Since the first case of Severe Acute Respiratory Syndrome-Coronavirus 2 (SARS-CoV-2) in China in December 2019, the World Health Organization has recorded more than 580 million infections and more than 6.4 million deaths worldwide by the end of August 20221.

As highlighted in a French study based on 66 million people, risk factors such as older age, male sex and numerous chronic conditions (including active cancers) are associated with a higher risk of hospitalization and death from SARS-CoV-2 infection2. Furthermore, vaccinated patients with active cancers have an increased risk of in-hospital death3. On an international level, numerous studies demonstrated that cancer patients are at higher risk of severe outcome when infected with SARS-CoV-24,5,6,7,8,9,10,11. For example, an English study (OpenSAFELY database, including more than 17 million individuals) showed that hematological cancer patients were up to 2.5 times more likely to die from a SARS-CoV-2 infection12. Sinha and colleagues explain the greater fragility to SARS-CoV-2 in cancer patients as a result of six main elements: older age, increased expression of the ACE2 receptor, increased expression of the TMPRSS2 protease, underlying immunosuppression (due to cancer and/or anti-cancer treatments), significant inflammatory response and cancer-induced pro-coagulant state13.

The question one might ask is whether the severity of SARS-CoV-2 infection for individuals hospitalized in intensive care unit (ICU) is a marker or a result of underlying immunosuppression. Regarding the link between immunosuppression and new cancers, a meta-analysis (including 444,172 HIV/AIDS and 31,977 transplant patients), showed that these patients had an increased risk of developing cancers because of immunosuppression, particularly for cancers of infectious etiology (for example Hodgkin's lymphoma)14. Other studies support these results, either for stem cell transplant patients15,16,17 or HIV patients18. As explained by Al-Adra and colleagues, several pathophysiological mechanisms could explain the increased susceptibility of transplant patients to cancer: the loss of immune system control over oncogenic viruses, the accumulation of mutations no longer recognized by the immune system, and the direct effect of some immunosuppressive treatments19.

With the evidence that cancer is a risk factor for severe SARS-Cov-2 infection and being aware of the link between immunosuppression and the development of new cancers, the research question which motivated this study was the following: Is severe SARS-CoV-2 infection a marker of an undiagnosed cancer already present at the time of SARS-CoV-2 infection?


Data sources

This study is based on the SNDS database (“Système National des Données de Santé”), a medico-administrative database which includes healthcare reimbursements data of the whole French population (67 million inhabitants) and which has been extensively used for pharmaco-epidemiology studies2,3,20. The SNDS database is subdivided into two distinct sub-databases, the DCIR database (“Datamart de Consommation Inter Régimes”) and the PMSI database (“Programme de Médicalisation des Systèmes d'Information”). Since 2006, the existence of a unique and anonymized identifier makes it possible to link the information contained in these two sub-databases. The DCIR contains information on the reimbursement of ambulatory medical care (including ambulatory medical care, laboratory tests and drugs according to the International Anatomical Therapeutic Chemical classification system). The PMSI contains information on the admission and discharge dates of any hospitalization in public or private hospital establishment in France, as well as the medical diagnoses related to the hospitalization (coded according to the ICD-10 classification, main medical or surgical procedures classified according to the “Classification Commune des Actes Médicaux”). Regarding the identification of individuals’ comorbidities, a specific tool was developed from the DCIR and the PMSI databases (“Cartographie des Pathologies et des Dépenses”), allowing the identification of pathologies from medical algorithms (based on the reasons for hospitalization, diagnoses of long-term illnesses and the reimbursement of specific treatments), for the previous 5 years21.

ICU hospitalized individuals were included from February 15, 2020 until August 31, 2021 (beginning of the SARS-CoV-2 pandemic, end of the 4th wave in France). Regarding the duration of follow-up, the index date was the admission date of hospitalization of ICU hospitalized individuals (same date for matched control individuals), with a follow-up end date on December 31, 2021, allowing a minimum follow-up of 4 months for individuals admitted to hospital in August 2021. All data was collected between March and June 2022 (reason why the follow-up end date was set for end of 2021).

