Implementation of the RRT was associated with a significant reduction in hospital-wide mortality. In the 3-year postintervention period, 139 fewer deaths occurred than would be expected from the preintervention mortality rate. Additionally, 132 fewer out-of-ICU codes and 126 fewer out-of-ICU code-related deaths were found after RRT implementation than would be expected from preintervention rates.
Attributing the reduction in hospital-wide mortality to the RRT is strongly supported by the concurrent 51% decrease in out-of-ICU cardiopulmonary arrest codes and 35% decrease in out-of-ICU mortality (Table ). Despite a substantial increase in the case-mix index, which indicated a secular trend toward increasing severity of illness among admitted patients, hospital-wide mortality decreased by 1.76 deaths per 1, 000 discharges. This mortality reduction remained statistically significant after adjusting for trends over time, and it was driven by a 39% decrease in mortality on the medicine service, where 75% of the RRTs took place. A commensurate reduction of 1.66 out-of-ICU cardiopulmonary-arrest codes and 1.60 out-of-ICU code-related deaths per 1, 000 discharges occurred, supporting attribution of the hospital-wide mortality reduction to the prevention of out-of-ICU arrests. The decreases in hospital-wide mortality, out-of-ICU mortality, and out-of-ICU cardiopulmonary-arrest codes all coincided temporally with introduction of the RRT in 2006 (Figure ).
Moreover, the reduction in hospital-wide mortality was nearly fully matched by a commensurate decrease in out-of-ICU codes and code-related mortality. The slightly greater reduction in hospital-wide mortality may be accounted for by additional lives saved during RRTs activated for patients with advanced directives specifying DNR. Any improvement in mortality among DNR patients due to the RRT would not be reflected by changes in code-related deaths, as DNR patients were explicitly excluded from the cardiopulmonary-arrest code response. Alternatively, this difference may be due to confounding influences that the study was not designed to capture.
The descriptive findings of RRT activations further support attribution of the reduction in hospital-wide mortality to the RRT. The most common diagnoses that immediately led to RRT activation--seizure, severe sepsis, arrhythmia, pneumonia, and aspiration--are high-acuity conditions that often require ICU-level care. The most common therapeutic interventions performed by the RRT centered on airway, respiratory, and cardiovascular support (Table ). Furthermore, no other concomitant patient-safety or quality-improvement interventions were introduced during the study period to account for the improved clinical outcomes. Although introducing hospitalists and intensivists previously was shown to reduce inpatient mortality [36
], no substantial changes in hospital staffing or work hours occurred during the study period.
Previously, most large studies assessing hospital-wide mortality failed to demonstrate reduced mortality after RRT implementation [23
], as did a recent meta-analysis evaluating hospital-wide mortality of more than 400, 000 patients [29
]. Two likely explanations exist for the mortality reduction seen here, but not in most previous RRT studies.
First, this RRT was widely used, with 10.8 activations per 1, 000 hospital-wide discharges and 21.6 activations per 1, 000 medicine service discharges. A comparable hospital-wide activation rate of 9.3 RRT activations per 1, 000 admissions was observed in the only other large study to date to find a significant reduction in postintervention hospital-wide mortality [30
]. By contrast, two of the three previous largest negative studies reported lower RRT utilization of 2.5 [28
] and 8.7 [27
] activations per 1, 000 admissions. In the MERIT trial [27
], the only large cluster-randomized study to date, just 41% of patients who had RRT-activation criteria present more than 15 minutes before an adverse event actually had an RRT activated. Yet, a subsequent post hoc
analysis of MERIT [40
] demonstrated a significant dose-response relation between rate of RRT activation and incidence of cardiac arrests and deaths.
