Ever since it was first reported in China in late 2019, coronavirus disease 2019 (COVID-19) has impacted millions of patients worldwide. It is now well documented that this disease, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, preferentially affects the elderly and those with comorbidities (1, 2). Among frail patients, those with cancer are reported to harbor an increased susceptibility to SARS-COV-2 infection (3). This vulnerability is thought to be related to cancer-related immune compromised status and cancer-directed treatments.

However, data regarding the impact of SARS-CoV-2 infection on mortality of cancer patients are conflicting (4, 5). Although data in the ICU setting are scarce, most studies have identified a higher risk of mortality in cancer patients (6). Several confounding factors, such as specific patient case-mix, cancer treatment, or progression status, may explain this discrepancy. Furthermore, considering cancer-related immunosuppression, data regarding ICU-acquired infections that could prolong ICU morbidity and mortality are scarce.

Therefore, this study aimed to determine the outcomes of COVID-19 patients with cancer in the ICU, compared to a cohort of patients without cancer, using the inverse of the propensity score (PS) as weight. This enabled the creation of homogeneous cohorts from a real-life hospital setting with limited confounding effects. Using a large multicenter cohort of severe patients, we aim to provide robust and clinically relevant data to assist clinicians in the emergency management of fragile patients.

Between 12 March 2020 and 8 March 2021, 2,608 patients were admitted with SARS-CoV-2 infection in our study, accounting for 2.8% of the total population of patients with SARS-CoV-2 admitted in all French ICUs during the same period. The main patient characteristics are presented in Table 1 . The prevalence of cancer at ICU admission was 4% (n=105), with 51 and 54 patients presenting with haematological malignancy and solid tumours, respectively ( Table 2 ). Given that 105 patients had cancer at admission among the 2608 patients admitted during the study period, and because we used a 1:3 ratio matching to build the propensity score, we obtained a final cohort of 420 patients which constitutes the core of this work for the mortality analyses ( Supplementary Figure 1 ).

In our study, solid organ transplant was significantly associated with less cancer (OR: 0·01, 95% CI: 0–0·17, p<0·0001), whereas immunosuppressive agents (OR: 11·56, 95% CI: 4·2–31·81, p<0·0001) and hospitalisation in the prior 3 months (OR: 4·03, 95% CI: 1·97–8·26, p>0·0001) were both significantly associated with a higher risk of cancer. From the 420 evaluable severe COVID-19 patients, 409 patients were included in the main analysis, of whom 307 patients were in the non-cancer group and 102 patients in the cancer group. After IPW, the characteristics of the populations were well balanced ( Supplementary Figure 2 ).

The mortality rate of patients without cancer was 27%, while that of patients with cancer was 56% (n=59/105). Upon performing logistic regression for death in the weighted cohort, including both patients with and without cancer, we found that cancer was significantly associated with death, with an OR estimate of 2·30 (95% CI: 1·24–4·27, p=0·0082). Accordingly, other factors being equal, the death risk was 2·3 fold greater for cancer patients than for those without cancer at ICU admission.

The extent of this association may be related to the type of cancer, i.e., haematological malignancies or solid tumours. Therefore, we performed a logistic regression analysis of death in the weighted cohort considering the type of cancer. We found that haematological malignancy was significantly associated with death (OR: 3·19, 95% CI: 1·38–7·37, p=0·0065), whereas solid tumours were not associated with death (OR: 1·88, 95% CI: 0·82–4·27, p=0·13). We obtained similar results when patients were balanced using a specific PS (data not shown).

We also assessed the robustness of the estimates by performing sensitivity analyses. Cancer was still associated with ICU mortality when computing the PS using a backward selection with only the significant variables (OR: 2·24, 95% CI: 1·16–4·33, p=0·016) or when the PS was added in the final logistic regression analysis (OR: 2·42, 95% CI: 1·22–10·33, p=0·0019). No parameter was significantly associated with ICU mortality when variables of the PS were included in the final logistic regression model of the weighted population. Including PS components in the final model, we observed a strong trend towards an association between age at admission and ICU mortality (OR: 1·04, 95% CI: 0·996–1·088, p=0·07).

