As of April 30, 2020, more than 3 million people worldwide contracted severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and close to 250,000 people died of coronavirus disease (COVID-19) complications. In Israel, by that date, 16,004 individuals had been infected by the virus and 223 died from the disease. This pandemic poses grave challenges to patients, health care providers, and policy makers. Many of these challenges may be better addressed with timely stratification of patients to risk groups, based on their past and current medical characteristics. For example, reliably identifying patients at increased (or decreased) risk could guide clinical decisions (eg, hospitalization vs home care), public health policies (eg, risk-based quarantine), and preparedness efforts (eg, expected medical equipment required).

Various algorithms for identifying patients at risk for COVID-19 (severe) complications have been proposed. The Centers for Disease Control and Prevention (CDC) identified individuals 65 years and older, living in a nursing home or long-term care facility, or suffering from underlying medical conditions, particularly if not well controlled, as being at high risk for severe illness from COVID-19 [1]. Similarly, the European Centre for Disease Prevention and Control (ECDC) lists the age category >70 years and some underlying conditions as risk factors for critical illness [2]. The United Kingdom National Health Service (NHS) included solid organ transplant recipients, patients with specific cancers or severe respiratory conditions, pregnant women with significant heart disease, and those with increased risk of infection (eg, due to immunosuppressive therapies) in the highest clinical COVID-19 risk group [3]. In April 2020, approximately 1.3 million people in this group were asked to “shield” by staying at home for a period of at least 12 weeks. In addition, patients >70 years and those suffering from some underlying health conditions (eg, chronic respiratory diseases, BMI ≥40, and pregnant women) were considered in a wider vulnerable group (also referred to as the “flu group”). Finally, a more quantitative risk model (derived from Barda et al [4]) was adopted by the Israeli Ministry of Health (MoH), assigning a point for each underlying condition from a predefined list, then considering age group and point count to identify high-risk patients.

Initially, these algorithms were derived from a quickly growing number of epidemiological characterization studies (eg, [5,6]), which report the prevalence of various conditions in a population of interest, typically severe, hospitalized COVID-19 patients. These studies provide timely and important information; however, identifying risk factors calls for a comparative analysis, contrasting the prevalence of conditions in case and control populations. To date, only a handful of studies implemented such an approach, using, for example, the general population [7] or a confirmed (and symptomatic) COVID-19 patient cohort [8]. Similar to these efforts, we analyze here the medical records of all SARS-CoV-2–positive patients in a nationwide health organization (covering 2.3 million individuals). We compare the prevalence of existing conditions in complicated and noncomplicated cohorts and identify those conditions associated with COVID-19 complications in various age and sex strata. Our analysis highlights stratum-specific risk factors and may allow better identification of patients at risk in different subpopulations.

The MHS SARS-CoV-2–positive cohort included 4353 individuals, of whom 173 deteriorated to moderate (n=87, 50%) or severe condition (n=45, 26%), were admitted to the intensive care unit (n=66, 38%, partly overlapping with other conditions), or died (n=21, 12%). This group of patients make up the complicated COVID-19 cohort. Overall, patients in the complicated COVID-19 cohort were older, suffered from more comorbidities, and were predominantly male (Table 1). Moreover, the prevalence of COVID-19 complications increased with age and more steeply for men than for women (Table 2). The risk of COVID-19 complications in men <70 years was significantly higher than in women (eg, P=.01 for patients 60-70 years old; see Table 2).

Table 1

Characteristics of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)–positive, complicated, and noncomplicated coronavirus disease (COVID-19) patient cohorts.

Table 2

Association of male sex and coronavirus disease (COVID-19) complications across age groups.

Comparing the prevalence of existing conditions in the three age groups between the complicated and noncomplicated COVID-19 cohorts revealed multiple risk factors, including obesity for patients 18-50 years old (OR 11.09, 95% CI 4.15-32.67; P<.001), chronic kidney disease for patients 50-65 years (OR 4.06, 95% CI 1.89-8.38; P=.005), and neurological disorders (OR 2.65, 95% CI 1.69-4.17; P<.001) for patients ≥65 years (for a complete list, see Table 3 and Multimedia Appendix 1).

Stratifying by age (below and above 65 years) and sex (Table 4 and Multimedia Appendix 1), we observed that kidney diseases are a risk factor in all strata (eg, OR 3.45, 95% CI 1.57-8.06; P=.02 in women ≥65 years). Additional risk factors included hypertension in males under 65 years (OR 4.56, 95% CI 2.35-8.55; P<.001); neurological disorders in females ≥65 years (OR 3.55, 95% CI 1.68-7.74; P=.008); cognitive impairment (OR 4.18, 95% CI 1.81-9.72; P=.009) and depression (OR 2.94, 95% CI 1.55-5.58; P=.01) in males ≥65 years. Respiratory diseases and smoking, while typically more prevalent in complicated COVID-19 patients, were not identified as significant risk factors (eg, chronic obstructive pulmonary disease in patients ≥65 years: OR 1.36, 95% CI 0.75-2.4; P=.63) (see Multimedia Appendix 1).

