23.9 (23.024.9)	31.8 (30.633.1)	24.2 (23.325.1)	20.9 (20.121.7)
Surgical
Total, N (%)	552971 (28.8)	64983 (11.8)	190225 (34.4)	254336 (46)	43426 (7.9)	295660 (53.5)	257287 (46.5)
Pneumonia, %*	34.9	33.0	34.0	35.0	40.5	37.1	32.2
CHF, %*	27.2	8.9	21.7	33.3	42.6	26.7	27.7
ARDS, %*	45.5	51.5	45.2	44.7	42.7	45.0	46.1
Sepsis, %*	25.1	22.8	25.4	25.2	26.1	25.4	24.7
COPD, %*	8.2	1.1	7.4	10.8	7.5	8.3	8.1
AMI, %*	16.9	4.9	17.0	19.8	17.9	19.1	14.4
Asthma, %*	6.1	7.6	7.2	5.4	3.6	4.1	8.5
Stroke, %*	8.9	6.6	9.2	9.4	7.2	8.9	8.8
Trauma or burns, %*	12.2	26.5	9.6	9.2	20.3	13.8	10.4
Cardiorespiratory arrest, %*	5.5	4.4	6.0	5.4	5.2	6.1	4.7
Drug, %*	0.5	1.3	0.7	0.2	0.2	0.4	0.6
IMV, %*	52.9	57.1	54.3	51.3	50.0	54.5	51.0
NIV, %*	5.8	3.5	5.5	6.4	6.4	5.6	6.0
In‐hospital mortality, % (CI)	18.6 (17.819.5)	10.7 (9.312.0)	15.5 (14.216.8)	20.8 (19.821.9)	29.4 (27.831.1)	19.0 (18.219.8)	18.3 (17.319.2)

When we examined ventilation practices among medical patients we found that patients older than 85 years, when compared to patients younger than 45 years, were less likely to be treated with IMV (25% vs 55%) and more likely to be treated with NIV (12.7% vs 7%). At the same time, the average cost per case was lowest among patients 85 years and older, and hospital costs per case fell sharply after age 70 years. Costs were considerably higher for those who did not survive during hospitalization, particularly for patients younger than 45 years (Figure 2).

Figure 2

DISCUSSION

In this large population‐based study, we found that the number of hospitalizations associated with a diagnosis of ARF almost doubled over a 9‐year period. In 2009 there were nearly 2 million hospitalizations with ARF in the United States, resulting in approximately 380,000 deaths and inpatient costs of over $54 billion. The population‐adjusted ARF hospitalization rates increased in all age groups, and patients 85 years and older had the highest age‐specific hospitalization rate. Although overall rates of mechanical ventilation (NIV or IMV) remained stable over the 9‐year period, there was an important shift away from IMV (which decreased from 48% in 2001 to 42% in 2009) toward NIV (which increased from 4% in 2001 to 10% in 2009). Overall, there was a significant increase in the number of total deaths despite a decline in adjusted in‐hospital mortality rates. In‐hospital mortality rates decreased for all cases of ARF regardless of ventilation choice.

The findings of this study mirror results of others that have shown that although the incidence of critical care illnesses like sepsis[17, 20, 21, 26] and acute renal failure[27] has increased over the last decade, in‐hospital mortality rates have decreased.[20, 21, 28] Our results also compliment the results of a recent study that looked at hospitalizations for noncardiogenic ARF, which observed a 3.7‐fold increase in the number of cases and a steady decline in case fatality.[11]

Most prior studies addressing the incidence of ARF have included only patients receiving mechanical ventilation. In 1994, the estimated number of cases of ARF requiring IMV was 329,766,[9] which increased to 790,257 in 2005.[10] In our study we found that in 2009, the number of patients with ARF hospitalizations with IMV increased to 806,538. The increase in the overall number of cases with ARF was mainly driven by a surge in cases of sepsis and pneumonia. Our findings are consistent with national trends over time in noncardiogenic ARF[11] and in conditions that predispose patients to ARF such as sepsis[17, 20, 28] and acute renal failure.[27] As the number of claims for ARF doubled and the number of deaths increased, we found that adjusted hospital mortality improved from 27.6% in 2001 to 20.6% in 2009. This decline in hospital mortality was observed among all patients groups, regardless of ventilation choice. The decline in overall case fatality is consistent with prior findings in noncardiogenic ARF,[11] sepsis,[17, 28] and CHF.[29]

