COVID-19 in patients with hepatobiliary and pancreatic diseases: A single-centre cross-sectional study in East London

Study setting and data sources

All data utilised for this study were collected and processed under the East London Pancreatic Cancer Epidemiology (EL-PaC-Epidem) study at BHNT. In brief, EL-PaC-Epidem is an ongoing study that ascertains patients diagnosed or reported with HPB diseases including cancers, as well as control patients (e.g., small intestine, hernia), within five BHNT hospital sites (The Royal London Hospital, Newham University Hospital, St Bartholomew’s Hospital, Whipps Cross University Hospital, Mile End Hospital) between 2008 and 2021. The EL-PaC-Epidem study was approved by the East of England - Essex Research Ethics Committee (19/EE/0163; 17 May 2019) and supported by the NHS Confidentiality Advisory Group for collecting and processing confidential patient information without consent (19/CAG/0219; 17 January 2020). The study is limited to the secondary use of a specified subset of patients’ retrospective EHR generated during the course of normal care of these patients. It links EHRs from different data sources (via UK unique individual NHS numbers), including primary care through General Practitioners (GP) (Discovery East London Programme data service [DDS]) and secondary or tertiary care through hospitals (BHNT Consolidated Data Extract [CDE]). Patients, who have previously informed their GPs or NHS to stop sharing their personal and health records for purposes other than their individual care, were automatically excluded. The current EL-PaC-Epidem study cohort consists of 27321 adult patients (aged 18 years or over), diagnosed or reported with at least one of the HPB conditions (supplemental table 1) between 1 April 2008 and 6 March 2020.

Study design and population

This is a single-centre cross-sectional study utilising the linked EHR data of patients with a history of HPB diseases. Within this specific patient group, the study focused on the incidence of COVID-19, and examined the association of SARS-CoV-2 infection with six common medical comorbidities (i.e., diabetes, hypertension, high cholesterol, cardiovascular disease, chronic respiratory disease, renal disease), lifestyle factors (i.e., smoking, alcohol use, substance misuse, obesity), and use of selected prescription medications.

As the first case of COVID-19 in London was reported on 12 February 2020, we used this as the start date for this study and extracted data on a subgroup of the EL-PaC-Epidem study cohort until 12 June 2020 (figure 1). Eligible individuals were a resident in East London and alive on the study start date (EL-HPB). Residency of East London was inferred if a patient had at least one appointment or prescription issued from a GP in East London boroughs or had a scheduled or unscheduled visit to one of the BHNT hospitals within the last six months (after 12 August 2019). Patients with confirmed SARS-CoV-2 infection were identified by: i) the presence of International Classification of Diseases 10^th edition (ICD-10) or Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) codes for confirmed COVID-19 or SARS diagnosis assigned in their hospital encounters or GP records during the observation period between February 12 and June 12, 2020 (supplemental table 2) OR ii) positive record of SARS-CoV-2 RNA through BHNT oral and/or nasal swabs test during the same period. For confirmed COVID-19 cases, the earliest date of diagnosis or positive swab test was considered as the index date, whereas 12 February 2020 was considered as index date for rest of the cohort. Patients, who were assigned an ICD-10 or SNOMED CT diagnosis code for suspected COVID-19 but were neither reassigned to confirmed diagnosis nor positive RNA test, were excluded from the analysis.

• Download figure
• Open in new tab

Figure 1.

Selection of patients for the cross-sectional study.

We also examined the onset-to-death distribution within the patient group with a confirmed COVID-19 diagnosis (EL-HPB-COVID). Mortality data was collected on 12 October 2020. Following the latest Public Health England (PHE) definition³⁰, the death of a patient within 28 days of the index date is considered as a COVID-19 related death. This is different from a 60-day window that was being used in the UK prior to 12 August 2020 to define COVID-19 related death. To ensure consistency, COVID-19 patients who survived beyond 60 days of index date are considered as survivor in the study; Nine patients who died between 29 and 60 days of diagnosis were excluded from the analysis. The onset-to-death distribution was analysed in the context of same set of comorbidities, lifestyle factors and medication use, as well as cardiovascular, respiratory and renal complications during hospital care.