Inclusion and exclusion criteria

We collected data on individuals aged ≥ 16 years, living in mainland France, having benefited from at least one health care reimbursement in the 2 years preceding the index date, with no history of cancer in the previous 5 years. Individuals living in nursing homes and twins < 22 years were excluded (Fig. 1; Supplementary Table S1 online). Individuals were included in two groups, the ICU hospitalized group (ICU-gr) and the matched control group (C-gr). The two groups were matched on the basis of year of birth, sex, and French department (n = 95). Each individual from the ICU-gr was matched with between six (minimum) and 20 (maximum) individuals belonging to the C-gr (average number 17). We matched to a large number of controls (a maximum of 20 controls was deliberately chosen) in order to increase the representativeness of the C-gr when compared to the ICU-gr.

Figure 1
figure 1

Flow Chart of the study. ICU-gr ICU hospitalized group, C-gr matched control group.

Sociodemographic characteristics and co-variables

The following sociodemographic characteristics were taken into account: age, sex, and region of residence. The social deprivation index was used as a measure of the socio-economic status. The following co-variables were analyzed: various comorbidities, addictive disorders, vaccination status for SARS-CoV-2, and immunosuppressive/oral corticoid treatment (Table 1). The variables are defined in a previous article2, as the social deprivation index22.

Table 1 Sociodemographic characteristics and comorbidities of the ICU hospitalized group and the matched control group.

Outcome and censoring criteria

The outcome was the incidence of cancers in the two groups during the follow-up period. A cancer case was defined as any hospitalization for cancer or any long-term cancer-like condition needing health care reimbursement (including in situ cancers). The censoring criteria which required the exclusion of the individual (or the end of the follow-up) after the initial inclusion were the death of the individual (ICU-gr and C-gr), the outcome occurrence (ICU-gr and C-gr), and the hospitalization due to a SARS-CoV-2 infection (C-gr; 5177 control individuals in the C-gr, i.e. < 1%, were censored because they were hospitalized for a SARS-CoV-2 infection; 694 of these control individuals were then re-included in the ICU-gr). The censoring criteria were applied at the individual level and censoring was done at the first event that occurred. The death was recorded through death certificates registered in the database, which therefore included deaths from any cause.

Statistical analysis

The categorical variables are reported as frequencies with percentages and the continuous variables reported as means with standard deviations. To study the association between severe SARS-CoV-2 infection and overall cancer, as well as the association with specific cancer sites, we conducted Cox proportional hazards models that were systematically adjusted on matching variables and with further adjustment for all the co-variables previously described. In secondary analyses, we excluded in situ cancers, lung cancers or events occurring during the SARS-CoV-2 hospital stay. The follow-up was also divided into two sub-periods, distinguishing the first 3 months from the rest of the period, to assess the consistency of the associations over time (Table 3). Analyses by sex and age groups were performed (Table 4). Analysis taking into account death as competing event was conducted using Cox cause-specific hazard method (Supplementary Table S5 online). Missing data in the database, which concerned only the social deprivation index, were analyzed as a separate group (small number of missing data: 1.8% in C-gr, 1.6% in ICU-gr). All statistical analyses were performed using SAS software, version 9.4 (SAS Institute Inc.).

Regulatory approval and ethical aspects

The French National Health Data System (SNDS) is a medico-administrative database containing all the healthcare reimbursements of the French population. EPI-PHARE has permanent regulatory access to the data from the SNDS via its constitutive bodies ANSM and CNAM. This permanent access is given in accordance with the French Decree No. 2016-1871 of December 26, 2016 relating to the processing of personal data called the "National Health Data System"23 and French law articles Art. R. 1461-1324 and 1425. All requests in the database were made by duly authorized people. In accordance with the permanent regulatory access granted to EPI-PHARE via ANSM and CNAM, this work did not require the approval from the French Data Protection Authority (CNIL) nor the approval from the ethics committee/institutional review board. The study was registered on the study register of EPI-PHARE concerning studies from SNDS data under the reference [EP-0376]. In accordance with data protection legislation and the French regulation, the authors cannot publicly release the data from the SNDS. However, any person or structure, public or private, for-profit or non-profit, is able to access SNDS data upon authorization from the French Data Protection Office (CNIL), in order to carry out a study, research or an evaluation in the public interest ( and


