Zero-Inflated Count Regression Models in Solving Challenges Posed by Outlier-Prone Data; an Application to Length of...

Area of Science:

• Biostatistics
• Health Services Research
• Data Science

Background:

• Length of stay (LOS) data frequently exhibit outliers and skewness, potentially compromising analytical accuracy.
• Traditional statistical models may produce misleading results when confronted with outlier-prone count data.
• Robust methodologies are essential for reliable analysis of complex healthcare datasets.

Purpose of the Study:

• To evaluate the efficacy of Zero-Inflated Poisson (ZIP) and robust Zero-Inflated Poisson (RZIP) models for analyzing outlier-prone LOS data.
• To identify significant predictors of LOS in intensive care unit (ICU) patients using robust statistical techniques.
• To demonstrate the advantages of RZIP over ZIP models in handling skewed and outlier-rich count data.

Main Methods:

• Utilized Zero-Inflated Poisson (ZIP) and robust Zero-Inflated Poisson (RZIP) models.
• Employed the Robust Expectation-Solution (RES) algorithm within the RZIP model for enhanced parameter estimation.
• Analyzed data from 254 ICU patients, including demographic and clinical variables.

Main Results:

• The RZIP model identified age, comorbidities, and insurance status as significant predictors of LOS.
• The RZIP model demonstrated superior performance compared to the ZIP model, indicated by lower Akaike Information Criteria (AIC) and Bayesian Information Criteria (BIC) values.
• Analysis revealed that 9.45% of cases displayed outliers, highlighting the need for robust methods.

Conclusions:

• The RZIP model provides a more accurate and reliable analysis of LOS data, especially in the presence of outliers.
• Findings from the RZIP model can inform hospital management and resource allocation strategies.
• Robust statistical modeling is crucial for uncovering meaningful insights from skewed and outlier-prone healthcare data.