Mapping the Dynamic Interplay of Mental Health and Weight Across Childhood: Data-Driven Explorations Using Causal Discovery

Using causal discovery on a large Danish birth cohort, this study reveals that while childhood weight strongly influences future weight status, there is no direct causal link between weight and mental health, suggesting their observed association is likely driven by confounding factors rather than a direct causal mechanism.

The Gist

Imagine you are trying to understand a massive, tangled ball of yarn. One end of the string represents a child's weight, and the other end represents their mental health. For years, doctors and scientists have been pulling on these ends, trying to figure out: Does being heavy cause sadness? Or does feeling sad make you gain weight?

Most studies have tried to untangle this by looking at just two or three strands at a time. But this new study, led by researchers in Denmark, decided to look at the entire ball of yarn at once. They used a special computer "detective" tool to map out how 67 different factors (like weight, stress, sleep, and family history) interacted over 18 years, from before a baby is born until they become a young adult.

Here is the story of what they found, explained simply:

1. The Detective Tool: "Causal Discovery"

Think of the researchers as detectives trying to solve a crime. Usually, they look at a suspect (like "eating too much sugar") and ask, "Did this cause the crime (obesity)?"

But this study used a new kind of detective work called Causal Discovery. Instead of guessing who the culprit is, they fed the computer a massive database of 67,000 lives. The computer then tried to build a map of cause-and-effect, like a subway map showing exactly which station leads to which. It asked: "If we know what happened at age 7, can we predict what happens at age 18?"

2. The Big Surprise: Boys and Girls Are More Alike Than Different

The researchers wondered if the "map" for boys was totally different from the map for girls. Maybe boys' weight affects their mental health in one way, while girls' weight affects it in another?

The verdict: The maps were almost identical. The computer found that the rules of the game are the same for both boys and girls. There was no statistically significant difference in how their bodies and minds interacted.

3. The Weight Story: It's a Long Chain, Not a Jump

The study looked at weight like a relay race.

• The Findings: A child's weight in early childhood directly influences their weight in early adolescence, which then influences their weight in late adolescence. It's a chain reaction.
• The Twist: They found a "shortcut" in the chain. A child's weight at age 7 had a direct link to their weight at age 18, skipping the middle step. This suggests that once a weight pattern is set in childhood, it's hard to break, even if things look okay for a few years in between.
• The Mom Factor: The mother's weight was a huge "upstream" influence. If the mother gained weight during pregnancy or was heavier, it set the stage for the child's weight later on. Interestingly, the father's weight didn't seem to have this same direct "push."

4. The Mental Health Story: Stress is the Hub

If weight is a long chain, mental health is more like a busy city intersection.

• The Findings: Mental health issues didn't jump from childhood straight to adulthood. Instead, they moved step-by-step.
• The Key Player: Stress in early adolescence (around age 11) was the "traffic hub." Almost all the roads of mental health problems (like anxiety or depression) went through this intersection. If you can fix the stress at age 11, you might prevent a traffic jam of mental health issues later in life.

5. The Big "Aha!" Moment: No Direct Link Between Weight and Mood

This is the most important part of the story.

For a long time, people thought: Heavy kids get bullied and become depressed OR Depressed kids eat too much and get heavy.

The computer said: "Nope."

When the researchers looked at the map, they found no direct road connecting "Weight" to "Mental Health Problems" (like anxiety or ADHD). They didn't cause each other directly.

The Analogy: Imagine a tree. The roots are deep underground (genetics, family environment, socio-economic status). The tree has two branches: one branch is "Weight" and the other is "Mental Health."
The branches look like they are touching, but they aren't growing from each other. They are both growing because of the same roots.

• If a family has high stress or poor nutrition (the roots), the child might gain weight and struggle with mental health.
• The weight didn't cause the stress, and the stress didn't cause the weight. They are just "siblings" caused by the same family environment.

