Can a machine learning model accurately predict patient resource utilization following lumbar spinal fusion?

Machine Learning in Spine Surgery: A Narrative Review

Artificial intelligence and machine learning (ML) can offer revolutionary advances in their application to the field of spine surgery. Within the past 5 years, novel applications of ML have assisted in surgical decision-making, intraoperative imaging and navigation, and optimization of clinical outcomes. ML has the capacity to address many different clinical needs and improve diagnostic and surgical techniques. This review will discuss current applications of ML in the context of spine surgery by breaking down its implementation preoperatively, intraoperatively, and postoperatively. Ethical considerations to ML and challenges in ML implementation must be addressed to maximally benefit patients, spine surgeons, and the healthcare system. Areas for future research in augmented reality and mixed reality, along with limitations in generalizability and bias, will also be highlighted.

The Use of Artificial Intelligence for the Prediction of Periprosthetic Joint Infection Following Aseptic Revision Total Knee Arthroplasty

Periprosthetic joint infection (PJI) following revision total knee arthroplasty (TKA) for aseptic failure is associated with poor outcomes, patient morbidity, and high health care expenditures. The aim of this study was to develop novel machine learning algorithms for the prediction of PJI following revision TKA for patients with aseptic indications for revision surgery. A single-institution database consisting of 1,432 consecutive revision TKA patients with aseptic etiologies was retrospectively identified. The patient cohort included 208 patients (14.5%) who underwent re-revision surgery for PJI. Three machine learning algorithms (artificial neural networks, support vector machines, k-nearest neighbors) were developed to predict this outcome and these models were assessed by discrimination, calibration, and decision curve analysis. This is a retrospective study. Among the three machine learning models, the neural network model achieved the best performance across discrimination (area under the receiver operating characteristic curve = 0.78), calibration, and decision curve analysis. The strongest predictors for PJI following revision TKA for aseptic reasons were prior open procedure prior to revision surgery, drug abuse, obesity, and diabetes. This study utilized machine learning as a tool for the prediction of PJI following revision TKA for aseptic failure with excellent performance. The validated machine learning models can aid surgeons in patient-specific risk stratifying to assist in preoperative counseling and clinical decision making for patients undergoing aseptic revision TKA.

Machine Learning Predicts Decompression Levels for Lumbar Spinal Stenosis Using Canal Radiomic Features from Computed Tomography Myelography

BACKGROUND

The accurate preoperative identification of decompression levels is crucial for the success of surgery in patients with multi-level lumbar spinal stenosis (LSS). The objective of this study was to develop machine learning (ML) classifiers that can predict decompression levels using computed tomography myelography (CTM) data from LSS patients.

METHODS

A total of 1095 lumbar levels from 219 patients were included in this study. The bony spinal canal in CTM images was manually delineated, and radiomic features were extracted. The extracted data were randomly divided into training and testing datasets (8:2). Six feature selection methods combined with 12 ML algorithms were employed, resulting in a total of 72 ML classifiers. The main evaluation indicator for all classifiers was the area under the curve of the receiver operating characteristic (ROC-AUC), with the precision-recall AUC (PR-AUC) serving as the secondary indicator. The prediction outcome of ML classifiers was decompression level or not.

RESULTS

The embedding linear support vector (embeddingLSVC) was the optimal feature selection method. The feature importance analysis revealed the top 5 important features of the 15 radiomic predictors, which included 2 texture features, 2 first-order intensity features, and 1 shape feature. Except for shape features, these features might be eye-discernible but hardly quantified. The top two ML classifiers were embeddingLSVC combined with support vector machine (EmbeddingLSVC_SVM) and embeddingLSVC combined with gradient boosting (EmbeddingLSVC_GradientBoost). These classifiers achieved ROC-AUCs over 0.90 and PR-AUCs over 0.80 in independent testing among the 72 classifiers. Further comparisons indicated that EmbeddingLSVC_SVM appeared to be the optimal classifier, demonstrating superior discrimination ability, slight advantages in the Brier scores on the calibration curve, and Net benefits on the Decision Curve Analysis.

CONCLUSIONS

ML successfully extracted valuable and interpretable radiomic features from the spinal canal using CTM images, and accurately predicted decompression levels for LSS patients. The EmbeddingLSVC_SVM classifier has the potential to assist surgical decision making in clinical practice, as it showed high discrimination, advantageous calibration, and competitive utility in selecting decompression levels in LSS patients using canal radiomic features from CTM.

Artificial Intelligence for Automated Implant Identification in Knee Arthroplasty: A Multicenter External Validation Study Exceeding 3.5 Million Plain Radiographs

BACKGROUND

Surgical management of complications following knee arthroplasty demands accurate and timely identification of implant manufacturer and model. Automated image processing using deep machine learning has been previously developed and internally validated; however, external validation is essential prior to scaling clinical implementation for generalizability.

