Received: 12 July 2021 Revised: 31 August 2021 Accepted: 20 September 2021 DOI: 10.1002/ijc.33846 CANCER EPIDEMIOLOGY Cancer and the risk of coronavirus disease 2019 diagnosis, hospitalisation and death: A population-based multistate cohort study including 4 618 377 adults in Catalonia, Spain Elena Roel1,2 | Andrea Pistillo1 | Martina Recalde1,2 | Sergio Fernández-Bertolín 1  n | Isabelle Soerjomataram3 | María Arago 1 | Mazda Jenab3 | Diana Puente1 | Daniel Prieto-Alhambra4 | Edward Burn1,4 | Talita Duarte-Salles1 1  Institut Universitari per a la recerca a l'Atencio Fundacio  Primària de Salut Jordi Gol i Gurina (IDIAPJGol), Barcelona, Spain 2 Universitat Autònoma de Barcelona, Barcelona, Spain 3 International Agency for Research on Cancer (IARC-WHO), Lyon, France 4 Centre for Statistics in Medicine, Nuffield Department of Orthopaedics, Rheumatology, and Musculoskeletal Sciences, University of Oxford, Botnar Research Centre, Oxford, UK Correspondence Talita Duarte-Salles, Fundacio  Institut Abstract Universitari per a la recerca a l'Atencio The relationship between cancer and coronavirus disease 2019 (COVID-19) infection Primària de Salut Jordi Gol i Gurina (IDIAPJGol), Gran Via Corts Catalanes, and severity remains poorly understood. We conducted a population-based cohort 587 àtic, 08007 Barcelona, Spain. study between 1 March and 6 May 2020 describing the associations between cancer Email: tduarte@idiapjgol.org and risk of COVID-19 diagnosis, hospitalisation and COVID-19-related death. Data Funding information were obtained from the Information System for Research in Primary Care (SIDIAP) This project was funded by the Health Department from the Generalitat de Catalunya database, including primary care electronic health records from 80% of the popula- with a grant for research projects on SARS- tion in Catalonia, Spain. Cancer was defined as any primary invasive malignancy CoV-2 and COVID-19 disease organised by the Direccio  General de Recerca i Innovacio  en excluding non-melanoma skin cancer. We estimated adjusted hazard ratios (aHRs) for Salut. This project has also received support the risk of COVID-19 (outpatient) clinical diagnosis, hospitalisation (with or without a from the European Health Data and Evidence Network (EHDEN) project. EHDEN received prior COVID-19 diagnosis) and COVID-19-related death using Cox proportional haz- funding from the Innovative Medicines ard regressions. Models were estimated for the overall cancer population and by Initiative 2 Joint Undertaking (JU) under grant agreement No 806968. The JU receives years since cancer diagnosis (<1 year, 1-5 years and ≥5 years), sex, age and cancer support from the European Union's Horizon type; and adjusted for age, sex, smoking status, deprivation and comorbidities. We 2020 research and innovation programme and EFPIA. The University of Oxford received a included 4 618 377 adults, of which 260 667 (5.6%) had a history of cancer. A total grant related to this work from the Bill & of 98 951 individuals (5.5% with cancer) were diagnosed, and 6355 (16.4% with can- Melinda Gates Foundation (Investment ID INV-016201), and partial support from the UK cer) were directly hospitalised with COVID-19. Of those diagnosed, 6851 were sub- National Institute for Health Research (NIHR) sequently hospitalised (10.7% with cancer), and 3227 died without being hospitalised Oxford Biomedical Research Centre. Elena Roel was supported by Instituto de Salud (18.5% with cancer). Among those hospitalised, 1963 (22.5% with cancer) died. Carlos III (grant number CM20/00174). Daniel Cancer was associated with an increased risk of COVID-19 diagnosis (aHR: 1.08; Abbreviations: 95% CI, 95% confidence interval; aHR, adjusted hazard ratio; BIC, Bayesian information criterion; CDM, Common Data Model; CI, cumulative incidence; COVID-19, coronavirus disease 2019; GP, general practitioner; ICD-10-CM, International Classification of Diseases, Tenth Revision, Clinical Modification; OHDSI, Observational Health Data Sciences and Informatics; OMOP, Observational Medical Outcomes Partnership; RT-PCR, reverse transcription polymerase chain reaction; SARS-CoV-2, severe acute respiratory syndrome coronavirus 2; SIDIAP, Information System for Research in Primary Care; SMD, standardised mean difference. Where authors are identified as personnel of the International Agency for Research on Cancer and World Health Organisation, the authors alone are responsible for the views expressed in our study, and they do not necessarily represent the decisions, policy or views of the International Agency for Research on Cancer and World Health Organisation. Int. J. Cancer. 2021;1–13. wileyonlinelibrary.com/journal/ijc © 2021 UICC. 1 2 ROEL ET AL. Prieto-Alhambra is funded through a National Institute for Health Research (NIHR) Senior 95% confidence interval [1.05-1.11]), direct COVID-19 hospitalisation (1.33 Research Fellowship (Grant number SRF- [1.24-1.43]) and death following hospitalisation (1.12 [1.01-1.25]). These associations 2018-11-ST2-004). The views expressed in this publication are those of the authors and were stronger for patients recently diagnosed with cancer, aged <70 years, and with not necessarily those of the NHS, the National haematological cancers. These patients should be prioritised in COVID-19 vaccina- Institute for Health Research or the Department of Health. The funders of the tion campaigns and continued non-pharmaceutical interventions. study had no role in study design, data collection, analysis and interpretation, or KEYWORDS writing of the report. cancer, COVID-19, electronic health record, fatality, SARS-CoV-2 What's new? Studies addressing associations between cancer and severity of coronavirus disease 2019 (COVID-19) have focused primarily on hospitalized patients. Findings have been inconsistent, however, owing to varying cancer criteria, lack of representative samples, and other factors. Here, the natural history of COVID-19 in cancer patients during the first wave of the pandemic in 2020 in Spain was investigated in a large, representative cohort with a heterogenous cancer population. Patients with cancer were at increased risk of severe COVID-19. Risk was notably high among those over age 70 and those with recent cancer diagnosis, hematological cancer, or lung and bladder cancer. 