A Comorbidity Model to Predict Inpatient Mortality Using Clinical Classifications Software with National Inpatient Sample Data 2020.

Authors' Affiliations

Hezborn Magacha, Department of Biostatistics and Epidemiology, College of Public Health, East Tennessee State University, Johnson City, TN. Adedeji Adenusi, Department of Biostatistics and Epidemiology, College of Public Health, East Tennessee State University, Johnson City, TN. Shimin Zheng, Department of Biostatistics and Epidemiology, College of Public Health, East Tennessee State University, Johnson City, TN. Sheryl Strasser, Department of Health Promotion & Behavior, Georgia State University School of Public Health, Atlanta, GA. Oluwatobi Adegbile, Department of Biostatistics and Epidemiology, College of Public Health, East Tennessee State University, Johnson City, TN.

Location

Culp Center Ballroom

Start Date

4-25-2023 9:00 AM

End Date

4-25-2023 11:00 AM

Poster Number

35

Faculty Sponsor’s Department

Biostatistics & Epidemiology

Name of Project's Faculty Sponsor

Shimin Zheng

Classification of First Author

Graduate Student-Master’s

Competition Type

Competitive

Type

Poster Presentation

Project's Category

Healthcare and Medicine

Abstract or Artist's Statement

Background.

In-hospital mortality is a measure recognized by US Agency for Healthcare Quality to represent quality of care within hospitals, that accounts for mortality based on three indicators: 1. select medical conditions and procedures; 2. procedures linked with questions of use (misuse, over/under use); 3. high volume procedures traditionally associated with lower mortality rates. Understanding how different comorbidity models measure in-hospital mortality is essential not only for determining patient health status in the hospital setting, but also help to regulating mortality risk and mortality risk predictions. One of the most widely used discriminatory models is the Charlson model, which predicts the risk of mortality within one year of hospitalization of patients with various comorbidities using CCSR codes for ICD-10 diagnoses which is quantified by the c-statistics, represented by the area under the curve (AUC).

Objectives.

To adapt a comorbidity index model to the National Inpatient Sample (NIS) database of 2020 to predict 1-year mortality for patients admitted with select ICD-10 codes of diagnoses.

Methods

Our study analysis examined mortality with comorbidity using the Charlson model in a sample population of estimated 5,533,477 adult inpatients (individuals ≥18 years of age). A multivariate logistic regression model was constructed with in-hospital mortality as the outcome variable and identifying predictor variables as defined by the Clinical Classifications Software Refined Variables (CCSR) codes for selected ICD-10 diagnoses (Table 3). Descriptive statistics and the base logistic regression analyses were conducted using SAS statistical software version 9.4. To avoid overpowering and avoid variables attaining statistical significance while only marginally changing the outcome, a subsample (n=100,000) was randomly selected from the original data set. Ultimately, 20 CCSR variables with p-values <0.20 from the base simple logistic regression models were included in the subsequent backward stepwise logistic regression analysis.

Results

Table 1 shows the prevalence of the selected diagnoses for our analysis. Anemia (28.32%), pulmonary disease (asthma, COPD, pneumoconiosis;21.88%), and diabetes without complications (19.47%) were the three most prevalent conditions among hospitalized patients. Table 2 shows the results of the base logistic regression analysis conducted, which excluded connective tissue/rheumatologic disorders, peptic ulcer disease, anemia, diabetes with complications, and human immunodeficiency as predictors of inpatient mortality. Results of the backward stepwise regression analysis revealed that severe liver disease/hepatic failure ([adjusted odds ratio (aOR): 10.50, (CI: 10.40-10.59)], acute myocardial infarction ([2.85, (2.83-2.87)] and malnutrition ([2.15, (2.14-2.16)] were three most important risk factors and had the highest impact on inpatient mortality (p-value <0.0001). However, smoking history, obesity, and liver disease were negatively associated with inpatient mortality. The c-statistic or the area under the curve (AUC) for the final model was 0.752.

