Risk factors for severe outcomes of COVID-19: a rapid review

Review Approach

The urgent need for empiric evidence to inform the prioritization of pandemic immunization strategies in Canada necessitated a rapid but rigorous approach to synthesizing the currently available data. Therefore, we performed a rapid review informed by traditional systematic review methodology, [12] with several modifications to allow for the evidence to be synthesized on an expedited timeline (e.g., single reviewer for study selection, data extraction, and assessment of risk of bias) and focusing on studies having high applicability to Canada (e.g., countries with universal health care system)

NACI’s High Consequence Infectious Disease Vaccine Working Group was consulted to develop and refine the scope of the review (i.e., priority population(s), risk condition(s)/factor(s), and outcomes of interest), but was not involved in the conduct of the review. The working group was not involved in selection of studies nor the synthesis of findings.

The review was conducted following an a-priori protocol (PROSPERO #CRD42020198001). Because there is not yet formal guidance on the reporting of rapid reviews, reporting adheres to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA).[13]

Literature Search

A health sciences librarian searched Ovid MEDLINE(R) ALL on 15 June 2020 using concepts related to COVID-19, P²ROGRESS And Other Factors, and severe outcomes (Supplemental File). The search was limited to studies published in English or French in 2020. Additionally, the search was limited to populations in countries that are members of the Organisation for Economic Cooperation and Development (OECD),[14] in an effort to include studies of highest relevance to the Canadian context. Editorials and letters were excluded. We supplemented the Medline search by hand-searching Epistemonikos COVID-19 in L^OVE Platform (https://app.iloveevidence.com/topics) and McMaster COVID-19 Evidence Alerts (https://plus.mcmaster.ca/COVID-19/) for relevant prognosis or aetiology studies up to 12 June 2020. A hand-search of relevant websites recommended by the NACI working group was also undertaken, as well as continual surveillance for publication of eligible pre-prints located by the search. Searches were exported to an Endnote Library (X9, Clarivate Analytics, Philadelphia, PA) and duplicates removed.

Eligibility Criteria

We included studies published in English or French since 1 January 2020 that reported on the magnitude of association between potential P²ROGRESS And Other Factors and severe outcomes of COVID-19 (Supplemental File). Eligible populations, in order of priority, were people (a) from a general/community sample, (b) with COVID-19 confirmed (by laboratory testing or epidemiologic linkage), (c) hospitalized with COVID-19, and d) with a risk factor of interest. To ensure relevance to the Canadian context, studies had to be conducted in OECD countries;[14] we included studies from countries that do not provide universal (or near universal) coverage for core medical services (i.e., Chile, Greece, Mexico, Poland, the Slovak Republic, and the United States)[15] but considered these to be less applicable to the Canadian context when interpreting the findings.

The exposures of interest were any P²ROGRESS And Other Factors believed to be associated with differential health outcomes across population groups (i.e., pre-existing conditions, place or state of residence, race/ethnicity/culture/language, immigration, refugee status, occupation, gender identity or sex, religion or belief system, education or literacy level, socioeconomic status, social capital, age, and other factors).[16, 17] Eligible comparators were population groups that did not have the P²ROGRESS And Other Factor, or experienced a P²ROGRESS And Other Factor to a different degree (e.g., older vs. younger). Factors could be present among a population with or without COVID-19. The infection must have been confirmed by laboratory testing or linked epidemiologically (e.g., household contact). Studies including populations with other pandemic-related infections (e.g., Severe Acute Respiratory Syndrome, Middle East Respiratory Syndrome) were excluded if data specific to COVID-19 cases could not be isolated. We also excluded studies of interventions and where the entire study population had severe disease (e.g., ICU settings).

Any length of follow-up for outcomes of interest was acceptable. Eligible studies reported on at least one primary outcome (i.e., rate of hospitalization, hospital length of stay, severe disease [as defined by study authors; for example, composite outcome of ICU transfer or death], ICU admission and length of stay, need for mechanical ventilation [MV], and mortality [case fatality or all-cause]). In order to prioritize the most rigorous and applicable evidence, we included only prospective and retrospective cohort studies that employed a multivariate analysis and provided results of the independent contribution of P²ROGRESS And Other Factors to severe outcomes, while accounting for potential confounders (minimally age and sex). Pre-prints were included only if they were accepted by a peer-reviewed journal; pre-prints that were later published (between the date of the search and manuscript submission) were included. Government reports from hand-searched websites were eligible.

