The management of postoperative pain is essential to perioperative care, and adequate postoperative analgesia has been associated with several key clinical benefits, including fewer postoperative complications, earlier patient ambulation, reduced costs due to shorter hospital stays, and improved rehabilitation.1, 2 While opioids have long been central to postoperative analgesia, they have been associated with various adverse effects, including sedation, dizziness, nausea, vomiting, constipation, dependence, tolerance, and respiratory depression.2, 3 Constipation, one of the most common adverse effects resulting from opioid therapy, can be debilitating. Indeed, opioid effects on gut motility can occur even after a single dose.3 The consequences of opioid‐induced constipation (OIC) may be severe enough to warrant a dosage reduction of the opioid; however, this may lead to compromised analgesia, which can hinder recovery.4, 5 Thus, effective treatment of OIC is an important clinical consideration in patients undergoing pain management with opioids. Unfortunately, laxatives and other treatment strategies can have unpredictable or suboptimal results for many patients with OIC; therefore, other options are needed for the treatment of OIC.6, 7

Opioid receptor agonists cause constipation by adversely altering many aspects of intestinal function, including fluid dynamics, gastric emptying, propulsive motor activity, and transit time.3 Opioid receptors are widely distributed in the central nervous system and throughout the intestinal system. The mechanism of OIC may have both peripherally and centrally mediated components.8 Nonselective opioid receptor antagonists block the undesired effects on the gut, but because they cross the blood‐brain barrier, they also interfere with analgesia and may lead to symptoms of withdrawal. Methylnaltrexone is a selective, peripherally acting mu‐opioid receptor antagonist,9 formed by the addition of a methyl group to the amine ring of the mu‐opioid receptor antagonist naltrexone. The resulting quarternary amine has greater polarity, lower lipid solubility, and restricted ability to cross the blood‐brain barrier.10 Thus, methylnaltrexone was designed to decrease the peripheral adverse effects of opioids without interfering with centrally mediated analgesia.

Investigations of methylnaltrexone effects in healthy volunteers showed that methylnaltrexone attenuated morphine‐induced delays in gastric emptying and oral‐cecal transit without affecting analgesia.1113 Further studies of methylnaltrexone for the treatment of constipation due to methadone use demonstrated rapid laxation response.1416 Two randomized, double‐blind, placebo‐controlled studies of methylnaltrexone in 288 patients with advanced illness and OIC showed that methylnaltrexone rapidly induced laxation without compromising analgesia.17, 18 Methylnaltrexone is currently approved for the treatment of OIC in patients with advanced illness who are receiving palliative care, when response to laxative therapy has not been sufficient.19

Recently, the use of methylnaltrexone for the treatment of OIC in patients with chronic, nonmalignant pain was assessed in a randomized, double‐blind, placebo‐controlled trial of more than 400 patients. Investigators found that methylnaltrexone induced laxation and was generally well tolerated (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA; Duerden et al., 29th Annual Scientific Meeting of the American Pain Society, May 6‐10, 2010, Baltimore, MD), supporting the safety and efficacy of methylnaltrexone in the setting of OIC resulting from chronic opioid treatment. The present study aimed to assess the activity of methylnaltrexone in patients receiving mu‐agonist opioid analgesics during rehabilitation, following an orthopedic surgical procedure, who were experiencing acute OIC.

METHODS

Study Design

This randomized, double‐blind, placebo‐controlled, parallel‐group, hypothesis‐generating phase 2 study was conducted from October 2007 to January 2009 at 16 US hospitals and rehabilitation facilities in accordance with the International Conference on Harmonisation Good Clinical Practice Guidelines and the Declaration of Helsinki, and was approved by the Institutional Review Board and/or Independent Ethics Committee at each of the participating investigational centers. All patients provided written informed consent prior to study participation.

Eligible patients were randomized by interactive voice response system in a 1:1 ratio to receive once‐daily subcutaneous (SC) injections of either 12 mg methylnaltrexone or placebo (Figure 1). The chosen 12‐mg unit dosing corresponds to approximately 0.15 mg/kg (assuming an 80‐kg patient) and was found to be both efficacious and well tolerated in the treatment of OIC in prior studies, including studies in advanced‐illness patients17, 18 and in patients with chronic, nonmalignant pain (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA; Duerden et al., 29th Annual Scientific Meeting of the American Pain Society, May 6‐10, 2010, Baltimore, MD.20 The first dose of study medication was administered on the day of randomization or on the next calendar day. Once enrolled, the patient received once‐daily doses of methylnaltrexone for up to 4 or 7 days. Dosing continued until the patient received the maximum number of doses allowed, no longer needed opioid medication, or was discharged from the medical facility. Each patient completed a follow‐up safety visit at 14 3 days following the last dose.

Figure 1

RESULTS

Patient Populations

The flow of patients through the study is summarized in Figure 2. A total of 51 patients were enrolled. Of these, 33 received at least 1 dose of study treatment following double‐blind randomization and comprised both the mITT and safety populations. Seventeen of these patients were enrolled under the original protocol and could receive study drug for up to 7 days, while 16 patients enrolled under a subsequent protocol revision could receive study drug for up to 4 days. This change from a 7‐day to a 4‐day treatment protocol allowed for the capture of more study patients in view of the time pressures of short lengths of stay in postoperative settings. In total, 31 patients received at least 2 doses, and 26 patients received at least 4 doses of study drug. A total of 27 patients completed the study. Baseline demographics and prestudy surgical procedures were similar in both treatment groups (Table 1).

Figure 2

Table 1.Baseline Patient Demographics

Characteristic	Methylnaltrexone (n = 18)	Placebo (n = 15)
NOTE: Modified intent‐to‐treat population. Abbreviations: BMI, body mass index.; SD, standard deviation. * Baseline opioid use was defined as total opioid use within 24 hours prior to randomization.
Mean age, yr (SD)	64.2 (9.0)	65.2 (11.6)
Mean weight, kg (SD)	92.5 (22.5)	91.0 (20.2)
Mean BMI, kg/m2 (SD)	32.3 (7.2)	34.2 (6.41)
Sex, n (%)
Female	11 (61.1)	11 (73.3)
Male	7 (38.9)	4 (26.7)
Race, n (%)
White	14 (77.8)	10 (66.7)
Black	4 (22.2)	5 (33.3)
Type of surgery, n (%)
Total knee replacement	8 (44.4)	7 (46.7)
Total hip replacement	6 (33.3)	6 (40.0)
Spinal fusion	2 (11.1)	0
Fracture reduction	2 (11.1)	2 (13.3)
Median opioid use,* mg (range)	28.00 (6.75‐168.01)	25.00 (9.00‐75.00)
Median time from surgery to study drug administration, days (range)	4 (3‐6)	4 (3‐6)

Efficacy

A significantly greater percentage of patients had a bowel movement within 2 hours (P = 0.021) and 4 hours (P = 0.046) of the first dose of methylnaltrexone compared with patients who received placebo (Figure 3). Within 2 hours, 6 patients (33.3%; 95% confidence interval [CI], 13.34‐59.01) who received methylnaltrexone achieved laxation, while laxation did not occur in any patient who received placebo. By 4 hours posttreatment, 7 patients (38.9%; 95% CI, 17.30‐64.25) in the methylnaltrexone group achieved laxation compared with only 1 patient (6.7%; 95% CI, 0.17‐31.95) on placebo. Three patients in each treatment group received rescue laxatives.

Figure 3

The time to first laxation (Figure 4) was significantly shorter in patients who received methylnaltrexone compared with those in the placebo group. Patients on methylnaltrexone achieved laxation in a median time of 15.8 hours, compared with a median time of 50.9 hours for patients in the placebo group (P = 0.02, log‐rank test). The median time to laxation was less than 1 hour in the 7 methylnaltrexone‐treated patients who experienced laxation within 4 hours following the first dose. Of the remaining 11 methylnaltrexone‐treated patients, one experienced no laxation after 6 doses, and the median time to laxation for the others was 29.9 hours (not shown in figure).

Figure 4

Analysis of the Global Satisfaction With Treatment Scale revealed that more patients expressed overall treatment satisfaction (defined as very satisfied, satisfied, or minimally satisfied) with methylnaltrexone assessed 4 hours ( 30 minutes) after the first dose, compared with patients on placebo (83.3% vs 60.0%, respectively). At the study endpoint, overall treatment satisfaction with methylnaltrexone remained high (83.3%), whereas satisfaction with placebo was 53.3%. Additionally, no patients in the methylnaltrexone group expressed any dissatisfaction with treatment (defined as minimally dissatisfied, dissatisfied, or very dissatisfied) at endpoint, compared with 26.7% of patients in the placebo group who expressed some degree of dissatisfaction.

DISCUSSION

This pilot study suggests that methylnaltrexone actively induces laxation and is generally well tolerated in patients receiving mu‐opioid analgesia, following orthopedic surgery, who develop OIC acutely. It was the first study, to our knowledge, to investigate the efficacy of methylnaltrexone for the treatment of OIC in an acute postoperative setting. The protocol amendment changing the duration of treatment from 7 days to 4 days did not materially affect the results of the study. The response to methylnaltrexone was rapid, with 33.3% experiencing laxation within 2 hours. The median time to laxation was nearly 1.5 days shorter in patients treated with methylnaltrexone compared with those receiving placebo. Correspondingly, overall patient satisfaction was high in the methylnaltrexone group. Efficacy was attained without diminishing opioid analgesia, and without inducing signs or symptoms of opioid withdrawal. The incidence of AEs was similar between groups, and no treatment‐emergent SAEs were reported in this study.

Previous clinical trials investigated the safety and efficacy of methylnaltrexone for the treatment of OIC in patients with advanced illness and with chronic, nonmalignant pain. The present study extends those findings to a population of patients experiencing acute OIC following orthopedic surgery. Previous studies showed that approximately 48% to 62% of advanced‐illness patients experienced laxation within 4 hours of receiving SC methylnaltrexone,17, 18 compared with 38.9% of acute OIC patients in this study. In a clinical trial of patients with chronic, nonmalignant pain, 34.2% of patients experienced laxation within 4 hours of SC methylnaltrexone injection (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA). The differences in laxation response between these trials may be attributable to differences in the patient populations or to methodologic differences between the studies.

Similar to findings demonstrated in a clinical study evaluating methylnaltrexone for OIC in a different patient population, those with advanced illness,22 this study supports the premise that future laxation response with prolonged use is most likely to occur when a laxation response was achieved after the first or second initial administrations of methylnaltrexone. In contradistinction, if laxation does not occur with these early doses, continued methylnaltrexone dosing is less likely to produce a response later.

This study has some limitations that must be considered. First, as this was a hypothesis‐generating study, all efficacy parameters investigated were exploratory in nature. The results reported herein warrant careful consideration, owing to a small sample size that may limit their generalizability, prior to replication in a more rigorously designed study with prespecified efficacy endpoints. Likewise, the assessment of health outcome parameters is limited. Another limitation is the small sample size utilized in this study, potentially resulting in a type II error.

Subcutaneous administration potentially offers a considerable benefit over oral therapies for OIC in this patient population post‐orthopedic surgery. Nausea and vomiting can occur as a consequence of anesthesia and of postoperative opioid analgesia, and may compromise adequate dosing of oral medications prescribed to treat OIC. Subcutaneous delivery of methylnaltrexone may circumvent this potential drawback while providing potentially rapid, effective treatment for OIC. Once‐daily dosing may also help to minimize caregiver burden and patient discomfort by preventing the need for more frequent or unpleasant treatments for OIC, such as enemas.

This study provides an initial positive signal for a broader, albeit off‐label use for methylnaltrexonethat being for the treatment of acute constipation that occurs as a consequence of postoperative opioid‐mediated analgesia in patients following orthopedic procedures. Adequate treatment of OIC, even in the acute postoperative setting, is likely to lead to better overall pain management and improved patient outcomes. Additionally, effective management of acute OIC is likely to be cost‐effective in terms of reducing the duration of hospital stays, reducing the need for nursing resources and the time spent administering rescue treatments for OIC (eg, enemas), and avoiding returns to an acute setting (eg, the emergency department) for treatment. The results presented herein suggest that methylnaltrexone may be effective and have a good safety profile in the treatment of acute OIC following orthopedic surgery. Validation of these results in larger well‐controlled trials would be welcome.

The management of postoperative pain is essential to perioperative care, and adequate postoperative analgesia has been associated with several key clinical benefits, including fewer postoperative complications, earlier patient ambulation, reduced costs due to shorter hospital stays, and improved rehabilitation.1, 2 While opioids have long been central to postoperative analgesia, they have been associated with various adverse effects, including sedation, dizziness, nausea, vomiting, constipation, dependence, tolerance, and respiratory depression.2, 3 Constipation, one of the most common adverse effects resulting from opioid therapy, can be debilitating. Indeed, opioid effects on gut motility can occur even after a single dose.3 The consequences of opioid‐induced constipation (OIC) may be severe enough to warrant a dosage reduction of the opioid; however, this may lead to compromised analgesia, which can hinder recovery.4, 5 Thus, effective treatment of OIC is an important clinical consideration in patients undergoing pain management with opioids. Unfortunately, laxatives and other treatment strategies can have unpredictable or suboptimal results for many patients with OIC; therefore, other options are needed for the treatment of OIC.6, 7

Opioid receptor agonists cause constipation by adversely altering many aspects of intestinal function, including fluid dynamics, gastric emptying, propulsive motor activity, and transit time.3 Opioid receptors are widely distributed in the central nervous system and throughout the intestinal system. The mechanism of OIC may have both peripherally and centrally mediated components.8 Nonselective opioid receptor antagonists block the undesired effects on the gut, but because they cross the blood‐brain barrier, they also interfere with analgesia and may lead to symptoms of withdrawal. Methylnaltrexone is a selective, peripherally acting mu‐opioid receptor antagonist,9 formed by the addition of a methyl group to the amine ring of the mu‐opioid receptor antagonist naltrexone. The resulting quarternary amine has greater polarity, lower lipid solubility, and restricted ability to cross the blood‐brain barrier.10 Thus, methylnaltrexone was designed to decrease the peripheral adverse effects of opioids without interfering with centrally mediated analgesia.

Investigations of methylnaltrexone effects in healthy volunteers showed that methylnaltrexone attenuated morphine‐induced delays in gastric emptying and oral‐cecal transit without affecting analgesia.1113 Further studies of methylnaltrexone for the treatment of constipation due to methadone use demonstrated rapid laxation response.1416 Two randomized, double‐blind, placebo‐controlled studies of methylnaltrexone in 288 patients with advanced illness and OIC showed that methylnaltrexone rapidly induced laxation without compromising analgesia.17, 18 Methylnaltrexone is currently approved for the treatment of OIC in patients with advanced illness who are receiving palliative care, when response to laxative therapy has not been sufficient.19

Recently, the use of methylnaltrexone for the treatment of OIC in patients with chronic, nonmalignant pain was assessed in a randomized, double‐blind, placebo‐controlled trial of more than 400 patients. Investigators found that methylnaltrexone induced laxation and was generally well tolerated (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA; Duerden et al., 29th Annual Scientific Meeting of the American Pain Society, May 6‐10, 2010, Baltimore, MD), supporting the safety and efficacy of methylnaltrexone in the setting of OIC resulting from chronic opioid treatment. The present study aimed to assess the activity of methylnaltrexone in patients receiving mu‐agonist opioid analgesics during rehabilitation, following an orthopedic surgical procedure, who were experiencing acute OIC.

METHODS

Study Design

This randomized, double‐blind, placebo‐controlled, parallel‐group, hypothesis‐generating phase 2 study was conducted from October 2007 to January 2009 at 16 US hospitals and rehabilitation facilities in accordance with the International Conference on Harmonisation Good Clinical Practice Guidelines and the Declaration of Helsinki, and was approved by the Institutional Review Board and/or Independent Ethics Committee at each of the participating investigational centers. All patients provided written informed consent prior to study participation.

Eligible patients were randomized by interactive voice response system in a 1:1 ratio to receive once‐daily subcutaneous (SC) injections of either 12 mg methylnaltrexone or placebo (Figure 1). The chosen 12‐mg unit dosing corresponds to approximately 0.15 mg/kg (assuming an 80‐kg patient) and was found to be both efficacious and well tolerated in the treatment of OIC in prior studies, including studies in advanced‐illness patients17, 18 and in patients with chronic, nonmalignant pain (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA; Duerden et al., 29th Annual Scientific Meeting of the American Pain Society, May 6‐10, 2010, Baltimore, MD.20 The first dose of study medication was administered on the day of randomization or on the next calendar day. Once enrolled, the patient received once‐daily doses of methylnaltrexone for up to 4 or 7 days. Dosing continued until the patient received the maximum number of doses allowed, no longer needed opioid medication, or was discharged from the medical facility. Each patient completed a follow‐up safety visit at 14 3 days following the last dose.

Figure 1

RESULTS

Patient Populations

The flow of patients through the study is summarized in Figure 2. A total of 51 patients were enrolled. Of these, 33 received at least 1 dose of study treatment following double‐blind randomization and comprised both the mITT and safety populations. Seventeen of these patients were enrolled under the original protocol and could receive study drug for up to 7 days, while 16 patients enrolled under a subsequent protocol revision could receive study drug for up to 4 days. This change from a 7‐day to a 4‐day treatment protocol allowed for the capture of more study patients in view of the time pressures of short lengths of stay in postoperative settings. In total, 31 patients received at least 2 doses, and 26 patients received at least 4 doses of study drug. A total of 27 patients completed the study. Baseline demographics and prestudy surgical procedures were similar in both treatment groups (Table 1).