1 | I N T RO DU CT I O N COVID-19 incidence and COVID-19-related outcomes could address the limitations of the previous evidence. In our study, we aimed to Cancer is a leading cause of morbidity and death worldwide, with an describe the associations between cancer and the risks of COVID-19 estimated 19 million new cases and 10 million deaths in 2020.1 diagnosis, hospitalisation with COVID-19 and COVID-19-related Patients with cancer are often older and have multiple comorbidities death, overall and by different population subgroups, using real-world and an impaired immunity due to the cancer itself and cancer thera- data from Catalonia, Spain. pies, thus increasing their susceptibility to infections.2 As a result, patients with cancer have been considered a high-risk population for the novel coronavirus disease 2019 (COVID-19) since the beginning 2 | M A T E R I A L S A N D M ET H O D S 3 of the pandemic. This disease, caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), manifests with a varying 2.1 | Study design, setting and data sources degree of severity, ranging from asymptomatic to severe disease and death.4 We conducted a population-based cohort study from 1 March 2020 Although there are a substantial number of publications until 6 May 2020 (last date of data available), using data from the addressing the relationship between cancer and COVID-19, these Information System for Research in Primary Care (SIDIAP; www. have shown conflicting results.5 Some studies have found that sidiap.org), a primary care database from Catalonia, a north-eastern patients with cancer have an increased risk of COVID-19 infection, region in Spain. Spain has a universal primary care-based health sys- 6-9 hospitalisation and death compared to patients without cancer, tem, in which general practitioners (GPs) are the first point of contact whereas others have reported null associations.10-12 The majority of for care. As a consequence, GPs have diagnosed and managed the these studies were small, used different criteria to identify patients majority of COVID-19 cases since the beginning of the pandemic.14 In with cancer (eg, only active cancers, or solid cancers) and did not addition, because GPs are responsible of issuing sick leaves, patients include representative samples (ie, restricted to hospital and/or diagnosed with COVID-19 in other settings (eg, hospital emergency laboratory-confirmed cases), which limits the generalizability of their departments) were also bound to contact primary care providers dur- findings and increases the risk of selection bias.13 ing study follow-up. Patients with cancer are a highly heterogeneous population that The SIDIAP database includes anonymized primary care elec- encompasses patients with different features, such as cancer type or tronic health records collected since 2006 covering approximately six phases of care since time of diagnosis (eg, under active treatment, million people (80% of the population in Catalonia, Spain) and is repre- active surveillance or cured). Understanding which patients with can- sentative in terms of age, sex and geographic distribution.15 SIDIAP cer are at the highest risk of COVID-19-infection of poor outcomes is includes data on demographics, lifestyle information and disease diag- essential to inform clinical care and to guide prevention strategies noses, among others and has been linked to SARS-CoV-2 reverse targeting this population. A large, population-based cohort study that transcription polymerase chain reaction (RT-PCR) test results and hos- includes a heterogeneous cancer population and that captures both pital records (both from the public sector), as well as to regional ROEL ET AL. 3 mortality data through unique ID linkage. In addition, SIDIAP has been this approach to provide a more comprehensive overview of patient's mapped to the Observational Medical Outcomes Partnership (OMOP) interactions with the health system, taking into account those who Common Data Model (CDM), allowing us to apply common analytical seek primary and hospital care. tools developed by the open-science Observational Health Data Sci- For all the transitions, individuals were followed until the occur- ences and Informatics (OHDSI) network.16 rence of a state of interest, the occurrence of a competing event or the end of the study period (6 May 2020). Because we were solely interested in COVID-19-related outcomes, we did not model the tran- 2.2 | Study participants sition from the general population to death. However, we reported deaths occurring in the general population, which were considered as We included all adults (aged 18 years or older) registered in the a competing event. SIDIAP database as of 1 March 2020 (index date for all participants) with at least 1 year of prior history observation available. We excluded patients who had a record of a secondary cancer before a 2.4 | Variables record of a primary cancer, patients with a clinical diagnosis or posi- tive test result for COVID-19 prior to index date and patients hos- The exposure of interest was cancer, which we defined as any diagno- pitalised or living in a nursing home at index date (to include only sis of a primary invasive solid or haematological cancer, excluding patients representative of the community population). non-melanoma skin cancer, prior to the index date. We used the Inter- national Classification of Diseases, Tenth Revision, Clinical Modifica- tion (ICD-10-CM) to identify cancer diagnoses: C00 to C96, except 2.3 | Multistate framework C44 (non-melanoma skin cancer) and C77-C79 (secondary cancers). Cancer types by anatomical location were identified using definitions To address our objectives, we employed a multistate framework that previously validated in the SIDIAP database.19 To avoid mis- we have previously utilised to describe the risks of COVID-19 diagno- classification of primary cancers, we only considered the earliest can- sis, hospitalisation and death.17 Multistate models can be used to cer type registered for each patient. We stratified patients with describe processes where individuals transition from one health status cancer according to the number of years since the diagnosis to the to another, while separating baseline risk and covariate effects associ- index date into three groups (<1 year, 1-5 years and ≥5 years), 18 ated with each transition. In our study, individuals started the because we lacked information on cancer status (ie, active, in remis- follow-up at the general population and then could transition to three sion) and cancer therapies. By doing this, we assumed that those diag- other states: diagnosed with COVID-19 (in an outpatient setting), nosed with cancer <5 years prior to the index date were more likely hospitalised with COVID-19 and death. Six different transitions were to have an active cancer and/or an ongoing cancer treatment (espe- possible: from the general population to either diagnosed with cially those diagnosed within 1 year prior), whereas those diagnosed COVID-19, hospitalised with COVID-19 (ie, direct hospitalisation) or ≥5 years prior would be mostly cancer survivors. death; from diagnosed to either hospitalised with COVID-19 or death The covariates of interest were sex, age, smoking status, depriva- and from hospitalised with COVID-19 to death (Figure 1). We used tion and comorbidities. We extracted participants' sex and age at n = 11 326 29.6% with cancer n = 6355 n = 3227 16.4% with cancer 18.5% with cancer n = 6851 n = 1963 Diagnosed Hospitalised General 10.7% with cancer 22.5% with cancer with with Death population COVID-19 COVID-19 n = 4 618 377 n = 98 951 n = 13 206 n = 16 516 5.6% with cancer 5.5% with cancer 13.5% with cancer 26.6% with cancer FIGURE 1 Overview of the multistate model used in our study 4 ROEL ET AL. index date. Smoking status (never, former or current smoker) was polynomial of degree 2 (ie, quadratic), and with restricted cubic splines assigned as the closest assessment to the index date recorded. Depri- (with three, four or five knots).23 We calculated the Bayesian informa- vation was assessed using the MEDEA deprivation index, which is cal- tion criterion (BIC) for each of those models, and we selected the culated