What can hospital emergency admissions prior to cancer diagnosis tell us about socio-economic inequalities in cancer diagnosis? Evidence from population-based data in England

Emergency Presentation (EP) is a common route to colon cancer diagnosis in England (~22%) and is associated with worse survival and patient experience than non-emergency routes. EP can reflect diagnostic delays due to both patient factors (e.g., symptom recognition) and healthcare system factors (e.g., access and delivery). Socioeconomic inequalities are pronounced: older, more deprived, female, non-white, and comorbid patients have higher EP risk. More deprived areas also show higher relative use of emergency versus elective care, only partly explained by comorbidity burden. The authors hypothesise that patterns of hospital emergency admissions (HEAs) in the two years before colon cancer diagnosis are linked to subsequent EP and may reflect differences in healthcare access and utilisation. The study aims to (1) compare HEA proportions and rates up to two years pre-diagnosis by deprivation and by EP vs non-EP, and (2) identify combinations of pre-diagnostic conditions most associated with HEAs, thereby illuminating mechanisms underlying socio-economic inequalities in EP.

Prior research shows high EP proportions in colorectal cancer in England and internationally, with EP linked to poorer survival and patient experience. Inequalities in EP are well-documented: higher among older, more deprived, women, non-white ethnicities, and those with comorbidities. Deprived areas exhibit greater emergency-to-elective hospital use, not fully explained by comorbidity, suggesting systemic care delivery factors. Patient-level explanations include lower symptom awareness, help-seeking delays, and complex comorbidity. Evidence also indicates sex differences (e.g., women’s symptoms more often attributed to benign causes pre-EP) and that certain comorbidities raise EP risk. Ambulatory Care Sensitive Conditions (ACSCs) like COPD, heart diseases, and some mental health conditions are associated with preventable HEAs, with deprivation influencing emergency admission rates for ACSCs. This context motivates examining pre-diagnostic HEAs to disentangle patient versus system drivers of inequalities in EP.

Design and setting: Population-based study of all patients diagnosed with colon cancer (ICD-10 C18.0–C18.9) in England in 2013. Data sources: English National Cancer Registry (patient and tumour characteristics, Routes to Diagnosis algorithm) deterministically linked to Hospital Episode Statistics (HES) Admitted Patient Care (APC) records for hospitalisations in the two years preceding cancer diagnosis. The Routes to Diagnosis algorithm assigns route based on the admission closest to diagnosis (typically up to 28 days before). Exclusions: 2,522 patients without any HES record; for visualisation/HEA trend analyses, excluded the index hospitalisation during which diagnosis was made and 2,596 patients who only had that single admission. Measures: Analysed primary diagnostic codes (first of up to 20 ICD-10 fields) from the discharge episode for each hospital spell; grouped 5,118 ICD-10 codes into 58 aggregate condition groups a priori, reducing to 42 after excluding low/zero-incidence groups by sex and deprivation. Socioeconomic deprivation: Income domain of the English Index of Multiple Deprivation (IMD) at LSOA of residence, categorised into national quintiles (1 least deprived to 5 most deprived); income domain used to avoid incorporating health/access components inherent in overall IMD. Descriptive analyses: Computed monthly rate of hospital admissions per patient and monthly proportion of patients with at least one HEA over the 24 months pre-diagnosis, stratified by route to diagnosis (EP vs non-EP), sex, and deprivation (most vs least deprived). Conditions predictive of HEA: Outcome was HEA (binary). Fitted generalised mixed-effects logistic regression models with random intercepts at patient level (logit link; R lme4::glmer), stratified by sex and deprivation. Predictors: grouped condition indicators and age at diagnosis. Variable selection: Adapted Purposeful Variable Selection (PVS) for mixed-effects models (Hosmer et al.), using likelihood ratio tests (p<0.01) across four steps—(1) univariable screening; (2) multivariable elimination to reduced model; (3) confounding assessment (>10% change in estimates) and re-inclusion; (4) testing previously excluded variables individually to form final main-effects model. Addressed complete separation by detecting problematic covariates with brglm2 (on fixed effects) and excluding those causing convergence issues (e.g., appendicitis, cognition/speech symptoms, musculoskeletal symptoms in specific strata). Estimated marginal effects (probabilities) using ggeffects; additionally presented population-average changes with approximate 95% CIs in supplementary materials. Post-hoc clinical grouping of conditions (potentially related, indirect/non-specific, unrelated to colon cancer) for interpretability. Sensitivity analyses: Compared PVS-selected models with GLMM LASSO (glmmmixedlasso) in least-deprived females across lambdas with BIC evaluation; results largely concordant (24/29 variables), with some discrepancies likely due to correlated covariates. Also fitted regression trees (rpart) on predicted probabilities to explore combinations of conditions; no dominant combinations identified. All analyses stratified by sex and deprivation.

