Long‐COVID causal impacts on cognitive ability of mild‐cognitive impaired population from large‐scale electronic health record using heterogeneous causal effect model

Background Identification of long‐COVID impacts on human cognitive ability is important but Randomized Controlled Trial (RCT) is not possible. Virtual RCT (VRCT) can be done from electronic health record (EHR) which requires to address heterogeneity appropriately. Method We design new method for VRCT on COVID intervention among Mild Cognitive Impaired (MCI) population. We select patients (n = 71,900) from National COVID Cohort Collaborative (N3C) data who meet our inclusion criteria (Table‐1 and Table‐2). We compute 28 Cognitive Impairment (CI)‐related covariates (Table‐3). Final outcome is CI and COVID‐19 diagnosis is intervention. To formulate a placebo like intervention virtually, we compute covariates: (i) For COVID positive patients, consider pre‐COVID histories, (ii) For Non‐COVID having transition MCI‐to‐CI, consider 6 months before first CI diagnosis, (iii) For Non‐COVID having no transition MCI‐to‐CI, consider 6 months before last recorded data. We utilize Heterogenous Causal Effects using Random Forest (HCE‐RF) model[1] to estimate COVID‐19 effects on 7 different CI outcomes. HCE‐RF, proven method in machine learning (ML)‐based VRCT, estimates conditional average causal effect (CATE) where 0 and 1 mean no and highest risk. Result COVID‐19 causes (details Table‐4) 54% (CATE:0.5489 CI[0.5428, 0.5618]) higher risk of developing any kind of dementia within 6‐months LongCOVID given the patient has prior MCI which is higher for severe (0.6498 CI: [0.6066, 0.6817]) and little bit lower, for mild COVID‐19 (0.5489 CI: [0.5428, 0.5618]). Similarly, COVID‐19 with different severity (mild‐to‐severe) causes significant higher risks of developing mixed dementia of any two types, Alzheimer’s disease or Vascular dementia without behavioral disturbance among MCI patients compared with COVID‐19 uninfected population. On the other hand, COVID‐19 infection causes no significant risk of developing presenile, senile and vascular dementia without behavioral disturbance for MCI patients in 6 months LongCOVID. Conclusion We proposed new method of developing VRCT cohort on observational data to identify disease intervention effect on disease progression in presence of heterogeneity of causes and effects and it on EHR data to identify LongCOVID (6 months) causal risks on MCI patients in developing different CIs. Our method can be utilized to uncover many hidden effects of COVID‐19 on observational data. [1] https://doi.org/10.3386/w24678