3 Risk of Bias and Limitations
When analysing data related to COVID-19 positivity it is important to
recognise the sensitivity of the test is variable. Bronchoalveolar
lavage is the most sensitive test (93%) whilst nasal swabs (63%) and
pharyngeal swabs (46%) have lower positive rates14.
Moein et al, who conducted the UPSIT, case control study and Mao et al
were the only authors to report the technique and anatomical location of
their Polymerase Chain Reaction (PCR) analysis of COVID-19 status. Moien
et al used nasal aspirates or washes and Mao et al’s group used throat
swabs . Due to the relatively low sensitivity of the test used there
will be a proportion of false negatives that will falsely lower the
incidence of COVID-19 positivity in the OD groups.
The majority of the responses to questionnaires were received using
remotely using electronic response forms of mobile based applications
which will cause selection bias. Younger more technologically
interactive cohort are more likely to interact and this sub-group seem
to be less affected by COVID-19 when compared to older age groups who
have a higher morbidity and mortality. For example, Menni et al(in an
unpublished paper made available before peer review) who used a mobile
based application, report an average age of 41.48 (CI = 13.77) for those
in their non-PCR-tested group, including over 1.5million people.
Hospitalised populations are also less likely to interact with these
methods due to their disease severity, internet connection or associated
interventional treatments.
Cross-sectional questionnaires and case series are prone to bias due to
influence of confounding variables, assessment of patients at different
time points relative to their exposure and reporting bias. In case
series specifically consecutive patients often missed in data
collection. In these studies, however the researchers are simply
presenting patient factors and associated symptoms rather than
treatments or interventions and their subsequent effects or outcomes and
this observational nature could help to reduce observer bias. In studies
that were conducted requiring historical data from the patients there is
a risk of recall bias and under-reporting or inaccuracies of symptoms
specifically where onset and duration of symptoms is involved.