In this systematic review of risk factors for COVID-19 disease, disease severity and mortality, we document 33 comparative studies examining sociodemographic, behavioural and clinical exposures. Age and sex were very commonly examined; a wide array of comorbidities have also been considered.
Within the synthesised evidence, risk factors for mortality were the clearest, plausibly partly because this outcome is easy to define. Increasing age (different studies presented different thresholds, but being over 50 years of age was common) was an uncontested risk factor. Five studies also presented evidence for the presence of any comorbidities being a risk factor,(20, 34, 45, 46, 50) with none demonstrating evidence against. Given the increasing prevalence of comorbidities with age, the lack of adjustment for confounding in these studies likely over-emphasises the effect size of each risk factor. We note that work subsequent to our literature search documents an independent effect of age on COVID-19 mortality from overall comorbidities, as measured by the Charlson Comorbidity Index Score, but not vice- versa.(51) Another study published outside of the time range of our search found both age and an array of comorbidities, each analysed separately (chronic cardiac disease, chronic pulmonary disease, chronic kidney disease, chronic neurological disease, dementia, malignancy, moderate/severe liver disease; and obesity), to be independent risk factors (as well as sex).(52)
Risk factors for severe disease were more complex to synthesise, likely due to the mixed array of outcome measures that can also be prone to observer bias. The impact of age was very commonly assessed, generally showing evidence in favour of this being a risk factor (with a similar age spectrum to the mortality data). Ethnicity was studied in two publications,(25, 29) with mixed results. We note that such findings are likely to be highly context-specific, given that ethnicity acts as a proxy for a series of sociodemographic factors that are highly relevant to the spread of an infectious condition (as well as, perhaps, some biological traits).
Studies of risk factors for COVID-19 disease have been complicated by testing strategies globally, which have largely been concentrated on severe disease. As our knowledge of the full symptom spectrum of the disease moves forward, it will be possible to have a broader case definition that does not solely focus on viral testing, and thus the ability for more generalised complementary studies. Additionally, serological surveys assessing the history of infection with SARS-CoV-2 in different population groups will allow the identification of risk factors for infection, whether symptomatic or not. Both ethnicity (Black and Asian individuals at higher risk; from a single study in England, Northern Ireland and Wales)(29) and higher BMI were found to be associated with disease severity within the included literature,(29, 36) again from descriptive studies only. While these studies were not eligible for our review, we note a series of reports from non-comparative studies documenting the potential influence of ethnicity on the likelihood of getting COVID-19 e.g. the work of Price-Haywood from the US.(51) Male sex was reasonably consistently shown to be a risk factor for presence of COVID-19 but not with severity of disease or mortality.(23, 29, 39) As with ethnicity, socioeconomic and behavioural factors make this association likely to vary between settings.
In considering the role of comorbidities in COVID-19, it is important to consider the underlying pathology of the virus. Respiratory coronaviruses associated with the common cold in immunocompetent people generally affect only cells in the upper respiratory tract (URT), whereas the previously discovered highly pathogenic coronaviruses SARS-CoV and MERS-CoV affect cells in the URT and lower respiratory tract (LRT). SARS-CoV-2 has been shown to do the same,(53) and one of the host cell receptors it targets is Angiotensin- Converting Enzyme 2 (ACE2), with a second major receptor being Transmembrane Serine Protease 2 (TMPRSS2).(54) SARS-CoV-2 can infect all the major cell types in the respiratory tract – type I and type II pneumocytes, alveolar macrophages and endothelial cells.(55, 56) This infection leads to cell death, with significant leaking of fluid into the alveolar spaces (pulmonary oedema), which compromises gas exchange,(57) eventually leading to ARDS. The inflammatory response adds aggregation of repair proteins such as fibrin, which can lead to creation of hyaline membranes which further reduces the surface available for gas exchange.(57) Subsequently, inflammatory cells are activated, recruited by release or exposure of cytokines such as the interleukins (IL) 1β and 6, monocyte chemoattractant protein-1,(55) and proteins of the extracellular matrix, as well as upregulation of the complement system. Inflammatory cells release cytokines which have systemic effects, eventually leading to disseminated intravascular coagulation (DIC), hypotensive shock and metabolic disturbances if not checked.(57)
This pathogenesis therefore offers several points where co-morbidities may exacerbate the process. The target receptor TMPRSS2 is modulated in response to air pollution and in autoimmune conditions such as asthma,(54) which may affect the number of receptors available for SARS-CoV-2 to target, and ACE2 is involved in the renin-angiotensin system (RAS) which controls blood pressure. Viral interference causes dysfunction, which leads to a pro-inflammatory state and increased vascular permeability in response to changes in vascular contraction and sodium homeostasis – exacerbating the effect from the physical damage to the affected cells. Conditions causing hypertension – both primary and secondary to renal disease, endocrine dysfunctions such as hypothyroidism, cardiovascular dysfunction such as arteriosclerosis, or neurological dysfunctions such as acute stress – also affect the RAS,(57) meaning that these conditions might be expected to exacerbate pathology caused by SARS-CoV-2. Any condition creating a pro-inflammatory state, such as type II diabetes or pre-existing infection, or involving autoimmunity, such as type I diabetes, might also be expected to contribute to increased pathology. There is also the direct effect of cell damage – if the target tissues are already damaged this reduces ‘spare’ capacity and therefore the leeway for adaptation to allow the host to continue to maintain homeostasis whilst still being able to eliminate the pathogen and repair the damage. The need for inflammatory cells to clear the infection is also a potential area of interface with comorbidities e.g. conditions such as unsuppressed HIV infection or congenital deficiencies, or the administration of immunosuppressant drugs.