at the census tract level in urban areas of Catalonia.20 MEDEA models with the lowest BIC values. deprivation index is categorised in quintiles, with the first quintile rep- Second, we estimated the relationship between cancer and resenting the least deprived group and the fifth the most deprived. It COVID-19 outcomes adjusting for age and sex; and adjusting for age, also includes a rural category for individuals living in rural areas. Our sex, the MEDEA deprivation index and smoking status. Third, we fur- comorbidities of interest were autoimmune conditions, chronic kidney ther estimated our main models separately for <1-year, 1 to 5-year disease, chronic obstructive pulmonary disease, dementia, heart dis- and ≥5-year cancer patients and stratified these models by sex ease, hyperlipidaemia, hypertension, obesity and type-2 diabetes. (women or men), age (<70 and ≥70 years, 70 years was the median Comorbidities were defined as previously described based on medical age of patients with cancer) and cancer type (haematological or solid diagnosis17 and selected due to their relevance to the COVID-19 cancer, as well as by solid cancer types). All models were relative to research field.21 The definitions for each comorbidity can be consul- patients without cancer (cancer-free). ted in a web application (“Index Event Breakdown” tab) available at As sensitivity analyses, we re-estimated our main models: https://livedataoxford.shinyapps.io/MultiStateCovidCohorts/. (a) stratifying by calendar time for transitions in which the proportion- Our outcomes of interest were an outpatient clinical diagnosis of ality assumption was violated, (b) restricting participants to never COVID-19, a hospitalisation with COVID-19 and COVID-19-related smokers, to avoid residual confounding by smoking and (c) after per- death. We defined COVID-19 diagnoses based on a recorded clinical forming a multiple imputation of missing data (smoking status and code for COVID-19 disease (ICD-10-CM: B34.2; B97.29). We did not MEDEA deprivation index) using predictive mean matching, with five require a positive RT-PCR test result in the definition of COVID-19 imputations drawn. We also compared baseline characteristics of diagnoses due to testing restrictions during the first months of the patients with and without missing data using standardised mean dif- pandemic.17 For instance, at that time, tests were exclusively available ferences (SMD). We considered SMD ≥j0.1j as a meaningful differ- at the hospital level, and only patients with severe symptoms and/or ence in the distribution of a given characteristic between the two with underlying conditions were tested. We defined hospitalisation groups.24 with COVID-19 as a hospital admission (with at least one-day hospital We used R version 3.6 for data analysis and visualisation. The R stay) where the patient had a COVID-19 diagnosis or a positive RT- packages used in the analysis included mstate25 and rms.26 The ana- PCR test result 21 days prior to admission up to 3 days after admis- lytic code is available at https://github.com/SIDIAP/COVID-19- sion (to allow for a delay in diagnosis and minimise the risk of includ- cancer-multi-state. ing hospital-acquired COVID-19 infections). We extracted deaths (from any cause) from region-wide mortality data, and by doing so, we included both deaths during hospitalisation and in the community. 3 | RE SU LT S Deaths occurring following a COVID-19 event (diagnosis or hospitalisation) were considered as COVID-19-related deaths. 3.1 | Population included A total of 4 618 377 adults were included. We excluded 104 022 indi- 2.5 | Statistical analyses viduals with less than a year of prior observation history; 1496 with a record of a secondary cancer before a record of a primary cancer, We described participants' baseline characteristics, participants' time 303 with a COVID-19 diagnosis or positive SARS-CoV-2 test before at risk at each state and numbers of events observed for each transi- index date, 40 421 living in a nursing home and 1138 hospitalised at tion by cancer status (with or without cancer). To assess the relation- the index date (Figure S2). Baseline characteristics of the population ship between the cancer and the risk of transitioning to a subsequent included are summarised in Table 1. In total, 260 667 (5.6%) patients state in the multistate model, we estimated adjusted cause-specific had a prior diagnosis of cancer. Of these, 167 053 (64.1% of the can- hazard ratios (aHRs), with 95% confidence intervals (CIs), using Cox cer population) were diagnosed ≥5 years, 72 033 (27.6%) 1 to 5 years proportional hazard regressions for each transition. and 21 581 (8.3%) <1 year prior to the index date. Compared to First, we estimated models for all patients with cancer compared cancer-free patients, those with cancer were older, more frequently to patients without cancer adjusting for age, sex, the MEDEA depriva- former smokers and living in the least deprived areas of Catalonia. In tion index, smoking status and all the comorbidities of interest (main addition, they had a higher burden of comorbidities, especially cardio- models). We used a directed acyclic graph to guide decisions on the vascular conditions (eg, 27.4% had heart disease vs 10.2% in cancer- control for confounding (Figure S1).22 To check the proportional haz- free patients). When stratifying patients by age categories, we ard assumptions for the variables included in the models, we visually observed that the burden of comorbidities increased with age for both inspected log-log survival curves. Missing data were handled as an groups (Figure S3). Among patients with cancer, 239 030 (91.7%) and additional category. Non-linearity in age and risks of transition was 21 637 (8.3%) had a solid and haematological cancer, respectively. considered by fitting models with age as a linear term, with a The most frequent solid cancer types were breast (n = 58 611, TABLE 1 Baseline characteristics of the population included, by cancer status With cancer ROEL ET AL. Total population Without cancer Overall ≥5 yearsa 1-5 yearsa <1 yeara n 4 618 377 4 357 710 260 667 167 053 72 033 21 581 Age (median [IQR]) 48 [36.0, 63.0] 47 [35.0, 61.0] 70 [59.0, 78.0] 71 [61.0, 79.0] 67 [57.0, 76.0] 66 [56.0, 76.0] Age categories (%) 18-39 1 437 236 (31.1) 1 427 705 (32.8) 9531 (3.7) 5555 (3.3) 2974 (4.1) 1002 (4.6) 40-59 1 785 495 (38.7) 1 727 443 (39.6) 58 052 (22.3) 32 909 (19.7) 19 019 (26.4) 6124 (28.4) 60-69 615 198 (13.3) 553 838 (12.7) 61 360 (23.5) 36 999 (22.1) 18 786 (26.1) 5575 (25.8) 70-79 468 286 (10.1) 393 504 (9.0) 74 782 (28.7) 50 205 (30.1) 19 197 (26.7) 5380 (24.9) 80 or older 312 162 (6.8) 255 220 (5.9) 56 942 (21.8) 41 385 (24.8) 12 057 (16.7) 3500 (16.2) Sex, female (%) 2 361 230 (51.1) 2 226 424 (51.1) 134 806 (51.7) 89 473 (53.6) 35 060 (48.7) 10 273 (47.6) MEDEA deprivation index (%) Quintile 1 (least deprived) 714 183 (15.5) 668 548 (15.3) 45 635 (17.5) 29 662 (17.8) 12 392 (17.2) 3581 (16.6) Quintile 2 703 921 (15.2) 662 113 (15.2) 41 808 (16.0) 26 971 (16.1) 11 534 (16.0) 3303 (15.3) Quintile 3 697 074 (15.1) 656 859 (15.1) 40 215 (15.4) 25 893 (15.5) 11 114 (15.4) 3208 (14.9) Quintile 4 692 844 (15.0) 654 775 (15.0) 38 069 (14.6) 24 488 (14.7) 10 445 (14.5) 3136 (14.5) Quintile 5 (most deprived) 687 062 (14.9) 653 878 (15.0) 33 184 (12.7) 21 149 (12.7) 9168 (12.7) 2867 (13.3) Rural 832 256 (18.0) 785 356 (18.0) 46 900 (18.0) 29 744 (17.8) 13 073 (18.1) 4083 (18.9) Missing 291 037 (6.3) 276 181 (6.3) 14 856 (5.7) 9146 (5.5) 4307 (6.0) 1403 (6.5) Smoking status (%) Never smoker 1 834 657 (39.7) 1 736 604 (39.9) 98 053 (37.6) 64 646 (38.7) 25 891 (35.9) 7516 (34.8) Former smoker 772 875 (16.7) 695 636 (16.0) 77 239 (29.6) 48 635 (29.1) 22 576 (31.3) 6028 (27.9) Current smoker 712 739 (15.4) 686 159 (15.7) 26 580 (10.2) 15 702 (9.4) 7901 (11.0) 2977 (13.8) Missing 1 298 106 (28.1) 1 239 311 (28.4) 58 795 (22.6) 38 070 (22.8) 15 665 (21.7) 5060 (23.4) Comorbidities (%) Autoimmune condition 259 234 (5.6) 235 347 (5.4) 23 887 (9.2) 15 474 (9.3) 6526 (9.1) 1887 (8.7) Chronic kidney disease 201 258 (4.4) 165 751 (3.8) 35 507 (13.6) 24 922 (14.9) 8339 (11.6) 2246 (10.4) Chronic obstructive pulmonary disease 119 532 (2.6) 98 365 (2.3) 21 167 (8.1) 13 281 (8.0) 6001 (8.3) 1885 (8.7) Dementia 42 504 (0.9) 36 026 (0.8) 6478 (2.5) 4817 (2.9) 1328 (1.8) 333 (1.5) Heart disease 516 140 (11.2) 444 733 (10.2) 71 407 (27.4) 47 851 (28.6) 18 145 (25.2) 5411 (25.1) Hyperlipidaemia 505 102 (10.9) 458 565 (10.5) 46 537 (17.9) 30 173 (18.1) 12 785 (17.7) 3579 (16.6) Hypertension 687 358 (14.9) 610 694 (14.0) 76 664 (29.4) 49 254 (29.5) 21 195 (29.4) 6215 (28.8) Obesity 1 144 442 (24.8) 1 045 689 (24.0) 98 753 (37.9) 64 148 (38.4) 26 800 (37.2) 7805 (36.2) Type-2 diabetes 317 005 (6.9) 275 132 (6.3) 41 873 (16.1) 26 913 (16.1) 11 560 (16.0) 3400 (15.8) (Continues) 5 6 TABLE 1 (Continued) With cancer Total population Without cancer Overall ≥5 yearsa 1-5 yearsa <1 yeara Age at cancer diagnosis, median [IQR] — — 61 [50.3, 70.2] 59 [48.1, 67.9] 65 [54.3, 73.6] 66 [55.5, 75.5] Cancer type [ICD-10-CM code] (%) Haematological 21 637 (0.5) 21 637 (8.3) 13 657 (8.2) 6148 (8.5) 1832 (8.5) Leukaemia [C91-C95] 7402 (0.2) — 7402 (2.8) 4744 (2.8) 2051 (2.8) 607 (2.8) Non-Hodgkin lymphoma [C82-C96] 5111 (0.1) — 5111 (2.0) 3776 (2.3) 1031 (1.4) 304 (1.4) Hodgkin's lymphoma [C81] 2724 (0.1) — 2724 (1.0) 2133 (1.3) 466 (0.6) 125 (0.6) Multiple myeloma [C90] 2249 (0.0) — 2249 (0.9) 1031 (0.6) 916 (1.3) 302 (1.4) Other haematological [C96] 4151 (0.1) — 4151 (1.6) 1973 (1.2) 1684 (2.3) 494 (2.3) Solid 239 030 (5.2) — 239 030 (91.7) 153 396 (91.8) 65 885 (91.5) 19 749 (91.5) Breast [C50] 58 611 (1.3) — 58 611 (22.5) 40 074 (24.0) 14 725 (20.4) 3812 (17.7) Prostate [C61] 37 141 (0.8) — 37 141 (14.2) 24 400 (14.6) 10 165 (14.1) 2576 (11.9) Colorectal [C18-C21] 36 071 (0.8) — 36 071 (13.8) 21 669 (13.0) 11 415 (15.8) 2987 (13.8) Bladder [C67] 20 592 (0.4) — 20 592 (7.9) 12 509 (7.5) 6293 (8.7) 1790 (8.3) Skin melanoma [C43] 12 956 (0.3) — 12 956 (5.0) 8490 (5.1) 3422 (4.8) 1044 (4.8) Kidney [C64] 7911 (0.2) — 7911 (3.0) 4522 (2.7) 2630 (3.7) 759 (3.5) Lung [C33-C34] 7569 (0.2) — 7569 (2.9) 3080 (1.8) 2948 (4.1) 1541 (7.1) Corpus uterus [C54-C55] 7353 (0.2) — 7353 (2.8) 4983 (3.0) 1855 (2.6) 515 (2.4) Thyroid [C73] 6449 (0.1) — 6449 (2.5) 4579 (2.7) 1500 (2.1) 370 (1.7) Head and neck [C00-C14] 5770 (0.1) — 5770 (2.2) 4042 (2.4) 1323 (1.8) 405 (1.9) Cervix [C53] 3979 (0.1) — 3979 (1.5) 3035 (1.8) 755 (1.0) 189 (0.9) Ovary [C56] 3889 (0.1) — 3889 (1.5) 2523 (1.5) 997 (1.4) 369 (1.7) Stomach [C16] 3628 (0.1) — 3628 (1.4) 2210 (1.3) 995 (1.4) 423 (2.0) Larynx [C32] 3317 (0.1) — 3317 (1.3) 2161 (1.3) 874 (1.2) 282 (1.3) Brain and central nervous system 3313 (0.1) — 3313 (1.3) 2216 (1.3) 750 (1.0) 347 (1.6) [C70-C72, C75.1-C75.3] Testis [C62] 2763 (0.1) — 2763 (1.1) 2073 (1.2) 562 (0.8) 128 (0.6) Liver [C22] 2051 (0.0) — 2051 (0.8) 852 (0.5) 818 (1.1) 381 (1.8) Bone and cartilage [C40-C41] 1944 (0.0) — 1944 (0.7) 1458 (0.9) 371 (0.5) 115 (0.5) Pancreas [C25] 1622 (0.0) — 1622 (0.6) 568 (0.3) 592 (0.8) 462 (2.1) Oesophagus [C15] 763 (0.0) — 763 (0.3) 349 (0.2) 270 (0.4) 144 (0.7) Gallbladder [C23-C24] 479 (0.0) — 479 (0.2) 214 (0.1) 181 (0.3) 84 (0.4) Other solid 10 859 (0.2) — 10 859 (4.2) 7389 (4.4) 2444 (3.4) 1026 (4.8) Note: — means not applicable. The MEDEA deprivation index is calculated at the census tract level in urban areas. Other solid cancers include other solid cancers, cancers of unspecified site [C76, C80] and more than one cancer (ie, patients that had more than one cancer recorded on the same date). Abbreviations: ICD-10-CM, International Classification for Diseases, 10th revision Clinical Modification; IQR, interquartile range. a Years since cancer diagnosis to the index date (1 March 2020). ROEL ET AL. ROEL ET AL. 7 22.5%), prostate (37 141, 14.2%), colorectal (36 071, 13.8%) and blad- 1963 (17.57%) 1522 (15.71%) 441 (29.34%) 293 (30.55%) 110 (25.75%) 38 (32.64%) (CI at 45 days) der (20 592, 7.9%). Number of To death events 3.2 | Occurrence of COVID-19 outcomes From hospitalised with COVID-19 36 (0.5, 22 to 43, 58) 36 (0.5, 23 to 43, 58) 37 (0, 27 to 43, 65) 37 (0, 28 to 43, 65) 36 (1, 22 to 43, 58) 35 (1, 22 to 42, 58) (min, IQR, max) Among the general population, 98 951 (2.1% cumulative incidence Follow-up [CI] at 67 days) individuals were diagnosed with COVID-19, 6355 Median (days) (0.1% CI) were directly hospitalised with COVID-19 and 11 326 (0.25% CI) died without a COVID-19 diagnosis/hospitalisation 1778 1134 13 206 11 428 479 165 (Figure 1, Table 2). Among individuals diagnosed with COVID-19, n 6851 (7.2% CI at 45 days) were hospitalised and 3227 (3.9% CI) died 596 (13.39%) without a hospitalisation. Among those hospitalised, 1963 (18% CI at 379 (13.13%) 149 (12.09%) 68 (20.15%) (CI at 45 days) 3227 (3.91%) 2631 (3.37%) 45 days) died. Among the total cancer population (n = 260 667), Number of To death events 5393 (2.1% CI at 67 days) patients were diagnosed with COVID-19, 1043 (0.4%) were directly hospitalised with COVID-19 and 3356 (1.3%) died without a COVID-19 diagnosis/hospitalisation. Among with COVID-19 To hospitalised 735 (14.14%) 464 (13.91%) 211 (14.85%) 60 (13.57%) (CI at 45 days) 6851 (7.19%) 6116 (6.79%) those diagnosed with COVID-19, 735 (14.1% CI at 45 days) were sub- Number of sequently hospitalised and 596 (13.4%) died without a hospitalisation. events Among those hospitalised, 441 (29.3% CI at 45 days) died. Descriptive characteristics by state and transition are shown in Table S1. In brief, 36 (0, 20 to 44, 66) 36 (0, 21 to 44, 66) 30 (0, 13 to 42, 65) 30 (0, 13 to 42, 65) 30.5 (0, 13 to 43, 65) 24 (0, 10 to 40, 64) From diagnosed with COVID-19 individuals diagnosed/hospitalised with COVID-19, as well as having a Follow-up (days) (min, IQR, max) COVID-19-related death, were older, more frequently male and for- mer smokers, and had more comorbidities than the general Median population. 