Identifying the fundamental structures and processes of care contributing to emergency general surgery quality using a mixed-methods Donabedian approach

BACKGROUND

Acute Care Surgery (ACS) was developed as a structured, team-based approach to providing round-the-clock emergency general surgery (EGS) care for adult patients needing treatment for diseases such as cholecystitis, gastrointestinal perforation, and necrotizing fasciitis. Lacking any prior evidence on optimizing outcomes for EGS patients, current implementation of ACS models has been idiosyncratic. We sought to use a Donabedian approach to elucidate potential EGS structures and processes that might be associated with improved outcomes as an initial step in designing the optimal model of ACS care for EGS patients.

METHODS

We developed and implemented a national survey of hospital-level EGS structures and processes by surveying surgeons or chief medical officers regarding hospital-level structures and processes that directly or indirectly impacted EGS care delivery in 2015. These responses were then anonymously linked to 2015 data from the American Hospital Association (AHA) annual survey, Medicare Provider Analysis and Review claims (MedPAR), 17 State Inpatient Databases (SIDs) using AHA unique identifiers (AHAID). This allowed us to combine hospital-level data, as reported in our survey or to the AHA, to patient-level data in an effort to further examine the role of EGS structures and processes on EGS outcomes. We describe the multi-step, iterative process utilizing the Donabedian framework for quality measurement that serves as a foundation for later work in this project.

RESULTS

Hospitals that responded to the survey were primarily non-governmental and located in urban settings. A plurality of respondent hospitals had fewer than 100 inpatient beds. A minority of the hospitals had medical school affiliations.

DISCUSSION

Our results will enable us to develop a measure of preparedness for delivering EGS care in the US, provide guidance for regionalized care models for EGS care, tiering of ACS programs based on the robustness of their EGS structures and processes and the quality of their outcomes, and formulate triage guidelines based on patient risk factors and severity of EGS disease.

CONCLUSIONS

Our work provides a template for team science applicable to research efforts combining primary data collection (i.e., that derived from our survey) with existing national data sources (i.e., SIDs and MedPAR).

Hepatopancreatobiliary Surgery: the Role of Clinical Resources and Variation in Performance of Hospitals Located in "Distressed" Communities

INTRODUCTION

The USA has one of the largest known income-based health disparities, with low-income adults being up to five times more likely to report being in poor health. We evaluated the association of hospital zip-code-based distressed communities index (DCI) with post-surgical outcomes following hepatopancreatobiliary (HPB) surgery.

METHODS

Adults undergoing HPB surgery were identified in the National Inpatient Sample. The association between hospital socioeconomic distress and outcomes including complications, mortality, failure to rescue (FTR), and never events were compared between high-distress facilities (HDF) and low-distress facilities (LDF).

RESULTS

A total of 11,119 (37.8%) patients underwent an operation at an HDF. Patients treated at HDF were younger (18-39 years, HDF: n = 1261, 11.3% vs. LDF: n = 966, 9.0%; p < 0.001), Black/Hispanic (HDF: n = 2060, 18.5% vs. LDF: n = 1440, 11.4%; p < 0.001) and in the lowest income quartile (HDF: n = 2825, 25.4% vs. LDF: n = 1116, 10.8%; p < 0.001). While complications were comparable at HDF versus LDF (HDF: n = 2483, 22.3% vs. LDF: n = 2370, 22.0%; p = 0.28), patients treated at HDF had higher odds of in-hospital mortality (OR, 1.31; 95% CI, 1.07-1.59), FTR (OR, 1.24; 95% CI, 1.02-1.52), and a never event (OR, 1.76; 95% CI, 1.29-2.39; all p < 0.001). Hospitals having advanced internal medicine services had reduced odds of mortality (OR, 0.61; 95% CI, 0.47-0.80) whereas high nurse-to-patient ratio was associated with reduced odds of a complication (OR, 0.89; 95% CI, 0.81-0.98).

CONCLUSION

Approximately 40% of patients were admitted to HDF. These patients were more likely to be Black/Hispanic and underinsured. Perioperative outcomes were worse at HDF following HPB surgery.

The impact of individual surgeon on the likelihood of minimal invasive surgery among Medicare beneficiaries undergoing pancreatic resection

BACKGROUND

The objective of the current study was to evaluate the impact of the individual surgeon on the use of minimally invasive pancreatic resection.

