National Cancer Diagnosis Audits for England 2018 versus 2014: a comparative analysis

Background Timely diagnosis of cancer in patients who present with symptoms in primary care is a quality-improvement priority. Aim To examine possible changes to aspects of the diagnostic process, and its timeliness, before and after publication of the National Institute for Health and Care Excellence's (2015) guidance on the referral of suspected cancer in primary care. Design and setting Comparison of findings from population-based clinical audits of cancer diagnosis in general practices in England for patients diagnosed in 2018 or 2014. Method GPs in 1878 (2018) and 439 (2014) practices collected primary care information on the diagnostic pathway of cancer patients. Key measures including patient characteristics, place of presentation, number of pre-referral consultations, use of primary care investigations, and referral type were compared between the two audits by descriptive analysis and regression models. Results Among 64 489 (2018) and 17 042 (2014) records of a new cancer diagnosis, the percentage of patients with same-day referral (denoted by a primary care interval of 0 days) was higher in 2018 (42.7% versus 37.7%) than in 2014, with similar improvements in median diagnostic interval (36 days versus 40 days). Compared with 2014, in 2018: fewer patients had ≥3 pre-referral consultations (18.8% versus 26.2%); use of primary care investigations increased (47.9% versus 45.4%); urgent cancer referrals increased (54.8% versus 51.8%); emergency referrals decreased (13.4% versus 16.5%); and recorded use of safety netting decreased (40.0% versus 44.4%). Conclusion In the 5-year period, including the year when national guidelines were updated (that is, 2015), there were substantial improvements to the diagnostic process of patients who present to general practice in England with symptoms of a subsequently diagnosed cancer.


INTRODUCTION
Timely diagnosis of cancer is a public and policy concern in many high-income countries worldwide, as longer time to diagnosis is associated with worse clinical and patient-reported outcomes. 1 Of patients diagnosed with cancer, >90% present with symptoms and most of them consult with a GP in the first instance. 2,3 As a consequence, successive UK cancer strategies have included early diagnosis initiatives specifically targeted at primary care, including the creation of urgent referral pathways, together with clinical guidelines supporting their use, 4 the development of clinical decision support tools, 5 greater use of safety netting, 6 and easier access to specialist investigations. 7 In 2015, guidance from the National Institute for Health and Care Excellence (NICE) on the management and referral of patients with suspected cancer was extensively revised and, for the first time, explicitly specified a 3% risk threshold (for as-yetundiagnosed cancer) for urgent referral. 8,9 Before 2018, two national populationbased audits of cancer diagnosis in primary care were undertaken in England; these covered patients diagnosed in 2009-2010 10 and 2014. 11 These audits provided unique insights into the diagnostic processes of cancer in primary care, identifying patient groups at greater risk of prolonged diagnostic intervals (defined as the time from first relevant presentation to primary care until diagnosis date) and providing benchmarks by which improvements in patient care could be assessed. In 2019-2020, a third national audit took place, covering patients diagnosed in 2018.
Using the same methodology and, for the most part, questions identical to those used in the 2014 audit, the 2018 audit aimed to characterise the quality of the diagnostic process for patients diagnosed with cancer in 2018, following the revision of NICE guidance in 2015. In this article, the principal findings of the 2018 audit are reported and compared with those from 2014. The authors' hypothesis was that the implementation of NICE guidance should be associated with an observable change in the audited process measures. Data are presented here in a format designed to facilitate a 'side-by-side' comparison with the previously reported findings of the 2014 audit.

METHOD
Data were collected using the same system as the 2014 audit. 11 Briefly, all incident malignant cancer cases among England residents in 2018 12 were assigned to the Abstract Background Timely diagnosis of cancer in patients who present with symptoms in primary care is a quality-improvement priority.

Aim
To examine possible changes to aspects of the diagnostic process, and its timeliness, before and after publication of the National Institute for Health and Care Excellence's (2015) guidance on the referral of suspected cancer in primary care.

Design and setting
Comparison of findings from population-based clinical audits of cancer diagnosis in general practices in England for patients diagnosed in 2018 or 2014.

Method
GPs in 1878 (2018) and 439 (2014) practices collected primary care information on the diagnostic pathway of cancer patients. Key measures including patient characteristics, place of presentation, number of prereferral consultations, use of primary care investigations, and referral type were compared between the two audits by descriptive analysis and regression models.