98 951 93 558 463 5393 3464 1466 Time at risk, absolute number of events and cumulative incidence, by cancer status 3.3 | Risks of COVID-19 diagnosis, hospitalisation n and death among patients with cancer 11 326 (0.25%) 7970 (0.18%) 3356 (1.29%) 1714 (1.03%) 911 (1.27%) 731 (3.39%) (CI at 67 days) Number of To death Compared to cancer-free patients, those with cancer had an increased events risk of COVID-19 diagnosis (overall aHR: 1.08; 95% CI [1.05-1.11]), direct COVID-19 hospitalisation (1.33 [1.24-1.43]) and death follow- with COVID-19 To hospitalised (CI at 67 days) ing a COVID-19 hospitalisation (1.12 [1.01-1.25]) (Figure 2). Models 6355 (0.14%) 5312 (0.12%) 1043 (0.40%) 670 (0.40%) 268 (0.37%) 105 (0.49%) Number of using different adjustment strategies showed similar results to our events main models (Figure S4). In models stratified by years since cancer diagnosis, the risk of with COVID-19 COVID-19 diagnosis was similar in <1-year, 1 to 5-year and ≥5-year 98 951 (2.14%) 93 558 (2.15%) 5393 (2.07%) 3464 (2.07%) 1466 (2.04%) 463 (2.15%) (CI at 67 days) To diagnosed Number of cancer patients (Figure 2). As for the risk of direct COVID-19 events hospitalisation, <1-year cancer patients had the highest risk (1.84 Abbreviations: CI, cumulative incidence; IRQ, interquartile range. Years since cancer diagnosis to the index date (1 March 2020). From general population [1.52-2.23]), followed by 1 to 5-year cancer patients (1.32 [1.17-1.50]) and ≥5-year cancer patients (1.27 [1.17-1.38]). Increased 67 (1, 67 to 67, 67) 67 (1, 67 to 67, 67) 67 (1, 67 to 67, 67) 67 (2, 67 to 67, 67) 67 (1, 67 to 67, 67) 67 (2, 67 to 67, 67) (min, IQR, max) risk of COVID-19-related death remained significant only in <1-year Follow-up cancer patients, for both deaths following a COVID-19 diagnosis Median (days) (1.81[1.42-2.31]) and following a COVID-19 hospitalisation (1.63 [1.18-2.26]). population Overall, in models stratified by sex, the associations between can- 4 618 377 4 357 710 260 667 167 053 72 033 21 581 General cer and risk of COVID-19 diagnosis and death (following a diagnosis/ Patients with cancer n hospitalisation) were moderately stronger in men, whereas the associ- ations with risk of direct hospitalisation were moderately stronger in population 1-5 yearsa TABLE 2 a without <1 yeara ≥5 years women (Figure 3, Table S2). In models stratified by age, we found a Overall cancer Patients Total stronger association between cancer and COVID-19 outcomes in the subgroup of patients aged <70 years compared to those aged a 8 ROEL ET AL. F I G U R E 2 Adjusted hazard ratios of COVID-19 outcomes in patients with cancer compared to patients without cancer, overall and by years since cancer diagnosis. Models are adjusted for age, sex, the MEDEA deprivation index, smoking status and comorbidities (autoimmune conditions, chronic kidney disease, chronic obstructive pulmonary disease, dementia, heart disease, hyperlipidaemia, hypertension, type-2 diabetes and obesity). aHR, adjusted hazard ratio; CI, confidence interval ≥70 years, aside from the risk of COVID-19 diagnosis (Figure 3, When stratifying patients by haematological or solid cancers, Table S3). Age differences were more pronounced in <1-year cancer those with haematological cancers had a higher risk of COVID-19 out- patients. In addition, the associations between cancer and COVID- comes (Figure 3, Table S4). These differences were more pronounced 19-related death (either following a COVID-19 diagnosis or a in <1-year cancer patients. For example, the overall aHR for having a hospitalisation) were only significant in the subgroup of patients aged direct COVID-19 hospitalisation was 2.51 (2.12-2.98) for patients <70 years. For example, the overall aHR for death following hospitalisation with haematological cancers and 1.24 (1.15-1.33) for those with solid was 1.49 (1.10-2.01) in <70-year patients and 1.07 (0.95-1.20) in ≥70-year cancers. Among <1-year cancer patients, aHRs were 6.18 (4.31-8.86) patients. In <1-year cancer patients, the aHR was 4.58 (2.47-8.50) in for haematological cancers and 1.49 (1.19-1.87) for solid cancers. <70-year patients and 1.30 (0.88-1.90) in ≥70-year patients. Patients with haematological cancers also had an increased risk of ROEL ET AL. 9 F I G U R E 3 Adjusted hazard ratios of COVID-19 outcomes in patients with cancer (overall and by years since cancer diagnosis) compared to patients without cancer, stratified by sex, age and cancer type (solid or haematological). Models are adjusted for sex (excepting models stratified by sex), age, the MEDEA deprivation index, smoking status and comorbidities (autoimmune conditions, chronic kidney disease, chronic obstructive pulmonary disease, dementia, heart disease, hyperlipidaemia, hypertension, type-2 diabetes and obesity). aHR, adjusted hazard ratio; CI, confidence interval F I G U R E 4 Adjusted hazard ratios of COVID-19 outcomes in patients with cancer (overall and by years since the cancer diagnosis) compared to patients without cancer, stratified by solid cancer type. Models for specific cancer types include patients without cancer and patients with the cancer type of interest; models for prostate and breast cancer include only males and females, respectively. Models are adjusted for sex, age, smoking status, the MEDEA deprivation index, smoking status and comorbidities (autoimmune conditions, chronic kidney disease, chronic obstructive pulmonary disease, dementia, heart disease, hyperlipidaemia, hypertension, type-2 diabetes and obesity). aHR, adjusted hazard ratio; CI, confidence interval 10 ROEL ET AL. COVID-19 hospitalisation following an outpatient diagnosis (overall infection in patients with cancer, which was higher in those recently 1.37 [1.10-1.71]; <1-year cancer patients: 2.24 [1.34-3.76]). diagnosed and/or with haematological cancers.6,27 Increased risk of We also estimated the associations between cancer and COVID- diagnosis could be related to higher levels of interaction with 19 outcomes by solid cancers. (Figure 4, Table S5). Due to small sam- healthcare services among patients with cancers, especially among ples, models were estimated for breast, prostate, colorectal, bladder those with a recent cancer diagnosis (thus, higher risk of being diag- and lung cancer; overall and for <5-year (<1-year and 1-5-year cate- nosed with COVID-19 but also higher exposure to healthcare- gories combined) and ≥5-year cancer patients. Four cancer types were associated infections), and to factors related to the cancer itself associated with having a direct COVID-19 hospitalisation: breast and/or cancer therapies (eg, haematological cancers, as well as (1.30 [1.10-1.54]), colorectal (1.28 [1.10-1.49]), bladder (1.50 treatment-related immunosuppression, thus increasing the risk of [1.26-1.79]) and lung (1.53 [1.13-2.08]) cancer; these associations infection).29 were stronger in <5-year cancer patients. Lung cancer was associated Patients with cancer have also been reported to be at increased with death following a COVID-19 diagnosis (1.68 [1.06-2.64]), with a risk of COVID-19 severity, including hospitalisation and death.6-9 We stronger association in <5-year cancer patients (2.57 [1.49-4.46]). found that cancer was associated with a higher risk of direct Bladder cancer was associated with death following a COVID-19 hospitalisation, especially among <1-year cancer patients. Conversely, hospitalisation only in <5-year cancer patients (1.70 [1.11-2.60]). <1-year cancer patients had not an increased risk of subsequent hospitalisation (following an outpatient diagnosis). This counterintui- tive finding could be explained by differences in care-seeking behav- 3.4 | Sensitivity analysis iours and/or in the clinical presentation of COVID-19. On the one hand, patients recently diagnosed with cancer have more interactions The