Machine-learning vs. logistic regression for preoperative prediction of medical morbidity after fast-track hip and knee arthroplasty-a comparative study.

BACKGROUND: Machine-learning models may improve prediction of length of stay (LOS) and morbidity after surgery. However, few studies include fast-track programs, and most rely on administrative coding with limited follow-up and information on perioperative care. This study investigates potential benefits of a machine-learning model for prediction of postoperative morbidity in fast-track total hip (THA) and knee arthroplasty (TKA). METHODS: Cohort study in consecutive unselected primary THA/TKA between 2014-2017 from seven Danish centers with established fast-track protocols. Preoperative comorbidity and prescribed medication were recorded prospectively and information on length of stay and readmissions was obtained through the Danish National Patient Registry and medical records. We used a machine-learning model (Boosted Decision Trees) based on boosted decision trees with 33 preoperative variables for predicting "medical" morbidity leading to LOS > 4 days or 90-days readmissions and compared to a logistical regression model based on the same variables. We also evaluated two parsimonious models, using the ten most important variables in the full machine-learning and logistic regression models. Data collected between 2014-2016 (n:18,013) was used for model training and data from 2017 (n:3913) was used for testing. Model performances were analyzed using precision, area under receiver operating (AUROC) and precision recall curves (AUPRC), as well as the Mathews Correlation Coefficient. Variable importance was analyzed using Shapley Additive Explanations values. RESULTS: Using a threshold of 20% "risk-patients" (n:782), precision, AUROC and AUPRC were 13.6%, 76.3% and 15.5% vs. 12.4%, 74.7% and 15.6% for the machine-learning and logistic regression model, respectively. The parsimonious machine-learning model performed better than the full logistic regression model. Of the top ten variables, eight were shared between the machine-learning and logistic regression models, but with a considerable age-related variation in importance of specific types of medication. CONCLUSION: A machine-learning model using preoperative characteristics and prescriptions slightly improved identification of patients in high-risk of "medical" complications after fast-track THA and TKA compared to a logistic regression model. Such algorithms could help find a manageable population of patients who may benefit most from intensified perioperative care.

Short-term surgical complications following fast-track medial unicompartmental knee arthroplasty.

Aims: Medial unicompartmental knee arthroplasty (mUKA) is an advised treatment for anteromedial knee osteoarthritis. While long-term survival after mUKA is well described, reported incidences of short-term surgical complications vary and the effect of surgical usage on complications is less established. We aimed to describe the overall occurrence and treatment of surgical complications within 90 days of mUKA, as well as occurrence in high-usage centres compared to low-usage centres. Methods: mUKAs performed in eight fast-track centres from February 2010 to June 2018 were included from the Lundbeck Foundation Centre for Fast-track Hip and Knee Replacement Database. All readmissions within 90 days of surgery underwent chart review and readmissions related to the surgical wound or the prosthesis were recorded. Centres were categorized as high-usage centres when using mUKA in ≥ 20% of annual knee arthroplasties. The occurrence of complications between high- and low-usage centres were compared using Fisher's exact test. Results: We included 3,757 mUKAs: 2,377 mUKAs from high-usage centres and 1,380 mUKAs from low-usage centres. Surgical complications within 90 days occurred in 69 cases (1.8%), 45 (1.9%) in high-usage centres and 24 (1.7%) in low-usage centres (odds ratio (OR) 1.1 (95% confidence interval (CI) 0.65 to 1.8)). The most frequent complications were periprosthetic joint infections (PJIs) (n = 18; 0.48%), wound-related issues (n = 14; 0.37%), and periprosthetic fractures (n = 13; 0.35%). Bearing dislocations (n = 7; 0.19%) occurred primarily in procedures from high-usage centres. In high-usage centres, seven periprosthetic fractures (0.29%) occurred compared to six (0.43%) in low-usage centres (OR 0.68 (95% CI 0.20 to 2.0)). In high-usage centres, nine PJIs (0.38%) occurred compared to nine (0.65%) in low-usage centres (OR 0.58 (95% CI 0.22 to 1.6)). Conclusion: Surgical complications are rare after fast-track mUKA surgery and with no difference in overall occurrence of surgical complications between high- and low-usage centres, although the risk of some specific surgical complications may favour high-usage centres.

