Delayed care for patients with newly diagnosed cancer due to COVID-19 and estimated impact on cancer mortality in France

Background The impact of the first coronavirus disease 2019 (COVID-19) wave on cancer patient management was measured within the nationwide network of the Unicancer comprehensive cancer centers in France. Patients and methods The number of patients diagnosed and treated within 17 of the 18 Unicancer centers was collected in 2020 and compared with that during the same periods between 2016 and 2019. Unicancer centers treat close to 20% of cancer patients in France yearly. The reduction in the number of patients attending the Unicancer centers was analyzed per regions and cancer types. The impact of delayed care on cancer-related deaths was calculated based on different hypotheses. Results A 6.8% decrease in patients managed within Unicancer in the first 7 months of 2020 versus 2019 was observed. This reduction reached 21% during April and May, and was not compensated in June and July, nor later until November 2020. This reduction was observed only for newly diagnosed patients, while the clinical activity for previously diagnosed patients increased by 4% similar to previous years. The reduction was more pronounced in women, in breast and prostate cancers, and for patients without metastasis. Using an estimated hazard ratio of 1.06 per month of delay in diagnosis and treatment of new patients, we calculated that the delays observed in the 5-month period from March to July 2020 may result in an excess mortality due to cancer of 1000-6000 patients in coming years. Conclusions In this study, the delays in cancer patient management were observed only for newly diagnosed patients, more frequently in women, for breast cancer, prostate cancer, and nonmetastatic cancers. These delays may result is an excess risk of cancer-related deaths in the coming years.


INTRODUCTION
The recent coronavirus disease 2019 (COVID-19) epidemic has resulted in a massive mobilization of heath care systems for the management of patients requiring active treatment and intensive care. Resulting delays in the management of other diseases, in particular cancers, have been reported and identified as a potential risk factor for an increased rate of mortality of cancer patients with curable disease in the coming years. [1][2][3][4][5][6] The magnitude of this increased mortality is not well known.
An extensive number of articles have reported that delays in the diagnosis and management of patients with cancer are associated with an increased risk of death at an advanced stage, an increased risk of relapse, and death in the localized phase. 7,8 Reducing the delays in early management is a general strategy proposed to improve patient outcome in low-and middle-income countries, 9 as illustrated for breast cancer. 10 Public health strategies of nationwide screening for breast, colorectal, cervix, and lung carcinoma are based on the reduction of the risk of cancerrelated deaths with earlier diagnosis. [11][12][13][14] While delayed diagnosis and cancer treatment increase the risk of death due to cancer, with few exceptions such as indolent lymphomas or low-risk prostate cancers, [15][16][17][18] the magnitude of the impact of 1-6-month delay in the initial management of curable cancer patients on relapse and death due to cancer remains less clear, and varies considerably across studies for almost all cancer types. [19][20][21][22] Newly diagnosed cancer patients are particularly at risk of a negative impact of delayed diagnosis and treatment. [19][20][21][22] In addition, cancer patients are a population at risk of major complications and death due to COVID-19, in all countries, including France. [23][24][25] France has a national security system based on free health service including national screening programs. Unicancer is the French Federation of comprehensive cancer centers with an exclusive public activity gathering 18 centers treating w23% of all cancer patients in this country (unicancer.fr).
We investigated the impact of the COVID-19 pandemic and the first national lockdown in France on the number of patients consulting within Unicancer centers for a new diagnosis of cancer, according to the cancer type.

Unicancer: new patients per center in 2019 and 2020
Unicancer is the French Federation of Comprehensive Cancer centers. Unicancer includes 18 different hospitals in all regions (http://www.unicancer.fr/en/unicancer-group/ key-figures). Each Unicancer center was interrogated for the number of total and newly diagnosed patients consulting from 1 January to 31 July of 2019 and 2020, as well as those recorded from 2016 to 2018. Seventeen of the 18 centers were able to contribute to the study. The 18th center did not contribute because it was engaged in a merger precluding a relevant comparison of 2019 and 2020. The total number of cancers treated as well as that of breast, digestive, thoracic, gynecological, head and neck, urological, and hematological malignancies and general characteristics were collected.

Incidence of cancers in France
We collected information on the overall incidence of cancers in France from the French National Cancer Institute (INCA) website. In 2018, there were 382 000 new diagnoses of solid tumor cancers in France, and 157 400 cancerrelated deaths (41%) (https://en.e-cancer.fr/). Raw nationwide data from the publicly available health care system were also used as of 24 November 2020 26 to extrapolate the reduction in new cases attending Unicancer centers at the national level.