Between February 15, 2020, and August 31, 2021, 41,302 individuals were hospitalized in the ICU in France due to a SARS-CoV-2 infection. These individuals were matched with 713,670 individuals who were not hospitalized for a SARS-CoV-2 infection (Fig. 1).

Sociodemographic characteristics and comorbidities

Sociodemographic characteristics are detailed in Table 1. The mean age was 60.8 years in the ICU-gr (standard deviation (SD) 12.8), 60.0 years in the C-gr (SD 12.8), and 67% of individuals were men. More individuals from the ICU-gr were in the most deprived class (ICU-gr 24.5%; C-gr 19.7%). Smoking cessation program was slightly more prevalent in the C-gr (ICU-gr 3.8%; C-gr 5.0%). The majority of individuals of the two groups were not vaccinated against SARS-Cov-2 (without any vaccination dose received at the index date), even if unvaccinated individuals were more prevalent in the ICU-gr (ICU-gr 95.3%; C-gr 86.5%). Only 0.8% of individuals in the ICU-gr (n = 324) and 6.6% of individuals in the C-gr (n = 47,100) had received at least two doses of vaccine against SARS-CoV-2 at that time. However, SARS-CoV-2 vaccines were available to the French population only since December 27, 2020. More than twice as many individuals in the ICU-gr were under immunosuppressive therapy compared to the C-gr (ICU-gr 2.4%; C-gr 0.9%) and more individuals of the ICU-gr were under oral corticoid treatment (which includes any oral dose) compared to the C-gr (ICU-gr 3.1%; C-gr 0.7%). Individuals in the ICU-gr had more comorbidities overall compared to individuals in the C-gr. The median follow-up period was 327 days for the ICU-gr (interquartile range (IQR) 257–444 days) and 340 days for the C-gr (IQR 267–457 days).

Cancer incidence in the two groups

In total, 897/41,302 individuals (2.2%) in the ICU-gr and 10,944/713,670 individuals (1.5%) in the C-gr were diagnosed with a cancer. The mean age at cancer diagnosis was 68.0 years (SD 9.3). The repartition of cancers according to cancer site is detailed in the Supplement (Supplementary Table S2 online). The median follow-up time for individuals to present the outcome was 168 days in the ICU-gr (IQR 73–270 days) and 200 days in the C-gr (IQR 99–322 days).

Using a Cox model adjusted only for age and sex, individuals in the ICU-gr had a 1.45 higher risk of being diagnosed with a cancer during the follow-up period compared to the C-gr (aHR 1.45, 95% CI 1.36–1.55). With a multivariable model (taking into account all the co-variables in Table 1), the adjusted HR was 1.31 for the ICU-gr (95% CI 1.22–1.41) (Table 2). The association obtained between the outcome and the exposure is relatively stable (same order of magnitude) between univariable and multivariable models, as is the range of the 95% confidence interval (Table 2). Similar results were observed when in situ cancers were excluded (aHR 1.32, 95% CI 1.23–1.42) or when lung cancers were excluded (aHR 1.27, 95% CI 1.18–1.37) (Supplementary Table S3 online).

Table 2 Occurrence of overall cancer in the ICU hospitalized group and the matched control group. The multivariable model (aHR) was adjusted for all variables presented in Table 1.

Stratification according to the follow-up period

The association between the risk of being diagnosed with a cancer and exposure (ICU-gr vs C-gr) was stronger in the first 3 months of follow-up, starting at the index date (aHR 1.65, 95% CI 1.45–1.88), compared to the rest of the follow-up period (aHR 1.21, 95% CI 1.11–1.33). This result was confirmed even when lung cancers were excluded from the multivariable analysis (Period 1: aHR 1.59, 95% CI 1.38–1.83; Period 2: aHR 1.18, 95% CI 1.07–1.30) (Table 3).