Gastrointestinal complications after fast-track total hip and knee replacement: an observational study in a consecutive 36,932 patient cohort.

INTRODUCTION: Gastrointestinal complications after total hip (THA) and knee arthroplasty (TKA) have been reported to be between 0.3 and 2.6% with bleeding and C. difficile infection in 0-1%, and 0.1-1.7%, respectively. The use of enhanced recovery or "fast-track" protocols have focused on optimizing all aspects of perioperative care resulting in reduced length of hospital stay (LOS) and potentially also gastrointestinal complications. This study is a detailed analysis on the occurrence of postoperative gastrointestinal complications resulting in increased hospital stay or readmissions in a large consecutive cohort of fast-track THA and TKA with complete 90 days follow-up. MATERIALS AND METHODS: This is an observational study on a consecutive cohort of primary unilateral THAs and TKAs performed between January 2010 and August 2017 in nine Danish high-volume fast-track centers. Discharge summaries and relevant patient records were reviewed in patients with readmissions within 90 days or LOS > 4 days caused by gastrointestinal complications. RESULTS: The cohort included 36,932 patients with 58.3% females and 54.1% THAs. Mean age and BMI were 68 years and 28. Median postoperative LOS was 2 days. Only n: 276 (0.75 %) had a LOS > 4 days or a readmission within 90 days due to a gastrointestinal complication (CI 0.67%-0.84%). Of these, only 34 (0.09%) were graded as severe ileus or gastrointestinal bleeding. CONCLUSIONS: The risk of GI-complications within the first 90 postoperative days after fast-track THA and TKA was low (0.75%).

Peripheral nerve-blocks and associations with length of stay and readmissions in fast-track total hip and knee arthroplasty.

BACKGROUND: Peripheral nerve blocks (PNB) have recently been recommended in total hip (THA) and knee (TKA) arthroplasty as they may reduce pain, morphine consumption, length of stay (LOS) and complications. However, whether PNBs are associated with early discharge within an enhanced recovery protocol including multimodal analgesia is uncertain. METHODS: An observational multicenter study from January to August 2017 in six Danish Arthroplasty Centers with established fast-track protocols. Prospective recording of preoperative characteristics and information on PNB, LOS and readmissions through the Danish National Patient Registry and medical records. Multiple logistic regression was used to investigate associations between PNB and a LOS >1 day, LOS >4 days, and 30-days readmissions. We also reported on mobilization, pain, opioid and fall-related complications leading to LOS >4 days or readmissions. RESULTS: A total of 2027 (58.6%) THA and 1432 (41.4%) TKAs with a median LOS of 1 day (IQR 1-2) and 5.3% (CI:4.6-6.1) 30-days readmission rate were identified. PNB was used in 40.7% (CI:38.2-43.3) of TKA and 2.7% (CI:2.0-3.5) of THA, but with considerable interdepartmental variation (0.0-89.0% for TKA). There was no association between PNB and LOS >1 day (OR:1.19 CI:0.82-1.72; p = .354), LOS >4 days (OR:1.4 CI:0.68-2.89; p = .359) or 30-days readmissions (OR:1.02 CI:0.63-1.65; p = .935) in TKA. Logistic regression in THA was not possible due to limited use of PNB. In TKA there were 12 (2.1% CI:1.2-3.6) with and 1 (0.1% CI:0.02-0.7) without a PNB, who had mobilization, pain or opioid-related complications, and 5 (0.9% CI:0.4-2.0) versus 4 (0.5% CI:0.2-1.2) who fell. Correspondingly, 2 (3.7% CI:1.0-12.6) and 11 (0.6% CI:0.3-1.0) of THA patients had these complications, while 0 (0.0% CI:0.0-6.6) and 17 (0.8% CI:0.5-1.3) fell. CONCLUSION: Routine use of peripheral nerve blocks was not associated with early discharge or 30-days readmissions in fast-track THA and TKA. Future studies should focus on benefits of PNB in high-risk patients.

