Scholarlink

Background <p>Infective endocarditis (IE) remains a diagnostically challenging disease with diverse symptoms. Whether IE is suspected frequently depends on clinical judgement. A reliable model to estimate the likelihood of IE at presentation across clinical departments is needed; however, no such model is currently available. We aimed to develop such a diagnostic prediction model using objective and reproducible variables and to validate its performance.</p> Methods <p>We included inpatients aged ≥ 20 years who had either a diagnosis code for IE or for undiagnosed fever during hospitalization. The model was developed using data from a single university hospital between 2007 and 2017 (derivation cohort) and validated with data from four university hospitals between 2018 and 2020 (validation cohort). IE was diagnosed according to the 2023 Duke-International Society for Cardiovascular Infectious Diseases (ISCVID) criteria. Variables were selected using the Boruta algorithm and Least Absolute Shrinkage and Selection Operator regression. Multivariable logistic regression analysis was used to estimate odds ratios and 95% confidence intervals and to construct the model. Model performance was assessed in both cohorts using the area under the curve (AUC). In the validation cohort, calibration was also assessed using the calibration slope, Hosmer–Lemeshow test, and stratified likelihood ratio.</p> Results <p>The derivation and validation cohorts included 105 (46 IE) and 286 (106 IE) patients, respectively. Predictors in the final model were the presence of a cardiac murmur, log-transformed platelet count, neutrophil percentage, presence of pleural effusion, and a quick sequential [sepsis-related] organ failure assessment score ≥ 2. The AUC was 0.918 in the derivation cohort. In the validation cohort, the AUC was 0.859, with a Hosmer– Lemeshow test p-value of 0.246 and a calibration slope of 0.759. The stratified likelihood ratio ranged from 0.04 to 9.71 and increased with higher model scores.</p> Conclusions <p>This model showed high discrimination and good calibration using objective variables that are readily available early after admission. Furthermore, this is the first model to predict IE based on the 2023 Duke-ISCVID criteria. Further multicenter validation in community hospitals would enhance generalizability.</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

A new diagnostic prediction model for infective endocarditis based on the 2023 duke–international society for cardiovascular infectious disease criteria: a multicenter observational study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月16日

查看全文
  • Shun Yamashita,
  • Ryosuke Osako,
  • Takanori Yamaguchi,
  • Yoshinori Tokushima,
  • Yukinori Harada,
  • Taiju Miyagami,
  • Yosuke Sasaki,
  • Yasutomo Oda,
  • Masaki Tago

摘要

Background

Infective endocarditis (IE) remains a diagnostically challenging disease with diverse symptoms. Whether IE is suspected frequently depends on clinical judgement. A reliable model to estimate the likelihood of IE at presentation across clinical departments is needed; however, no such model is currently available. We aimed to develop such a diagnostic prediction model using objective and reproducible variables and to validate its performance.

Methods

We included inpatients aged ≥ 20 years who had either a diagnosis code for IE or for undiagnosed fever during hospitalization. The model was developed using data from a single university hospital between 2007 and 2017 (derivation cohort) and validated with data from four university hospitals between 2018 and 2020 (validation cohort). IE was diagnosed according to the 2023 Duke-International Society for Cardiovascular Infectious Diseases (ISCVID) criteria. Variables were selected using the Boruta algorithm and Least Absolute Shrinkage and Selection Operator regression. Multivariable logistic regression analysis was used to estimate odds ratios and 95% confidence intervals and to construct the model. Model performance was assessed in both cohorts using the area under the curve (AUC). In the validation cohort, calibration was also assessed using the calibration slope, Hosmer–Lemeshow test, and stratified likelihood ratio.

Results

The derivation and validation cohorts included 105 (46 IE) and 286 (106 IE) patients, respectively. Predictors in the final model were the presence of a cardiac murmur, log-transformed platelet count, neutrophil percentage, presence of pleural effusion, and a quick sequential [sepsis-related] organ failure assessment score ≥ 2. The AUC was 0.918 in the derivation cohort. In the validation cohort, the AUC was 0.859, with a Hosmer– Lemeshow test p-value of 0.246 and a calibration slope of 0.759. The stratified likelihood ratio ranged from 0.04 to 9.71 and increased with higher model scores.

Conclusions

This model showed high discrimination and good calibration using objective variables that are readily available early after admission. Furthermore, this is the first model to predict IE based on the 2023 Duke-ISCVID criteria. Further multicenter validation in community hospitals would enhance generalizability.