Statistical Models for the Analysis of Zero-Inflated Pain Intensity Numeric Rating Scale Data

Area of Science:

• Pain research
• Biostatistics
• Musculoskeletal disorders

Background:

• Pain intensity is commonly measured using the 0-10 numeric rating scale (NRS).
• NRS data often exhibit a high proportion of zero scores and right-skewed distributions.
• Standard statistical methods for normally distributed data are frequently applied, potentially leading to inaccurate analyses.

Purpose of the Study:

• To compare various statistical models for analyzing NRS pain intensity data.
• To identify the most suitable models for NRS data with a substantial number of zero scores.
• To evaluate model fit, interpretability, and the impact on conclusions regarding predictor effects.

Main Methods:

• An observational cross-sectional study of 18,935 veterans with painful musculoskeletal disorders.
• Comparison of linear regression, generalized linear models (Poisson, negative binomial), zero-inflated models, hurdle models, and cumulative logit models.
• Assessment of model fit and interpretability for NRS pain intensity data.

Main Results:

• 34% of patients reported no pain (NRS=0).
• Models designed to handle zero-inflated data demonstrated a superior fit compared to other tested models.
• Conclusions regarding predictor effects remained consistent across models, but zero-inflated models offered better data representation.

Conclusions:

• Statistical models that accommodate zero inflation are recommended for analyzing NRS pain intensity data, especially when a large proportion of patients report no pain.
• These models provide a better fit and more accurate representation of NRS data characteristics.
• Consideration of these specialized models can improve the reliability of pain research and clinical assessments.