Figure 2

Table 1.Baseline Patient Demographics

Characteristic	Methylnaltrexone (n = 18)	Placebo (n = 15)
NOTE: Modified intent‐to‐treat population. Abbreviations: BMI, body mass index.; SD, standard deviation. * Baseline opioid use was defined as total opioid use within 24 hours prior to randomization.
Mean age, yr (SD)	64.2 (9.0)	65.2 (11.6)
Mean weight, kg (SD)	92.5 (22.5)	91.0 (20.2)
Mean BMI, kg/m2 (SD)	32.3 (7.2)	34.2 (6.41)
Sex, n (%)
Female	11 (61.1)	11 (73.3)
Male	7 (38.9)	4 (26.7)
Race, n (%)
White	14 (77.8)	10 (66.7)
Black	4 (22.2)	5 (33.3)
Type of surgery, n (%)
Total knee replacement	8 (44.4)	7 (46.7)
Total hip replacement	6 (33.3)	6 (40.0)
Spinal fusion	2 (11.1)	0
Fracture reduction	2 (11.1)	2 (13.3)
Median opioid use,* mg (range)	28.00 (6.75‐168.01)	25.00 (9.00‐75.00)
Median time from surgery to study drug administration, days (range)	4 (3‐6)	4 (3‐6)

Efficacy

A significantly greater percentage of patients had a bowel movement within 2 hours (P = 0.021) and 4 hours (P = 0.046) of the first dose of methylnaltrexone compared with patients who received placebo (Figure 3). Within 2 hours, 6 patients (33.3%; 95% confidence interval [CI], 13.34‐59.01) who received methylnaltrexone achieved laxation, while laxation did not occur in any patient who received placebo. By 4 hours posttreatment, 7 patients (38.9%; 95% CI, 17.30‐64.25) in the methylnaltrexone group achieved laxation compared with only 1 patient (6.7%; 95% CI, 0.17‐31.95) on placebo. Three patients in each treatment group received rescue laxatives.

Figure 3

The time to first laxation (Figure 4) was significantly shorter in patients who received methylnaltrexone compared with those in the placebo group. Patients on methylnaltrexone achieved laxation in a median time of 15.8 hours, compared with a median time of 50.9 hours for patients in the placebo group (P = 0.02, log‐rank test). The median time to laxation was less than 1 hour in the 7 methylnaltrexone‐treated patients who experienced laxation within 4 hours following the first dose. Of the remaining 11 methylnaltrexone‐treated patients, one experienced no laxation after 6 doses, and the median time to laxation for the others was 29.9 hours (not shown in figure).

Figure 4

Analysis of the Global Satisfaction With Treatment Scale revealed that more patients expressed overall treatment satisfaction (defined as very satisfied, satisfied, or minimally satisfied) with methylnaltrexone assessed 4 hours ( 30 minutes) after the first dose, compared with patients on placebo (83.3% vs 60.0%, respectively). At the study endpoint, overall treatment satisfaction with methylnaltrexone remained high (83.3%), whereas satisfaction with placebo was 53.3%. Additionally, no patients in the methylnaltrexone group expressed any dissatisfaction with treatment (defined as minimally dissatisfied, dissatisfied, or very dissatisfied) at endpoint, compared with 26.7% of patients in the placebo group who expressed some degree of dissatisfaction.

DISCUSSION

This pilot study suggests that methylnaltrexone actively induces laxation and is generally well tolerated in patients receiving mu‐opioid analgesia, following orthopedic surgery, who develop OIC acutely. It was the first study, to our knowledge, to investigate the efficacy of methylnaltrexone for the treatment of OIC in an acute postoperative setting. The protocol amendment changing the duration of treatment from 7 days to 4 days did not materially affect the results of the study. The response to methylnaltrexone was rapid, with 33.3% experiencing laxation within 2 hours. The median time to laxation was nearly 1.5 days shorter in patients treated with methylnaltrexone compared with those receiving placebo. Correspondingly, overall patient satisfaction was high in the methylnaltrexone group. Efficacy was attained without diminishing opioid analgesia, and without inducing signs or symptoms of opioid withdrawal. The incidence of AEs was similar between groups, and no treatment‐emergent SAEs were reported in this study.

Previous clinical trials investigated the safety and efficacy of methylnaltrexone for the treatment of OIC in patients with advanced illness and with chronic, nonmalignant pain. The present study extends those findings to a population of patients experiencing acute OIC following orthopedic surgery. Previous studies showed that approximately 48% to 62% of advanced‐illness patients experienced laxation within 4 hours of receiving SC methylnaltrexone,17, 18 compared with 38.9% of acute OIC patients in this study. In a clinical trial of patients with chronic, nonmalignant pain, 34.2% of patients experienced laxation within 4 hours of SC methylnaltrexone injection (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA). The differences in laxation response between these trials may be attributable to differences in the patient populations or to methodologic differences between the studies.

Similar to findings demonstrated in a clinical study evaluating methylnaltrexone for OIC in a different patient population, those with advanced illness,22 this study supports the premise that future laxation response with prolonged use is most likely to occur when a laxation response was achieved after the first or second initial administrations of methylnaltrexone. In contradistinction, if laxation does not occur with these early doses, continued methylnaltrexone dosing is less likely to produce a response later.

This study has some limitations that must be considered. First, as this was a hypothesis‐generating study, all efficacy parameters investigated were exploratory in nature. The results reported herein warrant careful consideration, owing to a small sample size that may limit their generalizability, prior to replication in a more rigorously designed study with prespecified efficacy endpoints. Likewise, the assessment of health outcome parameters is limited. Another limitation is the small sample size utilized in this study, potentially resulting in a type II error.

Subcutaneous administration potentially offers a considerable benefit over oral therapies for OIC in this patient population post‐orthopedic surgery. Nausea and vomiting can occur as a consequence of anesthesia and of postoperative opioid analgesia, and may compromise adequate dosing of oral medications prescribed to treat OIC. Subcutaneous delivery of methylnaltrexone may circumvent this potential drawback while providing potentially rapid, effective treatment for OIC. Once‐daily dosing may also help to minimize caregiver burden and patient discomfort by preventing the need for more frequent or unpleasant treatments for OIC, such as enemas.

This study provides an initial positive signal for a broader, albeit off‐label use for methylnaltrexonethat being for the treatment of acute constipation that occurs as a consequence of postoperative opioid‐mediated analgesia in patients following orthopedic procedures. Adequate treatment of OIC, even in the acute postoperative setting, is likely to lead to better overall pain management and improved patient outcomes. Additionally, effective management of acute OIC is likely to be cost‐effective in terms of reducing the duration of hospital stays, reducing the need for nursing resources and the time spent administering rescue treatments for OIC (eg, enemas), and avoiding returns to an acute setting (eg, the emergency department) for treatment. The results presented herein suggest that methylnaltrexone may be effective and have a good safety profile in the treatment of acute OIC following orthopedic surgery. Validation of these results in larger well‐controlled trials would be welcome.

The management of postoperative pain is essential to perioperative care, and adequate postoperative analgesia has been associated with several key clinical benefits, including fewer postoperative complications, earlier patient ambulation, reduced costs due to shorter hospital stays, and improved rehabilitation.1, 2 While opioids have long been central to postoperative analgesia, they have been associated with various adverse effects, including sedation, dizziness, nausea, vomiting, constipation, dependence, tolerance, and respiratory depression.2, 3 Constipation, one of the most common adverse effects resulting from opioid therapy, can be debilitating. Indeed, opioid effects on gut motility can occur even after a single dose.3 The consequences of opioid‐induced constipation (OIC) may be severe enough to warrant a dosage reduction of the opioid; however, this may lead to compromised analgesia, which can hinder recovery.4, 5 Thus, effective treatment of OIC is an important clinical consideration in patients undergoing pain management with opioids. Unfortunately, laxatives and other treatment strategies can have unpredictable or suboptimal results for many patients with OIC; therefore, other options are needed for the treatment of OIC.6, 7

Opioid receptor agonists cause constipation by adversely altering many aspects of intestinal function, including fluid dynamics, gastric emptying, propulsive motor activity, and transit time.3 Opioid receptors are widely distributed in the central nervous system and throughout the intestinal system. The mechanism of OIC may have both peripherally and centrally mediated components.8 Nonselective opioid receptor antagonists block the undesired effects on the gut, but because they cross the blood‐brain barrier, they also interfere with analgesia and may lead to symptoms of withdrawal. Methylnaltrexone is a selective, peripherally acting mu‐opioid receptor antagonist,9 formed by the addition of a methyl group to the amine ring of the mu‐opioid receptor antagonist naltrexone. The resulting quarternary amine has greater polarity, lower lipid solubility, and restricted ability to cross the blood‐brain barrier.10 Thus, methylnaltrexone was designed to decrease the peripheral adverse effects of opioids without interfering with centrally mediated analgesia.

Investigations of methylnaltrexone effects in healthy volunteers showed that methylnaltrexone attenuated morphine‐induced delays in gastric emptying and oral‐cecal transit without affecting analgesia.1113 Further studies of methylnaltrexone for the treatment of constipation due to methadone use demonstrated rapid laxation response.1416 Two randomized, double‐blind, placebo‐controlled studies of methylnaltrexone in 288 patients with advanced illness and OIC showed that methylnaltrexone rapidly induced laxation without compromising analgesia.17, 18 Methylnaltrexone is currently approved for the treatment of OIC in patients with advanced illness who are receiving palliative care, when response to laxative therapy has not been sufficient.19

Recently, the use of methylnaltrexone for the treatment of OIC in patients with chronic, nonmalignant pain was assessed in a randomized, double‐blind, placebo‐controlled trial of more than 400 patients. Investigators found that methylnaltrexone induced laxation and was generally well tolerated (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA; Duerden et al., 29th Annual Scientific Meeting of the American Pain Society, May 6‐10, 2010, Baltimore, MD), supporting the safety and efficacy of methylnaltrexone in the setting of OIC resulting from chronic opioid treatment. The present study aimed to assess the activity of methylnaltrexone in patients receiving mu‐agonist opioid analgesics during rehabilitation, following an orthopedic surgical procedure, who were experiencing acute OIC.

METHODS

Study Design

This randomized, double‐blind, placebo‐controlled, parallel‐group, hypothesis‐generating phase 2 study was conducted from October 2007 to January 2009 at 16 US hospitals and rehabilitation facilities in accordance with the International Conference on Harmonisation Good Clinical Practice Guidelines and the Declaration of Helsinki, and was approved by the Institutional Review Board and/or Independent Ethics Committee at each of the participating investigational centers. All patients provided written informed consent prior to study participation.

Eligible patients were randomized by interactive voice response system in a 1:1 ratio to receive once‐daily subcutaneous (SC) injections of either 12 mg methylnaltrexone or placebo (Figure 1). The chosen 12‐mg unit dosing corresponds to approximately 0.15 mg/kg (assuming an 80‐kg patient) and was found to be both efficacious and well tolerated in the treatment of OIC in prior studies, including studies in advanced‐illness patients17, 18 and in patients with chronic, nonmalignant pain (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA; Duerden et al., 29th Annual Scientific Meeting of the American Pain Society, May 6‐10, 2010, Baltimore, MD.20 The first dose of study medication was administered on the day of randomization or on the next calendar day. Once enrolled, the patient received once‐daily doses of methylnaltrexone for up to 4 or 7 days. Dosing continued until the patient received the maximum number of doses allowed, no longer needed opioid medication, or was discharged from the medical facility. Each patient completed a follow‐up safety visit at 14 3 days following the last dose.

Figure 1

RESULTS

Patient Populations

The flow of patients through the study is summarized in Figure 2. A total of 51 patients were enrolled. Of these, 33 received at least 1 dose of study treatment following double‐blind randomization and comprised both the mITT and safety populations. Seventeen of these patients were enrolled under the original protocol and could receive study drug for up to 7 days, while 16 patients enrolled under a subsequent protocol revision could receive study drug for up to 4 days. This change from a 7‐day to a 4‐day treatment protocol allowed for the capture of more study patients in view of the time pressures of short lengths of stay in postoperative settings. In total, 31 patients received at least 2 doses, and 26 patients received at least 4 doses of study drug. A total of 27 patients completed the study. Baseline demographics and prestudy surgical procedures were similar in both treatment groups (Table 1).

Figure 2

Table 1.Baseline Patient Demographics

Characteristic	Methylnaltrexone (n = 18)	Placebo (n = 15)
NOTE: Modified intent‐to‐treat population. Abbreviations: BMI, body mass index.; SD, standard deviation. * Baseline opioid use was defined as total opioid use within 24 hours prior to randomization.
Mean age, yr (SD)	64.2 (9.0)	65.2 (11.6)
Mean weight, kg (SD)	92.5 (22.5)	91.0 (20.2)
Mean BMI, kg/m2 (SD)	32.3 (7.2)	34.2 (6.41)
Sex, n (%)
Female	11 (61.1)	11 (73.3)
Male	7 (38.9)	4 (26.7)
Race, n (%)
White	14 (77.8)	10 (66.7)
Black	4 (22.2)	5 (33.3)
Type of surgery, n (%)
Total knee replacement	8 (44.4)	7 (46.7)
Total hip replacement	6 (33.3)	6 (40.0)
Spinal fusion	2 (11.1)	0
Fracture reduction	2 (11.1)	2 (13.3)
Median opioid use,* mg (range)	28.00 (6.75‐168.01)	25.00 (9.00‐75.00)
Median time from surgery to study drug administration, days (range)	4 (3‐6)	4 (3‐6)

Efficacy

A significantly greater percentage of patients had a bowel movement within 2 hours (P = 0.021) and 4 hours (P = 0.046) of the first dose of methylnaltrexone compared with patients who received placebo (Figure 3). Within 2 hours, 6 patients (33.3%; 95% confidence interval [CI], 13.34‐59.01) who received methylnaltrexone achieved laxation, while laxation did not occur in any patient who received placebo. By 4 hours posttreatment, 7 patients (38.9%; 95% CI, 17.30‐64.25) in the methylnaltrexone group achieved laxation compared with only 1 patient (6.7%; 95% CI, 0.17‐31.95) on placebo. Three patients in each treatment group received rescue laxatives.

Figure 3

The time to first laxation (Figure 4) was significantly shorter in patients who received methylnaltrexone compared with those in the placebo group. Patients on methylnaltrexone achieved laxation in a median time of 15.8 hours, compared with a median time of 50.9 hours for patients in the placebo group (P = 0.02, log‐rank test). The median time to laxation was less than 1 hour in the 7 methylnaltrexone‐treated patients who experienced laxation within 4 hours following the first dose. Of the remaining 11 methylnaltrexone‐treated patients, one experienced no laxation after 6 doses, and the median time to laxation for the others was 29.9 hours (not shown in figure).

Figure 4

Analysis of the Global Satisfaction With Treatment Scale revealed that more patients expressed overall treatment satisfaction (defined as very satisfied, satisfied, or minimally satisfied) with methylnaltrexone assessed 4 hours ( 30 minutes) after the first dose, compared with patients on placebo (83.3% vs 60.0%, respectively). At the study endpoint, overall treatment satisfaction with methylnaltrexone remained high (83.3%), whereas satisfaction with placebo was 53.3%. Additionally, no patients in the methylnaltrexone group expressed any dissatisfaction with treatment (defined as minimally dissatisfied, dissatisfied, or very dissatisfied) at endpoint, compared with 26.7% of patients in the placebo group who expressed some degree of dissatisfaction.

DISCUSSION

This pilot study suggests that methylnaltrexone actively induces laxation and is generally well tolerated in patients receiving mu‐opioid analgesia, following orthopedic surgery, who develop OIC acutely. It was the first study, to our knowledge, to investigate the efficacy of methylnaltrexone for the treatment of OIC in an acute postoperative setting. The protocol amendment changing the duration of treatment from 7 days to 4 days did not materially affect the results of the study. The response to methylnaltrexone was rapid, with 33.3% experiencing laxation within 2 hours. The median time to laxation was nearly 1.5 days shorter in patients treated with methylnaltrexone compared with those receiving placebo. Correspondingly, overall patient satisfaction was high in the methylnaltrexone group. Efficacy was attained without diminishing opioid analgesia, and without inducing signs or symptoms of opioid withdrawal. The incidence of AEs was similar between groups, and no treatment‐emergent SAEs were reported in this study.

Previous clinical trials investigated the safety and efficacy of methylnaltrexone for the treatment of OIC in patients with advanced illness and with chronic, nonmalignant pain. The present study extends those findings to a population of patients experiencing acute OIC following orthopedic surgery. Previous studies showed that approximately 48% to 62% of advanced‐illness patients experienced laxation within 4 hours of receiving SC methylnaltrexone,17, 18 compared with 38.9% of acute OIC patients in this study. In a clinical trial of patients with chronic, nonmalignant pain, 34.2% of patients experienced laxation within 4 hours of SC methylnaltrexone injection (Blonsky et al., 28th Annual Scientific Meeting of the American Pain Society, May 7‐9, 2009, San Diego, CA). The differences in laxation response between these trials may be attributable to differences in the patient populations or to methodologic differences between the studies.

Similar to findings demonstrated in a clinical study evaluating methylnaltrexone for OIC in a different patient population, those with advanced illness,22 this study supports the premise that future laxation response with prolonged use is most likely to occur when a laxation response was achieved after the first or second initial administrations of methylnaltrexone. In contradistinction, if laxation does not occur with these early doses, continued methylnaltrexone dosing is less likely to produce a response later.

This study has some limitations that must be considered. First, as this was a hypothesis‐generating study, all efficacy parameters investigated were exploratory in nature. The results reported herein warrant careful consideration, owing to a small sample size that may limit their generalizability, prior to replication in a more rigorously designed study with prespecified efficacy endpoints. Likewise, the assessment of health outcome parameters is limited. Another limitation is the small sample size utilized in this study, potentially resulting in a type II error.

Subcutaneous administration potentially offers a considerable benefit over oral therapies for OIC in this patient population post‐orthopedic surgery. Nausea and vomiting can occur as a consequence of anesthesia and of postoperative opioid analgesia, and may compromise adequate dosing of oral medications prescribed to treat OIC. Subcutaneous delivery of methylnaltrexone may circumvent this potential drawback while providing potentially rapid, effective treatment for OIC. Once‐daily dosing may also help to minimize caregiver burden and patient discomfort by preventing the need for more frequent or unpleasant treatments for OIC, such as enemas.

