A simple computational model of population substance use

Abstract

BackgroundSubstance use behaviors and their etiologies are complex and often not amenable to traditional statistical analysis. Computational models are an increasingly popular alternative approach for investigating substance use. However, cumulative progress has been difficult because of a lack of standardization. This study aims to develop and evaluate a simple computational model that could serve as a common starting point for future computation-based investigations of substance use.MethodsA two-state (“Using” a substance or “Not using” a substance) stochastic model with three manipulable parameters is used to reproduce the distributions of past 30-day alcohol, cannabis, and tobacco cigarette consumption frequencies (e.g., used on 5 days within the past 30 days) observed in the U.S. National Survey on Drug Use and Health (NSDUH) (years 2002-2019 combined). The model employs a path-dependent process: during each iteration (i.e., each “day”) of the simulation, each computational object chooses to use or not use a substance based on probabilities that are contingent on choices made in prior iterations. The Lempel-Ziv complexity measure was used to examine the resulting sequences of binary decisions (use ordon’t use) made by each computational object.ResultsThe model accurately reproduces the population-level “U-shaped” distributions of past 30-day alcohol, cannabis, and cigarette use in the U.S. The path dependence function was required for reproducing these distributions. The model also suggests an “arc” of behavioral complexity stages: as the frequency of use increases, the complexity of decision sequences increases, peaks, and then decreases. However, decision sequence complexity still varied considerably among objects with similar frequencies of use.ConclusionA simple computational model that simulates individual-level sequences of substance use can reproduce the population-level distributions of substance use observed in national survey data. The model also suggests that complexity measures are a potentially helpful tool for examining substance use behaviors.