The prediction of hospital length of stay using unstructured data

Abstract

Objective This study aimed to assess the performance improvement for machine learning-based hospital length of stay (LOS) predictions when clinical signs written in text are accounted for and compared to the traditional approach of solely considering structured information such as age, gender and major ICD diagnosis. Methods This study was an observational retrospective cohort study and analyzed patient stays admitted between 1 January to 24 September 2019. For each stay, a patient was admitted through the Emergency Department (ED) and stayed for more than two days in the subsequent service. LOS was predicted using two random forest models. The first included unstructured text extracted from electronic health records (EHRs). A word-embedding algorithm based on UMLS terminology with exact matching restricted to patient-centric affirmation sentences was used to assess the EHR data. The second model was primarily based on structured data in the form of diagnoses coded from the International Classification of Disease 10th Edition (ICD-10) and triage codes (CCMU/GEMSA classifications). Variables common to both models were: age, gender, zip/postal code, LOS in the ED, recent visit flag, assigned patient ward after the ED stay and short-term ED activity. Models were trained on 80% of data and performance was evaluated by accuracy on the remaining 20% test data. Results The model using unstructured data had a 75.0% accuracy compared to 74.1% for the model containing structured data. The two models produced a similar prediction in 86.6% of cases. In a secondary analysis restricted to intensive care patients, the accuracy of both models was also similar (76.3% vs 75.0%). Conclusions LOS prediction using unstructured data had similar accuracy to using structured data and can be considered of use to accurately model LOS.

Development and Validation of an Artificial Intelligence Electrocardiogram Recommendation System in the Emergency Department

“…In addition to cardiopulmonary symptoms and trauma, features of pain-associated conditions were also weighted higher in information gain among others in the prediction model, indicating that an early ECG survey was commonly needed in those with pain-related syndromes in the ED. There are growing numbers of studies on ED triage systems using machine learning techniques to predict various outcomes such as ICU or hospital admission [ 38 , 39 ], length of stay [ 40 ], mortality [ 38 ], ED revisits [ 41 ], and need for critical care [ 42 ], with the results mostly suggesting that predictive performance improves with machine learning techniques compared with conventional analysis. By integrating such a decision support tool, the ED may accelerate ECG diagnostic processes by activating an integrated AI-assisted ECG analysis to increase diagnostic proficiency and identify crucial abnormalities early.…”

Predicting medical events and ICU requirements using a multimodal multiobjective transformer network

“…Neural language models are the default choice for building predictive models that exploit text. [11][12][13][14][15] An end-to-end deep dynamic neural framework was proposed by Pham et al 11 to predict future medical outcomes such as the next diagnosis, current interventions from the current diagnoses, and future risks like unplanned readmission within a certain period. In 2021, Van Aken et al 12 proposed a transformerbased model for predicting multiple clinical outcomes such as the International Statistical Classification of Diseases and Related Health Problems-Ninth Edition (ICD-9) diagnosis, ICD-9 procedures, in-hospital mortality, and length of ICU stay, using the discharge summary from the MIMIC data set.…”

“…Several preprocessing steps were applied to obtain the relevant data. Chrusciel et al 13 proposed the use of random forest along with a word-embedding algorithm based on the Unified Medical Language System (UMLS) terminology, to predict hospital length of stay from unstructured EHRs. Huang et al 14 also investigated the use of physicians and nursing notes generated within the first 48 h of admissions, in predicting ICU length of stay and mortality.…”

Consequences of inequity in the neurosurgical workforce: Lessons from traumatic brain injury

“…The utilization of objective measures with machine learning (artificial intelligence) has the potential to reduce inequities in the neurosurgical field through automation, improved accuracy, speed, accessibility, and reduced costs (34)(35)(36). A major caveat is that we need to ensure that data elements incorporated into future algorithms do not perpetuate inequity (40)(41)(42). Yet, we find the implicit bias currently found in healthcare is further propagated by machine learning due to systemic inequities.…”