This study provides an initial positive signal for a broader, albeit off‐label use for methylnaltrexonethat being for the treatment of acute constipation that occurs as a consequence of postoperative opioid‐mediated analgesia in patients following orthopedic procedures. Adequate treatment of OIC, even in the acute postoperative setting, is likely to lead to better overall pain management and improved patient outcomes. Additionally, effective management of acute OIC is likely to be cost‐effective in terms of reducing the duration of hospital stays, reducing the need for nursing resources and the time spent administering rescue treatments for OIC (eg, enemas), and avoiding returns to an acute setting (eg, the emergency department) for treatment. The results presented herein suggest that methylnaltrexone may be effective and have a good safety profile in the treatment of acute OIC following orthopedic surgery. Validation of these results in larger well‐controlled trials would be welcome.

Obesity affects all segments of the American population. However, it imposes a larger burden and poses a greater threat to minority populations.1 The increase in overall prevalence of obesity and obesity‐related diseases are especially pronounced in ethnic minorities,2 and the outlook for minorities who develop obesity‐associated diseases such as stroke and chronic renal disease is worse than in their Caucasian counterparts.3

Despite the higher prevalence of obesity in ethnic minorities, the majority of research on the relationship between body mass index (BMI) and mortality has been conducted among Caucasians in the United States. This is due largely to the small number of minority participants in most studies, which makes for low statistical power.4

A curious epidemiologic paradox has been the observation, in many studies, that black adults have lower morbidity and mortality associated with obesity compared to Caucasians.5 In fact, some authors suggest that high BMI among black and other minorities may not be as strong a risk factor for mortality as it is in others.6, 7

Very few studies have specifically examined the contribution of BMI to postoperative outcome in a large cohort of minority patients.8, 9 Similarly, the authors are unaware of any previous studies describing the clinical relevance of being overweight and obese in minority patients undergoing surgery. Therefore, the primary objective of this observational study was to describe the prevalence of overweight and obesity in a large cohort of minority surgical patients, and the impact of BMI class on their postoperative outcome. Our a priori hypothesis was that obese minority surgical patients would have a poorer postoperative outcome and have higher 30‐day all‐cause mortality than normal weight individuals.

METHODS

Study data were derived from the Participant Use Data File of the American College of Surgeons (ACS) National Surgical Quality Improvement Program (NSQIP) for the calendar years 2005 to 2008. This multi‐institutional (186 participating centers) reporting system was designed to provide risk‐adjusted surgical outcome data from throughout the United States.

The methodology for collecting these data, including their accuracy and reproducibility, has been detailed in previous publications.10, 11 It is briefly reviewed here. Dedicated nurse clinical reviewers at each hospital prospectively enrolled patients and collected data in a standardized fashion according to strict ACS‐NSQIP definitions. A systematic sample was obtained by taking the first 40 cases per nurse reviewer on an 8‐day cycle from the operating room log, ensuring that no particular operating room day block time would bias the weighting of cases. Nurse reviewers had completed comprehensive training regarding definitions and data extraction, as well as continuing education and monitoring through the ACS‐NSQIP program. They are assessed for inter‐rater reliability during biennial site visits. Information was obtained from patient medical records, physician office records, and telephone interviews. Patients were followed through their hospital course and after discharge from hospital up to 30 days postoperatively. A high level of accuracy and reproducibility of the data have been previously demonstrated.12

Race was defined as African American, Hispanic, Asian or Pacific Islander, or American Indian/Alaskan native, as identified by the clinical care provider, or within the medical record. Patients were excluded if race was coded as white, or not recorded. We also excluded patients with missing record of height and/or weight. The final study cohort consisted of 119,619 minority patients. We then computed BMI as weight in kilograms divided by the square of the height in meters (BMI = kg/m²). Patients were classified as underweight (BMI 18.5 kg/m²), normal weight (BMI = 18.6‐24.9 kg/m²), overweight (BMI = 25‐29.9 kg/m²), obese (BMI = 30‐39.9 kg/m²), and morbidly obese (BMI 40 kg/m²) in accordance with National Institute of Health (NIH) standards.13

RESULTS

The study cohort included 119,619 surgical patients (Table 1). Their mean (standard deviation [SD]) age was 50.4 (16.9), and the mean (SD) BMI of 30.3 (8.9) kg/m² was in the obese range. The overall prevalence of high BMI (overweight or obese) was 70.8%. A very small proportion, 2.7%, of patients was classified as underweight. Sex‐stratified demographic and behavioral characteristics differed significantly across all the variables in our study cohort. Men were more likely to be overweight, whereas women were more likely to be obese or severely obese. A majority (77.4%) of the patients were non‐elderly adults (<65years) undergoing elective (85.9%) surgical procedures. The minority patients in this study were African American (50%), Hispanic (36%), Asian and Pacific Islander (10%), and American Indian and Alaskan native (4%).

The distribution of baseline preoperative clinical characteristics by BMI class revealed many significant differences (Table 2). Age was significantly different among the BMI classes, with the severely obese group being about 8 years younger than the underweight or normal weight group. Similarly, severely obese patients were more likely to be women, less likely to smoke, more likely to be hypertensive, diabetic, have a history of dyspnea at rest, and more likely to belong to high ASA class. On the other hand, underweight patients were more likely to have disseminated cancer, be current smokers, consume more than 2 alcoholic drinks per day, have active chronic obstructive pulmonary disease (COPD), and have ascites. They were also more likely to be on dialysis and have cardiac disease, as well as a history of stroke. Urgency of surgery also varied significantly across the BMI categories, with the underweight group having the highest incidence of emergency surgery (20.6%) and the severely obese group being the least likely to present for emergency surgery (8.2%).

Perioperative outcomes according to BMI classes documented significant differences (Table 3). Work relative value unit (Work RVU, a measure of surgical complexity), as well as total anesthesia and operation time decreased in a stepwise fashion across the BMI classes to the obese group, followed by increase in these parameters in the severely obese group. Following a decrease to the normal BMI category, there was a positive association between BMI and the incidence of postoperative superficial and deep wound infection, as well as wound disruption.

There was a negative association between BMI class and the likelihood of postoperative sepsis, septic shock, reintubation, delayed ventilator wean, and postoperative pneumonia. Similarly, the proportions of patients who developed postoperative acute renal failure, cardiac arrest, and those who required postoperative blood transfusion or needed reoperation, decreased significantly across the BMI classes, with the highest proportion of cases being in the underweight group and the lowest in the severely obese group. Overall composite morbidity was twice as high in the underweight compared to the severely obese group.

There were 1758 deaths among the study's 119,619 patients, resulting in an overall mortality rate of 1.5%. The overall major complication rate was 13.8%. The distribution of total mortality rate as well as mortality in patients with at least 1 major postoperative complication across BMI classes revealed consistent differences (Figure 1). Over the entire range of BMI classes, there was a progressive, stepwise decrease in the proportion of deaths with increasing BMI. This pattern also occurred among patients who developed at least 1 major postoperative complication, indicating a reduced likelihood of death after a major complication. This is reflective of a reduced likelihood of death after a major complication (failure to rescue) with increasing BMI.

Figure 1

Multivariate logistic regression defined a number of factors associated with 30‐day mortality (Table 4). The Hosmer and Lemeshow goodness‐of‐fit test for this model was not statistically significant (² = 17.8, df = 8, P = 0.23). High ASA physical status was associated with high odds of mortality. Specifically, when controlling for the other covariates in the model, ASA status was associated with a 5‐fold increased relative odds of death (adjusted odds ratio [OR] = 5.30; 95% confidence interval = 4.96‐5.79, P < 0.001). Similarly, occurrence of 1 or more major postoperative complication was associated with 6‐fold increased relative odds of mortality. The paradoxical effect of BMI category observed on univariate analysis was maintained in the multivariate model. Specifically, underweight patients had the highest relative odds of mortality, while severely obese patients had the lowest, compared with patients at a healthy weight (Table 4). Interestingly, smoking had no significant effect on the odds of mortality after controlling for other factors. Similarly, the specific racial group and the timing of the surgical intervention had no significant effect on mortality.

DISCUSSION

In this large, study of minority surgical patients, the impact of BMI on the 30‐day morbidity and mortality was unexpected. The working hypothesis was that overweight and obese patients would have a worse outcome after surgery. However, contrary to this hypothesis, the lowest all‐cause mortality rate was found in the severely obese (BMI 40 kg/m²) group in both men and women. Death rates decreased progressively in a stepwise fashion from the underweight to the severely obese group. Similarly, even in patients who developed at least 1 major postoperative complication, the likelihood of death was still negatively associated with BMI. This negative association of mortality with BMI was observed despite the higher prevalence of chronic diseases, such as hypertension, diabetes, and dyspnea at rest, in the increasing BMI classes.

Controversy remains regarding the association between BMI and mortality, in particular about the shape of the curve for the association between BMI and mortality. Epidemiologic studies have variously described J‐shaped, U‐shaped, monotonic, or linear relationships.16, 17 In the surgical population, a reverse J‐shaped relationship between BMI category and mortality has been described.18, 19 Sometimes this is referred to as obesity paradox or reverse epidemiology: a trend whereby overweight and moderately obese patients have better outcomes and lower risk of death than leaner patients.18 This phenomenon is particularly well described in adult20 and elderly heart failure and hypertensive patients.21 Many of these studies either had very few minority patients,21 or mortality pattern was not analyzed along ethnic lines.

Few studies10 have focused exclusively on minority surgical patients. Some investigators have determined that high BMI in black adults may not be as important a risk factor for mortality6, 7 as in whites. Our data suggest that among minority surgical patients, the relationship appeared to be a downward trend in mortality from low to high BMI, thus revealing the obesity paradox. This pattern was evident even in patients who developed 1 major complication in the postoperative period, suggesting that high BMI also protects against failure to rescueor death after a major complication.

Despite decades of research, the mechanisms underlying the obesity paradox remain speculative.20, 22. Many have posited that adiposity may confer protection against cytokines and various inflammatory mediators in heart failure patients by the production of buffering lipoproteins.23, 24 It is conceivable that similar protection against inflammatory response to surgical tissue trauma is operational in minority patients with high BMI. Another possible reason for the obesity paradox is the clinical presentation and disease progression at the time of surgery. Perhaps, similar to the observation in obese patients with heart failure,25 obese minority patients are symptomatic at an earlier stage of their disease than lean patients, making for earlier diagnosis and treatment. Thus, obesity may simply be a marker of less severe disease at the time of presentation.

Obese patients may also be more aggressively monitored and treated in the perioperative period than lean patients, because of the general perception that they are a high‐risk group.10 This may partly explain the decreased likelihood of failure to rescue with increasing BMI in our patients. Increased vigilance and prompt treatment of complications should reduce the overall morbidity and mortality rate in this group. It is also conceivable that a therapeutic selection bias is operational in the patients we studied. This describes scenarios where relatively healthy obese minority patients were operated upon, while sicker, morbidly obese patients were denied surgery due to perceived prohibitive risks. However, we would have expected a higher proportion of severely obese patients to present for emergency surgery, which is contrary to our finding of the lowest incidence of emergency surgery in the severely obese group. It is also possible that severe obesity may be associated with a higher attrition rate, such that the extremely obese patients did not live long enough to present for surgery. This is somewhat likely, given the significantly younger age of the severely obese patients in our study cohort. It is, however, impossible to determine survival effect from a cross‐sectional hospital‐based study design. Clearly, mechanisms used to explain the obesity paradox in minority surgical patients are likely to remain speculative, owing to the interaction of several factors such as concomitant comorbidities, disease progression at the time of presentation, patients' weight history, and regional fat distribution.

The current study confirms the findings of previous investigators26 about the importance of reducing major postoperative complications in surgical patients. While this may seem axiomatic, it deserves reiteration because the risk of postoperative mortality increases considerably in all the BMI categories following 1 or more major postoperative complication. However, it is not clear why obese and morbidly obese patients had a lower incidence of failure to rescue. This may be related to greater physiologic reserve in the obese and morbidly obese group, especially because patients in the higher BMI groups were significantly younger than the normal weight or lean patients. For the same reason, these younger, severely obese patients may have been more aggressively monitored and treated, thereby increasing the likelihood of being rescued following a major complication. It is also possible that the lower proportion of emergency procedures performed in obese and severely obese groups was somewhat protective, especially because emergency surgery was an independent predictor of overall mortality in this cohort of patients. In fact, when we stratified the patients according to urgency of surgery and explored the bivariate relationship between BMI category and mortality (data not shown) among those undergoing urgent surgery, the geometrical distribution of mortality did show a reverse‐J pattern with the highest proportion of cases in the underweight group, declining in the normal BMI and overweight group, and increasing steadily in obese and severely obese group. To this end, caution should be exercised when interpreting the association of BMI group with postoperative mortality for procedures performed as an emergency.

Smoking and antecedent illness are 2 confounding factors commonly criticized in studies attempting to associated BMI with mortality. This is because smokers tend to weigh less and have higher mortality rates than nonsmokers. The present investigation did not find a significant contribution of smoking to mortality when other factors included in a logistic regression model were considered. The current study's findings are consistent with those of previous data in African American patients,27 and contrasts with the excess mortality described in currently smoking Caucasian men and women.28 It is possible that smoking is not an important effect modifier when considering the relevance of BMI to postoperative mortality in minority patients.

Obesity affects all segments of the American population. However, it imposes a larger burden and poses a greater threat to minority populations.1 The increase in overall prevalence of obesity and obesity‐related diseases are especially pronounced in ethnic minorities,2 and the outlook for minorities who develop obesity‐associated diseases such as stroke and chronic renal disease is worse than in their Caucasian counterparts.3

Despite the higher prevalence of obesity in ethnic minorities, the majority of research on the relationship between body mass index (BMI) and mortality has been conducted among Caucasians in the United States. This is due largely to the small number of minority participants in most studies, which makes for low statistical power.4

A curious epidemiologic paradox has been the observation, in many studies, that black adults have lower morbidity and mortality associated with obesity compared to Caucasians.5 In fact, some authors suggest that high BMI among black and other minorities may not be as strong a risk factor for mortality as it is in others.6, 7

Very few studies have specifically examined the contribution of BMI to postoperative outcome in a large cohort of minority patients.8, 9 Similarly, the authors are unaware of any previous studies describing the clinical relevance of being overweight and obese in minority patients undergoing surgery. Therefore, the primary objective of this observational study was to describe the prevalence of overweight and obesity in a large cohort of minority surgical patients, and the impact of BMI class on their postoperative outcome. Our a priori hypothesis was that obese minority surgical patients would have a poorer postoperative outcome and have higher 30‐day all‐cause mortality than normal weight individuals.

METHODS

Study data were derived from the Participant Use Data File of the American College of Surgeons (ACS) National Surgical Quality Improvement Program (NSQIP) for the calendar years 2005 to 2008. This multi‐institutional (186 participating centers) reporting system was designed to provide risk‐adjusted surgical outcome data from throughout the United States.

The methodology for collecting these data, including their accuracy and reproducibility, has been detailed in previous publications.10, 11 It is briefly reviewed here. Dedicated nurse clinical reviewers at each hospital prospectively enrolled patients and collected data in a standardized fashion according to strict ACS‐NSQIP definitions. A systematic sample was obtained by taking the first 40 cases per nurse reviewer on an 8‐day cycle from the operating room log, ensuring that no particular operating room day block time would bias the weighting of cases. Nurse reviewers had completed comprehensive training regarding definitions and data extraction, as well as continuing education and monitoring through the ACS‐NSQIP program. They are assessed for inter‐rater reliability during biennial site visits. Information was obtained from patient medical records, physician office records, and telephone interviews. Patients were followed through their hospital course and after discharge from hospital up to 30 days postoperatively. A high level of accuracy and reproducibility of the data have been previously demonstrated.12

Race was defined as African American, Hispanic, Asian or Pacific Islander, or American Indian/Alaskan native, as identified by the clinical care provider, or within the medical record. Patients were excluded if race was coded as white, or not recorded. We also excluded patients with missing record of height and/or weight. The final study cohort consisted of 119,619 minority patients. We then computed BMI as weight in kilograms divided by the square of the height in meters (BMI = kg/m²). Patients were classified as underweight (BMI 18.5 kg/m²), normal weight (BMI = 18.6‐24.9 kg/m²), overweight (BMI = 25‐29.9 kg/m²), obese (BMI = 30‐39.9 kg/m²), and morbidly obese (BMI 40 kg/m²) in accordance with National Institute of Health (NIH) standards.13

RESULTS

The study cohort included 119,619 surgical patients (Table 1). Their mean (standard deviation [SD]) age was 50.4 (16.9), and the mean (SD) BMI of 30.3 (8.9) kg/m² was in the obese range. The overall prevalence of high BMI (overweight or obese) was 70.8%. A very small proportion, 2.7%, of patients was classified as underweight. Sex‐stratified demographic and behavioral characteristics differed significantly across all the variables in our study cohort. Men were more likely to be overweight, whereas women were more likely to be obese or severely obese. A majority (77.4%) of the patients were non‐elderly adults (<65years) undergoing elective (85.9%) surgical procedures. The minority patients in this study were African American (50%), Hispanic (36%), Asian and Pacific Islander (10%), and American Indian and Alaskan native (4%).

The distribution of baseline preoperative clinical characteristics by BMI class revealed many significant differences (Table 2). Age was significantly different among the BMI classes, with the severely obese group being about 8 years younger than the underweight or normal weight group. Similarly, severely obese patients were more likely to be women, less likely to smoke, more likely to be hypertensive, diabetic, have a history of dyspnea at rest, and more likely to belong to high ASA class. On the other hand, underweight patients were more likely to have disseminated cancer, be current smokers, consume more than 2 alcoholic drinks per day, have active chronic obstructive pulmonary disease (COPD), and have ascites. They were also more likely to be on dialysis and have cardiac disease, as well as a history of stroke. Urgency of surgery also varied significantly across the BMI categories, with the underweight group having the highest incidence of emergency surgery (20.6%) and the severely obese group being the least likely to present for emergency surgery (8.2%).