METHODS

The Medicare 100% Standard Analytic Files were reviewed to identify Medicare beneficiaries who underwent pancreatic resection between 2013 and 2017. The impact of patient- and procedure-related factors on the likelihood of minimally invasive pancreatic resection was investigated.

RESULTS

A total of 12,652 (85.4%) patients underwent open pancreatic resection, whereas minimally invasive pancreatic resection was performed in 2,155 (14.6%) patients. Unadjusted rates of minimally invasive pancreatic resection ranged from 0% in the bottom volume tertile to 35.3% in the top tertile. Although patients with emergency admission were less likely to undergo minimally invasive pancreatic resection (odds ratio = 0.43, 95% confidence interval 0.32-0.58), patients operated on more recently had a higher chance of minimally invasive pancreatic resection (year 2017; odds ratio = 1.51, 95% confidence interval 1.28-1.79). On multivariable analysis, there was over a 3-fold variation in the odds that a patient underwent minimally invasive versus open pancreatic resection based on the individual surgeon (median odds ratio = 3.27, 95% confidence interval 2.98-3.56). Patients who underwent pancreatectomy by a low-volume, minimally invasive pancreatic resection surgeon had higher odds of 90-day mortality after surgery (odds ratio = 1.33, 95% confidence interval: 1.16-1.59), as well as higher observed/expected mortality compared with individuals treated by high-volume surgeons.

CONCLUSION

The likelihood of undergoing minimally invasive pancreatic resection among Medicare beneficiaries was markedly influenced by the individual treating surgeon rather than patient- or procedure-level factors.

Analyzing Outcomes Among Older Adults With Necrotizing Soft-Tissue Infections in the United States

BACKGROUND

Necrotizing soft-tissue infections (NSTIs) encompass a group of severe, life-threatening diseases with high morbidity and mortality. Evidence suggests advanced age is associated with worse outcomes. To date, no large data sets exist describing outcomes in older individuals, and risk factor identification is lacking.

METHODS

Retrospective data were obtained from the 2015 Medicare 100% sample. Included in the analysis were those aged ≥65 y with a primary diagnosis of an NSTI (gas gangrene, necrotizing fasciitis, cutaneous gangrene, or Fournier's gangrene). Risk factors for in-hospital mortality and discharge disposition were examined. Continuous variables were assessed using central tendency, t-tests, and Wilcoxon rank-sum tests. Categorical variables were assessed using the chi-squared and Fisher's exact tests. Statistical significance was defined as P < 0.05.

RESULTS

1427 patient records were reviewed. 59% of patients were male, and the overall mean age was 75.4±8.6 y. 1385 (97.0%) patients required emergency surgery for their NSTI diagnosis. The overall mortality was 5.3%. Several underlying comorbidities were associated with higher rates of mortality including cancer (OR: 3.50, P = 0.0009), liver disease (OR: 2.97, P = 0.03), and kidney disease (OR: 2.15, P = 0.01). While associated with high in-hospital mortality, these diagnoses were not associated with a difference in the rate of discharge to home compared with skilled nursing or rehab. Overall, patients discharged to skilled nursing facilities or rehab had higher rates of underlying comorbidities than patients who were discharged home (3 or more comorbid illness 84.3% versus 68.6%, P < 0.0001); however, no individual comorbid illness was associated with discharge location.

CONCLUSIONS

In our Medicare data set, we identified several medical comorbidities that are associated with increased rates of in-hospital mortality. Patients with underlying cancers had the highest odds of increased mortality. The effect on outcomes of the potentially immunosuppressive cancer treatments in these patients is unknown. These data suggest that patients with underlying illnesses, especially cancer, kidney disease, or liver disease have higher mortalities and are more likely to be discharged to skilled nursing facilities or rehab. It is unclear why these illnesses were associated with these worse outcomes while others including diabetes and heart disease were not. These data suggest that these particular comorbid illnesses may have special prognostic implications, although further analysis is necessary to identify the causative factors.

Characterizing Pastoral Care Utilization by Cancer Patients

PURPOSE

To assess the rate of and characterize the utilization of pastoral care (PC) among patients on their cancer treatment trajectory.

METHODS

Patients included in the present study were diagnosed with cancer 01/2015-08/2019 at The Ohio State Wexner Medical Center-The James. To determine which patient demographic and clinical factors were independently associated with PC utilization, a multivariable logistic regression was performed.