Study Selection

All records retrieved by the searches were exported to a Microsoft Office Excel (Microsoft Corporation, Redmond, WA) spreadsheet for screening. After piloting the eligibility criteria on a sample of 70 records, one reviewer independently screened records for inclusion by title/abstract, and those deemed to be potentially relevant were assessed by full text. Uncertainties about the inclusion of any full text study were resolved through consultation with a second reviewer.

Data Extraction

Following a pilot round, one reviewer independently extracted data from each included study into an Excel workbook. We extracted data on (a) population size and demographics, (b) setting, (c) dates of data collection, (d) COVID-19 ascertainment method, (e) co-infections, (f) outcomes reported with definitions for composite outcomes (e.g., severe disease), (f) number of participants analysed, (h) relevant outcome data related to P²ROGRESS factors of interest. For both continuous and dichotomous outcomes, we extracted adjusted relative effect sizes (i.e., odds ratio [OR], risk ratio [RR], hazard ratio [HR]) and measures of variability (95% confidence interval [CI]). A second reviewer was consulted in the event of uncertainty about any of the extracted data. Given the expedited approach, we extracted only data that were reported within the included studies and made no attempt to contact authors for missing or unclear data.

Quality Assessments

To expedite quality assessments, we did not use a formal tool; instead we focused on key variables that were considered to be most relevant to the topic, and that would allow for meaningful stratification of studies by quality. The key variables that we used to assess the quality of the included studies were (a) the extent of adjustment for relevant covariates (i.e., basic adjustment for age and sex, versus more extensive adjustment for numerous potential confounders including comorbidities), (b) follow-up duration and extent of censorship for some outcomes (e.g., ≥2 weeks for mortality), and (c) inappropriate or large exclusions from the study and/or analysis (e.g., missing data on risk factor status or analytical variables). Following assessment of these key variables by a single reviewer, studies without concerns for all three criteria were rated good while others were rated fair. A second reviewer was consulted in the case of uncertainty about the assessment of any individual study.

Synthesis

Given substantial clinical (e.g., risk factors and/or comparators examined, outcome definitions) and methodological (varying covariates included in the adjusted analyses, different measures of association) heterogeneity, it was not appropriate to pool the studies statistically. Instead, we present a narrative summary of the results across studies for each risk factor. When making conclusions about the association between a P²ROGRESS And Other Factor and an outcome, we focused primarily on the magnitude of effect rather than statistical significance, which is heavily dependent on sample size. We categorized associations to be small/unimportant (odds ratio [OR] or risk ratio [RR] ≤1.70), moderate (1.71 to 1.99), large (≥2.00), or very large (≥5.00). [18] When determining the magnitude, we compared findings across all relevant studies and often relied heavily on the findings of the largest and/or good quality studies.

Certainty of Evidence

The expedited approach to evidence synthesis did not allow for a formal appraisal of the certainty of evidence across studies for each P²ROGRESS And Other Factor-Outcome association. Instead, a single reviewer assessed the certainty of the evidence for each association considering relevant components of the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach: [19, 20] (a) directness in terms of country (presence of universal healthcare) and source population (community sample vs. hospitalized patients), (b) sample size (n<500 considered small) and magnitude of association, (c) study quality, and (d) consistency of associations (in direction and magnitude) across studies. Bodies of evidence started at high certainty[21] and were rated down for weaknesses in any of the aforementioned characteristics. The level of certainty in associations are referred to using the terms ‘uncertain’ (no or very low certainty), ‘may’ (low or some certainty), and ‘probably’ (moderate certainty).[22] At least two other reviewers confirmed the certainty of evidence appraisals, with disagreements resolved by discussion.

Characteristics of Studies

Of 3,740 unique records identified by the searches, 949 were screened at full text, and 34 studies that reported on 32 unique populations were included in the review (Figure 1; Supplemental File shows studies excluded by full text, with reasons). [23-56] Three studies conducted in the United Kingdom (UK)[39, 44, 47] used overlapping cohorts from a single medical/research database, and were considered as a single population in the analysis. Another large UK study[56] is likely to also be overlapping with these populations, but the degree of overlap is not known.