Procedures

All patient data were obtained from retrospective EHR, harmonised across hospital and GP coding systems where applicable, and organised into 40 primary variables across seven categories corresponding to the focus of the study (table 1). BHNT CDE uses 2011 UK census grouping to record ethnicity, ICD-10 or SNOMED diagnosis codes for clinically relevant diagnoses, and Office of the Population Censuses and Surveys Classification of Interventions and Procedures version 4 (OPCS-4) procedural codes for treatments and procedures. Physiological observations (weight, body-mass index [BMI], blood pressure) and laboratory tests results are available in locally developed terms. Semi-structured text entries such as discharge summaries, past medical history and a lifestyle questionnaire collected during the pre-operative assessment, and presenting symptoms from scheduled or unscheduled hospital visits are also available. All GP records via DDS were available in Read Codes v2 or Clinical Terminology Version 3 (CTV3) codes, except the prescribed medication records and COVID-19 diagnosis which were available in SNOMED codes. For each variable, we consulted ICD-10, SNOMED, Read, CTV3 or OPCS-4 dictionaries as appropriate to construct the mapping codelists. For some variables, codelists also included keywords to conduct automated sub-string search within semi-structured text as well as local laboratory test and physiological observation terms.

Rule-based phenotyping algorithms were developed for each categorical variable to characterise patients, integrating information from multiple sources where available to counteract bias. HPB diseases were grouped into four categories (supplemental table 1): any malignant disease, and non-malignant diseases of liver, pancreas or biliary tract. Non-malignant liver diseases were further divided into mild and moderate to severe subgroups, extending the definition from CDMF Charlson Comorbidity Index,³¹ whereas non-malignant pancreas or biliary diseases were divided into acute and chronic disease subgroups (supplemental table 1). Within each disease category, a patient was assigned to chronic (or more severe) subgroup, when data indicated the history of both acute (or mild) and chronic (more severe) conditions. A patient can either be assigned to a malignant disease category or any of the non-malignant disease subgroups. Ethnicity was grouped into four categories - White, South Asian, Black, and Other. White and Black ethnic groups were defined based on the 2011 UK census classification; Indian, Pakistani and Bangladeshi origin from the Asian group represented South Asian, while the rest (i.e., Mixed, Chinese, other Asian and other ethnic group) were represented in the Other group. The ethnic category recorded at the GP took precedence over hospital records.

Phenotyping algorithms defining the comorbidities were based on diagnosis codes (presence) or semi-structured text search (presence or absence), with the additional inclusion of procedural codes (presence), some observation or laboratory test results (presence) and related medication use (at least three prescriptions). Patients were considered to have or have had a specific medical condition if they met at least one criterion indicating the presence of the condition before the index date, otherwise they were considered negative for the condition.

Phenotyping algorithms defining the lifestyle factors were based on the longitudinal entries (current, past or never) derived from diagnosis codes and free text search, with the additional inclusion of BMI observation for obesity. Obesity was defined as BMI of 30 kg/m² or more. Patients assigned never status at any point but having a record of current or past status before that date were reassigned to past status. The most recent lifestyle record before or on the index date was then used to assign current, past or never status to the patients. Patients with no record of a specific lifestyle factor were classified as having missing data. Patients were assigned current, past or non-user status for medication use variables based on the number of GP prescriptions issued in the last two years for the medicines under specific medication groups. Patients with no record of prescription for particular medications were assigned non-user status. With at least three prescriptions issued, a patient was assigned current user status if the latest issue was within three months preceding the index date, and past user status otherwise. Patients with record of less than three prescriptions were classified as non-user. Patients with COVID-19 were considered to have a specific complication during admitted patient care if at least one of the hospital diagnosis codes from the complications codelist was recorded during the observation period after index date, otherwise they were considered negative for the complication. A patient was considered to have a recurrent complication if they had a history of that particular comorbidity, otherwise it was considered as a novel complication.

Selection of study variables, codelist construction, and phenotyping algorithm development were done in consultation with a panel of clinicians and scientists (HMK, CC, LS). A comprehensive list of codelists and phenotyping algorithms for the study variables are available on the EL-PaC-Epidem portal (https://pac-epidem-el.bcc.qmul.ac.uk/covid-19/).