METHODS

We trained, validated, and externally tested a deep learning system to classify knee arthroplasty systems as one of the 9 models from 4 manufacturers derived from 4,724 original, retrospectively collected anteroposterior plain knee radiographs across 3 academic referral centers. From these radiographs, 3,568 were used for training, 412 for validation, and 744 for external testing. Augmentation was applied to the training set (n = 3,568,000) to increase model robustness. Performance was determined by the area under the receiver operating characteristic curve, sensitivity, specificity, and accuracy. Implant identification processing speed was calculated. The training and testing sets were drawn from statistically different populations of implants (P < .001).

RESULTS

After 1,000 training epochs by the deep learning system, the system discriminated 9 implant models with a mean area under the receiver operating characteristic curve of 0.989, accuracy of 97.4%, sensitivity of 89.2%, and specificity of 99.0% in the external testing dataset of 744 anteroposterior radiographs. The software classified implants at a mean speed of 0.02 seconds per image.

CONCLUSION

An artificial intelligence-based software for identifying knee arthroplasty implants demonstrated excellent internal and external validation. Although continued surveillance is necessary with implant library expansion, this software represents a responsible and meaningful clinical application of artificial intelligence with immediate potential to globally scale and assist in preoperative planning prior to revision knee arthroplasty.

Artificial Intelligence for Automated Implant Identification in Total Hip Arthroplasty: A Multicenter External Validation Study Exceeding Two Million Plain Radiographs

BACKGROUND

The surgical management of complications after total hip arthroplasty (THA) necessitates accurate identification of the femoral implant manufacturer and model. Automated image processing using deep learning has been previously developed and internally validated; however, external validation is necessary prior to responsible application of artificial intelligence (AI)-based technologies.

METHODS

We trained, validated, and externally tested a deep learning system to classify femoral-sided THA implants as one of the 8 models from 2 manufacturers derived from 2,954 original, deidentified, retrospectively collected anteroposterior plain radiographs across 3 academic referral centers and 13 surgeons. From these radiographs, 2,117 were used for training, 249 for validation, and 588 for external testing. Augmentation was applied to the training set (n = 2,117,000) to increase model robustness. Performance was evaluated by area under the receiver operating characteristic curve, sensitivity, specificity, and accuracy. Implant identification processing speed was calculated.

RESULTS

The training and testing sets were drawn from statistically different populations of implants (P < .001). After 1,000 training epochs by the deep learning system, the system discriminated 8 implant models with a mean area under the receiver operating characteristic curve of 0.991, accuracy of 97.9%, sensitivity of 88.6%, and specificity of 98.9% in the external testing dataset of 588 anteroposterior radiographs. The software classified implants at a mean speed of 0.02 seconds per image.

CONCLUSION

An AI-based software demonstrated excellent internal and external validation. Although continued surveillance is necessary with implant library expansion, this software represents responsible and meaningful clinical application of AI with immediate potential to globally scale and assist in preoperative planning prior to revision THA.

Machine Learning Applications in Spine Surgery

This literature review sought to identify and evaluate the current applications of artificial intelligence (AI)/machine learning (ML) in spine surgery that can effectively guide clinical decision-making and surgical planning. By using specific keywords to maximize search sensitivity, a thorough literature research was conducted in several online databases: Scopus, PubMed, and Google Scholar, and the findings were filtered according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A total of 46 studies met the requirements and were included in this review. According to this study, AI/ML models were sufficiently accurate with a mean overall value of 74.9%, and performed best at preoperative patient selection, cost prediction, and length of stay. Performance was also good at predicting functional outcomes and postoperative mortality. Regression analysis was the most frequently utilized application whereas deep learning/artificial neural networks had the highest sensitivity score (81.5%). Despite the relatively brief history of engagement with AI/ML, as evidenced by the fact that 77.5% of studies were published after 2018, the outcomes have been promising. In light of the Big Data era, the increasing prevalence of National Registries, and the wide-ranging applications of AI, such as exemplified by ChatGPT (OpenAI, San Francisco, California), it is highly likely that the field of spine surgery will gradually adopt and integrate AI/ML into its clinical practices. Consequently, it is of great significance for spine surgeons to acquaint themselves with the fundamental principles of AI/ML, as these technologies hold the potential for substantial improvements in overall patient care.

Accounting for age in prediction of discharge destination following elective lumbar fusion: a supervised machine learning approach

BACKGROUND CONTEXT

The number of elective spinal fusion procedures performed each year continues to grow, making risk factors for post-operative complications following this procedure increasingly clinically relevant. Nonhome discharge (NHD) is of particular interest due to its associations with increased costs of care and rates of complications. Notably, increased age has been found to influence rates of NHD.

PURPOSE

To identify aged-adjusted risk factors for nonhome discharge following elective lumbar fusion through the utilization of Machine Learning-generated predictions within stratified age groupings.

