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

Association between type of index complication and outcomes after noncardiac surgery

BACKGROUND

Failure to rescue, or the death of a patient after a surgical complication, largely occurs in patients who develop a cascade of postoperative complications. However, it is unclear whether there are specific types of index complications that are more strongly associated with failure to rescue, additional secondary complications, or other types of postoperative outcomes. This is a national cohort study of veterans who underwent noncardiac surgery at Veterans Affairs hospitals using data from the Veterans Affairs Surgical Quality Improvement Program (January 1, 2016 to September 30, 2021). Index complications were grouped into categories (cardiovascular, venous thromboembolism, pulmonary, bleeding/transfusion, renal, central nervous system, wound, sepsis, Clostridium difficile colitis, graft, or minor [defined as complications having an associated mortality rate <1%]). The association between type of index complication and failure to rescue, secondary complications, reoperation, and postoperative length of stay was evaluated with multivariable, hierarchical regression, and risk of death assessed with shared frailty modeling.

RESULTS

Among 574,195 patients, 5.3% had at least 1 complication (of which 26.1% had secondary complications, and 8.2% had failure to rescue), and 4.5% had a reoperation. Secondary complication (5.0%-61.4%) and failure to rescue (0.8%-34.2%) rates varied by the type of index complication. Relative to index minor complications, index bleeding was most associated with secondary complication (subdistribution hazard ratio 1.4, 95% confidence interval [1.1-1.8]), index cardiac complications were most associated with failure to rescue (odds ratio 45.4 [34.5-59.7]), index graft complications were most associated with reoperation (odds ratio 96.0 [79.5-115.8]), and index pulmonary complications were associated with 2.6 times longer length of stay (incident rate ratio 2.6 [2.6-2.7]). Index cardiac and central nervous system complications were most strongly associated with risk of death (cardiac-hazard ratio 2.45, 95% confidence interval [2.14-2.81]; central nervous system-hazard ratio 1.84 [1.49-2.27]).

CONCLUSION

Different types of index complications are associated with different outcome profiles. This suggests surgical quality improvement efforts should be tailored not only to the type of index complication to be addressed but also to the desired outcome to improve.

Association of preoperative cholangitis with outcomes and expenditures among patients undergoing pancreaticoduodenectomy

BACKGROUND

This study aimed to characterize the association of preoperative acute cholangitis (PAC) with surgical outcomes and healthcare costs.

METHODS

Patients who underwent pancreaticoduodenectomy (PD) between 2013 and 2021 were identified using 100% Medicare Standard Analytic Files. PAC was defined as the occurrence of at least 1 episode of acute cholangitis within the year preceding surgery. Multivariable regression analyses were used to compare postoperative outcomes and costs relative to PAC.

RESULTS

Among 23,455 Medicare beneficiaries who underwent PD, 2,217 patients (9.5%) had at least 1 episode of PAC. Most patients (n = 14,729 [62.8%]) underwent PD for a malignant indication. On multivariable analyses, PAC was associated with elevated odds of surgical site infection (odds ratio [OR], 1.14; 95% CI, 1.01-1.29), sepsis (OR, 1.17; 95% CI, 1.01-1.37), extended length of stay (OR, 1.13; 95% CI, 1.01-1.26), and readmission within 90 days (OR, 1.14; 95% CI, 1.04-1.26). Patients with a history of PAC before PD had a reduced likelihood of achieving a postoperative textbook outcome (OR, 0.83; 95% CI, 0.75-0.92) along with 87.8% and 18.4% higher associated preoperative and postoperative healthcare costs, respectively (all P < .001). Overall costs increased substantially among patients with more than 1 PAC episode ($59,893 [95% CI, $57,827-$61,959] for no episode vs $77,922 [95% CI, $73,854-$81,990] for 1 episode vs $101,205 [95% CI, $94,871-$107,539] for multiple episodes).

CONCLUSION

Approximately 1 in 10 patients undergoing PD experienced an antecedent PAC episode, which was associated with adverse surgical outcomes and greater healthcare expenditures.

Length of Stay Prediction Models for Oral Cancer Surgery: Machine Learning, Statistical and ACS-NSQIP

OBJECTIVE

Accurate prediction of hospital length of stay (LOS) following surgical management of oral cavity cancer (OCC) may be associated with improved patient counseling, hospital resource utilization and cost. The objective of this study was to compare the performance of statistical models, a machine learning (ML) model, and The American College of Surgeons National Surgical Quality Improvement Program's (ACS-NSQIP) calculator in predicting LOS following surgery for OCC.

MATERIALS AND METHODS

A retrospective multicenter database study was performed at two major academic head and neck cancer centers. Patients with OCC who underwent major free flap reconstructive surgery between January 2008 and June 2019 surgery were selected. Data were pooled and split into training and validation datasets. Statistical and ML models were developed, and performance was evaluated by comparing predicted and actual LOS using correlation coefficient values and percent accuracy.

RESULTS

Totally 837 patients were selected with mean patient age being 62.5 ± 11.7 [SD] years and 67% being male. The ML model demonstrated the best accuracy (validation correlation 0.48, 4-day accuracy 70%), compared with the statistical models: multivariate analysis (0.45, 67%) and least absolute shrinkage and selection operator (0.42, 70%). All were superior to the ACS-NSQIP calculator's performance (0.23, 59%).

CONCLUSION

We developed statistical and ML models that predicted LOS following major free flap reconstructive surgery for OCC. Our models demonstrated superior predictive performance to the ACS-NSQIP calculator. The ML model identified several novel predictors of LOS. These models must be validated in other institutions before being used in clinical practice.

LEVEL OF EVIDENCE

Level 3 Laryngoscope, 2024.