RESULTS

A total of 14,322 patients were included in the study and 5,166 (36.1%) had at least one visit with PC. Cancers such as brain (n = 232, 4.5% vs. n = 159, 1.7%), liver/pancreas (n = 733, 14.2% vs. 686, 7.5%), and lung (n = 1,288 vs. 24.9% vs. n = 1,113, 12.2%) were more commonly noted among patients who utilized PC services (all p < 0.001). Furthermore, compared with patients diagnosed with Stage 1 cancer, patients with more advanced disease stages had higher odds of utilizing PC services (Stage III: OR 2.37, 95% CI 2.07-2.70; Stage IV OR 2.31, 95% CI: 2.04-2.61; both p < 0.05). Interestingly, patients who had a DNR order had a markedly higher odds (OR 4.18, 95%CI 3.76-4.65, p < 0.001) of utilizing PC services.

DISCUSSION

One in three patients with cancer utilized PC services. Patients with more severe prognoses and individuals with a DNR order were more likely to utilize PC. The data suggest that PC services are an important resource for many patients and should be integrated into the treatment approach for cancer.

A Novel Machine-Learning Approach to Predict Recurrence After Resection of Colorectal Liver Metastases

BACKGROUND

Surgical resection of hepatic metastases remains the only potentially curative treatment option for patients with colorectal liver metastases (CRLM). Widely adopted prognostic tools may oversimplify the impact of model parameters relative to long-term outcomes.

METHODS

Patients with CRLM who underwent a hepatectomy between 2001 and 2018 were identified in an international, multi-institutional database. Bootstrap resampling methodology used in tandem with multivariable mixed-effects logistic regression analysis was applied to construct a prediction model that was validated and compared with scores proposed by Fong and Vauthey.

RESULTS

Among 1406 patients who underwent hepatic resection of CRLM, 842 (59.9%) had recurrence. The full model (based on age, sex, primary tumor location, T stage, receipt of chemotherapy before hepatectomy, lymph node metastases, number of metastatic lesions in the liver, size of the largest hepatic metastases, carcinoembryonic antigen [CEA] level and KRAS status) had good discriminative ability to predict 1-year (area under the receiver operating curve [AUC], 0.693; 95% confidence interval [CI], 0.684-0.704), 3-year (AUC, 0.669; 95% CI, 0.661-0.677), and 5-year (AUC, 0.669; 95% CI, 0.661-0.679) risk of recurrence. Studies analyzing validation cohorts demonstrated similar model performance, with excellent model accuracy. In contrast, the AUCs for the Fong and Vauthey scores to predict 1-year recurrence were only 0.527 (95% CI, 0.514-0.538) and 0.525 (95% CI, 0.514-0.533), respectively. Similar trends were noted for 3- and 5-year recurrence.

CONCLUSION

The proposed clinical score, derived via machine learning, which included clinical characteristics and morphologic data, as well as information on KRAS status, accurately predicted recurrence after CRLM resection with good discrimination and prognostic ability.

Development and validation of a real-time mortality risk calculator before, during and after hepatectomy: an analysis of the ACS NSQIP database

BACKGROUND

Although most conventional risk prediction models have been based on preoperative information, intra- and post-operative events may be more relevant to mortality after surgery. We sought to develop a mortality risk calculator based on real time characteristics associated with hepatectomy.

METHODS

Patients who underwent hepatectomy between 2014 and 2017 were identified in the ACS-NSQIP dataset. Three prediction models (pre-, intra-, post-operative) were developed and validated using perioperative data.

RESULTS

Among 14,720 patients, 197 (1.3%) experienced 30-day mortality. The predictive ability of the real-time mortality risk calculator was very good based on only preoperative factors (AUC; training cohort: 0.813, validation cohort: 0.731). Incorporating intra-operative variables into the model increased the AUC (training: 0.838, validation: 0.777), while the post-operative model achieved an AUC of 0.922 in the training and 0.885 in the validation cohorts, respectively. While patients with low preoperative risk had only very small fluctuations in the estimated 30-day mortality risk during the intraoperative (Δ0.4%) and postoperative (Δ0.6%) phases, patients who were already deemed high risk preoperatively had additional increased mortality risk based on factors that occurred in the intraoperative (Δ5.4%) and postoperative (Δ9.3%) periods.

