Covariate-Adjusted Optimal PMA Analysis

This analysis repeats the optimal PMA protocol methods with adjustment for identified confounders. By comparing these adjusted results with the unadjusted analyses, we can determine whether the statistical relationships are robust to confounding.

Covariates included in all adjusted models: - gestational_age_weeks: Gestational age at birth measured in weeks - birth_weight_grams: Birth weight measured in grams - mechanical_ventilation: History of mechanical ventilation - o2_device_at_36_weeks: Respiratory support requirement at 36 weeks

Note: Birth weight missing values are imputed with median.

Data Completeness

Sample Size

• Original sample (post-exclusions): 228
• Complete cases for adjusted analysis: 228
• Cases dropped due to missing covariates: 0

Method 1: Adjusted Slope Comparison

Comparing adjusted slopes before and after 35 weeks PMA, controlling for confounders.

Adjusted Regression Results

Early Group (< 35 weeks PMA, n=166):

• Adjusted slope: 0.532 (SE: 0.176)
• p-value: 0.0029

Late Group (≥ 35 weeks PMA, n=62):

• Adjusted slope: 0.971 (SE: 0.228)
• p-value: 0.0001

Adjusted Slope Difference Test

• Interaction term coefficient: 0.502
• Interaction p-value: 0.0517
• Statistical significance: Not significant difference in adjusted slopes

Coefficient Comparison:

• Unadjusted slopes: <35w = 0.66, ≥35w = 1.23
• Adjusted slopes: <35w = 0.53, ≥35w = 0.97
• Change after adjustment: Early slope Δ = -0.13, Late slope Δ = -0.26

Model diagnostics note: Potential issues with multicollinearity, non-normal residuals detected in one or both groups.

Adjusted Slope Visualization

Comparison of adjusted slopes and their relationship to unadjusted estimates.

Method 2: Adjusted Cutpoint Analysis

Testing cutpoints with adjustment for confounders to find optimal PMA threshold.

Adjusted Cutpoint Comparison

Cutpoint	N Early	N Late	Adj Mean Early	Adj Mean Late	Difference	p-value
33	35	193	17.2	11.82	-5.38	0.0188
33.5	68	160	14.36	11.98	-2.38	0.1549
34	110	118	14.08	10.94	-3.13	0.0545
34.5	138	90	13.41	11.33	-2.08	0.2356
35	166	62	14.33	9.17	-5.16	0.0099
35.5	191	37	13.98	7.92	-6.06	0.0123

Optimal Cutpoint: 34.5 weeks PMA

• Minimizes adjusted mean time to full feeding for early starters
• Adjusted difference: -2.08 days
• Non-significant after adjustment (p = 0.2356)

Cutpoint Comparison:

• Unadjusted optimal: 35.5 weeks
• Adjusted optimal: 34.5 weeks
• Difference: -1.0 weeks

Adjusted Cutpoint Visualization

Adjusted mean outcomes by cutpoint after controlling for confounders.

Method 3: Adjusted Continuous Optimization

Polynomial model with covariate adjustment to find optimal PMA.

Adjusted Model Results

• Linear coefficient: -37.445 (p = 0.0217)
• Quadratic coefficient: 0.513 (p = 0.0265)
• Model R²: 0.154

Adjusted Optimal Timing

• Optimal PMA: 36.5 weeks
• Expected days to FOF at optimal: 9.2 days
• Optimal zone: 32.8 - 39.8 weeks PMA

Optimal PMA Comparison:

• Unadjusted optimal: 35.2 weeks
• Adjusted optimal: 36.5 weeks
• Difference: 1.3 weeks

Model diagnostics note: Multicollinearity, non-normal residuals detected in polynomial model.

Adjusted Optimization Visualization

Polynomial model with optimal zone after covariate adjustment.

Comparison Framework

Analysis Comparison Framework

Methodological Differences:

1. Sample size: Complete cases analysis after covariate imputation
2. Statistical models: All models include confounder adjustment
3. Coefficient estimates: Direct comparison of adjusted vs unadjusted parameters
4. Significance testing: P-values and confidence intervals for all comparisons

Statistical Comparison Points:

• Slope coefficients before and after adjustment
• Optimal cutpoint identification with and without covariates
• Model fit statistics (R²) across adjustment levels