Figure - available from: BMC Musculoskeletal Disorders
This content is subject to copyright. Terms and conditions apply.
Source publication
Background:
Decreasing the length of hospital stay is an ideal course of action to appropriately allocate medical resources. The aim of this retrospective study was to identify perioperative factors that may decrease the length of hospital stay (LOS).
Methods:
In this study, we collected the data on 1112 patients who underwent primary total knee...
Citations
... The association between postoperative respiratory complications and longer surgical times has also been documented [14]. Similarly, it has been observed in a retrospective trial that prolonged operative time is associated with longer hospital stays [15]. ...
Background
Liver surgery is a major and challenging procedure for the surgeon, the anesthetist, and the patient. The objective of this study was to evaluate the postoperative nonhepatic complications of patients undergoing liver resection surgery with perioperative factors.
Methods
We retrospectively analyzed 79 patients who underwent liver resection surgeries at the Shaukat Khanum Memorial Cancer Hospital and Research Centre in Lahore, Pakistan, from July 2015 to December 2022.
Results
The mean age at the time of surgery was 53 years (range: 3-77 years), and the mean BMI was 26.43 (range: 15.72-38.0 kg/m2). Of the total patients, 44.3 % (n = 35) had no comorbidities, 26.6% (n=21) had one comorbidity, and 29.1% (n=23) had two or more comorbidities. Patients in whom the blood loss was more than 375 ml required postoperative oxygen inhalation with a significant relative risk of 2.6 (p=0.0392) and an odds ratio of 3.5 (p=0.0327). Similarly, patients who had a surgery time of more than five hours stayed in the hospital for more than seven days, with a statistically significant relative risk of 2.7 (p=0.0003) and odds
ratio of 7.64 (p=0.0001). The duration of surgery was also linked with the possibility of requiring respiratory support, with a relative risk of 5.0 (p=0.0134) and odds ratio of 5.73 (p=0.1190).
Conclusion
Patients in our cohort who had a prolonged duration of surgery received an increased amount of fluids, and a large volume of blood loss was associated with prolonged stay in the ICU (>2 days), hospital admission (>7 days), ICU readmission, and increased incidence of cardiorespiratory, neurological, and renal disturbances
postoperatively.
... The average length of inpatient hospital stay (LOS) following primary total knee arthroplasty (TKA) has reduced from 8.4 days to 2.4 days over the decades due to the successful implementation of fast-track protocols [1][2][3][4], which focuses on improving the healthcare coordination, perioperative patient management, and early ambulation after the surgery [5]. Reducing LOS was effective in controlling the overall hospital costs [6] as well as the risks of adverse events after surgery [7]. ...
... Reducing LOS would be one course of action to increase the accessibility of healthcare services and cost efficiency [10]. Although previous studies showed that prolonged LOS following primary TKA was predictable by a series of patient factors [3,5,[10][11][12][13][14], the challenge related to traditional statistical methods in predicting prolonged LOS following primary TKA is that they were unable to identify individuals predisposed to lengthened hospitalization from a source patient cohort [15]. Besides, statistical models do not routinely quantify the importance of a patient feature in influencing the risks of prolonged LOS. ...
... The database of the American College of Surgeons National Surgical Quality Improvement Program was queried for unilateral TKA procedures from 2013 to 2020 using the CPT code 27,447. Exclusion criteria were (1) emergency admission or revision surgery; (2) unknown, empty, or unclear information in any candidate patient feature category; (3) outliers (values that were considered physical extremes or obvious input errors) in a feature category (age > 100 years, BMI < 12 or > 100, white blood cell > 100 thousands/mm 3 , Hematocrit > 100%, platelet count > 800 thousands/mm 3 , and operation time > 800 min). The study protocol was approved by the institutional review board. ...
Introduction
The total length of stay (LOS) is one of the biggest determinators of overall care costs associated with total knee arthroplasty (TKA). An accurate prediction of LOS could aid in optimizing discharge strategy for patients in need and diminishing healthcare expenditure. The aim of this study was to predict LOS following TKA using machine learning models developed on a national-scale patient cohort.
Methods
The ACS-NSQIP database was queried to acquire 267,966 TKA cases from 2013 to 2020. Four machine learning models—artificial neural network (ANN), random forest, histogram-based gradient boosting, and k-nearest neighbor were trained and tested on the dataset for the prediction of prolonged LOS (LOS exceeded the 75th of all values in the cohort). The model performance was assessed by discrimination (area under the receiver operating characteristic curve [AUC]), calibration, and clinical utility.
