Population-based cancer registry data resources, such as the National Cancer Registration and Analysis Service (NCRAS) in England [15], are a vital source of research data that are used across many research studies for epidemiological research; both in terms of monitoring trends in descriptive studies and to undertake comparative, analytical studies of risk factors and treatments. These same data resources are also increasingly used as an approach for long-term follow-up data for clinical trials [16, 17].

There will be an undoubted impact of the pandemic on these data and further the representativeness of the individuals that were added to these data resources due to the changes in clinical interactions during the pandemic. There have been reports of decreased incidence, a shift towards later stage cancer and changes in recorded patient information such as a higher proportion of unknown/not recorded cancer stage. The impact on cancer registration processes may also impact measures of cancer incidence, diagnosis, and survival at the population level.

There were also necessary changes to treatment and diagnostic practice over periods of 2020 and 2021 that will further impact on potential cancer patient survival prospects during this period and looking forwards in calendar time. Others have begun to explore the level of impact that this has had, but how to utilise these data resources in a consistent manner given the massive changes in recorded data is yet to be studied or discussed in detail. These reports of counts, descriptive statistics and estimates of short-term survival provide a critical starting point in understanding the changes to cancer incidence and mortality during the pandemic, but it is essential to also consider the impact on the estimation of long-term survival metrics, and how we fairly compare these metrics. These concerns are not limited to COVID-19 specific studies, but apply to any study using data from the period. For instance, international and risk group comparison studies utilise these electronic health databases to unpick and explain disparities, with the view to these being removed. To make fair comparisons, it is essential to account for differences due to the pandemic.

An obvious impact for cancer incidence monitoring is the potential displacement of cases from their hypothetical incidence date should the pandemic had not have happened. This has been evidenced across several cancer sites and countries [18,19,20,21,22], and furthermore some evidence of a ‘catching up’ of cases has also been observed in certain settings [5, 8]. The implications for monitoring of cancer incidence in general are not particularly stark in this case – it is obvious some care must be taken to account for variation across regions and countries – but it is likely that many of these cancer cases will still have been diagnosed at some point, with the exception of those who died due to COVID-19. However, studies using the years spanning the pandemic as part of their study period do need to consider the likely case-mix of individuals that still were diagnosed with cancer during those time periods and whether that has implications for their research question of interest. It is likely that those that would have been detected through screening, with a potentially better stage profile overall, contribute a proportion of the ‘delayed’ cases, although there is also strong evidence that those with symptomatic disease were also not being diagnosed through GP referral pathways to the same extent in England [23].

Even in a hypothetical scenario where cancer-specific mortality is unaffected by changes to healthcare delivery due to the COVID-19 pandemic, all-cause mortality may increase as a result of COVID-19 mortality among the population with cancer.

Cancer mortality rates among the whole population, as opposed to the population with cancer, may also have been affected. Cancer mortality rates rely on accurate recording of death certification to ascertain the underlying cause of death. For the most part, it is likely that deaths due to cancer would still be recorded consistently during this period. However, other causes of death, known as competing mortality, influence cancer mortality rates. For example, age and site-specific cancer mortality rates that decrease in 2020 may reflect the excess deaths in the population due to COVID-19, rather than a direct change in cancer mortality.

Given the real and artefactual changes to survival time discussed in this text, we must consider how to fairly compare and monitor cancer survival over time periods spanning the pandemic with metrics from pre and post-pandemic periods. In addition to the comparison of metrics over time, we must also consider how to make fair comparisons between population groups and other risk factor groups when using data spanning the pandemic period. Often our analyses concern data spanning several calendar years and hence the suitability of existing methods will require consideration for many years.

A key consideration when calculating cancer patient survival is the impact of delays in diagnosis caused by the pandemic. Care must be taken when interpreting changes in survival time during this period.

For example, the cessation of screening and other cancer services in many countries has resulted in diagnostic delays. Scenario A in Fig. 2 depicts what we may typically see in the absence on the pandemic. Scenarios B and C depict two potential changes in observed survival time for a delayed diagnosis. Scenario B illustrates a shorter observed survival time even if the delayed diagnosis does not affect the date of death. This is synonymous to lead time in a screening setting, but in reverse. It is well-known that lead time bias complicates the interpretation of survival-based metrics. Scenario C illustrates an even greater reduction in observed survival time in the case where delayed diagnosis negatively impacts the date of death. We would expect this to be the case if the delay in diagnosis (or access to treatment) resulted in disease progression to a more advanced stage with worse prognosis. It is necessary to consider that some of the shortening in survival time is potentially artificial (as in Scenario B), and will further worsen prognosis when comparing to unimpacted calendar periods or registries with a lesser overall impact from the pandemic.

