Personalizing post-stroke symptom management: integrating network analysis and in silico intervention to identify subgroup-specific targets

Stroke is a leading cause of death and disability worldwide. Survivors often experience multiple co-occurring symptoms that interact in complex ways. These interactions create significant symptom burden and reduce quality of life. However, substantial heterogeneity exists in how patients experience post-stroke symptoms. Traditional approaches often treat patients as a homogeneous profile. They also focus on individual symptoms in isolation. This limits the development of tailored symptom management strategies. This study aimed to identify distinct symptom burden based on patients’ reported symptom experiences, compare their symptom network structures, and simulate potential intervention targets using computational modeling. A total of 451 stroke survivors were recruited from three hospitals in Zhejiang Province, China. Symptoms were assessed using the Stroke Symptom Experience Scale. Exploratory factor analysis was used to identify symptom domains. Latent profile analysis was applied to categorize patients based on these domains. Network analysis with EBICglasso regularization was employed to estimate symptom networks for the full sample and each identified profile. The stability of centrality indices was assessed via the case-dropping bootstrap, and network structures were compared using a Network Comparison Test. Finally, the NodeIdentifyR algorithm was used to simulate the potential impact of aggravating and alleviating interventions on each symptom. The goal was to generate hypotheses about potential high-risk and high-benefit targets for future study. Four symptom domains were identified: Motor Dysfunction, Emotional Disorders, Cognitive and Language Disorders, and Pain and Foot Morphological Abnormalities. Latent profile analysis revealed two distinct patient profiles based on symptom experience: a low symptom burden profile (Profile 1, 65.4%) and a high symptom burden profile (Profile 2, 34.6%). In the overall network, limited limb mobility (S1), impaired memory (S9), and impaired attention (S10) showed the highest centrality. The two profiles exhibited significantly different network structures (M = 0.285, p = 0.026). Core symptoms shifted from motor and self-care deficits (S1, S19) in Profile 1 to cognitive symptoms (S9, S10) in Profile 2. In simulated interventions, targeting central symptoms such as S1 was associated with the greatest predicted alleviating impact. This intervention corresponded to a reduction in overall burden by 11.8% in the model. However, the simulations also suggested a dissociation between centrality and risk. In Profile 1, worsening a non-central symptom (foot varus, S6) was predicted to be associated with an elevation in total burden by 26.2%. This effect size was larger than that of most central symptoms. In this cohort of 451 stroke survivors, symptom experience was not homogeneous. Patients could be grouped into low and high symptom burden profiles with different network structures. Simulation analyses showed that targeting high-centrality symptoms was associated with the largest predicted burden reductions, while worsening a non-central symptom (foot varus, S6) in the low-burden subgroup was predicted to increase burden by 26.2%. A sensitivity analysis confirmed that motor, cognitive, and emotional symptom domains play important roles, though the precise ranking of individual symptoms should be interpreted with caution. We introduce a benefit–risk framework as a hypothesis-generating tool for symptom management. Future longitudinal studies are needed to validate these predictions.