STUDY DESIGN

Retrospective Database Study.

PATIENT SAMPLE

The American College of Surgeons National Quality Improvement Program (ACS-NSQIP) database years 2008 to 2018.

OUTCOME MEASURES

Postoperative discharge destination.

METHODS

ACS-NSQIP was queried to identify adult patients undergoing elective lumbar spinal fusion from 2008 to 2018. Patients were then stratified into the following age ranges: 30 to 44 years, 45 to 64 years, and ≥65 years. These groups were then analyzed by eight ML algorithms, each tasked with predicting post-operative discharge destination.

RESULTS

Prediction of NHD was performed with average AUCs of 0.591, 0.681, and 0.693 for those aged 30 to 44, 45 to 64, and ≥65 years respectively. In patients aged 30 to 44, operative time (p<.001), African American/Black race (p=.003), female sex (p=.002), ASA class three designation (p=.002), and preoperative hematocrit (p=.002) were predictive of NHD. In ages 45 to 64, predictive variables included operative time, age, preoperative hematocrit, ASA class two or class three designation, insulin-dependent diabetes, female sex, BMI, and African American/Black race all with p<.001. In patients ≥65 years, operative time, adult spinal deformity, BMI, insulin-dependent diabetes, female sex, ASA class four designation, inpatient status, age, African American/Black race, and preoperative hematocrit were predictive of NHD with p<.001. Several variables were distinguished as predictive for only one age group including ASA Class two designation in ages 45 to 64 and adult spinal deformity, ASA class four designation, and inpatient status for patients ≥65 years.

CONCLUSIONS

Application of ML algorithms to the ACS-NSQIP dataset identified a number of highly predictive and age-adjusted variables for NHD. As age is a risk factor for NHD following spinal fusion, our findings may be useful in both guiding perioperative decision-making and recognizing unique predictors of NHD among specific age groups.

Artificial intelligence-based applications in shoulder surgery leaves much to be desired: a systematic review

Background

Artificial intelligence (AI) aims to simulate human intelligence using automated computer algorithms. There has been a rapid increase in research applying AI to various subspecialties of orthopedic surgery, including shoulder surgery. The purpose of this review is to assess the scope and validity of current clinical AI applications in shoulder surgery literature.

Methods

A systematic literature review was conducted using PubMed for all articles published between January 1, 2010 and June 10, 2022. The search query used the terms as follows: (artificial intelligence OR machine learning OR deep learning) AND (shoulder OR shoulder surgery OR rotator cuff). All studies that examined AI application models in shoulder surgery were included and evaluated for model performance and validation (internal, external, or both).

Results

A total of 45 studies were included in the final analysis. Eighteen studies involved shoulder arthroplasty, 13 rotator cuff, and 14 other areas. Studies applying AI to shoulder surgery primarily involved (1) automated imaging analysis including identifying rotator cuff tears and shoulder implants (2) risk prediction analyses including perioperative complications, functional outcomes, and patient satisfaction. Highest model performance area under the curve ranged from 0.681 (poor) to 1.00 (perfect). Only 2 studies reported external validation.

Conclusion

Applications of AI in the field of shoulder surgery are expanding rapidly and offer patient-specific risk stratification for shared decision-making and process automation for resource preservation. However, model performance is modest and external validation remains to be demonstrated, suggesting increased scientific rigor is warranted prior to deploying AI-based clinical applications.

Social Media Presence Is Associated With Diversity and Application Volume for Orthopedic Surgery Residency Programs

The purpose of this study was to assess the association between social media presence (Twitter and Instagram), diversity in orthopedic surgery residency programs, and the number of applications received by a program. Data from Twitter and Instagram for 179 orthopedic residency programs accredited by the Accreditation Council for Graduate Medical Education were collected, including the presence of a social media account, date of first post, number of posts, and number of followers. Residency program data were collected from the Association of American Medical Colleges Residency Explorer Tool and included percentage of Whiteresidents, percentage of male residents, residency ranking, and number of applications submitted during the 2019 application cycle. Bivariate and multivariable analyses were performed with adjustment for program ranking. Of 179 residency programs, 34.6% (n=62) had Twitter, and 16.7% (n=30) had Instagram. Overall, 39.7% (n=71) had a social media presence, defined as having at least one of the two forms of social media. Programs with social media presences had higher average rankings (48.1 vs 99.6 rank, P<.001). After adjusting for program ranking, social media presence was associated with increased applications during the 2019 application cycle (odds ratio [OR]=2.76, P=.010). Social media presence was associated with increased odds of gender diversity (OR=3.07, P=.047) and racial diversity (OR=2.21, P=.041). Individually, Twitter presence was associated with increased odds of gender (OR=4.81, P=.018) and racial diversity (OR=4.00, P=.021), but Instagram was not (P>.05). Social media presence is associated with more residency program applications and increased resident diversity. Social media can be used to highlight inclusivity measures and related opportunities. [Orthopedics. 2023;46(1):47-53.].