Statistical analysis

We conducted descriptive analyses for the EL-HPB cohort as a whole, by group for patients with confirmed SARS-CoV-2 infection and the rest (herein referred to as COVID-19 and non-COVID-19 respectively). Differences in demographic and clinical characteristics between the groups were assessed with Pearson’s Chi-square test, Fisher’s Exact test and Kruskal-Wallis rank sum test, as appropriate. P values less than 0.05 were considered significant. Similar descriptive analyses were performed for the EL-HPB-COVID cohort, and by survivor and deceased groups.

To explore the risk factors associated with COVID-19 susceptibility and subsequent survival, the effect size for each variable under investigation was evaluated with odds ratios (ORs) with 95% confidence intervals (CI), using regression models with a binomial distribution. Crude ORs were obtained from univariable regression models, and then simultaneously controlled for a fixed set of potential confounders (gender, ethnicity, age group) using multivariable regression models with Benjamini-Hochberg correction for P values adjustment. The median age of the overall EL-HPB cohort being 57, a simplified binary age grouping (18-60, 61+) was used in multivariable regression models for comorbidity, lifestyle, medication use and post-diagnosis complication analyses. Since a participant with non-malignant HPB diagnoses for multiple organs can be represented in multiple HPB subgroups, the effect estimation for individual HPB disease variables was further mutually controlled for other HPB diseases.

We also conducted more in depth post hoc analysis to evaluate the confounding effect of pre-existing medical conditions by adding comorbidity covariates individually in the multivariable regression models. Finally, effect modification by non-malignant HPB disease subgroups was evaluated by adding interaction terms in the models and comparing them with models lacking this interaction via the likelihood ratio test. Any potential association between HPB diseases and COVID-19 susceptibility/mortality risk factors were further evaluated in stratified analyses according to the disease subgroups.

Patients with missing data for individual categorical variables were included in the descriptive analyses and in regression models for effect estimation. All statistical analyses and visualisations were performed in R (version 3.5.1).

Patient and public involvement

Patients and the public were involved in evaluating the design of the umbrella study (EL-PaC-Epidem), particularly the notion of collection and processing of retrospective patient data without their consent. The support from NHS Confidentiality Advisory Group was obtained based on the positive opinion posed by patient and the public.

Population characteristics

The final EL-HPB cohort consisted of 15 540 patients, after applying the eligibility criteria and excluding 168 suspected but unconfirmed COVID-19 cases. By 12 June 2020, 226 (1.5%; 145 per 10 000 adult population) confirmed cases of COVID-19 were reported in this cohort (figure 1). This was more than three-times higher than in the general population of East London where prevalence of COVID-19 at the same time was 41 per 10 000 adult population.²⁵ More than half of the COVID-19 cases had some form of non-malignant liver diseases (N=138, 62.8%); however, when comparing confirmed COVID-19 cases with the non-COVID-19 cases, we observed a disproportionate infection frequency in patients with chronic pancreatic conditions (14.1% vs 8.8%) and moderate to severe liver conditions (11.4% vs 6.9%). We also observed differences in gender, ethnic origin, and age group between COVID-19 and non-COVID-19 cases (table 2). The proportion of males was higher in the COVID-19 group compared to the baseline non-COVID-19 group (53.5% vs 43.7%, P=0.005). The same trend was observed for Black population (17.7% vs 10.7%). COVID-19 patients were older than non-COVID-19 patients (median 67 years, interquartile range 55.1 to 80.9 years vs 57.1 years, 44.8 to 69.2 years, P<0.001), with a steady increase in infection frequency with age. Some 78.3% of COVID-19 patients had three or more comorbidities, with hypertension being the most common comorbidity (85.4%), followed by high cholesterol and diabetes (table 2). Only eight COVID-19 patients had none of the six medical conditions. In general, COVID-19 patients had a higher rate of past history of smoking, drinking, substance mis-use and obesity compared to the non-COVID-19 group. Consistent with the underlying prevalent comorbidities of the COVID-19 group, history of prescription drugs use associated with managing hypertension or cardiovascular disease (ACE inhibitor, calcium channel blocker, β-blocker, aldosterone antagonists, antiarrhythmic, antiplatelet, anticoagulant), cholesterol (statin), inflammation (glucocorticoid, β2-agonists) or background HPB condition (proton pump inhibitor) were higher in COVID-19 patients (table 2). Intake of vitamin D was also significantly higher in COVID-19 patients.

