Table 3. Recommendations for mitigating biases in Long COVID studies.

	Primary Studies	Reviews
Confounding	Document common confounders including age, sex, comorbidities, race/ethnicity, severity of acute infection, hospitalization status, timing of infection, duration of follow-up, variants of concern, and pre-infection vaccination status	Define confounders in study protocol for data extraction, ROB assessment and subgroup/sensitivity analysis
	Exploit EHR and ongoing cohort studies with robust data on patient characteristics that cannot be prospectively measured at the time of infection (e.g., pre-infection vaccination status, predominant variant of concern, pre-infection patient phenotype)	Use or revise an ROB tool to enable precise definition of the most important confounders a study should be designed to address
	Include comparator/control cohort of people without COVID-19 enrolled concurrently with SARS-CoV-2 positive individuals	Conduct subgroup or sensitivity analysis as specifically as the number of studies and participants allow (e.g., not just hospitalized vs. non-hospitalized but also hospitalized with comparator cohort vs. non-hospitalized with comparator cohort)
Selection bias	PICO criteria of population and comparator group should be clearly specified and outcome-dependent. For example, methods of measuring long COVID based on periodic symptom-monitoring or electronic health records should have exposure-negative control groups; patient self-report of long COVID as the outcome should prioritize representative sampling to ensure inclusion of people with different healthcare access and education.	PICO criteria of the review should be clearly specified. PICO definition of the primary studies should be extracted for each study. Specify the applicability and comparability of the non-exposed cohort for every study in data extraction.
	Recruit patients consecutively and report response rate or attrition for both case and comparator cohorts*	Differences in PICO criteria used in the primary studies should inform subgroup/sensitivity analysis. For instance, studies that did and those that did not use a comparator cohort should be analyzed as separate subgroups
	Use PSM or IPW for matching cases with comparators especially if regression analysis is planned	Use an ROB tool with clear metrics for assessing sample selection or the suitability of comparator groups such as NOS or ROBINS-E
Measurement bias	Clearly define and document time zero and duration of follow-up, including range, measures of central tendency and variation*	Extract timing and duration of follow-up for all studies and conduct subgroup analysis accordingly as the number of studies and participants allow
	Define or corroborate exposure status with confirmatory testing using RT-PCR or test of similar sensitivity and specify*	Extract method of exposure assessment for all studies and conduct subgroup analysis accordingly as the number of studies and participants allow
	Administer established symptom scales such as PROMIS or EuroQol, ideally with both case and comparator groups, rather than relying on self-report of long COVID while the condition does not have a clear symptom-based definition	Extract method of outcome assessment (i.e., interview, survey, electronic health records) and conduct subgroup analysis accordingly as the number of studies and participants allow
Outcomes selection	Register protocol to include the tool and method used to ascertain Long COVID or a set of signs and symptoms that may qualify as Long COVID*	Specify the case definition used to include the set of symptoms and conditions measured
	Refer to consensus definitions supplemented by existing definitions for related illnesses including ME/CFS or other post-viral conditions*	Report study-level differences in the types of outcomes selected and aggregated to represent Long COVID
	Report symptom severity and degree of functional impairment experienced by patients*	Specify which outcomes of which studies are included in composite outcomes such as "at-least 1 symptom," including underlying sample size and population characteristics

EHR, electronic health record; IPW, inverse probability weighing; ME/CFS, myalgic encephalomyelitis/chronic fatigue syndrome; NOS, Newcastle Ottawa Scale; PICO, population, intervention, comparator, outcome; PROMIS, Patient-Reported Outcomes Measurement Information System; PSM, propensity score matching; ROB, risk of bias; ROBINS-E, Risk of Bias In Non-randomized Studies—of Exposure; RT-PCR, reverse transcriptase polymerase chain reaction

*: Adapted from Nasserie T, Hittle M, Goodman SN. Assessment of the Frequency and Variety of Persistent Symptoms Among Patients With COVID-19: A Systematic Review. JAMA Netw Open. 2021;4(5):e2111417