Table 1_Predicting the hospitalization burdens of patients with mental disease: a multiple model comparison.docx

BackgroundMental disorders represent a growing public health challenge, with rising hospitalization rates worldwide. Despite their significant impact, systematic investigations into the hospitalization burden (HB) of mental disorders remain notably lacking in current studies.
ObjectiveThis study aims to employ machine learning (ML) techniques to predict the HB among patients with mental disorders. By doing so, we seek to optimize the allocation of medical resources and enhance the efficiency of healthcare services for this specific population.
MethodsHistorical hospitalization data were collected, encompassing patient demographics, diagnostic details, length of stay, costs, and other relevant information. The data were then cleaned to remove missing values and outliers, and key features related to the HB were extracted. A statistical analysis of the basic characteristics of the HB was conducted. Subsequently, prediction models for the HB were developed based on the historical data and identified key features, including time series models and regression models. The predictive ability of these models was evaluated by comparing the actual values with the predicted values.
ResultsHB was influenced by diagnosis, age, and seasonality, with schizophrenia (A3) and personality disorders (A7) incurring the highest burdens. ML models demonstrated task-specific efficacy: ridge regression for hospitalization frequency, long short-term memory/categorical boosting regression for length of stay, and seasonal autoregressive integrated moving average with exogenous regressors/light gradient boosting machine regression for hospitalization costs. The findings support tailored resource allocation and early intervention for high-risk groups.
ConclusionThis study showcased the effectiveness of machine learning methods in predicting the hospitalization burden of inpatients with mental disorders, thereby offering scientific decision support for medical institutions. This approach contributes to enhancing the quality of patient care and optimizing the efficiency of medical resource utilization.

背景
精神障碍（mental disorders）已成为日益严峻的公共卫生挑战，全球范围内的住院率持续攀升。尽管其影响深远，但当前研究中针对精神障碍住院负担（hospitalization burden, HB）的系统性调查仍显著匮乏。
研究目的
本研究旨在运用机器学习（machine learning, ML）技术预测精神障碍患者的住院负担，以期优化医疗资源配置，提升该特定人群的医疗服务效率。
研究方法
本研究收集了包含患者人口学特征、诊断详情、住院时长、住院费用及其他相关信息的历史住院数据。随后对数据进行清洗，以去除缺失值与异常值，并提取与住院负担相关的关键特征。同时开展了住院负担基本特征的统计分析。基于上述历史数据与提取的关键特征，本研究构建了住院负担预测模型，涵盖时间序列模型与回归模型两类。通过对比实际值与预测值，评估了各模型的预测性能。
研究结果
住院负担受诊断类型、年龄及季节因素影响，其中精神分裂症（A3）与人格障碍（A7）对应的住院负担最高。机器学习模型展现出针对特定任务的有效性：岭回归（ridge regression）适用于住院频次预测，长短期记忆（long short-term memory, LSTM）/分类增强回归（categorical boosting regression）适用于住院时长预测，带外生变量的季节自回归积分移动平均模型（seasonal autoregressive integrated moving average with exogenous regressors, SARIMAX）/轻量梯度提升机（light gradient boosting machine, LightGBM）回归适用于住院费用预测。本研究结果可为高风险人群制定针对性的资源配置方案与早期干预措施提供依据。
研究结论
本研究证实了机器学习方法在预测精神障碍住院患者住院负担方面的有效性，可为医疗机构提供科学的决策支持，有助于提升患者护理质量与医疗资源利用效率。