Results
ANN delivered the best performance among the four models. ANN distinguished prolonged LOS in the study cohort with an AUC of 0.71 and accurately predicted the probability of prolonged LOS for individual patients (calibration slope: 0.82; calibration intercept: 0.03; Brier score: 0.089). All models demonstrated clinical utility by generating positive net benefits in decision curve analyses. Operation time, pre-operative transfusion, pre-operative laboratory tests (hematocrit, platelet count, and white blood cell count), and BMI were the strongest predictors of prolonged LOS.
Conclusion
ANN demonstrated modest discrimination capacity and excellent performance in calibration and clinical utility for the prediction of prolonged LOS following TKA. Clinical application of the machine learning models has the potential to improve care coordination and discharge planning for patients at high risk of extended hospitalization after surgery. Incorporating more relevant patient factors may further increase the models’ prediction strength.
... /frai. . 2020), schizophrenia (Kirchebner et al., 2020), knee arthroplasty (Song et al., 2020), COVID-19 (Vekaria et al., 2021;Etu et al., 2022;Zeleke et al., 2022), abdominal pain (Dadeh and Phunyanantakorn, 2020), mental health (Wolff et al., 2015), cardiovascular diseases (Almashrafi et al., 2016;Alsinglawi et al., 2020a), or in specific discipline areas or specialties such as spine surgery (Basil and Wang, 2019) and cancer surgeries (Laky et al., 2010;Gohil et al., 2014;Jo et al., 2021). However, most of these studies have had limited sample sizes and have not considered a wide range of clinical factors. ...
... LoS is calculated as the number of days between admission and discharge. We defined PLoS threshold as any LoS that is longer than the reported average LoS (i.e., 6 days; Zoller et al., 2014;Song et al., 2020;Wu et al., 2020). The LoS was reclassified as binary (i.e., either "without PLoS< 6 'days' or with PLoS" ≥6 "days") for classification analysis, and LoS as a continuous outcome for regression analysis. ...
Objective
This study aims to develop and compare different models to predict the Length of Stay (LoS) and the Prolonged Length of Stay (PLoS) of inpatients admitted through the emergency department (ED) in general patient settings. This aim is not only to promote any specific model but rather to suggest a decision-supporting tool (i.e., a prediction framework).
Methods
We analyzed a dataset of patients admitted through the ED to the “Sant”Orsola Malpighi University Hospital of Bologna, Italy, between January 1 and October 26, 2022. PLoS was defined as any hospitalization with LoS longer than 6 days. We deployed six classification algorithms for predicting PLoS: Random Forest (RF), Support Vector Machines (SVM), Gradient Boosting (GB), AdaBoost, K-Nearest Neighbors (KNN), and logistic regression (LoR). We evaluated the performance of these models with the Brier score, the area under the ROC curve (AUC), accuracy, sensitivity (recall), specificity, precision, and F1-score. We further developed eight regression models for LoS prediction: Linear Regression (LR), including the penalized linear models Least Absolute Shrinkage and Selection Operator (LASSO), Ridge and Elastic-net regression, Support vector regression, RF regression, KNN, and eXtreme Gradient Boosting (XGBoost) regression. The model performances were measured by their mean square error, mean absolute error, and mean relative error. The dataset was randomly split into a training set (70%) and a validation set (30%).
Results
A total of 12,858 eligible patients were included in our study, of whom 60.88% had a PloS. The GB classifier best predicted PloS (accuracy 75%, AUC 75.4%, Brier score 0.181), followed by LoR classifier (accuracy 75%, AUC 75.2%, Brier score 0.182). These models also showed to be adequately calibrated. Ridge and XGBoost regressions best predicted LoS, with the smallest total prediction error. The overall prediction error is between 6 and 7 days, meaning there is a 6–7 day mean difference between actual and predicted LoS.
Conclusion
Our results demonstrate the potential of machine learning-based methods to predict LoS and provide valuable insights into the risks behind prolonged hospitalizations. In addition to physicians' clinical expertise, the results of these models can be utilized as input to make informed decisions, such as predicting hospitalizations and enhancing the overall performance of a public healthcare system.
... Halawi reported an average hospitalization period of 3 days among a group of patients from the United States after TKR [25]. In a group of Chinese patients after TKR, the average hospital stay was 8.3 days [26]. The period of hospitalization after TKR reported by other researchers [7,18,[23][24][25][26] from different countries was similar to our results. ...