- Cohort: 15,263 colon cancer patients diagnosed in 2013 with at least one NHS hospital admission in the prior two years (74% of all cases). About 21% lived in the least deprived and ~15% in the most deprived areas. Most patients were >65 years; age distribution skewed slightly younger with increasing deprivation. - Volume and type of admissions: Excluding the diagnostic admission, there were 38,859 inpatient hospitalisations in the two years pre-diagnosis; 37% were emergency admissions. Approximately 80% of patients had at least one HEA and 20% had two or more HEAs. - Route to diagnosis: Patients diagnosed via EP had consistently higher HEA rates than non-EP patients. Up to 7 months pre-diagnosis, the HEA rate difference was ~0.01; from 7 months before diagnosis, HEAs rose disproportionately for EP patients, reaching a rate of 0.29 per patient in the final month versus 0.11 in non-EP (difference 0.18). Overall hospital admissions (elective and emergency) also rose around the 7-month mark in both groups. - Deprivation gradients: Higher proportions of multiple admissions with increasing deprivation. Males: >3 HAs in least vs most deprived, 17% vs 21%; >2 HEAs, 7% vs 11%. Females: >3 HAs, 16% vs 21%; >2 HEAs, 7% vs 14%. HEA rates were consistently higher in the most deprived, with a notable widening from 7 months pre-diagnosis and an approximate 20% gap in monthly HEA rates in the last month. However, the monthly proportion of patients with at least one HEA differed by <5% between most and least deprived across most months. - Conditions associated with HEA: Among 42 grouped conditions, 22–26 were retained per sex-deprivation stratum as most predictive. Conditions potentially related to colon cancer that increased HEA probability included abdominal/pelvic pain, appendicitis, digestive disorders, and peritoneal disorders; appendicitis was highly predictive across strata. Upper GI diseases and inflammatory bowel diseases decreased HEA probability. Indirect/non-specific drivers included urinary tract disorders and general symptoms. Unrelated but acute or systemic conditions increasing HEA probability included infectious/parasitic diseases, injury/poisoning, COPD, heart diseases, and mental/behavioural disorders (with probabilities often >0.6). Neoplasms (malignant and in situ/benign/other) and some general conditions reduced HEA probability. - Baseline HEA probabilities (age at mean; no conditions present): Males—most deprived 0.62 (95% CI 0.57–0.66) vs least deprived 0.33 (0.30–0.37). Females—most deprived 0.50 (0.45–0.55) vs least deprived 0.45 (0.40–0.51). - Discrepancies by sex/deprivation: Fewer predictive conditions among most deprived males (22) vs least deprived (26). Some conditions (e.g., mental/behavioural disorders, digestive/peritoneal disorders, urinary tract disorders) increased HEA probability in least but not most deprived males. In females, urinary diseases and general symptoms increased HEA in least but not most deprived; gynaecological conditions (pregnancy/perinatal, female genital disorders) increased HEA in most but not least deprived. Anaemia decreased HEA probability in most deprived males and in least deprived females only. - No dominant condition combinations: Regression trees on predicted probabilities did not identify outstanding condition combinations driving high HEA probabilities; patterns were broadly similar across sex and deprivation, with fewer selected conditions among most deprived males.

Findings support the hypothesis that socio-economic inequalities in emergency presentation of colon cancer are mirrored by patterns of increased and repeated emergency service use before diagnosis among more deprived patients. While overall hospital admission rates were similar across deprivation groups, HEA rates were higher in the most deprived, especially in the last seven months pre-diagnosis. The proportion of patients with any HEA differed little, implying that higher HEA rates in deprived groups stem from repeated emergency use by a subset of patients, often for non-specific or indirectly related conditions. The conditions most predictive of HEA were broadly similar across groups, suggesting that disease-specific factors alone do not explain the inequalities; rather, system-level barriers and patterns of access/use may drive repeated emergency utilisation. Ambulatory Care Sensitive Conditions (e.g., COPD, heart disease, some mental health conditions) were prominent contributors to HEAs, with some showing stronger associations in deprived patients, highlighting preventable opportunities via better ambulatory care and chronic disease management. Sex-specific nuances (e.g., higher baseline HEA in women; benign attribution of women’s abdominal symptoms; differential association of anaemia) suggest that clinical pathways and diagnostic attribution may contribute to delays and emergency use. Overall, the results underline the role of healthcare system factors—accessibility, primary care consultation constraints, and diagnostic access—in shaping pre-diagnostic emergency use and EP risk, beyond patient awareness/comorbidity alone.

More disadvantaged populations exhibit higher and more frequent emergency hospital use in the two years preceding colon cancer diagnosis, particularly in the last seven months, without major differences in the specific conditions triggering HEAs. This points to system-related barriers along the care pathway—such as challenges in primary care consultations, digital access, and timely diagnostics—contributing to delays and emergency presentations. Addressing inequalities in cancer diagnosis should therefore prioritise healthcare system improvements that reduce barriers for deprived groups, enhance management of ambulatory care sensitive conditions, and streamline access to diagnostics and specialist care. Future research should disentangle patient versus system drivers across cancers, evaluate interventions targeting repeated emergency users, and assess policy changes (e.g., extended consultations, facilitated diagnostic pathways) for their impact on reducing HEAs and EPs among deprived populations.

- Data coverage: 2,522 (12%) cancer patients lacked HES records, likely due to non-NHS care, and were excluded; generalisability to those patients is uncertain. - Outcome and coding: Used primary diagnosis of the discharge episode to define admission reason; misclassification is possible, though >95% concordance with admission episode was noted. Grouping of ICD-10 codes into condition categories may obscure heterogeneity within groups. - Analytical exclusions: For HEA trend analyses, excluded the diagnostic admission and patients with only that admission, potentially affecting absolute rates though aiming to avoid bias. - Variable selection: Purposeful Variable Selection tests previously excluded variables one at a time (not jointly), risking omission of jointly important predictors; complete separation issues required dropping some covariates in specific strata. - Model scope: Stratification by sex and deprivation reduces sample within strata; despite large overall size, some condition–stratum cells were sparse. - Observational design: Causal inference about system vs patient factors is limited; unmeasured confounding (e.g., primary care access measures, health literacy) may persist. - Time frame: Focus on two years pre-diagnosis; patterns outside this window were not assessed.