<aside>

After preprocessing, we input the z-scores into SuStaIn. This page highlights three key points:

1. Main SuStaIn results — the optimal number of subtypes and the ordering of lobes
2. Baseline differences — covariates and NIH Toolbox measures compared between the two subtypes
3. Follow-up trajectories — longitudinal patterns of BMI and CBCL across subtypes </aside>

1. Main Results

The plots below illustrate the reconstructed progression sequences of regional brain changes at the lobe level, derived using the SuStaIn model.

On the left, case participants (i.e., children exposed to GDM) are stratified into two distinct subtypes, each characterized by a different temporal ordering of affected lobes. The optimal number of subtypes was determined using cross-validation and information criteria.

On the right, the ordering derived from healthy controls is shown for comparison. Together, these visualizations highlight both the heterogeneity among exposed individuals and their divergence from the control trajectory

🔸 Subtype 1 ( n = 372, 65%):

insula → occipital_lobe → parietal_lobe → temporal_lobe → frontal_lobe → cingulate 🔸 Subtype 2 ( n = 202, 35%):

frontal_lobe → parietal_lobe → temporal_lobe → occipital_lobe → cingulate → insula

🔸 HC ( n = 3557)

occipital_lobe → parietal_lobe → frontal_lobe → temporal_lobe → insula → cingulate

2. Baseline Distribution

Covariates

The only baseline covariate that differed between the two subtypes was gestational age : subtype 1, 38.4 (2.9); subtype 2, 38.9 (2.3).

The total is larger than subtype 1 + subtype 2 bc we hide the HCs distribution.

NIH

Since the NIH follow-up data have a lot of missing values, we only looked at the baseline distribution. The CBCL results will be covered in the next section.

No significant difference was observed.

The total is larger than subtype 1 + subtype 2 bc we hide the HCs distribution.

3. Follow-up Distribution

The bar plot on the right shows the sample size at baseline (event = 0) and across annual follow-up visits. Starting with 4,131 participants in total (574 cases vs. 3,557 controls), the sample size in each category decreased over time.

Subtype 1 and Subtype 2 groups:

• Begin with moderate numbers (~350 and ~200 at baseline, respectively).
• Experience gradual attrition over time.
• By event 4, sample sizes reduce to 163 (Subtype 1) and 93 (Subtype 2).

In the following sections, based on the SuStaIn results and the assigned subtype labels, which was built solely using baseline data. We further examine how key measures differ across subtypes over time. Specifically, we ask whether, beyond the observed ordering differences in the case group, additional disparities emerge in other clinical measures, and whether such differences persist or even diverge as time progresses. In other words, we hypothesize that the distinct ordering of brain structural changes is associated with different clinical trajectories.