Development of Machine Learning Algorithms to Predict Patient Dissatisfaction After Primary Total Knee Arthroplasty

BackgroundPostoperative dissatisfaction after primary total knee arthroplasty (TKA) that requires additional care or readmission may impose a significant financial burden to healthcare systems. The purpose of the current study is to develop machine learning algorithms to predict dissatisfaction after TKA.MethodsA retrospective review of consecutive TKA patients between 2014 and 2016 from 1 large academic and 2 community hospitals was performed. Preoperative variables considered for prediction included demographics, medical history, flexion contracture, knee flexion, and outcome scores (patient-reported health state, Knee Society Score [KSS], and KSS-Function [KSS-F]). Recursive feature elimination was used to select features that optimized algorithm performance. Five supervised machine learning algorithms were developed by training with 10-fold cross-validation 3 times. These algorithms were subsequently applied to an independent testing set of patients and assessed by discrimination, calibration, Brier score, and decision curve analysis.ResultsOf 430 patients, a total of 40 (9.0%) were dissatisfied with their outcome after primary TKA at a minimum of 2 years postoperatively. The random forest algorithm achieved the best performance in the independent testing set not used for algorithm development (c-statistic: 0.77, calibration intercept: 0.087, calibration slope: 0.74, Brier score: 0.082). The most important factors for predicting dissatisfaction were age, number of medical comorbidities, presence of one or more drug allergies, preoperative patient-reported health state score, and preoperative KSS.ConclusionThe current study developed machine learning algorithms based on partially modifiable risk factors for predicting dissatisfaction after TKA. This model demonstrates good discriminative capacity for identifying those at greatest risk for dissatisfaction and may allow for preoperative health optimization.     

Deep Learning Preoperatively Predicts Value Metrics for Primary Total Knee Arthroplasty: Development and Validation of an Artificial Neural Network Model

BackgroundThe objective is to develop and validate an artificial neural network (ANN) that learns and predicts length of stay (LOS), inpatient charges, and discharge disposition before primary total knee arthroplasty (TKA). The secondary objective applied the ANN to propose a risk-based, patient-specific payment model (PSPM) commensurate with case complexity.MethodsUsing data from 175,042 primary TKAs from the National Inpatient Sample and an institutional database, an ANN was developed to predict LOS, charges, and disposition using 15 preoperative variables. Outcome metrics included accuracy and area under the curve for a receiver operating characteristic curve. Model uncertainty was stratified by All Patient Refined comorbidity indices in establishing a risk-based PSPM.ResultsThe dynamic model demonstrated “learning” in the first 30 training rounds with areas under the curve of 74.8%, 82.8%, and 76.1% for LOS, charges, and discharge disposition, respectively. The PSPM demonstrated that as patient comorbidity increased, risk increased by 2.0%, 21.8%, and 82.6% for moderate, major, and severe comorbidities, respectively.ConclusionOur deep learning model demonstrated “learning” with acceptable validity, reliability, and responsiveness in predicting value metrics, offering the ability to preoperatively plan for TKA episodes of care. This model may be applied to a PSPM proposing tiered reimbursements reflecting case complexity.     

Predictors of Patient Satisfaction Following Primary Total Knee Arthroplasty: Results from a Traditional Statistical Model and a Machine Learning Algorithm