Analysis of the literature
We analyzed the published literature on PubMed describing the correlation between cancer treatment (surgery, radiotherapy, and chemotherapy) delay and survival. The keywords used for the enquiry on PubMed were 'delay', 'cancer', 'diagnostic', 'time to treatment', 'survival'. This research was conducted in November 2020 and we identified 3840 articles. These articles were further selected with manual screening focusing on delays for treatment initiation in localized phase. Meta-analysis, retrospective reviews, and trials were selected for all cancer types: breast, colorectal, lung, prostate, head and neck, ovarian, uterine, renal cell carcinoma, bladder, lymphoma, and leukemia; a selection is presented on Supplementary Table S1, available at https:// doi.org/10.1016/j.esmoop.2021.100134. 21, The level of increase in the risk of cancer death associated with 1 week to 6 months of delay (depending on the studies) was investigated. With the exception of studies in indolent lymphomas and low-risk prostate cancer, 15

Statistics
Comparisons of the different proportions and numbers were conducted using the chi-square test or the nonparametric ManneWhitney U test. In general, the chi-square test was used to compare the number of patients with different characteristics (e.g. sex) in 2020 versus 2019 (for the total first seven months of these 2 years, or for an individual month).
To calculate the potential impact of delays on the risk of cancer-specific deaths, we conducted three different analyses. . The increased risk of death related to 1 month or multiple months was calculated as follows: If the rate of death due to delay is 'RDdel'; if the rate of death without delay is 'RD'; HR corresponds to the HR of death related to a 1-month delay, while the HR related to an n-month delay is HR n . The risk of death due to delay RDdel can be calculated as follows: RDdel ¼ 1eexp [ln(1 À RD) Â HR n ]. If N is the difference between the number of newly diagnosed patients seen in 2019 minus the number of newly diagnosed patients seen in 2020 per month, the estimated excess number of deaths related to a delay of 1 month (NDexc) is therefore NDexc ¼ N (RDdel À RD) for each individual month of the study.
2. Second, we applied the same calculation with two extreme HRs of 1.02 and 1.1 with a similar standard death rate of 41% to describe different hypotheses. 3. Third, we then applied the same calculation described in the first point with (i) specific mortality rates for five individual tumor types [colorectal, head and neck carcinoma, bladder carcinoma, breast carcinoma, and lung carcinoma (as published in the meta-analysis 15 ] and (ii) specific mortality rates of these five individual tumors reported in this country in 2018 (e-cancer.fr).
The statistical analyses were conducted with SPSS 23.0 package (IBM, Paris, France).
The median reduction from January to July was 9% (ranging from 13% to 5%) versus 6% reduction (ranging from 13% to 3%) in the five centers in the higher versus lower COVID-19 incidence zone (U test, P ¼ 0.203).

Reduced incidence according to stage, gender, and cancer types
The reduction of new cancers in women was greater as compared with men, for both patients with and without metastasis ( Figure 2A). It was significant from January to July 2020 versus 2019, and specifically in April (chi-square test, P ¼ 0.03), May (P < 0.0001), June (P < 0.0001), and July (P ¼ 0.007; Figure 2A). This difference was exclusively observed between 2020 and 2019, in men as well as in women ( Figure 2B).
The proportion of reduction was greater for breast cancer versus other gynecological cancers (uterine or ovarian carcinoma) from January to July 2020 versus the same period in 2019 (P < 0.001). Similarly, it was larger for prostate cancers versus than other urological malignancies in the same period (P < 0.001).
We then assessed the disease stage (metastatic versus nonmetastatic) at diagnosis of newly diagnosed cancer

Estimation of the impact of delayed diagnosis on longterm patient survival
We then estimated the impact of delayed diagnosis and treatment occurring in this period on cancer-specific survival. The impact of the time to initiation of treatment on the survival of specific cancer histotypes and primary sites varies across studies (see Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2021.100134).
We first used the HR of 1.06 to estimate the impact of these delays on patient survival in the Unicancer network based on a recent meta-analysis. 20 This estimation included the reduced number of new patients from March to July 2020 versus 2019, per month, as well as the absence of compensation in the months following the first lockdown, and using a cumulative rate of increased risk of death for patients not consulting for several consecutive months. If we consider a À1.4%, À20.6%, À21.8%, À1.8%, and À11.3% decrease from March to July (Figure 1), respectively, the estimated additional number of deaths per month was 9, 83, 64, 4, and 14 from March to July, with a total of 174 additional cancer-related deaths within Unicancer centers. Nationwide, extrapolation of these estimates for the months of March to July lead to additional deaths per month of 42, 502, 393, 20, and 67 (i.e. an estimated total number of deaths due to delayed diagnosis from March to July 2020 of 1024; Table 2, see footnote a). These tables also include different hypotheses for the additional percentage of patients with delays (on the horizontal bars), and the number of additional months of delays (in the vertical axis, with 1 additional month of delay representing, for instance, all patients diagnosed and treated in September).
These estimates are dependent on three different parameters, namely, (i) the actual percentage of patients with delayed treatment, (ii) the number of months of delay, and (iii) the HR for a 1-month delay.
In a second step, we evaluated the impact of varying this HR. Table 2 shows estimates obtained using an HR of 1.06 ( Table 2, first section), a lower HR of 1.02 (Table 2, second section), or a higher HR of 1.1 ( Table 2, third section).
On the national social security website, 26 a reduction of newly diagnosed cancers of 23.3% for the first 7 months of 2020 was described as compared with 2019 (consulted by the authors on the national website 24 November). 26 This represents a further decrease of 15% in the number of cases

