Jesús D. López

Study overview

Association Between Gestational Diabetes and Congenital Hearing Loss

A comprehensive analysis of risk factors for congenital hearing loss in Queensland health data.

My contribution

I was responsible for the statistical work on this unpublished cohort study (~330,000 pregnancies from Queensland, Australia): data processing, regression modeling, diagnostics, and quantitative result reporting.

This page presents those quantitative outputs in an interactive format.

Study status

This study remains unpublished. The content here reflects my statistical contribution and should not be interpreted as a final peer-reviewed publication.

Loading study data…

Analytical approach

A pipeline from profiling to adjusted modeling.

The workflow moves in four steps: population profiling, risk screening, group comparison, and confounder-adjusted modeling.

Analysis strategy

Tap a card to flip

Statistical methods

What I used.

Each method shows the rationale on the front and the actual code on the back.

Methods

Tap a card to flip

Study population

Demographics

Characteristics of mothers and infants in the analysis.

Key finding

The cohort distribution is balanced across major demographic groups, supporting stable downstream association estimates.

Population overview

Loading demographics…

Clinical characteristics

Loading clinical data…

Additional variables — maternal BMI, ethnicity, hypertension status, delivery complications — are examined in the risk factor analysis below. Not every variable analyzed is presented here.

Risk factor landscape

Hearing loss rates by risk factor.

Univariate odds-ratio screening across maternal, infant, and pregnancy-related factors.

Analysis focus

This section reports univariate logistic-regression estimates. In later adjusted models, gestational diabetes remains the primary exposure variable for the main analysis.

Key finding

Several factors show elevated univariate odds ratios. These estimates are screening results and require multivariable adjustment before interpreting independent effects.

Hearing loss rates

Pick a view

Loading risk factors…

Primary predictor· Focal exposure
High risk· OR > 5
Moderate· OR 2–5
Low / none· OR ≤ 2

Primary predictor

1

All Types Diabetes — focus of study.

High risk

0

OR > 5 · strong associations.

Moderate risk

0

OR 2–5 · moderate associations.

Low / none

0

OR ≤ 2 · weak or no associations.

Interpretation context

The strongest univariate associations are concentrated in established congenital or genetic factors. Gestational diabetes shows a modest univariate association and therefore requires adjusted modeling to determine whether an independent signal remains after accounting for confounding variables.

Bivariate comparison

No diabetes vs. gestational diabetes.

Unadjusted hearing-loss rates between groups, with 95% confidence intervals.

Key finding

There is no statistically significant difference in hearing loss rates between gestational diabetes (0.330%) and no diabetes (0.320%) groups (p = 0.345).

Rate comparison

Chart + summary table

Loading comparison data…

Test statistic

χ² = 0.89

Chi-square test of independence.

P-value

p = 0.345

Probability of observing this difference by chance.

Conclusion

No association

p > 0.05 — no significant difference detected.

Study context

Hearing loss affects approximately 3.2 per 1,000 births in the general population. Both groups show rates consistent with this baseline in the unadjusted comparison. The chi-square statistic summarizes how far observed counts are from those expected if groups were truly similar, while the p-value quantifies how likely that difference is under the null model. Here p = 0.345 indicates the observed gap is compatible with random variation, so this bivariate step supports a no-association conclusion. Independent-effect assessment still requires multivariable adjustment for confounding variables.

Multivariable analysis

The complete picture.

Adjusted odds ratios accounting for confounding variables and their interactions.

Approach

Multivariable logistic regression controls for potential confounding variables and isolates the independent effect of gestational diabetes on hearing loss risk. Variables were selected using backward stepwise selection on AIC, keeping only statistically meaningful predictors while avoiding overfitting. This simultaneous adjustment reveals the true independent association between gestational diabetes and hearing loss.

Key finding

After controlling for all significant risk factors in a multivariable model, gestational diabetes shows no independent association with hearing loss (adjusted OR = 1.093, p = ).

Adjusted odds ratios

Forest plot · 95% CI

Loading forest plot…

Primary predictor· Focal exposure
Strong risk· OR > 5
Moderate· OR 2–5
Non-significant· p > 0.05

Primary predictor

0

Gestational diabetes — focus of study.

Strong risk

0

OR > 5 · highly significant associations.

Moderate risk

0

OR 2–5 · moderate associations.

Non-significant

0

p > 0.05 · no significant association.

Primary finding & interpretation

Primary finding. Gestational diabetes (OR = 1.093, p = ) shows no independent association with hearing loss after adjusting for 0 other potential risk factors. This null finding persists after controlling for confounding, confirming the bivariate result.

Significant risk factors. 0 factors show significant associations in the adjusted model. The strongest are . These predominantly represent non-modifiable genetic and congenital conditions rather than preventable maternal factors.

Model validation

Diagnostics & validation.

Comprehensive evaluation of model performance, discrimination, and calibration.

Approach

Validating a multivariable logistic regression model means checking three things: that it captures the relationship between predictors and outcome (fit), that it distinguishes positive from negative cases (discrimination), and that its predicted probabilities match reality (calibration).

Three pillars

Tap a card to flip

Key finding

The final multivariable model shows good calibration (Hosmer-Lemeshow p = 0.131) and moderate discrimination (C-statistic = 0.582), with a low Pseudo R² of 0.000 reflecting the rarity and complexity of congenital hearing loss. Despite modest predictive power, the model reliably confirms that gestational diabetes shows no independent association with hearing loss after controlling for confounders.

Model development & comparison

Loading model statistics…

AIC improvement

−0.0

Lower is better. Drop from baseline GDM model to final.

C-statistic gain

+0.0%

Discrimination ability gained.

Pseudo R²

0.000

Typical for rare outcomes.

Diagnostic tests

Calibration

Hosmer-Lemeshow test

χ² statistic
P-value
Interpretation

Generalization

Cross-validation

Mean AUC
95% CI
Interpretation
Consistent performance across validation folds.

Statistical interpretation & study context

Primary findings

    Study limitations

      Study conclusion

      Gestational diabetes does not independently increase the risk of congenital hearing loss after accounting for other risk factors. The evidence supports current clinical guidelines that do not recommend additional hearing screening for infants born to mothers with gestational diabetes.