assumption of proportionality was violated for age and years with hospital services and, therefore, could be more prone to seek since cancer diagnosis for the risk of COVID-19 diagnosis (Figure S5). care directly at the hospital level than the general population.30 On Thus, we stratified our model by years since cancer diagnosis and cal- the other hand, these patients might have a higher risk of rapidly endar time (Figure S6). The overall association was similar in March developing severe COVID-19 symptoms due to their impaired immu- and April. However, in <1-year cancer patients, cancer was associated nity, thus more likely to be directly hospitalised. It is worth noting that with a significant increased risk of COVID-19 diagnosis in April (1.41 although <1-year cancer patients had the highest risk of [1.23-1.60]) but not in March (0.91 [0.80-1.05]). hospitalisation, this association remained significant in >5-year cancer In models restricted to never smokers (n = 1 834 657), the results patients (which mostly represent cancer survivors). This is consistent were similar to those including all the population (Figure S7). Patients with a study showing that cancer survivors have higher risks of with missing data (n = 1 502 442) were younger and had fewer com- hospitalisation and death from influenza than cancer-free patients,31 orbidities than patients without missing data, but the distribution of and could be related to long-term effects on the immune system of cancer types was similar in both groups (Table S6). Despite these dif- cancer therapies. ferences, imputed models showed similar results to the main models Conversely, the risk of COVID-19-related death was only signifi- (Figure S8). cantly higher in <1-year cancer patients. Again, this could be due to factors related to the cancer itself (ie, the group of <1-year cancer patients might include individuals with more aggressive and active 4 | DISCUSSION cancers) and/or cancer therapies. However, while some studies have shown that active cancer therapies increase the risk of COVID-19 In our population-based cohort study including 4 618 377 adults, a death,9 others have not.8,32 These studies included different prior diagnosis of cancer was associated with an increased risk of populations, cancer types or considered all different cancer therapies COVID-19 outpatient (clinical) diagnosis, direct COVID-19 combined, which might have a different impact on COVID-19 out- hospitalisation (without a prior outpatient diagnosis) and COVID- comes. For instance, two meta-analyses reported an association 19-related death during the first wave of the COVID-19 pandemic in between recent chemotherapy and increased COVID-19-related Catalonia, Spain. Overall, these associations were stronger in patients death, but a null association with recent surgery, radiotherapy, immu- with a recent cancer diagnosis (<1 year), younger than 70 years and notherapy and targeted therapies.33,34 with haematological cancers. Lung and bladder cancers were also We found that the associations between cancer and direct associated with higher risk of COVID-19 hospitalisation and death. hospitalisation and COVID-19-related deaths were more pronounced Prior studies investigating the risk of contracting SARS-CoV-2 in in patients younger than 70 years or with haematological cancers. patients with cancer have reported conflicting results.6,10,27,28 Even Given that age is strongly associated with severe COVID-19 out- though we did not analyse the risk of COVID-19 infection per se, comes, cancer in older patients might not have a significantly worse patients with cancer had a modestly increased risk of having an outpa- impact as compared to cancer-free patients. In a study including 1187 tient COVID-19 diagnosis, which was higher in <1-year cancer patients with solid cancers and COVID-19, younger patients patients with haematological cancers. This is consistent with two (<60 years) were also those with the highest risk of in-hospital mortal- studies from the United States (US) showing an increased risk of ity when compared to cancer-free patients.35 Furthermore, increasing ROEL ET AL. 11 evidence shows that patients with haematological cancers have a previously validated cancer codes,19 and we included only individuals higher risk of poor COVID-19 outcomes.6,7,9 The OpenSAFELY study with at least 1 year of prior history available to comprehensively cap- reported an association between cancer and increased COVID-19 ture baseline characteristics. death, which was stronger in <1-year cancer patients and in those Despite these weaknesses, our results highlight that patients with with haematological cancers.7 Estimated aHRs for <1-year cancer cancer are a vulnerable population for COVID-19 and, therefore, patients were similar to ours for death following a COVID-19 diagno- should be prioritised for vaccination against SARS-CoV-2. Unfortu- sis, with an aHR of 1.72 (1.50-1.96) (vs 1.69 [1.30-2.19] in our study) nately, the efficacy and effectiveness of COVID-19 vaccines in this for solid cancer patients and an aHR of 2.80 (2.08-3.78) (vs 3.11 subgroup population remain unknown. Indeed, patients with active [1.67-5.81]) for haematological cancer patients. We also found a cancer were excluded from randomised clinical trials,38 and, to our higher risk of hospitalisation and COVID-19-related death for lung knowledge, observational studies describing vaccine's effectiveness and bladder cancers, both of which are strongly linked to tobacco among patients with cancer are lacking to date. Emerging data suggest smoking. While lung cancer has already been associated with poor that these patients might have a weakened response to COVID-19 COVID-19 outcomes,36 to our knowledge, our study is the first show- vaccines,39,40 and recent studies have shown that COVID-19 vaccines ing an association with bladder cancer. However, these findings are less effective among individuals immunocompromised.41,42 As a should be interpreted with caution considering the small sample sizes, result, the Centers for Disease Control and Prevention recently rec- which prevented us from performing analysis restricted to never ommended a third mRNA-based vaccine dose among individuals smokers by specific cancer types. immunocompromised, which include patients with ongoing treatment Our study has several strengths. First, we used prospective data for haematological cancers or who have received a stem cell trans- from a large and representative population covering almost all the plant within the last 2 years.43 Further studies are needed to assess population in Catalonia, and we included a heterogeneous cancer pop- the effectiveness of COVID-19 vaccines among patients with cancer, ulation. Second, by including patients with a clinical COVID-19 diag- overall and by oncologic features (eg, cancer type, cancer treatment), nosis, we avoided selection bias due to testing restrictions, or to as well as to elucidate the utility of antibody testing44 and booster (hypothetically) different testing patterns (ie, higher rates of testing in vaccine doses. Meanwhile, these patients should also be protected patients with cancer), although some cases might be false positives. with continued non-pharmaceutical interventions, infection