Ensuring No Death Is Ignored: Development of a Hospital Mortality Review and Notification System

C.S. Mott Children’s Hospital is a 250-bed pediatric hospital in the University of Michigan Health System. In early 2018, the hospital associate chief clinical officer for quality (K.B.) formed a workgroup with representatives from clinical areas where most inpatient mortalities occurred (eg, ICUs). This workgroup included physicians, physician trainees, registered nurses, the patient safety program manager, and 1 nurse practitioner. This work was deemed not regulated by the University of Michigan’s institutional review board.

The workgroup began by characterizing our institution’s current state of mortality data and M&M conferences. The structure, format, and documentation of M&M conferences varied widely across divisions, as did identification of cases for discussion and expectations for formation of action plans after the discussion. Some medical divisions had no M&M discussions, and many surgical divisions (eg, pediatric otolaryngology) participated in departmental M&Ms rather than pediatric-specific discussions.

Next, the workgroup designed a future state process map for identification and review of inpatient mortalities (Fig 1). We drafted a mortality review form to be completed by designated physician reviewers from the discharge service, including questions about plans for further review in an M&M conference or other forum. Using a weekly electronic medical record report, the patient safety program manager e-mailed each deceased patient’s demographic information and an online mortality review form link to the local reviewer. Beginning in October 2018, we used Plan-Do-Study-Act cycles to pilot the form with the services represented in our workgroup for 3 months. We disseminated the process to all pediatric services starting in January 2019. On the basis of feedback from participants, the survey was modified to improve survey flow and add clarity (eg, Question 16: Rationale for further review of a case). In December 2019, we expanded the scope of reviews to include postdischarge mortalities within 30 days of hospital discharge utilizing data from the Michigan Health Information Network. We used the same review form and notification system for these postdischarge mortalities.

We developed 2 notification mechanisms to increase awareness of mortalities. First, we added a graph displaying the number of deaths by hospital unit to our monthly hospital safety newsletter, along with the first names of the deceased to personalize the numbers (Fig 2). Second, a monthly notification which included a brief “one-liner” summary for each patient was created to improve awareness of inpatient deaths for rotating residents, faculty, and other clinicians. The notification provided information about planned local reviews, such as M&M conferences, so recipients could participate if interested. The notification was originally e-mailed to pediatric division chiefs and residents with the request that it be shared with those faculty and staff who cared for the included patients. After receiving feedback, we expanded the mailing list to include nursing leaders and other faculty who requested to be added. We considered sending notifications only to those physicians who had cared for a patient during the current admission; however, this was not practical for patients with prolonged hospitalizations or with multiple admissions over longer periods of time. At the same time, given the potential sensitivity of the information, we did not want to send it to all hospital faculty and staff. The notification includes our institutional confidentiality statement because the information contained within was gathered for the purpose of QI activities.

The primary outcome measure was the percentage of pediatric inpatient mortalities that had a completed mortality review form. Our process measure was days between date of mortality and the survey completion date, summarized by month of mortality to reflect the timeliness of our review process. If >1 entry was made per mortality, the earliest review was selected for the data set (n = 13 duplications).

We constructed a P chart using QI Macros (KnowWare International, Denver) to analyze the outcome measure, using 20 data points to establish baseline upper and lower control limits (set at 3 times the SD) and centerline (mean). We did not calculate this measure during the pilot period because the denominator was limited by the pilot. For the process measure, we used an XbarS chart, a 2-part chart that uses the same x-axis to show the Xbar chart (mean days between the mortality and the review) and the S chart (SD) summarized monthly. For the XbarS chart, the baseline upper and lower control and centerline were calculated using the first 16 data points rather than the typical 20 to 30 data points on the basis of signal from the coronavirus disease 2019 pandemic. If standard health care rules for special cause variation were met in any chart, such as the 8-point rule, a new centerline and control limits were calculated and extended to evaluate for future special cause variation. If special cause variation was noted in either the Xbar or the S chart, a new centerline and control limits were calculated for both charts.¹³  Because this measure was not affected by the smaller number of mortalities included in the pilot, we calculated the time to review completion during the pilot period (October–December 2018) to have more points for our control chart.

In addition to ongoing workgroup discussions about causes of mortality, 4 coauthors (K.B., K.L., A.M., R.V.) reviewed narrative information submitted on all completed forms in this data set (eg, Question 20: Brief description of patient) to identify themes and opportunities for systemwide improvement. We also performed a formal qualitative analysis of responses to questions about rationale for the type of further review chosen for all mortalities between January 2019 and June 2021 that had completed review forms. Responses were coded by 2 researchers (B.P. and M.M.) using principles of grounded theory analysis.¹²  Discrepancies were resolved via consensus or using a third coauthor (K.B.).

