Characterizing Sleep Disturbances in Long COVID: A Retrospective Analysis from a Safety Net Hospital (P8-4.007), 2025, Sohng et al

Characterizing Sleep Disturbances in Long COVID: A Retrospective Analysis from a Safety Net Hospital (P8-4.007)

Angela Sohng, Shruti Misra, Neoreet Braha, Ariana Rauch, Andrew Tsao, Ta’Kiya Moore, Michael Alosco, Jai Marathe, and Sanford Auerbach

[Line breaks added]

Objective
To characterize risk factors of Long COVID sleep disturbance at a city hospital.

Background
Millions globally suffer from long-term functional and cognitive impairment from post-acute SARS-CoV-2 infection (PASC), or “Long COVID.” Sleep disturbances (SDs), either new or exacerbated, are common but under-researched in this population despite their impact on immune function, chronic organ illness, and quality of life.

Design/Methods
Retrospective chart review of patients seen at a hospital-based Long COVID Clinic (01/2022–12/2023). SDs experienced during PASC were classified into four phenotypes: insomnia, hypersomnia, mixed, and other. Binomial and multinomial regression identified predictors, with no SD as the reference and Area Deprivation Index (ADI) as the covariate. P-values were adjusted for multiplicity using the Benjamini-Hochberg method.

Results
Of 452 patients (median age 47, 70.4% Female, 46% White), 71.9% reported SDs, of which 34.5% experienced new-onset and 23.2% reported exacerbation of prior disturbances.

ADI was the only demographic predictor (p=.016) of SD. Vaccination status, number of infections, and hospitalization were not predictive.

Key PASC risk factors for SDs (all p<.01 unless specified) included post-exertional malaise (OR=7.98), fatigue (OR=6.39), dysautonomia/POTS (OR=5.96), gait instability (OR=5.36, p=.019), depression (OR=4.95), brain fog (OR=3.85), myalgias/arthralgias (OR=3.19), anxiety (OR=3.06), SOB/DOE (OR=2.17).

Key prescription risk factors included antidepressants, albuterol, and benzodiazepines (OR>2.0, p<.01).

PASC neurologic symptoms of paresthesia, weakness, dizziness, and migraines predicted insomnia (OR>2.39, p<.027). Insomnia was also associated with pre-existing pain, anxiety, and depression.

Benzodiazepines were stronger predictors of insomnia than anxiety (OR=3.53 vs. 2.96, p< .01). Beta blockers and PPIs predicted insomnia, while stimulants did not. Muscle relaxants predicted insomnia (OR=5.30, p=.016), while opioids predicted hypersomnia (OR=11.16, p=.018). Sleep aids like melatonin, mirtazapine, trazodone, doxepin, and zolpidem were protective, unlike antihistamines.

Conclusions
PASC symptoms, comorbidities, and medications predicted sleep disturbances, revealing distinct patterns for insomnia and hypersomnia. Identifying these risk factors could enable earlier interventions and inform future treatment strategies.

Link (Neurology)

 

Can they say that something was predictive, protective, etc. based on a retrospective chart review?

Wouldn’t they be limited to correlation?

 

It is bizarre—I suspect it’s a very messy use of what “predict” means in a regression context. “Predictor” is often used to refer to your X variables, so saying X predicts Y really just means “X was statistically significantly associated with Y in the regression equation.”

For a publication they should’ve just said “associated with” unless it was literally a time course data set where the drugs were administered first and then the before vs. after incidence rate of those symptoms was being compared.

Which doesn’t seem to be the case here. I think they just had some information on drug usage and some information on reported symptoms. We don’t know if they started taking the drugs before or after the symptoms happened.