Table 3

Most statistically significant conditions associated with increased risk of coronavirus disease (COVID-19) complications in age-stratified patient groups.

We compared the prevalence of dozens of existing conditions in Israeli SARS-CoV-2–positive and complicated COVID-19 patient cohorts to highlight conditions associated with a high risk of complications. A few other studies have employed a similar study design to identify risk factors for COVID-19 complications. For example, Ebinger et al [8] studied a cohort of symptomatic COVID-19 individuals (N=442) and examined the association of existing conditions with disease severity; and the OpenSAFELY Collaborative explored the risk of COVID-19–related hospital death in the general population (N>17 million). We emphasize that cohort composition dictates the research question it can address: our analysis focuses on SARS-CoV-2–positive individuals, hence searches for risk factors of complications in patients who already contracted the virus (but are potentially asymptomatic), while studying the general population may combine risk factors for infection and severe COVID-19 outcome. Additionally, cohorts that consider only a subset of patients, defined based on disease outcome (eg, symptomatic or hospitalized) or otherwise nonrepresentative of the entire population (eg, demographically skewed) may introduce biases to the analysis [21]; instead, we study here all SARS-CoV-2–infected patients in a large, nationwide health organization.

Multiple studies (eg, [7,22]) have shown that COVID-19 complications are most strongly associated with age and sex. Stratifying by these factors provides readily interpretable insights on the supplemental associations (in addition to older age and male sex) between pre-existing conditions and disease complications.

Many conditions highlighted by our analysis have been previously reported [5,6,8] and are part of commonly used at-risk definitions [1,3], including hypertension, obesity, as well as kidney and cardiovascular diseases. We do, however, identify a few additional risk factors, notably depression in patients aged 18-50 years and males ≥65 years; and cognitive and neurological disorders in patients ≥65 years. These additions may be, in part, associated with the different age distribution in the ≥65 years group (median 76 years, IQR 70-83.5 years versus 72 years, IQR 68-78 years, in the complicated and noncomplicated COVID-19 cohorts, respectively) and rely on small sample size (only 7 patients aged 18-50 years with depression in the complicated COVID-19 cohort; Table 3). Nonetheless, with some preliminary support [7], they may deserve more consideration in future studies. Our analysis also points out to the reduced importance of respiratory diseases and smoking. Both conditions appear as factors in most at-risk definitions [3,5]: chronic obstructive pulmonary disease has been associated with severe COVID-19 in multiple studies [23] (though not all [6]), while the role of smoking has been somewhat controversial [23,24]. The discrepancies between our analysis and previous reports likely stem from the different cohorts analyzed: SARS-CoV-2–positive individuals, ranging from asymptomatic to severe COVID-19 versus hospitalized COVID-19 patients, respectively. Other study-related attributes (eg, country-specific characteristics) may also contribute to the varying importance of the studied risk factors.

In parallel to the COVID-19 epidemiological characterization efforts, researchers have also attempted to use retrospective observational data to derive risk models for severe COVID-19 patients [25]. Such models require ample data of COVID-19 patients for both model training and performance assessment. As such data are scarce at present, some models compromised on using data for other diseases with, supposedly, similar clinical trajectory and complications. For example, DeCapprio et al [26] trained models on US Medicare claims data to predict inpatient visits with a primary diagnosis of either pneumonia, influenza, acute bronchitis, or other specified upper respiratory infections as proxy for COVID-19 complications. However, as previously reported (eg, [27]), and in agreement with our analysis, severe COVID-19 patient characteristics differ considerably from that of other diseases, thus limiting the generalizability of such models to COVID-19 and requiring adjustments to their parameters [4].

Our study has several limitations. First and foremost, it relies on routinely maintained electronic health records, which may be inaccurate and incomplete [28]. Second, the number of complicated COVID-19 patients in the MHS data is below 200, limiting the statistical power of our analysis. Third, health care policies and, in particular, testing criteria, may systematically bias the composition of the SARS-CoV-2–positive cohort. Fourth, asymptomatic and patients with mild symptoms of COVID-19 (currently in the noncomplicated cohort) may deteriorate and eventually be part of the complicated cohort, potentially modifying the results of the analysis. Fifth, our analysis is univariate in nature, testing the association of individual conditions with COVID-19 complications; as such, it is unable to uncover more complex relations (eg, interdependencies between existing conditions and COVID-19 complications), which may be discovered by multivariate analysis. Finally, we focused on data from Israel; characteristics in other geographies may differ [27]. We attempted to mitigate some of these limitations by age and sex stratification and robust estimations of statistical significance. We also note that, at the current point in time, many of these shortcomings are shared by all published research on COVID-19.

Notwithstanding these limitations, our work adopts a novel vantage point to the problem of identifying patients at increased risk for COVID-19 complications. Importantly, as SARS-CoV-2 containment efforts focus on patients at risk for severe complications (eg, shielding vulnerable population in the United Kingdom [3]), changes in the list of considered conditions may have a substantial effect on a large number of individuals, thus calling for continuous fine-tuning of the corresponding definitions.