Prediction of Admission Costs Following Anterior Cervical Discectomy and Fusion Utilizing Machine Learning

STUDY DESIGN

Retrospective case series.

OBJECTIVE

Predict cost following anterior cervical discectomy and fusion (ACDF) within the 90-day global period using machine learning models.

BACKGROUND

The incidence of ACDF has been increasing with a disproportionate decrease in reimbursement. As bundled payment models become common, it is imperative to identify factors that impact the cost of care.

MATERIALS AND METHODS

The Nationwide Readmissions Database (NRD) was accessed in 2018 for all primary ACDFs by the International Classification of Diseases 10th Revision (ICD-10) procedure codes. Costs were calculated by utilizing the total hospital charge and each hospital's cost-to-charge ratio. Hospital characteristics, such as volume of procedures performed and wage index, were also queried. Readmissions within 90 days were identified, and cost of readmissions was added to the total admission cost to represent the 90-day healthcare cost. Machine learning algorithms were used to predict patients with 90-day admission costs >1 SD from the mean.

RESULTS

There were 42,485 procedures included in this investigation with an average age of 57.7±12.3 years with 50.6% males. The average cost of the operative admission was $24,874±25,610, the average cost of readmission was $25,371±11,476, and the average total cost was $26,977±28,947 including readmissions costs. There were 10,624 patients who were categorized as high cost. Wage index, hospital volume, age, and diagnosis-related group severity were most correlated with the total cost of care. Gradient boosting trees algorithm was most predictive of the total cost of care (area under the curve=0.86).

CONCLUSIONS

Bundled payment models utilize wage index and diagnosis-related groups to determine reimbursement of ACDF. However, machine learning algorithms identified additional variables, such as hospital volume, readmission, and patient age, that are also important for determining the cost of care. Machine learning can improve cost-effectiveness and reduce the financial burden placed upon physicians and hospitals by implementing patient-specific reimbursement.

Current Applications of Machine Learning in Spine: From Clinical View

STUDY DESIGN

Narrative review.

OBJECTIVES

This review aims to present current applications of machine learning (ML) in spine domain to clinicians.

METHODS

We conducted a comprehensive PubMed search of peer-reviewed articles that were published between 2006 and 2020 using terms (spine, spinal, lumbar, cervical, thoracic, machine learning) to examine ML in spine. Then exclude research of other domain, case report, review or meta-analysis, and which without available abstract or full text.

RESULTS

Total 1738 articles were retrieved from database, and 292 studies were finally included. Key findings of current applications were compiled and summarized in this review. Main clinical applications of those techniques including image processing, diagnosis, decision supporting, operative assistance, rehabilitation, surgery outcomes, complications, hospitalization and cost.

CONCLUSIONS

ML had achieved excellent performance and hold immense potential in spine. ML could help clinical staff to improve medical level, enhance work efficiency, and reduce adverse events. However more randomized controlled trials and improvement of interpretability are essential to clinicians accepting models' assistance in real work.

Artificial Learning and Machine Learning Applications in Spine Surgery: A Systematic Review

OBJECTIVES

This current systematic review sought to identify and evaluate all current research-based spine surgery applications of AI/ML in optimizing preoperative patient selection, as well as predicting and managing postoperative outcomes and complications.

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 guidelines.

RESULTS

After application of inclusion and exclusion criteria, 41 studies were included in this review. Bayesian networks had the highest average AUC (.80), and neural networks had the best accuracy (83.0%), sensitivity (81.5%), and specificity (71.8%). Preoperative planning/cost prediction models (.89,82.2%) and discharge/length of stay models (.80,78.0%) each reported significantly higher average AUC and accuracy compared to readmissions/reoperation prediction models (.67,70.2%) (P < .001, P = .005, respectively). Model performance also significantly varied across postoperative management applications for average AUC and accuracy values (P < .001, P < .027, respectively).

CONCLUSIONS

Generally, authors of the reviewed studies concluded that AI/ML offers a potentially beneficial tool for providers to optimize patient care and improve cost-efficiency. More specifically, AI/ML models performed best, on average, when optimizing preoperative patient selection and planning and predicting costs, hospital discharge, and length of stay. However, models were not as accurate in predicting postoperative complications, adverse events, and readmissions and reoperations. An understanding of AI/ML-based applications is becoming increasingly important, particularly in spine surgery, as the volume of reported literature, technology accessibility, and clinical applications continue to rapidly expand.

Healthcare resource utilization and costs 2 years pre- and post-lumbar spine surgery for stenosis: a national claims cohort study of 22,182 cases

BACKGROUND CONTEXT

Improved understanding of the pre- and postoperative trends in costs and healthcare resource utilization (HCRU) is needed to better inform patient expectations and aid in the development of strategies to minimize the significant healthcare burden associated with lumbar spine surgery.