Development and Application of a Novel Machine Learning Model Predicting Pancreatic Cancer-Specific Mortality

Precise prognostication is vital for guiding treatment decisions in people diagnosed with pancreatic cancer. Existing models depend on predetermined variables, constraining their effectiveness. Our objective was to explore a novel machine learning approach to enhance a prognostic model for predicting pancreatic cancer-specific mortality and, subsequently, to assess its performance against Cox regression models. Datasets were retrospectively collected and analyzed for 9,752 patients diagnosed with pancreatic cancer and with surgery performed. The primary outcomes were the mortality of patients with pancreatic carcinoma at one year, three years, and five years. Model discrimination was assessed using the concordance index (C-index), and calibration was assessed using Brier scores. The Survival Quilts model was compared with Cox regression models in clinical use, and decision curve analysis was done. The Survival Quilts model demonstrated robust discrimination for one-year (C-index 0.729), three-year (C-index 0.693), and five-year (C-index 0.672) pancreatic cancer-specific mortality. In comparison to Cox models, the Survival Quilts models exhibited a higher C-index up to 32 months but displayed inferior performance after 33 months. A subgroup analysis was conducted, revealing that within the subset of individuals without metastasis, the Survival Quilts models showcased a significant advantage over the Cox models. In the cohort with metastatic pancreatic cancer, Survival Quilts outperformed the Cox model before 24 months but exhibited a weaker performance after 25 months. This study has developed and validated a novel machine learning-based Survival Quilts model to predict pancreatic cancer-specific mortality that outperforms the Cox regression model.

Comparing super-utilizers and lower-utilizers among commercial- and Medicare-insured adults with sickle cell disease

ABSTRACT

Sickle cell disease (SCD) is a rare but costly condition in the United States. Super-utilizers have been defined as a subset of the population with high health care encounters or expenditures. Although super-utilizers have been described in other disease states, little is known about super-utilizers among adults with SCD. This study aimed to characterize the differences in expenditures, overall health care encounters, and pain episode encounters between super-utilizers (top 10% expenditures) and lower-utilizers with SCD (high, top 10%-24.9%; moderate, 25%-49.9%; and low, bottom 50% expenditures). A retrospective longitudinal cohort of adults with SCD were identified using validated algorithms in MarketScan and Medicare claim databases from 2016 to 2020. Encounters and expenditures were analyzed from inpatient, outpatient, and emergency department settings. Differences in encounters and expenditures between lower-utilizers and super-utilizers were compared using logistic regression. Among super-utilizers, differences in encounters and expenditures were compared according to incidences of pain episode encounters. The study population included 5666 patients with commercial insurance and 8600 with Medicare. Adjusted total annual health care expenditure was 43.46 times higher for super-utilizers than for low-utilizers among commercial-insured and 13.37 times higher in Medicare-insured patients. Among super-utilizers, there were patients with few pain episode encounters who had higher outpatient expenditures than patients with a high number of pain episode encounters. Our findings demonstrate the contribution of expensive outpatient care among SCD super-utilizers, in which analyses of high expenditure have largely focused on short-term care. Future studies are needed to better understand super-utilizers in the SCD population to inform the effective use of preventive interventions and/or curative therapies.

Revolutionizing hysteroscopy outcomes: AI-powered uterine myoma diagnosis algorithm shortens operation time and reduces blood loss

Background

The application of artificial intelligence (AI) powered algorithm in clinical decision-making is globally popular among clinicians and medical scientists. In this research endeavor, we harnessed the capabilities of AI to enhance the precision of hysteroscopic myomectomy procedures.

Methods

Our multidisciplinary team developed a comprehensive suite of algorithms, rooted in deep learning technology, addressing myomas segmentation tasks. We assembled a cohort comprising 56 patients diagnosed with submucosal myomas, each of whom underwent magnetic resonance imaging (MRI) examinations. Subsequently, half of the participants were randomly designated to undergo AI-augmented procedures. Our AI system exhibited remarkable proficiency in elucidating the precise spatial localization of submucosal myomas.

Results

The results of our study showcased a statistically significant reduction in both operative duration (41.32 ± 17.83 minutes vs. 32.11 ± 11.86 minutes, p=0.03) and intraoperative blood loss (10.00 (6.25-15.00) ml vs. 10.00 (5.00-15.00) ml, p=0.04) in procedures assisted by AI.

Conclusion

This work stands as a pioneering achievement, marking the inaugural deployment of an AI-powered diagnostic model in the domain of hysteroscopic surgery. Consequently, our findings substantiate the potential of AI-driven interventions within the field of gynecological surgery.

The application of artificial intelligence in health financing: a scoping review

INTRODUCTION

Artificial Intelligence (AI) represents a significant advancement in technology, and it is crucial for policymakers to incorporate AI thinking into policies and to fully explore, analyze and utilize massive data and conduct AI-related policies. AI has the potential to optimize healthcare financing systems. This study provides an overview of the AI application domains in healthcare financing.

METHOD

We conducted a scoping review in six steps: formulating research questions, identifying relevant studies by conducting a comprehensive literature search using appropriate keywords, screening titles and abstracts for relevance, reviewing full texts of relevant articles, charting extracted data, and compiling and summarizing findings. Specifically, the research question sought to identify the applications of artificial intelligence in health financing supported by the published literature and explore potential future applications. PubMed, Scopus, and Web of Science databases were searched between 2000 and 2023.

RESULTS

We discovered that AI has a significant impact on various aspects of health financing, such as governance, revenue raising, pooling, and strategic purchasing. We provide evidence-based recommendations for establishing and improving the health financing system based on AI.

CONCLUSIONS

To ensure that vulnerable groups face minimum challenges and benefit from improved health financing, we urge national and international institutions worldwide to use and adopt AI tools and applications.

Association of Minority-Serving Hospital Status with Post-Discharge Care Utilization and Expenditures in Gastrointestinal Cancer

BACKGROUND

Disparities in utilization of post-discharge care and overall expenditures may relate to site of care and race/ethnicity. We sought to define the impact of minority-serving hospitals (MSHs) on postoperative outcomes, discharge disposition, and overall expenditures associated with an episode of surgical care.