• Download figure
• Open in new tab

Figure 1. PRISMA flow of study selection

Table 1 shows the characteristics of the included studies (full details about individual studies in Supplemental File). Briefly, all of the included studies were prospective or retrospective cohorts. The studies were published between 23 April and 6 July 2020, and half (17/34, 50%) reported on populations in the United States.[23, 24, 31, 32, 36-38, 40-43, 45, 46, 49, 51, 53, 54] The remaining countries represented (Italy,[25, 27-30, 35, 50, 55] Spain,[26] UK[33, 39, 44, 47, 48, 52, 56]) all have universal or universal-like healthcare (one study used data from 17 countries). All studies reported on adults, and the overall median was 596 participants (range 44 to 418,794). The mean or median age of the populations studied ranged from 42 to 84 years (32/34 [94%] 54 to 71 years). Most studies (16/34, 47%) examined the association between risk factors and outcomes in a hospitalized population. Studies most commonly reported on the independent association of pre-existing conditions (n=27 studies), gender identity or sex (n=18), and race or ethnicity (n=12) with severe outcomes (most commonly hospitalization, n=9). P²ROGRESS And Other Factors not examined in the included studies were immigration or refugee status, religion or belief system, social capital, and substance use disorders. There were also no data specific to pregnant women, indigenous populations, people with disabilities, nor different ages in children.

Study Quality

The majority of studies (19/34, 56%) were rated as good quality[23, 24, 29, 31, 33-36, 40, 42, 45, 46, 48-53, 55] because they adjusted for age, sex, and pre-existing disease in their analysis, had adequate follow-up of outcomes, and few or no missing data. The remaining studies had flaws in one or more of the three domains that we considered to be most important for this review.

Association Between Risk Factors and Outcomes

Table 2 shows a summary of findings for associations between each reported risk factor and outcomes of interest; all contributing data are shown in the Supplemental File.

There was low or moderate certainty of evidence for important/large associations with increased risk of hospitalization in people having confirmed COVID-19, for the following risk factors: obesity class III (body mass index >40 kg/m²; 1 study, n=5,297),[49] heart failure (2 studies, n=6,331),[23, 49] diabetes (2 studies, n=6,331),[23, 49] chronic kidney disease (confirmed COVID-19 or community sample; 2 studies, n=424,073),[47, 49] dementia (1 study, n=418,794),[47] age over 45 years (vs. 45 or younger; 2 studies, n=6,331),[23, 49] male gender (3 studies, n=3,812),[23, 49, 51] black race/ethnicity (vs. non-Hispanic white; confirmed COVID-19 and community samples, 5 studies in 4 populations, n=428,606), [23, 44, 47, 49, 51] homelessness (1 study, n=1,052),[23] and low income (<25^th vs. >50^th percentile; 1 study, n=1,052).[23] Age over 60 and over 70 years may be associated with important increases in the rate of mechanical ventilation (1 study, n=486)[40] and severe disease (1 study, n=2,725), [49] respectively.

For mortality, important associations with increased risk may exist for liver disease (2 studies, n=20,597),[33, 53] Bangladeshi ethnicity (vs. British white; 1 study, n=130,091), [56] and age over 45 years (vs. <45 years; 3 studies, n=87,819).[33, 49, 56] The data were somewhat inconsistent for gender, with most studies showing moderate certainty of no important effect, but one large fair quality study (n=130,091)[56] from the UK that stratified its analysis by age showed that hospitalized males aged 2064 years (but not older) may be at about two-fold increased risk of mortality compared to females. Associations with hospitalization and mortality may be very large for those aged over 60 years (2 studies, n=6,331 for hospitalization;[23, 49] 3 studies, n=24,163 for mortality[33, 41, 49]) and are probably very large for those over 70 years (2 studies, n=6,331 for hospitalization; [23, 49] 2 studies, n=22,858 for mortality[33, 49]). One study (n=63,094)[56] directly compared subgroups of older adults, showing that compared to those aged 65-69 years, there may be no important increased risk of mortality among hospitalized adults aged 70-79 years, but risk may increase about 2-fold for those 80 years and older. Studies treating age on a continuum or across small increments consistently found that risks for hospitalization and mortality increased with increasing age (e.g., approximately 2-6% and 5-10% relative increase in risk per year) (3 studies in 2 populations, n=422,275 for hospitalization; [44, 47, 51] 11 studies, n=6,877 for mortality).[25-27, 31, 35, 38, 45, 46, 48, 51, 55]