6. What About Lifestyle?

The study did find one clear lifestyle culprit: Screen Time.
Spending too much time on screens in early adolescence was a direct cause of mental health struggles later on. This is a clear target for parents and teachers to focus on.

The Takeaway for Everyday Life

This study is like a GPS for parents and doctors. It tells us:

1. Don't blame the child: If a child is struggling with weight and mood, it's likely not because one caused the other. It's likely because of the environment they grew up in (the "roots").
2. Focus on the "Hub": If you want to help a child's mental health, pay extra attention to stress levels around age 11. That is the critical turning point.
3. Start Early with Weight: Weight patterns are set very early. Helping a mother manage her weight during pregnancy and ensuring a child has a healthy start can prevent issues decades later.
4. Cut the Screen Time: Reducing screen time is a proven way to help mental health.

In short, the researchers didn't just find a link; they found that the link we thought existed might be an illusion caused by other hidden factors. By understanding the whole map, we can stop chasing the wrong clues and start fixing the real problems.

Technical Summary

1. Problem Statement

Childhood overweight and obesity are increasingly prevalent and frequently co-occur with mental health issues (e.g., anxiety, depression, ADHD, conduct problems). While associations are well-documented, the dynamic, causal mechanisms linking weight and mental health across development remain unclear. Specifically, it is unknown:

• Whether causal pathways differ between sexes (boys vs. girls).
• Whether the relationship is bidirectional (does weight cause mental health issues or vice versa?) or driven by unmeasured confounding.
• How these processes evolve over time from pregnancy to late adolescence.

Traditional studies often focus on single exposure-outcome pairs or pre-specified models, failing to capture the complex, high-dimensional, and time-dependent nature of these systems.

2. Methodology

The study employed a fully data-driven causal discovery approach using longitudinal data from the Danish National Birth Cohort (DNBC).

• Study Population:

  • Source: DNBC (born 1996–2003).
  • Sample: $n = 67,593$ (after exclusions for congenital anomalies, CNS disorders, etc.).
  • Stratification: Analyses were conducted separately for boys ($n \approx 38,645$) and girls ($n \approx 36,458$).
  • Handling Missing Data: Used single imputation (random forests) for partial missingness and data cloning to account for non-random dropout/skipping of data collections.

• Variables & Temporal Structure:

  • 67 Variables categorized into four domains: Weight (offspring/parental), Mental Health, Lifestyle Risk Factors, and Other (confounders).
  • 6 Temporal Tiers: Pregnancy, Birth, Infancy, Childhood (age 7), Early Adolescence (age 11), and Late Adolescence (age 18).

• Causal Discovery Algorithm:

  • Algorithm: Temporal Greedy Equivalence Search (TGES).
  • Assumptions: No unobserved confounders; strict temporal ordering (causes must precede effects).
  • Output: A Temporal Maximally Oriented Partially Directed Acyclic Graph (TMPDAG). This represents a family of DAGs consistent with the data. Directed edges indicate causal direction; undirected edges indicate ambiguity in direction given the data.
  • Scoring: Discrete Bayesian Information Criterion (BIC). The penalty term was tuned to yield a graph with exactly 300 edges to ensure plausibility and informativeness.
  • Stability: Validated via 100 bootstrap repetitions. Edges were only retained if present in $\ge 80\%$ of bootstrap samples.

• Comparative Analysis:

  • Sex Differences: Compared TMPDAGs for boys and girls using Structural Hamming Distance (SHD) and a permutation test to determine statistical significance of structural differences.
  • Subgraph Extraction: Derived specific subgraphs (Weight-only, Mental Health-only, and Weight/Mental Health/Lifestyle) to visualize flow of causal information.

• Tools: Implemented in R (v4.4) using the causalDisco package. An interactive web application was developed to visualize results.