... In a group of Chinese patients after TKR, the average hospital stay was 8.3 days [26]. The period of hospitalization after TKR reported by other researchers [7,18,[23][24][25][26] from different countries was similar to our results. Our study showed no significant differences between the first and second surgery of staged bilateral TKR procedures in terms of the duration of hospital stay. ...
Background:
Bilateral osteoarthritis of the knee is an indication for a bilateral total knee replacement (TKR) procedure. The goal of our study was to assess the sizes of the implants used during the first and second stages of TKR procedures in order to compare their size and identify the prognostic factors for the second procedure.
Methods:
We evaluated 44 patients who underwent staged bilateral TKR procedures. We assess the following prognostic factors from the first and second surgery: duration of anesthesia, femoral component size, tibial component size, duration of hospital stay, tibial polyethylene insert size, and the number of complications.
Results:
All assessed prognostic factors did not differ statistically between the first and second TKR. A strong correlation was found between the size of femoral components and the size of tibial components used during the first and second total knee arthroplasty. The mean duration of the hospital stay associated with the first TKR surgery was 6.43 days, whereas the mean duration of the second hospital stay was 5.5 days (p = 0.211). The mean sizes of the femoral components used during the first and second procedures were 5.43 and 5.2, respectively (p = 0.54). The mean sizes of the tibial components used during the first and second TKR procedures were 5.36 and 5.25, respectively (p = 0.382). The mean sizes of the tibial polyethylene inserts used during the first and second procedures were 9.45 and 9.34 (p = 0.422), respectively. The mean duration of anesthesia during the first and second knee arthroplasty was 117.04 min and 118.06 min, respectively (p = 0.457). The mean rates of recorded complications associated with the first and second TKR procedures were 0.13 and 0.06 per patient (p = 0.371).
Conclusions:
We observed no differences between the two stages of treatment in terms of all analyzed parameters. We observed a strong correlation between the size of femoral components used during the first and second total knee arthroplasty. We noted a strong correlation between the size of tibial components used during the first and second procedure. Slightly weaker prognostic factors include the number of complications, duration of anesthesia and tibial polyethylene insert size.
... Day-case surgeries have become quite challenging in joint arthroplasties. However, the implementation of "fast-track" pathways in Chinese hospitals has been infrequent, and the LOS after TKA has not been extensively researched (13,14). This study aimed to develop a prediction model to identify factors that can predict a longer LOS following TKA. ...
Background:
Total knee arthroplasty (TKA) is the ultimate option for end-stage osteoarthritis, and the demand of this procedure are increasing every year. The length of hospital stay (LOS) greatly affects the overall cost of joint arthroplasty. The purpose of this study was to develop and validate a predictive model using perioperative data to estimate the risk of prolonged LOS in patients undergoing TKA.
Methods:
Data for 694 patients after TKA collected retrospectively in our department were analyzed by logistic regression models. Multi-variable logistic regression modeling with forward stepwise elimination was used to determine reduced parameters and establish a prediction model. The discrimination efficacy, calibration efficacy, and clinical utility of the prediction model were evaluated.
Results:
Eight independent predictors were identified: non-medical insurance payment, Charlson Comorbidity Index (CCI) ≥ 3, body mass index (BMI) > 25.2, surgery on Monday, age > 67.5, postoperative complications, blood transfusion, and operation time > 120.5 min had a higher probability of hospitalization for ≥6 days. The model had good discrimination [area under the curve (AUC), 0.802 95% CI, 0.754-0.850]] and good calibration (p = 0.929). A decision curve analysis proved that the nomogram was clinically effective.
Conclusion:
This study identified risk factors for prolonged hospital stay in patients after TKA. It is important to recognize all the factors that affect hospital LOS to try to maximize the use of medical resources, optimize hospital LOS and ultimately optimize the care of our patients.
... Αναφορικά με τους παράγοντες που βρέθηκαν να σχετίζονται με τις συνολικές ημέρες νοσηλείας, διαπιστώθηκε πως η αύξηση της ηλικίας σχετίζεται με αυξημένη διάρκεια νοσηλείας (r=0,215, p<0,001), εύρημα που βρίσκεται σε συμφωνία με εκείνα άλλων μελετών. [25][26][27] Σύμφωνα βέβαια με άλλη μελέτη που διεξήχθη σε δείγμα μόνο ηλικιωμένων ασθενών με κάταγμα του ισχίου, διαπιστώθηκε πως για κάθε αύξηση της ηλικίας κατά 1 έτος η διάρκεια νοσηλείας μειώθηκε κατά μέσο όρο 0,085 ημέρες (p < 0,0041). 28 Επίσης, η μη προσαρμοσμένη ανάλυση σε μια πρόσφατη μελέτη έδειξε μια στατιστικά σημαντική διαφορά στη διάρκεια νοσηλείας με την ηλικία των ασθενών, με το μοντέλο παλινδρόμησης να μην αναδεικνύει όμως την ηλικία ως ένα στατιστικά σημαντικό ανεξάρτητο παράγοντα της διάρκειας νοσηλείας. ...