Machine learning classifiers do not improve prediction of hospitalization > 2 days after fast-track hip and knee arthroplasty compared with a classical statistical risk model.

Background and purpose: Prediction of postoperative outcomes and length of hospital stay (LOS) of patients is vital for allocation of healthcare resources. We investigated the performance of prediction models based on machinelearning algorithms compared with a previous risk stratification model using traditional multiple logistic regression, for predicting the risk of a LOS of > 2 days after fast-track total hip and knee replacement. Patients and methods: 3 different machine learning classifiers were trained on data from the Lundbeck Centre for Fast-track Hip and Knee Replacement Database (LCDB) collected from 9,512 patients between 2016 and 2017. The chosen classifiers were a random forest classifier (RF), a support vector machine classifier with a polynomial kernel (SVM), and a multinomial Naïve-Bayes classifier (NB). Results: Comparing performance measures of the classifiers with the traditional model revealed that all the models had a similar performance in terms of F1 score, accuracy, sensitivity, specificity, area under the receiver operating curve (AUC), and area under the precision-recall curve (AUPRC). A feature importance analysis of the RF classifier found hospital, age, use of walking aid, living alone, and joint operated on to be the most relevant input features. None of the classifiers reached a clinically relevant performance with the input data from the LCDB. Interpretation: Despite the promising prospects of machine-learning practices for disease and risk prediction, none of the machine learning models tested outperformed the traditional multiple regression model in predicting which patients in this cohort had a LOS > 2 days.

Postoperative Morbidity and Mortality in Diabetic Patients After Fast-Track Hip and Knee Arthroplasty: A Prospective Follow-up Cohort of 36,762 Procedures.

BACKGROUND: Diabetes increases the risk of adverse outcomes in surgical procedures, including total hip and knee arthroplasty (THA/TKA), and the prevalence of diabetic patients undergoing these procedures is high, ranging from approximately 8% to 20%. However, there is still a need to clarify the role of diabetes and antihyperglycemic treatment in a fast-track THA/TKA setting, which otherwise may decrease morbidity. Consequently, we investigated the association between diabetes and antihyperglycemic treatment on length of stay (LOS) and complications following fast-track THA/TKA within a multicenter fast-track collaboration. METHODS: We used an observational study design on data from a prospective multicenter fast-track collaboration on unselected elective primary THA/TKA from 2010 to 2017. Complete follow-up (>99%) was achieved through The Danish National Patient Registry, antihyperglycemic treatment established through the Danish National Database of Reimbursed Prescriptions and types of complications leading to LOS >4 days, 90-day readmission or mortality obtained by scrutinizing health records and discharge summaries. Patients were categorized as nondiabetic and if diabetic into insulin-, orally, and dietary-treated diabetic patients. RESULTS: A total of 36,762 procedures were included, of which 837 (2.3%) had insulin-treated diabetes, 2615 (7.1%) orally treated diabetes, and 566 (1.5%) dietary-treated diabetes. Overall median LOS was 2 (interquartile range [IQR]: 1-3) days, and mean LOS was 2.4 (standard deviation [SD], 2.5) days. The proportion of patients with LOS >4 days was 6.0% for nondiabetic patients, 14.7% for insulin-treated, 9.4% for orally treated, and 9.5% for dietary-treated diabetic patients. Pharmacologically treated diabetes (versus nondiabetes) was independently associated with increased odds of LOS >4 days (insulin-treated: odds ratio [OR], 2.2 [99.6% confidence interval {CI}, 1.3-3.7], P < .001; orally treated: OR, 1.5 [99.6% CI, 1.0-2.1]; P = .002). Insulin-treated diabetes was independently associated with increased odds of "diabetes-related" morbidity (OR, 2.3 [99.6% CI, 1.2-4.2]; P < .001). Diabetic patients had increased renal complications regardless of antihyperglycemic treatment, but only insulin-treated patients suffered significantly more cardiac complications than nondiabetic patients. There was no increase in periprosthetic joint infections or mortality associated with diabetes. CONCLUSIONS: Patients with pharmacologically treated diabetes undergoing fast-track THA/TKA were at increased odds of LOS >4 days. Although complication rates were low, patients with insulin-treated diabetes were at increased odds of postoperative complications compared to nondiabetic patients and to their orally treated counterparts. Further investigation into the pathogenesis of postoperative complications differentiated by antihyperglycemic treatment is needed.