There are a number of potential explanations for the reduction in mortality observed over the study period, including improvements in hospital management of the underlying conditions leading to ARF, an increase in the proportion of patients being treated with NIV,[30] and advances in the care of critically ill patients such as the use of low‐tidal volume ventilation.[31, 32] Another contributor may be an increase in the proportion of discharges to nursing facilities, although this change in discharge disposition cannot fully explain our findings. For example, from 2007 to 2009, mortality decreased by 2 percentage points, and nursing home discharges increased by only 0.4 percentage points. Growth and aging of the US population only partially explain the increase we observed in the incidence of ARF, as age‐ and sex‐adjusted population rates increased by 56% from 2001 to 2009. In addition, the NIS captures data on hospital discharges and not individual patients; thus, a patient may have had multiple admissions. Over the last decade adoption of a more intensive practice style has been associated with improved in‐hospital mortality,[33, 34] and although these patients may be living longer they may have multiple readmissions.[35, 36]

We also observed that older patients were less likely to be treated with IMV, had a higher mortality rate, and less expensive care. These results are consistent with other studies and suggest that the intensity of treatment decreases with increasing age, and decisions to withhold or withdraw life‐supporting treatments are more frequent in the elderly.[26, 37] Prior research has shown that severity of illness is more important than age on patients' prognosis,[38, 39] and aggressive treatment strategies are not less cost‐effective when provided to older patients.[40]

Another important finding of this study is the marked increase in the use of NIV paired with a modest reduction in the use of IMV in the treatment of patients with ARF. This finding adds to evidence from other studies, which have similarly reported a dramatic increase in the use of NIV and a decrease in the use of IMV in patients with COPD as well as in ARF of other etiologies.[30, 41]

Our work has several limitations. First, we identified ARF based on ICD‐9‐CM codes and therefore cannot exclude disease misclassification. We did not find any studies in the literature addressing the accuracy and the completeness of ARF coding. However, we employed the same codes used to define ARF as has been used to define organ dysfunction in studies of severe sepsis,[17, 20] and the ICD‐9‐CM codes that we used to identify cases of ARDS have been used in prior studies.[11, 22, 23] Another limitation is that it is not clear to what extent the trends we observed may be due to changes over time in documentation and coding practices. Although this should be considered given the additional reimbursement associated with the diagnosis of ARF, our observation that rates of assisted ventilation have remained almost flat over the 9‐year period of the study suggest that would not wholly account for the rise in ARF. Second, because we did not have access to physiological data such as results of blood gas testing, we could not determine whether the threshold for applying the diagnosis of ARF or for delivering ventilatory support has changed over time. Third, for the purpose of this study we employed a broad definition of ARF, not limiting cases to those requiring mechanical ventilation, and this led to a more heterogeneous cohort including less severe cases of ARF. However, this is not dissimilar to the heterogeneity in disease severity observed among patients who receive a diagnosis of heart failure or acute renal failure. Fourth, survivors of ARF remain at high risk of death in the months after hospitalization,[42] but we assessed only in‐hospital mortality. It is possible that although in‐hospital mortality has improved, 30‐day mortality remained stable. Finally, as the NIS contains only discharge‐level data, we could not distinguish between patients admitted for ARF from those who developed ARF (potentially iatrogenic) after admission.

In summary, over the period of 2001 to 2009, there was a large increase in the number of patients given a diagnosis of ARF and a concomitant reduction in inpatient mortality. Although rates of mechanical ventilation remained relatively constant, there was a significant shift toward greater use of NIV at the expense of IMV.