Using Multilevel Negative Binomial Modeling to Detect Active Smoking in Colorectal Cancer Screening

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Multilevel analysis is widely used, but its application in colorectal cancer (CRC) screening, particularly at the community level, requires clarification.
• This study addresses the need for a practical guide on using multilevel negative binomial analysis for CRC screening data.

Purpose of the Study:

• To explain the application of multilevel negative binomial analysis in CRC screening.
• To compare multilevel negative binomial analysis with standard negative binomial methods using real-world data.

Main Methods:

• Analysis of 2,475 fecal immunochemical test (FIT) cases from a CRC screening randomized controlled trial in Thailand.
• Comparison of standard negative binomial and multilevel negative binomial approaches, considering active smoking and fecal hemoglobin (f-Hb) concentration.

Main Results:

• The standard negative binomial method showed active smoking was significantly associated with f-Hb concentration (IRRadj = 1.47).
• The multilevel negative binomial approach yielded non-significant results for active smoking and f-Hb concentration (IRRadj = 1.30).
• Significant differences in effect magnitude and value were observed between the two statistical methods.

Conclusions:

• Standard statistical approaches may be inadequate for CRC screening data with zero-inflated or over-dispersed outcomes.
• Hierarchical data structures and contextual factors necessitate the use of multilevel modeling.
• Over-dispersion in f-Hb concentration suggests the utility of multilevel modeling for improved statistical analysis in future studies.