Conclusion
In the 5-year period, including the year when national guidelines were updated (that is, 2015), there were substantial improvements to the diagnostic process of patients who present to general practice in England with symptoms of a subsequently diagnosed cancer. In line with research guidance, 13 the authors reported on the following timescales using the definitions given:

Keywords
• patient interval (PI) -the number of days between symptom onset and the first presentation to a health professional; • primary care interval (PCI) -the number of days from the first relevant presentation to the date of first referral; • diagnostic interval (DI) -the number of days from first relevant presentation to the date of diagnosis; and • the number of days from referral to the date the patient was informed that they had cancer by a specialist.
Interval times of <0 days and >730 days were excluded, consistent with previous literature. 14 The distribution of gender, age, stage at diagnosis, and cancer site of the NCDA cohort was compared with the 2018 national cancer registration statistics. 15 Participating and non-participating practices were compared in terms of their key characteristics and key aspects of patients' experience of primary care, similarly to Swann et al. 11

Regression
To account for potential confounding between other patient-level exposure variables, the relationship between NCDA year, and binary diagnostic process variables was assessed using mixed-effects logistic regression (reference: 2014 audit), with a random effect for general practice. As the continuous diagnostic process variables, pre-referral consultation number, PCI, and DI have a skewed distribution, quantile regression was used as commonly used in relevant

How this fits in
There is ongoing national monitoring of elements of the cancer-referral process from primary care, including the proportion of urgent referrals ('2-week waits') and emergency presentations. The 2014 National Cancer Diagnosis Audit provided a richer picture of this process, as reported by GPs themselves. This research presents a direct comparison of that audit to the more recent one carried out on patients diagnosed in 2018, with revised National Institute for Health and Care Excellence guidance on referral of suspected cancer having been published in 2015. It shows improvements in practice over time such as fewer pre-referral consultations and a shortened time to referral and diagnosis. Crude and adjusted models are reported. Models were adjusted for gender, age group, ethnicity (Asian, Black, mixed, other, White, and not known), IMD deprivation quintile, and cancer site. Ethnicity data from the National Cancer Registration and Analysis Service (NCRAS) were used, as these were available for both audits. Individuals with 'not known' studied diagnostic process measures were excluded from respective analyses. As it was not mandatory to report variables for screening cases in the 2014 audit, the NCRAS-verified screening cases (n = 3896) were excluded from the 2018 audit prior to regression analysis so a direct comparison could be made.
In order to examine whether there was selection bias in the included audit cancer cases, the authors carried out a sensitivity analysis of the characteristics of practices that completed <95% versus ≥95% of their NCDA cases. In that analysis, the authors aimed to include a random effect for the practice in order to examine the likely influence of clustering of observations by general practice. Quantile regression models (outcome being interval length) did not converge when including a practice random effect, however, so the sensitivity analysis that included a practice random effect was carried out after dichotomising the PCI and DI (above/below the median and 90th percentile) and using mixedeffects logistic regression models. Analyses were carried out using RStudio (version 2021.09.1+372) and R (version 4.1.2).  Figure S1; in total, in 2018, 15.3% of patients were diagnosed with prostate cancer compared with 12.5% in the 2014 audit. The patient cohort in 2018 was similar to the cancer registration data from the same year for age, gender, cancer type, and stage (Supplementary Figure S2).  Table S12). The proportion of routine referrals was 5.1% and 5.4% points lower in 2018 than in 2014 for patients with multiple myeloma (9.2% versus 14.3%) and prostate cancer (6.7% versus 12.1%).

Intervals and avoidable delays
The percentage of patients with sameday referral (denoted by a PCI of 0 days) There was large variation in DI by cancer site, consistently, in both audits. The overall decrease in intervals was driven by a shortening of intervals of those cancer sites that had longer-than-average intervals in the 2014 audit, such as multiple myeloma (47 days versus 54 days) or stomach cancer (32 days versus 42 days).
Cases with GP-ordered investigations had longer median PCIs compared with those without (10 days versus 0 days), which was shorter than in the 2014 audit (15 days versus 0 days). For patients who had investigations, compared with those who did not, the DI was longer in the 2018 round (49 days versus 20 days), although this was not as great as the difference observed in the 2014 audit (57 days versus 22 days).
The overall median PI in 2018 (not measured in 2014) was 2 days; however, there was substantial variation by cancer site, with the median PI for oesophageal and rectal cancers approaching 2 weeks. The median interval from referral to the date the patient was informed they had cancer by a specialist was 33 days (interquartile range [IQR] 20-55 days) in the 2018 audit, which was longer than in 2014 (28 days, IQR 16-49 days).
GPs perceived an avoidable delay to a patient's diagnosis in 15 060 (23.4%) cases (Supplementary Table S15). The proportion of patients with a perceived avoidable delay occurring at any one of the pre-presentation, primary care, or postreferral phases ranged from 10.5% to 12.4%. Patients aged 0-24 years had more reported delays between presentation and referral (17.2%), and there was variation by cancer site. For breast cancer, avoidable delay was reported in 9.2% of patients before presentation, and 2.7% and 3.2%, respectively, of patients between presentation and referral, and after referral; however, for patients with rectal cancer, these figures were 20.7%, 14.2%, and 11.9%, respectively. The data on avoidable delays are not directly comparable with the 2014 audit, because of changes in the audit question.

