Factors associated with cervical cancer among women in northeastern Iran: a registry-linked case–control study

Cervical cancer remains an important public health problem among women worldwide. Although several demographic, reproductive, lifestyle, and clinical characteristics have been associated with cervical cancer, evidence from northeastern Iran remains limited. This study aimed to investigate correlates of cervical cancer in a registry-linked, population-based case–control study restricted to registry-identified cases with available pre-diagnosis primary-care records, conducted in northeastern Iran; because absolute predictive performance cannot be validly estimated from a case–control design with a fixed sampling ratio (see Methods), this study is explanatory and hypothesis-generating in nature rather than a clinical risk-prediction study. This registry-linked, population-based case–control study, restricted to registry-identified cases with available pre-diagnosis primary-care records, included women covered by Mashhad University of Medical Sciences between March 21, 2017, and March 19, 2021. Cases comprised 74 women with registry-defined cervical cancer identified from the regional Population-Based Cancer Registry who had a pre-diagnosis primary-care health record (60.8% pathology-confirmed, 35.1% clinically diagnosed, 4.1% death certificate-ascertained), representing 26.1% of all registry-identified cases during the study period; 148 controls without cervical cancer were randomly selected from the Primary Health Care Family Physician Information System. Univariable and multivariable logistic regression analyses were performed to estimate crude and adjusted odds ratios (ORs) with 95% confidence intervals (CIs); the four-variable model was designated as the primary parsimonious specification based on clinical considerations, lower missingness, and a more favorable events-per-variable ratio; this designation reflects the analysis process rather than prospective specification. A seven-variable model adjusting for additional covariates was examined as an expanded sensitivity analysis. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), Hosmer–Lemeshow goodness-of-fit test, Brier score, and Nagelkerke R². Internal validation was performed using bootstrap resampling. A total of 222 women (74 cases and 148 controls) were included in the analysis. A primary parsimonious four-variable model and a seven-variable model, examined as an expanded sensitivity analysis, were fitted: a four-variable model (age, parity, age at first pregnancy, hypertension; N = 197, events-per-variable ratio 17.8) and a seven-variable model additionally adjusting for BMI, age at menarche, and smoking ( N = 185, events-per-variable ratio 10.0). In the four-variable model, increasing age was statistically associated with cervical cancer (adjusted OR [aOR] = 1.050, 95% CI: 1.013–1.088; P = 0.008; aOR ≈ 1.63 per 10-year difference), while age at first pregnancy showed a directionally consistent inverse trend that did not reach conventional statistical significance (aOR = 0.936 per one-year increase, 95% CI: 0.875–1.002; P = 0.057). In the seven-variable model, increasing age remained statistically associated with cervical cancer (aOR = 1.043, 95% CI: 1.004–1.083; P = 0.030), and the inverse estimate for age at first pregnancy was similar in direction and magnitude (aOR = 0.918, 95% CI: 0.850–0.992; P = 0.030). In a multiple-imputation sensitivity analysis using the full analytic sample, the age association was materially unchanged (aOR = 1.034, P = 0.032), but the age-at-first-pregnancy association was attenuated and no longer statistically significant (aOR = 0.938, 95% CI: 0.871– 1.010; P = 0.090). Restricting cases to pathology-confirmed invasive histology only ( n = 39) likewise confirmed the age association (aOR = 1.076, P = 0.004) but not the age-at-first-pregnancy association (aOR = 0.922, P = 0.120). Across all analyses, age was consistently associated with cervical cancer. The inverse association between age at first pregnancy and cervical cancer was directionally consistent but was sensitive to model specification and missing-data handling and was not statistically significant in the primary model or pathology-restricted analysis. No statistically significant associations were detected for body mass index, parity, age at menarche, smoking, or hypertension; confidence intervals were wide, and the study had limited power to exclude modest associations for these variables. Both models demonstrated modest discriminatory ability (AUC = 0.670 [four-variable] and 0.666 [seven-variable]) and non-significant Hosmer–Lemeshow tests; because of the case–control sampling design, calibration and Brier-score estimates should not be interpreted as population-level absolute-risk performance. Internal validation using 1,000 bootstrap resamples with formal optimism correction indicated meaningful overfitting in both models, more pronounced in the seven-variable model (optimism-corrected AUC = 0.607, calibration slope 0.643) than the four-variable model (optimism-corrected AUC = 0.640, calibration slope 0.802). Among women with a pre-diagnosis primary-care record in northeastern Iran, increasing age was consistently associated with registry-defined cervical cancer across both the four-variable and seven-variable model specifications and the multiple-imputation analysis. Because HPV infection and screening history were unavailable, this association should not be interpreted as a causal or fully adjusted independent effect. The inverse association between age at first pregnancy and cervical cancer was directionally consistent but was sensitive to model specification and missing-data handling and was not statistically significant in the primary model or pathology-restricted analysis, so this finding should be regarded as fragile and interpreted with greater caution than the age association. While age and, more tentatively, age at first pregnancy were associated with cervical cancer, the apparent discrimination of both models was modest and, after optimism correction, further attenuated to little better than chance (optimism-corrected AUC 0.607–0.640); these associations should therefore be used to generate hypotheses, not to guide individual risk assessment or clinical decision-making in their current form.