We performed exploratory analyses to determine the association between death and patient characteristics, and ICU management. Although the management of SARS-CoV-2 infection varied among patients ( Table 3 ), some parameters were associated with ICU mortality ( Table 4 ). In patients who required the use of norepinephrine during the ICU stay (OR: 2·61, 95% CI: 1·10-6·16, p=0·028) and in those who required the use of antiobiotic therapy (OR: 6·85, 95% CI: 2·13-22·02; p=0·0013), mortality was significantly higher. Conversely, treatment with hydroxychloroquine and azithromycin was associated with a better outcome (OR: 0·187, 95% CI: 0·047–0·74; p=0·0177). Prone positioning had a trend towards an association with an unfavourable outcome (OR: 1·99, 95% CI: 0·89–4·44, p=0·092). Of note, we have analysed ICU mortality between pre-treated cancer, non pre-treated cancer and non-cancer patients. Based on the main logistic model, we obtained an estimate of OR equal to 2.55 [1.15; 5.65] for the pre-treated cancer patients. For the cancer patients without pre-treatment within 4 weeks, the OR is estimated at 1.68 [0.48; 5.87]. However, the overall effect of this variable is not associated with mortality.

Finally, we computed a multivariate model including parameters significantly associated with mortality (cancer type, norepinephrine use, antimicrobial treatment use, and treatment with hydroxychloroquine and azithromycin). We added the age at admission in this model because this parameter has been widely and robustly reported in the literature as an independent parameter associated with mortality in patients with severe SARS-CoV-2 infection. Moreover, we added the period of admittance of the patient in this model based on the definition of the first and second waves of SARS-CoV-2 in France, with 1 August 2020 as the cut-off date ( Table 5 ). The multivariate model identified that the presence of solid tumours (OR: 2·344, 95% CI: 0·87–6·31, p=0·062) and haematological malignancies (OR: 4·144, 95% CI: 1·24–13·83, p=0·062) were independently associated with mortality. Antimicrobial use and admission during the second wave were also independently associated with increased mortality in this study.

Among patients admitted to the ICU with severe SARS-COV-2 infection, patients with cancer represented 4% of the entire population in our study. Our analysis suggests that the risk of ICU death was 2·3-fold greater in cancer patients than in those without cancer after adjustment for confounding factors. Among cancer patients, those with haematological malignancies had a higher risk in comparison with solid tumours.

From the outset of the COVID-19 pandemic, cancer patients have been reported to be highly vulnerable (7, 8).. However, they account for a relatively low proportion, ranging from 2·4% to 9% of SARS-CoV-2 infection-related ICU admissions in series reporting this comorbidity (9–11).. This proportion is lower than the rate of patients admitted to a general ICU with cancer outside the COVID-19 context, which varies between 10·9% and 21·5% (12–14).. This difference does not seem to be related to a lower incidence of SARS-CoV-2 infection in patients with cancer, given that age-adjusted rates of SARS-CoV-2 infection in the population with cancer are reported to be similar to those in the population without cancer (15). Moreover, several studies have reported that patients with cancer are more likely to need ICU admission (16). Therefore, our result (i.e. the lower rate of cancer should be considered as an unfavorable comorbidity in the triage process. Moreover, in addition to prognosis, level of physiological compromise, and the existence of advanced directives among cancer patients; experts have also stated that ICU transfer decisions must include an assessment of the strain on ICU resources (17). However, a recent French study reported that differences in mortality rates at a regional level were not significantly associated with cancer status (18).

Owing to cancer-related immune alterations, immunosuppressive treatment, and cancer-related organ failure, patients with cancer have a higher risk of mortality in ICUs (19). Before the pandemic era, a meta-analysis of individual data of patients reported an overall ICU mortality rate of 49% in cancer patients, whereas the ICU mortality rate of patients without cancer ranges from 10% to 29% (20). In the context of COVID-19, a recent meta-analysis reported an overall mortality of 60% in 1,276 cancer patients admitted to the ICU with severe SARS-COV-2 infection, with an almost 2-fold increased risk of mortality than in patients without cancer (6).

Several parameters could play a role in the highly fatal ICU course of cancer patients with severe SARS-COV-2 infection. First, recent cancer treatment or comorbidities could arguably affect the course of SARS-CoV-2 infection. In our study, neither anticancer therapy in the 4 weeks before ICU admission nor comorbidities were significantly associated with the poor outcome of cancer patients, suggesting that the cancer itself exerts a deleterious effect on patients’ outcomes. Second, one could argue that cancer-related immune suppression could promote the occurrence of secondary infections and therefore contribute to a higher mortality rate. In line with this hypothesis, we identified a higher number of ICU-acquired infections in cancer patients in our study, particularly among patients with haematological malignancies. However, these nosocomial complications were considered responsible for only a low proportion of deaths, as assessed by physicians. These results must be interpreted cautiously, given controversial data regarding both the relatively low attributable mortality of nosocomial infections in septic patients and the variability in the definition of causes of death (21). Third, an increased risk of thromboembolic events could explain higher mortality rates among cancer patients. Cancer patients with SARS-CoV-2 infection accumulate cancer-related inflammatory pro-thrombotic risk along with the risk of thromboembolic events known to accompany viral infections (22). However, we noted only a few cases of pulmonary embolism and deep vein thrombosis. Moreover, cancer is not associated with a higher risk of a COVID-19-associated hypercoagulable state (23, 24). Finally, the impact of specific types of malignancy could influence the risk of death in patients with cancer. Although the sample size of this study does not allow the provision of robust statistical data regarding solid cancer subgroups, we found no differences in the repartition of solid cancer type. However, we observed a statistically higher risk of death in patients with haematological malignancies.