Between 12 February and 12 August 2020, the all-cause mortality rate in the non-COVID-19 group was 2.4%, whereas the rate in the COVID-19 group during the same period was 27.4% (table 2). When analysing the 53 deceased and 164 surviving patients with confirmed SARS-CoV-2 infection, we found differences in gender (P=0.005) and age (P<0.001); deceased patients were older than the survivors (median 80.4 years, interquartile range 71.7 to 85.1 years vs 62.9 years, 49.8 to 77.4 years) with steady increase in death with age becoming prominent in those above 70 years of age (table 3). We observed a higher mortality amongst South Asian (34% vs 29.3%) and Black (26.4% vs 13.4%) populations, which were even more pronounced when comparing with the all-cause mortality in the non-COVID-19 group (supplemental table 3). Higher mortality was observed for pancreatic and biliary disease patients in general, whereas liver disease patients had higher survival rate (table 3). The median survival period for the deceased patients from the date of confirmed COVID-19 diagnosis was 11 days (interquartile range 2 to 18 days). After stratifying patients according to the comorbidities investigated, the mortality for HPB patients with COVID-19 was at least six-times higher than that of HPB patients without COVID-19 (supplemental table 3). Diabetes, hypertension, cardiovascular and renal conditions, in particular, were associated with mortality in COVID-19 patients (table 3). All except one deceased patient had at least three additional comorbidities, compared to 71.3% of patients who survived. The deceased group had higher proportion of patients with a history of past smoking and current substance mis-use, but no overall differences were observed for drinking and obesity. Notable differences were observed in the use of glucocorticoid, β2-agonists, and statins. Recurrent complications were more common in the deceased group compared to survivors, however frequency of novel respiratory complications was notably higher in the deceased group (39.6% vs 21.3%).

Odds of SARS-CoV-2 infection

The risk analyses showed a greater odds of COVID-19 for men, the Black community, and those with moderate to severe liver disease (figure 2). Patients with chronic pancreatic and mild liver conditions were also associated with a higher odds of infection (OR 1.89, 95% CI 1.25 to 2.85, P=0.007; 1.52, 1.07 to 2.15, P=0.039); however, post-hoc adjustment for comorbidities returned a reduced non-significant positive odds (1.57, 1.04 to 2.38, P=0.084; 1.32, 0.93 to 1.88, P=0.237), with diabetes principally responsible for this reduction (supplemental table 4). The similar association was observed for elderly patients (over 70) with underlying multimorbidity as confounding factor. Patients with pre-existing renal conditions had the highest odds of COVID-19 (2.93, 2.2 to 3.89, P<0.001), followed by a more than two-fold increased odds for patients with hypertension, diabetes, cardiovascular or chronic respiratory disease (figure 2). However, the independent effects of hypertension and high cholesterol were absent when controlled for other comorbidities (supplemental table 4).

• Download figure
• Open in new tab

Figure 2.

Odds ratio estimates of COVID-19 for HPB patients with specific demographic, comorbidity, lifestyle and medication use characteristics. Odds ratio estimates for demographic characteristics are mutually controlled for each other, i.e., gender, ethnicity, and age group. Estimates for HPB disease subgroups are further controlled for each other. For comorbidity, lifestyle and medication use characteristics, estimates are controlled for gender, ethnicity, and dichotomous age group (under and over 60).

Substance mis-users had higher odds of infection, but the higher odds observed for those with history of smoking or obesity were due to underlying comorbidities. Patients on Vitamin D treatment and past users of ACE inhibitors were associated with higher odds of infection. The slightly reduced yet significantly high odds remained after controlling for comorbidities, with renal (for Vitamin D users) and cardiovascular (for ACE inhibitor users) diseases being the principal source of the reduced estimates (supplemental table 4). Higher odds were also observed for users of proton pump inhibitors, glucocorticoid, β-blockers, β2-agonists, aldosterone antagonist, muscarinic antagonist, antiplatelet, antiarrhythmic, and statin compared to the non-users of these respective drugs; however, post-hoc adjustment for comorbidities returned non-significant positive odds for those.

