Mapping risk of gambling disorder in older adults: a neural network approach.

BACKGROUND AND

OBJECTIVES

Gambling disorder (GD) in older adults remains insufficiently understood, despite growing concerns about gambling-related harm in older adulthood. Risk profiles and clinical presentations in this age group may differ from those typically described in younger or mixed-age samples, reflecting age-specific psychosocial contexts, health conditions, and life transitions. Traditional statistical approaches may be limited in their ability to model the complex, non-linear interplay among psychological, behavioral, and contextual factors. The present study aimed to apply neural network-based models to predict the presence and severity of GD in older adults.

METHODS

The study used a cross-sectional, single-center, case-control design. Data were obtained from adults aged 50 to 85 years, with 50 years adopted as a pragmatic lower threshold given the lack of consensus on defining older adulthood, comprising a clinical sample of 745 patients seeking treatment for gambling-related problems at a specialized treatment hospital unit, and a healthy control (HC) group of 136 participants without gambling-related problems, recruited at non-psychiatric hospital outpatient services.

RESULTS

A neural network (multilayer perceptron, MLP) with one hidden layer composed of eight neurons showed excellent discrimination between individuals with GD and healthy controls in the complete sample (AUC = 0.987, 90.4% specificity, 97.7% sensitivity, and 96.6% classification accuracy). In the independent holdout subset, the model yielded an AUC of 0.890, with 75.0% specificity, 96.2% sensitivity, and 93.3% classification accuracy (although these estimates should be interpreted cautiously because the holdout included only 12 healthy controls). Novelty seeking emerged as the strongest contributor to the model, followed by emotion regulation difficulties, particularly non-acceptance of emotional responses and reduced impulse control. Focusing exclusively on the clinical GD subsample, a second MLP including six hidden neurons indicated that overall psychological distress was the primary factor associated with disorder severity. Additional relevant contributors to GD included novelty seeking, gambling-related cognitive distortions, and lower self-directedness.

CONCLUSION

The neural network models effectively classified GD and its severity in older adults, identifying a multifactorial pattern characterized by affective distress, emotion regulation difficulties, and distorted gambling-related beliefs. These findings highlight the potential of non-linear modeling to characterize patterns associated with GD in older adults, while the clinical utility of machine learning tools for age-sensitive detection and intervention remains preliminary and requires further validation.