Perioperative outcomes according to BMI classes documented significant differences (Table 3). Work relative value unit (Work RVU, a measure of surgical complexity), as well as total anesthesia and operation time decreased in a stepwise fashion across the BMI classes to the obese group, followed by increase in these parameters in the severely obese group. Following a decrease to the normal BMI category, there was a positive association between BMI and the incidence of postoperative superficial and deep wound infection, as well as wound disruption.

There was a negative association between BMI class and the likelihood of postoperative sepsis, septic shock, reintubation, delayed ventilator wean, and postoperative pneumonia. Similarly, the proportions of patients who developed postoperative acute renal failure, cardiac arrest, and those who required postoperative blood transfusion or needed reoperation, decreased significantly across the BMI classes, with the highest proportion of cases being in the underweight group and the lowest in the severely obese group. Overall composite morbidity was twice as high in the underweight compared to the severely obese group.

There were 1758 deaths among the study's 119,619 patients, resulting in an overall mortality rate of 1.5%. The overall major complication rate was 13.8%. The distribution of total mortality rate as well as mortality in patients with at least 1 major postoperative complication across BMI classes revealed consistent differences (Figure 1). Over the entire range of BMI classes, there was a progressive, stepwise decrease in the proportion of deaths with increasing BMI. This pattern also occurred among patients who developed at least 1 major postoperative complication, indicating a reduced likelihood of death after a major complication. This is reflective of a reduced likelihood of death after a major complication (failure to rescue) with increasing BMI.

Figure 1

Multivariate logistic regression defined a number of factors associated with 30‐day mortality (Table 4). The Hosmer and Lemeshow goodness‐of‐fit test for this model was not statistically significant (² = 17.8, df = 8, P = 0.23). High ASA physical status was associated with high odds of mortality. Specifically, when controlling for the other covariates in the model, ASA status was associated with a 5‐fold increased relative odds of death (adjusted odds ratio [OR] = 5.30; 95% confidence interval = 4.96‐5.79, P < 0.001). Similarly, occurrence of 1 or more major postoperative complication was associated with 6‐fold increased relative odds of mortality. The paradoxical effect of BMI category observed on univariate analysis was maintained in the multivariate model. Specifically, underweight patients had the highest relative odds of mortality, while severely obese patients had the lowest, compared with patients at a healthy weight (Table 4). Interestingly, smoking had no significant effect on the odds of mortality after controlling for other factors. Similarly, the specific racial group and the timing of the surgical intervention had no significant effect on mortality.

DISCUSSION

In this large, study of minority surgical patients, the impact of BMI on the 30‐day morbidity and mortality was unexpected. The working hypothesis was that overweight and obese patients would have a worse outcome after surgery. However, contrary to this hypothesis, the lowest all‐cause mortality rate was found in the severely obese (BMI 40 kg/m²) group in both men and women. Death rates decreased progressively in a stepwise fashion from the underweight to the severely obese group. Similarly, even in patients who developed at least 1 major postoperative complication, the likelihood of death was still negatively associated with BMI. This negative association of mortality with BMI was observed despite the higher prevalence of chronic diseases, such as hypertension, diabetes, and dyspnea at rest, in the increasing BMI classes.

Controversy remains regarding the association between BMI and mortality, in particular about the shape of the curve for the association between BMI and mortality. Epidemiologic studies have variously described J‐shaped, U‐shaped, monotonic, or linear relationships.16, 17 In the surgical population, a reverse J‐shaped relationship between BMI category and mortality has been described.18, 19 Sometimes this is referred to as obesity paradox or reverse epidemiology: a trend whereby overweight and moderately obese patients have better outcomes and lower risk of death than leaner patients.18 This phenomenon is particularly well described in adult20 and elderly heart failure and hypertensive patients.21 Many of these studies either had very few minority patients,21 or mortality pattern was not analyzed along ethnic lines.

Few studies10 have focused exclusively on minority surgical patients. Some investigators have determined that high BMI in black adults may not be as important a risk factor for mortality6, 7 as in whites. Our data suggest that among minority surgical patients, the relationship appeared to be a downward trend in mortality from low to high BMI, thus revealing the obesity paradox. This pattern was evident even in patients who developed 1 major complication in the postoperative period, suggesting that high BMI also protects against failure to rescueor death after a major complication.

Despite decades of research, the mechanisms underlying the obesity paradox remain speculative.20, 22. Many have posited that adiposity may confer protection against cytokines and various inflammatory mediators in heart failure patients by the production of buffering lipoproteins.23, 24 It is conceivable that similar protection against inflammatory response to surgical tissue trauma is operational in minority patients with high BMI. Another possible reason for the obesity paradox is the clinical presentation and disease progression at the time of surgery. Perhaps, similar to the observation in obese patients with heart failure,25 obese minority patients are symptomatic at an earlier stage of their disease than lean patients, making for earlier diagnosis and treatment. Thus, obesity may simply be a marker of less severe disease at the time of presentation.

Obese patients may also be more aggressively monitored and treated in the perioperative period than lean patients, because of the general perception that they are a high‐risk group.10 This may partly explain the decreased likelihood of failure to rescue with increasing BMI in our patients. Increased vigilance and prompt treatment of complications should reduce the overall morbidity and mortality rate in this group. It is also conceivable that a therapeutic selection bias is operational in the patients we studied. This describes scenarios where relatively healthy obese minority patients were operated upon, while sicker, morbidly obese patients were denied surgery due to perceived prohibitive risks. However, we would have expected a higher proportion of severely obese patients to present for emergency surgery, which is contrary to our finding of the lowest incidence of emergency surgery in the severely obese group. It is also possible that severe obesity may be associated with a higher attrition rate, such that the extremely obese patients did not live long enough to present for surgery. This is somewhat likely, given the significantly younger age of the severely obese patients in our study cohort. It is, however, impossible to determine survival effect from a cross‐sectional hospital‐based study design. Clearly, mechanisms used to explain the obesity paradox in minority surgical patients are likely to remain speculative, owing to the interaction of several factors such as concomitant comorbidities, disease progression at the time of presentation, patients' weight history, and regional fat distribution.

The current study confirms the findings of previous investigators26 about the importance of reducing major postoperative complications in surgical patients. While this may seem axiomatic, it deserves reiteration because the risk of postoperative mortality increases considerably in all the BMI categories following 1 or more major postoperative complication. However, it is not clear why obese and morbidly obese patients had a lower incidence of failure to rescue. This may be related to greater physiologic reserve in the obese and morbidly obese group, especially because patients in the higher BMI groups were significantly younger than the normal weight or lean patients. For the same reason, these younger, severely obese patients may have been more aggressively monitored and treated, thereby increasing the likelihood of being rescued following a major complication. It is also possible that the lower proportion of emergency procedures performed in obese and severely obese groups was somewhat protective, especially because emergency surgery was an independent predictor of overall mortality in this cohort of patients. In fact, when we stratified the patients according to urgency of surgery and explored the bivariate relationship between BMI category and mortality (data not shown) among those undergoing urgent surgery, the geometrical distribution of mortality did show a reverse‐J pattern with the highest proportion of cases in the underweight group, declining in the normal BMI and overweight group, and increasing steadily in obese and severely obese group. To this end, caution should be exercised when interpreting the association of BMI group with postoperative mortality for procedures performed as an emergency.

Smoking and antecedent illness are 2 confounding factors commonly criticized in studies attempting to associated BMI with mortality. This is because smokers tend to weigh less and have higher mortality rates than nonsmokers. The present investigation did not find a significant contribution of smoking to mortality when other factors included in a logistic regression model were considered. The current study's findings are consistent with those of previous data in African American patients,27 and contrasts with the excess mortality described in currently smoking Caucasian men and women.28 It is possible that smoking is not an important effect modifier when considering the relevance of BMI to postoperative mortality in minority patients.

Obesity affects all segments of the American population. However, it imposes a larger burden and poses a greater threat to minority populations.1 The increase in overall prevalence of obesity and obesity‐related diseases are especially pronounced in ethnic minorities,2 and the outlook for minorities who develop obesity‐associated diseases such as stroke and chronic renal disease is worse than in their Caucasian counterparts.3

Despite the higher prevalence of obesity in ethnic minorities, the majority of research on the relationship between body mass index (BMI) and mortality has been conducted among Caucasians in the United States. This is due largely to the small number of minority participants in most studies, which makes for low statistical power.4

A curious epidemiologic paradox has been the observation, in many studies, that black adults have lower morbidity and mortality associated with obesity compared to Caucasians.5 In fact, some authors suggest that high BMI among black and other minorities may not be as strong a risk factor for mortality as it is in others.6, 7

Very few studies have specifically examined the contribution of BMI to postoperative outcome in a large cohort of minority patients.8, 9 Similarly, the authors are unaware of any previous studies describing the clinical relevance of being overweight and obese in minority patients undergoing surgery. Therefore, the primary objective of this observational study was to describe the prevalence of overweight and obesity in a large cohort of minority surgical patients, and the impact of BMI class on their postoperative outcome. Our a priori hypothesis was that obese minority surgical patients would have a poorer postoperative outcome and have higher 30‐day all‐cause mortality than normal weight individuals.

METHODS

Study data were derived from the Participant Use Data File of the American College of Surgeons (ACS) National Surgical Quality Improvement Program (NSQIP) for the calendar years 2005 to 2008. This multi‐institutional (186 participating centers) reporting system was designed to provide risk‐adjusted surgical outcome data from throughout the United States.

The methodology for collecting these data, including their accuracy and reproducibility, has been detailed in previous publications.10, 11 It is briefly reviewed here. Dedicated nurse clinical reviewers at each hospital prospectively enrolled patients and collected data in a standardized fashion according to strict ACS‐NSQIP definitions. A systematic sample was obtained by taking the first 40 cases per nurse reviewer on an 8‐day cycle from the operating room log, ensuring that no particular operating room day block time would bias the weighting of cases. Nurse reviewers had completed comprehensive training regarding definitions and data extraction, as well as continuing education and monitoring through the ACS‐NSQIP program. They are assessed for inter‐rater reliability during biennial site visits. Information was obtained from patient medical records, physician office records, and telephone interviews. Patients were followed through their hospital course and after discharge from hospital up to 30 days postoperatively. A high level of accuracy and reproducibility of the data have been previously demonstrated.12

Race was defined as African American, Hispanic, Asian or Pacific Islander, or American Indian/Alaskan native, as identified by the clinical care provider, or within the medical record. Patients were excluded if race was coded as white, or not recorded. We also excluded patients with missing record of height and/or weight. The final study cohort consisted of 119,619 minority patients. We then computed BMI as weight in kilograms divided by the square of the height in meters (BMI = kg/m²). Patients were classified as underweight (BMI 18.5 kg/m²), normal weight (BMI = 18.6‐24.9 kg/m²), overweight (BMI = 25‐29.9 kg/m²), obese (BMI = 30‐39.9 kg/m²), and morbidly obese (BMI 40 kg/m²) in accordance with National Institute of Health (NIH) standards.13

RESULTS

The study cohort included 119,619 surgical patients (Table 1). Their mean (standard deviation [SD]) age was 50.4 (16.9), and the mean (SD) BMI of 30.3 (8.9) kg/m² was in the obese range. The overall prevalence of high BMI (overweight or obese) was 70.8%. A very small proportion, 2.7%, of patients was classified as underweight. Sex‐stratified demographic and behavioral characteristics differed significantly across all the variables in our study cohort. Men were more likely to be overweight, whereas women were more likely to be obese or severely obese. A majority (77.4%) of the patients were non‐elderly adults (<65years) undergoing elective (85.9%) surgical procedures. The minority patients in this study were African American (50%), Hispanic (36%), Asian and Pacific Islander (10%), and American Indian and Alaskan native (4%).

The distribution of baseline preoperative clinical characteristics by BMI class revealed many significant differences (Table 2). Age was significantly different among the BMI classes, with the severely obese group being about 8 years younger than the underweight or normal weight group. Similarly, severely obese patients were more likely to be women, less likely to smoke, more likely to be hypertensive, diabetic, have a history of dyspnea at rest, and more likely to belong to high ASA class. On the other hand, underweight patients were more likely to have disseminated cancer, be current smokers, consume more than 2 alcoholic drinks per day, have active chronic obstructive pulmonary disease (COPD), and have ascites. They were also more likely to be on dialysis and have cardiac disease, as well as a history of stroke. Urgency of surgery also varied significantly across the BMI categories, with the underweight group having the highest incidence of emergency surgery (20.6%) and the severely obese group being the least likely to present for emergency surgery (8.2%).

Perioperative outcomes according to BMI classes documented significant differences (Table 3). Work relative value unit (Work RVU, a measure of surgical complexity), as well as total anesthesia and operation time decreased in a stepwise fashion across the BMI classes to the obese group, followed by increase in these parameters in the severely obese group. Following a decrease to the normal BMI category, there was a positive association between BMI and the incidence of postoperative superficial and deep wound infection, as well as wound disruption.

There was a negative association between BMI class and the likelihood of postoperative sepsis, septic shock, reintubation, delayed ventilator wean, and postoperative pneumonia. Similarly, the proportions of patients who developed postoperative acute renal failure, cardiac arrest, and those who required postoperative blood transfusion or needed reoperation, decreased significantly across the BMI classes, with the highest proportion of cases being in the underweight group and the lowest in the severely obese group. Overall composite morbidity was twice as high in the underweight compared to the severely obese group.

There were 1758 deaths among the study's 119,619 patients, resulting in an overall mortality rate of 1.5%. The overall major complication rate was 13.8%. The distribution of total mortality rate as well as mortality in patients with at least 1 major postoperative complication across BMI classes revealed consistent differences (Figure 1). Over the entire range of BMI classes, there was a progressive, stepwise decrease in the proportion of deaths with increasing BMI. This pattern also occurred among patients who developed at least 1 major postoperative complication, indicating a reduced likelihood of death after a major complication. This is reflective of a reduced likelihood of death after a major complication (failure to rescue) with increasing BMI.

Figure 1

Multivariate logistic regression defined a number of factors associated with 30‐day mortality (Table 4). The Hosmer and Lemeshow goodness‐of‐fit test for this model was not statistically significant (² = 17.8, df = 8, P = 0.23). High ASA physical status was associated with high odds of mortality. Specifically, when controlling for the other covariates in the model, ASA status was associated with a 5‐fold increased relative odds of death (adjusted odds ratio [OR] = 5.30; 95% confidence interval = 4.96‐5.79, P < 0.001). Similarly, occurrence of 1 or more major postoperative complication was associated with 6‐fold increased relative odds of mortality. The paradoxical effect of BMI category observed on univariate analysis was maintained in the multivariate model. Specifically, underweight patients had the highest relative odds of mortality, while severely obese patients had the lowest, compared with patients at a healthy weight (Table 4). Interestingly, smoking had no significant effect on the odds of mortality after controlling for other factors. Similarly, the specific racial group and the timing of the surgical intervention had no significant effect on mortality.

DISCUSSION

In this large, study of minority surgical patients, the impact of BMI on the 30‐day morbidity and mortality was unexpected. The working hypothesis was that overweight and obese patients would have a worse outcome after surgery. However, contrary to this hypothesis, the lowest all‐cause mortality rate was found in the severely obese (BMI 40 kg/m²) group in both men and women. Death rates decreased progressively in a stepwise fashion from the underweight to the severely obese group. Similarly, even in patients who developed at least 1 major postoperative complication, the likelihood of death was still negatively associated with BMI. This negative association of mortality with BMI was observed despite the higher prevalence of chronic diseases, such as hypertension, diabetes, and dyspnea at rest, in the increasing BMI classes.

Controversy remains regarding the association between BMI and mortality, in particular about the shape of the curve for the association between BMI and mortality. Epidemiologic studies have variously described J‐shaped, U‐shaped, monotonic, or linear relationships.16, 17 In the surgical population, a reverse J‐shaped relationship between BMI category and mortality has been described.18, 19 Sometimes this is referred to as obesity paradox or reverse epidemiology: a trend whereby overweight and moderately obese patients have better outcomes and lower risk of death than leaner patients.18 This phenomenon is particularly well described in adult20 and elderly heart failure and hypertensive patients.21 Many of these studies either had very few minority patients,21 or mortality pattern was not analyzed along ethnic lines.

Few studies10 have focused exclusively on minority surgical patients. Some investigators have determined that high BMI in black adults may not be as important a risk factor for mortality6, 7 as in whites. Our data suggest that among minority surgical patients, the relationship appeared to be a downward trend in mortality from low to high BMI, thus revealing the obesity paradox. This pattern was evident even in patients who developed 1 major complication in the postoperative period, suggesting that high BMI also protects against failure to rescueor death after a major complication.

Despite decades of research, the mechanisms underlying the obesity paradox remain speculative.20, 22. Many have posited that adiposity may confer protection against cytokines and various inflammatory mediators in heart failure patients by the production of buffering lipoproteins.23, 24 It is conceivable that similar protection against inflammatory response to surgical tissue trauma is operational in minority patients with high BMI. Another possible reason for the obesity paradox is the clinical presentation and disease progression at the time of surgery. Perhaps, similar to the observation in obese patients with heart failure,25 obese minority patients are symptomatic at an earlier stage of their disease than lean patients, making for earlier diagnosis and treatment. Thus, obesity may simply be a marker of less severe disease at the time of presentation.

Obese patients may also be more aggressively monitored and treated in the perioperative period than lean patients, because of the general perception that they are a high‐risk group.10 This may partly explain the decreased likelihood of failure to rescue with increasing BMI in our patients. Increased vigilance and prompt treatment of complications should reduce the overall morbidity and mortality rate in this group. It is also conceivable that a therapeutic selection bias is operational in the patients we studied. This describes scenarios where relatively healthy obese minority patients were operated upon, while sicker, morbidly obese patients were denied surgery due to perceived prohibitive risks. However, we would have expected a higher proportion of severely obese patients to present for emergency surgery, which is contrary to our finding of the lowest incidence of emergency surgery in the severely obese group. It is also possible that severe obesity may be associated with a higher attrition rate, such that the extremely obese patients did not live long enough to present for surgery. This is somewhat likely, given the significantly younger age of the severely obese patients in our study cohort. It is, however, impossible to determine survival effect from a cross‐sectional hospital‐based study design. Clearly, mechanisms used to explain the obesity paradox in minority surgical patients are likely to remain speculative, owing to the interaction of several factors such as concomitant comorbidities, disease progression at the time of presentation, patients' weight history, and regional fat distribution.