METHODS

Patients who underwent resection for esophageal, colon, rectal, pancreatic, and liver cancer were identified from Medicare Standard Analytic Files (2013-2017). A MSH was defined as the top decile of facilities treating minority patients (Black and/or Hispanic). The impact of MSH on outcomes of interest was analyzed using multivariable logistic regression and generalized linear regression models. Textbook outcome (TO) was defined as no postoperative complications, no prolonged length of stay, and no 90-day mortality or readmission.

RESULTS

Among 113,263 patients, only a small subset of patients underwent surgery at MSHs (n = 4404, 3.9%). While 52.3% of patients achieved TO, rates were lower at MSHs (MSH: 47.2% vs. non-MSH: 52.5%; p < 0.001). On multivariable analysis, receiving care at an MSH was associated with not achieving TO (odds ratio [OR] 0.81, 95% confidence interval [CI] 0.76-0.87) and concomitantly higher odds of additional post-discharge care (OR 1.10, 95% CI 1.01-1.20). Patients treated at an MSH also had higher median post-discharge expenditures (MSH: $8400, interquartile range [IQR] $2300-$22,100 vs. non-MSH: $7000, IQR $2200-$17,900; p = 0.002). In fact, MSHs remained associated with a 11.05% (9.78-12.33%) increase in index expenditures and a 16.68% (11.44-22.17%) increase in post-discharge expenditures.

CONCLUSIONS

Patients undergoing surgery at a MSH were less likely to achieve a TO. Additionally, MSH status was associated with a higher likelihood of requiring post-discharge care and higher expenditures.

The Viability of an Artificial Intelligence/Machine Learning Prediction Model to Determine Candidates for Knee Arthroplasty

BACKGROUND

The purpose of this study is to assess the viability of a knee arthroplasty prediction model using 3-view X-rays that helps determine if patients with knee pain are candidates for total knee arthroplasty (TKA), unicompartmental knee arthroplasty (UKA), or are not arthroplasty candidates.

METHODS

Analysis was performed using radiographic and surgical data from a high-volume joint replacement practice. The dataset included 3 different X-ray views (anterior-posterior, lateral, and sunrise) for 2,767 patients along with information of whether that patient underwent an arthroplasty surgery (UKA or TKA) or not. This resulted in a dataset including 8,301 images from 2,707 patients. This dataset was then split into a training set (70%) and holdout test set (30%). A computer vision model was trained using a transfer learning approach. The performance of the computer vision model was evaluated on the holdout test set. Accuracy and multiclass receiver operating characteristic area under curve was used to evaluate the performance of the model.

RESULTS

The artificial intelligence model achieved an accuracy of 87.8% on the holdout test set and a quadratic Cohen's kappa score of 0.811. The multiclass receiver operating characteristic area under curve score for TKA was calculated to be 0.97; for UKA a score of 0.96 and for No Surgery a score of 0.98 was achieved. An accuracy of 93.8% was achieved for predicting Surgery versus No Surgery and 88% for TKA versus not TKA was achieved.

CONCLUSION

The artificial intelligence/machine learning model demonstrated viability for predicting which patients are candidates for a UKA, TKA, or no surgical intervention.

Global research landscape on artificial intelligence in arthroplasty: A bibliometric analysis

Background

Artificial intelligence (AI) has promising applications in arthroplasty. In response to the knowledge explosion resulting from the rapid growth of publications, we applied bibliometric analysis to explore the research profile and topical trends in this field.

Methods

The articles and reviews related to AI in arthroplasty were retrieved from 2000 to 2021. The Java-based Citespace, VOSviewer, R software-based Bibiometrix, and an online platform systematically evaluated publications by countries, institutions, authors, journals, references, and keywords.

Results

A total of 867 publications were included. Over the past 22 years, the number of AI-related publications in the field of arthroplasty has grown exponentially. The United States was the most productive and academically influential country. The Cleveland Clinic was the most prolific institution. Most publications were published in high academic impact journals. However, collaborative networks revealed a lack and imbalance of inter-regional, inter-institutional, and inter-author cooperation. Two emerging research areas represented the development trends: major AI subfields such as machine learning and deep learning, and the other is research related to clinical outcomes.

Conclusion

AI in arthroplasty is evolving rapidly. Collaboration between different regions and institutions should be strengthened to deepen our understanding further and exert critical implications for decision-making. Predicting clinical outcomes of arthroplasty using novel AI strategies may be a promising application in this field.

Preoperative Serum Markers and Risk Classification in Intrahepatic Cholangiocarcinoma: A Multicenter Retrospective Study

Accurate risk stratification selects patients who are expected to benefit most from surgery. This retrospective study enrolled 225 Japanese patients with intrahepatic cholangiocellular carcinoma (ICC) who underwent hepatectomy between January 2009 and December 2020 and identified preoperative blood test biomarkers to formulate a classification system that predicted prognosis. The optimal cut-off values of blood test parameters were determined by ROC curve analysis, with Cox univariate and multivariate analyses identifying prognostic factors. Risk classifications were established using classification and regression tree (CART) analysis. CART analysis revealed decision trees for recurrence-free survival (RFS) and overall survival (OS) and created three risk classifications based on machine learning of preoperative serum markers. Five-year rates differed significantly (p < 0.001) between groups: 60.4% (low-risk), 22.8% (moderate-risk), and 4.1% (high-risk) for RFS and 69.2% (low-risk), 32.3% (moderate-risk), and 9.2% (high-risk) for OS. No difference in OS was observed between patients in the low-risk group with or without postoperative adjuvant chemotherapy, although OS improved in the moderate group and was prolonged significantly in the high-risk group receiving chemotherapy. Stratification of patients with ICC who underwent hepatectomy into three risk groups for RFS and OS identified preoperative prognostic factors that predicted prognosis and were easy to understand and apply clinically.