CONCLUSION

A real-time mortality risk calculator may better help clinicians identify patients at risk of death at the different stages of the surgical episode.

Use of Machine Learning for Prediction of Patient Risk of Postoperative Complications After Liver, Pancreatic, and Colorectal Surgery

BACKGROUND

Surgical resection is the only potentially curative treatment for patients with colorectal, liver, and pancreatic cancers. Although these procedures are performed with low mortality, rates of complications remain relatively high following hepatopancreatic and colorectal surgery.

METHODS

The American College of Surgeons (ACS) National Surgical Quality Improvement Program was utilized to identify patients undergoing liver, pancreatic and colorectal surgery from 2014 to 2016. Decision tree models were utilized to predict the occurrence of any complication, as well as specific complications. To assess the variability of the performance of the classification trees, bootstrapping was performed on 50% of the sample.

RESULTS

Algorithms were derived from a total of 15,657 patients who met inclusion criteria. The algorithm had a good predictive ability for the occurrence of any complication, with a C-statistic of 0.74, outperforming the ASA (C-statistic 0.58) and ACS-Surgical Risk Calculator (C-statistic 0.71). The algorithm was able to predict with high accuracy thirteen out of the seventeen complications analyzed. The best performance was in the prediction of stroke (C-statistic 0.98), followed by wound dehiscence, cardiac arrest, and progressive renal failure (all C-statistic 0.96). The algorithm had a good predictive ability for superficial SSI (C-statistic 0.76), organ space SSI (C-statistic 0.76), sepsis (C-statistic 0.79), and bleeding requiring transfusion (C-statistic 0.79).

CONCLUSION

Machine learning was used to develop an algorithm that accurately predicted patient risk of developing complications following liver, pancreatic, or colorectal surgery. The algorithm had very good predictive ability to predict specific complications and demonstrated superiority over other established methods.

Association of Neighborhood Characteristics with Utilization of High-Volume Hospitals Among Patients Undergoing High-Risk Cancer Surgery

INTRODUCTION

As high-risk cancer surgery continues to become more centralized, it is important to understand the association of neighborhood characteristics relative to access to surgical care. We sought to determine the neighborhood level characteristics that may be associated with travel patterns and utilization of high-volume hospitals.

METHODS

The California Office of Statewide Health Planning database was used to identify patients who underwent pancreatectomy (PD), esophagectomy (ES), proctectomy (PR), or pneumonectomy (PN) for cancer between 2014 and 2016. Total minutes (m) traveled as well as whether a patient bypassed the nearest hospital that performed the operation to get to a higher-volume center was assessed. Data were merged with the Centers for Disease control social vulnerability index (SVI).

RESULTS

Overall, 26,937 individuals (ES: 4.7%; PN: 53.5% PD: 13.9% PR: 27.9%) underwent a complex oncologic operation. Median travel time was 16 m (interquartile range [IQR] 8.3-30.24) [ES: 21.8 m (IQR 10.6-46.9); PN: 14 m (IQR 7.8-27.0); PD: 21.2 m (IQR 10.6-42.6); PR: 15 m (IQR 8.1-28.4)]. Nearly three-quarter of patients (ES: 34%; PN: 73%; PD: 72%; LR: 81%) underwent an operation at a high-volume hospital. For all four operations, patients who resided in a county with a high overall SVI were less likely to have surgery at a high-volume hospital (ES: odds ratio [OR] 0.39, 95% confidence interval [CI] 0.24-0.65; PN: OR: 0.67, 95% CI 0.51-0.88; PD: OR 0.61, 95% CI 0.44-0.84; PR: OR 0.76, 95% CI 0.58-0.98).

CONCLUSIONS

Patients residing in communities of high social vulnerability were less likely to undergo high-risk cancer surgery at a high-volume hospital. The identification of society-based contextual disparities in access to complex surgical care should serve to inform targeted strategies to direct additional resources toward these vulnerable communities.

Prognostic factors differ according to KRAS mutational status: A classification and regression tree model to define prognostic groups after hepatectomy for colorectal liver metastasis

BACKGROUND

Although KRAS mutation status is known to affect the prognosis of patients with colorectal liver metastasis, the hierarchical association between other prognostic factors and KRAS status is not fully understood.

