This study is a population-based cohort study using routinely collected electronic health records from UK Clinical Practice Research Datalink (CPRD) GOLD. CPRD GOLD contains anonymised patient-level information on demographics, lifestyle data, clinical diagnoses, prescriptions and preventive care contributed by general practitioners from the UK (33). The use of CPRD data for this study was approved by the Independent Scientific Advisory Committee (22_002331). This database has previously been mapped to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) (34). People were eligible if they were registered between January, 2017 and December, 2021 with at least one year of prior clinical history. Additional criteria for the incident cancer diagnosis cohorts were including individuals who had a diagnosis or record of cancer, specifically for breast, colorectal, lung, or prostate cancer; excluding individuals diagnosed with the same type of cancer at any point in their clinical history and excluding those with metastases. We did not specifically interrogate (or exclude on the basis of) whether individuals had intact organs.

The ‘exposure’ was the date of the first UK national lockdown (23rd March, 2020), which was used to dissect the full study period into three distinct time-periods: pre-pandemic (January, 2017 to February, 2020), during lockdown (March, 2020 to June, 2020), and post-lockdown (July, 2020 to December, 2021). Additionally, we further dissected the extended post-lockdown periods distinguished by the changing social restrictions according to the specific dates shown in Figure 1, and lockdown periods categorized as follows: Lockdown (March, 2020 to June, 2020); post-first lockdown (July, 2020 to October, 2020); second lockdown (Nov, 2020 to Dec, 2020); third lockdown (Jan, 2021 to March, 2021); easing of restrictions (April, 2021 to June, 2021); and most legal restrictions removed (July, 2021 to December, 2021).

For aim 1, frequencies of screening/diagnostic tests and referrals relevant to each cancer were selected: they constitute the primary tools used in the cancer diagnostic pathways in the UK (see Supplementary Table S1, Additional File 1). For aim 1 and 3, IR of cancer diagnoses included first-ever (incident) diagnoses of breast, colorectal, lung and prostate cancer. For aim 2, cancer patients were characterised on all comorbidities and medication usage available within CPRD GOLD across the study periods. All diagnoses, observations, measurements, procedures and medications were defined based on SNOMED/Rx Norm/LOINC codes (as appropriate), in the OMOP-mapped data. A list of all codes used to define each outcome can be found in our associated shiny app: https://dpa-pde-oxford.shinyapps.io/CancerCovid_CohortDiagnosticsShiny_paper1/

Overall, there were 5,191,650 people eligible to be included in the denominator population from January, 2017. Total counts of patients excluded after applying the exclusion criteria for incidence estimates, are shown in Supplementary Table S2. The population structure of CPRD GOLD, in terms of age and sex was similar across the three time-periods (see Supplementary Tables S3, S4), but the proportion of practices from the different regions in the UK changed over time with fewer practices in England and greater proportion in Scotland during and after lockdown (Supplementary Figure S3). Demographics and total number of patients registered in CPRD in each of the time-periods, and diagnoses of first-ever breast, colorectal, lung and prostate cancer, are shown in Table 1. Mean age at date of diagnosis and sex distribution of cancer patients were largely the same across the three lockdown periods. Interactions with the healthcare system and routes to diagnosis were substantially reduced for patients receiving their cancer diagnosis during lockdown compared to those diagnosed pre-pandemic. Patients diagnosed after lockdown had fewer interactions with the healthcare system than pre-pandemic, though to a lesser extent than those diagnosed during lockdown (see Supplementary Tables S4–8). Across the lockdown periods there were no notable differences in comorbidities and medication prescriptions for those diagnosed with breast, colorectal, lung or prostate cancer (see Supplementary Tables S9–S12).

