Application of accelerated time models to compare performance of two comorbidity-adjusting methods with APACHE II in predicting short-term mortality among the critically ill

George Mnatzaganian, Melanie Bish, Jason Fletcher, Cameron Knott, John Stephenson

Research output: Contribution to journalArticleResearchpeer-review

Abstract

Objective: This study aimed to determine how the abilities of the Charlson Index and Elixhauser comorbidities compared with the chronic health components of the Acute Physiology and Chronic Health Evaluation (APACHE II) to predict in-hospital 30 day mortality among adult critically ill patients treated inside and outside of Intensive Care Unit (ICU). Methods: A total of 701 critically ill patients, identified in a prevalence study design on four randomly selected days in five acute care hospitals, were followed up from the date of becoming critically ill for 30 days or until death, whichever occurred first. Multiple data sources including administrative, clinical, pathology, microbiology and laboratory patient records captured the presence of acute and chronic illnesses. The exponential, Gompertz, Weibull, and log-logistic distributions were assessed as candidate parametric distributions available for the modelling of survival data. Of these, the log-logistic distribution provided the best fit and was used to construct a series of parametric survival models. Results: Of the 701 patients identified in the initial prevalence study, 637 (90.9%) had complete data for all fields used to calculate APACHE II score. Controlling for age, sex and Acute Physiology Score (APS), the chronic health components of the APACHE II score, as a group, were better predictors of survival than Elixhauser comorbidities and Charlson Index. Of the APACHE II chronic health components, only the relatively uncommon conditions of liver failure (3.4%) and immunodeficiency (9.6%) were statistically associated with inferior patient survival with acceleration factors of 0.35 (95% CI 0.17, 0.72) for liver failure, and 0.42 (95% CI 0.26, 0.72) for immunodeficiency. Sensitivity analyses on an imputed dataset that also included the 64 individuals with imputed APACHE II score showed identical results. Conclusion: Our study suggests that, in acute critical illness, most co-existing comorbidities are not major determinants of short-term survival, indicating that observed variations in ICU patient 30-day mortality may not be confounded by lack of adjustment to pre-existing comorbidities.

Original languageEnglish
Pages (from-to)81-88
Number of pages8
JournalMethods of Information in Medicine
Volume57
Issue number1
DOIs
Publication statusPublished - 1 Jan 2018

Keywords

  • Accelerated time modelling
  • APACHE II
  • Charlson index
  • Critically ill
  • Elixhauser comorbidities
  • In-hospital mortality

Cite this

@article{385a95834ad74ee6b6810cab01b5f0a7,
title = "Application of accelerated time models to compare performance of two comorbidity-adjusting methods with APACHE II in predicting short-term mortality among the critically ill",
abstract = "Objective: This study aimed to determine how the abilities of the Charlson Index and Elixhauser comorbidities compared with the chronic health components of the Acute Physiology and Chronic Health Evaluation (APACHE II) to predict in-hospital 30 day mortality among adult critically ill patients treated inside and outside of Intensive Care Unit (ICU). Methods: A total of 701 critically ill patients, identified in a prevalence study design on four randomly selected days in five acute care hospitals, were followed up from the date of becoming critically ill for 30 days or until death, whichever occurred first. Multiple data sources including administrative, clinical, pathology, microbiology and laboratory patient records captured the presence of acute and chronic illnesses. The exponential, Gompertz, Weibull, and log-logistic distributions were assessed as candidate parametric distributions available for the modelling of survival data. Of these, the log-logistic distribution provided the best fit and was used to construct a series of parametric survival models. Results: Of the 701 patients identified in the initial prevalence study, 637 (90.9{\%}) had complete data for all fields used to calculate APACHE II score. Controlling for age, sex and Acute Physiology Score (APS), the chronic health components of the APACHE II score, as a group, were better predictors of survival than Elixhauser comorbidities and Charlson Index. Of the APACHE II chronic health components, only the relatively uncommon conditions of liver failure (3.4{\%}) and immunodeficiency (9.6{\%}) were statistically associated with inferior patient survival with acceleration factors of 0.35 (95{\%} CI 0.17, 0.72) for liver failure, and 0.42 (95{\%} CI 0.26, 0.72) for immunodeficiency. Sensitivity analyses on an imputed dataset that also included the 64 individuals with imputed APACHE II score showed identical results. Conclusion: Our study suggests that, in acute critical illness, most co-existing comorbidities are not major determinants of short-term survival, indicating that observed variations in ICU patient 30-day mortality may not be confounded by lack of adjustment to pre-existing comorbidities.",
keywords = "Accelerated time modelling, APACHE II, Charlson index, Critically ill, Elixhauser comorbidities, In-hospital mortality",
author = "George Mnatzaganian and Melanie Bish and Jason Fletcher and Cameron Knott and John Stephenson",
year = "2018",
month = "1",
day = "1",
doi = "10.3414/ME17-01-0097",
language = "English",
volume = "57",
pages = "81--88",
journal = "Methods of Information in Medicine",
issn = "0026-1270",
number = "1",

}

Application of accelerated time models to compare performance of two comorbidity-adjusting methods with APACHE II in predicting short-term mortality among the critically ill. / Mnatzaganian, George; Bish, Melanie; Fletcher, Jason; Knott, Cameron; Stephenson, John.