Direct comparison of 2018 and 2014 audits
Regression modelling confirmed that, after accounting for possible confounding between exposure variables, all observed changes in aspects of the diagnostic process and intervals prevailed (Table 3), except for the median reduction of 3 days in the PCI to 0 days (that is, no change after adjustment). This is a result of zero-inflated distributions causing a lack of resolution at the median, and should be interpreted alongside the increase in the observed proportion of same-day referred patients with a PCI value of 0 days (see above), and the reduction after adjustment at the 90th percentile (observed -19 days, adjusted -16 days). In 2018, there was a statistically significant reduction (P<0.001) in patients who had ≥3 pre-referral consultations, safety netting, and emergency referrals (odds ratios [ORs] 0.70, 0.76, and 0.83, respectively) and a statistically significant increase (P<0.001) in patients who had any investigations, blood tests, or a TWW referral (ORs 1.1, 1.2, and 1.2, respectively) ( Table 3). Adjusted quantile regression results showed a median difference of 3 days in the DI in 2018 from 2014. A sensitivity analysis, allowing a random effect for the practice, also provided concordant findings (Supplementary Table S16).

DISCUSSION Summary
For patients diagnosed with cancer in 2018, compared with those diagnosed in 2014, most diagnostic process measures had improved, with notable reductions in PCI length and the proportion of patients experiencing multiple pre-referral consultations. Patients diagnosed in 2018 were statistically significantly more likely to have been referred urgently for suspected cancer (TWW) and less likely to have had an emergency referral compared with those in the 2014 audit. The modest reduction in the DI, compared with the substantial reduction in the length of the PCI, suggests considerable progress in primary care diagnostic processes, not matched by capacity increases in secondary care diagnostic pathways. The apparent increase in the interval between referral and the date the patient was informed of their cancer status should be interpreted cautiously as both audits predate the formal implementation of this measure.
Use of safety netting was advocated in 2015 by the Independent Cancer Taskforce; 6 however, recorded use of safety netting was lower in 2018 than in 2014. Further discussion points are included in Supplementary Box S3.

Strengths and limitations
This was a large-scale population-based audit with more than one-quarter of practices in England participating, and cases identified using the English cancer registration system. Detailed information was added from the primary care patient record by practice staff, able to read freetext entries, and able to apply their clinical judgement when completing the audit data fields.
Both audit rounds used similar methods for data collection, enabling -for the most part -between-round comparisons. Further, the characteristics of audited patients and participating practices were highly representative of the national incident cohort in respect of sex, age, and cancer site, and of all English practices in both rounds. There were minor differences in ethnicity case mix reflecting both general demographic trends and the greater participation in the 2018 audit of London practices, and a greater proportion of prostate cancers in 2018 in England. 18 In spite of similarities in terms of other characteristics, there is still potential for differences in the diagnostic processes between participating and nonparticipating practices, for example, practices interested in cancer diagnosis and care -which, therefore, may be more likely to be performing well -may be overrepresented among participating practices.
In principle, some of the observed changes in the key outcomes of the diagnostic process (the percentage of patients who were referred at first consultation, and the median length of the PCI and the DI) may reflect changes in the distribution of stage at diagnosis between 2014 and 2018. However, answering this question requires causal inferences to be made, which is challenging because of the changing completeness of data on stage at diagnosis, and because the changes in the diagnostic-process outcomes may, themselves, mediate changes in the stage distribution. Nonetheless, as the betweenaudit waves increase in stage completeness was relatively small, and as the distribution of stage among observed cases was highly similar between the two audits, the findings are consistent with the hypothesis that most of the observed changes in the examined diagnostic outcomes are unrelated to stage distribution and reflect changes in diagnostic processes per se.
The reduction in safety netting could reflect fewer patients consulting ≥3 times pre-referral and should be interpreted with caution, as safety-netting advice may be under-recorded. 19 Comparison with existing literature To the authors' knowledge, international experience with population-based audits of the diagnosis of cancer in primary care is limited. Population-based audits have been conducted in sub-national populations of different sizes previously in England, Scotland, and Denmark. 11,[20][21][22][23] The findings presented here concord with

Funding
The National Cancer Diagnosis Audit received enabling support from Cancer Research UK, NHS England, and the National Cancer Registration and Analysis Service. Gary A Abel was supported by the National Institute for Health Research (NIHR) Applied Research Collaboration South West Peninsula (NIHR200167). Greg Rubin is chair, Georgios Lyratzopoulos is associate director, and Gary A Abel is co-investigator of the CanTest research collaborative funded by Cancer Research UK (reference: C8640/A23385). Georgios Lyratzopoulos is supported by an Advanced Clinician Scientist Fellowship by Cancer Research UK (reference: C18081/A18180). The views expressed in this publication are those of the author(s) and not necessarily those of the NIHR, the Department of Health and Social Care, or any funder.