control Third, we performed our analysis across different cancer population measures in healthcare settings and increased vaccination uptake groups, allowing us to identify those at highest risk of poor COVID-19 among their caregivers and close contacts. outcomes. Finally, our results were robust after restricting participants In conclusion, our population-based cohort study including a het- to never smokers and after multiple imputation of missing data, which erogeneous cancer population provides a comprehensive analysis of lends credibility to our findings. the associations between cancer and COVID-19 outcomes during the However, our study also has weaknesses. First, we did not have first wave of the pandemic in a Southern European region. Cancer information on cancer stage nor specific-cancer therapy receipt and was associated with an increased risk of COVID-19 diagnosis, used instead years since cancer diagnosis as a proxy for active/ hospitalisation and COVID-19-related death, with higher risks for inactive cancer. We also did not have information on the cause of patients diagnosed with cancer within the year prior, as well as those death and considered as COVID-19-related deaths those occurring younger than 70 years and those with haematological cancers. following a COVID-19 state. However, in patients with cancer, occur- Research is needed to address potential risk differences by specific rence of death was substantially higher in those diagnosed (11.1%) cancer types, such as lung or bladder cancer, as well as to analyse the and hospitalised (24.8%) with COVID-19 than in those without effect of subsequent COVID-19 waves. Notwithstanding that, our COVID-19 (1.3%), which suggests that we did capture deaths due to results highlight that patients with cancer are a vulnerable population COVID-19. In addition, the proportion of deaths among hospitalised for COVID-19. These patients, as well as their caregivers, should be patients was in line with prior studies.37 On the other hand, we cannot prioritised in preventive strategies, including vaccination campaigns discard that some deaths in the general population might have and continued non-pharmaceutical interventions. occurred in undiagnosed COVID-19 cases, especially at the beginning of the pandemic. Second, due to the nature of our database, our AC KNOW LEDG EME NT S results are not representative of asymptomatic or pauci-symptomatic The authors would like to acknowledge the patients who suffered COVID-19 cases that did not seek medical care. Third, our data from or died of this devastating disease, and their families and carers. spanned to May 2020, and therefore, our results are generalizable to They would also like to thank the healthcare professionals involved in the first wave of the pandemic. While changes over time might have the management of COVID-19 during these challenging times, from changed SARS-CoV-2 virulence (eg, the emergence of new variants), primary care to intensive care units in the Catalan healthcare system. it is unlikely that such changes have decreased the risk of severe dis- The analysis has made use of a range of open-source, free available ease among patients with cancer when compared to patients without tools provided by the OHDSI community. The phenotypes used in our cancer. Finally, routinely collected data often raise concerns about study were developed or adapted from work performed during the data quality, and some conditions, including cancer itself, may have OHDSI COVID-19 studyathon. The mapping of data to the OMOP been incompletely or inaccurately recorded. However, we used CDM has been supported by a taskforce from EHDEN. 12 ROEL ET AL. CONF LICT OF IN TE RE ST 3. Liang W, Guan W, Chen R, et al. Cancer patients in SARS-CoV-2 infec- All authors have completed the ICMJE uniform disclosure form at tion: a nationwide analysis in China. Lancet Oncol. 2020;21(3):335-337. 4. Guan W, Ni Z, Hu Y, et al. Clinical characteristics of coronavirus dis- www.icmje.org/coi_disclosure.pdf and declare: DPA reports grant ease 2019 in China. N Engl J Med. 2020;382:1708-1720. support from Les Laboratoires Servier; that his research group has 5. Zhang Y, Han H, Tian Y, et al. Impact of cancer on mortality and received grants and advisory or speaker fees from Amgen, Astellas, severity of corona virus disease 2019: a protocol for systematic AstraZeneca, Chesi-Taylor, Johnson and Johnson and UCD; and that review and meta-analysis. Medicine (Baltimore). 2020;99(44):e23005. 6. Wang Q, Berger NA, Xu R. Analyses of risk, racial disparity, and out- Janssen, on behalf of Innovative Medicines Initiative-funded comes among US patients with cancer and COVID-19 infection. European Health Data Evidence Network and European Medical JAMA Oncol. 2020;7(2):220-227. Information Framework consortium and Synapse Management Part- 7. Williamson EJ, Walker AJ, Bhaskaran K, et al. Factors associated with ners, have supported training programs, open to external participants, COVID-19-related death using OpenSAFELY. Nature. 2020;584: 430-436. organised by his department. No other relationships or activities that 8. Kuderer NM, Choueiri TK, Shah DP, et al. Clinical impact of COVID- could appear to have influenced the submitted work. 19 on patients with cancer (CCC19): a cohort study. Lancet. 2020; 6736:1-13. DATA AVAI LAB ILITY S TATEMENT 9. Lee LYW, Cazier JB, Starkey T, et al. COVID-19 prevalence and mor- tality in patients with cancer and the effect of primary tumour sub- In accordance with current European and national law, the data used type and patient demographics: a prospective cohort study. Lancet in our study are only available for the researchers participating in our Oncol. 2020;21:1309-1316. study. Thus, we are not allowed to distribute or make publicly avail- 10. Angelis V, Tippu Z, Joshi K, Reis S, et al. Defining the true impact of able the data to other parties. Researchers from public institutions can coronavirus disease 2019 in the at-risk population of patients with request data from SIDIAP if they comply with certain requirements. cancer. Eur J Cancer. 2020;136:99-106. 11. Reilev M, Kristensen KB, Pottegård A, et al. Characteristics and pre- Further information is available online (https://www.sidiap.org/index. dictors of hospitalization and death in the first 11 122 cases with a php/menu-solicitudesen/application-proccedure) or by contacting positive RT-PCR test for SARS-CoV-2 in Denmark: a nationwide Anna Moleras (amoleras@idiapjgol.org). cohort. Int J Epidemiol. 2020;49:1468-1481. 12. Brar G, Pinheiro LC, Shusterman M, et al. COVID-19 severity and out- comes in patients with cancer: a matched cohort study. J Clin Oncol. E TH I CS S T A TE M E N T 2020;38:3914-3924. Our study was approved by the Clinical Research Ethics Committee of 13. Griffith GJ, Morris TT, Tudball MJ, et al. Collider bias undermines our the IDIAPJGol (project code: 20/070-PCV). Informed consent of indi- understanding of COVID-19 disease risk and severity. Nat Commun. vidual patients was not required as anonymised information was 2020;11:1-12. 14. Prieto-Alhambra D, Ballo  E, Coma E, et al. Filling the gaps in the char- obtained from medical records. acterization of the clinical management of COVID-19: 30-day hospital admission and fatality rates in a cohort of 118 150 cases diagnosed in TRANSPARENCY STATEMENT outpatient settings in Spain. Int J Epidemiol. 