(S-109) Mixed Effects Modeling of Diurnal Cortisol Dysregulation in Long COVID

Hari Prabhath Tummala – Clinical Pharamcy – University of California San Francisco (UCSF); Danny (Hoi Tsun) Chu – Clinical Pharmacy – University of California San Francisco (UCSF); Thomas Dalhuisen – University of California San Francisco (UCSF); Chang Song Celina – University of California San Francisco (UCSF); Khamal Anglin – University of California San Francisco (UCSF); Beatrice Huang – University of California San Francisco (UCSF); Brent Abel – University of California San Francisco (UCSF); Emily Fehrman – University of California San Francisco (UCSF); Morrie Schambelan – University of California San Francisco (UCSF); Elizabeth Murphy – University of California San Francisco (UCSF); Michael Peluso – University of California San Francisco (UCSF); Amelia Deitchman – Clinical Pharmacy – University of California San Francisco (UCSF)

Objectives: Long COVID (LC) is a disabling infection-associated chronic condition following SARS-CoV-2 infection that has been associated with cortisol dysregulation. Cortisol studies in LC to date have been limited to analyses at a single time point and do not account for known fluctuations in cortisol throughout the day. In this study we compare diurnal variations in cortisol patterns between individuals with LC and those who fully recovered from COVID-19 via nonlinear mixed effects modeling.

Methods: Plasma cortisol was measured every 20 minutes from 0800 to 1600 hours in 16 individuals with confirmed history of SARS CoV 2 infection ≥ 3 months prior who either (1) met the National Academies of Sciences, Engineering, and Medicine (NASEM) case definition for LC including prominent fatigue and post-exertional symptoms (n = 7), or (2) had fully recovered from COVID-19 (“fully recovered,” n = 9). We excluded individuals who used any medication known to affect cortisol metabolism in the preceding 4 weeks, and those with sleep disorders or other conditions that could potentially alter their circadian rhythm. One participant who had a documented stressful occurrence on the day of study was also excluded from the model. To capture the hormone’s negative feedback regulation without suppressing circadian oscillation, we defined an inhibitory cortisol concentration (Cinh) driving release dynamics using a dual exponential function, one term representing production and one term representing feedback suppression. The irregular pulses across groups were defined using an exponential probability density function (PDF). Model parameters and inter individual variability were then estimated via nonlinear mixed effects modeling. Phenotype (LC versus recovered) was used as a covariate. Goodness-of-fit, visual predictive checks, and change in objective function value (OFV) were used to guide model selection.

Results: The dataset comprised of 400 cortisol concentrations. A combined error structure (13.4 % proportional; 0.38 µg/dL additive) provided the optimal description of residual variability. Adding an exponential PDF for discrete cortisol pulses reduced OFV from 1151.8 to 600.5. The exponential PDF consisted of scaling parameter, rate parameter with a decay-shape modifier to describe the irregular cortisol secretion throughout the day. Introducing phenotype as an exponential covariate on the pulse decay rate (θ = –0.376) further improved fit (ΔOFV = -22.18), describing 25% of interindividual variability on pulse-decay rate. This covariate effect corresponds to approximately 31% slower decay in recovered versus LC participants.

Conclusions: Our integrated circadian–pulsatile cortisol model accurately captures both the time of day pattern and discrete secretory events. The identified phenotype effect on pulse decay rate suggests altered HPA axis regulation in LC participants.

Citations: [1] Rohatagi S, Bye A, et al., Mathematical modeling of cortisol circadian rhythm and cortisol suppression. European Journal of Pharmaceutical Sciences. 1996 January 5; 341-350.

[2] Veldhuis J, Johnson M. Deconvolution analysis of hormone data. Methods in Enzymology. 1992; 210:539-575

Keywords: Long COVID, Cortisol Dysregulation, Myalgic Encephalomyelitis / Chronic Fatigue Syndrome