The findings of this study revealed a reduction in number of routinely performed screening, diagnostic tests, and referrals during the period from March, 2020 to December, 2021 compared to data from January, 2017 to February, 2020. Particularly during the first lockdown, there was a substantial decrease in mammograms, sigmoidoscopies, colonoscopies, and visits to breast surgeons by 77%, 70%, 65%, and 63% respectively, compared to pre-pandemic rates. Similar findings were reported in other countries, such as Slovenia (11) and Argentina (39).

Although some rates of screening, diagnostic tests, and referrals increased in the post-first lockdown period, they remained below pre-pandemic levels. For instance, mammograms, colonoscopies, and visits to breast surgeons were still reduced by 26%, 21%, and 80% respectively between July, 2021 and December, 2021. These findings contradict studies from Catalonia, Spain, and Canada, where mammograms and colonoscopies returned to expected levels by December, 2021 (18, 31). Similarly, data from Canada shows that breast cancer screening returned to pre-pandemic levels by December, 2020; and faecal occult blood tests for colorectal cancer by September, 2020 (7). In the UK, the data from CPRD GOLD did not show a recovery to pre-pandemic levels for screening and diagnostic tests. Possible explanations include the fact that the UK was the only European country to have additional lockdowns after the first, and that the NHS has experienced staff shortages and strikes over recent years impacting on its capacity to catch up.

Lockdown had varying effects on screening and referral procedures compared to diagnostic procedures in CPRD GOLD. Diagnostic procedures were not deprioritized after lockdown, except for visits to breast surgeons, indicating efforts to reduce the backlog. Screening, on the other hand, was more susceptible to postponement or lower prioritization, as it is used for asymptomatic individuals, as shown in the data.

There are multiple explanations for the persistent reductions in screening, diagnostic tests, and referrals during extended lockdown periods. Variations in screening reductions may be related to the prevalence of COVID-19 restrictions/infections across countries (4). Reports indicate that the UK’s response to the pandemic was inadequate, resulting in a significant impact on the country and the need for subsequent lockdowns. The UK faced high infection rates, hospitalizations, and a substantial death toll (40). Healthcare resources were diverted from standard care, affecting cancer diagnostic pathways until December, 2021. Although rates showed some increase from March, 2020 to December, 2021, they were inconsistent. Data from a systematic review predicted a clearance of the screening backlog (specifically mammograms) in the USA within 12-24 weeks (4, 41), whereas our data suggest that even after 52-73 weeks, the queue was not cleared in the UK. That said, changes in screening methods (such as a switch from direct appointments to open invitations for routine mammograms) may have affected number of patients screened at least for breast cancer across this time period (42, 43).

Reduced screening and diagnostic tests lead to decreased cancer detection and diagnosis. Breast, colorectal, and prostate cancers were significantly underdiagnosed during lockdown and remained below expected levels until June, 2021 for breast cancer and until December, 2021 for prostate cancer. The expected effect of these reductions in rates is that diagnoses will be delayed, and prognosis worsened by these backlogs in diagnosis and treatment. These findings contradict a study from Catalonia, Spain, where breast cancer diagnoses recovered to pre-pandemic levels within this time-frame (18, 31). Belgium’s cancer registry data also showed recovery by June, 2020. Although incidence rates for colorectal and lung cancer returned to pre-pandemic levels, these rates likely represent missed diagnoses during lockdown, requiring substantial catch-up to compensate for the shortfall.

Our model predicts that prostate cancer had the highest number of missed cases, with an estimated 21,525 (25%) of expected cases missed from March, 2020 to December, 2021. Similar reductions in prostate cancer diagnosis were observed in other studies (18). Lung cancer was the least affected, with 14.2% of expected cases missed, which aligns with other reports (18, 31). It is possible this is because we have limited screening tools for lung cancer, leading to comparatively smaller diagnosis rates compared to other cancers. Though the increased use of chest radiography during COVID-19 infections may have inadvertently led to the identification of potential lung cancer symptoms and subsequent diagnosis (18). Stratification analyses revealed consistent underdiagnosis in specific age groups: women aged 60 to 79 for breast and colorectal cancer, men aged 60 to 79 for lung cancer, and men aged 40 to 59 years for prostate cancer. It is already known that we see a steep rise in risk for these cancers from these ages onwards (1, 44). These findings emphasize the urgency of prioritizing screening and diagnostics in these populations to detect the missed cases.

