The proper application of logistic regression model in complex survey data: a systematic review

Area of Science:

• Statistical methodology
• Survey data analysis
• Biostatistics

Background:

• Logistic regression is a widely used statistical method for binary outcomes in healthcare, marketing, and finance.
• Complex survey data, characterized by intricate sampling designs, present unique methodological challenges for logistic regression analysis.
• Overlooking these specific issues can lead to inaccurate insights and conclusions.

Purpose of the Study:

• To systematically review the application of logistic regression in complex survey data analysis.
• To identify common methodological shortcomings in studies utilizing Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS).
• To highlight the need for improved standards in reporting and analysis.

Main Methods:

• A systematic review was conducted using PubMed and ScienceDirect databases (January 2015 - December 2021).
• The review adhered to PRISMA 2020 guidelines, focusing on studies using DHS and MICS data.
• Methodological issues examined included model adequacy, data dependencies, survey design utilization, missing values, outliers, and goodness-of-fit assessments.

Main Results:

• A significant majority of studies (94.8%) failed to report model validation techniques, and 75.3% omitted goodness-of-fit assessments.
• Proper handling of complex survey design elements (weights, PSUs, strata) was inconsistent, with 41.7% of studies not using these variables, potentially biasing results.
• Inadequate attention was paid to data dependencies (only 19.7% used multilevel models), outliers (95.8% not mentioned), and missing data (41.0% corrected, 2.7% imputed).

Conclusions:

• Significant gaps exist in the rigorous application of logistic regression to complex survey data, particularly concerning data dependencies, survey design, and validation.
• Neglect of outliers, missing data handling, and goodness-of-fit assessments further compromises the reliability of published findings.
• There is a critical need for enhanced methodological standards and transparent reporting to ensure the robustness of results derived from complex survey data.