Factors Associated with Late-Stage Cervical Cancer Diagnosis in Kazakhstan (2015-2023)

Abstract

Background. Cervical cancer is the second most common cancer among women in Kazakhstan. Early detection reduces mortality and treatment burden. Late-stage diagnosis remains common despite a national screening programme in place since 2008. Evidence on which patient- and facility-level factors are associated with late-stage presentation in Kazakhstan is limited. Aim. To identify factors associated with late-stage cervical cancer diagnosis in Kazakhstan between 2015 and 2023, and to examine regional and temporal variation. Methods. Retrospective cross-sectional analysis of patient-level records from the Electronic Registry of Oncological Patients (EROP) with ICD-10 code C53. Records were deduplicated to one observation per patient (earliest admission) giving 12,130 unique patients. The outcome was late-stage (III-IV) and early-stage (I-II) diagnosis. Predictors were age, body mass index category, rural residence, emergency admission, first hospitalisation in the year, region, and year of diagnosis. Descriptive statistics, bivariate tests, and multivariable logistic regression were performed in Stata 19. Model discrimination (AUC), calibration (Hosmer-Lemeshow), and six sensitivity analyses were performed. Results. Of 12,130 patients, 17.07% were diagnosed at late stage. Regional late-stage proportion ranged from 1.90% (Atyrau) to 35.33% (Akmola). In the fully adjusted model, emergency admission was the strongest individual predictor (adjusted odds ratio [aOR] = 5.20, 95% CI 3.98-6.80). Age was positively associated (aOR = 1.014 per year). BMI showed an inverse dose-response: underweight aOR = 1.30, overweight aOR = 0.77, obese aOR = 0.70 (normal as reference). First hospitalisation of the year was protective (aOR = 0.62). The rural main effect was null (aOR = 0.999), but the rural × region interaction was statistically significant (χ²(19) = 32.14, p = 0.030), with rural residence associated with higher late-stage risk in northern agricultural regions and lower late-stage risk in several southern and urban regions. Late-stage odds were elevated in 2020, 2021, and 2023 relative to 2016. Model discrimination was modest (AUC = 0.699); calibration was adequate (Hosmer-Lemeshow p = 0.363). Conclusion. Late-stage cervical cancer diagnosis in Kazakhstan is strongly regionally patterned and is more than tripled in emergency admissions. Region-level variation is not fully explained by patient-level variables, which points to differences in screening coverage, referral pathways, and stage-at-diagnosis recording that are not captured in the present dataset. Strengthening screening outreach in high-burden regions, auditing stage recording in regions with implausibly low late-stage proportions, and linking oncology records to screening history and socioeconomic data should be priorities.