BackgroundIt is well-documented in the orthopedic literature that 1 in 5 patients are dissatisfied following total knee arthroplasty (TKA). However, multiple statistical models have failed to explain the causes of dissatisfaction. Furthermore, payers are interested in using patient-reported satisfaction scores to adjust surgeon reimbursement rates without a full understanding of the influencing parameters. The purpose of this study was to more comprehensively identify predictors of satisfaction and compare results using both a statistical model and a machine learning (ML) algorithm.MethodsA retrospective review of consecutive TKAs performed by 2 surgeons was conducted. Identical perioperative protocols were utilized by both surgeons. Patients were grouped as satisfied or unsatisfied based on self-reported satisfaction scores. Fifteen variables were correlated with satisfaction using binary logistic regression and stochastic gradient boosted ML models.ResultsIn total, 1325 consecutive TKAs were performed. After exclusions, 897 TKAs were available with minimum 1-year follow-up. Overall, 85.3% of patients were satisfied. Older age generation and performing surgeon were predictors of satisfaction in both models. The ML model also retained cruciate-retaining/condylar-stabilizing implant; lack of inflammatory conditions, preoperative narcotic use, depression, and lumbar spine pain; female gender; and a preserved posterior cruciate ligament as predictors of satisfaction which allowed for a significantly higher area under the receiver operator characteristic curve compared to the binary logistic regression model (0.81 vs 0.60).ConclusionFindings indicate that patient satisfaction may be multifactorial with some factors beyond the scope of a surgeon’s control. Further study is warranted to investigate predictors of patient satisfaction particularly with awareness of differences in results between traditional statistical models and ML algorithms.Level of EvidenceTherapeutic Level III.     

Development and Validation of a Machine Learning Algorithm After Primary Total Hip Arthroplasty: Applications to Length of Stay and Payment Models

Background

Value-based payment programs in orthopedics, specifically primary total hip arthroplasty (THA), present opportunities to apply forecasting machine learning techniques to adjust payment models to a specific patient or population. The objective of this study is to (1) develop and validate a machine learning algorithm using preoperative big data to predict length of stay (LOS) and patient-specific inpatient payments after primary THA and (2) propose a risk-adjusted patient-specific payment model (PSPM) that considers patient comorbidity.

Patient-Specific Versus Conventional Instrumentation for Total Knee Arthroplasty: Peri-Operative and Cost Differences

The role of patient-specific instrumentation in total knee arthroplasty (TKA) is yet to be clearly defined. Current evidence evaluating peri-operative and cost differences against conventional TKA is unclear. We reviewed 356 TKAs between July 2008 and April 2013; 306 TKAs used patient-specific instrumentation while 50 had conventional instrumentation. The patient-specific instrumentation cohort averaged 20.4 min less surgical time (P < 0.01) and had a 42% decrease in operating room turnover time (P = 0.022). At our institution, the money saved through increased operating room efficiency offset the cost of the custom cutting blocks and pre-operative advanced imaging. Routine use of patient-specific TKA can be performed with less surgical time, no increase in peri-operative morbidity, and at no increased cost when compared to conventional TKA.     

Outcomes of a First Total Knee Arthroplasty Are Associated With Outcomes of the Subsequent Contralateral Total Knee Arthroplasty

BackgroundTo determine if preoperative characteristics and postoperative outcomes of a first total knee arthroplasty (TKA) were predictive of characteristics and outcomes of the subsequent contralateral TKA in the same patient.MethodsRetrospective administrative claims data from (SPARCS) database were analyzed for patients who underwent sequential TKAs from September 2015 to September 2017 (n = 5,331). Hierarchical multivariable Poisson regression (length of stay [LOS]) and multivariable logistic regression (all other outcomes), controlling for sex, age, and Elixhauser comorbidity scores were performed.ResultsThe cohort comprised 65% women, with an average age of 66 years and an average duration of 7.3 months between surgeries (SD: 4.7 months). LOS was significantly shorter for the second TKA (2.6 days) than for the first TKA (2.8 days; P < .001). Patients discharged to a facility after their first TKA had a probability of 76% of discharge to facility after the second TKA and were significantly more likely to be discharged to a facility compared with those discharged home after the first TKA (odds ratio [OR]: 63.7; 95% confidence interval [CI]: 52.1-77.8). The probability of a readmission at 30 and 90 days for the second TKA if the patient was readmitted for the first TKA was 1.0% (OR: 3.70; 95% CI: 0.98-14.0) and 6.4% (OR: 9; 95% CI: 5.1-16.0), respectively. Patients with complications after their first TKA had a 27% probability of a complication after the second TKA compared with a 1.6% probability if there was no complication during the first TKA (OR: 14.6; 95% CI: 7.8.1-27.2).ConclusionThe LOS, discharge disposition, 90-day readmission rate, and complication rate for a second contralateral TKA are strongly associated with the patient’s first TKA experience. The second surgery was found to be associated with an overall shorter LOS, fewer readmissions, and higher likelihood of home discharge.Level of EvidenceLevel 3-retrospective cohort study.     

