Patterns Contributing to Severe Complications After Liver Resection: An Aggregate Root Cause Analysis of a Prospective Cohort

a are considered The reporting The Aggregate Root Cause Analysis was designed to improve the understanding of system vulnerabilities contributing to patient harm, including surgical complications. It remains poorly used due to methodological complexity and resource limitations. This study aimed to identify the main patterns contributing to severe complications after liver resection (SC) using an AggRCA. contributory factors were in in Disrupted perioperative process, Unplanned intraoperative change, Ineffective


Results
Among 105 consecutive liver resections, 15 cases (14.3%) including 5 deaths (4.8%) met the inclusion criteria. AggRCA resulted in the identi cation of 36 contributory factors. Eight contributory factors were reported in more than half of the cases and were compiled in three entangled patterns: (1) Disrupted perioperative process, (2) Unplanned intraoperative change, (3) Ineffective communication.

Conclusion
A pragmatic aggregated RCA process improved our understanding of system vulnerabilities based on the analysis of a limited number of events and a reasonable resource intensity. The identi cation of patterns contributing to SC lay the rationale of future contextualized safety interventions beyond the scope of liver resections.

Background
For the past two decades, the global diffusion of modern liver resection techniques and evidence-based perioperative care practices has contrasted with the persistence of signi cant safety outcome disparities, including the incidence of severe postoperative complications and death rates [1]. To re ne our understanding of the mechanisms contributing to patient harm, the adoption of strategies beyond the quantitative factors that revolve around the patient, the surgeon, and the surgical procedure is advocated [2].
Initially derived from high-hazard engineering industries, the Root Cause Analysis (RCA) is a quality improvement strategy that focuses on system vulnerabilities that contribute to the likelihood of errors, rather than individual errors themselves. For that, it has become a standard tool to review single-case reports of adverse events across all healthcare specialties [3]. Recently, the aggregation of data from single-case RCAs (Aggregate RCA) was proposed to enhance insight into system functioning [4] and to re ne the prioritization of interventions that would prevent the occurrence of similar events [5].
Despite its promising design to improve overall patient safety, the aggregate RCA remains rarely used to investigate surgical outcomes. In the speci c eld of hepatobiliary and pancreatic surgery, RCAs of postoperative deaths have revealed consistent patterns of contributory factors, including complication management delays, intraoperative technical incidents, and gaps in compliance with guidelines [6,7]. These ndings that were based on retrospective data aggregation from multiple centers should be challenged by the use of frameworks that extend and deepen the analysis of adverse outcomes to a wide scope of possible in uences, including human factors [8,9]. The ALARM framework, originally inspired by Reason's model of organizational accidents [10,11], was adapted to medicine to enable researchers to formalize such an approach [12].
The aim of this study was to use an aggregate RCA based on the ALARM framework to identify the main patterns of contributory factors associated with severe complications after liver resection in the setting of a North African anticancer center.
consists of 50 questions (Q) selected from a large set of examples proposed by the French High Authority for Health https://www.hassante.fr/upload/docs/application/pdf/2017-07/dir152/2017-alarm-commente.pdf. Each question investigates one of the contributory factors related to the six following ALARM categories: "Patient", "Tasks", "Individual staff", "Team", "Work environment", and "Management and Institutional context". This latter was obtained from the merge of two categories "Organizational and management factors" and "Institutional context factors", as it was suggested by Vincent et al ( [12,15]). Answers incriminating a contributing factor ("yes" or "no" depending on the context) are referred to as "triggered answer" or "triggered contributory factor", indifferently. A "refuted" option or a "non-applicable (NA)" option (when information is judged lacking) is offered otherwise. Justi cations and comments regarding triggered contributory factors, recovery factors, and corrective measures are included in the nal report of the MMR. The set of 50 questions of the MMR reporting tool and their ALARM categorization is presented in Appendix 1.

Interventions a. Design
In the current study, an aggregate RCA (AggRCA) based on the ALARM framework [12] was used as a method to identify the main patterns of contributory factors associated with severe complications after liver resection. A pattern was de ned as a regular sequence of factors contributing to the prede ned outcome (vs. single root cause [5]).
In order to limit data overwriting, we chose to analyze aggregated data from independent RCAs of single cases, rather than making a root cause analysis directly from a summary of the cohort.
All the cases of severe complications after elective liver resection that were consecutively performed at an academic surgical department between January 1st, 2018 and December 31st, 2019 were included. Severe complications were de ned as complications grade > IIIa according to the Clavien-Dindo classi cation within the rst 90 postoperative days (PODs) [16,17].
In order to overcome selection and disponibility biases associated with voluntary reporting of adverse events [18], cases were identi ed from a prospective electronic database including all liver resections performed at the department,

Research team and participant selection
The research team included the surgeon in charge of the liver surgery program at the NIO (BA), a surgical resident (LO), and a research fellow that acted as a third party (HK). Six clinicians (4 surgeons, 2 intensivists) and 2 nurses, were purposively selected among surgical and intensive care staff given their involvement in the management of liver resections and their experience with Mortality and Morbidity reviews (MMRs). Characteristics and roles of participants and research team members are detailed in Table 1. Step 2