The current study confirms the findings of previous investigators26 about the importance of reducing major postoperative complications in surgical patients. While this may seem axiomatic, it deserves reiteration because the risk of postoperative mortality increases considerably in all the BMI categories following 1 or more major postoperative complication. However, it is not clear why obese and morbidly obese patients had a lower incidence of failure to rescue. This may be related to greater physiologic reserve in the obese and morbidly obese group, especially because patients in the higher BMI groups were significantly younger than the normal weight or lean patients. For the same reason, these younger, severely obese patients may have been more aggressively monitored and treated, thereby increasing the likelihood of being rescued following a major complication. It is also possible that the lower proportion of emergency procedures performed in obese and severely obese groups was somewhat protective, especially because emergency surgery was an independent predictor of overall mortality in this cohort of patients. In fact, when we stratified the patients according to urgency of surgery and explored the bivariate relationship between BMI category and mortality (data not shown) among those undergoing urgent surgery, the geometrical distribution of mortality did show a reverse‐J pattern with the highest proportion of cases in the underweight group, declining in the normal BMI and overweight group, and increasing steadily in obese and severely obese group. To this end, caution should be exercised when interpreting the association of BMI group with postoperative mortality for procedures performed as an emergency.

Smoking and antecedent illness are 2 confounding factors commonly criticized in studies attempting to associated BMI with mortality. This is because smokers tend to weigh less and have higher mortality rates than nonsmokers. The present investigation did not find a significant contribution of smoking to mortality when other factors included in a logistic regression model were considered. The current study's findings are consistent with those of previous data in African American patients,27 and contrasts with the excess mortality described in currently smoking Caucasian men and women.28 It is possible that smoking is not an important effect modifier when considering the relevance of BMI to postoperative mortality in minority patients.

Recognition that healthcare carries considerable risks of patient injury has focused efforts on identifying problems before they occur, and understanding the root causes of those problems that do occur to prevent them from happening again.1 To further these efforts, a Joint Commission (JC) standard requires hospitals to review sentinel events (SE).2 Reviews must develop a timely, thorough, and credible root cause analysis (RCA), implement action plans to reduce risk, and monitor the effectiveness of implemented improvements.3

Ideally, hospitals would summarize their experiences with SE reviews, identify high‐risk activities and patients, institute system changes to prevent SE recurrences, and share their findings with other healthcare organizations to help them avoid similar patient injuries.1 In support of this last goal, the JC maintains a voluntary database system that allows hospitals to report their SE analyses for other facilities to review and institute preventative actions.

Unfortunately, the reality of SE reviews does not match their ideals for improving patient safety.4 Healthcare organizations often describe their review process as less than credible and note a need for ongoing oversight to maintain the reviews' effectiveness.5 The JC voluntary reporting system captures less than 1% of the SEs that occur nationally,2 because hospitals perceive barriers to external reporting.1 If healthcare organizations decide against reporting externally, they can create their own internal systems to aggregate and summarize SEs, but few such systems exist. A major impediment to designing internal systems is the absence of universally endorsed nomenclature for safety‐related events.6, 7 Poorly aligned terminology and subjective conceptualizations for safety incidents impede the aggregation of SEs, comparisons between facilities, and trend analyses for tracking SE patterns.

In 2005, the World Health Organization (WHO) World Alliance for Patient Safety, in collaboration with the JC, began developing an International Classification for Patient Safety (ICPS) to provide healthcare organizations a consistent conceptual model for safety incidents and promote their classification by a standardized taxonomy.810 Although this system has promise for allowing standardization, data aggregation, analysis, and learning between institutions,11 integration of the ICPS conceptual model into an SE decision support tool with summarizing and reporting features has not been reported.

This report describes our development of an intranet‐based SE reporting system, called Incident Tracker (I‐Tracker), based on the ICPS model. For our SE review groups from the 4 Providence Health Systems (PHS) Portland Service Area (PSA) hospitals, the I‐Tracker system offers a tool to guide efforts in developing RCAs and action plans in alignment with the ICPS framework. The system includes scripts that automatically generate and distribute standardized reports of individual and aggregated SEs. The objectives of this project were to report our experience with developing a flexible and accessible intranet‐based system that assists RCA participants in conforming to the ICPS framework and oversight safety staff in summarizing and reporting root cause analyses.

METHODS

The 4 PSA hospitals have 1083 licensed beds and perform SE reviews with a centralized process that reports results to a Community Governing Board. An ad hoc team for each SE performs the RCAs. The SE groups report RCAs and action plans in an unstructured format that varies for each event. A paper file is maintained for each SE report, but a system for aggregating reports to track trends, disseminating SE trends, or monitoring the completion or effectiveness of action plans is not available.

We designed a system to achieve the following objectives:

• 1

  Apply the ICPS framework (Figure 1) and taxonomy of terms to SE analyses;

• 2

  Provide a computer‐based tool to assist review groups and quality staff to perform their SE reviews and data collection in alignment with the ICPS framework;

• 3

  Create an intranet‐based database that captures elements of the reviews, RCAs, and action plans with the use of drop‐down lists, help windows, windows with live access to Internet educational resources and tools, decision support tools, default entries, and audio prompts to streamline data entry;

• 4

  Generate a suite of standardized reports customized for different audiences that can be accessed online and printed from the database with automated scripts;

• 5

  Produce intranet‐based summaries of aggregated events to identify common causes and disseminate observed patterns and action plans to other PSA hospitals.

Figure 1

We selected FileMaker Pro 11 Advanced (FMP11) for authoring and maintaining the decision support tool and database, and FileMaker Pro Server 11 Advanced (FMPS11) (Filemaker, Inc, Santa Clara, CA) for hosting the system, because it provides intranet access and tools for updating the system by personnel with minimal programming experience. End users can view and enter data through layouts that display only the information allowed by the user's login password and access privileges, with external authentication by Active Directory and Open Directory technology. Staff who author and manage the database do so through client FMP11 software loaded on a computer that provides remote server access.

The I‐Tracker system was authored using the ICPS definitions for the 48 preferred terms for safety incidents and the ICPS conceptual framework.8 The conceptual framework consists of 10 major incident domains, that include incident type, patient outcomes, patient characteristics, incident characteristics, contributing factors and hazards, organizational outcomes, detection, mitigating factors, ameliorating actions, and actions taken to reduce risk (Figure 1).11 The framework is applicable to all hospital safety incidents, but we limited I‐Tracker to SEs because our hospitals had completed comprehensive reviews and action plans only for these more serious events. The literature on the ICPS framework812 was carefully reviewed to identify the specific data fields that were recommended by ICPS developers to be included under each of the 10 major classification domains. In most instances, data fields existed only in the body of these reports. Article texts, however, provided sufficient descriptions of these data fields to allow their translation into data entry fields in I‐Tracker with accompanying help windows and explanations to guide I‐Tracker users. Sixty ICPS data fields were programmed into I‐Tracker, with another 120 fields that allowed entry of descriptions and explanations of the ICPS data field entries. For instance, an entry of Yes into an ICPS data field that queried Was there a systems problem of an external organization other than a Providence entity opens a Describe field that allows a brief description of the problem, and an additional Details field that allows a longer explanation of the problem if necessary. The brief Describe field contents populate an automatically generated fishbone diagram.

The authors and quality staff translated the most recent 15 SE reviews into ICPS terms and classifications, and entered the results into I‐Tracker as it was being developed, to assist system design and programming of the system. The authors noted during data entry which of the 10 ICPS major domains had not been analyzed by the previous 15 reviews. Because existing reports were unstructured with considerable variation in style and usage of terms, the authors and quality staff made group decisions regarding how to cross‐walk existing information into the standardized ICPS data fields.

RESULTS

In developing I‐Tracker, the authors and quality staff observed that the ICPS framework and recommended data fields were logical and straightforward to learn. Although it was difficult to find the definitions of specific ICPS data fields within the 10 major domains in the text of retrieved articles, these fields could be readily cross‐walked into I‐Tracker data entry fields. Translating existing SE reports into I‐Tracker classifications, however, presented considerable challenges because of the unstructured, discursive, and variable nature of our SE review and reports. The authors and staff spent 1 to 2 hours conferring over each report to make judgments as to which elements of the review would be entered into which I‐Tracker data fields. Once the authors and staff translated existing reports into ICPS terms, actual data entry into I‐Tracker took typically less than 30 minutes for each review. We found that none of our 15 SE reviews included information on the following ICPS major domains: detection, mitigating factors, and ameliorating actions. We also observed that many ICPS data fields were not assessed, such as patient contributions to errors and external organization's contributions to a safety incident.

The latest version of I‐Tracker receives and displays information at the individual patient level. Records are shown onscreen with different screen layouts depending on the viewer's login security clearance. Hospital safety staff have full access to view and enter data on the initial layout, which displays patient demographic information and folder tabs that navigate when clicked to other database fields (Figure 2). Viewers with lower security clearance either view the same opening screen, but have limited access to other screens, or view a different opening screen designed to meet their specific needs. All screens provide definitions of terms and information to assist data entry, buttons that navigate to help pages, pop‐up windows that provide tips, and buttons that trigger brief audio explanations. Most fields use drop‐down lists to standardize data entry around the ICPS definitions, with default values entered into many fields to streamline data entry. A list view allows review of all patients and quick access to an individual patient's record. All fields and combinations of fields with Boolean rules are searchable within the database.

Figure 2

I‐Tracker has features that support SE review groups in beginning an SE review by providing them a paper form or electronic interface by way of a portable computer or tablet device, that guides their discussions and analyses toward providing conclusions that can be entered into the database fields, thereby aligning their deliberations with the ICPS conceptual framework. The same resource is available within the database online for those groups who would prefer to use computer prompts and enter data directly into the database as they proceed through their analyses. Some layouts contain windows that port live views from external Web sites (eg, JC RCA resources) that provide participants of RCA groups with tools to assist their work. FMP11 allows users to access the database by portable computers or handheld tablet devices using the hospitals' WiFi network.

A report screen allows automatic generation of different printouts of individual or aggregate summary reports. A Comprehensive Report includes all of the data fields included in the ICPS conceptual framework. Other reports present subsets of data depending on the user's needs and access privileges. The FMPS11 database allows printing the reports to paper or Portable Document Format (pdf), exporting data into an Excel spreadsheet, or e‐mailing reports to recipients from within I‐Tracker.

Additionally, I‐Tracker functionality facilitates follow‐up and monitoring of action items developed during the RCA process in a manner that conforms to the ICPS framework. We are now developing educational resources for RCA team members to investigate the implementation of I‐Tracker into future RCAs.

DISCUSSION

I‐Tracker provides an intranet‐based tool that met the objectives of the present project. The process of entering 15 existing SE reviews and action plans from our healthcare system into I‐Tracker allowed an incremental development of the database and identified gaps in our existing RCA process. For instance, none of the previous RCAs critically appraised detection, mitigating factors, or ameliorating actions; defined the specific nature or quantified severity of patient injuries using standardized terms; distinguished between human errors and negligence; or comprehensively reported the full spectrum of underlying causes of Tracker's use of standardized terms based on the ICPS conceptual framework provided a potential resource for focusing SE reviews and producing more comprehensive RCAs and action plans in the future. I‐Tracker has additional potential to facilitate dissemination of RCAs to other facilities, both as individual incident reports and aggregated summaries as recommended by experts in patient safety.13

The deficiencies in our existing RCA analyses, identified during data entry into I‐Tracker, represent common shortcomings experienced by other healthcare organizations and summarized in a report by the Agency for Healthcare Research and Quality.4 Considerable hindsight bias and prevailing concerns of the day taint the RCA process, which is time‐consuming and labor intensive, and thereby hinders comprehensive reviews. Also, our SE reviews, like others reviewed in the literature,14 focused on biologic injury to patients and omitted assessment of psychologic, organizational, social, and economic injury domains. Although SE review teams benefit from involvement of quality improvement staff who are trained in techniques and goals of RCA,15 many hospitals like ours have limited resources for fully staffing all SE reviews with trained facilitators. These SE reviews generate both quantitative and qualitative data, the latter of which hinders standardized data entry in the absence of a conceptual framework. A structured database with formative tools to guide RCAs in conformance with the ICPS framework in organizations without sufficient numbers of trained facilitators offers opportunities to produce more comprehensive, standardized, and actionable reports. To date, our quality staff and leadership have responded positively to presentations of the functional features of I‐Tracker (Table 1).

Limitations of our report include its focus solely on the development and programming phase of I‐Tracker and the absence of information on its actual implementation. We believe, however, the development phase is important to report because it demonstrates that the ICPS framework and specific ICPS data fields are amenable to incorporation into a decision support and reporting tool, which to our knowledge has not been previously reported. We begin implementation of I‐Tracker within our organization this year and will have observations on its feasibility, acceptability, and staff training needs. As an additional limitation, we emphasize that we do not propose I‐Tracker as a solution for other organizations, because we have no plans for its commercial or public domain development. This report is intended to demonstrate, however, that commercially available software, such as FileMaker, can readily support the ICPS Framework and thereby has potential to assist RCAs and SE reporting. Other organizations may develop similar systems on other database platforms that incorporate the ICPS system into their reviews.

To implement I‐Tracker, we are now working with nursing and pharmacy leadership focus groups to develop formative tools, data collection forms, and other resources to assist their RCA efforts and data entry into the database. We also plan to apply the database tool to our residency training program to promote resident involvement in SE reviews by providing standardized, reproducible, and structured processes.16 Our 5‐state healthcare system has funded an evaluation of the implementation phase of I‐Tracker to other Providence facilities. Because the ICPS framework applies to all safety incidents beyond SEs (Table 2), a successful implementation of I‐Tracker for SEs will allow its eventual application to other types of critical incidents.

The strength of this project derives from its innovative development of an intranet‐based tool that allows groups to conform their RCAs to the ICPS framework. Because the absence of a standardized classification for patient safety concepts has hindered advances in patient safety,11 we believe I‐Tracker, or decision support tools like it that use the ICPS framework, can standardize RCAs and promote dissemination and adoption of action plans.

Recognition that healthcare carries considerable risks of patient injury has focused efforts on identifying problems before they occur, and understanding the root causes of those problems that do occur to prevent them from happening again.1 To further these efforts, a Joint Commission (JC) standard requires hospitals to review sentinel events (SE).2 Reviews must develop a timely, thorough, and credible root cause analysis (RCA), implement action plans to reduce risk, and monitor the effectiveness of implemented improvements.3

Ideally, hospitals would summarize their experiences with SE reviews, identify high‐risk activities and patients, institute system changes to prevent SE recurrences, and share their findings with other healthcare organizations to help them avoid similar patient injuries.1 In support of this last goal, the JC maintains a voluntary database system that allows hospitals to report their SE analyses for other facilities to review and institute preventative actions.

Unfortunately, the reality of SE reviews does not match their ideals for improving patient safety.4 Healthcare organizations often describe their review process as less than credible and note a need for ongoing oversight to maintain the reviews' effectiveness.5 The JC voluntary reporting system captures less than 1% of the SEs that occur nationally,2 because hospitals perceive barriers to external reporting.1 If healthcare organizations decide against reporting externally, they can create their own internal systems to aggregate and summarize SEs, but few such systems exist. A major impediment to designing internal systems is the absence of universally endorsed nomenclature for safety‐related events.6, 7 Poorly aligned terminology and subjective conceptualizations for safety incidents impede the aggregation of SEs, comparisons between facilities, and trend analyses for tracking SE patterns.

In 2005, the World Health Organization (WHO) World Alliance for Patient Safety, in collaboration with the JC, began developing an International Classification for Patient Safety (ICPS) to provide healthcare organizations a consistent conceptual model for safety incidents and promote their classification by a standardized taxonomy.810 Although this system has promise for allowing standardization, data aggregation, analysis, and learning between institutions,11 integration of the ICPS conceptual model into an SE decision support tool with summarizing and reporting features has not been reported.

This report describes our development of an intranet‐based SE reporting system, called Incident Tracker (I‐Tracker), based on the ICPS model. For our SE review groups from the 4 Providence Health Systems (PHS) Portland Service Area (PSA) hospitals, the I‐Tracker system offers a tool to guide efforts in developing RCAs and action plans in alignment with the ICPS framework. The system includes scripts that automatically generate and distribute standardized reports of individual and aggregated SEs. The objectives of this project were to report our experience with developing a flexible and accessible intranet‐based system that assists RCA participants in conforming to the ICPS framework and oversight safety staff in summarizing and reporting root cause analyses.

METHODS

The 4 PSA hospitals have 1083 licensed beds and perform SE reviews with a centralized process that reports results to a Community Governing Board. An ad hoc team for each SE performs the RCAs. The SE groups report RCAs and action plans in an unstructured format that varies for each event. A paper file is maintained for each SE report, but a system for aggregating reports to track trends, disseminating SE trends, or monitoring the completion or effectiveness of action plans is not available.

We designed a system to achieve the following objectives:

• 1

  Apply the ICPS framework (Figure 1) and taxonomy of terms to SE analyses;

• 2

  Provide a computer‐based tool to assist review groups and quality staff to perform their SE reviews and data collection in alignment with the ICPS framework;

• 3

  Create an intranet‐based database that captures elements of the reviews, RCAs, and action plans with the use of drop‐down lists, help windows, windows with live access to Internet educational resources and tools, decision support tools, default entries, and audio prompts to streamline data entry;

• 4

  Generate a suite of standardized reports customized for different audiences that can be accessed online and printed from the database with automated scripts;

• 5

  Produce intranet‐based summaries of aggregated events to identify common causes and disseminate observed patterns and action plans to other PSA hospitals.