Preoperative Prediction of Value Metrics and a Patient-Specific Payment Model for Primary Total Hip Arthroplasty: Development and Validation of a Deep Learning Model

BackgroundThe primary objective was to develop and test an artificial neural network (ANN) that learns and predicts length of stay (LOS), inpatient charges, and discharge disposition for total hip arthroplasty. The secondary objective was to create a patient-specific payment model (PSPM) accounting for patient complexity.MethodsUsing 15 preoperative variables from 78,335 primary total hip arthroplasty cases for osteoarthritis from the National Inpatient Sample and our institutional database, an ANN was developed to predict LOS, charges, and disposition. Validity metrics included accuracy and area under the curve of the receiver operating characteristic curve. Predictive uncertainty was stratified by All Patient Refined comorbidity cohort to establish the PSPM.ResultsThe dynamic model demonstrated “learning” in the first 30 training rounds with areas under the curve of 82.0%, 83.4%, and 79.4% for LOS, charges, and disposition, respectively. The proposed PSPM established a risk increase of 2.5%, 8.9%, and 17.3% for moderate, major, and severe comorbidities, respectively.ConclusionThe deep learning ANN demonstrated “learning” with good reliability, responsiveness, and validity in its prediction of value-centered outcomes. This model can be applied to implement a PSPM for tiered payments based on the complexity of the case.     

The Cost of an Episode of Care in a Total Knee Arthroplasty Patient Is More Than a Total Hip Arthroplasty Patient Within an Alternative Payment Model

BackgroundAlternative payment models were set up to increase the value of care for total joint arthroplasty. Currently, total knee arthroplasty (TKA) and total hip arthroplasty (THA) are reimbursed within the same bundle. We sought to determine whether it was appropriate for these cases to be included within the same bundle.MethodsThe data were collected from consecutive patients in a bundled payment program at a single large academic institution. All payments for 90 days postoperatively were included in the episode of care. Readmission rates, demographics, and length of stay were collected for each episode of care.ResultsThere was a significant difference in cost of episode of care between TKA and THA, with the average TKA episode-of-care cost being higher than the average THA episode-of-care cost ($25803 vs $23805, P < .0001). There was a statistically significant difference between the 2 groups between gender, race, medical complexity, disposition outcome, and length of stay. The TKA group trended toward a lower readmission rate (5.3%) compared to the THA group (6.6%).ConclusionThe cost of an episode of care for patients within the bundled payment model is significantly higher for patients undergoing TKA compared with those undergoing a THA. This should be taken into consideration when determining payment plans for patients in alternative payment plans, along with other aspects of risk that need to be considered in order to allow for hospitals to be successful under the bundled payment model.     

Increased Patient-Level Payment After Removal of Total Knee Arthroplasty From the Inpatient-Only List

BackgroundIn January 2018, the Centers for Medicare and Medicaid Services removed total knee arthroplasty (TKA) from the Inpatient Only (IPO) list. This study aimed to compare patient-level payments in TKA cases with a length of stay (LOS) <2 midnights before and after removal of TKA from IPO list.MethodsIn this retrospective cohort study, all Medicare patients who received a primary elective TKA from 2016-2019 with a LOS <2 midnights at an academic tertiary center were identified. Total and itemized charges and patient-level payments were compared between eligible TKA cases performed in 2016-2017 and those in 2018-2019. There were 351 eligible TKA cases identified: 151 in 2016-2017 and 200 in 2018-2019.ResultsThe percentage of patients making any out-of-pocket payment increased in 2018-2019 from 2016-2017 (51.0% versus 10.6%), as did median patient-level payment ($7.30 [range, $0.00-$3,389] versus $0.00 [range, $0.00-$1,248], P < .001 for both). A greater proportion of patients in 2018-2019 paid $1-$50 than in 2016-2017 (37.5% versus 1.3%, P < .001) with no change in the proportion of patients who made payments >$50. Total charges were less in 2018-2019 than in 2016-2017 (P = .001). Charges for drugs, laboratory tests, admissions/floor, and therapies decreased in 2018-2019, whereas charges for the operating room and radiology increased (P < .001 for all).ConclusionPatients receiving outpatient TKA in 2018-2019 were more likely to have out-of-pocket payments than patients with comparable hospital stay who were designated as inpatients, although most of these payments were less than $50.     