Conclusion

Our findings, based on Charlson modeling procedures, indicate that independent variables representative of comorbidity with the strongest 1-year risk of mortality were among patients with ICD-10 codes relating to: severe liver disease/hepatic failure, acute myocardial infarction, and malnutrition. Hence, relevant stakeholders (patients, family members, and healthcare providers) can utilize this knowledge to advance models of care and prevention strategies that limit disease progression and improve patient outcomes.

This document is currently not available here.

Share

COinS
 
Apr 25th, 9:00 AM Apr 25th, 11:00 AM

A Comorbidity Model to Predict Inpatient Mortality Using Clinical Classifications Software with National Inpatient Sample Data 2020.

Culp Center Ballroom

Background.

In-hospital mortality is a measure recognized by US Agency for Healthcare Quality to represent quality of care within hospitals, that accounts for mortality based on three indicators: 1. select medical conditions and procedures; 2. procedures linked with questions of use (misuse, over/under use); 3. high volume procedures traditionally associated with lower mortality rates. Understanding how different comorbidity models measure in-hospital mortality is essential not only for determining patient health status in the hospital setting, but also help to regulating mortality risk and mortality risk predictions. One of the most widely used discriminatory models is the Charlson model, which predicts the risk of mortality within one year of hospitalization of patients with various comorbidities using CCSR codes for ICD-10 diagnoses which is quantified by the c-statistics, represented by the area under the curve (AUC).

Objectives.

To adapt a comorbidity index model to the National Inpatient Sample (NIS) database of 2020 to predict 1-year mortality for patients admitted with select ICD-10 codes of diagnoses.

Methods

Our study analysis examined mortality with comorbidity using the Charlson model in a sample population of estimated 5,533,477 adult inpatients (individuals ≥18 years of age). A multivariate logistic regression model was constructed with in-hospital mortality as the outcome variable and identifying predictor variables as defined by the Clinical Classifications Software Refined Variables (CCSR) codes for selected ICD-10 diagnoses (Table 3). Descriptive statistics and the base logistic regression analyses were conducted using SAS statistical software version 9.4. To avoid overpowering and avoid variables attaining statistical significance while only marginally changing the outcome, a subsample (n=100,000) was randomly selected from the original data set. Ultimately, 20 CCSR variables with p-values <0.20 from the base simple logistic regression models were included in the subsequent backward stepwise logistic regression analysis.

Results

Table 1 shows the prevalence of the selected diagnoses for our analysis. Anemia (28.32%), pulmonary disease (asthma, COPD, pneumoconiosis;21.88%), and diabetes without complications (19.47%) were the three most prevalent conditions among hospitalized patients. Table 2 shows the results of the base logistic regression analysis conducted, which excluded connective tissue/rheumatologic disorders, peptic ulcer disease, anemia, diabetes with complications, and human immunodeficiency as predictors of inpatient mortality. Results of the backward stepwise regression analysis revealed that severe liver disease/hepatic failure ([adjusted odds ratio (aOR): 10.50, (CI: 10.40-10.59)], acute myocardial infarction ([2.85, (2.83-2.87)] and malnutrition ([2.15, (2.14-2.16)] were three most important risk factors and had the highest impact on inpatient mortality (p-value <0.0001). However, smoking history, obesity, and liver disease were negatively associated with inpatient mortality. The c-statistic or the area under the curve (AUC) for the final model was 0.752.

Conclusion

Our findings, based on Charlson modeling procedures, indicate that independent variables representative of comorbidity with the strongest 1-year risk of mortality were among patients with ICD-10 codes relating to: severe liver disease/hepatic failure, acute myocardial infarction, and malnutrition. Hence, relevant stakeholders (patients, family members, and healthcare providers) can utilize this knowledge to advance models of care and prevention strategies that limit disease progression and improve patient outcomes.