METHODS

Patients who underwent a hepatectomy for colorectal liver metastasis were identified in a multi-institutional international database. A classification and regression tree model was constructed to investigate the hierarchical association between prognostic factors and overall survival relative to KRAS status.

RESULTS

Among 1,123 patients, 29.9% (n = 336) had a KRAS mutation. Among wtKRAS patients, the classification and regression tree model identified presence of metastatic lymph nodes as the most important prognostic factor, whereas among mtKRAS patients, carcinoembryonic antigen level was identified as the most important prognostic factor. Among patients with wtKRAS, the highest 5-year overall survival (68.5%) was noted among patients with node negative primary colorectal cancer, solitary colorectal liver metastases, size <4.3 cm. In contrast, among patients with mtKRAS colorectal liver metastases, the highest 5-year overall survival (57.5%) was observed among patients with carcinoembryonic antigen <6 mg/mL. The classification and regression tree model had higher prognostic accuracy than the Fong score (wtKRAS [Akaike's Information Criterion]: classification and regression tree model 3334 vs Fong score 3341; mtKRAS [Akaike's Information Criterion]: classification and regression tree model 1356 vs Fong score 1396).

CONCLUSION

Machine learning methodology outperformed the traditional Fong clinical risk score and identified different factors, based on KRAS mutational status, as predictors of long-term prognosis.

Novel Machine Learning Approach to Identify Preoperative Risk Factors Associated With Super-Utilization of Medicare Expenditure Following Surgery

Importance

Typically defined as the top 5% of health care users, super-utilizers are responsible for an estimated 40% to 55% of all health care costs. Little is known about which factors may be associated with increased risk of long-term postoperative super-utilization.

Objective

To identify clusters of patients with distinct constellations of clinical and comorbid patterns who may be associated with an elevated risk of super-utilization in the year following elective surgery.

Design, Setting, and Participants

A retrospective longitudinal cohort study of 1 049 160 patients who underwent abdominal aortic aneurysm repair, coronary artery bypass graft, colectomy, total hip arthroplasty, total knee arthroplasty, or lung resection were identified from the 100% Medicare inpatient and outpatient Standard Analytic Files at all inpatient facilities performing 1 or more of the evaluated surgical procedures from 2013 to 2015. Data from 2012 to 2016 were used to evaluate expenditures in the year preceding and following surgery. Using a machine learning approach known as Logic Forest, comorbidities and interactions of comorbidities that put patients at an increased chance of becoming a super-utilizer were identified. All comorbidities, as defined by the Charlson (range, 0-24) and Elixhauser (range, 0-29) comorbidity indices, were used in the analysis. Higher scores indicated higher comorbidity burden. Data analysis was completed on November 16, 2018.

Main Outcome and Measures

Super-utilization of health care in the year following surgery.

Results

In total, 1 049 160 patients met inclusion criteria and were included in the analytic cohort. Their median (interquartile range) age was 73 (69-78) years, and approximately 40% were male. Super-utilizers comprised 4.8% of the overall cohort (n = 79 746) yet incurred 31.7% of the expenditures. Although the difference in overall expenditures per person between super-utilizers ($4049) and low users ($2148) was relatively modest prior to surgery, the difference in expenditures between super-utilizers ($79 698) vs low users ($2977) was marked in the year following surgery. Risk factors associated with super-utilization of health care included hemiplegia/paraplegia (odds ratio, 5.2; 95% CI, 4.4-6.2), weight loss (odds ratio, 3.5; 95% CI, 2.9-4.2), and congestive heart failure with chronic kidney disease stages I to IV (odds ratio, 3.4; 95% CI, 3.0-3.9).

Conclusions and Relevance

Super-utilizers comprised only a small fraction of the surgical population yet were responsible for a disproportionate amount of Medicare expenditure. Certain subpopulations were associated with super-utilization of health care following surgical intervention despite having lower overall use in the preoperative period.

The association between self-declared acute care surgery services and critical care resources: Results from a national survey

PURPOSE

We examined differences in critical care structures and processes between hospitals with Acute Care Surgery (ACS) versus general surgeon on call (GSOC) models for emergency general surgery (EGS) care.

METHODS

2811 EGS-capable hospitals were surveyed to examine structures and processes including critical care domains and ACS implementation. Differences between ACS and GSOC hospitals were compared using appropriate tests of association and logistic regression models.