Table 3 Stratification according to the follow-up period for overall cancers and without lung cancers. The multivariable model (aHR) was adjusted for all variables presented in Table 1.

Stratification according to age and sex

The association between exposure and the risk of cancer was stronger in women compared to men (aHR 1.69, 95% CI 1.48–1.93, and aHR 1.20, 95% CI 1.10–1.30, respectively) and among individuals younger than 60 years old compared to older individuals (aHR 1.78, 95% CI 1.52–2.09, and aHR 1.22, 95% CI 1.12–1.32, respectively). The strongest association was found in women under 60 years old (aHR 2.15, 95% CI 1.65–2.80) (Table 4).

Table 4 Estimated risk of cancer diagnosis stratified by age (all, < 60 years, ≥ 60 years) and sex (all, male, female). The multivariable model (aHR) was adjusted for all variables presented in Table 1.

Occurrence of cancer according to cancer site

An analysis of cancer distribution according to cancer site is detailed in Table 5. The risk of being diagnosed with a cancer was significantly higher in the ICU-gr than in the C-gr regarding the following categories: renal cancer (aHR 3.16, 95% CI 2.33–4.27), hematological cancer (aHR 2.54, 95% CI 2.07–3.12), colon cancer (aHR 1.72, 95% CI 1.34–2.21), lung cancer (aHR 1.70, 95% CI 1.39–2.08), and other malignancies (aHR 1.18, 95% CI 1.04–1.35). Among the hematological cancer, ICU-gr had a significantly higher risk of being diagnosed with a leukemia (aHR 3.28, 95% CI 2.41–4.46), a myeloma (aHR 2.21, 95% CI 1.36–3.59), or a non-Hodgkin’s lymphoma (aHR 2.15, 95% CI 1.53–3.04), compared to the C-gr. No difference could be found between the two groups for the following cancers: Hodgkin’s lymphoma, melanoma, breast, prostate, rectal, liver, bladder, and uterine cancers (Table 5). Regarding the “other malignancies” category, more details can be found in the Supplement (Supplementary Table S2 online).

Table 5 Occurrence of cancer according to cancer site. The multivariable model (aHR) was adjusted for all variables presented in Table 1.

The same analysis was performed, but with the follow-up starting only after hospital discharge. The ICU-gr had then a 1.17 higher risk of being diagnosed with a cancer compared to the C-gr (aHR 1.17, 95% CI 1.08–1.26). The results showed a similar trend for each category of cancer, apart for myeloma (aHR 1.21, 95% CI 0.67–2.21) (Supplementary Table S4 online). A final analysis was performed taking into account the competing risk of death with the multivariable model. The ICU-gr had then a 1.25 higher risk of being diagnosed with a cancer compared to the C-gr (aHR 1.25, 95% CI 1.16–1.34). The overall categories showed a similar trend, apart from other malignancies category (aHR 1.13, 95% CI 0.99–1.29) (Supplementary Table S5 online).


This large population-based study included 41,302 individuals hospitalized in ICU due to SARS-CoV-2 infection (between February 15, 2020 and August 31, 2021) and 713,670 control individuals. Among these individuals, 2.2% of the ICU-gr was diagnosed with a cancer compared to 1.5% in the C-gr. Individuals in the ICU-gr had a 1.31 higher risk of being diagnosed with a cancer compared to the C-gr. The association was stronger by limiting the follow-up period to the first 3 months, and among women. The ICU-gr had a significant higher risk of being diagnosed with a renal, hematological, colon, or a lung cancer, compared to the C-gr. No significant differences were found for the other sites of cancers.

To the best of our knowledge, to date no studies have been conducted on this issue. However, studies with similar design aiming to assess the risk of cancer following other diseases, such as herpes zoster, have already been conducted. For example, a study conducted in the United Kingdom, using the General Practice Research Database (including 74,029 individuals), demonstrated the link between individuals having had herpes zoster and the risk of them being diagnosed with cancer in the following years26.