PURPOSE

Examine the time course of costs and HCRU in the 2 years preceding and following elective lumbar spine surgery for stenosis in a large national claims cohort.

STUDY DESIGN/SETTING

Retrospective analysis of an administrative claims database (IBM® Marketscan® Research Databases 2007-2015).

PATIENT SAMPLE

Adult patients undergoing elective primary single-level lumbar surgery for stenosis with at least 2 years of continuous health plan enrollment pre- and postoperatively.

OUTCOME MEASURES

Functional measures, including monthly rates of HCRU (15 categories), monthly gross covered payments (including payments made by the health plan and deductibles and coinsurance paid by the patient) overall, by HCRU category, and by spine versus non-spine-related.

METHODS

All available patients were utilized for analysis of HCRU. For analysis of payments, only patients on noncapitated health plans providing accurate financial information were analyzed. Payments were converted to 2015 United States dollars using the medical care component of the consumer price index. Trends in payments and HCRU were plotted on a monthly basis pre- and post-surgery and assessed with regression models. Relationships with demographics, surgical factors, and comorbidities were assessed with multivariable repeated measures generalized estimating equations.

RESULTS

Median monthly healthcare payments 2 years prior to surgery were $275 ($22, $868). Baseline HCRU at 2 years preoperatively was stable or only gradually rising (office visits, prescription drug use), but began an increasingly steep rise in many categories 6 to 12 months prior to surgery. Monthly payments began an increasingly steep rise 6 months prior to surgery, reaching a peak of $1,402 ($634, $2,827) in the month prior to surgery. This was driven by an increase in radiology, office visits, PT, injections, prescription medications, ER encounters, and inpatient admissions. Payments dropped dramatically immediately following surgery. Over the remainder of the 2 years, the median total payments declined only slightly, as a continued decline in spine-related payments was offset by gradually increased non-spine related payments as patients aged. By 2 years postoperatively, the percentage of patients using PT and injections returned to within 1% of the baseline levels observed 2 years preoperatively; however, spine-related prescription medication use remained elevated, as did other categories of HCRU (radiology, office visits, lab/diagnostic services, and also rare events such as inpatient admissions, ER encounters, and SNF/IRF). Patients with a fusion component to their surgeries had higher payments and HCRU preoperatively, and this did not resolve postoperatively. Variations in payments and HCRU were also evident among plan types, with patients on comprehensive medical plans-predominantly employer-sponsored supplemental Medicare coverage-utilizing more inpatient, ER, and inpatient rehabilitation & skilled nursing facilities. Patients on high-deductible plans had fewer payments and HCRU across all categories; however, we are unable to distinguish whether this is because they used fewer of these services or if they were paying for these services out of pocket without submitting to the payer. By 2 years postoperatively, 51% of patients had no spine-related monthly payments, while 33% had higher and 16% had lower monthly payments relative to 2 years preoperatively.

CONCLUSIONS

This is the first study to characterize time trends in direct healthcare payments and HCRU over an extended period preceding and following spine surgery. Differences among plan types potentially highlight disparities in access to care and plan-related financial mediators of patients' healthcare resource utilization.

The Ability of Deep Learning Models to Identify Total Hip and Knee Arthroplasty Implant Design From Plain Radiographs

INTRODUCTION

The surgical management of patients with failed total hip or knee arthroplasty (THA and TKA) necessitates the identification of the implant manufacturer and model. Failure to accurately identify implant design leads to delays in care, increased morbidity, and healthcare costs. The automated identification of implant designs has the potential to assist in the surgical management of patients with failed arthroplasty. This study aimed to develop and validate a convolutional neural network deep learning model for the identification of primary and revision hip and knee total joint arthroplasty designs from plain radiographs.

METHODS

This study trained a convolutional neural network deep learning model to automatically identify 24 THA designs and 14 TKA designs from 11,204 anterior-posterior radiographs obtained from 8,763 patients. From these radiographs, 8,963 radiographs (80%) were used for model training and 2,241 radiographs (20%) were used for model validation. Model performance was assessed through receiver operating curve characteristics.

RESULTS

After 1,000 training epochs by the convolutional neural network deep learning model, the computational model discriminated 17 primary THA designs with an area under the receiver operating curve (AUC) of 0.98, sensitivity of 95.8%, and specificity of 98.6%. The deep learning model discriminated eight primary TKA designs with an AUC of 0.97, sensitivity of 94.9%, and specificity of 97.8%. The deep learning model demonstrated an AUC of 0.98 and 0.96 for the identification of seven revision THA and six revision TKA designs, respectively.