= ESMO Open
diagnosed compared with that measured in the Unicancer centers. Based on this national figure and Unicancer estimates, we then extrapolated nationwide excess mortality due to cancer ( Table 2) from 1872 to 9756 excess cancer deaths with the different HRs. Next, we calculated the number of cancer-specific deaths due to observed delays in Unicancer centers using the specific mortality rate (e-cancer.fr) and the HR of 1.06 per month of delays 20 for colorectal adenocarcinoma, head and neck carcinoma, bladder carcinoma, and lung carcinoma, and an HR of 1.08 for breast carcinoma. The calculated number of excess cancer-related deaths due to delays in the Unicancer network obtained with the 'tumor-specific' method and the 'one-size-fits-all' approach was very close (N ¼ 120 versus N ¼ 132, Table 3). Of note these five tumors accounted for a large proportion of the calculated excess cancer-related deaths aforementioned (132 of the 174).
It is important to note that these two estimates are based on the assumptions that from August 2020, all patients with delayed diagnosis would have been managed without further delay. Currently (as of 28 February 2021) the reported number of new patients seen in the Unicancer network in 2020 was 28 900 versus 28 955 from August to November in 2019 (the 2020 figures being 99.8% of that of 2019). No further reduction in new diagnoses was thus observed in this period, but no increased activity was measured that would have compensated the reduced activity during the lockdown.