Scenario A given in (a) illustrates a typical timeline from onset of disease to death. Scenarios B and C, given in (b, c), illustrate two potential changes in observed survival time due to a delayed diagnosis. Scenario B illustrates a shorter observed survival time even if the delayed diagnosis does not affect the date of death. Scenario C illustrates an even greater reduction in observed survival time in the case where delayed diagnosis negatively impacts the date of death.

As a result of delays in diagnosis and changes to healthcare delivery the stage at diagnosis of patients diagnosed in 2020 and 2021 must also be considered to fairly compare cancer survival. There is evidence of a shift in distribution of stage at diagnosis and together with potential sub-optimal treatment plans, marginal measures of cancer survival may not be a fair representation of overall cancer survival in the population. Stage-specific survival comparisons across calendar year and between populations can provide an overview of differences in survival, and the use of stage-standardisation can provide an approach to fairly compare survival across calendar year and between populations.

A study in Wales reported not only a shift in stage distribution from early stage to late stage, but also an increase in missing stage information for some cancer sites [24]. The authors explored possible reasons for the change in completeness of the recording of stage information including decreased diagnostic and pathology activity. England has also seen an increase in missing stage information for many cancer sites in 2019 due to reduced capacity in 2020 and 2021 when the data was being processed.

In addition to patients diagnosed in or shortly after March 2020, we must consider those diagnosed beforehand who were still receiving treatment. It is possible that these patients also experienced competing mortality due to COVID-19, and changes to their treatment plans and healthcare interactions. A reduction in the number of patients treated by adjuvant therapy is likely to increase recurrence rates and reduce survival [25], but this may take several years or more to be detectable. Changes in mortality rates among the whole population may affect estimates of cancer survival. This is discussed in more detail in the next section.

Cancer patient survival from population-based cancer registry data is typically reported using net survival metrics, and these are usually estimated in the relative survival framework [26,27,28]. Relative survival is estimated using the all-cause survival of cancer patients, whilst accounting for the expected survival of the population in the absence of cancer. The expected survival is obtained from mortality rates given in population lifetables. Relative survival estimates assume these mortality rates are derived from a population which reflects the survival likely to be seen by the cohort of cancer patients in the absence of cancer [29, 30]. The COVID-19 pandemic has resulted in larger expected mortality rates and disproportionately affected cancer patients. Hence, it is important to assess the need to adjust standard lifetables, and how to best make these adjustments. Individuals that were considered vulnerable to COVID-19 were strongly advised to shield for prolonged periods for reasons such as age and morbidities such as a cancer diagnosis [31, 32]. Individuals who shielded successfully may have been at a lower risk of COVID-19 mortality, whilst others may have been at an increased risk due to immune deficiency and the need to attend healthcare appointments. Hence, it is possible that cancer patients experienced different COVID-19 mortality and therefore different non-cancer mortality than the general population. If we believe patients with cancer experienced greater non-cancer mortality than the general population and standard lifetables continued to be used, this would lead to an underestimate of relative survival. Furthermore, older patients are at an increased risk of mortality due to both COVID-19 and cancer, potentially resulting in non-negligible bias from inappropriate lifetables [33]. If we believe patients with cancer experienced lower non-cancer mortality than the general population and used standard lifetables this could lead to an overestimate of relative survival.

It is necessary to investigate the suitability of existing lifetables spanning 2020-2021 and subsequent years, with the possibility of adjusted lifetables for this period improving estimates of cancer survival. Patients diagnosed prior to the COVID-19 pandemic may also be impacted by these lifetables in the net survival setting when analysing survival beyond the years immediately following diagnosis.

Characteristics such as age and socioeconomic status may influence the case-mix of patients. COVID-19 mortality rates vary by age and socioeconomic status, therefore removing a disproportionate number of elderly and more deprived patients from the denominator pool of patients at risk of cancer diagnosis. Hence, it is possible that the pool of patients diagnosed with cancer immediately post-COVID-19 may be somewhat younger and somewhat less deprived. It is possible that some populations may not see a full ‘catching up’ of expected annual cancer cases due to this.