Underutilization is commonly reported in other RRT studies and may minimize improvements in clinical outcomes gained with adopting an otherwise effective RRT [26
]. Delayed activation similarly has been associated with increased mortality [34
]. Whereas failure to activate an RRT promptly may reflect insufficient staff awareness, most studies highlight the great lengths taken to promote their system. Thus, failure of prompt activation may instead reflect reluctance by nurses, junior physicians, and allied health professionals to go outside the traditional hierarchic model for referrals of clinical management (that is, junior nurse to senior nurse to junior physician to senior physician), even for acutely decompensating patients who meet criteria for RRT activation [21
In this case, several features of RRT design and hospital culture promoted greater utilization. First, the RRT was run by the medical consult resident, who was already widely recognized by nurses as leader of the cardiopulmonary-arrest code team. Conversely, introduction of a new and unfamiliar staff member, such as a senior attending or intensivist, to lead the team might have introduced a psychological barrier to nurses activating the RRT in borderline cases. Second, as a public teaching institution staffed by house-officer trainees and salaried teaching attending physicians, the hospital had a preexisting culture of shared responsibility for patient care, leading to wide acceptance of the RRT model by the physician and nurse staffs. Third, the training of hospital staff explicitly encouraged heavy use of the RRT with a low threshold for activation and emphasized clinical judgment of nurses as a key activation criterion, creating an avenue for nursing empowerment. As a result, nurses took ownership of the RRT, accounting for 85% of all RRT activations and a higher rate of hospital-wide utilization than found in most other studies.
A second plausible explanation for the reduction in hospital-wide mortality is the emphasis that was placed on clinical judgment as a key criterion for RRT activation during staff training sessions, which may have prompted earlier RRT activations before intractable clinical deterioration. Although most previous studies have included clinical judgment as an activation criterion, the RRT in these studies was only infrequently activated for this reason [22
]. By contrast, 43% of RRTs in the present study were activated for reasons other than vital-sign derangements. Furthermore, vital signs--and in particular respiratory rate--are often inaccurately measured and recorded [42
], which may lead to further underutilization of the RRT. Potential failures to trigger RRT activation through vital-signs criteria may have been circumvented by emphasizing clinical judgment, as reflected in the high rate of RRT activation for "staff worried: clinical judgment that patient does not look right."
Interpretation of these findings shares limitations similar to those of most other RRT studies. Common to any cohort study with historical controls, it is possible that improvements in mortality and code rates were due to differences in the preintervention and postintervention populations rather than to the RRT itself. However, the trend toward increased severity of illness throughout the study period, as measured by the case-mix index, likely would have increased the risk of death during the postintervention period. Also considered was whether improvements in mortality reflected a secular downward trend that predated the RRT, yet the mortality rate actually trended upward during the 3-year preintervention period, coinciding with the increasing case-mix index. Adjusting for this time trend thus increased the estimated postintervention mortality improvement. However, the adjusted results had much wider confidence intervals and therefore provide a less reliable point estimate, compared with the unadjusted results, of the decreased mortality after RRT implementation. Regardless, hospital-wide mortality remained significantly lower after RRT implementation after controlling for underlying trends in mortality over time.
Several outcomes measures, including out-of-ICU mortality and cardiopulmonary-arrest codes, could be favorably biased by excluding mortality of patients transferred to the ICU. This study instead focused on hospital-wide mortality, which avoids this bias by counting all deaths regardless of where they occur in the hospital. Still, other unmeasured confounders, which our study was not designed to capture, may have favorably biased the study results. For example, it is possible that the transfer of patients to outside hospice or other long-term care facilities increased during the study period. Yet, no changes to palliative care services occurred throughout the study period, and the improvement in overall mortality was nearly completely accounted for by the decrease in out-of-ICU codes and associated deaths.
The low rate of RRT activation and low baseline mortality among nonmedical services may limit generalizability of these results to hospitals with a large nonmedical patient composition or differences in service-specific mortality. Additionally, these findings reflect a single tertiary referral public teaching hospital's experience and may not be generalizable to nonteaching or lower-acuity hospitals. Finally, difficulty in standardizing clinical judgment as a criterion for RRT activation may limit generalizability across institutions with different levels of staff experience and out-of-ICU monitoring. It may be precisely this flexibility, to activate an RRT for any reasonable clinical judgment without threat of repudiation or reprisal, that resulted in high RRT utilization and led to the mortality reduction observed.