Ethical approval
This study uses data collected as part of a clinical audit and collated by the National Cancer Registration and Analysis Service under regulation 2 of the Health Service (Control of Patient Information) Regulations 2002.

Data
Data for this audit are based on patient-level information collected by the NHS, as part of the care and support of patients with cancer. The data are collated, maintained, and quality assured by the National Cancer Registration and Analysis Service, which is now part of NHS Digital (it was part of Public Health England during the data-collection phase of the audit). Data for the National Cancer Diagnosis Audit are held by NHS Digital, and can be made available for approved users through the Data Access Request Service: https://digital.nhs.uk/services/data-accessrequest-service-dars. other evidence indicating secular trends in increasing numbers of TWW referrals and a decreasing use of emergency diagnosis. 24,25 In comparison with the 2018 Scottish NCDA, 26 patients in England had fewer multiple consultations (19% versus 25%), more urgent referrals for suspected cancer (55% versus 43%), fewer emergency referrals (13% versus 19%), fewer primary care-led investigations (48% versus 51%), and fewer reported avoidable delays (23% versus 33%).
The 2018 audit also included information on the PI, as recorded in the GP record. This measure was not included in the 2014 audit, but was collected in 2009-2010; 27 compared with the results of the 2009-2010 audit, there has been a substantial reduction in the measured median PI (from 10 days to 2 days). There are wellrecognised difficulties in measuring intervals between symptom onset and presentation (that is, patient intervals) in population studies; 28 nonetheless, the pattern of variation arising from the two audits is highly similar -oesophageal, rectal, and oropharyngeal cancer had some of the longest PIs, whereas bladder and kidney cancer had some of the shortest.
The diagnostic pathway and routes to diagnosis of patients with lung, colorectal, and ovarian cancer have also been described in bespoke samples of patients in several jurisdictions. [29][30][31] Our findings concord with the overall pattern of variation in diagnostic intervals observed between different cancers, but provide a wider appreciation of key measures of the diagnostic process in a larger populationbased sample including all cancer sites, albeit in a single jurisdiction.

Implications for practice
Variation in PI by cancer site may be useful in directing symptom awareness interventions. The proportion of patients urgently referred with suspected cancer rose from 51.8% to 54.8% between the two audit rounds. Although diagnosis after such a referral is associated with earlier stage and better outcomes than emergency presentation, the continued rise in the use of this pathway may become unsustainable without sufficient uplift in capacity in diagnostic services to meet this rising demand. 32,33 Moreover, the optimal referral rate per practice is unknown and health economic data to underpin its provision are lacking.
The use of diagnostic tests in primary care in 2018 had changed very little from 2014, despite the recommendations made by NICE in 2015. Some diagnostic tests (for example, faecal immunochemical tests) were used in only a small percentage of cases. Whether these findings represent under-use of this test, and how their use may have changed since 2018, needs to be further examined.
Regarding the longer PCI and DI for patients who had primary care ledinvestigations, it should not be assumed that primary care testing necessarily prolongs the diagnostic process -lack of primary care testing, particularly among patients with non-specific symptoms, could have resulted in even longer intervals in the absence of information from these tests. In 2018, there was only a small increase in the proportion of patients with diagnostic tests initiated in primary care, compared with 2014. This was despite the government's commitment, in 2015, to spend an additional £300 million on diagnostic tests by 2020, 34 which may have been expected to support and encourage more primary care direct access and diagnostics use. Patients with primary care-led investigations had longer PCIs and DIs than those without. It should not be assumed that primary care testing necessarily prolongs the diagnostic process -lack of primary care testing, particularly among patients with nonspecific symptoms, could have resulted in even longer intervals in the absence of information from these tests.
The organisation and funding of largescale audits such as NCDA is challenging, requiring substantial resources for informatics support, as well as general practice engagement and facilitation. The COVID-19 pandemic, which post-dates the findings reported here, has not only had a considerable impact on patients' responses to symptoms of possible cancer, the nature of GP consultations, and referral patterns, but also facilitated the greater sharing of primary care data. 35 Further audit rounds would be invaluable for the richness of insight they could provide; greater data availability could be leveraged by incorporating elements of newly available data as part of a routine standing audit of cancer diagnosis.

Provenance
Freely submitted; externally peer reviewed.