This study cannot conclude on a causal effect of a severe SARS-COV-2 infection on the risk of developing a cancer in the following months. Cancer screening and diagnosis may indeed have been different between the two groups, leading to a detection bias. Individuals hospitalized in the ICU-gr may have benefited from more lung scans, used as a screening tool for lung cancers, and from more repetitive blood tests that allowed screening of hematological diseases. On the other hand, screening by PSA or mammography may have been less frequent during the ICU stay or at discharge, as this was not necessarily a priority for these patients. For the control group, individuals were probably able to benefit from a better screening for certain cancers as they did not experience serious health events and were in better health condition to receive these screenings. However, since individuals hospitalized in ICU for a SARS-CoV-2 infection had a 31% higher risk of being diagnosed with a cancer in an average of 168 days following the index date, a severe SARS-CoV-2 infection may represent a marker of an underlying undiagnosed cancer, especially as the association with the risk of being diagnosed with a cancer was stronger in the first 3 months following hospitalization. Therefore, a more systematic screening could be more efficient during this period of time. It should also be noted that identical multivariate analyses were performed taking into account follow-up starting only from hospital discharge. These additional results showed a 17% increased risk of being diagnosed with a cancer in the ICU-gr compared to the C-gr, which underlines the fact that even when the follow-up does not include the hospitalization period, a similar trend is confirmed despite the possible detection bias previously described. Furthermore, multivariate analyses were performed taking into account the competing risk of death, highlighting a global similar trend with a 25% increased risk of being diagnosed with a cancer in the ICU-gr compared to the C-gr.

Regarding cancer sites, renal, hematological, colon and lung cancers were most likely to be diagnosed following a severe SARS-CoV-2 infection. While it may be more intuitive to understand why some type of hematological cancer might impact the immune system, it may be more difficult to understand the link between renal or colon cancer and higher frailty to SARS-Cov-2 infection. Nevertheless, some recent studies have already confirmed the immune dysfunction associated with renal and colon cancers27,28,29, as well as the fact that any type of cancer may promote immune dysfunction30. This could represent one explanation to our findings.

Strengths of the study

The main strength of this study is that the SNDS is a claims database that allowed us to analyze the risk of being diagnosed with a cancer from the comprehensive population without cancer history, thus limiting selection bias. Furthermore, a large number of individuals were included in the study, as the database includes the whole French population. In addition, all analyses were adjusted with a multivariable model to minimize confounding factors.

Limitations of the study

This study had several limitations. Firstly, the definition of a severe SARS-CoV-2 infection was limited to individuals hospitalized in ICU. However, this allowed us to focus on the most severe cases of SARS-CoV-2 infections. Secondly, information was potentially wrongly classified for certain variables (obesity, tobacco dependence, alcohol related disorders), which are significantly underestimated in this database. For instance, it is possible that some patients who smoke were misclassified as non-smokers in the database, thus underestimating this variable. However, this should not substantially modify the association between the risk of being diagnosed with a cancer and the group of exposure, except probably for obesity. Thirdly, we did not have information on the medication of residents in nursing homes, which have their own pharmacy, and therefore did not identify their comorbidities exhaustively. For this reason and knowing that many of these patients were not admitted to hospital during the first wave of SARS-CoV-2 pandemic because of the hospital restrictions in place at this time in France, we excluded this subpopulation. Finally, our study may also have been affected by residual confounding factors due to differences between the two groups, although matching and adjustment for a high range of comorbidities have been done.


In conclusion, this study is the first to suggest an association between severe SARS-CoV-2 infection and cancer diagnosis in the following months, suggesting that a severe SARS-CoV-2 infection may represent a marker of undiagnosed cancer. More research is needed to determine the nature of the relationship between an underlying cancer and a severe SARS-CoV-2 infection. Based on this future research, it would be necessary to discuss whether more targeted screening should be offered or not to this population of individuals.