Machine learning clustering of psychological response trajectories across the first and second waves of the COVID-19 pandemic

Résumé

Background The COVID-19 pandemic disrupted daily life globally, leading to increased psychological distress. While many studies have documented mental health trends, few have tracked how person-centered psychological clusters emerge and evolve across pandemic waves. Methods This longitudinal study assessed 338 adults (mean age, 38.3 years; 55% female) from 53 countries at two pandemic waves (2020 and 2021). Post-traumatic stress symptoms (PTSS; 15-item scale, α = 0.93), coping strategies (11 binary items), and pandemic stress (15 binary items) were measured using validated instruments; composite scores were constructed by summing item responses. Machine learning clustering (K-means and Gaussian mixture models) evaluated solutions k = 2–6 using the silhouette coefficient, Davies–Bouldin index, Calinski–Harabasz score, BIC/AIC, and the gap statistic. The majority of the criteria supported a three-cluster solution. Cross-wave cluster alignment was achieved using the Hungarian algorithm. Person-centered transitions, stability rates, and demographic associations were examined. Results Three statistically defined clusters emerged at both waves: a low-distress, high-coping cluster (labeled “resilient”; Wave 1: 35.5%; Wave 2: 29.00%); a low-coping cluster (40.04%; 49.11%); and a high-distress cluster (25.06%; 21.89%). Overall, 55.3% (95% CI [50.0%, 60.5%]) remained in the same cluster across waves. Among those initially in the resilient cluster, 55.8% transitioned to higher-distress clusters by Wave 2. The low-coping cluster showed the greatest stability (68.9%), and 50.6% of those initially in the high-distress cluster transitioned to lower-distress clusters. Net flows shifted toward the low-coping cluster (31 participants). Country income level was associated with cluster membership at both waves ( p < 0.001) but did not predict individual transitions. Older age, higher education, and employment were associated with the resilient cluster, whereas younger age and unemployment were associated with high distress. Conclusion Psychological responses to prolonged crises are dynamic and heterogeneous. Coping capacity and socioeconomic factors are associated with cluster membership, but causal inferences cannot be drawn due to the observational two-wave design and convenience sampling. These findings suggest that interventions could benefit from strengthening coping skills while also addressing structural factors such as education and employment, though further research with more frequent assessments and experimental designs is needed.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0