A novel machine learning approach to identify social risk factors associated with textbook outcomes after surgery

BACKGROUND

Identifying social determinants of health has become a priority for many researchers, health care providers, and payers. The vast amount of patient and population-level data available on social determinants creates, however, both an opportunity and a challenge as these data can be difficult to synthesize and analyze.

METHODS

Medicare beneficiaries who underwent 1 of 4 common operations between 2013 and 2017 were identified. Using a machine learning algorithm, the primary independent variable, surgery social determinants of health index, was derived from 15 common, publicly available social determents of health measures. After development of a surgery social determinants of health index, multivariable logistic regression was used to estimate the association of this index with textbook outcomes, as well as the component metrics of textbook outcomes.

RESULTS

A novel surgery social determinants of health index was developed with factor component weights that varied relative to their impact on postoperative outcomes. Factors with the highest weight in the algorithm relative to postoperative outcomes were the proportion of noninstitutionalized civilians with a disability and persons without high school diploma, while components with the lowest weights were the proportion of households with more people than rooms and persons below poverty. Overall, an increase in surgery social determinants of health index was associated with 6% decreased odds (95% confidence interval: 0.93-0.94) of achieving a textbook outcome. In addition, an increase in surgery social determinants of health index was associated with increased odds of each of the individual components of textbook outcome; ranging from 3% increased odds (95% confidence interval: 1.03-1.04) for 90-day readmission to 10% increased odds (95% confidence interval: 1.09-1.11) for 90-day mortality. Further, there was 6% increased odds (95% confidence interval: 1.05-1.07) of experiencing a complication and 7% increased odds (95% confidence interval: 1.06-1.07) of having an extended length of stay. Minority patients from a high surgery social determinants of health index had 38% lower odds (95% confidence interval: 0.60-0.65) of achieving a textbook outcome compared with White/non-Hispanic patients from a low surgery social determinants of health index area.

CONCLUSION

Using a machine learning approach, we developed a novel social determents of health index to predict the probability of achieving a textbook outcome after surgery.

Efficacy and Applications of Artificial Intelligence and Machine Learning Analyses in Total Joint Arthroplasty: A Call for Improved Reporting

BACKGROUND

There has been a considerable increase in total joint arthroplasty (TJA) research using machine learning (ML). Therefore, the purposes of this study were to synthesize the applications and efficacies of ML reported in the TJA literature, and to assess the methodological quality of these studies.

METHODS

PubMed, OVID/MEDLINE, and Cochrane libraries were queried in January 2021 for articles regarding the use of ML in TJA. Study demographics, topic, primary and secondary outcomes, ML model development and testing, and model presentation and validation were recorded. The TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) guidelines were used to assess the methodological quality.

RESULTS

Fifty-five studies were identified: 31 investigated clinical outcomes and resource utilization; 11, activity and motion surveillance; 10, imaging detection; and 3, natural language processing. For studies reporting the area under the receiver operating characteristic curve (AUC), the median AUC (and range) was 0.80 (0.60 to 0.97) among 26 clinical outcome studies, 0.99 (0.83 to 1.00) among 6 imaging-based studies, and 0.88 (0.76 to 0.98) among 3 activity and motion surveillance studies. Twelve studies compared ML to logistic regression, with 9 (75%) reporting that ML was superior. The average number of TRIPOD guidelines met was 11.5 (range: 5 to 18), with 38 (69%) meeting greater than half of the criteria. Presentation and explanation of the full model for individual predictions and assessments of model calibration were poorly reported (<30%).

CONCLUSIONS

The performance of ML models was good to excellent when applied to a wide variety of clinically relevant outcomes in TJA. However, reporting of certain key methodological and model presentation criteria was inadequate. Despite the recent surge in TJA literature utilizing ML, the lack of consistent adherence to reporting guidelines needs to be addressed to bridge the gap between model development and clinical implementation.

Applications of artificial intelligence and machine learning for the hip and knee surgeon: current state and implications for the future

BACKGROUND

Artificial Intelligence (AI)/Machine Learning (ML) applications have been proven efficient to improve diagnosis, to stratify risk, and to predict outcomes in many respective medical specialties, including in orthopaedics.

CHALLENGES AND DISCUSSION

Regarding hip and knee reconstruction surgery, AI/ML have not made it yet to clinical practice. In this review, we present sound AI/ML applications in the field of hip and knee degenerative disease and reconstruction. From osteoarthritis (OA) diagnosis and prediction of its advancement, clinical decision-making, identification of hip and knee implants to prediction of clinical outcome and complications following a reconstruction procedure of these joints, we report how AI/ML systems could facilitate data-driven personalized care for our patients.

Fragmentation of practice: The adverse effect of surgeons moving around

BACKGROUND

Whether surgical team familiarity is associated with improved postoperative outcomes remains unknown. We sought to characterize the impact of fragmented surgical practice on the likelihood that a patient would experience a textbook outcome, which is a validated patient-centric composite outcome representing an "ideal" postoperative outcome.

METHOD

Medicare beneficiaries aged 65 and older who underwent elective inpatient abdominal aortic aneurysm repair, coronary artery bypass graft, cholecystectomy, colectomy, or lung resection were identified. Rate of fragmented practice was calculated based on the total number of surgical procedures of interest performed over the study period (2013-2017) divided by the number of different hospitals in which the surgeon operated. Surgeons were categorized into "low," "average," "above average," or "high" rate of fragmented practice categories using an unsupervised machine learning technique known k-medians cluster analysis.

