Design of a Clinical Decision Support Model for Predicting Pneumonia Readmission

Abstract

Readmission is considered as an indicator for evaluating the overall health care environment of a hospital. Existing models developed to predict readmissions for pneumonia lack discriminative ability. In this study, we aim to determine the risk factors to predict readmission and to design a clinical decision support system (CDSS) to predict if a patient will be readmitted for pneumonia within 30 days after discharge. The data of 17,222 patients who had been admitted within the period from January to December 2013 were used for analysis. The study cohort consisted of 520 index admissions for pneumonia in a general hospital situated at Tao-Yuan area of Taiwan. Variables including demographic information, treatment and clinical factors, and health care utilization factors were collected. The selected variables are then applied for CDSS design using the RBF-SVM. Of the 520 index admissions for pneumonia, 86 (16.2%) patients were readmitted within 30 day. Six variables, including age, gender, number of medication, length of admission, number of comorbidities, and total admission cost were observed to be significant (p<0.05) in predicting readmission. The predictive model was constructed using the RBF-SVM with 6 significant variables and all 20 variables, respectively. By testing models with different combinations of SVM parameters, C and γ, the predictive accuracy for two different models achieved 83.85% and 82.24% respectively. The model can be effective in identifying pneumonia patients at high risk for readmission.

FasterRib: A deep learning algorithm to automate identification and characterization of rib fractures on chest computed tomography scans

“…Despite seemingly disparate tasks (e.g., natural image and medical image classification), fine-tuning pretrained algorithms often outperforms algorithms trained from scratch 18 . Transfer learning has been successfully deployed for various medical tasks, from natural language processing (ClinicalBERT, a fine-tuned version of Google's general language model, was used to predict 30-day hospital readmissions from medical records 19 ) to computer vision (a model pretrained on nonmedical images was fine-tuned to identify metastatic lymph nodes in magnetic resonance imaging scans 20 ). Code transparency and readability are critical for transfer learning: we made FasterRib's code open source to support other groups fine-tuning the model on external data sets, conducting heterogeneous transfer learning (i.e., inference on other CT scan tasks) or exploring architectural breakthroughs in object detection 21 …”

Predicting hospital-acquired pneumonia among schizophrenic patients: a machine learning approach

“…In our study, random forest, a widely recognized learner with well-performance across a broad range of problems [ 37 ], outperformed the study of Caruana, Lou, Gehrke, Koch, Sturm and Elhadad [ 19 ], but the logistic regression in our model performed poorer than that of Caruana, Lou, Gehrke, Koch, Sturm and Elhadad [ 19 ]. Huang, Chen and Hsu [ 26 ] constructed a clinical decision model via a support vector machine learner for predicting pneumonia readmission, and their model achieved 83.9% predictive accuracy with six predictors, including age, gender, number of medication, length of admission, number of comorbidities, and total admission cost. Our proposed model utilizing a support vector machine learner provided a slightly higher predictive accuracy (87.1%) than that of Huang, Chen and Hsu [ 26 ].…”

“…Huang, Chen and Hsu [ 26 ] constructed a clinical decision model via a support vector machine learner for predicting pneumonia readmission, and their model achieved 83.9% predictive accuracy with six predictors, including age, gender, number of medication, length of admission, number of comorbidities, and total admission cost. Our proposed model utilizing a support vector machine learner provided a slightly higher predictive accuracy (87.1%) than that of Huang, Chen and Hsu [ 26 ].…”

“…In order to predict risk factors for schizophrenia inpatients, we adopted R 3.5.1 [ 22 ], an open source statistical platform for data analysis. Prior pneumonia- related studies have adopted several machine-learning algorithms such as decision tree [ 23 ], k-nearest neighbors [ 24 ], naïve Bayes [ 25 ], random forest [ 19 ], logistic regression [ 19 ], and support vector machine [ 25 , 26 ]. We therefore chose seven algorithms, including classification and regression tree (CART), decision tree (C5.0), k-nearest neighbors (KNN), naïve Bayes (NB), random forest (RF), support vector machine (SVM), and logistic regression (LGR), to build the predictive model and also to compare the performance of those machine learning algorithms along with prior studies.…”

Untitled

“…We analyzed the capabilities of different types of LLMs to generate clinically relevant synthetic data. We tested nine different LLMs ranging from medical domainspecific models (such as BioGPT [24], ClinicalBERT [16], BioClinicalBERT [1], and Medtiron [9]) to more general models (Mistral [18], Zehyr [32], Llama2 [17], GPT-3.5, and GPT-4). Figure S2 of Supplementary Material presents an example of a prompt to generate values for existing features based on the patient's clinical context.…”