2020;49:1930-1939. 15. Bolíbar B, Fina Avilés F, Morros R, et al. Base de datos SIDIAP: La Elena Roel and Talita Duarte-Salles as guarantors of the study affirm historia clínica informatizada de Atencio  n Primaria como fuente de that the study is an honest, accurate and transparent account of the  n para la investigacio  n epidemiolo  gica. Med Clin (Barc). informacio study being reported; that no important aspects of the study have 2012;138:617-621. been omitted and that any discrepancies from the study as planned 16. Hripcsak G, Duke JD, Shah NH, et al. Observational health data sci- have been explained. ences and informatics (OHDSI): opportunities for observational researchers. Stud Health Technol Inform. 2015;216:574-578. 17. Burn E, Tebé C, Fernandez-Bertolin S, et al. The natural history of ORCID symptomatic COVID-19 during the first wave in Catalonia. Nat Commun. 2021;12:1-12. Elena Roel https://orcid.org/0000-0002-1964-3546 18. Putter H, Fiocco M, Gekus RB. Tutorial in biostatistics: competing risk Martina Recalde https://orcid.org/0000-0003-2256-618X and multi-state models. Stat Med. 2007;26:2389-2430. Sergio Fernández-Bertolín https://orcid.org/0000-0002-0029-6564 19. Recalde M, Manzano-Salgado CB, Díaz Y, et al. Validation of cancer Isabelle Soerjomataram https://orcid.org/0000-0002-6017-741X diagnoses in electronic health records: results from the information system for research in primary care (SIDIAP) in Northeast Spain. Clin Diana Puente https://orcid.org/0000-0003-4725-537X Epidemiol. 2019;11:1015-1024. Daniel Prieto-Alhambra https://orcid.org/0000-0002-3950-6346 20. Domínguez-Berjo  n MF, Borrell C, Cano-Serral G, et al. Construccio n Edward Burn https://orcid.org/0000-0002-9286-1128 de un índice de privacio  n a partir de datos censales en grandes ciudades españolas (Proyecto MEDEA). Gac Sanit. 2008;22:179-187. Talita Duarte-Salles https://orcid.org/0000-0002-8274-0357 21. Richardson S, Hirsch JS, Narasimhan M, et al. Presenting characteristics, comorbidities, and outcomes among 5700 patients hospitalized with RE FE R ENC E S COVID-19 in the New York City area. JAMA. 2020;323:2052-2059. 1. Sung H, Ferlay J, Siegel RL, et al. Global cancer statistics 2020: 22. Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic GLOBOCAN estimates of incidence and mortality worldwide for research. Epidemiology. 1999;10:37-48. 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209-249. 23. Harrell FE. Regression Modeling Strategies. New York, NY: Springer; 2015. https://doi.org/10.3322/caac.21660 24. Austin PC. Using the standardized difference to compare the preva- 2. Kumar D, Humar A. Respiratory viral infections in transplant and lence of a binary variable between two groups in observational oncology patients. Infect Dis Clin North Am. 2010;24:395-412. research. Commun Stat Simul Comput. 2009;38:1228-1234. ROEL ET AL. 13 25. de Wreede LC, Fiocco M, Putter H. mstate: an R package for the anal- 37. Desai A, Gupta R, Advani S, et al. Mortality in hospitalized ysis of competing risks and multi-state models. J Stat Softw. 2011;38: patients with cancer and coronavirus disease 2019: a systematic 1-30. review and meta-analysis of cohort studies. Cancer. 2020;127(9): 26. Harrell Jr FE. rms: Regression modeling strategies. R package version 1459-1468. 5.1-0. [Internet]. 2017 [cited April 6, 2021]; Available from: https:// 38. Corti C, Curigliano G. Commentary: SARS-CoV-2 vaccines and cancer CRAN.R-project.org/package=rms patients. Ann Oncol. 2021;32:569-571. 27. Wang QQ, Berger NA, Xu R. When hematologic malignancies meet 39. Monin L, Laing AG, Muñoz-Ruiz M, et al. Safety and immunogenicity COVID-19 in the United States: infections, death and disparities. of one versus two doses of the COVID-19 vaccine BNT162b2 for Blood Rev. 2021;47:100775. patients with cancer: interim analysis of a prospective observational 28. Bertuzzi AF, Marrari A, Gennaro N, et al. Low incidence of SARS- study. Lancet Oncol. 2021;22(6):765-778. CoV-2 in patients with solid tumours on active treatment: an obser- 40. Goshen-Lago T, Waldhorn I, Holland R, et al. Serologic status and vational study at a tertiary cancer centre in Lombardy, Italy. Cancers toxic effects of the SARS-CoV-2 BNT162b2 vaccine in patients (Basel). 2020;12(9):2352. undergoing treatment for cancer. JAMA Oncol. 2021;1-8. 29. Seth G, Sethi S, Bhattarai S, Saini G, Singh CB, Aneja R. SARS-CoV-2 41. Chodick G, Tene L, Rotem RS, et al. The effectiveness of the two- infection in cancer patients: effects on disease outcomes and patient dose BNT162b2 vaccine: analysis of real-world data. Clin Infect Dis. prognosis. Cancers (Basel). 2020;12:1-16. 2021;ciab438. 30. Lash RS, Bell JF, Reed SC, et al. A systematic review of emergency 42. Tenforde MW, Patel MM, Ginde AA, et al. Effectiveness of severe department use among cancer patients. Cancer Nurs. 2017;40: acute respiratory syndrome coronavirus 2 messenger RNA vaccines 135-144. for preventing coronavirus disease 2019 Hospitalizations in the 31. Carreira H, Strongman H, Peppa M, et al. Prevalence of COVID- United States. Clin Infect Dis. 2021;ciab687. 19-related risk factors and risk of severe influenza outcomes in 43. Centers for Disease Control and Prevention. COVID-19 Vaccines for cancer survivors: a matched cohort study using linked English Moderately to Severely Immunocompromised People j CDC [Inter- electronic health records data. EClinicalMedicine. 2020;29–30: net]. [cited August 23, 2021]; Available from: https://www.cdc.gov/ 100656. coronavirus/2019-ncov/vaccines/recommendations/immuno.html 32. Fillmore NR, La J, Szalat RE, et al. Prevalence and outcome of 44. Sun L, Warner JL, Parikh RB. Immune responses to SARS-CoV-2 COVID-19 infection in cancer patients: a National Veterans Affairs among patients with cancer: what can seropositivity tell us? JAMA Study. J Natl Cancer Inst. 2021;113(6):691-698. Oncol. 2021;7:1123-1125. 33. Park R, Lee SA, Kim SY, de Melo AC, Kasi A. Association of active oncologic treatment and risk of death in cancer patients with COVID- 19: a systematic review and meta-analysis of patient data. Acta Oncol SUPPORTING INF ORMATION (Madr). 2021;60:13-19. Additional supporting information may be found in the online version 34. Yekedüz E, Utkan G, Ürün Y. A systematic review and meta-analysis: of the article at the publisher's website. the effect of active cancer treatment on severity of COVID-19. Eur J Cancer. 2020;141:92-104. 35. de Azambuja E, Brand~ao M, Wildiers H, et al. Impact of solid cancer on in-hospital mortality overall and among different subgroups of How to cite this article: Roel E, Pistillo A, Recalde M, et al. patients with COVID-19: a nationwide, population-based analysis. Cancer and the risk of coronavirus disease 2019 diagnosis, ESMO Open. 2020;5:e000947. hospitalisation and death: A population-based multistate 36. Garassino MC, Whisenant JG, Huang LC, et al. COVID-19 in cohort study including 4 618 377 adults in Catalonia, Spain. patients with thoracic malignancies (TERAVOLT): first results of an international, registry-based, cohort study. Lancet Oncol. 2020;21: Int. J. Cancer. 2021;1-13. doi:10.1002/ijc.33846 914-922.