Development of a Novel,Potentially Universal Machine Learning Algorithm for Prediction of Complications After Total Hip Arthroplasty

BackgroundAs the prevalence of hip osteoarthritis increases, the number of total hip arthroplasty (THA) procedures performed is also projected to increase. Accurately risk-stratifying patients who undergo THA would be of great utility, given the significant cost and morbidity associated with developing perioperative complications. We aim to develop a novel machine learning (ML)-based ensemble algorithm for the prediction of major complications after THA, as well as compare its performance against standard benchmark ML methods.MethodsThis is a retrospective cohort study of 89,986 adults who underwent primary THA at any California-licensed hospital between 2015 and 2017. The primary outcome was major complications (eg infection, venous thromboembolism, cardiac complication, pulmonary complication). We developed a model predicting complication risk using AutoPrognosis, an automated ML framework that configures the optimally performing ensemble of ML-based prognostic models. We compared our model with logistic regression and standard benchmark ML models, assessing discrimination and calibration.ResultsThere were 545 patients who had major complications (0.61%). Our novel algorithm was well-calibrated and improved risk prediction compared to logistic regression, as well as outperformed the other four standard benchmark ML algorithms. The variables most important for AutoPrognosis (eg malnutrition, dementia, cancer) differ from those that are most important for logistic regression (eg chronic atherosclerosis, renal failure, chronic obstructive pulmonary disease).ConclusionWe report a novel ensemble ML algorithm for the prediction of major complications after THA. It demonstrates superior risk prediction compared to logistic regression and other standard ML benchmark algorithms. By providing accurate prognostic information, this algorithm may facilitate more informed preoperative shared decision-making.     

A Validated Pre-operative Risk Prediction Tool for Extended Inpatient Length of Stay Following Primary Total Hip or Knee Arthroplasty

BackgroundAs value-based reimbursement models mature, understanding the potential trade-off between inpatient lengths of stay and complications or need for costly postacute care becomes more pressing. Understanding and predicting a patient’s expected baseline length of stay may help providers understand how best to decide optimal discharge timing for high-risk total joint arthroplasty (TJA) patients.MethodsA retrospective review was conducted of 37,406 primary total hip (17,134, 46%) and knee (20,272, 54%) arthroplasties performed at two high-volume, geographically diverse, tertiary health systems during the study period. Patients were stratified by 3 binary outcomes for extended inpatient length of stay: 72 + hours (29%), 4 + days (11%), or 5 + days (5%). The predictive ability of over 50 sociodemographic/comorbidity variables was tested. Multivariable logistic regression models were created using institution #1 (derivation), with accuracy tested using the cohort from institution #2 (validation).ResultsDuring the study period, patients underwent an extended length of stay with a decreasing frequency over time, with privately insured patients having a significantly shorter length of stay relative to those with Medicare (1.9 versus 2.3 days, P < .0001). Extended stay patients also had significantly higher 90-day readmission rates (P < .0001), even when excluding those discharged to postacute care (P < .01). Multivariable logistic regression models created from the training cohort demonstrated excellent accuracy (area under the curve (AUC): 0.755, 0.783, 0.810) and performed well under external validation (AUC: 0.719, 0.743, 0.763). Many important variables were common to all 3 models, including age, sex, American Society of Anesthesiologists (ASA) score, body mass index, marital status, bilateral case, insurance type, and 13 comorbidities.DiscussionAn online, freely available, preoperative clinical decision tool accurately predicts risk of extended inpatient length of stay after TJA. Many risk factors are potentially modifiable, and these validated tools may help guide clinicians in preoperative patient counseling, medical optimization, and understanding optimal discharge timing.     