A small number of factors appeared to modify the association between HPB disease subgroups and risk of COVID-19 infection (supplemental table 6). In patients with mild liver disease, the odds of COVID-19 infection doubled for chronic pancreatic disease patients compared to patients with no pancreatic condition (P value for heterogeneity, P-het, by liver disease=0.02). A history of substance-misuse was associated with significantly higher odds of infection, particularly for patients with chronic biliary conditions (P-het by biliary disease=0.03), and mild liver conditions (P-het by liver disease=0.04).

Odds of COVID-19 related death

The risk analyses showed an increased odds of COVID-19 related death for men, individuals from the Black community and patients who had acute respiratory complications during admitted care without a history of long-standing respiratory problems (figure 3). Increased odds of death were also observed for the glucocorticoid and β2-agonists. No HPB disease subgroups were particularly more vulnerable to COVID-related death, although patients with chronic pancreatic condition showed a trend towards significance. Elderly patients (over 70), and recent users of ACE inhibitors and non-steroidal anti-inflammatory drugs were associated with a higher odds of death; however, post-hoc adjustment for comorbidities returned a non-significant positive odds for these risk factors (supplemental table 6). Stratified analyses according to HPB disease subtypes did not reveal any meaningful effect modification, principally due to small EL-HPB-COVID sample size (data not shown).

• Download figure
• Open in new tab

Figure 3.

Odds ratio estimates of COVID-19 related death for HPB patients with specific demographic, comorbidity, lifestyle, medication use and post COVID-19 diagnosis complication characteristics. Odds ratio estimates for demographic characteristics are mutually controlled for each other, i.e., gender, ethnicity, and age group. Estimates for HPB disease subgroups are further controlled for each other. For comorbidity, lifestyle, medication use and post diagnosis complication characteristics, estimates are controlled for gender, ethnicity, and dichotomous age group (under and over 60). Categories with odds ratio P>0.95 are not shown.

Comparison with other studies

We noted a higher odds of COVID-19 in patients with prior pancreatic and liver conditions. The higher odds associated with liver conditions is consistent with earlier findings.^6,19 Patients with moderate to severe liver conditions had higher susceptibility to SARS-CoV2 infection than those with milder conditions, which could be due to the increase in abnormalities of immune function with severity of this disease group³². We speculate that reduced pancreatic function, particularly in individuals with chronic pancreatic conditions, leading to altered digestion, and therefore gut flora, may make patients more susceptible to pathogens with an enteric route of SARS-CoV2 infection^33,34, and also contribute to the magnitude of COVID-19 severity via modulating host immune responses³⁵. Surprisingly the most vulnerable patients with cancer had a low COVID-19 incidence rate, which may reflect the effectiveness of public health interventions such as shielding.³⁶ However, at the same time, we noted a 17% death rate in this cohort (not due to COVID-19) in six months (supplemental table 3), perhaps indicating the unintended, but potentially inevitable, negative sequelae of social distancing and reduced healthcare provisions for this group of patients as resources were diverted to COVID-19 affected patients.

Men had a higher odds of infection and mortality than women, which is consistent with previous reports,^1,14 and could be due to a favourable genetic predisposition to the virus,³⁷ and/or gender differences in risk behaviours. Our study also affirms older age, particularly over 70, as an established risk factor for COVID-19 incidence and mortality;^2,4,5 however, this can be largely explained by the presence of multiple comorbidities in the older age groups.³⁸

COVID-19 statistics have highlighted a disproportionate effect on BAME ethnic groups with an increased risk of infection and poor outcomes.^13–15 Our results confirm that people from Black community are at a higher risk of both COVID-19 infection and related mortality compared to the White ethnic group. Only a small part of the excess risk in the Black community is explained by multiple comorbidities. Therefore, further variables such as deprivation, occupational exposure, and living conditions might be useful to explore as potential factors behind the apparent vulnerability of the Black population to COVID-19.

All comorbidities such as diabetes, hypertension, high cholesterol, cardiovascular disease, kidney, and respiratory disease, were independently associated with an increased risk of COVID-19, whereas the presence of cardiovascular disease contributed to an added risk of death, concurring with previously reported cohort studies.^4–6–11 Interestingly, our results highlight that for patients without underlying respiratory issues, an acute respiratory episode due to SARS-CoV-2 infection could be indicative of a worse outcome. This is in line with previous reports describing an unexpectedly lower prevalence of chronic respiratory conditions among those who had been admitted to hospital due to COVID-19;^39,40 whereas severe outcomes are often a result of respiratory complications, ^41,42 such as acute respiratory distress syndrome (ARDS) and respiratory failure.