Influence of day of surgery and prediction of LOS > 2 days after fast-track hip and knee replacement.

Background and purpose - Enhanced recovery programs have reduced length of stay (LOS) after hip and knee arthroplasty (THA/TKA). Although risk factors disposing to prolonged LOS are well documented, there is limited information on the role of weekday of surgery. This study analyzed the role of weekday of surgery and other potential risk factors for LOS > 2 days.Patients and methods - We included 10,576 unselected consecutive procedures between January 2016 and August 2017 within a multicenter fast-track THA/TKA collaboration with prospective collection of preoperative characteristics. We used multiple regression analysis of potential risk factors for LOS > 2 days followed by construction of a simple risk score from 0 to 15 points based on the calculated odds ratios.Results - Mean LOS was 1.9 (SD 1.8) days, with 80% of patients having surgery from Monday to Wednesday. Of these, 17% (95% CI 16-18) had a LOS > 2 days vs. 19% (CI 17-21) in those operated on Thursday and Friday. Patients were scheduled evenly throughout the week regardless of risk of LOS > 2 days and despite the fact that 38% (CI 35-40) of patients with ≥ 6 points (16% of the total population) had a LOS > 2 days compared with 14% (CI 13-14) in those with < 6 points. In these "high-risk" patients, the fraction with LOS > 2 days increased when having surgery on Thursdays or Fridays (43% CI 38-49) compared with Monday to Wednesday (37% CI 34-39).Interpretation - A detailed preoperative risk assessment may be helpful to plan the weekday of surgery in order to decrease LOS and weekend hospitalization.

Unicompartmental knee arthroplasty undertaken using a fast-track protocol.

AIMS: The aim of this prospective multicentre study was to describe trends in length of stay and early complications and readmissions following unicompartmental knee arthroplasty (UKA) performed at eight different centres in Denmark using a fast-track protocol and to compare the length of stay between centres with high and low utilization of UKA. METHODS: We included data from eight dedicated fast-track centres, all reporting UKAs to the same database, between 2010 and 2018. Complete ( > 99%) data on length of stay, 90-day readmission, and mortality were obtained during the study period. Specific reasons for a length of stay of > two days, length of stay > four days, and 30- and 90-day readmission were recorded. The use of UKA in the different centres was dichotomized into ≥ 20% versus < 20% of arthroplasties which were undertaken being UKAs, and ≥ 52 UKAs versus < 52 UKAs being undertaken annually. RESULTS: A total of 3,927 procedures were included. Length of stay (mean 1.1 days (SD 1.1), median 1 (IQR 0 to 1)) was unchanged during the study period. The proportion of procedures with a length of stay > two days was also largely unchanged during this time. The percentage of patients discharged on the day of surgery varied greatly between centres (0% to 50% (0 to 481)), with centres with high UKA utilization (both usage and volume) having a larger proportion of same-day discharges. The 30- and 90-day readmissions were 166 (4.2%) and 272 (6.9%), respectively; the 90-day mortality was 0.08% (n = 3). CONCLUSION: Our findings suggest general underutilization of the potential for quicker recovery following UKA in a fast-track setup. Cite this article: Bone Joint J 2020;102-B(9):1167-1175.