This study benefits from the strengths of CPRD GOLD, known for its extensive UK population coverage and comprehensive healthcare records (33), facilitating thorough phenotyping of screening, diagnostics, and cancer cases. The longitudinal nature of the database enabled an extended observation period beyond the typical one-year post-lockdown timeframe. Unlike most studies, our analysis covers screening and diagnostic rates up to December, 2021. In addition, our study uses innovative time-series forecasting to estimate the shortfall in cancer diagnosis rates, whilst the majority of research investigating the ramifications of COVID-19 on screening, diagnostics and incidence rates are descriptive in nature. Further research should explore additional data to assess if the UK has fully recovered from the rate shortfalls.

Although this study has many strengths it does have some limitations. First, as these data are derived from primary care and not linked to cancer registry data there were many screening and diagnostic tests of relevance to this study that were not captured in the database. This is common of studies using primary care data, as many diagnostic tests and procedures occur in hospital settings. Furthermore, cancer diagnoses may have shifted to hospital settings during the pandemic, and there may be a time-lag in recording cancer diagnoses in primary care records. Thus, it is likely that the estimated shortfall in screening/diagnostic tests, and cancer diagnosis rates in the present study, are underestimated. Relatedly, it would be informative to obtain information on cancer stage at time of diagnosis in order to determine the impact of the pandemic on disease severity. Linkage to cancer registry data would enable this analysis, but is beyond the scope of the current report. Similarly, it would be informative to investigate how the pandemic impacted on time to treatment. Not surprisingly, other reports have indicated that delayed treatment initiation was associated with increased mortality rates at 5- and 10-years after diagnosis, for all the cancers of focus here, particularly colorectal cancer (45). Linkage to cancer registry with treatment and cause of death data would allow us to replicate such analyses. That said, our own analyses have indicated that the pandemic impacted negatively on colorectal survival, with reductions in survivorship for patients diagnosed during the pandemic equivalent to returning to mortality seen in the first decade of the, 2000s (46). Second, the composition of patients and practices in the database have changed over time. Indeed, with the advent of the CPRD AURUM database, some practices were transferred out of GOLD and into AURUM, thus accounting for the reduced source population counts across time-points. Reassuringly, the IR of the cancers in the three broad time-periods across regions were largely similar, except for slightly smaller IR in England across some time-points, likely reflecting the change in population composition. Thirdly, the generalizability of findings is predominantly limited to Scotland and Wales, with less representation from England and Northern Ireland. Finally, as real-world evidence, causal inference is challenging, and other factors could have influenced the reduction in cancer diagnoses during lockdown, such as pre-existing trends in screening/diagnostic tests and cancer diagnoses, seasonality patterns, or COVID-19-related deaths. Whilst we did observe a pre-existing downward trend for visits to breast surgeons, there were no other observed pre-existing trends in our data, and our modelling statistically accounted for seasonal variability. However, data from Catalonia suggest a small proportion of missed diagnoses were attributed to COVID-19 deaths (47).

Delays in diagnosis are likely to impact on cancer stage at time of diagnosis, treatment initiation, mortality rates and total years of life lost. To effectively tackle the existing backlog and potential long-term consequences on cancer survival, it may be necessary to implement strategies to identify those potential ~62,000 cancer cases missed. These could include raising public awareness through targeted campaigns aimed at particular age groups that have been most affected, encouraging participation in screening programs, and enhancing the coordination between primary care facilities and hospitals. Increases in screening and diagnostic testing may need to be increased in the months following December, 2021 to account for the observed shortfall in the UK. These measures are vital for effective public health intervention and reducing the impact of delayed diagnoses on cancer outcomes.