DISCUSSION

This study developed and validated a convolutional neural network deep learning model for the identification of hip and knee total joint arthroplasty designs from plain radiographs. The study findings demonstrate excellent accuracy of the deep learning model for the identification of 24 THA and 14 TKA designs, illustrating the great potential of the deep learning model to assist in preoperative surgical planning of failed arthroplasty patients.

Artificial Intelligence-Driven Prediction Modeling and Decision Making in Spine Surgery Using Hybrid Machine Learning Models

Healthcare systems worldwide generate vast amounts of data from many different sources. Although of high complexity for a human being, it is essential to determine the patterns and minor variations in the genomic, radiological, laboratory, or clinical data that reliably differentiate phenotypes or allow high predictive accuracy in health-related tasks. Convolutional neural networks (CNN) are increasingly applied to image data for various tasks. Its use for non-imaging data becomes feasible through different modern machine learning techniques, converting non-imaging data into images before inputting them into the CNN model. Considering also that healthcare providers do not solely use one data modality for their decisions, this approach opens the door for multi-input/mixed data models which use a combination of patient information, such as genomic, radiological, and clinical data, to train a hybrid deep learning model. Thus, this reflects the main characteristic of artificial intelligence: simulating natural human behavior. The present review focuses on key advances in machine and deep learning, allowing for multi-perspective pattern recognition across the entire information set of patients in spine surgery. This is the first review of artificial intelligence focusing on hybrid models for deep learning applications in spine surgery, to the best of our knowledge. This is especially interesting as future tools are unlikely to use solely one data modality. The techniques discussed could become important in establishing a new approach to decision-making in spine surgery based on three fundamental pillars: (1) patient-specific, (2) artificial intelligence-driven, (3) integrating multimodal data. The findings reveal promising research that already took place to develop multi-input mixed-data hybrid decision-supporting models. Their implementation in spine surgery may hence be only a matter of time.

Systematic review of prediction models for postacute care destination decision-making

OBJECTIVE

This article reports a systematic review of studies containing development and validation of models predicting postacute care destination after adult inpatient hospitalization, summarizes clinical populations and variables, evaluates model performance, assesses risk of bias and applicability, and makes recommendations to reduce bias in future models.

MATERIALS AND METHODS

A systematic literature review was conducted following PRISMA guidelines and the Cochrane Prognosis Methods Group criteria. Online databases were searched in June 2020 to identify all published studies in this area. Data were extracted based on the CHARMS checklist, and studies were evaluated based on predictor variables, validation, performance in validation, risk of bias, and applicability using the Prediction Model Risk of Bias Assessment Tool (PROBAST) tool.

RESULTS

The final sample contained 28 articles with 35 models for evaluation. Models focused on surgical (22), medical (5), or both (8) populations. Eighteen models were internally validated, 10 were externally validated, and 7 models underwent both types. Model performance varied within and across populations. Most models used retrospective data, the median number of predictors was 8.5, and most models demonstrated risk of bias.

DISCUSSION AND CONCLUSION

Prediction modeling studies for postacute care destinations are becoming more prolific in the literature, but model development and validation strategies are inconsistent, and performance is variable. Most models are developed using regression, but machine learning methods are increasing in frequency. Future studies should ensure the rigorous variable selection and follow TRIPOD guidelines. Only 14% of the models have been tested or implemented beyond original studies, so translation into practice requires further investigation.

The Influence of Baseline Clinical Status and Surgical Strategy on Early Good to Excellent Result in Spinal Lumbar Arthrodesis: A Machine Learning Approach

The study aims to create a preoperative model from baseline demographic and health-related quality of life scores (HRQOL) to predict a good to excellent early clinical outcome using a machine learning (ML) approach. A single spine surgery center retrospective review of prospectively collected data from January 2016 to December 2020 from the institutional registry (SpineREG) was performed. The inclusion criteria were age ≥ 18 years, both sexes, lumbar arthrodesis procedure, a complete follow up assessment (Oswestry Disability Index-ODI, SF-36 and COMI back) and the capability to read and understand the Italian language. A delta of improvement of the ODI higher than 12.7/100 was considered a "good early outcome". A combined target model of ODI (Δ ≥ 12.7/100), SF-36 PCS (Δ ≥ 6/100) and COMI back (Δ ≥ 2.2/10) was considered an "excellent early outcome". The performance of the ML models was evaluated in terms of sensitivity, i.e., True Positive Rate (TPR), specificity, i.e., True Negative Rate (TNR), accuracy and area under the receiver operating characteristic curve (AUC ROC). A total of 1243 patients were included in this study. The model for predicting ODI at 6 months' follow up showed a good balance between sensitivity (74.3%) and specificity (79.4%), while providing a good accuracy (75.8%) with ROC AUC = 0.842. The combined target model showed a sensitivity of 74.2% and specificity of 71.8%, with an accuracy of 72.8%, and an ROC AUC = 0.808. The results of our study suggest that a machine learning approach showed high performance in predicting early good to excellent clinical results.