Εισαγωγή: Οι μυοσκελετικές διαταραχές αποτελούν την πιο κοινή αιτία σοβαρού, μακροχρόνιου πόνου και σωματικής αναπηρίας επηρεάζοντας εκατοντάδες εκατομμύρια ανθρώπους, ανεξαρτήτως ηλικίας, φύλου και κοινωνικοδημογραφικού στρώματος παγκοσμίως. Αντιπροσωπεύ- ουν δε μια σημαντική επιβάρυνση για το άτομο, την κοινωνία και τις υπηρεσίες υγείας και είναι ιδιαίτερα δαπανηρές για αυτές. Διάφοροι παράγοντες έχουν συσχετιστεί με τη συνολική διάρκεια νοσηλείας των ασθενών με μυοσκελετικές διαταραχές, τόσο μεμονωμένοι παράγοντες σχετιζόμενοι με τον ίδιο τον ασθενή όσο και παράγοντες που αφορούν τις υπηρεσίες υγείας.
Σκοπός: Η παρούσα εργασία έχει ως σκοπό την αναγνώριση των χαρα- κτηριστικών των νοσηλευόμενων με μυοσκελετικές διαταραχές ασθενών και στον εντοπισμό των χαρακτηριστικών εκείνων των ασθενών και του νοσοκομείου που σχετίζονται με τη διάρκεια και το κόστος νοσηλείας. Υλικό και Μέθοδος: Το δείγμα αποτέλεσαν το 10% του συνόλου των εισαχθέντων στην ορθοπεδική κλινική ελληνικού δημόσιου νοσοκομείου από τις 1/1/2012 έως και τις 31/12/2017 (ισχύς 90%, επίπεδο σημαντικό- τητας 95%), η επιλογή των οποίων έγινε μέσω τυχαίας δειγματοληψίας δια λογισμικού δημιουργίας τυχαίων αριθμών. Η στατιστική επεξεργασία του εμπειρικού υλικού της έρευνας πραγματοποιήθηκε με το λογισμικό πρόγραμμα Statistical Package for the Social Sciences, 20.0 (S.P.S.S. Inc., Chicago, IL, USA), χρησιμοποιώντας τις μεθόδους της Περιγραφικής (Descriptive) και της Επαγωγικής (Inferential) Στατιστικής.
Αποτελέσματα: Το δείγμα της παρούσας μελέτης αποτέλεσαν 634 ασθε- νείς, μέσης ηλικίας 62.8±19.4 ετών. Το 59,8% των νοσηλευθέντων ήταν γυναίκες. Τα κατάγματα αποτελούσαν την αιτία για το 52,2% των νοση- λειών στην ορθοπεδική κλινική. Η μέση διάρκεια νοσηλείας των ασθενών στην ορθοπεδική κλινική ανήλθε σε 4,58±4,57, ενώ το μέσο συνολικό κόστος αυτής σε 2083,32±1756,66 €. Ποικίλοι παράγοντες συσχετίστη- καν στατιστικά σημαντικά με το συνολικό πραγματικό κόστος νοσηλείας, συμπεριλαμβανομένων του φύλου (t(554)=-3.834, p<0,001), της ηλικί- ας (r=0.288, p<0,001), της οικογενειακής κατάστασης (F(4,626)=4.781, p<0,001), του επαγγέλματος (F(5,626)=2.408, p<0,05), της διάγνωσης (t(627)=-4.671, p<0,001), καθώς και της διενέργειας ιατρικής πράξης (t(629)=6.903, p<0,001) ή χειρουργικής επέμβασης (t(628)=9.388, p<0,001). Αντίστοιχα, η διάρκεια νοσηλείας συσχετίστηκε με την ηλι- κία (r=0.215, p<0,001), με τη διάγνωση κατάγματος (t=(627)=2.894, p<0,001), καθώς και με τη διενέργεια χειρουργικής επέμβασης κατά τη διάρκεια της νοσηλείας (t=(628)=3.500, p<0,001).