In: Methods of Information in Medicine, Vol. 57, No. 1, 01.01.2018, p. 81-88.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Application of accelerated time models to compare performance of two comorbidity-adjusting methods with APACHE II in predicting short-term mortality among the critically ill

AU - Mnatzaganian, George

AU - Bish, Melanie

AU - Fletcher, Jason

AU - Knott, Cameron

AU - Stephenson, John

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Objective: This study aimed to determine how the abilities of the Charlson Index and Elixhauser comorbidities compared with the chronic health components of the Acute Physiology and Chronic Health Evaluation (APACHE II) to predict in-hospital 30 day mortality among adult critically ill patients treated inside and outside of Intensive Care Unit (ICU). Methods: A total of 701 critically ill patients, identified in a prevalence study design on four randomly selected days in five acute care hospitals, were followed up from the date of becoming critically ill for 30 days or until death, whichever occurred first. Multiple data sources including administrative, clinical, pathology, microbiology and laboratory patient records captured the presence of acute and chronic illnesses. The exponential, Gompertz, Weibull, and log-logistic distributions were assessed as candidate parametric distributions available for the modelling of survival data. Of these, the log-logistic distribution provided the best fit and was used to construct a series of parametric survival models. Results: Of the 701 patients identified in the initial prevalence study, 637 (90.9%) had complete data for all fields used to calculate APACHE II score. Controlling for age, sex and Acute Physiology Score (APS), the chronic health components of the APACHE II score, as a group, were better predictors of survival than Elixhauser comorbidities and Charlson Index. Of the APACHE II chronic health components, only the relatively uncommon conditions of liver failure (3.4%) and immunodeficiency (9.6%) were statistically associated with inferior patient survival with acceleration factors of 0.35 (95% CI 0.17, 0.72) for liver failure, and 0.42 (95% CI 0.26, 0.72) for immunodeficiency. Sensitivity analyses on an imputed dataset that also included the 64 individuals with imputed APACHE II score showed identical results. Conclusion: Our study suggests that, in acute critical illness, most co-existing comorbidities are not major determinants of short-term survival, indicating that observed variations in ICU patient 30-day mortality may not be confounded by lack of adjustment to pre-existing comorbidities.

AB - Objective: This study aimed to determine how the abilities of the Charlson Index and Elixhauser comorbidities compared with the chronic health components of the Acute Physiology and Chronic Health Evaluation (APACHE II) to predict in-hospital 30 day mortality among adult critically ill patients treated inside and outside of Intensive Care Unit (ICU). Methods: A total of 701 critically ill patients, identified in a prevalence study design on four randomly selected days in five acute care hospitals, were followed up from the date of becoming critically ill for 30 days or until death, whichever occurred first. Multiple data sources including administrative, clinical, pathology, microbiology and laboratory patient records captured the presence of acute and chronic illnesses. The exponential, Gompertz, Weibull, and log-logistic distributions were assessed as candidate parametric distributions available for the modelling of survival data. Of these, the log-logistic distribution provided the best fit and was used to construct a series of parametric survival models. Results: Of the 701 patients identified in the initial prevalence study, 637 (90.9%) had complete data for all fields used to calculate APACHE II score. Controlling for age, sex and Acute Physiology Score (APS), the chronic health components of the APACHE II score, as a group, were better predictors of survival than Elixhauser comorbidities and Charlson Index. Of the APACHE II chronic health components, only the relatively uncommon conditions of liver failure (3.4%) and immunodeficiency (9.6%) were statistically associated with inferior patient survival with acceleration factors of 0.35 (95% CI 0.17, 0.72) for liver failure, and 0.42 (95% CI 0.26, 0.72) for immunodeficiency. Sensitivity analyses on an imputed dataset that also included the 64 individuals with imputed APACHE II score showed identical results. Conclusion: Our study suggests that, in acute critical illness, most co-existing comorbidities are not major determinants of short-term survival, indicating that observed variations in ICU patient 30-day mortality may not be confounded by lack of adjustment to pre-existing comorbidities.

KW - Accelerated time modelling

KW - APACHE II

KW - Charlson index

KW - Critically ill

KW - Elixhauser comorbidities

KW - In-hospital mortality

UR - http://www.scopus.com/inward/record.url?scp=85046161564&partnerID=8YFLogxK

U2 - 10.3414/ME17-01-0097

DO - 10.3414/ME17-01-0097

M3 - Article

VL - 57

SP - 81

EP - 88

JO - Methods of Information in Medicine

JF - Methods of Information in Medicine

SN - 0026-1270

IS - 1

ER -