Study population and setting
The study included 236,379 patients over the age of 10 years (mean age 46 years, 56% female) with a confirmed diagnosis of COVID-19 from January 20 to December 13, 2020 from the TriNetX network of anonymized electronic health records from 62 healthcare organizations in the United States. Two cohorts, matched on 55 variables considered risk factors for COVID-19 outcomes, were considered as controls: one of patients diagnosed with influenza (n=105,579), and one of patients diagnosed with any respiratory tract infection, inclusive of influenza (n=236,038). The outcomes of interest were 14 neurologic and psychiatric diagnoses, diagnosed at any time up to 180 days after the index health event (i.e. COVID-19, influenza, or respiratory tract infection diagnosis): intracranial hemorrhage; ischemic stroke; Parkinson’s disease or Parkinsonism; Guillain-Barre syndrome; nerve, nerve root, and plexus disorders; myoneural junction and muscle disease; encephalitis; dementia; psychotic, mood, or anxiety disorder (any and each separately); substance use disorder; and insomnia. Each outcome was assessed separately, and a composite outcome of any event was also assessed. For chronic conditions, those with a diagnosis prior to the index event were excluded. Subgroup analyses were also conducted by 3 overlapping groups: those requiring hospitalization (n=46,302), those not requiring hospitalization (n=190,077), and those exhibiting encephalopathy (n=6,229; defined as a diagnosis of altered mental state or delirium around time of diagnosis). Four additional matched cohorts of patients with different index events (skin infection, urolithiasis, large bone fracture, and pulmonary embolism) were also assessed in sensitivity analyses. Hazard ratios (HRs) were calculated with Cox models.
Summary of Main Findings
In the cohort of patients with COVID-19 who had never had a prior diagnosis of any of the 14 neurologic/psychiatric outcomes, 13% had a first diagnosis of at least one neurologic/psychiatric outcome within 180 days, and 32% of the full cohort had such a diagnosis. Hospitalized patients had a higher incidence of each outcome, and those with encephalopathy had a much higher incidence of each outcome, relative to non-hospitalized COVID-19 patients. The proportion of patients with a first diagnosis of any mood/anxiety/psychotic disorder was 8.7%, with corresponding first-diagnosis figures of 2.9% for nerve disorders, 0.8% for ischemic stroke, 0.7% for dementia, and 0.3% for intracranial hemorrhage. When comparing patients with COVID-19 to the influenza cohort, HRs were significantly (p<0.05) above one for all neurologic/psychiatric conditions except for Parkinsonism and Guillain-Barre syndrome. Relative to the cohort with any other respiratory tract infection, COVID-19 patients had hazard ratios significantly above one for all outcomes. For sensitivity analyses with the additional four control cohorts, results were similar except when the index event was a known predictor of the outcome (e.g., pulmonary embolism and ischemic stroke). In subgroup analysis, these results remained similar when restricting to non-hospitalized patients. Hazards of all outcomes were larger in magnitude among COVID-19 patients who were hospitalized, and who had encephalopathy, compared to matched cohorts of COVID-19 patients who were not hospitalized and who did not exhibit encephalopathy, respectively.
This study drew from a large and demographically diverse database; the sample size was very large, permitting control matching with respect to many potential confounding variables. Multiple matched cohorts were considered in sensitivity analyses.
All data were drawn from electronic health records with an unknown degree of missing data. It is unclear how analyses dealt with data that was missing (e.g., exclusion, imputation, or coding as zero); results may be sensitive to these methods, particularly if the degree of missingness was large. Reliance on ICD-10 codes for diagnosis may result in under-ascertainment of some conditions. This is a concern with encephalopathy, for example, since the reported prevalence in COVID-19 patients was 11%, which is considerably lower than the authors expected. Inclusion in the study was conditional on being alive at the time of analysis; this may have created a bias if those patients who died (COVID-19 patients would have a higher risk of death than those in the control groups) had a different risk of neurologic or psychiatric outcomes relative to those who survived. Additionally, although control cohorts were matched on multiple possible confounders, the analyses may have required additional adjustment for these confounders if there was censoring (for example, by the end of follow-up). There was little information available on other potential confounding variables such as health behavior; omission of these variables from analysis could have biased the results in unpredictable ways. Finally, it is likely that COVID-19 cases were distributed later in time than influenza and respiratory tract infection cases. If so, it is possible that psychological stress from extended social isolation during 2020 (unrelated to COVID-19 diagnoses) could confound the results for some outcomes (e.g., mood disorders).
This is the largest study to date on longer-term (up to 6 months) neurologic and psychiatric outcomes related to COVID-19.
This review was posted on: 12 February 2021