Study design and setting: This retrospective cohort study took place from May to November 2022 at Montefiore Medical Center and Montefiore Medical Group (hereafter referred to as “Montefiore”), an academic medical center comprising three hospitals and a group of primary and specialty care clinics serving Bronx and Westchester Counties, New York. The majority of Montefiore patients live in the Bronx, which has a predominantly Hispanic and non-Hispanic Black population [20]. The study was approved by the Montefiore-Einstein Institutional Review Board (IRB), all methods were performed in accordance with the relevant guidelines/recommendations, and all study participants provided verbal informed consent.
Participants: Inclusion and exclusion criteria were selected to ensure all study participants were eligible to be prescribed nirmatrelvir/ritonavir, and that long COVID symptoms reported were temporally related to their COVID-19 infection four months ago. The four month timeframe was chosen as the first nirmatrelvir/ritonavir prescriptions were sent on December 29th, 2021, and this study was approved by the IRB in late April 2022. Inclusion criteria were age greater than or equal to 18 years old and meeting criteria to be prescribed nirmatrelvir/ritonavir under the Food and Drug Administration’s EUA [21, 22]. Exclusion criteria were: 1) taking a medication with a clinically relevant drug-drug interaction with nirmatrelvir/ritonavir that could not be held or dose-reduced for the duration of the nirmatrelvir/ritonavir course [23], 2) advanced kidney disease (eGFR < 30 mL/min), 3) advanced liver disease (Child-Pugh class C), 4) unable to consent and/or unable to reasonably recount symptoms accurately (e.g., patients with dementia, intellectual disability or altered mental status), 5) having a subsequent episode of COVID-19 since the infection four months prior, 6) experiencing long COVID symptoms from a prior SARS-CoV-2 infection that had not resolved fully before the infection four months prior, 7) being treated with chemotherapy in the month prior to enrollment, as it may cause symptoms which overlap with long COVID, and 8) being treated with molnupiravir, an antiviral alternative to nirmatrelvir/ritonavir with a similar mechanism of action.
The prevalence of long COVID among people who survive acute COVID-19 varies among different cohorts which have used different definitions. Population-based surveys and the largest meta-analysis to date have reported a prevalence between 6.2% and 35.1% [10, 24, 25]. To determine the sample size for the current study, we therefore estimated the expected prevalence to be 20%. With this assumption, enrolling 250 patients who were treated with nirmatrelvir/ritonavir and 250 patients who were not treated would yield at least 80% power at the 0.05 level of significance to detect an absolute difference in having long COVID of 10% or greater, which we deemed to be clinically significant.
Screening and enrollment: Using the Montefiore EMR, we identified patients prescribed nirmatrelvir/ritonavir, and patients who were not prescribed nirmatrelvir/ritonavir, during the period of December 29th, 2021, going forward in time until 500 study participants were enrolled. To identify nirmatrelvir/ritonavir patients, we generated a list of all consecutive patients prescribed nirmatrelvir/ritonavir from a Montefiore medical provider; the date of prescription was considered the index date. To identify control patients, we created a list of patients who tested positive for SARS-CoV-2 on a polymerase chain reaction (PCR) test done at Montefiore on the index dates for the nirmatrelvir/ritonavir-prescribed patients.
To screen for eligibility, we first reviewed patients’ medical records. Next, we attempted to contact potential participants via phone. We attempted to contact everyone who was prescribed nirmatrelvir/ritonavir. Since there were many more patients in the control group, it was not feasible to attempt to contact them all. The list of people with a positive SARS-CoV-2 PCR for each index date was randomized; going down that list, we attempted to contact potential participants until one or more were enrolled or we reached the end of the list. Patients were considered “unable to be contacted” if they were not successfully contacted on two separate occasions, or were contacted but requested to be called back at a future time, and then did not answer the phone when called at that future time. Individuals who were successfully contacted and agreed to participate in the study were then screened for the full inclusion and exclusion criteria.
Data collection: Participants completed a single structured telephone interview conducted by a physician or medical student four months (120 to 150 days) after their positive SARS-CoV-2 test. Participants were asked to identify which, if any, treatments they received to treat their acute COVID-19 episode four months ago. If participants were prescribed nirmatrelvir/ritonavir, they were asked if they completed a full course, partial course, or did not take it. For participants that took a partial course, they were asked to estimate the number of doses taken out of the ten total doses. Individuals were then asked to self-identify as having long COVID by responding “yes” or “no” to whether they were currently experiencing any new or worsened symptoms since developing COVID-19 four months prior. They were then asked if they were experiencing eleven common long COVID symptoms: dyspnea or change in how their breathing feels; change in smell or taste; headaches; dizziness or lightheadedness; chest pain or pressure; palpitations; generalized fatigue or tiredness; exertional intolerance; nausea/vomiting or abdominal discomfort; brain fog; and paresthesias [4, 26, 27, 28, 29]. It was emphasized that they should only answer “yes” if the symptom was either new or had worsened since developing COVID-19 four months prior and had not fully resolved. The full study survey is included in the supplemental materials.
Outcomes: We considered long COVID to be present if participants were experiencing any of the long COVID symptoms listed in the survey. The primary outcome was the presence of long COVID. Secondary outcomes were: the presence of each individual long COVID symptom, the number of long COVID symptoms per participant, and the number of participants who self-identified as having long COVID.
Other variables assessed: Baseline characteristics that may influence the risk of developing long COVID [12] were assessed through manual EMR chart review and verbal confirmation with patients. Predefined covariates considered to be potential confounders were age; race/ethnicity; sex; body mass index (BMI); COVID-19 vaccination status (primary series plus booster, primary series without booster, or unvaccinated); smoking status (current, former, never); and preexisting conditions at the time of developing COVID-19 four months prior: hypertension, diabetes, asthma or chronic obstructive pulmonary disease, or a mood disorder (anxiety disorder, depressive disorder, bipolar disorder, posttraumatic stress disorder). We also assessed the number of nirmatrelvir/ritonavir doses taken, and if individuals received prespecified treatments during their acute COVID-19 episode (monoclonal antibodies, remdesivir, corticosteroids and convalescent plasma; confirmed by EMR review).
Statistical Analysis: We used descriptive statistics to report characteristics and frequency of any and each of the long COVID symptoms in the treated and not treated groups. We used chi-square tests or t-tests to compare between the demographics between the two groups. To determine whether receipt of nirmatrelvir/ritonavir was associated with long COVID symptoms, we conducted separate bivariate chi-square tests or t-tests for each of the primary and secondary outcomes, comparing between the two groups. To visualize the incidence of individual long COVID symptoms in the two groups, we created a Forest Plot presenting the odds ratio for each symptom. Finally, we conducted a series of separate multivariable logistic regression models where the main independent variable was treatment group (treated or not treated), and the dependent variable was the primary outcome. We included the following potential confounders as covariates in the model, as they have been associated with higher incidence of long COVID in prior studies: female sex, obesity, elderly age, smoking history, diabetes, lung disease, mood disorder, and unvaccinated status [12]. We also included Hispanic race/ethnicity as a covariate, as compared to other racial/ethnic groups this group has reported higher rates of long COVID on the United States Census Bureau’s Long COVID Household Pulse Survey [10].