Choice of models for the analysis and forecasting of hospital beds.
Choice of models for the analysis and forecasting of hospital beds.
复制标题
DOI：
10.1007/s10729-005-2013-y
复制
发表时间：
2005-08-01
影响因子：
3.6
通讯作者：
Lee, Michael
中科院分区：
文献类型：
--
作者：
Mackay, Mark;Lee, Michael
人们越来越担心，目前的保健服务是不可持续的。隔室流动模型提供了改善床位占用决策的机会，特别是那些具有战略性质的决策。这种建模方法可用于补充基础设施和劳动力规划方法。关于模型复杂程度的适当性、拟合程度以及使用分室流动模型进行推广和预测的能力的讨论一直缺乏。作者研究了与医院病床间室流量模型相关的模型选择和评估。创建了一个适用于一系列情景的分区模型。培训和测试数据分别涉及1998和1999日历年。测试的大多数情景都是基于描述时间段的常用时期。通过优化实现的拟合优度针对训练和测试数据进行测量。正如预期的那样，随着复杂性的增加，模型拟合得到了改善。对模型与试验数据的拟合分析表明，增加模型的复杂性确实会导致过拟合，并且使用相对简单的模型可以实现更好的预测。在推广方面，季节性模型表现最好。米勒德和他的同事们使用的单日人口普查类型模型也被生成。这些模型的性能相似，但不如从一整年的训练数据生成的模型。额外的数据使模型能够更好地捕捉全年活动的变化。
There is growing concern that current health care services are not sustainable. The compartmental flow model provides the opportunity for improved decision-making about bed occupancy decisions, particularly those of a strategic nature. This modelling can be applied to complement infrastructure and workforce-planning methods. Discussion about appropriateness of the level of model complexity, the degree of fit and the ability to use compartmental flow models for generalization and forecasting has been lacking. The authors investigated model selection and assessment in relation to hospital bed compartment flow models. A compartment model for a range of scenarios was created. The training and test data related to the 1998 and 1999 calendar years, respectively. The majority of scenarios tested were based upon commonly used periods that describe periods of time. The goodness-of-fit achieved by optimisation was measured against the training and test data. Model fit improved with increasing complexity as expected. The analysis of model fit against the test data showed that increasing model complexity did result in over-fitting, and better prediction was achieved with a relatively simple model. In terms of generalisation, the seasonal models performed best. Single day census type models, which have been used by Millard and his colleagues, were also generated. The performance of these models was similar, but inferior to that of the models generated from a full year of training data. The additional data make the models better able to capture the variation across the year in activity.