Single-case RCAs
Step 3 Step 2 consolidation Step 4 Focus group Step 3 validation Step 5 The AggRCA was conducted through a ve-step process over the period from December 2019 to March 2020. The MMR reporting tool was used for data collection and aggregation.
Step 1: Event storyline. For each case, a storyline depicting the timeframe of the perioperative care was created. Data were collected from the electronic database and completed from the patients' respective hard copy les: case history, physical examination, results and/or copies of documented preoperative medical imaging, pre-anesthetic consultation reports, treatment plan decisions, procedure reports, monitoring, and complication diagnosis and management. Interviews with staff members were conducted in case of missing information to obtain the most comprehensive case reports.
Step 2. Single-case RCAs Anonymized storylines were emailed to the six participating clinicians individually for review. Each participant was asked to ll out the MMR reporting tool for every single case independently to determine potential contributory factors, recovery factors, and corrective measures. A deadline was set for two weeks after receipt of the storylines.
Step 3. Consolidation of single-case RCAs (workshop): For each question, the research team consolidated the respondents' answers into a single option: "Triggered", "Refuted" and "NA". Consolidation was based on the congruence of responses between at least three (half) of the respondents, unless a justi cation that was presented brought unique insights: individual staff perception (stress, fatigue, moral support) and/or speci c perspective of the context of events (direct involvement in the management in the ICU or the OR). Con icting justi cations were identi ed and deferred to the next step for resolution. Recovery factors and corrective measures were pooled and rephrased using verbs.
Step 4. Validation of single-case RCAs (focus group) All the participants were gathered one month after the individual analysis was completed. The consolidation process and results were presented. Con icting justi cations were discussed and settled on a case by case basis until a consensus was reached for all triggered contributory factors, recovery factors, and corrective measures.
Step 5. Aggregate RCA. Combinations of triggered contributory factors across the single cases were visualized according to the ALARM categories. Validated data from single cases were aggregated to obtain a distribution (percentage) of triggered contributory factors and their respective categories among the whole cohort.
A network of relationships (contribution to harm and/or failure to prevent harm) was established between the contributory factors that were triggered in more than half of the cases. The main patterns were suggested from the network analysis and re ned from the integration of less frequent contributory factors, as well as insights from recovery factors and corrective measures.

Statistical analysis
Aggregated data from single-case answers were summarized into descriptive statistics tables, including median, percentages, standard deviation, and quartiles when appropriate. Analyses were performed using Google sheet.