RESULTS

Among 546,422 Medicare beneficiaries who underwent an elective surgical procedure of interest (coronary artery bypass graft: n = 156,384, 28.6%; lung resection: n = 83,164, 15.2%; abdominal aortic aneurysm: n = 112,578, 20.6%; cholecystectomy: n = 42,955, 7.9%; colectomy: n = 151,341, 27.7%), median patient age was 74 years (interquartile range: 69-80), and most patients were male (n = 319,153, 58.4%). Machine learning identified 3 cutoffs to categorize rate of fragmented practice: 2.8%, 5.6%, and 10.6%. Overall, the majority of surgical procedures were performed by surgeons with a low rate of fragmented practice (n = 382,504, 70.0%); other surgical procedures were performed by surgeons with average (n = 109,141, 20.0%), above average (n = 44,249, 8.1%), or high (n = 10,528, 1.9%) rate of fragmented practice. On multivariable analyses, after controlling for patient demographics, individual surgeon volume, procedure type, and a random effect for hospital, patients who underwent a surgical procedure by a high versus low rate of fragmented practice surgeon had lower odds to achieve a postoperative textbook outcome (odds ratio 0.71, 95% confidence interval 0.77-0.84). Patients who underwent a procedure by a high rate of fragmented practice surgeon also had increased odds of a perioperative complication (odds ratio 1.30, 95% confidence interval: 1.23-1.37), extended length of stay (odds ratio 1.17, 95% confidence interval: 1.11-1.24), 90-day readmission (odds ratio 1.17, 95% confidence interval: 1.11-1.23), and 90-day mortality (odds ratio 1.29, 95% confidence interval: 1.17-1.42) (all P < .05).

CONCLUSION

Patients undergoing a surgical procedure by a surgeon with a high rate of fragmented practice had lower odds of achieving an optimal postoperative textbook outcome. Surgical team familiarity, measured by a surgeon rate of fragmented practice, may represent a modifiable mechanism to improve surgical outcomes.

Machine learning in knee arthroplasty: specific data are key-a systematic review

PURPOSE

Artificial intelligence (AI) in healthcare is rapidly growing and offers novel options of data analysis. Machine learning (ML) represents a distinct application of AI, which is capable of generating predictions and has already been tested in different medical specialties with various approaches such as diagnostic applications, cost predictions or identification of risk factors. In orthopaedics, this technology has only recently been introduced and the literature on ML in knee arthroplasty is scarce. In this review, we aim to investigate which predictions are already feasible using ML models in knee arthroplasty to identify prerequisites for the effective use of this novel approach. For this reason, we conducted a systematic review of ML algorithms for outcome prediction in knee arthroplasty.

METHODS

A comprehensive search of PubMed, Medline database and the Cochrane Library was conducted to find ML applications for knee arthroplasty. All relevant articles were systematically retrieved and evaluated by an orthopaedic surgeon and a data scientist on the basis of the PRISMA statement. The search strategy yielded 225 articles of which 19 were finally assessed as eligible. A modified Coleman Methodology Score (mCMS) was applied to account for a methodological evaluation.

RESULTS

The studies presented in this review demonstrated fair to good results (AUC median 0.76/range 0.57-0.98), while heterogeneous prediction models were analysed: complications (6), costs (4), functional outcome (3), revision (2), postoperative satisfaction (2), surgical technique (1) and biomechanical properties (1) were investigated. The median mCMS was 65 (range 40-80) points.

CONCLUSION

The prediction of distinct outcomes with ML models applying specific data is already feasible; however, the prediction of more complex outcomes is still inaccurate. Registry data on knee arthroplasty have not been fully analysed yet so that specific parameters have not been sufficiently evaluated. The inclusion of specific input data as well as the collaboration of orthopaedic surgeons and data scientists are essential prerequisites to fully utilize the capacity of ML in knee arthroplasty. Future studies should investigate prospective data with specific and longitudinally recorded parameters.

LEVEL OF EVIDENCE

III.

Potential benefits, unintended consequences, and future roles of artificial intelligence in orthopaedic surgery research : a call to emphasize data quality and indications

Artificial intelligence and machine-learning analytics have gained extensive popularity in recent years due to their clinically relevant applications. A wide range of proof-of-concept studies have demonstrated the ability of these analyses to personalize risk prediction, detect implant specifics from imaging, and monitor and assess patient movement and recovery. Though these applications are exciting and could potentially influence practice, it is imperative to understand when these analyses are indicated and where the data are derived from, prior to investing resources and confidence into the results and conclusions. In this article, we review the current benefits and potential limitations of machine-learning for the orthopaedic surgeon with a specific emphasis on data quality.

Artificial intelligence's role in vascular surgery decision-making

Artificial intelligence (AI) is the next great advance informing medical science. Several disciplines, including vascular surgery, use AI-based decision-making tools to improve clinical performance. Although applied widely, AI functions best when confronted with voluminous, accurate data. Consistent, predictable analytic technique selection also challenges researchers. This article contextualizes AI analyses within evidence-based medicine, focusing on "big data" and health services research, as well as discussing opportunities to improve data collection and realize AI's promise.

Artificial intelligence in orthopaedics: A scoping review

There is a growing interest in the application of artificial intelligence (AI) to orthopaedic surgery. This review aims to identify and characterise research in this field, in order to understand the extent, range and nature of this work, and act as springboard to stimulate future studies. A scoping review, a form of structured evidence synthesis, was conducted to summarise the use of AI in orthopaedics. A literature search (1946-2019) identified 222 studies eligible for inclusion. These studies were predominantly small and retrospective. There has been significant growth in the number of papers published in the last three years, mainly from the USA (37%). The majority of research used AI for image interpretation (45%) or as a clinical decision tool (25%). Spine (43%), knee (23%) and hip (14%) were the regions of the body most commonly studied. The application of artificial intelligence to orthopaedics is growing. However, the scope of its use so far remains limited, both in terms of its possible clinical applications, and the sub-specialty areas of the body which have been studied. A standardized method of reporting AI studies would allow direct assessment and comparison. Prospective studies are required to validate AI tools for clinical use.