The effect of ageing was particularly strong within our review, both in terms of the magnitude of effect estimates and the number of studies presenting evidence. As well as the above impact of comorbidities, we note that the host’s age may influence pathogenesis, both in terms of the likelihood of having various comorbidities, and also due to its effect on the immune system. Indeed, the immune system becomes less effective over time (immunosenescence), which affects the quality and number of immune system cells generated.(58) Given the scale of the impact of age documented within this review, it seems unlikely that its effect can be explained by a single or a small number of comorbidities which are yet to be detected. This opens up the need to explore biological markers, for example ACE2,(59) and markers of immunosenescence.
The strengths of our review include its systematic approach and broad use of search terms to avoid missing studies. We additionally present a quality assessment to aid the interpretation of the strength of the evidence. In some instances, included publications may have focussed on one specific outcome, whereas our quality assessment took the perspective of the outcomes extracted for this review. We were unable to detect instances where two publications used the same patient populations for their analyses, potentially over-emphasising certain findings. Given the global nature of the pandemic, our review includes studies from around the world, albeit with a large preponderance from China, including studies conducted early after the emergence of SARS-CoV-2 when the at-risk population was predominantly those who had contact with Huanan seafood market and their contacts, and not necessarily representative of the general population. We note a particular lack of studies from the African continent and the Americas, which may have implications for generalisability. Given the rapidly evolving literature on COVID-19, we also note our exclusion of studies published after April 2020, and our exclusion of preprints (which was undertaken to ensure that all included studies had undergone an external quality assessment prior to inclusion).
Across the included publications, variability in study design, exposure and outcome measurement, and analyses made exact syntheses of effect sizes across different risk factors very difficult. Measures of disease severity varied, e.g. admission to ICUs or clinical parameters such as percentage oxygen saturation of the blood. Even measures such as admission to ICU can be subjective and may be time-, clinician-, and health systems- dependent. If severity is recorded at admission, risk factors may reflect issues associated with delayed access to healthcare, which may differ between settings and healthcare systems. It is also important to note that, in some studies of disease severity, mild disease included both people who were hospitalised with symptoms and asymptomatic individuals identified through contact tracing. Generally, analyses were descriptive or univariable and thus did not control for confounding. As documented above, this may be particularly problematic when it comes to separating the impact of age and the presence of comorbidities, as well as for identifying which comorbidities truly increase risk, given that many patients may have multi-morbidity.
The implications of our findings are two-fold for COVID-19, firstly for current public health practice and secondly for the design of future studies. We flag a number of factors of interest that should be considered by governments and public health agencies when designing shielding strategies and the targeting of future vaccines, as well as in mathematical modelling projecting the likely impact of the pandemic over time. We note, however, the need for sensitive handling of population groups deemed to be at higher risk, and how such labelling does not devolve responsibility from public bodies to these individuals for their own welfare.(8) Some public health agencies are now including reporting of potential risk factors in their routine outputs, including ICNARC (included in this review)(29) and the newer European Centre for Disease Prevention and Control reports, which were released after this review was conducted.(60)
Our review demonstrates both the volume of literature that can be published within only a few months since the appearance of an emerging infectious disease, and the need for co- ordinated approaches to such pathogens. Global efforts using national datasets are hugely valuable in systematically determining the aetiology of a disease, particularly to detect smaller effect sizes. Determination of the exact threshold of important risk depends on public perceptions of the disease,(61) as well as policy needs. Data collection should be standardised where possible, e.g. by using consistent definitions of outcomes and the treatment of exposures (for example for hypertension, given that blood pressure is continuous). (For COVID-19 we note both the valuable World Health Organization interim guidelines on its management in providing consistent approaches for testing and the definition of ARDS,(14) and that platforms such as the International Severe Acute Respiratory and Emerging Infections Consortium (ISARIC) have aimed to facilitate such standardisation.(62)) The choice of comparison groups should also merit careful consideration; comparison to other forms of the same condition (e.g. SARS and MERS for COVID-19), although interesting, provide little information about risk groups to be currently acted upon. Where key potential risk factors of interest, such as deprivation, are linked to both the disease of interest and the comparator condition, this limits the inferences possible. Saying this, studies of COVID-19 with the comparator group of other forms of viral pneumonia are a useful complement to studies using a general population comparator, as they show whether people with particular risk factors are at risk over and above what they might experience from ‘normal’ respiratory viruses, which might inform the level of additional precautions they could consider taking.
Finally, appropriately adjusted multivariable analyses should be prioritised, in order to separate the implications of different risk factors and to infer true causal relationships, for example exploring specific markers of comorbidity severity and control, such as the use of specific medications. Early clinical studies during pandemics are critically important and published rapidly under extremely difficult circumstances, but we would argue that high- quality epidemiological studies should also be seen as a priority, and that emergency response plans should include provision of appropriate epidemiological and statistical expertise.