Predicting Length of Stay using machine learning for total joint replacements performed at a rural community hospital

Abstract

Background Predicting patient’s Length of Stay (LOS) before total joint replacement (TJR) surgery is vital for hospitals to optimally manage costs and resources. Many hospitals including in rural areas use publicly available models such as National Surgical Quality Improvement Program (NSQIP) calculator which, unfortunately, performs suboptimally when predicting LOS for TJR procedures. Objective The objective of this research was to develop a Machine Learning (ML) model to predict LOS for TJR procedures performed at a Perioperative Surgical Home implemented rural community hospital for better accuracy and interpretation than the NSQIP calculator. Methods A total of 158 TJR patients were collected and analyzed from a rural community hospital located in Montana. A random forest (RF) model was used to predict patient’s LOS. For interpretation, permuted feature importance and partial dependence plot methods were used to identify the important variables and their relationship with the LOS. Results The root mean square error for the RF model (0.7) was lower than the NSQIP calculator (1.21). The five most important variables for predicting LOS were BMI, Duke Activity Status-Index, diabetes, patient’s household income, and patient’s age. Conclusion This pilot study is the first of its kind to develop an ML model to predict LOS for TJR procedures that were performed at a small-scale rural community hospital. This pilot study contributes an approach for rural hospitals, making them more independent by developing their own predictions instead of relying on public models.

Prediction of Complications and Prognostication in Perioperative Medicine: A Systematic Review and PROBAST Assessment of Machine Learning Tools

“…Readmission and discharge timing: Several studies predicted the risk of hospital readmission within 30 days of surgery, 114 when patients would be ready for hospital discharge 98,102 or length of stay after orthopedic surgery. 99,103–105…”

“…Previously validated scores such as perioperative medicine-related scores ( e.g. , ASA status, POSSUM, Charlson Comorbidity Index, or National Surgical Quality Improvement Program calculator scores) 30 , 32 , 49 , 61 , 100 , 104 or other scores 58 , 62 ( e.g. , Bariclot tool, STOP-BANG score, Mallampati test, various frailty indexes, and the acute kidney injury score) 34 , 49 , 58 , 72 , 79 , 82 , 86 , 114 , 121 …”

“…(1) Multivariate logistic regression 30,31,36,43,48,54,55,60,62,65,67,69,75,76,78,80,95,106,107,110 (2) Previously validated scores such as perioperative medicine-related scores (e.g., ASA status, POSSUM, Charlson Comorbidity Index, or National Surgical Quality Improvement Program calculator scores) 30,32,49,61,100,104 or other scores 58,62 (e.g., Bariclot tool, STOP-BANG score, Mallampati test, various frailty indexes, and the acute kidney injury score) 34,49,58,72,79,82,86,114,121 (3) Clinical assessment 42,52,88 Overall, the machine learning models described in these articles outperformed their technical or clinical comparator, with an average increase in AUC and accuracy between 0.2-0.3, except for that of Chen et al 38 where the ASA score alone, despite a lower AUC, had higher accuracy compared to neural network and logistic regression models.…”

Rural implementation of the perioperative surgical home: A case-control study

“…Future research should include analysis of outcomes including same-day surgery cancellation, surgical cost, postoperative recovery measures, and postoperative opioid consumption. Other future research should also include advanced analytics and predictive modeling such as machine learning and deep learning to predict patient risk and improve the performance of surgical systems at rural and frontier hospitals[ 47 ].…”

Prediction of Intensive Care Unit Length of Stay in the MIMIC-IV Dataset

“…Practical progress in clinical ML has only been incremental and difficult to measure due to the vast diversity of clinical datasets and objectives, as well as the lack of common standard benchmarks for evaluating models [24,25]. Many ICU prediction models, for example, have been developed using data collected by local healthcare providers in a strictly private settings with limited external access, and are therefore not generalizable [20,26]. Moreover, most health data features (attributes) are highly variable across institutions, reflecting different underlying populations, clinical conditions, or even global health issues [24].…”