Logistic Regression and Machine Learning Models Cannot Discriminate Between Satisfied and Dissatisfied Total Knee Arthroplasty Patients

BackgroundApproximately 20% of total knee arthroplasty (TKA) patients are found to be dissatisfied or unsure of their satisfaction at 1-year post-surgery. This study attempted to predict 1-year post-surgery dissatisfied/unsure TKA patients with pre-surgery and surgical variables using logistic regression and machine learning methods.MethodsA retrospective analysis of patients who underwent primary TKA for osteoarthritis between 2012 and 2016 at a single institution was completed. Patients were split into satisfied and dissatisfied/unsure groups. Potential predictor variables included the following: demographic information, patella re-surfaced, posterior collateral ligament sacrificed, and subscales from the Knee Society Knee Scoring System, the Knee Society Clinical Rating System, the Western Ontario and McMaster Universities Osteoarthritis Index, and the 12-Item Short Form Health Survey version 2. Logistic regression and 6 different machine learning methods were used to create prediction models. Model performance was evaluated using discrimination (AUC [area under the receiver operating characteristic curve]) and calibration (Brier score, Cox intercept, and Cox slope) metrics.ResultsThere were 1432 eligible patients included in the analysis, 313 were considered to be dissatisfied/unsure. When evaluating discrimination, the logistic regression (AUC = 0.736) and extreme gradient boosted tree (AUC = 0.713) models performed best. When evaluating calibration, the logistic regression (Brier score = 0.141, Cox intercept = 0.241, and Cox slope = 1.31) and gradient boosted tree (Brier score = 0.149, Cox intercept = 0.054, and Cox slope = 1.158) models performed best.ConclusionThe models developed in this study do not perform well enough as discriminatory tools to be used in a clinical setting. Further work needs to be done to improve the performance of pre-surgery TKA dissatisfaction prediction models.     

Artificial Intelligence to Identify Arthroplasty Implants From Radiographs of the Knee

BackgroundRevisions and reoperations for patients who have undergone total knee arthroplasty (TKA), unicompartmental knee arthroplasty (UKA), and distal femoral replacement (DFR) necessitates accurate identification of implant manufacturer and model. Failure risks delays in care, increased morbidity, and further financial burden. Deep learning permits automated image processing to mitigate the challenges behind expeditious, cost-effective preoperative planning. Our aim was to investigate whether a deep-learning algorithm could accurately identify the manufacturer and model of arthroplasty implants about the knee from plain radiographs.MethodsWe trained, validated, and externally tested a deep-learning algorithm to classify knee arthroplasty implants from one of 9 different implant models from retrospectively collected anterior-posterior (AP) plain radiographs from four sites in one quaternary referral health system. The performance was evaluated by calculating the area under the receiver-operating characteristic curve (AUC), sensitivity, specificity, and accuracy when compared with a reference standard of implant model from operative reports.ResultsThe training and validation data sets were comprised of 682 radiographs across 424 patients and included a wide range of TKAs from the four leading implant manufacturers. After 1000 training epochs by the deep-learning algorithm, the model discriminated nine implant models with an AUC of 0.99, accuracy 99%, sensitivity of 95%, and specificity of 99% in the external-testing data set of 74 radiographs.ConclusionsA deep learning algorithm using plain radiographs differentiated between 9 unique knee arthroplasty implants from four manufacturers with near-perfect accuracy. The iterative capability of the algorithm allows for scalable expansion of implant discriminations and represents an opportunity in delivering cost-effective care for revision arthroplasty.     