Figure 1

We selected FileMaker Pro 11 Advanced (FMP11) for authoring and maintaining the decision support tool and database, and FileMaker Pro Server 11 Advanced (FMPS11) (Filemaker, Inc, Santa Clara, CA) for hosting the system, because it provides intranet access and tools for updating the system by personnel with minimal programming experience. End users can view and enter data through layouts that display only the information allowed by the user's login password and access privileges, with external authentication by Active Directory and Open Directory technology. Staff who author and manage the database do so through client FMP11 software loaded on a computer that provides remote server access.

The I‐Tracker system was authored using the ICPS definitions for the 48 preferred terms for safety incidents and the ICPS conceptual framework.8 The conceptual framework consists of 10 major incident domains, that include incident type, patient outcomes, patient characteristics, incident characteristics, contributing factors and hazards, organizational outcomes, detection, mitigating factors, ameliorating actions, and actions taken to reduce risk (Figure 1).11 The framework is applicable to all hospital safety incidents, but we limited I‐Tracker to SEs because our hospitals had completed comprehensive reviews and action plans only for these more serious events. The literature on the ICPS framework812 was carefully reviewed to identify the specific data fields that were recommended by ICPS developers to be included under each of the 10 major classification domains. In most instances, data fields existed only in the body of these reports. Article texts, however, provided sufficient descriptions of these data fields to allow their translation into data entry fields in I‐Tracker with accompanying help windows and explanations to guide I‐Tracker users. Sixty ICPS data fields were programmed into I‐Tracker, with another 120 fields that allowed entry of descriptions and explanations of the ICPS data field entries. For instance, an entry of Yes into an ICPS data field that queried Was there a systems problem of an external organization other than a Providence entity opens a Describe field that allows a brief description of the problem, and an additional Details field that allows a longer explanation of the problem if necessary. The brief Describe field contents populate an automatically generated fishbone diagram.

The authors and quality staff translated the most recent 15 SE reviews into ICPS terms and classifications, and entered the results into I‐Tracker as it was being developed, to assist system design and programming of the system. The authors noted during data entry which of the 10 ICPS major domains had not been analyzed by the previous 15 reviews. Because existing reports were unstructured with considerable variation in style and usage of terms, the authors and quality staff made group decisions regarding how to cross‐walk existing information into the standardized ICPS data fields.

RESULTS

In developing I‐Tracker, the authors and quality staff observed that the ICPS framework and recommended data fields were logical and straightforward to learn. Although it was difficult to find the definitions of specific ICPS data fields within the 10 major domains in the text of retrieved articles, these fields could be readily cross‐walked into I‐Tracker data entry fields. Translating existing SE reports into I‐Tracker classifications, however, presented considerable challenges because of the unstructured, discursive, and variable nature of our SE review and reports. The authors and staff spent 1 to 2 hours conferring over each report to make judgments as to which elements of the review would be entered into which I‐Tracker data fields. Once the authors and staff translated existing reports into ICPS terms, actual data entry into I‐Tracker took typically less than 30 minutes for each review. We found that none of our 15 SE reviews included information on the following ICPS major domains: detection, mitigating factors, and ameliorating actions. We also observed that many ICPS data fields were not assessed, such as patient contributions to errors and external organization's contributions to a safety incident.

The latest version of I‐Tracker receives and displays information at the individual patient level. Records are shown onscreen with different screen layouts depending on the viewer's login security clearance. Hospital safety staff have full access to view and enter data on the initial layout, which displays patient demographic information and folder tabs that navigate when clicked to other database fields (Figure 2). Viewers with lower security clearance either view the same opening screen, but have limited access to other screens, or view a different opening screen designed to meet their specific needs. All screens provide definitions of terms and information to assist data entry, buttons that navigate to help pages, pop‐up windows that provide tips, and buttons that trigger brief audio explanations. Most fields use drop‐down lists to standardize data entry around the ICPS definitions, with default values entered into many fields to streamline data entry. A list view allows review of all patients and quick access to an individual patient's record. All fields and combinations of fields with Boolean rules are searchable within the database.

Figure 2

I‐Tracker has features that support SE review groups in beginning an SE review by providing them a paper form or electronic interface by way of a portable computer or tablet device, that guides their discussions and analyses toward providing conclusions that can be entered into the database fields, thereby aligning their deliberations with the ICPS conceptual framework. The same resource is available within the database online for those groups who would prefer to use computer prompts and enter data directly into the database as they proceed through their analyses. Some layouts contain windows that port live views from external Web sites (eg, JC RCA resources) that provide participants of RCA groups with tools to assist their work. FMP11 allows users to access the database by portable computers or handheld tablet devices using the hospitals' WiFi network.

A report screen allows automatic generation of different printouts of individual or aggregate summary reports. A Comprehensive Report includes all of the data fields included in the ICPS conceptual framework. Other reports present subsets of data depending on the user's needs and access privileges. The FMPS11 database allows printing the reports to paper or Portable Document Format (pdf), exporting data into an Excel spreadsheet, or e‐mailing reports to recipients from within I‐Tracker.

Additionally, I‐Tracker functionality facilitates follow‐up and monitoring of action items developed during the RCA process in a manner that conforms to the ICPS framework. We are now developing educational resources for RCA team members to investigate the implementation of I‐Tracker into future RCAs.

DISCUSSION

I‐Tracker provides an intranet‐based tool that met the objectives of the present project. The process of entering 15 existing SE reviews and action plans from our healthcare system into I‐Tracker allowed an incremental development of the database and identified gaps in our existing RCA process. For instance, none of the previous RCAs critically appraised detection, mitigating factors, or ameliorating actions; defined the specific nature or quantified severity of patient injuries using standardized terms; distinguished between human errors and negligence; or comprehensively reported the full spectrum of underlying causes of Tracker's use of standardized terms based on the ICPS conceptual framework provided a potential resource for focusing SE reviews and producing more comprehensive RCAs and action plans in the future. I‐Tracker has additional potential to facilitate dissemination of RCAs to other facilities, both as individual incident reports and aggregated summaries as recommended by experts in patient safety.13

The deficiencies in our existing RCA analyses, identified during data entry into I‐Tracker, represent common shortcomings experienced by other healthcare organizations and summarized in a report by the Agency for Healthcare Research and Quality.4 Considerable hindsight bias and prevailing concerns of the day taint the RCA process, which is time‐consuming and labor intensive, and thereby hinders comprehensive reviews. Also, our SE reviews, like others reviewed in the literature,14 focused on biologic injury to patients and omitted assessment of psychologic, organizational, social, and economic injury domains. Although SE review teams benefit from involvement of quality improvement staff who are trained in techniques and goals of RCA,15 many hospitals like ours have limited resources for fully staffing all SE reviews with trained facilitators. These SE reviews generate both quantitative and qualitative data, the latter of which hinders standardized data entry in the absence of a conceptual framework. A structured database with formative tools to guide RCAs in conformance with the ICPS framework in organizations without sufficient numbers of trained facilitators offers opportunities to produce more comprehensive, standardized, and actionable reports. To date, our quality staff and leadership have responded positively to presentations of the functional features of I‐Tracker (Table 1).

Limitations of our report include its focus solely on the development and programming phase of I‐Tracker and the absence of information on its actual implementation. We believe, however, the development phase is important to report because it demonstrates that the ICPS framework and specific ICPS data fields are amenable to incorporation into a decision support and reporting tool, which to our knowledge has not been previously reported. We begin implementation of I‐Tracker within our organization this year and will have observations on its feasibility, acceptability, and staff training needs. As an additional limitation, we emphasize that we do not propose I‐Tracker as a solution for other organizations, because we have no plans for its commercial or public domain development. This report is intended to demonstrate, however, that commercially available software, such as FileMaker, can readily support the ICPS Framework and thereby has potential to assist RCAs and SE reporting. Other organizations may develop similar systems on other database platforms that incorporate the ICPS system into their reviews.

To implement I‐Tracker, we are now working with nursing and pharmacy leadership focus groups to develop formative tools, data collection forms, and other resources to assist their RCA efforts and data entry into the database. We also plan to apply the database tool to our residency training program to promote resident involvement in SE reviews by providing standardized, reproducible, and structured processes.16 Our 5‐state healthcare system has funded an evaluation of the implementation phase of I‐Tracker to other Providence facilities. Because the ICPS framework applies to all safety incidents beyond SEs (Table 2), a successful implementation of I‐Tracker for SEs will allow its eventual application to other types of critical incidents.

The strength of this project derives from its innovative development of an intranet‐based tool that allows groups to conform their RCAs to the ICPS framework. Because the absence of a standardized classification for patient safety concepts has hindered advances in patient safety,11 we believe I‐Tracker, or decision support tools like it that use the ICPS framework, can standardize RCAs and promote dissemination and adoption of action plans.

Recognition that healthcare carries considerable risks of patient injury has focused efforts on identifying problems before they occur, and understanding the root causes of those problems that do occur to prevent them from happening again.1 To further these efforts, a Joint Commission (JC) standard requires hospitals to review sentinel events (SE).2 Reviews must develop a timely, thorough, and credible root cause analysis (RCA), implement action plans to reduce risk, and monitor the effectiveness of implemented improvements.3

Ideally, hospitals would summarize their experiences with SE reviews, identify high‐risk activities and patients, institute system changes to prevent SE recurrences, and share their findings with other healthcare organizations to help them avoid similar patient injuries.1 In support of this last goal, the JC maintains a voluntary database system that allows hospitals to report their SE analyses for other facilities to review and institute preventative actions.

Unfortunately, the reality of SE reviews does not match their ideals for improving patient safety.4 Healthcare organizations often describe their review process as less than credible and note a need for ongoing oversight to maintain the reviews' effectiveness.5 The JC voluntary reporting system captures less than 1% of the SEs that occur nationally,2 because hospitals perceive barriers to external reporting.1 If healthcare organizations decide against reporting externally, they can create their own internal systems to aggregate and summarize SEs, but few such systems exist. A major impediment to designing internal systems is the absence of universally endorsed nomenclature for safety‐related events.6, 7 Poorly aligned terminology and subjective conceptualizations for safety incidents impede the aggregation of SEs, comparisons between facilities, and trend analyses for tracking SE patterns.

In 2005, the World Health Organization (WHO) World Alliance for Patient Safety, in collaboration with the JC, began developing an International Classification for Patient Safety (ICPS) to provide healthcare organizations a consistent conceptual model for safety incidents and promote their classification by a standardized taxonomy.810 Although this system has promise for allowing standardization, data aggregation, analysis, and learning between institutions,11 integration of the ICPS conceptual model into an SE decision support tool with summarizing and reporting features has not been reported.

This report describes our development of an intranet‐based SE reporting system, called Incident Tracker (I‐Tracker), based on the ICPS model. For our SE review groups from the 4 Providence Health Systems (PHS) Portland Service Area (PSA) hospitals, the I‐Tracker system offers a tool to guide efforts in developing RCAs and action plans in alignment with the ICPS framework. The system includes scripts that automatically generate and distribute standardized reports of individual and aggregated SEs. The objectives of this project were to report our experience with developing a flexible and accessible intranet‐based system that assists RCA participants in conforming to the ICPS framework and oversight safety staff in summarizing and reporting root cause analyses.

METHODS

The 4 PSA hospitals have 1083 licensed beds and perform SE reviews with a centralized process that reports results to a Community Governing Board. An ad hoc team for each SE performs the RCAs. The SE groups report RCAs and action plans in an unstructured format that varies for each event. A paper file is maintained for each SE report, but a system for aggregating reports to track trends, disseminating SE trends, or monitoring the completion or effectiveness of action plans is not available.

We designed a system to achieve the following objectives:

• 1

  Apply the ICPS framework (Figure 1) and taxonomy of terms to SE analyses;

• 2

  Provide a computer‐based tool to assist review groups and quality staff to perform their SE reviews and data collection in alignment with the ICPS framework;

• 3

  Create an intranet‐based database that captures elements of the reviews, RCAs, and action plans with the use of drop‐down lists, help windows, windows with live access to Internet educational resources and tools, decision support tools, default entries, and audio prompts to streamline data entry;

• 4

  Generate a suite of standardized reports customized for different audiences that can be accessed online and printed from the database with automated scripts;

• 5

  Produce intranet‐based summaries of aggregated events to identify common causes and disseminate observed patterns and action plans to other PSA hospitals.

Figure 1

We selected FileMaker Pro 11 Advanced (FMP11) for authoring and maintaining the decision support tool and database, and FileMaker Pro Server 11 Advanced (FMPS11) (Filemaker, Inc, Santa Clara, CA) for hosting the system, because it provides intranet access and tools for updating the system by personnel with minimal programming experience. End users can view and enter data through layouts that display only the information allowed by the user's login password and access privileges, with external authentication by Active Directory and Open Directory technology. Staff who author and manage the database do so through client FMP11 software loaded on a computer that provides remote server access.

The I‐Tracker system was authored using the ICPS definitions for the 48 preferred terms for safety incidents and the ICPS conceptual framework.8 The conceptual framework consists of 10 major incident domains, that include incident type, patient outcomes, patient characteristics, incident characteristics, contributing factors and hazards, organizational outcomes, detection, mitigating factors, ameliorating actions, and actions taken to reduce risk (Figure 1).11 The framework is applicable to all hospital safety incidents, but we limited I‐Tracker to SEs because our hospitals had completed comprehensive reviews and action plans only for these more serious events. The literature on the ICPS framework812 was carefully reviewed to identify the specific data fields that were recommended by ICPS developers to be included under each of the 10 major classification domains. In most instances, data fields existed only in the body of these reports. Article texts, however, provided sufficient descriptions of these data fields to allow their translation into data entry fields in I‐Tracker with accompanying help windows and explanations to guide I‐Tracker users. Sixty ICPS data fields were programmed into I‐Tracker, with another 120 fields that allowed entry of descriptions and explanations of the ICPS data field entries. For instance, an entry of Yes into an ICPS data field that queried Was there a systems problem of an external organization other than a Providence entity opens a Describe field that allows a brief description of the problem, and an additional Details field that allows a longer explanation of the problem if necessary. The brief Describe field contents populate an automatically generated fishbone diagram.

The authors and quality staff translated the most recent 15 SE reviews into ICPS terms and classifications, and entered the results into I‐Tracker as it was being developed, to assist system design and programming of the system. The authors noted during data entry which of the 10 ICPS major domains had not been analyzed by the previous 15 reviews. Because existing reports were unstructured with considerable variation in style and usage of terms, the authors and quality staff made group decisions regarding how to cross‐walk existing information into the standardized ICPS data fields.

RESULTS

In developing I‐Tracker, the authors and quality staff observed that the ICPS framework and recommended data fields were logical and straightforward to learn. Although it was difficult to find the definitions of specific ICPS data fields within the 10 major domains in the text of retrieved articles, these fields could be readily cross‐walked into I‐Tracker data entry fields. Translating existing SE reports into I‐Tracker classifications, however, presented considerable challenges because of the unstructured, discursive, and variable nature of our SE review and reports. The authors and staff spent 1 to 2 hours conferring over each report to make judgments as to which elements of the review would be entered into which I‐Tracker data fields. Once the authors and staff translated existing reports into ICPS terms, actual data entry into I‐Tracker took typically less than 30 minutes for each review. We found that none of our 15 SE reviews included information on the following ICPS major domains: detection, mitigating factors, and ameliorating actions. We also observed that many ICPS data fields were not assessed, such as patient contributions to errors and external organization's contributions to a safety incident.

The latest version of I‐Tracker receives and displays information at the individual patient level. Records are shown onscreen with different screen layouts depending on the viewer's login security clearance. Hospital safety staff have full access to view and enter data on the initial layout, which displays patient demographic information and folder tabs that navigate when clicked to other database fields (Figure 2). Viewers with lower security clearance either view the same opening screen, but have limited access to other screens, or view a different opening screen designed to meet their specific needs. All screens provide definitions of terms and information to assist data entry, buttons that navigate to help pages, pop‐up windows that provide tips, and buttons that trigger brief audio explanations. Most fields use drop‐down lists to standardize data entry around the ICPS definitions, with default values entered into many fields to streamline data entry. A list view allows review of all patients and quick access to an individual patient's record. All fields and combinations of fields with Boolean rules are searchable within the database.

Figure 2

I‐Tracker has features that support SE review groups in beginning an SE review by providing them a paper form or electronic interface by way of a portable computer or tablet device, that guides their discussions and analyses toward providing conclusions that can be entered into the database fields, thereby aligning their deliberations with the ICPS conceptual framework. The same resource is available within the database online for those groups who would prefer to use computer prompts and enter data directly into the database as they proceed through their analyses. Some layouts contain windows that port live views from external Web sites (eg, JC RCA resources) that provide participants of RCA groups with tools to assist their work. FMP11 allows users to access the database by portable computers or handheld tablet devices using the hospitals' WiFi network.

A report screen allows automatic generation of different printouts of individual or aggregate summary reports. A Comprehensive Report includes all of the data fields included in the ICPS conceptual framework. Other reports present subsets of data depending on the user's needs and access privileges. The FMPS11 database allows printing the reports to paper or Portable Document Format (pdf), exporting data into an Excel spreadsheet, or e‐mailing reports to recipients from within I‐Tracker.

Additionally, I‐Tracker functionality facilitates follow‐up and monitoring of action items developed during the RCA process in a manner that conforms to the ICPS framework. We are now developing educational resources for RCA team members to investigate the implementation of I‐Tracker into future RCAs.

