Testing for treatment related trend with partially exchangeable clustered data

Area of Science:

• Toxicology
• Biostatistics
• Developmental Biology

Background:

• Developmental toxicity studies often prioritize primary outcomes (e.g., skeletal malformations) over secondary effects (e.g., reduced fetal weight).
• Secondary effects like weight reduction can occur in both malformed and normal fetuses, complicating analysis.
• Existing methods may not fully capture treatment-related trends when considering multiple outcome types.

Purpose of the Study:

• To develop and present a likelihood ratio procedure for constructing treatment-related trend tests in developmental experiments.
• To analyze both primary (malformations) and secondary (weight) outcomes simultaneously.
• To provide a robust statistical framework for developmental toxicity studies.

Main Methods:

• Utilized a likelihood ratio procedure for trend testing.
• Assumed partial exchangeability of data within clusters.
• Employed categorical covariates for malformation presence/absence.
• Considered two covariance models: general (cluster-specific matrices) and homogeneous (common matrix).
• Obtained Maximum Likelihood Estimates (MLEs) via direct maximization or iterative procedures.

Main Results:

• The proposed likelihood ratio procedure effectively constructs treatment-related trend tests for developmental toxicity data.
• Both general and homogeneous covariance models were successfully applied.
• Maximum likelihood estimates were obtained for model parameters under different covariance assumptions.

Conclusions:

• The developed statistical method offers a more comprehensive approach to analyzing developmental toxicity studies.
• It allows for the detection of treatment effects on both primary and secondary outcomes.
• The methodology is applicable to real-world developmental toxicity data, as demonstrated by two case studies.