Days alive and out of hospital after fast-track total hip and knee arthroplasty: an observational cohort study in 16 137 patients.

BACKGROUND: Days alive and out of hospital (DAH) has been proposed as a pragmatic outcome measure of surgical quality. However, there is a lack of procedure specific data or data within an optimised fast-track protocol. Furthermore, information about influence of follow-up duration and types of complications on DAH is limited. METHODS: Observational multicentre cohort study of patients undergoing fast-track total hip (THA) and knee arthroplasty (TKA). Prospective information on comorbidity and complete 90 days follow-up was undertaken through the Danish National Patient Register and chart review. RESULTS: For 16 137 procedures, of which 18.6% were high-risk (≥2 preoperative risk factors), the median length of stay was 2 days (inter-quartile range [IQR], 2-3), and 30- and 90-day readmission rates were 5.7% and 8.1%, respectively. Median DAH30 and DAH90 days were 27 (26-28) and 87 (85-88) vs 28 (27-28) and 88 (87-89) (P<0.001) in high-vs low-risk patients, respectively. The fraction with DAH ≤25 at 30 days and DAH ≤85 at 90 days was increased in high-vs low-risk patients: 23.3% vs 6.8% (odds ratio [OR]=4.16; 95% confidence interval [CI], 3.73-4.65) and 26.0% vs 8.6% (OR=3.75; 95% CI, 3.38-4.16). There were relatively fewer 'surgical' complications in high- vs low-risk patients with DAH30 ≤25 (14.6% vs 25.8%) (OR=0.49; 95% CI, 0.37-0.65) and DAH90 ≤85 (16.9% vs 31.89%) (OR=0.43; 95% CI, 0.34-0.56). About 2% of patients had readmissions, but DAH was >25 and >85 at 30 and 90 days after operation, respectively. CONCLUSION: Median DAH in fast-track THA/TKA patients is 28 at 30 days and 88 at 90 days after surgery. DAH in high-risk patients was only slightly reduced compared with low-risk patients, but they have relatively more 'medical' complications.

Incidence and Risk Factors for Stroke in Fast-Track Hip and Knee Arthroplasty-A Clinical Registry Study of 24,862 Procedures.

BACKGROUND: Postoperative stroke is a severe complication with a reported 30-day incidence of 0.4%-0.6% after total hip (THA) and knee arthroplasty (TKA). However, most data are based on diagnostic codes and with limited details on perioperative care, including the use of fast-track protocols. We investigated the incidence of and preoperative and postoperative factors for stroke after fast-track THA/TKA. METHODS: We used an observational study design of elective fast-track THA/TKA patients with prospective collection of comorbidity and complete 90-day follow-up. Medical records were evaluated for events potentially disposing to stroke. Identification of relevant preoperative risk factors was done by multivariable logistic regression. Incidence of stroke was compared with a Danish background population. RESULTS: Of 24,862 procedures with a median length of stay of 2 (interquartile range, 2-3) days, we found 27 (0.11%; 95% confidence interval [CI], 0.08%-0.16%) and 43 strokes (0.17%, 95% CI, 0.13%-0.23%) ≤30 and ≤90 days after surgery, respectively. Preoperative risk factors for stroke ≤30 days were age ≥ 85 years (odds ratio [OR], 4.3; 95% CI, 1.1-16.3) and anticoagulant treatment (OR, 3.1; 95% CI, 1.2-7.9). Preoperative anemia was near significant (OR, 2.1; 95% CI, 0.98-4.6, P = .055). Eight strokes ≤30 days were preceded by a cardiovascular event within the second postoperative day. Incidence of stroke after postoperative day 30 was similar to a Danish background population. CONCLUSION: Risk of postoperative stroke in fast-track THA and TKA was low but may be further reduced with increased focus on avoiding perioperative cardiovascular events and in patients with preoperative anticoagulants or anemia.