Moderate associations may exist for increased risk of mechanical ventilation (4 studies, n=1,559)[38, 40, 42, 46] and ICU admission (2 studies, n=873),[38, 42] and severe disease (1 study, n=2,725)[49] with obesity (body mass index >30 or 40 kg/m²); severe disease with heart failure (1 study, n=2,725);[49] mortality with haematological malignancy (1 study, n=1,183);[52] mechanical ventilation with male gender (4 studies, n=881);[27, 40, 42, 46] and hospitalization with social deprivation (highest vs. lowest quintile; 1 study, n=340,996).[44]

There was moderate certainty evidence for no important increase in risk of hospitalization with chronic respiratory conditions (4 studies in 3 populations, n=425,125), [23, 44, 47, 49] cardiovascular disease apart from heart failure (i.e., coronary artery disease, hypertension, hyperlipidaemia; 4 studies in 3 populations, n=425,125),[23, 44, 47, 49] non-specific cancer (2 studies, n=6,331), [23, 49] Asian race/ethnicity other than Bangladeshi (vs. non-Hispanic white; 3 studies in 2 populations, n=424,073), [44, 47, 49] and current or former smoking (5 studies in 3 populations, n=425,125). [23, 39, 44, 47, 49] Additionally, there was moderate certainty evidence for no important increase in severe disease with chronic respiratory conditions (1 study, n=2,725), [49] chronic kidney disease (2 studies, n=2,922),[24, 49] nonspecific cancer (2 studies, n=2,769), [29, 49] and Black race/ethnicity (vs. nonHispanic white; 2 studies, n=3,030);[36, 49] and no important increase in risk of mortality with obesity (body mass index >30 kg/m²; 6 studies, n=8,716),[35, 38, 43, 46, 49, 51] chronic respiratory conditions (4 studies, n=23,315),[31, 33, 46, 49] diabetes (4 studies, n=23,315), [31, 33, 46, 49] chronic kidney disease (3 studies, n=23,058), nonspecific cancer (3 studies, n=24,041),[33, 49, 52] male gender (9 studies, n=27,875),[25-27, 31, 33, 35, 46, 49, 51] Black (5 studies, n=135,418)[38, 48, 49, 51, 56] or Asian race/ethnicity (vs. non-Hispanic white; 3 studies, n=4,015),[38, 48, 49] and social deprivation (lowest vs. highest quintile; 1 study, n=130,091).[56] Overall, there were few data for the ICU and mechanical ventilation outcomes.

Strengths and Limitations

The expedited methods used in this review allowed for a rapid but comprehensive synthesis of the highest quality evidence available on multiple risk factors associated with severe COVID-19 outcomes that is applicable to OECD countries. Generalizationsto other countries should be made with caution, as high risk groups in these populations may differ. We excluded studies only examining patients with severe COVID-19 (i.e., in ICU settings), and therefore our findings for mechanical ventilation and mortality are applicable to people with COVID-19 or in general populations, but not necessarily all those with severe infection. Most studies of patients in the ICU setting that we located were relatively small and descriptive in nature, such that many would have been excluded due to lack of adjustment or only have been able to provide low or very low certainty evidence due to their lack of precision. As described previously, many available studies do not control for any important confounding variables which limited the number of studies and risk factors included in this review. Given the rapid emergence of new evidence on the topic, potential associations (or lack of association) for which only low or very low certainty of evidence is available should continue to be reviewed as new primary research is published. There is a need for high quality primary research (accounting for multiple confounders) to better understand the level of risk that might be associated with immigration or refugee status, religion or belief system, social capital, substance abuse disorders, pregnancy, indigenous identity, living with a disability, and differing levels of risk among children in various age groups.