Accessing Artificial Intelligence for Clinical Decision-Making

Advancements in computing and data from the near universal acceptance and implementation of electronic health records has been formative for the growth of personalized, automated, and immediate patient care models that were not previously possible. Artificial intelligence (AI) and its subfields of machine learning, reinforcement learning, and deep learning are well-suited to deal with such data. The authors in this paper review current applications of AI in clinical medicine and discuss the most likely future contributions that AI will provide to the healthcare industry. For instance, in response to the need to risk stratify patients, appropriately cultivated and curated data can assist decision-makers in stratifying preoperative patients into risk categories, as well as categorizing the severity of ailments and health for non-operative patients admitted to hospitals. Previous overt, traditional vital signs and laboratory values that are used to signal alarms for an acutely decompensating patient may be replaced by continuously monitoring and updating AI tools that can pick up early imperceptible patterns predicting subtle health deterioration. Furthermore, AI may help overcome challenges with multiple outcome optimization limitations or sequential decision-making protocols that limit individualized patient care. Despite these tremendously helpful advancements, the data sets that AI models train on and develop have the potential for misapplication and thereby create concerns for application bias. Subsequently, the mechanisms governing this disruptive innovation must be understood by clinical decision-makers to prevent unnecessary harm. This need will force physicians to change their educational infrastructure to facilitate understanding AI platforms, modeling, and limitations to best acclimate practice in the age of AI. By performing a thorough narrative review, this paper examines these specific AI applications, limitations, and requisites while reviewing a few examples of major data sets that are being cultivated and curated in the US.

Machine learning in gastrointestinal surgery

Machine learning (ML) is a collection of algorithms allowing computers to learn directly from data without predetermined equations. It is used widely to analyze "big data". In gastrointestinal surgery, surgeons deal with various data such as clinical parameters, surgical videos, and pathological images, to stratify surgical risk, perform safe surgery and predict patient prognosis. In the current "big data" era, the accelerating accumulation of a large amount of data drives studies using ML algorithms. Three subfields of ML are supervised learning, unsupervised learning, and reinforcement learning. In this review, we summarize applications of ML to surgical practice in the preoperative, intraoperative, and postoperative phases of care. Prediction and stratification using ML is promising; however, the current overarching concern is the availability of ML models. Information systems that can manage "big data" and integrate ML models into electronic health records are essential to incorporate ML into daily practice. ML is fundamental technology to meaningfully process data that exceeds the capacity of the human mind to comprehend. The accelerating accumulation of a large amount of data is changing the nature of surgical practice fundamentally. Artificial intelligence (AI), represented by ML, is being incorporated into daily surgical practice.

Artificial Learning and Machine Learning Decision Guidance Applications in Total Hip and Knee Arthroplasty: A Systematic Review

Background

Artificial intelligence (AI) and machine learning (ML) modeling in hip and knee arthroplasty (total joint arthroplasty [TJA]) is becoming more commonplace. This systematic review aims to quantify the accuracy of current AI- and ML-based application for cognitive support and decision-making in TJA.

Methods

A comprehensive search of publications was conducted through the EMBASE, Medline, and PubMed databases using relevant keywords to maximize the sensitivity of the search. No limits were placed on level of evidence or timing of the study. Findings were reported according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Analysis of variance testing with post-hoc Tukey test was applied to compare the area under the curve (AUC) of the models.

Results

After application of inclusion and exclusion criteria, 49 studies were included in this review. The application of AI/ML-based models and average AUC is as follows: cost prediction-0.77, LOS and discharges-0.78, readmissions and reoperations-0.66, preoperative patient selection/planning-0.79, adverse events and other postoperative complications-0.84, postoperative pain-0.83, postoperative patient-reported outcomes measures and functional outcome-0.81. Significant variability in model AUC across the different decision support applications was found (P < .001) with the AUC for readmission and reoperation models being significantly lower than that of the other decision support categories.

Conclusions

AI/ML-based applications in TJA continue to expand and have the potential to optimize patient selection and accurately predict postoperative outcomes, complications, and associated costs. On average, the AI/ML models performed best in predicting postoperative complications, pain, and patient-reported outcomes and were less accurate in predicting hospital readmissions and reoperations.

Association of Depression with In-Patient and Post-Discharge Disposition and Expenditures Among Medicare Beneficiaries Undergoing Resection for Cancer

BACKGROUND

The impact of depression on utilization of post-discharge care and overall episode of care expenditures remains poorly defined. We sought to define the impact of depression on postoperative outcomes, including discharge disposition, as well as overall expenditures associated with the global episode of surgical care.

METHOD

The Medicare 100% Standard Analytic Files were used to identify patients undergoing resection for esophageal, colon, rectal, pancreatic, and liver cancer between 2013 and 2017. The impact of depression on inpatient outcomes, as well as home health care and skilled nursing facilities utilization and expenditures, was analyzed.

RESULTS

Among 113,263 patients, 14,618 (12.9%) individuals had depression. Patients with depression were more likely to experience postoperative complications (odds ratio [OR] 1.36, 95% confidence interval [CI] 1.31-1.42), extended length of stay (LOS) (OR 1.41, 95% CI 1.36-1.47), readmission within 90 days (OR 1.20, 95% CI 1.14-1.25), as well as 90-day mortality (OR 1.35, 95% CI 1.27-1.42) (all p < 0.05). In turn, the proportion of patients who achieved a textbook outcome following cancer surgery was lower among patients with depression (no depression: 53.3% vs. depression: 45.3%; OR 0.70, 95% CI 0.68-0.73). Patients with a preexisting diagnosis of depression had higher odds of additional post-discharge expenditures compared with individuals without a diagnosis of depression (OR 1.42; 95% CI 1.35-1.50); patients with a preexisting diagnosis of depression ($10,500, IQR $3,200-$22,500) had higher median post-discharge expenditures versus patients without depression ($6600, IQR $2100-$17,400) (p < 0.001). On multivariable analysis, after controlling for other factors, depression remained associated with a 19.0% (95% confidence interval [CI] 15.7-22.3%) increase in post-discharge expenditures.

