Integrated approach to prediction of cervical intra-epithelial neoplasia status through a case-control study

A case-control study on cervical intra-epithelial neoplasia (CIN) was carri ed out on 398 subjects in the state of West Bengal, India. These samples we re taken from mass screening programs organized by the authors, maintaining the uniformity of sampling to the extent possible. The cervical smears wer e tested by the Papanicolaou (PAP) method, following the Bethesda system fo r reporting of CIN status. Odds ratios and correlation coefficients among d ifferent variables, assumed to produce carcinoma of the cervix, show that 6 out of 11 variables, i.e., age, education, socio-economic status, duration of marriage, age at marriage and body surface, are associated with CIN. Mu ltivariate analysis of logistic regression was carried out using BMDP-LR wi th dichotomized response variables considering CIN (0 and 1) in one group a nd CIN (2 and 3) in the other group. The outcome of the analysis indicated that age and educational level are 2 contributing factors for GIN. The perc entage of correct classification in this analysis has improved to 74.5%, wi th a probability of 0.90. Polychotomous regression analysis was carried out using BMDP-PR in the next step. This analysis showed that parity was a con tributing factor, in addition to age and educational level. These 3 factors provide a predictive model for identifying the high-risk group in a ration al way. This approach would restrict screening to approximately 10% of the population. Subsequently, the model has been validated in a confirmatory tr ial among 85 new cases and was found to work satisfactorily. (C) 1999 Wiley -Liss, Inc.