One of the more consistent findings in the literature is that there is a higher risk among patients with haematologic malignancies (25–27). In our study, patients with lymphoma and myeloma accounted for more than 80% of the population, with a low proportion of acute myeloid leukaemia. These patients exhibit a less pronounced immune response to the SARS-CoV-2 virus, including heterogeneous humoral responses and a high prevalence of prolonged virus shedding (28). However, in the specific context of lymphoma and chronic lymphoid leukaemia, it is still unclear whether altered B-cell immune response or B-cell depletion is the only cause of this increased mortality. Our results suggest that in the process of assessing whether a patient should be admitted to the ICU with severe SARS-CoV-2 infection, the existence of a haematological malignancy should be conceptualized differently from the presence of a solid tumor, as their prognosis varies significantly. Dan s

Using multivariate analysis, we also identified antimicrobial use and admission during the second wave as being independently associated with increased mortality. The association between antibiotics and mortality may reflect an increased incidence of secondary ICU-acquired pneumonia that was previously reported to exert a deleterious effect on the ICU course of COVID-19 (11). On the other hand, the independent deleterious effect of admission after 1 August 2020 was difficult to interpret. Though the case fatality rate of SARS-COV-2 infection between the first and second waves varies among countries and regions (29), studies suggest that changes in the proportion of SARS-COV-2 variant affect COVID-19 mortality (30).

This study had several limitations. First, the discrepancy we observed may involve characteristics of the cancer population that were not collected in this study, such as socioeconomic status or control of the underlying comorbidities. An additional limitation of this study relates to mortality attributable to SARS-CoV-2 infection. Mortality in cancer patients was assumed to be related to SARS-CoV-2 infection, whereas certain malignancies, such as acute myeloid leukaemia, are known to be life-threatening. Another limitation concerns the fact that despite the inverse probability of cancer weighting, some residual confounding factors could remain. Finally, the relatively small size of our cohort does not allow a statistically robust assessment of the effect of cancer subtypes on mortality.

We identified a 2·3-fold increase in death rate between COVID-19 patients with and without cancer admitted to the ICU. Patients with haematological malignancies had a higher risk of death. Recent cancer-specific therapies, comorbidities, ICU management, ICU-acquired infections, and thromboembolic events were not associated with a higher risk of death, suggesting that the cancer itself exerts a deleterious effect on the clinical course of SARS-CoV-2 infection. Our results suggest reinforcing vaccination and prevention strategies in this specific population. Furthermore, strategies for ICU admission triage must consider this susceptibility and involve oncologists to deliver patient-centred care.

Clinical study report are available upon request, after approval by the study principal investigator (corresponding author). Deidentified individual participant data from this clinical trial, as well as data dictionary, can be requested by filling out the data request form for Gustave Roussy clinical trials at https://redcap.gustaveroussy.fr/redcap/surveys/?s=DYDTLPE4AM. The process is similar for every study sponsored by Gustave Roussy. The study steering committee and the sponsor will review the requests on a case-by-case basis. In case of approval, a specific agreement between the sponsor and the researcher may be required for data transfer.

The studies involving human participants were reviewed and approved by Comité de protection des personnes Ile-de-Frande IV, IRB number 00003835. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

J-FL, OM, and AB were responsible for trial conception and design. MLe, TC, PN, GP, RT, MJo, HP, LL, CD, LA, MJa, CG, KM, ML, BS, SS, AS, and J-FL recruited and cared for patients. Collection and assembly of data was done by J-FL, HP and AB, while the data analysis and interpretation was done by AB, NI, and J-FL. AB, NI, and J-FL had access to and verified the raw data. J-FL wrote the first draft of the manuscript and all authors participated in its writing and gave approval for publication. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication. All authors read and approved the final manuscript.

AB declare consulting fees for ROCHE SAS. J-FL is employed by bioMérieux but was employed by Institut Gustave Roussy when that study was performed.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.