Long-Term Trends in Post-COVID Severity: A Machine Learn-ing Analysis from the POP/COVIDOM cohort of the German NAPKON Cohort Network

Background: Post-COVID syndrome (PCS) affects many survivors with varying symptom profiles driven by acute disease severity (PCS S) or individual resilience (PCS R). While cross sectional studies have identified risk factors and gender differences, long term trajectories remain unclear.Objective: This study investigates the stability and progression of PCS-S and PCS-R scores after 9, 24 and 36 months from initial diagnosis, identifying key predictive factors stratified by gender.Methods: We analyzed data from 1,526 participants of the German National Pandemic Cohort Network (NAPKON), modeling symptom-based PCS-score trajectories over time with linear mixed effects mod-els. Data were split into training (n=944), test (n=233), and two-site external validation (n=349) sets. Gender stratified elastic net regression used nine month clinical and psychosocial measures to predict PCS scores at 24 and 36 months.Results: PCS-S and PCS-R scores showed small but significant declines between 9 and 36 months (β = –0.054 and –0.065, respectively; p < .001), indicating persistent symptom burden despite gradual im-provement. Predictive models explained 16.7–52.6% of variance in later PCS severity. Fatigue after 9 months and age predicted later PCS-S; quality of life and depression added predictive value in females. Fatigue and sleep issues predicted PCS-R, with living/employment status relevant in females and cog-nitive deficits in males.Conclusion: The severity of PCS subtype manifest after 9 months remains relatively stable over time, with distinct gender-specific predictors shaping symptom progression. Tailored interventions are essen-tial for long term management of PCS pathways.Funding: The COVIDOM study is funded by the Network University Medicine as part of the NAP-KON.