Background
National cancer plans and cancer control strategies increasingly incorporate indicators of stage at diagnosis, given its strong prognostic implications, and thanks to the increasing availability of high-quality and complete stage information in population-based cancer registries. Improvements in early diagnosis can potentially be supported by interventions targeted for groups with worse stage at diagnosis.

Objectives
To describe inequalities in stage at diagnosis for 21 cancer sites currently staged by the English National Disease Registration Service (and forming the current national indicator for early stage at diagnosis proportion in England), and estimate the potential for reducing advanced-stage diagnosis by removing inequalities between socio-demographic groups.

Methods
The English National Disease Registration Services provided data on all registrations for any of 21 cancers (in order of incidence: prostate, breast, lung, colon, melanoma, rectum, non-Hodgkin lymphoma, kidney, pancreas, oesophagus, bladder, uterus, ovary, oropharynx, stomach, thryoid, oral, cervix, testis, larynx, and Hodgkin lymphoma) diagnosed in 2021 and 2022 in England, excluding rare sex/cancer combinations (notably male breast cancer). We examined differences in proportions of cancers by stage, with particular focus on deprivation, ethnicity, and geographic area, adjusting for cancer site, age, sex, and comorbidity score. We used binary logistic regression to summarise differences, comparing non-advanced (i.e., TNM stages 1-2) to advanced (stages 3-4) cancer. Data were largely complete, but missing values for stage at diagnosis (20.5%) and ethnicity (37.6%) were accounted for using multiple imputation by chained equations.

Results
Our sample included 569,654 tumour registrations (in 550,817 different people). The most common cancers being breast (99,939 registrations) and prostate (98,492), and the least common being testis (3,747) and Hodgkin lymphoma (3,403). In imputed estimates, 33%, 20%, 20%, and 27% of tumours were diagnosed at stages, 1, 2, 3 and 4, respectively. Stage variation by site was considerable, with 79% of testis cancers at stage 1 vs 7% of oesophageal. We similarly observed substantial and site-dependent variation by age, and some variation by sex (women generally less likely to have advanced stage) and comorbidity (more comorbid patients generally more likely to have advanced stage).

Overall, patients in more deprived groups were slightly more likely to have been diagnosed at advanced stage (adjusted OR [aOR] comparing most deprived fifth to least deprived fifth: 1.25, 95% confidence interval 1.22 to 1.27). Deprivation gradients varied by site, with little or no evidence of variation for many sites including lung and ovarian cancer.

Across all sites combined, there was little evidence of variation in stage at diagnosis by ethnicity following adjustment for other variables. For individual sites, patients of black ethnicity appeared more likely to be diagnosed with advanced stage cancer for uterus (aOR vs White ethnicity: 2.06, 1.71 to 2.49), cervix (aOR: 2.03, 1.40 to 2.94) and breast cancer (aOR: 1.42, 1.29 to 1.56).

Implications for practice and policy
Cancer registry data can be used to identify groups at higher risk of diagnosis at an advanced stage, helping to guide targeted interventions to improve cancer control.