CONCLUSIONS

Patients with depression undergoing resection for cancer had worse in-patient outcomes and were less likely to achieve a TO. Patients with depression were more likely to require post-discharge care and had higher post-discharge expenditures.

Surgeon experience influences patient characteristics and outcomes in spine deformity surgery

PURPOSE

To characterize differences in patient demographics and outcomes by surgeon experience in a cohort of patients undergoing adult spinal deformity surgery.

METHODS

Patients undergoing degenerative spinal deformity were included. Patients whose surgeons graduated from fellowship ≤ 5 years prior to surgery versus > 5 years were compared. Multivariable linear and logistic regression, controlling for age, sex, comorbidity burden, number of segments fused, blood loss and operative time were used to evaluate differences in outcomes. Characteristics of operative invasiveness were plotted against surgeons' level of experience, and trends in these measures were assessed with univariate linear regression.

RESULTS

Three-hundred sixty-three patients were included. 147 patients' surgeons had ≤ 5 years of experience. Patient demographics were evenly matched. Patients with junior surgeons had more pre-existing medical complications, and senior surgeons were less likely to take care of patients with Medicare/Medicaid (p < 0.001). Junior surgeons were more likely to operate on non-elective patients (p < 0.001). Patients of junior surgeons received larger fusions (9.6 vs. 7.6 segments fused, p < 0.001). There were no differences in complication rates or death. Patients of junior surgeons had longer overall length of stays (p = 0.037) and higher rates of nonhome discharge (OR 2.0, p < 0.001), 30- and 90-day (p < 0.005) ED visits, and higher costs (+ $8548, 95% CI: $1596 to $15,502; p = 0.016).

CONCLUSION

Junior surgeons tend to perform more extensive deformity operations on more medically complex patients compared to senior surgeons, associated with higher costs and more resource utilization than senior surgeons.

Characterizing health care utilization following hospitalization for a traumatic brain injury: a retrospective cohort study

Objective: The purpose of this study was to characterize health services utilization among individuals hospitalized with a traumatic brain injury (TBI) 1-year post-injury.Methods: Using a retrospective cohort design, adult patients (n = 32, 042) hospitalized with a traumatic brain injury between 2005 and 2014 were selected from a statewide traumatic brain injury registry. Data on health services utilization for 1-year post-injury were extracted from electronic medical and administrative records. Descriptive statistics and logistic regression were used to characterize the cohort and a subgroup of superutilizers of health services.Results: One year after traumatic brain injury, 56% of participants used emergency department services, 80% received inpatient services, and 93% utilized outpatient health services. Superutilizers had ≥3 emergency department visits, ≥3 inpatient admissions, or ≥26 outpatient visits 1-year post-injury. Twenty-six percent of participants were superutilizers of emergency department services, 30% of inpatient services, and 26% of outpatient services. Superutilizers contributed to 81% of emergency department visits, 70% of inpatient visits, and 60% of outpatient visits. Factors associated with being a superutilizer included sex, race, residence, and insurance type.Conclusions: Several patient characteristics and demographic factors influenced patients' healthcare utilization post-TBI. Findings provide opportunities for developing targeted interventions to improve patients' health and traumatic brain injury-related healthcare delivery.

Are We at the Bottom? BPCI Programs Now Disincentivize Providers Who Maintain Quality Despite Caring for Increasingly Complex Patients

BACKGROUND

The Bundled Payments for Care Improvement (BPCI) initiative has been successful at reducing Medicare costs after total joint arthroplasty (TJA). Target pricing is based on each institution's historical performance and is periodically reset. The purpose of this study was to examine the performance of our BPCI program accounting for patient complexity, quality, and resource utilization.

METHODS

We reviewed a consecutive series of 9195 Medicare patients undergoing primary TJA from 2015 to 2018. Demographics, comorbidities, and readmissions by year were compared. We then examined 90-day episode-of-care costs, changes in target price, and financial margins during the duration of the BPCI program using Medicare claims data.

RESULTS

Patients undergoing TJA in 2018 had a higher prevalence of diabetes and cardiac disease (all P < .001) as compared with those in 2015. From 2015 to 2018, there was a decrease in the rate of discharge to rehabilitation facilities (23% vs 14%, P < .001) and length of stay (2.1 vs 1.7 days, P < .001) with no difference in readmissions (6% vs 6%, P = .945). There was a reduction in postacute care costs ($6076 vs $4,890, P < .001) and 90-day episode-of-care costs ($19,954 vs $18,449, P < .001). However, the target price also decreased ($22,280 vs $18,971, P < .001), and the per-patient margin diminished ($2683 vs $522, P < .001).

CONCLUSION

Surgeons have maintained quality of care at a reduced cost despite increasing patient complexity. The target price adjustments resulted in declining margins during the course of our BPCI experience. Policy makers should consider changes to target price methodology to encourage participation in these successful cost-saving programs.

The Removal of Total Hip and Total Knee Arthroplasty From the Inpatient-Only List Increases the Administrative Burden of Surgeons and Continues to Cause Confusion

BACKGROUND

Several studies have shown that the removal of total knee arthroplasty (TKA) from the Centers for Medicare and Medicaid Services (CMS) inpatient-only (IPO) list has caused confusion among surgeons, hospitals, and patients. The purpose of this study is to determine whether similar confusion was present after CMS recently removed total hip arthroplasty (THA) from the IPO list.