Smoking leads to severe health consequences, which explains the greater risk observed in our cohort of past smokers with high prevalence of respiratory and cardiovascular diseases. However, current smoking status appeared to have a protective effect in our cohort after adjusting for comorbidities, as has been observed by others, an aspect which cannot be mechanistically explained.^5,43 Carefully designed analyses are needed to explore the association and causality between smoking status (both current and past), associated comorbidities and COVID-19.

Although substance mis-use leads to a plethora of cardio-respiratory and metabolic problems, its role in COVID-19 remains unexplored. To date, this is the first study providing a concrete measure of the risk of COVID-19 for substance mis-users. Our initial results showing that substance mis-users are at a heightened risk for COVID-19 irrespective of the comorbidities warrants a strong case for considering it as an independent risk factor for COVID-19, and may be related to high-risk behavioural patterns. ^44,45

Previous studies have found a significant relationship between obesity and an increased risk of COVID-19,⁷ and subsequent hospitalisation,⁴⁶ advanced levels of treatment,¹⁵ and death.^4,6 However, our study does not suggest any particular effect of obesity on COVID-19 for patients with HPB conditions, who have a higher prevalence rate of obesity compared to the UK general population (38.8% vs 26%).⁴⁷ Our study suggest that the difference in effects for potential susceptibility to COVID-19 for patients with history of obesity are attributed more to other prevalent factors – such as cardiovascular or chronic renal disease ^47,48 – which in turn might be the consequences of obesity in these patients’ lifetime.

Concerns have been raised regarding the use of various medications with respect to the risk of COVID-19 and the subsequent outcome; and, our analyses contribute to that discussion for some of the widely used prescription drugs. The higher odds observed for the history of various prescription drugs use are consistent with the management of underlying prevalent comorbidities of the study cohort: cardiovascular conditions (ACE inhibitor, β-blocker, aldosterone antagonists, antiplatelet, antiarrhythmic), cholesterol (statin), chronic respiratory diseases (glucocorticoid, β2-agonists, muscarinic antagonist), or background HPB condition (proton pump inhibitor). An important finding from our study is the significant risk observed for vitamin D users, supportive of the possible association between development of COVID-19 and vitamin D deficiency, ^49,50 or specific medical conditions (such as kidney failure) where Vitamin D prescription is prevalent. Given that BAME communities are observed to be at a high risk of COVID-19, and there is evidence that vitamin D deficiency is particularly common in these ethnic groups,^49,50 further research on the relationship between vitamin D and COVID-19 is required, with a need to exclude confounding factors such as patients’ vitamin D level. Our result also suggest that patients currently taking PPIs are more susceptible to SARS-CoV-2 infection, which concurs with a large population-based online survey conducted in the US.⁵¹ The use of PPIs is highly prevalent in HPB patients for the management of gastrointestinal acid-related disorders, and the finding here supports the hypothesis that current use of PPIs might influence the susceptibility to SARS-CoV-2 infection in the gastrointestinal tract through reduction of stomach acid.^51,52

The literature is conflicted on the potential impact of antihypertensive drugs on COVID-19, particularly those that act as inhibitors to the renin–angiotensin– aldosterone system (RAAS) and upregulate ACE2 expression, suggesting these drugs may be potential risk factors for infection,^53,54 but also as having a protective effect on outcome.⁵⁵ However, recent studies found no underlying association between the use of different classes of antihypertensive drugs and the risk of developing COVID-19.¹⁶ With a high percentage of patients with hypertension in the study cohort, our finding that a high risk of COVID-19 is associated with past intake of ACE inhibitors or aldosterone agonists is suggestive of the potential risk of switching from one class of antihypertensive drug to another. This contributes to the debate of whether discontinuation of RAAS inhibitors and considering alternative antihypertensive therapy in times of COVID-19 would be a good practice or not.⁵⁶ A marginal association of current use of ACE inhibitors with COVID-19 related death suggests that any increased risk of mortality is likely to be small and will need to be scrutinised in future as more data accumulates.

