Jiang, Yiqun

Intensive care units (ICUs) serve patients with life-threatening conditions. However, the availability of ICU resources remains scarce worldwide. Therefore, it is critical to conduct ICU outcome predictions at an early stage and promote efficient use of ICU resources. However, the current ICU outcome prediction methods have limitations such as unsatisfactory accuracy and dependence on resource-demanding laboratory tests or expert domain knowledge. In this research, we design a wavelet transform based additive convolutional neural network (WT-A-CNN) that requires only patients’ vital sign series and information at ICU admission for real-time ICU outcome predictions. The model is evaluated using a large real-world ICU database and outperforms state-of-the-art baselines on both ICU mortality and length-of-stay prediction tasks. Furthermore, the additive structure of the model can be used for model interpretation and analyzing which input signal is more useful for ICU outcome prediction for different patient cohorts. Our work provides an efficient tool for ICU outcome predictions, allowing healthcare providers to act promptly on patients at risk and reduce the negative impacts on patient outcomes.

Abstract Background Provider burnout due to workload is a significant concern in primary care settings. Workload for primary care providers encompasses both scheduled visit care and non–visit care interactions. These interactions are highly influenced by patients' health conditions or acuity, which can be measured by the Adjusted Clinical Group (ACG) score. However, new patients typically have minimal health information beyond social determinants of health (SDOH) to determine ACG score. Objectives This study aims to assess new patient workload by first predicting the ACG score using SDOH, age, and gender and then using this information to estimate the number of appointments (scheduled visit care) and non–visit care interactions. Methods Two years of appointment data were collected for patients who had initial appointment requests in the first year and had the ACG score, appointment, and non–visit care counts in the subsequent year. State-of-art machine learning algorithms were employed to predict ACG scores and compared with current baseline estimation. Linear regression models were then used to predict appointments and non–visit care interactions, integrating demographic data, SDOH, and predicted ACG scores. Results The machine learning methods showed promising results in predicting ACG scores. Besides the decision tree, all other methods performed at least 9% better in accuracy than the baseline approach which had an accuracy of 78%. Incorporating SDOH and predicted ACG scores also significantly improved the prediction for both appointments and non–visit care interactions. The R 2 values increased by 95.2 and 93.8%, respectively. Furthermore, age, smoking tobacco, family history, gender, usage of injection birth control, and ACG were significant factors for determining appointments. SDOH factors such as tobacco usage, physical exercise, education level, and group activities were strongly correlated with non–visit care interactions. Conclusion The study highlights the importance of SDOH and predicted ACG scores in predicting provider workload in primary care settings.