Determinants of Early-Stage Breast Cancer Presentation in Western Kazakhstan: A Population-Based Analysis of Urban-Rural Disparities and Diagnostic Pathways (2015-2025).

Background and Objectives: Breast cancer is the leading oncological diagnosis among women in Kazakhstan, yet a substantial proportion of cases are still detected beyond the earliest stages, particularly in peripheral regions such as Aktobe. Despite a national mammography screening programme covering women aged 40-70 years since 2018, structural differences in access to early diagnosis-related to geography, socioeconomic circumstances, and the diagnostic pathway-may compromise outcomes. We aimed to identify factors independently associated with early-stage presentation and to characterise the temporal pattern of early-stage presentation without assuming a monotonic trend. Methods and Materials: We conducted a retrospective, population-based analytical study of all confirmed breast cancer cases (ICD-10 C50) registered in the Aktobe regional cancer registry and diagnosed between 1 January 2015 and 31 December 2025 (n = 2232). The outcome was early-stage presentation, defined literally as Stages I-IIa at diagnosis, with Stages IIb-IV as the comparator; this is a stage-at-presentation classification and is not intended to indicate surgical operability or treatment sequence. Multivariable logistic regression estimated adjusted odds ratios (aORs). Two models were used: Model A included age, sex, residence, administrative nationality, employment/social status, and calendar year; Model B additionally included the diagnostic pathway, which may lie on the causal pathway between structural determinants and stage. Calendar year was modelled as a categorical variable, and a complementary phase-based model (2015-2017, 2018-2019, 2020-2022, 2023-2025) was fitted. A residence-by-year interaction; a multinomial sensitivity analysis separating Stages IIb, III, and IV; discrimination (AUC, Brier score); and calibration (Hosmer-Lemeshow test, calibration plot) were also assessed. Results: Of 2232 patients (99.1% female; mean age 57.0 ± 12.5 years), 1323 (59.3%) presented at Stages I-IIa. In Model A, rural residence (aOR 0.77, 95% CI 0.64-0.93; p = 0.006), unemployment relative to employment (aOR 0.69, 95% CI 0.52-0.90; p = 0.007), Russian administrative nationality (aOR 0.64, 95% CI 0.50-0.82; p < 0.001), and other non-Kazakh nationalities (aOR 0.73, 95% CI 0.58-0.94; p = 0.012) were independently associated with lower odds of Stage I-IIa presentation. In Model B, patient-initiated (self-referral) presentation was associated with lower odds relative to clinical examination room detection (aOR 0.46, 95% CI 0.30-0.72; p < 0.001), whereas organised screening was not significantly associated (aOR 1.52, 95% CI 0.95-2.43; p = 0.082). The calendar-year pattern was clearly non-linear (categorical vs. linear year: likelihood-ratio χ2 = 76.1, df = 9, p < 0.001): odds of Stage I-IIa presentation peaked in 2018 (aOR 2.49, 95% CI 1.57-3.93 vs. 2015), were lowest in 2022 (aOR 0.64, 95% CI 0.42-0.97), and partially recovered thereafter. In the phase-based model (reference 2015-2017), the aORs were 1.62 (95% CI 1.22-2.15) for 2018-2019, 0.54 (95% CI 0.41-0.69) for 2020-2022, and 0.62 (95% CI 0.46-0.84) for 2023-2025. Model discrimination was limited (AUC 0.648, 95% CI 0.625-0.671; Brier score 0.226) with acceptable calibration (Hosmer-Lemeshow χ2 = 8.38, df = 8, p = 0.40). Conclusions: In this registry-based cohort, rural residence, unemployment, non-Kazakh administrative nationality, and patient-initiated presentation were independently associated with lower odds of early-stage breast cancer presentation. The temporal pattern was non-monotonic, with the highest odds around the 2018 screening expansion, a marked reduction during 2020-2022, and only partial recovery thereafter. These are observational associations rather than causal or programme-evaluation findings; they should be interpreted as hypothesis-generating and require confirmation with screening-process, service-capacity, and patient-level access data.