METHODS

We surveyed the American Association of Hip and Knee Surgeons membership via an online web-based questionnaire in February 2020. The 12-question form asked about practice type and the impact that having both THA and TKA removed from the IPO list has had on each surgeon's practice. Responses were tabulated and descriptive statistics of each question reported.

RESULTS

Of the 2847 American Association of Hip and Knee Surgeons members surveyed, 419 responded (14.7% response rate). Three hundred forty-one surgeons (81%) stated that changes to IPO status have increased their practice's administrative burden. Fifty-four percent of surgeons reported that they have needed to obtain preauthorization or appeal a denial of preauthorization for an inpatient total joint arthroplasty at least monthly, while 257 surgeons (61%) have had patients contact their office regarding an unexpected copayment. Despite the commitment of CMS to waiving certain audits for 2 years, 43 respondents (10%) stated they had undergone an audit regarding a patient's inpatient status.

CONCLUSION

The removal of THA and TKA from the IPO list continues to be an administrative burden for arthroplasty surgeons and a source of confusion among patients. CMS should provide additional guidance to address surgeons' concerns about preauthorization for inpatient stays, unexpected patient copayments, and CMS audits.

Patient generated health data and electronic health record integration in oncologic surgery: A call for artificial intelligence and machine learning

In this review, we aim to assess the current state of science in relation to the integration of patient-generated health data (PGHD) and patient-reported outcomes (PROs) into routine clinical care with a focus on surgical oncology populations. We will also describe the critical role of artificial intelligence and machine-learning methodology in the efficient translation of PGHD, PROs, and traditional outcome measures into meaningful patient care models.

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.

A Machine-Based Approach to Preoperatively Identify Patients with the Most and Least Benefit Associated with Resection for Intrahepatic Cholangiocarcinoma: An International Multi-institutional Analysis of 1146 Patients

BACKGROUND

Accurate risk stratification and patient selection is necessary to identify patients who will benefit the most from surgery or be better treated with other non-surgical treatment strategies. We sought to identify which patients in the preoperative setting would likely derive the most or least benefit from resection of intrahepatic cholangiocarcinoma (ICC).

METHODS

Patients who underwent curative-intent resection for ICC between 1990 and 2017 were identified from an international multi-institutional database. A machine-based classification and regression tree (CART) was used to generate homogeneous groups of patients relative to overall survival (OS) based on preoperative factors.

RESULTS

Among 1146 patients, CART analysis revealed tumor number and size, albumin-bilirubin (ALBI) grade and preoperative lymph node (LN) status as the strongest prognostic factors associated with OS among patients undergoing resection for ICC. In turn, four groups of patients with distinct outcomes were generated through machine learning: Group 1 (n = 228): single ICC, size ≤ 5 cm, ALBI grade I, negative preoperative LN status; Group 2 (n = 708): (1) single tumor > 5 cm, (2) single tumor ≤ 5 cm, ALBI grade 2/3, and (3) single tumor ≤ 5 cm, ALBI grade 1, metastatic/suspicious LNs; Group 3 (n = 150): 2-3 tumors; Group 4 (n = 60): ≥ 4 tumors. 5-year OS among Group 1, 2, 3, and 4 patients was 60.5%, 35.8%, 27.5%, and 3.8%, respectively (p < 0.001). Similarly, 5-year disease-free survival (DFS) among Group 1, 2, 3, and 4 patients was 47%, 27.2%, 6.8%, and 0%, respectively (p < 0.001).

CONCLUSIONS

The machine-based CART model identified distinct prognostic groups of patients with distinct outcomes based on preoperative factors. Survival decision trees may be useful as guides in preoperative patient selection and risk stratification.

Can We Improve Prediction of Adverse Surgical Outcomes? Development of a Surgical Complexity Score Using a Novel Machine Learning Technique

BACKGROUND

An optimal method to quantify surgical complexity using patient comorbidities derived from administrative billing data is lacking. We sought to develop a novel, easy-to-use surgical Complexity Score to accurately predict adverse outcomes among patients undergoing elective surgery.

STUDY DESIGN

A novel surgical Complexity Score was developed using 100% Medicare Inpatient and Outpatient Standard Analytic Files (SAFs) from years 2012 to 2016 (n = 1,049,160). Comorbid conditions were entered into a machine learning algorithm to assign weights to maximize the correlation with multiple postoperative outcomes including morbidity, readmission, mortality, and postoperative super-use. Predictive ability was compared against 3 of the most commonly used risk adjustment indices: the Charlson Comorbidity Index (CCI), Elixhauser Comorbidity Index (ECI), and the Centers for Medicare and Medicaid Service's Hierarchical Condition Category (CMS-HCC).

RESULTS

Patients underwent colectomy (12.6%), abdominal aortic aneurysm repair (4.4%), coronary artery bypass grafting (13.0%), total hip replacement (22.0%), total knee replacement (43.0%), or lung resection (5.0%). The Complexity Score had a good to very good predictive ability for all adverse outcomes. The Complexity Score had the highest accuracy in predicting perioperative morbidity (area under the curve [AUC]: 0.868, 95% CI 0.866 to 0.869); this performed better than the CCI (AUC: 0.717, 95% CI 0.715 to 0.719), ECI (AUC: 0.799, 95% CI 0.797 to 0.800), and similar to the CMS-HCC (AUC: 0.862, 95% CI 0.861 to 0.863). Similarly, the Complexity Score outperformed each of the 3 other comorbidity indices in predicting 90-day readmission (AUC: 0.707, 95% CI 0.705 to 0.709), 30-day readmission (AUC: 0.717, 95% CI 0.715 to 0.720), and postoperative super-use (AUC: 0.817, 95% CI 0.814 to 0.820).

CONCLUSIONS

Compared with the most commonly used comorbidity and surgical risk scores, the novel surgical Complexity Score outperformed the CCI, ECI, and CMS-HCC in predicting postoperative morbidity, 30-day readmission, 90-day readmission, and postoperative super-use.