Συμπεράσματα: Τα χαρακτηριστικά των ορθοπεδικών ασθενών και η θεραπευτική προσέγγιση αυτών αποτελούν σημαντικούς καθοριστικούς παράγοντες τόσο του κόστους όσο και της διάρκειας νοσηλείας.
... This agreed with Garcia et al., [29] who stated that ASA classification proved useful in estimating LOS. Furthermore, this was in accordance with Song et al., [30] who found that the ASA classification was one of the factors contributing to a prolonged LOS. ...
Background: Prediction scores could help to timely identify patients at risk. More intense care monitoring outside the pediatric intensive care unit (PICU) could improve outcomes. These concepts of critical care without borders could be implemented shortly with local resources and improve patient safety. Predict more, do less in PICUs, and more in the ward. The aim of this work was to evaluate the impact of prediction scores for identification of postoperative high risk surgical patients and their need for pediatric intensive care unit admission. Methods: This prospective study was carried out on 40 pediatric patients. The studied patients were divided into three groups. LR Group: 14 cases with ASA = 1, 2 PEWS ≤ 2, pSOFA ≤ 7, LqSOFA < 2. IR Group: 10 cases with PEWS = 3, 4 pSOFA = 8 -11 HiR Group: 16 cases with ASA = 3 -5, PEWS ≥ 5, pSOFA ≥ 12, LqSOFA ≥ 2. American Society of Anaesthesiologists (ASA) scoring was obtained from each patient preop. The Pediatric Early Warning (PEWS) Scoring was obtained from each patient immediately postop. The Pediatric Sequential Organ Failure Assessment (pSOFA) Scoring was obtained from patients admitted to PICU on day 1 and day 7. Liverpool quick Sequential Organ Failure Assessment (LqSOFA) scoring was obtained from all patients on admission. Results: Regarding prognostic performance of different scores to predict mortality. For ASA score: it was statistically significant with AUC = .882, cut off values > 3, sensitivity = 75%, specificity = 88.89%, PPV = 42.9% and NPV = 97%. -For PEW score: it was statistically significant with AUC = .892, cut off values > 5, sensitivity = 79%, specificity = 80.56%, PPV =30% and NPV = 96.7%. -For pSOFA score: it was statistically significant with AUC = .931, cut off values > 14, sensitivity = 85%, specificity = 86.11%, PPV =37.5%, and NPV = 96.9%. All ASA, PEW, and pSOFA were statistically significant as univariate but none was significant as multivariate. Conclusions: ASA score, pSOFA score and PEWS score were significant predictor to length of stay (>21 days). ASA score, PEW score and pSOFA score were significant as predictor to mortality. ASA score and PEWS score were highly significant as to predict PICU Admission postop. PEWS score was highly significant as to predict PICU Admission postop. ASA, PEWs, and pSOFA were predictors for LOS for more than 21 days, predictors of mortality and predictors for PICU admission Postop. Decreased platelets and increased WBCs, urea, creatinine, AST, and RBG were significant with HiR. LqSOFA is a simple variable bedside tool for identifying septic patients at high risk for poor outcomes.
... The results were quite different from the findings of many previous studies, which showed comorbidities were crucial predictors for prolonged LOS. [29][30][31] We believe the perioperative management of comorbidities in our team can minimize their influence on LOS. First, patients were strictly screened for comorbidities and assessed to determine if they were eligible for the TJA surgery. ...
Objective:
To investigate the safety, efficiency and cost of total joint arthroplasty (TJA) under the enhanced recovery after surgery (ERAS) program and identify predictors facilitating further decrease in length of stay (LOS).
Methods:
We retrospectively collected the information of patients who underwent primary unilateral TJA by a single surgical team between January 2017 and June 2019. A total of 604 patients with LOS ≤ 3 was enrolled in this study. All patients completed 12-month or longer follow-up. Patients received the same ERAS protocol, mainly including preoperative preparation (patient education, preoperative functional exercises, nutritional support), blood management, pain management, sleep management, prevention of infection, prevention of thrombosis and strict discharge criteria. Preoperative characteristics of patients were collected from the medical record system and were compared between the LOS ≤ 2 group and the LOS = 3 group. Factors with significant difference were included in multivariate logistic regression analysis to find independent preoperative predictors for LOS. Joint function at the latest follow-up, adverse events rate and hospitalization costs were compared between the LOS ≤ 2 group and the LOS = 3 group.