Variation of Medicare Payments for Total Knee Arthroplasty

We analyzed the 2009 Medicare inpatient claims data and other databases to estimate Medicare payments for primary or revision total knee arthroplasty (TKA). The average Medicare hospital payment per procedure was $13,464 for primary TKA (n = 227,587) and $17,331 for revision TKA (n = 18,677). For both primary and revision TKAs Medicare payments varied substantially across patients, hospitals and healthcare markets. Less than one percent of primary TKA cases but seven percent of revision TKA cases triggered Medicare “outlier” payments, which were $10,000 or higher per case beyond regular diagnosis-related-group payments. Urban and major teaching hospitals were more likely to treat these unusually expensive cases. Hospitals in the Northeast and West regions tended to receive higher Medicare payments than hospitals in the Midwest.     

Internal Validation of a Predictive Model for Satisfaction After Primary Total Knee Arthroplasty

Background

As the number of total knee arthroplasty (TKA) procedures continues to rise in the context of bundled payment models, patients dissatisfied postoperatively that require additional care will impose additional cost to the healthcare system. The purpose of this study is to internally validate a predictive model for postoperative patient satisfaction after TKA.

Outcomes of Total Knee Arthroplasty Revisions in Obese and Morbidly Obese Patient Populations

BackgroundThe obese population is at higher risk for complications following primary total knee arthroplasty (TKA), but little data is available regarding revision outcomes. This study aimed to investigate the role of body mass index (BMI) in the cause for revision TKA and whether BMI classification is predictive of outcomes.MethodsA multi-institutional database was generated, including revision TKAs from 2012 to 2019. Data collection included demographics, comorbidities, surgery types (primary revision, repeat revision), reasons for revision, lengths of hospital stay, and surgical times. Patients were compared using 3 BMI categories: nonobese (18.5 to 29.9), obese (30 to 39.9), and morbidly obese (≥40). Categorical and continuous variables were analyzed using chi-square and 1-way analysis of variance tests, respectively. Regression analyses were used to compare reasons for revision among weight classes.ResultsObese and morbidly obese patients showed significant risk for repeat revision surgery in comparison to normal weight patients. Obese patients were at higher risk for primary revision due to stiffness/fibrosis and repeat revision due to malposition. In comparison to the obese population, morbidly obese patients were more likely to require primary revision for dislocation and implant loosening.ConclusionSignificant differences in primary and repeat revision etiologies exist among weight classes. Furthermore, obese and morbidly obese patients have a greater risk of requiring repeat revision surgery. These patients should be informed of their risk for multiple operations, and surgeons should be aware of the differences in revision etiologies when anticipating complications following primary TKA.     

Readmission,Reoperation, and Complications: Total Hip vs Total Knee Arthroplasty

Background

Total hip arthroplasty (THA) and total knee arthroplasty (TKA) are currently grouped under the same Diagnosis-Related Group (DRG). With the introduction of bundled payments, providers are accountable for all the costs incurred during the episode of care, including the costs of readmissions and management of complications. However, it is unclear whether readmission rates and short-term complications are similar in primary THA and TKA.

Rotating Hinge Implants for Complex Primary and Revision Total Knee Arthroplasty

Background

Contemporary rotating hinge knee (RHK) prosthesis has shown improved survival rates over earlier generations. However, reports of high complication and mechanical failure rates highlight the need for more clinical outcome data in the complex primary and revision setting. The purpose of this study is to report our results of using a contemporary rotating hinge for complex primary and revision total knee arthroplasty.

Improved Radiographic Outcomes With Patient-Specific Total Knee Arthroplasty

Patient-specific guides can improve limb alignment and implant positioning in total knee arthroplasty, although not all studies have supported this benefit. We compared the radiographs of 100 consecutively-performed patient-specific total knees to a similar group that was implanted with conventional instruments instead. The patient-specific group showed more accurate reproduction of the theoretically ideal mechanical axis, with fewer outliers, but implant positioning was comparable between groups. Our odds ratio comparison showed that the patient-specific group was 1.8 times more likely to be within the desired + 3° from the neutral mechanical axis when compared to the standard control group. Our data suggest that reliable reproduction of the limb mechanical axis may accrue from patient-specific guides in total knee arthroplasty when compared to standard, intramedullary instrumentation.