Our study also shows that recent users of anti-inflammatory drugs, namely glucocorticoid and β2-agonists, had increased odds for COVID-19 and subsequent poor outcome. Controlling for comorbidities resulted in non-significant odds of infection for these patient subgroups, indicating underlying medical conditions - particularly those of respiratory system - to be responsible for the increased susceptibility. However, the observed harmful associations between these drugs and COVID-19-related death could not be explained by a simplified binary representation of underlying six common health conditions. Glucocorticoid drugs, for instance, are used to treat many other inflammatory conditions, notably inflammatory bowel disease (IBD), whereas HPB diseases constitute some of the most common extraintestinal manifestations of IBD. It has been shown that use of corticosteroids is associated with adverse COVID-19 outcomes among patients with IBD.¹⁷ Had it been possible to successfully control for differences in respiratory disease severity or other medical comorbidities, we speculate to see different and possibly non-significant odds of death in these patient subgroups.

Strengths and limitations of the study

A key strength of our study is that we have systematically identified the effect, or the lack of it, of individual demographic and clinical factors on the infection and mortality of COVID-19 in a cohort of over 15000 patients, robustly corrected for potential confounders in their evaluation. Our large population is highly representative of HPB patients from diverse ethnic groups, which contributes to the generalisability of our findings. Another strength is our use of linked electronic health records, harmonised for variations in coding that exist between different EHR systems. We ascertained patient demographics, lifestyle, comorbidities and medications by linking hospital records with pseudo-anonymized longitudinal primary care records, which substantially enrich the data that are recorded on hospital visits.

Retrospective EHR-based COVID-19 studies often suffer from incomplete or missing data on patient characteristics, including key variables such as BMI, ethnicity, smoking or pre-existing comorbidities.^4,58 The missing data is particularly applicable to otherwise healthy COVID-19 patients with low use of healthcare services in the past. However, our patient cohort had already been treated or managed at BHNT hospitals at least once, and often referred through primary care, which led to near-complete data for this study, an added advantage of this study. For instance, ethnicity, a common demographic feature, is missing only for 2.8% of cases in our cohort while the rate is significantly higher in other studies (up to 20% of cases).^4,57 The only variable with missing data frequency over 20% in our study is substance mis-use behaviour (50.4%). This is a unique lifestyle risk variable which is not yet explored - understandably due to a lack of recorded data as people often do not disclose this information to their clinicians,⁵⁸ unless manifested in physical or mental disorders. Yet, the substance mis-use history of over 7600 patients included in this study provide a good indication of the impact of COVID-19 on this under-studied group.

Our study also has some important limitations. One limitation is the risk of residual confounding or confounding by indication due to unmeasured or simplified binary representation of potential confounding variables. For example, the observed association between Vitamin D users and risk of COVID-19 may be different if participants’ Vitamin D level/deficiency status had been taken into consideration. Similarly, the observed association between COVID-19 related death and recent use of glucocorticoid or β2-agonists may reduce or get amplified if respiratory disease severity or other indications for corticosteroid use were considered.

Another critical limitation is associated with the confirmation of East London residency for the study cohort. Patients’ addresses (current or historic) are not collected under the umbrella study, which considers patients with HPB conditions (with the exception of cancer) treated or managed at BHNT hospitals as East London residents during the time of their care. The Royal London Hospital hosts one of the largest HPB centres in England, and supports suspected or confirmed HPB cancer patients from nearby geographical areas. As the umbrella study cohort is historic, we acknowledged the probability of people moving away from East London in the meantime. In absence of a patient’s current address to confirm their residency at the outset of COVID-19 pandemic in the UK, we relied on an indirect measure to infer residency. We used a strict six-month window preceding the study to identify a patient’s interaction with East London GPs or BHNT hospitals. Thus, we believe that any supposed reduction in the cohort size due to unaccounted change of residency within that window should have affected the COVID and non-COVID group in equal proportion, and hence unlikely to alter the findings we report here.

Due to the rarity of the outcome (SARS-CoV-2 infection) in the full HPB cohort, the effects reported in the study could be influenced by the smaller cohort size of COVID-19 cases. We recognise that larger sample sizes of COVID-19 patients are needed to fully understand the effect of SARS-CoV-2 in patients with HPB conditions. Our results are the first step towards this and require validation in similar national and international cohorts.