Scholarlink

Introduction <p>Population ageing increases long-term care (LTC) needs. Understanding which hospitalised patients are more likely to be discharged to LTC is essential for adapting services and policies. This study aims to assess clusters of hospitalised patients with a higher proportion of discharges to LTC (LTCD) in Portugal and to&#xa0;test clustering methods as a solution for the early identification of potential users using different approaches.</p> Methods <p>This nationwide Portuguese study used inpatient data from Portuguese hospitals between 2012 and 2017. The variables considered in this study were age, sex, principal diagnosis, comorbidities (identified using secondary diagnoses), admission type and hospital transfer. The main outcome of this analysis is discharge to long-term and maintenance units (<i>Unidades de Longa Duração e Manutenção</i>—ULDM). Different approaches were applied to categorise the&#xa0;principal diagnosis for each inpatient episode, using ICD-9-CM and ICD-10-CM main groups, ICD-9-CM and ICD-10-CM more detailed categories, Clinical Classification Software (CCS) and CCS Refined (CCSR). Subsequently, hierarchical clustering techniques were applied to determine the number of clusters in each dataset and decision tree methods were used to characterise each cluster.</p> Results <p>A total of 4427 inpatient episodes (0.23%) were discharged to LTC. Across clustering methods, the proportion of patients discharged to LTC varied widely, from 0.7% to 60.8%. Certain categorisation methods, such as CCSM2, showed more concentrated high-risk groups compared to other methods with more categories. The models showed high performance (F1 score &gt; 0.97).</p> Conclusion <p>The clustering results exhibit considerable variability when comparing the different approaches to categorising principal diagnoses. The “quality” of the principal diagnosis categorisation (i.e. the grouping method)&#xa0;overcomes the “quantity” (i.e., the number of categories). This can have important implications for health services and hospital management. Clustering methods serve as effective options for identifying high-risk groups, although different approaches should be assessed and compared.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Determinating clusters with a higher proportion of long-term care discharges from hospitals: a nationwide Portuguese study using clustering and decision tree methods

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月29日

查看全文
  • Ana Carreira,
  • Diogo Martinho,
  • Vítor Crista,
  • Júlio Souza,
  • Filipa Santos Martins,
  • Goreti Marreiros,
  • Alberto Freitas,
  • João Vasco Santos

摘要

Introduction

Population ageing increases long-term care (LTC) needs. Understanding which hospitalised patients are more likely to be discharged to LTC is essential for adapting services and policies. This study aims to assess clusters of hospitalised patients with a higher proportion of discharges to LTC (LTCD) in Portugal and to test clustering methods as a solution for the early identification of potential users using different approaches.

Methods

This nationwide Portuguese study used inpatient data from Portuguese hospitals between 2012 and 2017. The variables considered in this study were age, sex, principal diagnosis, comorbidities (identified using secondary diagnoses), admission type and hospital transfer. The main outcome of this analysis is discharge to long-term and maintenance units (Unidades de Longa Duração e Manutenção—ULDM). Different approaches were applied to categorise the principal diagnosis for each inpatient episode, using ICD-9-CM and ICD-10-CM main groups, ICD-9-CM and ICD-10-CM more detailed categories, Clinical Classification Software (CCS) and CCS Refined (CCSR). Subsequently, hierarchical clustering techniques were applied to determine the number of clusters in each dataset and decision tree methods were used to characterise each cluster.

Results

A total of 4427 inpatient episodes (0.23%) were discharged to LTC. Across clustering methods, the proportion of patients discharged to LTC varied widely, from 0.7% to 60.8%. Certain categorisation methods, such as CCSM2, showed more concentrated high-risk groups compared to other methods with more categories. The models showed high performance (F1 score > 0.97).

Conclusion

The clustering results exhibit considerable variability when comparing the different approaches to categorising principal diagnoses. The “quality” of the principal diagnosis categorisation (i.e. the grouping method) overcomes the “quantity” (i.e., the number of categories). This can have important implications for health services and hospital management. Clustering methods serve as effective options for identifying high-risk groups, although different approaches should be assessed and compared.