DISCUSSION

I‐Tracker provides an intranet‐based tool that met the objectives of the present project. The process of entering 15 existing SE reviews and action plans from our healthcare system into I‐Tracker allowed an incremental development of the database and identified gaps in our existing RCA process. For instance, none of the previous RCAs critically appraised detection, mitigating factors, or ameliorating actions; defined the specific nature or quantified severity of patient injuries using standardized terms; distinguished between human errors and negligence; or comprehensively reported the full spectrum of underlying causes of Tracker's use of standardized terms based on the ICPS conceptual framework provided a potential resource for focusing SE reviews and producing more comprehensive RCAs and action plans in the future. I‐Tracker has additional potential to facilitate dissemination of RCAs to other facilities, both as individual incident reports and aggregated summaries as recommended by experts in patient safety.13

The deficiencies in our existing RCA analyses, identified during data entry into I‐Tracker, represent common shortcomings experienced by other healthcare organizations and summarized in a report by the Agency for Healthcare Research and Quality.4 Considerable hindsight bias and prevailing concerns of the day taint the RCA process, which is time‐consuming and labor intensive, and thereby hinders comprehensive reviews. Also, our SE reviews, like others reviewed in the literature,14 focused on biologic injury to patients and omitted assessment of psychologic, organizational, social, and economic injury domains. Although SE review teams benefit from involvement of quality improvement staff who are trained in techniques and goals of RCA,15 many hospitals like ours have limited resources for fully staffing all SE reviews with trained facilitators. These SE reviews generate both quantitative and qualitative data, the latter of which hinders standardized data entry in the absence of a conceptual framework. A structured database with formative tools to guide RCAs in conformance with the ICPS framework in organizations without sufficient numbers of trained facilitators offers opportunities to produce more comprehensive, standardized, and actionable reports. To date, our quality staff and leadership have responded positively to presentations of the functional features of I‐Tracker (Table 1).

Limitations of our report include its focus solely on the development and programming phase of I‐Tracker and the absence of information on its actual implementation. We believe, however, the development phase is important to report because it demonstrates that the ICPS framework and specific ICPS data fields are amenable to incorporation into a decision support and reporting tool, which to our knowledge has not been previously reported. We begin implementation of I‐Tracker within our organization this year and will have observations on its feasibility, acceptability, and staff training needs. As an additional limitation, we emphasize that we do not propose I‐Tracker as a solution for other organizations, because we have no plans for its commercial or public domain development. This report is intended to demonstrate, however, that commercially available software, such as FileMaker, can readily support the ICPS Framework and thereby has potential to assist RCAs and SE reporting. Other organizations may develop similar systems on other database platforms that incorporate the ICPS system into their reviews.

To implement I‐Tracker, we are now working with nursing and pharmacy leadership focus groups to develop formative tools, data collection forms, and other resources to assist their RCA efforts and data entry into the database. We also plan to apply the database tool to our residency training program to promote resident involvement in SE reviews by providing standardized, reproducible, and structured processes.16 Our 5‐state healthcare system has funded an evaluation of the implementation phase of I‐Tracker to other Providence facilities. Because the ICPS framework applies to all safety incidents beyond SEs (Table 2), a successful implementation of I‐Tracker for SEs will allow its eventual application to other types of critical incidents.

The strength of this project derives from its innovative development of an intranet‐based tool that allows groups to conform their RCAs to the ICPS framework. Because the absence of a standardized classification for patient safety concepts has hindered advances in patient safety,11 we believe I‐Tracker, or decision support tools like it that use the ICPS framework, can standardize RCAs and promote dissemination and adoption of action plans.

In‐hospital cardiopulmonary arrests are often preceded by signs of clinical instability, such as changes in vital signs or mental status.1 Nearly 85% of patients who suffer from cardiopulmonary arrests have documented observations of deterioration in the 8 hours before arrest.2 A Medical Emergency Team (MET), sometimes known as Rapid Response Team (RRT), can rapidly assess and manage unstable patients, with the goal that early intervention will prevent adverse outcomes. In 2004, the Institute for Healthcare Improvement (IHI), as part of its 100,000 Lives Campaign, called for hospitals to implement rapid response systems as 1 of 6 strategies to reduce deaths in hospital.3 Since this recommendation, hundreds of hospitals in the United States have invested substantial financial and personnel resources to implement some form of a rapid response system, which is comprised of a varying array of healthcare providers who bring critical care expertise to the patient's bedside.4, 5

Despite the intuitive appeal of the approach, and early observational data which suggested that METs could reduce both codes and unexpected in‐hospital mortality,2, 6 the largest randomized controlled trial found that METs failed to reduce unplanned intensive care unit (ICU) admissions, cardiac arrests, or unexpected deaths.7 More recently, in a prospective observational cohort study at 1 US hospital, Chan et al. found that a nurse‐led RRT did not impact hospital‐wide code rates or mortality.4

The study of rapid response systems is further complicated by a lack of standard definition, and the many types of hospitals in which they may be implemented. In 2006, a consensus conference suggested that MET be used to refer to teams led by physicians (usually intensivists), and RRT be used for teams led by nurses.8 Many studies have been conducted at non‐US institutions, and follow‐up periods have generally been 1 year or less. We report on almost 4 years of experience with a hospitalist‐led MET implemented in a major US academic medical center, and examine the subsequent changes in code calls, cardiac arrests, deaths following cardiopulmonary resuscitation, and overall hospital mortality. Because the MET did not operate in the critical care units, and because cardiac arrest may occur without prior signs of deterioration, we hypothesized that implementation of the MET would correspond to a small drop in total code calls, no change in codes called inside of critical care units, no change in cardiac arrest outside of critical care, and a marked drop in other medical crises (mostly respiratory distress) outside critical care. We also hypothesized that there would be no change in the rate of fatal codes, because most deaths occur in patients who were found to be pulseless on arrival of the code team.

METHODS

RESULTS

Implementation of the MET

The MET was introduced in the first and second quarters of 2006, with 2717 calls logged through the end of 2009 (out of 154,382 admissions). The rate of MET calls increased during the first 6 months of implementation from 5.95 per 1000 admissions in the first quarter of the intervention, to 15.8 calls per 1000 admissions in the second quarter. Call rates peaked in the first half of 2009, at 20.9 calls per 1000 admissions, leveling off to 17.9 calls per 1000 admissions in the last half of 2009 (Figure 1). Of calls with time recorded, 40% occurred on the day shift, 35% on the evening shift, and 25% on the night shift. The most common reason to call the MET was respiratory distress (33%), followed by cardiovascular instability (25%), and neurological abnormality (20%). In 15% of cases, concern about a patient's condition prompted the nurse to call. Calls came primarily from medical floors (75%) and surgical units (20%). The median response time was 4 minutes (interquartile range [IQR], 2.8 to 5.2 minutes) with no meaningful trend during the study period. The median call duration was 50 minutes (IQR, 38 to 72 minutes). Again, there was no trend over time. The most common interventions were arterial blood gas, fluid resuscitation, and electrocardiogram (see Supporting Web Appendix Table 1 in the online version of this article). A physician was present at 52% of the calls in the first year, which rose to 93% of calls in the final year. Approximately 25% of calls resulted in the patient being transferred to a critical care unit. Staff evaluations were overwhelmingly positive. Nurses rated the teams on the following points: whether the critical care nurse and respiratory therapist were efficient and respectful (mean rating 98%, SD 5.6%); promptness (98%, SD 5.6%); whether the patient's needs were addressed appropriately (mean 98%, SD 4.2%); whether the nurse felt supported by the MET (99.5%, SD 1.7%); and whether they would call the MET again (99.7%, SD 1.4%).

Figure 1

Effect of MET on Code Calls and Mortality

Between January 2004 and December of 2009, the hospital case mix index remained constant, and there were a total of 1202 codes called. The majority (62%) took place outside of critical care units. Linear contrasts of pre‐piecewise and post‐piecewise intercepts revealed that overall code calls declined significantly between pre‐implementation and post‐implementation of the MET from 7.30 (95% CI 5.81, 9.16) codes called per 1000 admissions to 4.21 (95% CI 3.42, 5.18) calls per 1000 admissions (Figure 1; also see Supporting Web Appendix Table 2 in the online version of this article). Outside of critical care, code calls declined from 4.70 (95% CI 3.92, 5.63) before the MET was implemented to 3.11 (95% CI 2.44, 3.97) afterwards (Figure 2); this was due primarily to a decrease in medical crises, which averaged 3.29 events per 1000 admissions (95% CI 2.70, 4.02) before implementation and decreased to 1.72 (95% CI 1.28, 2.31) afterwards, whereas cardiac arrests did not change significantly (Figure 3). Following implementation, code calls within critical care also declined significantly, from 2.59 events per 1000 admissions (95% CI 1.82, 3.69) before to 1.24 events per 1000 admissions (95% CI 0.94, 1.63) afterwards. The change in codes called within critical care was smaller, however, and included reductions in both cardiac arrests ( 0.84 events, P = 0.01) and medical crises ( 0.55, P = 0.08). There was no significant change in the rate of fatal codes per 1000 admissions ( +0.06, P = 0.65) (Figure 4). Overall hospital mortality remained steady at 22.0 deaths per 1000 admissions throughout the study period.

Figure 2

Figure 3

Figure 4

DISCUSSION

In this report, we detail the implementation of a novel hospitalist‐led medical emergency team at a large academic medical center over a period of 4 years. The team, which consisted of the patient's physician, a critical care nurse, a respiratory therapist, and an intravenous therapist, achieved full implementation within 6 months, was well received by the nursing staff, and was associated with a 42% decrease in code calls hospital‐wide. Most of the overall reduction was due to a reduction in codes called for medical crises outside of critical care, accompanied by a lesser reduction in codes called for cardiac arrests and medical crises within critical care units. There was no significant effect on the rate of cardiac arrest outside critical care. More importantly, there was no change in the rate of fatal codes or overall hospital mortality.

The idea of early intervention to prevent deterioration among hospitalized patients appeals to the concept that an ounce of prevention is worth a pound of cure. Like many other preventive interventions, rapid response systems have not always delivered on this promise. Since several early reports from Australia2 suggested that medical emergency teams could reduce not only cardiopulmonary arrests, but overall hospital mortality, there has been a rapid proliferation in their implementation, spurred on by the IHI's 100,000 Lives Campaign, which incorporated rapid response systems as one of 6 hospital‐wide interventions aimed at reducing harm and mortality.13 Subsequent randomized trials have both reproduced and refuted the early observational results. A ward‐randomized trial within 1 British hospital found a 50% reduction in hospital mortality for wards assigned to have an RRT,14 while a cluster randomized trial conducted at 23 Australian hospitals found no difference in rates of cardiac arrest or mortality between hospitals implementing METs and those continuing with usual care.7 Interestingly, in the Australian trial, the rates of cardiac arrest and mortality declined for both groups compared to historical controls, an important limitation to observational trials. Reports from single‐institution observational trials are also divided between those that found a reduction in mortality following implementation and those that did not. A recent meta‐analysis reported that there was too much heterogeneity among these trials to reach a conclusion about the benefits of rapid response systems.15

Our study adds to this literature in several ways. First, our MET design, which included the patient's physician (as opposed to an intensive care physician), was different from those previously studied. Including the patient's physician increases the team's knowledge of the patient and disease, and may improve physician acceptance of METs. In addition, our study provides 4 full years of follow‐up. Second, our rate of MET activation (18 calls/1000 admissions) was 2 to 3 times higher than that seen in most other studies,16 thus, the lack of mortality benefit was not likely the result of underuse. Third, our hospital employs a large number of hospitalists whose continuous presence might be expected to attenuate the benefits of an MET. Indeed, our initial rate of codes (7.5/1000 admissions) was similar to the post‐intervention rate in other studies.4 Nevertheless, the decrease in the overall rate of code calls following implementation of our MET was similar to that observed by others.17 Finally, our stratification of code calls inside critical care (where the MET was not deployed) and outside critical care, as well as the division of codes into cardiac arrest (where intervention is often unsuccessful) and other medical crises (primarily respiratory distress), gives further insight into how METs might work. As expected, we found that outside critical care only, codes called for medical crises declined, implying that the main effect of the MET was to provide early interventions for patients who were not likely to die anyway (eg, respiratory care for patients with respiratory distress or intravenous fluids for hypotensive patients). Instead of intervening to prevent death, MET may avoid emergent intubation by providing respiratory therapy and/or urgent intubation. In addition, it represents a less‐intense option for responding to nonlife‐threatening emergencies, such as seizures or syncope. As codes were no longer called for these types of crises, the rate of code calls necessarily fell. The reason that code calls declined inside critical care is less clear. It could be that patients transferred to critical care by the MET were less likely to code than those transferred before implementation, or the decline might be due to other factors that were not evaluated. Regardless, it is clear that the MET did not simply relocate codes to critical care units.

Our study has a number of limitations. First, it is an observational study and cannot account for other confounders relating to temporal trends in the hospital. However, our long time window allowed us to examine trends over several years. For 2 years prior to implementation of the MET, there was no decline at all in the rate of code calls, followed by an immediate and sustained drop after implementation. Other interventions, including ventilator‐associated pneumonia bundles, sepsis bundles, and advanced cardiac life support simulation training were also implemented at different times during the study period. However, the stark demarcation in code call rates coinciding with MET implementation makes it less likely that these other interventions were responsible for the observed decline. Second, our study was limited to a single institution and a single type of MET. Our findings may not apply to other types of institutions with different staffing arrangements or a different hospital culture, nor would they necessarily apply to different types of MET. Third, our nurse surveys were not collected anonymously, and this may have affected the nurses' responses. Finally, we did not collect physiological parameters on our patients, so we cannot state with certainty what the MET intervention accomplished.

Since initial studies suggested that METs could reduce hospital mortality rates, the Joint Commission has effectively mandated implementation of rapid response systems in all hospitals. Newer evidence, however, has been less convincing of mortality or other benefit. Our study adds to the literature in that we also did not find a mortality benefit. However, there were 2 clear benefits that we did identify. Our MET did appear to substantially reduce total numbers of code calls, particularly codes called for medical crises. Also, our nurses had a very positive response to the MET, which empowered them to get help for a patient when the patient's physician was unavailable or did not take their concerns seriously. Clearly, additional study is needed to better understand the effects of METs on mortality, codes, and other indicators of patient outcomes. However, in the current regulatory environment, such studies will be difficult to perform. Instead, additional studies can clarify which models deliver best outcomes and optimal use of our limited resources.

In‐hospital cardiopulmonary arrests are often preceded by signs of clinical instability, such as changes in vital signs or mental status.1 Nearly 85% of patients who suffer from cardiopulmonary arrests have documented observations of deterioration in the 8 hours before arrest.2 A Medical Emergency Team (MET), sometimes known as Rapid Response Team (RRT), can rapidly assess and manage unstable patients, with the goal that early intervention will prevent adverse outcomes. In 2004, the Institute for Healthcare Improvement (IHI), as part of its 100,000 Lives Campaign, called for hospitals to implement rapid response systems as 1 of 6 strategies to reduce deaths in hospital.3 Since this recommendation, hundreds of hospitals in the United States have invested substantial financial and personnel resources to implement some form of a rapid response system, which is comprised of a varying array of healthcare providers who bring critical care expertise to the patient's bedside.4, 5

Despite the intuitive appeal of the approach, and early observational data which suggested that METs could reduce both codes and unexpected in‐hospital mortality,2, 6 the largest randomized controlled trial found that METs failed to reduce unplanned intensive care unit (ICU) admissions, cardiac arrests, or unexpected deaths.7 More recently, in a prospective observational cohort study at 1 US hospital, Chan et al. found that a nurse‐led RRT did not impact hospital‐wide code rates or mortality.4

The study of rapid response systems is further complicated by a lack of standard definition, and the many types of hospitals in which they may be implemented. In 2006, a consensus conference suggested that MET be used to refer to teams led by physicians (usually intensivists), and RRT be used for teams led by nurses.8 Many studies have been conducted at non‐US institutions, and follow‐up periods have generally been 1 year or less. We report on almost 4 years of experience with a hospitalist‐led MET implemented in a major US academic medical center, and examine the subsequent changes in code calls, cardiac arrests, deaths following cardiopulmonary resuscitation, and overall hospital mortality. Because the MET did not operate in the critical care units, and because cardiac arrest may occur without prior signs of deterioration, we hypothesized that implementation of the MET would correspond to a small drop in total code calls, no change in codes called inside of critical care units, no change in cardiac arrest outside of critical care, and a marked drop in other medical crises (mostly respiratory distress) outside critical care. We also hypothesized that there would be no change in the rate of fatal codes, because most deaths occur in patients who were found to be pulseless on arrival of the code team.

METHODS

RESULTS

Implementation of the MET

The MET was introduced in the first and second quarters of 2006, with 2717 calls logged through the end of 2009 (out of 154,382 admissions). The rate of MET calls increased during the first 6 months of implementation from 5.95 per 1000 admissions in the first quarter of the intervention, to 15.8 calls per 1000 admissions in the second quarter. Call rates peaked in the first half of 2009, at 20.9 calls per 1000 admissions, leveling off to 17.9 calls per 1000 admissions in the last half of 2009 (Figure 1). Of calls with time recorded, 40% occurred on the day shift, 35% on the evening shift, and 25% on the night shift. The most common reason to call the MET was respiratory distress (33%), followed by cardiovascular instability (25%), and neurological abnormality (20%). In 15% of cases, concern about a patient's condition prompted the nurse to call. Calls came primarily from medical floors (75%) and surgical units (20%). The median response time was 4 minutes (interquartile range [IQR], 2.8 to 5.2 minutes) with no meaningful trend during the study period. The median call duration was 50 minutes (IQR, 38 to 72 minutes). Again, there was no trend over time. The most common interventions were arterial blood gas, fluid resuscitation, and electrocardiogram (see Supporting Web Appendix Table 1 in the online version of this article). A physician was present at 52% of the calls in the first year, which rose to 93% of calls in the final year. Approximately 25% of calls resulted in the patient being transferred to a critical care unit. Staff evaluations were overwhelmingly positive. Nurses rated the teams on the following points: whether the critical care nurse and respiratory therapist were efficient and respectful (mean rating 98%, SD 5.6%); promptness (98%, SD 5.6%); whether the patient's needs were addressed appropriately (mean 98%, SD 4.2%); whether the nurse felt supported by the MET (99.5%, SD 1.7%); and whether they would call the MET again (99.7%, SD 1.4%).

