Statistical Test Selection for Pediatric Obesity Research
For a study examining the relationship between pediatric groups and obesity with multiple objectives and more than 3 diagnostic criteria, multinomial logistic regression (Option D) is the most appropriate statistical test.
Rationale for Multinomial Logistic Regression
When the outcome variable has more than two categories (as indicated by "more than 3 diagnostic criteria"), multinomial logistic regression is specifically designed to handle multiple categorical outcomes simultaneously. 1
Key Considerations:
Multiple outcome categories: The question explicitly states "more than 3 criteria to diagnose," which indicates a polytomous (multi-category) dependent variable rather than a simple binary outcome 1
Multiple risk factors: The expanded question mentions "multiple risk factors," and multinomial regression efficiently analyzes the effect of multiple independent variables on categorical outcomes by quantifying each variable's unique contribution 1
Pediatric obesity classification: Clinical obesity assessment in children involves multiple classification categories (normal weight, overweight, obesity, severe obesity) based on BMI percentiles, which naturally creates more than two outcome categories 2, 3
Why Other Options Are Inappropriate:
T-test (Option A) - Incorrect
- T-tests compare means between two groups and cannot handle multiple outcome categories or multiple predictor variables simultaneously 4
- Inappropriate for examining relationships between multiple risk factors and categorical obesity outcomes
Linear Regression (Option B) - Incorrect
- Linear regression is designed for continuous dependent variables, not categorical outcomes 1
- Obesity classification based on BMI percentiles creates categorical groups (normal, overweight, obese, severely obese), not a continuous outcome 2
Logistic Regression (Option C) - Insufficient
- Standard logistic regression handles only binary outcomes (two categories: disease present vs. absent) 1, 5
- While powerful for dichotomous outcomes, it cannot accommodate the "more than 3 diagnostic criteria" specified in the question 1
- Would require collapsing multiple obesity categories into just two groups, losing important clinical distinctions 2
Statistical Framework for Multinomial Analysis:
Multinomial logistic regression extends binary logistic regression principles to handle multiple outcome categories by:
- Creating separate logistic regression equations for each outcome category relative to a reference category 1
- Allowing simultaneous assessment of how multiple independent variables (pediatric risk factors) relate to each obesity classification category 1
- Providing odds ratios with 95% confidence intervals for each predictor-outcome relationship 1
Essential Model Requirements:
- Adequate sample size: Minimum 10-20 events per independent variable to avoid overfitting 1
- Independence of observations: Each child should be counted only once 1
- Absence of multicollinearity: Risk factors should not be highly correlated with each other 1
- Appropriate reference category selection: Typically the "normal weight" category serves as the reference 2
Clinical Application Context:
The meta-analyses examining pediatric obesity used various statistical approaches including pooled risk differences for obesity prevalence changes, demonstrating the importance of selecting methods appropriate to the outcome structure 4. When obesity is classified into multiple categories based on BMI percentiles (85th-94th percentile for overweight, ≥95th percentile for obesity, ≥120% of 95th percentile for severe obesity), multinomial regression becomes essential 2.