To evaluate the monthly notification, we surveyed recipients, including questions about how recipients interacted with the e-mail and whether the notification improved their awareness of inpatient deaths, clinical learning, and practice. We tested the survey with representative respondents and revised questions on the basis of their feedback (Appendix 1). The survey link was included in the monthly e-mail for 3 months and thus was sent to ∼350 recipients, primarily physicians (as outlined in Methods). Additionally, a pediatric chief resident sent a targeted e-mail with the survey link to pediatric residents. Because recipients of the monthly notifications are encouraged to share them as appropriate, we asked that this survey be shared with colleagues who may have previously received the notifications. As stated in the survey introduction, no identifiable information was collected from respondents. Survey responses were summarized using descriptive statistics. Of note, the survey design did not require answers to all questions and some questions allowed for multiple selections, so we received different numbers of answers for specific questions.

There were 361 deaths recorded in our system between October 2018 and February 2022, including 42 postdischarge mortalities after we began tracking them in December 2019. Of the 361 mortalities, 295 were reviewed, including 11 during the pilot period (October–December 2018). After the pilot, 284 out of 328 mortalities (86.6%) were reviewed. There were 286 inpatient mortalities with 277 completed reviews (96.9%). Only 7 of 42 (17%) postdischarge mortalities had completed review forms. Thus, 35 of 44 (79.5%) mortalities that did not have completed review forms were postdischarge mortalities. For the qualitative analysis, we analyzed 229 completed review forms for mortalities that occurred between January 1, 2019, and June 30, 2021.

As a QI effort, the mortality review form was iteratively adapted on the basis of feedback. The current version of the review form is shown in Appendix 2.

The outcome measure of mortality review form completion P chart had an initial centerline of 80.6%. We identified special cause variation in April 2021 and recalculated the centerline at 86.4% (Fig 3). For our process measure (days between mortality and review), the initial Xbar centerline was 37.6 days, with multiple points of special cause variation on the S chart, indicating considerable variability in timeliness to review between January 2018 and January 2020 (Fig 4). Starting in February 2020, the centerline increases to 137.1 days (365% increase). After the initial 12 months of the pandemic, the Xbar centerline decreased to 19.2 days, a 48.6% reduction. Similarly, the S chart demonstrated special cause variation in this period, indicating that mortality reviews were more consistently completed earlier compared with baseline.

Review of the completed forms for system-level opportunities for improvement revealed relatively few preventable mortalities on the basis of available information. Many patients had poor prognoses on admission or high-risk conditions (eg, extreme prematurity). Of note, 1 potentially preventable condition identified in 35.4% (84 of 326) of mortalities reviewed was sepsis. Although most of these patients had sepsis-associated organ dysfunction present (71 of 84, 84.5%), the vast majority had fully recovered from organ dysfunction at the time of death (65 of 71, 91.5%).

Of 229 mortality review forms that answered the question about further review and rationale for this decision, 154 (67.2%) indicated that they would undergo some type of further review. Of these, 80 of 154 (51.9%) were local M&M conferences, 7 of 154 (4.5%) were undergoing root cause analysis through the hospital patient safety program (which focuses primarily on serious safety events⁹ ), 6 of 154 (3.9%) had a staff debrief, and 62 of 154 (40.3%) answered other. In terms of rationale for the further review type, 88 of 154 (57.1%) were included in a standardized review process for all mortalities in the division, 33 of 154 (21.4%) had opportunities for improvement identified, 18 of 154 (11.7%) were chosen for other reasons, and 15 of 154 (9.7%) were reviewed in at least 2 local review processes (eg, standardized review and review because of opportunities for improvement). Of the 66 mortalities which did not undergo any further review, the rationale was identified as being a nonpreventable death in 36 of 66 (54.5%), palliative care death in 24 of 66 (36.4%), death of an extremely preterm infant in 5 of 66 (7.6%), and 1 of 66 (1.5%) occurred after discharge with insufficient information for further review.

Ninety-one respondents completed at least part of the mortality notification survey. Although it was impossible to calculate an exact denominator because the survey e-mail may have been forwarded, this represented a maximum response rate of ∼26% (91 of 350). Of the 91 survey respondents, 46 (51%) were resident physicians, 25 (28%) were attending physicians, 11 (12%) were fellow physicians, and the remaining 7 (8%) were advanced practice providers, nursing leaders, and other job categories. Most respondents (79 of 86, 92%) reported they review the e-mail and 8% (7 of 86) had forwarded it to colleagues. None of the respondents replied that they delete the e-mail without reading it. Most agreed they had learned about the death of at least 1 of the patients they had cared for (70 of 79, 89%). Respondents reported seeking out additional information using the following sources: discussion with a colleague in their division or program (46 of 79, 58%), reviewing the electronic medical record (45 of 79, 57%), attending an M&M conference or other discussion (18 of 79, 23%), or discussion with a colleague in a different division or program (17 of 79, 22%). Most respondents agreed the e-mail notifications add value to their personal clinical learning (71 of 86, 83%) and practice (62 of 86, 72%), and the notifications help them process grief associated with losing a patient (65 of 86, 76%) (Fig 5). Sample comments are reported in Table 1.