Figure 1

Effect of MET on Code Calls and Mortality

Between January 2004 and December of 2009, the hospital case mix index remained constant, and there were a total of 1202 codes called. The majority (62%) took place outside of critical care units. Linear contrasts of pre‐piecewise and post‐piecewise intercepts revealed that overall code calls declined significantly between pre‐implementation and post‐implementation of the MET from 7.30 (95% CI 5.81, 9.16) codes called per 1000 admissions to 4.21 (95% CI 3.42, 5.18) calls per 1000 admissions (Figure 1; also see Supporting Web Appendix Table 2 in the online version of this article). Outside of critical care, code calls declined from 4.70 (95% CI 3.92, 5.63) before the MET was implemented to 3.11 (95% CI 2.44, 3.97) afterwards (Figure 2); this was due primarily to a decrease in medical crises, which averaged 3.29 events per 1000 admissions (95% CI 2.70, 4.02) before implementation and decreased to 1.72 (95% CI 1.28, 2.31) afterwards, whereas cardiac arrests did not change significantly (Figure 3). Following implementation, code calls within critical care also declined significantly, from 2.59 events per 1000 admissions (95% CI 1.82, 3.69) before to 1.24 events per 1000 admissions (95% CI 0.94, 1.63) afterwards. The change in codes called within critical care was smaller, however, and included reductions in both cardiac arrests ( 0.84 events, P = 0.01) and medical crises ( 0.55, P = 0.08). There was no significant change in the rate of fatal codes per 1000 admissions ( +0.06, P = 0.65) (Figure 4). Overall hospital mortality remained steady at 22.0 deaths per 1000 admissions throughout the study period.

Figure 2

Figure 3

Figure 4

DISCUSSION

In this report, we detail the implementation of a novel hospitalist‐led medical emergency team at a large academic medical center over a period of 4 years. The team, which consisted of the patient's physician, a critical care nurse, a respiratory therapist, and an intravenous therapist, achieved full implementation within 6 months, was well received by the nursing staff, and was associated with a 42% decrease in code calls hospital‐wide. Most of the overall reduction was due to a reduction in codes called for medical crises outside of critical care, accompanied by a lesser reduction in codes called for cardiac arrests and medical crises within critical care units. There was no significant effect on the rate of cardiac arrest outside critical care. More importantly, there was no change in the rate of fatal codes or overall hospital mortality.

The idea of early intervention to prevent deterioration among hospitalized patients appeals to the concept that an ounce of prevention is worth a pound of cure. Like many other preventive interventions, rapid response systems have not always delivered on this promise. Since several early reports from Australia2 suggested that medical emergency teams could reduce not only cardiopulmonary arrests, but overall hospital mortality, there has been a rapid proliferation in their implementation, spurred on by the IHI's 100,000 Lives Campaign, which incorporated rapid response systems as one of 6 hospital‐wide interventions aimed at reducing harm and mortality.13 Subsequent randomized trials have both reproduced and refuted the early observational results. A ward‐randomized trial within 1 British hospital found a 50% reduction in hospital mortality for wards assigned to have an RRT,14 while a cluster randomized trial conducted at 23 Australian hospitals found no difference in rates of cardiac arrest or mortality between hospitals implementing METs and those continuing with usual care.7 Interestingly, in the Australian trial, the rates of cardiac arrest and mortality declined for both groups compared to historical controls, an important limitation to observational trials. Reports from single‐institution observational trials are also divided between those that found a reduction in mortality following implementation and those that did not. A recent meta‐analysis reported that there was too much heterogeneity among these trials to reach a conclusion about the benefits of rapid response systems.15

Our study adds to this literature in several ways. First, our MET design, which included the patient's physician (as opposed to an intensive care physician), was different from those previously studied. Including the patient's physician increases the team's knowledge of the patient and disease, and may improve physician acceptance of METs. In addition, our study provides 4 full years of follow‐up. Second, our rate of MET activation (18 calls/1000 admissions) was 2 to 3 times higher than that seen in most other studies,16 thus, the lack of mortality benefit was not likely the result of underuse. Third, our hospital employs a large number of hospitalists whose continuous presence might be expected to attenuate the benefits of an MET. Indeed, our initial rate of codes (7.5/1000 admissions) was similar to the post‐intervention rate in other studies.4 Nevertheless, the decrease in the overall rate of code calls following implementation of our MET was similar to that observed by others.17 Finally, our stratification of code calls inside critical care (where the MET was not deployed) and outside critical care, as well as the division of codes into cardiac arrest (where intervention is often unsuccessful) and other medical crises (primarily respiratory distress), gives further insight into how METs might work. As expected, we found that outside critical care only, codes called for medical crises declined, implying that the main effect of the MET was to provide early interventions for patients who were not likely to die anyway (eg, respiratory care for patients with respiratory distress or intravenous fluids for hypotensive patients). Instead of intervening to prevent death, MET may avoid emergent intubation by providing respiratory therapy and/or urgent intubation. In addition, it represents a less‐intense option for responding to nonlife‐threatening emergencies, such as seizures or syncope. As codes were no longer called for these types of crises, the rate of code calls necessarily fell. The reason that code calls declined inside critical care is less clear. It could be that patients transferred to critical care by the MET were less likely to code than those transferred before implementation, or the decline might be due to other factors that were not evaluated. Regardless, it is clear that the MET did not simply relocate codes to critical care units.

Our study has a number of limitations. First, it is an observational study and cannot account for other confounders relating to temporal trends in the hospital. However, our long time window allowed us to examine trends over several years. For 2 years prior to implementation of the MET, there was no decline at all in the rate of code calls, followed by an immediate and sustained drop after implementation. Other interventions, including ventilator‐associated pneumonia bundles, sepsis bundles, and advanced cardiac life support simulation training were also implemented at different times during the study period. However, the stark demarcation in code call rates coinciding with MET implementation makes it less likely that these other interventions were responsible for the observed decline. Second, our study was limited to a single institution and a single type of MET. Our findings may not apply to other types of institutions with different staffing arrangements or a different hospital culture, nor would they necessarily apply to different types of MET. Third, our nurse surveys were not collected anonymously, and this may have affected the nurses' responses. Finally, we did not collect physiological parameters on our patients, so we cannot state with certainty what the MET intervention accomplished.

Since initial studies suggested that METs could reduce hospital mortality rates, the Joint Commission has effectively mandated implementation of rapid response systems in all hospitals. Newer evidence, however, has been less convincing of mortality or other benefit. Our study adds to the literature in that we also did not find a mortality benefit. However, there were 2 clear benefits that we did identify. Our MET did appear to substantially reduce total numbers of code calls, particularly codes called for medical crises. Also, our nurses had a very positive response to the MET, which empowered them to get help for a patient when the patient's physician was unavailable or did not take their concerns seriously. Clearly, additional study is needed to better understand the effects of METs on mortality, codes, and other indicators of patient outcomes. However, in the current regulatory environment, such studies will be difficult to perform. Instead, additional studies can clarify which models deliver best outcomes and optimal use of our limited resources.

In‐hospital cardiopulmonary arrests are often preceded by signs of clinical instability, such as changes in vital signs or mental status.1 Nearly 85% of patients who suffer from cardiopulmonary arrests have documented observations of deterioration in the 8 hours before arrest.2 A Medical Emergency Team (MET), sometimes known as Rapid Response Team (RRT), can rapidly assess and manage unstable patients, with the goal that early intervention will prevent adverse outcomes. In 2004, the Institute for Healthcare Improvement (IHI), as part of its 100,000 Lives Campaign, called for hospitals to implement rapid response systems as 1 of 6 strategies to reduce deaths in hospital.3 Since this recommendation, hundreds of hospitals in the United States have invested substantial financial and personnel resources to implement some form of a rapid response system, which is comprised of a varying array of healthcare providers who bring critical care expertise to the patient's bedside.4, 5

Despite the intuitive appeal of the approach, and early observational data which suggested that METs could reduce both codes and unexpected in‐hospital mortality,2, 6 the largest randomized controlled trial found that METs failed to reduce unplanned intensive care unit (ICU) admissions, cardiac arrests, or unexpected deaths.7 More recently, in a prospective observational cohort study at 1 US hospital, Chan et al. found that a nurse‐led RRT did not impact hospital‐wide code rates or mortality.4

The study of rapid response systems is further complicated by a lack of standard definition, and the many types of hospitals in which they may be implemented. In 2006, a consensus conference suggested that MET be used to refer to teams led by physicians (usually intensivists), and RRT be used for teams led by nurses.8 Many studies have been conducted at non‐US institutions, and follow‐up periods have generally been 1 year or less. We report on almost 4 years of experience with a hospitalist‐led MET implemented in a major US academic medical center, and examine the subsequent changes in code calls, cardiac arrests, deaths following cardiopulmonary resuscitation, and overall hospital mortality. Because the MET did not operate in the critical care units, and because cardiac arrest may occur without prior signs of deterioration, we hypothesized that implementation of the MET would correspond to a small drop in total code calls, no change in codes called inside of critical care units, no change in cardiac arrest outside of critical care, and a marked drop in other medical crises (mostly respiratory distress) outside critical care. We also hypothesized that there would be no change in the rate of fatal codes, because most deaths occur in patients who were found to be pulseless on arrival of the code team.

METHODS

RESULTS

Implementation of the MET

The MET was introduced in the first and second quarters of 2006, with 2717 calls logged through the end of 2009 (out of 154,382 admissions). The rate of MET calls increased during the first 6 months of implementation from 5.95 per 1000 admissions in the first quarter of the intervention, to 15.8 calls per 1000 admissions in the second quarter. Call rates peaked in the first half of 2009, at 20.9 calls per 1000 admissions, leveling off to 17.9 calls per 1000 admissions in the last half of 2009 (Figure 1). Of calls with time recorded, 40% occurred on the day shift, 35% on the evening shift, and 25% on the night shift. The most common reason to call the MET was respiratory distress (33%), followed by cardiovascular instability (25%), and neurological abnormality (20%). In 15% of cases, concern about a patient's condition prompted the nurse to call. Calls came primarily from medical floors (75%) and surgical units (20%). The median response time was 4 minutes (interquartile range [IQR], 2.8 to 5.2 minutes) with no meaningful trend during the study period. The median call duration was 50 minutes (IQR, 38 to 72 minutes). Again, there was no trend over time. The most common interventions were arterial blood gas, fluid resuscitation, and electrocardiogram (see Supporting Web Appendix Table 1 in the online version of this article). A physician was present at 52% of the calls in the first year, which rose to 93% of calls in the final year. Approximately 25% of calls resulted in the patient being transferred to a critical care unit. Staff evaluations were overwhelmingly positive. Nurses rated the teams on the following points: whether the critical care nurse and respiratory therapist were efficient and respectful (mean rating 98%, SD 5.6%); promptness (98%, SD 5.6%); whether the patient's needs were addressed appropriately (mean 98%, SD 4.2%); whether the nurse felt supported by the MET (99.5%, SD 1.7%); and whether they would call the MET again (99.7%, SD 1.4%).

Figure 1

Effect of MET on Code Calls and Mortality

Between January 2004 and December of 2009, the hospital case mix index remained constant, and there were a total of 1202 codes called. The majority (62%) took place outside of critical care units. Linear contrasts of pre‐piecewise and post‐piecewise intercepts revealed that overall code calls declined significantly between pre‐implementation and post‐implementation of the MET from 7.30 (95% CI 5.81, 9.16) codes called per 1000 admissions to 4.21 (95% CI 3.42, 5.18) calls per 1000 admissions (Figure 1; also see Supporting Web Appendix Table 2 in the online version of this article). Outside of critical care, code calls declined from 4.70 (95% CI 3.92, 5.63) before the MET was implemented to 3.11 (95% CI 2.44, 3.97) afterwards (Figure 2); this was due primarily to a decrease in medical crises, which averaged 3.29 events per 1000 admissions (95% CI 2.70, 4.02) before implementation and decreased to 1.72 (95% CI 1.28, 2.31) afterwards, whereas cardiac arrests did not change significantly (Figure 3). Following implementation, code calls within critical care also declined significantly, from 2.59 events per 1000 admissions (95% CI 1.82, 3.69) before to 1.24 events per 1000 admissions (95% CI 0.94, 1.63) afterwards. The change in codes called within critical care was smaller, however, and included reductions in both cardiac arrests ( 0.84 events, P = 0.01) and medical crises ( 0.55, P = 0.08). There was no significant change in the rate of fatal codes per 1000 admissions ( +0.06, P = 0.65) (Figure 4). Overall hospital mortality remained steady at 22.0 deaths per 1000 admissions throughout the study period.

Figure 2

Figure 3

Figure 4

DISCUSSION

In this report, we detail the implementation of a novel hospitalist‐led medical emergency team at a large academic medical center over a period of 4 years. The team, which consisted of the patient's physician, a critical care nurse, a respiratory therapist, and an intravenous therapist, achieved full implementation within 6 months, was well received by the nursing staff, and was associated with a 42% decrease in code calls hospital‐wide. Most of the overall reduction was due to a reduction in codes called for medical crises outside of critical care, accompanied by a lesser reduction in codes called for cardiac arrests and medical crises within critical care units. There was no significant effect on the rate of cardiac arrest outside critical care. More importantly, there was no change in the rate of fatal codes or overall hospital mortality.

The idea of early intervention to prevent deterioration among hospitalized patients appeals to the concept that an ounce of prevention is worth a pound of cure. Like many other preventive interventions, rapid response systems have not always delivered on this promise. Since several early reports from Australia2 suggested that medical emergency teams could reduce not only cardiopulmonary arrests, but overall hospital mortality, there has been a rapid proliferation in their implementation, spurred on by the IHI's 100,000 Lives Campaign, which incorporated rapid response systems as one of 6 hospital‐wide interventions aimed at reducing harm and mortality.13 Subsequent randomized trials have both reproduced and refuted the early observational results. A ward‐randomized trial within 1 British hospital found a 50% reduction in hospital mortality for wards assigned to have an RRT,14 while a cluster randomized trial conducted at 23 Australian hospitals found no difference in rates of cardiac arrest or mortality between hospitals implementing METs and those continuing with usual care.7 Interestingly, in the Australian trial, the rates of cardiac arrest and mortality declined for both groups compared to historical controls, an important limitation to observational trials. Reports from single‐institution observational trials are also divided between those that found a reduction in mortality following implementation and those that did not. A recent meta‐analysis reported that there was too much heterogeneity among these trials to reach a conclusion about the benefits of rapid response systems.15

Our study adds to this literature in several ways. First, our MET design, which included the patient's physician (as opposed to an intensive care physician), was different from those previously studied. Including the patient's physician increases the team's knowledge of the patient and disease, and may improve physician acceptance of METs. In addition, our study provides 4 full years of follow‐up. Second, our rate of MET activation (18 calls/1000 admissions) was 2 to 3 times higher than that seen in most other studies,16 thus, the lack of mortality benefit was not likely the result of underuse. Third, our hospital employs a large number of hospitalists whose continuous presence might be expected to attenuate the benefits of an MET. Indeed, our initial rate of codes (7.5/1000 admissions) was similar to the post‐intervention rate in other studies.4 Nevertheless, the decrease in the overall rate of code calls following implementation of our MET was similar to that observed by others.17 Finally, our stratification of code calls inside critical care (where the MET was not deployed) and outside critical care, as well as the division of codes into cardiac arrest (where intervention is often unsuccessful) and other medical crises (primarily respiratory distress), gives further insight into how METs might work. As expected, we found that outside critical care only, codes called for medical crises declined, implying that the main effect of the MET was to provide early interventions for patients who were not likely to die anyway (eg, respiratory care for patients with respiratory distress or intravenous fluids for hypotensive patients). Instead of intervening to prevent death, MET may avoid emergent intubation by providing respiratory therapy and/or urgent intubation. In addition, it represents a less‐intense option for responding to nonlife‐threatening emergencies, such as seizures or syncope. As codes were no longer called for these types of crises, the rate of code calls necessarily fell. The reason that code calls declined inside critical care is less clear. It could be that patients transferred to critical care by the MET were less likely to code than those transferred before implementation, or the decline might be due to other factors that were not evaluated. Regardless, it is clear that the MET did not simply relocate codes to critical care units.

Our study has a number of limitations. First, it is an observational study and cannot account for other confounders relating to temporal trends in the hospital. However, our long time window allowed us to examine trends over several years. For 2 years prior to implementation of the MET, there was no decline at all in the rate of code calls, followed by an immediate and sustained drop after implementation. Other interventions, including ventilator‐associated pneumonia bundles, sepsis bundles, and advanced cardiac life support simulation training were also implemented at different times during the study period. However, the stark demarcation in code call rates coinciding with MET implementation makes it less likely that these other interventions were responsible for the observed decline. Second, our study was limited to a single institution and a single type of MET. Our findings may not apply to other types of institutions with different staffing arrangements or a different hospital culture, nor would they necessarily apply to different types of MET. Third, our nurse surveys were not collected anonymously, and this may have affected the nurses' responses. Finally, we did not collect physiological parameters on our patients, so we cannot state with certainty what the MET intervention accomplished.

Since initial studies suggested that METs could reduce hospital mortality rates, the Joint Commission has effectively mandated implementation of rapid response systems in all hospitals. Newer evidence, however, has been less convincing of mortality or other benefit. Our study adds to the literature in that we also did not find a mortality benefit. However, there were 2 clear benefits that we did identify. Our MET did appear to substantially reduce total numbers of code calls, particularly codes called for medical crises. Also, our nurses had a very positive response to the MET, which empowered them to get help for a patient when the patient's physician was unavailable or did not take their concerns seriously. Clearly, additional study is needed to better understand the effects of METs on mortality, codes, and other indicators of patient outcomes. However, in the current regulatory environment, such studies will be difficult to perform. Instead, additional studies can clarify which models deliver best outcomes and optimal use of our limited resources.