DISCUSSION
In this work, we (i) present a description of the reduction of the number of new cases of cancers observed during the first lockdown in France, with an analysis on the type of tumors, sex, and stage, and (ii) propose an estimation of the potential impact on long-term cancer-free survival of patients impacted by these delays. We describe first the observed reduction in the number of cancer patients seen in the first seven months, including the first lockdown, of the first COVID-19 outbreak in France, from March to July 2020. This was done for 17 of the 18 comprehensive cancer centers in France (Unicancer centers) treating close to one-fourth of all cancer patients in France.
The total number of previously diagnosed cancer patients treated in the Unicancer centers increased in 2020 similar to levels observed in the previous year. Conversely from January to July 2020, there was a reduction of 6.8% of newly diagnosed cancers in this network, versus þ4.5% between 2019 and 2018. The reduction was detectable as early as March 2020, while an increase in the first two months of 2020 was recorded in agreement with the trends of the past years. This decrease reached >20% in the 2 months (April and May) of the first lockdown and was observed in July 2020 (À11%). It was not significantly different in centers located in regions with a high incidence of COVID-19 during this period.   The observed reduction was more pronounced in women, for both patients with and without metastases at diagnosis. It was also higher in breast cancers, prostate cancers, and skin cancers. The number of new patients presenting metastases decreased less than those with localized diseases.
These results show that only patients not previously diagnosed with a cancer were those who postponed their consultation to medical services during the first lockdown and in the following months. This was observed for not only tumors with national screening plans (breast, colorectal, cervix cancers), which were interrupted during lockdown, but also for other tumors, indicating that this is not only related to delayed access to screening. The reduced breast cancer screening during this period is likely to have contributed to the level of reduction in new breast cancer diagnosis. The fear of contracting COVID-19 and difficulties in accessing primary care centers or general practitioners are mentioned as possible causes. [1][2][3][4][5][6] The impact of delayed access to diagnosis and treatment for newly diagnosed patients with cancer on the risk of cancer-related death was the second question explored in this work.
A large number of studies have estimated the increased risk of death due to delays in diagnosis and for the initiation of treatment. 21, These studies are heterogenous in their methodology, and report a variable magnitude of impact across cancer types, depending on the timing of the delay and the nature of the treatment. A selection, by no means exhaustive, of studies is presented in the Reference section and in Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2021.100134. Delays, ranging from 7 days to >6 months, for surgery and delays to the administration of chemotherapy or radiotherapy 21, were all reported as being associated with a reduced cure rate and survival in both localized and advanced phases of a magnitude ranging from 0.5% per weeks of delay up to þ169% with 12 weeks of delays in similar cancers.
A recent meta-analysis conducted for breast, colorectal, lung, bladder, and head and neck cancers 20 showed that a 1-month delay in surgery is associated with an HR of 1.06-1.08 for the risk of cancer-specific death, 1.09 for delay in radiotherapy in head and neck cancer, and variable for adjuvant chemotherapy. In other cancer types, 1-month delays in surgery or neoadjuvant chemotherapy are associated with variable HR often superior in magnitude (summarized in Supplementary Table S1, available at https://doi. org/10.1016/j.esmoop.2021.100134).
Herein, we first selected an HR of 1.06 as a reasonable estimate matching the analysis reported, with the knowledge that lower and higher HRs are reported in various tumor types. The mortality rate for cancer of 41% reported in the national population in France was used for this calculation (e-cancer.fr).
With this uniform HR of 1.06, the number of excess cancer deaths calculated for the first months of 2020 was 1024. This number supposes that no further delays were observed after August 2020, and that the percentage of patients with delays was similar across France. These two hypotheses are, however, unlikely in view of the larger 23.3% reduction in number of new cases observed at the national level on the same period. 26 For these reasons, a more realistic number of excess cancer death of 5791 is calculated for the first 7 months of 2020. These estimates again suppose that all treatments were initiated shortly after July 2020 which is not known yet. The number of new patients observed between August and November 2020 was stable versus 2019 in Unicancer centers, but no compensation for the backlog of the first 7 months of the year was observed.
Differences in the percentages of patients with delayed diagnosis and treatment, as well as the number of months of delay, and the HR associated with delay all strongly influence the impact of COVID-19 on cancer-specific survival. An HR of 1.02 for a 1-month delay is below that generally reported for most cancer types, whereas a 1-month delay associated with an HR >1.1 is reported in several tumor types ( 21,27-55 and Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2021.100134). Excess cancer deaths calculated with these HRs were intended to provide only a possible estimate of the level of the impact of cancer-related deaths, not an accurate description.
Finally, these estimates do not take in account the yearly increase in new cancer diagnoses, which was þ0.9% from 2018 to 2019 on the nationwide social security platform.
The number observed in Unicancer centers thus represents an 'optimistic' evaluation of the percentage of patients with delayed diagnosis and treatments which may be greater in nonspecialized centers nationwide.
One of the limitations of this work is to have selected a unique HR for all cancer types. Ideally, a specific HR should be provided for each cancer type and for different stages, histotypes, and ages. The impact of delay per stage and cancer types was reported recently at the national level in the United States, 19 showing variable HR across different cancer types and stages. In this study, increased time to treatment initiation was associated with poorer survival for stages I and II breast, lung, renal, pancreas, and colorectal cancers, with HRs ranging from 1.005 to 1.030 per week of increased time to treatment initiation. 19 It is important to note that stage, age, and histotypes are by definition not documented in patients who did not attend to the hospital for diagnosis, and therefore stage-adapted calculations are not possible in this work. The recently reported online tool (OncCOVID) calculates the risk of death due to cancer depending on stage, age, tumor type, and duration of delay to treatment. 56 While accurate for a single patient, this tool may not be applied to this work for the same reasons, and also because the delays for patient attending the Unicancer centers were outside of our scope. Importantly, the COVID-19-specific cause of death [23][24][25] is not calculated in this present work which focuses on cancerspecific deaths.
However, we tested cancer-specific HR for five different cancers. Interestingly, the excess mortality calculated with these five tumor-specific parameters was very similar, inferior only by <10% (120 versus 132) to those obtained with the unique mortality rate of 41% and HR of 1.06, as shown in Table 3. Because these five tumors were shown to contribute to 132 of the 174 deaths in excess due to delays, this observation supported the rough estimate of the excess cancer deaths calculated in this work.
Altogether, these results indicate that cancer patients attended hospital consultations for initial diagnosis and treatment with delays in this period, even in comprehensive cancer centers. This was observed for the majority of cancer types, more frequently in women, in breast cancers, and for prostate cancer in particular. It was more frequently observed for nonmetastatic patients. It will result in a significant increase in the number of cancer death in the future for which an estimation is provided in this study. Maintaining access to rapid diagnostic and treatment procedures is an important medical priority in the months to come before the normalization of the health care system expected following the implementation of vaccination.