Agent-Based Modeling and Simulation of Depression and Its Impact on Student Success and Academic Retention

Sherif E. Abdelhamid, Chris J. Kuhlman, Madhav Marathe, Sahana Ravi, Kenneth Reid2016 · DOI

Abstract: Sherif's research work lies at the intersection of computation, biology and education: in particular, he is interested in designing and building software systems to enable domain experts to easily access and effectively use high performance computing.

Introduction

College student retention, particularly in demanding fields like engineering, is a critical challenge for higher education. Beyond academic rigor, mental health—specifically depression—is a primary driver of student attrition. Research indicates that even a modest increase on standard depression scales correlates with significant drops in GPA and a nearly 5% increase in the probability of a student leaving their institution.

Traditional models often view depression as a static, binary state. However, this research argues that depression is a dynamic process characterized by internal symptom cascades (e.g., insomnia leading to fatigue) and social "contagion" through peer interactions. To capture these complexities, the researchers developed a multi-level Agent-Based Model (ABM) to simulate depression dynamics among a large student population.

The Two-Level Network Architecture

The core of this research is a dual-layered Graph Dynamical System (GDS) that models both the individual and the collective:

  1. Internal Symptom Network ($G^1$): Each student agent is modeled with an internal network of 14 symptoms derived from clinical literature, such as depressed mood, fatigue, and suicidal thoughts. These symptoms are interconnected; the activation of one increases the probability of others activating.
  2. Social Interaction Network ($G^2$): This layer represents the student body, where edges denote daily face-to-face interactions.

The model incorporates a "contagion" mechanism based on homophily. Rather than spreading like a virus, depression in this model spreads through peer influence on specific symptoms. For example, if a student's peer suffers from insomnia, the probability of that student also developing insomnia increases.

Multi-level network model showing internal symptom interactions (S0-S13) within individual students and their connections within a larger social network.

Synthetic Population and Clinical Parameters

To ensure the simulation's relevance to real-world scenarios, the researchers utilized a high-fidelity synthetic population. They extracted a sub-network of 18,866 students from a validated social contact network of Virginia, matching the demographic and geographic attributes of a major public university.

Structural properties of the synthetic student social network used in the simulation.

The internal dynamics of the symptom networks were parameterized using data from the Virginia Adult Twin Study of Psychiatric and Substance Use Disorders (VATSPUD), providing a clinically grounded basis for how symptoms trigger one another.

Key Insights and Results

The simulation revealed that depression prevalence is highly sensitive to small changes in network connectivity and influence:

  • Non-Linear Social Impact: A 67% increase in the strength of peer influence led to a six-fold increase in the number of depressed students (from 400 to 2,400) in the simulation.
  • Internal Connectivity: Increasing the strength of internal symptom-to-symptom influence by 2.5x resulted in a 16x increase in population prevalence, jumping from 0.8% to 12.7%.
  • Symptom Centrality: The model highlighted that "high-degree" symptoms—those with many connections to others, like depressed mood or fatigue—act as critical accelerators for the spread of depression across the population.
  • The "Seed" Effect: Students entering college with pre-existing symptoms act as initial seeds. The speed at which depression spreads through the campus is highly dependent on the initial state and activation probability of these students.

Conclusion and Future Directions

This research provides a framework for understanding depression not just as an individual struggle, but as an emergent property of social and internal networks. By identifying high-impact symptoms and the multiplier effect of social circles, universities can better design targeted interventions.

Future work aims to refine the model by including more diverse demographic factors (gender, ethnicity, income) and moving beyond "symptom homophily" to model how different symptoms might influence one another across social links. Validating these simulations against longitudinal university health data remains a primary goal for the research team.

Summary updated

Abstract

Sherif's research work lies at the intersection of computation, biology and education: in particular, he is interested in designing and building software systems to enable domain experts to easily access and effectively use high performance computing
BibTeX
@misc{abdelhamid2016agent,
  author = {Sherif E. Abdelhamid and Chris J. Kuhlman and Madhav Marathe and Sahana Ravi and Kenneth Reid},
  title = {{Agent-Based Modeling and Simulation of Depression and Its Impact on Student Success and Academic Retention}},
  year = {2016},
  doi = {10.18260/p.26545},
}
Figure