We identified a gap in our hospital’s knowledge of inpatient mortalities and successfully created a process to identify, review, and increase awareness of all inpatient mortalities using a distributed model of review. This process was feasible with high rates of review completion, particularly for mortalities that occurred in our hospital, and did not require excessive time from any single committee. The vast majority of the mortalities that were not reviewed occurred after discharge, reflecting the difficulty in obtaining information on patients who die outside of our hospital. Over time, we have decreased the average time to review completion through repeated reminders and other improvements. Having a standard review process provides better data to improve our understanding of the population of children who die in our hospital or shortly after discharge. Additionally, our mortality notification system has been appreciated by clinicians.

Although some publications report hospitalwide mortality review processes and notification systems,^4–8,10  there are few specific to pediatrics. Because inpatient pediatric mortalities occur less frequently than adult mortalities, it was feasible to establish a process for review by local reviewers. Using a small group of specialty-specific reviewers promoted familiarity and internal consistency with the process, but may also have delayed review completion as compared with having a single central review process.⁴  Increased awareness provided by the weekly mortality report has also been helpful for hospital leadership to provide emotional support to unit leaders and staff.

Iterative adaptation of the mortality review form and the review request system led to decreased average time to mortality review completion. Engaging a dedicated group of local reviewers, many of whom participated on our committee, enabled us to adapt the review form more efficiently. We noted key exceptions to timely completion of reviews, namely during the first year of the coronavirus disease 2019 pandemic. This is likely because of many factors, including the stress of the pandemic and, more specifically for our institution, changes in our hospital’s operations as a children’s hospital within a larger health system. After this period, mean time to complete reviews decreased significantly because of enhancements in our notification and reminder process and additional reviewers for services with high numbers of mortalities. This decrease in time to review is important because we believe timely reviews are optimal for just-in-time ongoing education and continual improvements.

The two mechanisms we developed for mortality notifications have been well received. Most respondents to our e-mail survey replied that they learned about at least 1 patient death from the notification, which likely reflects the fact that patients in a tertiary care children’s hospital receive care from multiple primary and consulting services during a given episode of care. Many respondents indicated they had sought additional information about a patient’s course, which is consistent with most respondents agreeing the notifications add value to their learning and practice of medicine. Finally, most respondents somewhat or strongly agreed the notifications helped process grief. Although the notifications themselves may not be able to fully assist with the grieving process, they can at least serve as an important first step by providing awareness of a patient’s death. Future work will focus on providing more constructive support for providers.

Our initial experience with this system of distributed mortality reviews has validated our ongoing quality efforts while raising new questions. We previously had no mechanism to track how often mortalities were reviewed within divisions. Our finding that 70% of mortalities undergo some type of further review provides a useful starting point, although we do not know what the “correct” rate of further reviews should be. Because only 36% of mortalities were reviewed in an M&M conference, having a standard centralized review for all inpatient mortalities provides a significant increase in our awareness and ability to learn from them. We found that sepsis was common during inpatient encounters that ended in death, which supports our efforts to improve timely sepsis recognition and treatment, as well as efforts to prevent health care-acquired infections. Aside from sepsis, it is difficult to identify themes of preventable harm in the mortality reviews using our current form. It is possible that a more structured review form asking respondents to categorize types of issues identified and a system that includes multiple perspectives from various clinicians involved would identify more generalizable themes to drive QI efforts. Despite this limitation, local leaders (eg, unit medical director) have found significant opportunities for improvement from engaging in local reviews. For example, the NICU has identified the need to update existing policies on safely intubating neonates and write a new guideline for massive transfusion on the basis of these reviews. The PICU has recognized variability in end-of-life symptom management and postmortem support provided to staff, and is thus actively working to standardize these practices.

Sharing within the committee also allowed for dissemination of best practices: For example, our pediatric critical care medicine group formed a multidisciplinary committee to review all mortalities and major morbidities to determine which should be reviewed at their division M&M or referred to other committees. This practice was then adopted by our neonatology group and adapted by the pediatric heart center. Sharing our pediatric inpatient mortality review process across our health system has informed a new standard mechanism to review adult mortalities. We recommend that other institutions consider establishing a similar system for regular reviews of data and sharing of best local practices, which should include early discussion of any notification system with compliance and legal teams.

Limitations of our approach include lack of generalizability if other institutions’ cultures are not supportive. Reviews are typically completed by a few physician reviewers per division, which may bias responses and may not reliably include a multidisciplinary perspective as described by other authors.^4–6  Encouraging reviewers to discuss cases with physician, nurse, and other colleagues or the formation of multidisciplinary review committees may mitigate this problem. We were unable to calculate a response rate for the monthly mortality notification survey because we do not know how many people received it as a forwarded invitation. The responses we received were likely biased toward recipients who read the monthly mortality notifications and thus may overestimate their value. Finally, most survey respondents were physicians, so our findings may not apply to other staff.

In conclusion, we successfully created a system for identification, review, and notification of inpatient pediatric mortalities. Our process is feasible and adds value to clinicians’ clinical practice, and thus may be useful for other pediatric institutions. Future work includes further refinement of this mortality review form, enhancement of our processes to review postdischarge mortalities, and consideration of how to align existing M&M conferences with hospital safety event review processes.

We thank the clinicians who have completed mortality reviews, as well as all clinicians who have provided feedback on the mortality review and notification system.

M&M

morbidity and mortality

QI

quality improvement