RESULTS

272/1497 hospitals eligible for analysis (18.2%) reported they use an ACS model. EGS patients at ACS hospitals were more likely to be admitted to a combined trauma/surgical ICU or a dedicated surgical ICU. GSOC hospitals had lower adjusted odds of having 24-h ICU coverage, in-house intensivists or respiratory therapists, and 4/6 critical-care protocols.

CONCLUSIONS

Critical care delivery is a key component of EGS care. While harnessing of critical care structures and processes varies across hospitals that have implemented ACS, overall ACS models of care appear to have more robust critical care practices.

Impact of Surgeon Volume on Outcomes and Expenditure Among Medicare Beneficiaries Undergoing Liver Resection: the Effect of Minimally Invasive Surgery

BACKGROUND

Although the role of annual surgeon volume on perioperative outcomes after liver resection (LR) has been investigated, there is a paucity of data regarding the impact of surgeon volume on outcomes of minimally invasive LR (MILR) versus open LR (OLR).

METHODS

Patients undergoing LR between 2013 and 2015 were identified in the Medicare inpatient Standard Analytic Files. Patients were classified into three groups based on surgeons' annual caseload: low (≤ 2 cases), medium (3-5 cases), or high (≥ 6 cases). Short-term outcomes and expenditures of LR, stratified by surgeon volume and minimally invasive surgery (MIS), were examined.

RESULTS

Among 3403 surgeons performing LR on 7169 patients, approximately 90% of surgeons performed less than 5 liver resections per year for Medicare patients. Only 7.1% of patients underwent MILR (n = 506). After adjustment, the likelihood of experiencing a complication and death within 90 days decreased with increasing surgeon volume. Outcomes of open and MILR among low- or high-volume surgeon groups, including rates of complications, 30- and 90-day readmission and mortality were similar. However, the difference of average total episode payment between open and MIS was higher in the high-volume surgeon group (low volume: $2929 vs. medium volume: $2333 vs. high volume: $7055).

CONCLUSION

Annual surgeon volume was an important predictor of outcomes following LR. MILR had comparable results to open LR among both the low- and high-volume surgeons.

A Novel Classification of Intrahepatic Cholangiocarcinoma Phenotypes Using Machine Learning Techniques: An International Multi-Institutional Analysis

INTRODUCTION

Patients with intrahepatic cholangiocarcinoma (ICC) generally have a poor prognosis, yet there can be heterogeneity in the patterns of presentation and associated outcomes. We sought to identify clusters of ICC patients based on preoperative characteristics that may have distinct outcomes based on differing patterns of presentation.

METHODS

Patients undergoing curative-intent resection of ICC between 2000 and 2017 were identified using a multi-institutional database. A cluster analysis was performed based on preoperative variables to identify distinct patterns of presentation. A classification tree was built to prospectively assign patients into cluster assignments.

RESULTS

Among 826 patients with ICC, three distinct presentation patterns were noted. Specifically, Cluster 1 (common ICC, 58.9%) consisted of individuals who had a small-size ICC (median 4.6 cm) and median carbohydrate antigen (CA) 19-9 and neutrophil-to-lymphocyte ratio (NLR) levels of 40.3 UI/mL and 2.6, respectively; Cluster 2 (proliferative ICC, 34.9%) consisted of patients who had larger-size tumors (median 9.0 cm), higher CA19-9 levels (median 72.0 UI/mL), and similar NLR (median 2.7); Cluster 3 (inflammatory ICC, 6.2%) comprised of patients with a medium-size ICC (median 6.2 cm), the lowest range of CA19-9 (median 26.2 UI/mL), yet the highest NLR (median 13.5) (all p < 0.05). Median OS worsened incrementally among the three different clusters {Cluster 1 vs. 2 vs. 3; 60.4 months (95% confidence interval [CI] 43.0-77.8) vs. 27.2 months (95% CI 19.9-34.4) vs. 13.3 months (95% CI 7.2-19.3); p < 0.001}. The classification tree used to assign patients into different clusters had an excellent agreement with actual cluster assignment (κ = 0.93, 95% CI 0.90-0.96).

CONCLUSION

Machine learning analysis identified three distinct prognostic clusters based solely on preoperative characteristics among patients with ICC. Characterizing preoperative patient heterogeneity with machine learning tools can help physicians with preoperative selection and risk stratification of patients with ICC.