Results:
Of the enrolled 604 patients, 271 patients (44.9%) had a LOS of 2 days or less while 333 patients (55.1%) had a LOS of 3 days. Pittsburgh Sleep Quality Index score (odds ratio [OR] = 1.084, 95% confidence interval [CI] = 1.024-1.147, P = 0.005), preoperative albumin level (OR = 0.945, 95% CI = 0.905-0.988, P = 0.012), digestive diseases (OR = 1.084, 95% CI = 1.024-1.147, P = 0.005) and total hip arthroplasty (THA) (OR = 0.273, 95% CI = 0.170-0.439, P < 0.001) were predictors of LOS ≤ 2 in the multivariate logistic analysis model. The postoperative joint function scores and adverse event rates were comparable between the LOS ≤ 2 group and the LOS = 3 group. The hospital costs were lower in the LOS ≤ 2 group than the LOS = 3 group.
Conclusion:
Under the rigorous ERAS program, 2-day discharge in unselected TJA patients can be routinely applied. Patients with high preoperative sleep quality, high preoperative albumin level, free of digestive disease and undergoing THA procedure are more likely to be discharged within 2 days.
... [1] In addition to the economic burden there is an increase in morbidity to the patients because of rehabilitation and delay in returning back to work or doing normal day to day activities. [3] ...
Background: Total knee replacement (TKR) is done for severe degenerative arthritis of the knee joint, with the increasing number of the cases in developing countries to undergo TKR. There is an increase in economic burden in patients who go through these procedures in terms of surgery cost, implant cost and the cost of stay in hospitals, the latter being the most common variation. Various preoperative, intraoperative and postoperative factors decide the length of hospital stay and the economic burden in patients. The aim of our study was to determine all intraoperative parameters which are responsible for increased stay in hospital leading to increased cost burden in patients and ways to reduce the hospital stay by optimizing the patients preoperatively and to effectively manage them intra-operatively, so as to reduce the length of stay but still providing best of care to the patients. Material and methods: This study was done in our institute. It's a retrospective study done in 1022 patients who underwent elective total knee replacement for arthritis of knee joint and fractures that may need TKR. Various intraoperative parameters have been studied in this study such as surgery time, intraoperative blood loss, intraoperative blood transfusion and types of anesthesia and its effect on the length of stay of the patients in terms of economic burden. Results: We looked into the intraoperative parameters during total knee replacement and found that all these parameters have a significant role to alter the duration of hospital stay. Conclusion: Our study concluded that most of the intraoperative parameters responsible for total knee replacement are non-modifiable as one cannot predict the intraoperative complication which can lead to increased surgery time which ultimately leads to increased intraoperative blood loss and need for intraoperative blood transfusion. Type of anesthesia given at the time of surgery was also responsible for the increase in length of stay. No surgery is free of complications and so surgeons should be mentally prepared for all the events. Better preoperative optimization of patients and preoperative planning may reduce these events.
... The primary outcome of this study was having pLOS-ICU. pLOS-ICU is defined as a length of stay longer than the reported average LOS-ICU [16,19,36], which is three days for general ICU patients in the United States [37]. ...
This study aimed to construct machine learning (ML) models for predicting prolonged length of stay (pLOS) in intensive care units (ICU) among general ICU patients. A multicenter database called eICU (Collaborative Research Database) was used for model derivation and internal validation, and the Medical Information Mart for Intensive Care (MIMIC) III database was used for external validation. We used four different ML methods (random forest, support vector machine, deep learning, and gradient boosting decision tree (GBDT)) to develop prediction models. The prediction performance of the four models were compared with the customized simplified acute physiology score (SAPS) II. The area under the receiver operation characteristic curve (AUROC), area under the precision-recall curve (AUPRC), estimated calibration index (ECI), and Brier score were used to measure performance. In internal validation, the GBDT model achieved the best overall performance (Brier score, 0.164), discrimination (AUROC, 0.742; AUPRC, 0.537), and calibration (ECI, 8.224). In external validation, the GBDT model also achieved the best overall performance (Brier score, 0.166), discrimination (AUROC, 0.747; AUPRC, 0.536), and calibration (ECI, 8.294). External validation showed that the calibration curve of the GBDT model was an optimal fit, and four ML models outperformed the customized SAPS II model. The GBDT-based pLOS-ICU prediction model had the best prediction performance among the five models on both internal and external datasets. Furthermore, it has the potential to assist ICU physicians to identify patients with pLOS-ICU risk and provide appropriate clinical interventions to improve patient outcomes.