Results
Description of the study population 15 single cases of severe complications out of 105 consecutive elective liver resections met the inclusion criteria and were therefore analyzed according to the ve-step process. The severe complication rate within 90 days of index surgery was 14.3%, including a mortality rate of 4.8%.
Step 1 and 2. Single-case RCAs All six participating clinicians sent back fully lled out RMM reporting tools regarding all the 15 cases.
Step 3. Consolidation of single-case RCAs The workshop lasted ve hours. Consolidation (from 4500 to 750 answers) was based on the congruence of responses between at least three respondents in 662 (88.2%) cases and a unique insight in 76 (10.2%) cases. In 12 (1.6%) cases, consolidation was not resolved at this step because of con icting justi cations.
Step 4. Validation of single-case RCAs The focus group that gathered all the participants lasted three hours. All con icting justi cations were resolved. The consensus was reached for each triggered contributory factor, recovery factor, and corrective measure. For each case, a median of 10 contributory factors was triggered (extremes 4-20).
24 corrective measures within all six ALARM categories were suggested. Only two of them (8.3%) were speci cally related to liver resection techniques. Ten (42%) measures including all those related to the "Patient factors" consisted of recommendations to implement or reinforce protocols. Details of the characteristics of recovery factors and corrective measures are presented in Table 5.  Pattern 1: Disrupted perioperative process. Nonoptimal protocol availability and/or use was consistently reported (Q10, 93.3%), especially in the assessment of patients with preexisting health conditions (Q3, 53.3%), such as advanced age, obesity, altered nutritional status, and mental health issues. Delayed diagnosis and/or treatment of complications (Q42, 53.3%) was attributed to intra-team factors such as clinical hesitations (imaging indication, revision surgery indication) and systemic factors such as regional blood shortage, lab test dependency upon a distant hospital, and senior radiology staff unavailability. Intra-team contributory factors may have been in uenced by individual stress and/or fatigue (Q21 33.3%) and heavy clinical workloads (Q39, 26.6%) that were reported when many complex cases and complications were dealt with during the same period. This called for workforce management and surgical scheduling adaptation upon security standards.
Pattern 2: Unplanned intraoperative change. An unplanned extension of the resection to liver parenchyma or another organ was performed as a technical adaptation in order to achieve tumor-free resection margins (Q17, 60%). This was associated with postoperative deaths in more than half of the cases.
Although re ecting the complexity (Q4, 100%) of many cases (large and/or multiple in ltrative tumors with borderline resectability), the performance of unplanned procedures called for the implementation of a formal intraoperative decision-making process. In two cases (unplanned portal resection and In this prospective cohort, the 90-day postoperative mortality rate was < 5%, which is in line with the results of expert centers and western national registries [19,20]. A high prevalence of sepsis among severe complications (versus liver failure) and a short interval between index surgery and death compared to series from expert centers were reported [21]. This should be investigated, while acknowledging latent factors that are common to developing countries, such as disruption to intensive care services and blood shortage [22].
In this study, we chose to identify patterns of contributory factors rather than a single or a small number of root causes. This allows us to better describe the entanglement of active failures and latent conditions [23], and consider the dynamics of their interactions [24].
In the medical literature, most studies that tackle "Disruptions in perioperative processes" (Pattern 1) refer to deviations in the respect of guidelines and/or local protocols [6]. In a previous RCA study of 86 postoperative deaths after liver resection, guidelines and postoperative management protocols were not respected in 57% and 22% of the cases, respectively [21]. Although it is critical, the implementation of measures to ensure compliance with evidence-based practice may be challenged by systemic factors depending on the context (eg: blood and drug shortages, limited access to imaging, and lab tests). These factors may lead to individual and team compensation mechanisms and may exacerbate the stress and burnout associated with the management of complications and the second victim syndrome [25,26]. In the present study, individual staff proactivity and family support (regardless of socioeconomic status) were the most frequently reported recovery factors that may re ect compensation mechanisms to systemic failures [27].
Unplanned intraoperative change (Pattern 2) involves a cascade of events that favor the violation of intraoperative guidelines and the occurrence of technical errors. Tumor progression leading to a more extensive procedure than planned is a typical pattern of postoperative complication and death [21]. Stress, cognitive biases such as the sunk cost fallacy and the anchoring effect [18], and overcon dence in one's intuition [28] may explain why it is challenging for surgeons to process signi cant updates in the balance between safety and potential oncological bene ts. This underscores the importance of preoperative planning including up-to-date imaging, multidisciplinary assessment, and accurate evaluation of remnant liver volume when indicated. In the event of unforeseen intraoperative ndings, a break in the operative course [21] and discussion with colleagues (surgeons, intensivists, oncologists) have been suggested to prevent futile and potentially lethal surgeries [29][30][31][32].
Ineffective communication (Pattern 3) covers a spectrum of situations that extends from oral interpersonal communication to written traceability in the medical records. It may be maintained by a culture of blame and low empowerment to notify disagreements and institutional failures, [8,33]. The communication of RCA results to key stakeholders and other staff [3] is a lever for tackling systemic factors and promoting a safety culture. It may scale up harm mitigation and support the sustainability of effective corrective measures [34].
In the current study, actionable system vulnerabilities were revealed by a collaborative methodology that allowed us to draw a maximum of relevant information from a limited number of events.
Effective multidisciplinary staff participation was favored by capitalizing on a pre-existing MMR process including a structured reporting tool. As it was already suggested, the MMR, which is a regulatory obligation in many countries, may represent an alternative to overcome the limited methods and intensity of resources (time, human and nancial) to conduct a formal RCA [35,36]; [37]. The use of a common taxonomy (ALARM framework) for the contributory factors, the recovery factors, and the corrective measures supported a comprehensive approach to patterns identi cation and improvement strategies recommendations.

Limitations
This study has some limitations. First, the inclusion of random cases of liver resection that were not followed by a complication may have overcome hindsight bias and reveal more latent contributory factors [38]. Second, the aggregation of a limited number of cases across a two-year period may have overlooked other contributory factors and potential evolutions of patterns. This invites us to keep an open mind on system changes including the collective learning curve and consider the need to update our interpretations. Third, emotional bias due to the involvement of participants in the management of the cases could not be totally excluded, despite the participation of a third party (HK). However, we believe on the contrary that the inclusion of experts in the concrete functioning of the studied system associated with a methodology based on formal justi cation added value to our approach.
Finally, an improved analysis may have been limited by inaccuracies related to the selection of questions of the MMR reporting tool and their formulation. Inclusion of a more relevant guidelines/protocols subdivision (cancer-related, patient-related and procedure-related) and human behavior categorization (knowledge-based, rule-based, and skill-based) [39,40] should be undertaken.

Conclusion
In this study, a pragmatic aggregated RCA methodology resulted in the identi cation of patterns contributing to severe complications after liver resection, based on the study of a limited number of events and a reasonable resource intensity. It revealed system vulnerabilities and potential safety interventions that may be exploited beyond the scope of liver surgery.
Future studies from different settings and subspeciality backgrounds are needed to examine the applicability of current methodology for conducting, aggregating, and analyzing data from RCAs of postoperative complications.

Abbreviations
Declarations Ethics approval and consent to participate This study is approved by the Mohammed V university ethics review board CERB (Comitéd'éthique pour la recherche biomédicale) in Rabat.

Consent for publication
Not applicable Availability of data and materials Not applicable

Competing interests
The authors declare that they have no competing interests Processing of the answers to the MMR reporting tool across the aggregate RCA steps. Figure 2