Prediction of outcome after spinal surgery-using The Dialogue Support based on the Swedish national quality register

PURPOSE

To evaluate the predictive precision of the Dialogue Support, a tool for additional help in shared decision-making before surgery of the degenerative spine.

METHODS

Data in Swespine (Swedish national quality registry) of patients operated between 2007 and 2019 found the development of prediction algorithms based on logistic regression analyses, where socio-demographic and baseline variables were included. The algorithms were tested in four diagnostic groups: lumbar disc herniation, lumbar spinal stenosis, degenerative disc disease and cervical radiculopathy. By random selection, 80% of the study population was used for the prediction of outcome and then tested against the actual outcome of the remaining 20%. Outcome measures were global assessment of pain (GA), and satisfaction with outcome.

RESULTS

Calibration plots demonstrated a high degree of concordance on a group level. On an individual level, ROC curves showed moderate predictive capacity with AUC (area under the curve) values 0.67-0.68 for global assessment and 0.6-0.67 for satisfaction.

CONCLUSION

The Dialogue Support can serve as an aid to both patient and surgeon when discussing and deciding on surgical treatment of degenerative conditions in the lumbar and cervical spine.

LEVEL OF EVIDENCE

I.

Development and Internal Validation of Supervised Machine Learning Algorithms for Predicting Clinically Significant Functional Improvement in a Mixed Population of Primary Hip Arthroscopy

PURPOSE

To (1) develop and validate a machine learning algorithm to predict clinically significant functional improvements after hip arthroscopy for femoroacetabular impingement syndrome and to (2) develop a digital application capable of providing patients with individual risk profiles to determine their propensity to gain clinically significant improvements in function.

METHODS

A retrospective review of consecutive hip arthroscopy patients who underwent cam/pincer correction, labral preservation, and capsular closure between January 2012 and 2017 from 1 large academic and 3 community hospitals operated on by a single high-volume hip arthroscopist was performed. The primary outcome was the minimal clinically important difference (MCID) for the Hip Outcome Score (HOS)-Activities of Daily Living (ADL) at 2 years postoperatively, which was calculated using a distribution-based method. A total of 21 demographic, radiographic, and patient-reported outcome measures were considered as potential covariates. An 80:20 random split was used to create training and testing sets from the patient cohort. Five supervised machine learning algorithms were developed using 3 iterations of 10-fold cross-validation on the training set and assessed by discrimination, calibration, Brier score, and decision curve analysis on an independent testing set of patients.

RESULTS

A total of 818 patients with a median (interquartile range) age of 32.0 (22.0-42.0) and 69.2% female were included, of whom 74.3% achieved the MCID for the HOS-ADL. The best-performing algorithm was the stochastic gradient boosting model (c-statistic = 0.84, calibration intercept = 0.20, calibration slope = 0.83, and Brier score = 0.13). Of the initial 21 candidate variables, the 8 most important features for predicting the MCID for the HOS-ADL included in model training were body mass index, age, preoperative HOS-ADL score, preoperative pain level, sex, Tönnis grade, symptom duration, and drug allergies. The algorithm was subsequently transformed into a digital application using local explanations to provide customized risk assessment: https://orthoapps.shinyapps.io/HPRG_ADL/.

CONCLUSIONS

The stochastic boosting gradient model conferred excellent predictive ability for propensity to gain clinically significant improvements in function after hip arthroscopy. An open-access digital application was created, which may augment shared decision-making and allow for preoperative risk stratification. External validation of this model is warranted to confirm the performance of these algorithms, as the generalizability is currently unknown.

LEVEL OF EVIDENCE

IV, Case series.

Artificial Intelligence for the Orthopaedic Surgeon: An Overview of Potential Benefits, Limitations, and Clinical Applications

Artificial intelligence (AI), along with its subset technology machine learning, has transformed numerous industries through newfound efficiencies and supportive decision-making. These technologies have similarly begun to find application within United States healthcare, particularly orthopaedics. Although these modalities have the potential to similarly transform health care, there exist limitations that must also be recognized and understood. Unfortunately, most clinicians do not have an understanding of the fundamentals of AI and therefore may have challenges in contextualizing its impact in modern healthcare. The purpose of this review was to provide an overview of the key concepts of AI and machine learning with the orthopaedic surgeon in mind. The review further highlights the potential benefits and limitations of AI, along with an overview of its applications, in orthopaedics.

Machine Learning Predicts Femoral and Tibial Implant Size Mismatch for Total Knee Arthroplasty

Background

Despite reasonable accuracy with preoperative templating, the search for an optimal planning tool remains an unsolved dilemma. The purpose of the present study was to apply machine learning (ML) using preoperative demographic variables to predict mismatch between templating and final component size in primary total knee arthroplasty (TKA) cases.