3. Key Results

• Sex Differences:

  • The permutation test yielded $P = 0.079$, indicating no statistically significant difference in the overall causal graph structures between boys and girls at the 5% level.
  • While 142 edges were shared with identical orientation, 20 edges were shared but with different orientations, and some edges were sex-specific.

• Weight Development:

  • Continuity: Offspring weight is causally connected from childhood through early adolescence to late adolescence.
  • Lagged Effects: A direct causal pathway exists from Childhood Weight $\rightarrow$ Late Adolescent Weight, suggesting a "sensitive period" not fully mediated by early adolescence weight.
  • Maternal Influence: Maternal weight (pregnancy and gestational weight gain) acts as a cause for paternal and offspring weight. Paternal weight did not show direct causal influence on offspring.
  • Birth/Infancy: No direct causal pathways were found from "Weight for Gestational Age" or "Infancy Weight" to subsequent weight; these relationships appear driven by common causes (e.g., maternal weight).

• Mental Health Development:

  • Temporal Dynamics: Causal pathways occurred primarily within or between adjacent time tiers, suggesting mental health dynamics are more rapid and malleable than weight development.
  • Key Node: Stress in Early Adolescence (Age 11) emerged as a critical hub, particularly for boys, acting as a source or sink for numerous causal pathways.
  • No Long-Term Direct Paths: No direct causal paths were found from childhood to late adolescence that bypassed early adolescence.

• Interplay Between Weight and Mental Health:

  • No Direct Causality: The analysis found no direct causal pathways between weight and internalizing/externalizing problems (or lifestyle factors) in either direction.
  • Confounding: The observed associations between weight and mental health are likely attributable to confounding factors (e.g., genetic or shared environmental factors) rather than direct causation.
  • Lifestyle: Screen time in early adolescence was the only lifestyle factor with a direct causal path to internalizing/externalizing problems in late adolescence.

4. Key Contributions

1. Methodological Innovation:

   • First application of TGES to a large birth cohort (67k participants) with 67 variables across 6 time points to model complex health systems.
   • Developed a novel subgraph construction procedure to visualize causal flow within complex TMPDAGs without losing information from the full model.
   • Created an interactive web application allowing researchers to query the estimated causal graphs, test specific pathways, and explore subgroup differences dynamically.

2. Subgroup Comparison Framework:

   • Demonstrated a rigorous statistical framework (SHD + Permutation Test) to test for structural differences in causal mechanisms across subgroups (sex), moving beyond simple association comparisons.

3. Etiological Insights:

   • Challenged the assumption of direct causality between weight and mental health, suggesting that interventions targeting one to fix the other may be ineffective without addressing shared confounders.
   • Identified maternal weight/gestational gain and early adolescent stress as high-priority leverage points for intervention.

5. Significance and Limitations

• Significance:

  • Shifts the paradigm from confirmatory, single-pathway studies to exploratory, system-level understanding of child development.
  • Provides evidence that weight and mental health trajectories, while correlated, may operate via distinct causal mechanisms linked by common upstream factors.
  • Offers a scalable template for causal discovery in other large-scale longitudinal cohorts.

• Limitations:

  • Unobserved Confounding: The TGES algorithm assumes no unmeasured confounders. If this assumption is violated (e.g., unmeasured genetic factors), spurious edges may exist.
  • Variable Scope: The study is limited to available variables; for instance, diet was not included, limiting insights into nutritional pathways.
  • Ambiguity: Some edge orientations remain undirected (undirected edges in TMPDAG), meaning the exact direction of causality for those specific links cannot be determined without further assumptions.
  • Generalizability: Findings are based on a Danish cohort; external validation in more diverse populations is required.

Conclusion:
This study utilizes advanced causal discovery to map the complex, time-dynamic landscape of childhood weight and mental health. It concludes that while these domains are linked, they do not directly cause one another; instead, they are likely driven by shared upstream factors. The findings highlight maternal weight and adolescent stress as critical intervention targets and provide a robust methodological framework for future causal research in epidemiology.