Myocardial infarction following fast-track total hip and knee arthroplasty-incidence, time course, and risk factors: a prospective cohort study of 24,862 procedures.

Background and purpose - Acute myocardial infarction (MI) is a leading cause of mortality following total hip and knee arthroplasty (THA/TKA). The reported 30-day incidence of MI varies from 0.3% to 0.9%. However, most data derive from administration and insurance databases or large RCTs with potential confounding factors. We studied the incidence of and potential modifiable risk factors for postoperative MI in a large, multicenter optimized "fast-track" THA/TKA setting. Patients and methods - A prospective cohort study was conducted on consecutive unselected elective primary unilateral THA and TKA, using prospective information on comorbidities and complete 90-day follow-up from the Danish National Patient Registry. Evaluation of discharge summaries and medical records was undertaken in cases of suspected MI. Logistic regression analyses were carried out for identification of preoperative risk factors. Results - Of 24,862 procedures with a median length of stay 2 (IQR 2-3) days, 30- and 90-day incidence of MI was 31 (0.12%) and 48 (0.19%). Preoperative risk factors for MI ≤30 days were age >85 years (OR 7.4, 95% CI 2.3-24), insulin-dependent diabetes mellitus (IDDM) (3.6, CI 1.1-12), cardiovascular disease (2.4, CI 1.1-5.0) and hypercholesterolemia (2.3, CI 1.1-5.1). Of 31 patients with MI ≤30 days 9 were treated with vasopressors for intraoperative hypotension and 27 had postoperative anemia. Interpretation - Fast-track THA and TKA had a low 30-day MI incidence. Focus on patients with age >85, IDDM, cardiovascular disease, and hypercholesterolemia may further reduce the 30-day incidence of MI. The role of postoperative anemia and intraoperative hypotension are other areas for further improvement.

Risk Factors for Post-treatment Complex Regional Pain Syndrome (CRPS): An Analysis of 647 Cases of CRPS from the Danish Patient Compensation Association.

OBJECTIVES: Complex regional pain syndrome is a challenging condition that includes a broad spectrum of sensory, autonomic, and motor features predominantly in extremities recovering from a trauma. Few large-scale studies have addressed occurrence of and factors associated with complex regional pain syndrome (CRPS) following orthopedic treatment. The present study aimed to identify factors associated with post-treatment development of CRPS. METHODS: Using the Danish Patient Compensation Association's database, we identified 647 patients claiming post-treatment CRPS between 1992 and 2015. Age, gender, initial diagnosis, treatment, and amount of compensation were extracted. Multivariate logistic regressions were performed to identify variables associated with approval of the claim. For carpal tunnel syndrome (CTS) patients, we registered whether symptoms were bilateral or unilateral and if neurophysiology prior to treatment was pathologic. RESULTS: The following ratios were found: women:men was 4:1, primary diagnosis to the upper limb:lower limb was 2.5:1, and surgical:nonsurgical treatment was 3:1. Mean age was 47.5 ± 13.7 years, and no intergender difference was detected. Antebrachial fracture (23%) and CTS (9%) were the most common primary conditions. Surgical treatment was associated with approval of the claim (odds ratio 3.5, 95% confidence interval 2.3 to 5.3; P < 0.001). Half of CTS patients had normal neurophysiology prior to surgery; among patients with unilateral symptoms, 71.4% had normal neurophysiology. CONCLUSIONS: Female gender, surgical treatment, and treatment to the upper limb were risk factors. Elective surgery accounted for a large number of post-treatment CRPS patients. In CTS patients developing CRPS, normal neurophysiological examination findings were common, and it could be suspected that these patients were suffering from an pre-clinical stage of CRPS, not CTS.