Methods

This was a retrospective case-control study of primary TKA patients between September 2012 and April 2018. The primary outcome was mismatch between the templated and final implanted component sizes extracted from the operative database. The secondary outcome was mismatch categorized as undersized and oversized. Five supervised ML algorithms were trained using 6 demographic features. Prediction accuracies were obtained as a metric of performance for binary mismatch (yes/no) and multilevel (undersized/correct/oversized) classifications.

Results

A total of 1801 patients were included. For binary classification, the best-performing algorithm for predicting femoral and tibial mismatch was the stochastic gradient boosting model (area under the curve: 0.76/0.72, calibration intercepts: 0.05/0.05, calibration slopes: 0.55/0.7, and Brier scores: 0.20/0.21). For multiclass classification, the best-performing algorithms had accuracies of 83.9% and 82.9% for predicting the concordance/mismatch of the femoral and tibial implant, respectively. Model predictions of greater than 51.0% and 47.9% represented high-risk thresholds for femoral and tibial sizing mismatch, respectively.

Conclusions

ML algorithms predicted templating mismatch with good accuracy. External validation is necessary to confirm the performance and reliability of these algorithms. Predicting sizing mismatch is the first step in using ML to aid in the prediction of final TKA component sizes. Further studies to optimize parameters and predictions for the algorithms are ongoing.

The value of artificial neural networks for predicting length of stay, discharge disposition, and inpatient costs after anatomic and reverse shoulder arthroplasty

HYPOTHESIS/PURPOSE

The objective is to develop and validate an artificial intelligence model, specifically an artificial neural network (ANN), to predict length of stay (LOS), discharge disposition, and inpatient charges for primary anatomic total (aTSA), reverse total (rTSA), and hemi- (HSA) shoulder arthroplasty to establish internal validity in predicting patient-specific value metrics.

METHODS

Using data from the National Inpatient Sample between 2003 and 2014, 4 different ANN models to predict LOS, discharge disposition, and inpatient costs using 39 preoperative variables were developed based on diagnosis and arthroplasty type: primary chronic/degenerative aTSA, primary chronic/degenerative rTSA, primary traumatic/acute rTSA, and primary acute/traumatic HSA. Models were also combined into diagnosis type only. Outcome metrics included accuracy and area under the curve (AUC) for a receiver operating characteristic curve.

RESULTS

A total of 111,147 patients undergoing primary shoulder replacement were included. The machine learning algorithm predicting the overall chronic/degenerative conditions model (aTSA, rTSA) achieved accuracies of 76.5%, 91.8%, and 73.1% for total cost, LOS, and disposition, respectively; AUCs were 0.75, 0.89, and 0.77 for total cost, LOS, and disposition, respectively. The overall acute/traumatic conditions model (rTSA, HSA) had accuracies of 70.3%, 79.1%, and 72.0% and AUCs of 0.72, 0.78, and 0.79 for total cost, LOS, and discharge disposition, respectively.

CONCLUSION

Our ANN demonstrated fair to good accuracy and reliability for predicting inpatient cost, LOS, and discharge disposition in shoulder arthroplasty for both chronic/degenerative and acute/traumatic conditions. Machine learning has the potential to preoperatively predict costs, LOS, and disposition using patient-specific data for expectation management between health care providers, patients, and payers.

A novel online calculator predicting short-term postoperative outcomes in patients with metastatic brain tumors

Purpose

Establishing predictors of hospital length of stay (LOS), discharge deposition, and total hospital charges is essential to providing high-quality, value-based care. Though previous research has investigated these outcomes for patients with metastatic brain tumors, there are currently no tools that synthesize such research findings and allow for prediction of these outcomes on a patient-by-patient basis. The present study sought to develop a prediction calculator that uses patient demographic and clinical information to predict extended hospital length of stay, non-routine discharge disposition, and high total hospital charges for patients with metastatic brain tumors.

Methods

Patients undergoing surgery for metastatic brain tumors at a single academic institution were analyzed (2017–2019). Multivariate logistic regression was used to identify independent predictors of extended LOS (> 7 days), non-routine discharge, and high total hospital charges (> $ 46,082.63). p < 0.05 was considered statistically significant. C-statistics and the Hosmer–Lemeshow test were used to assess model discrimination and calibration, respectively.

Results

A total of 235 patients were included in our analysis, with a mean age of 62.74 years. The majority of patients were female (52.3%) and Caucasian (76.6%). Our models predicting extended LOS, non-routine discharge, and high hospital charges had optimism-corrected c-statistics > 0.7, and all three models demonstrated adequate calibration (p > 0.05). The final models are available as an online calculator (https://neurooncsurgery.shinyapps.io/brain_mets_calculator/).

Conclusions

Our models predicting postoperative outcomes allow for individualized risk-estimation for patients following surgery for metastatic brain tumors. Our results may be useful in helping clinicians to provide resource-conscious, high-value care.