Compare the estimated odds ratios from logistic regression and conditional logistic regression in the case - control study determination risk factors for unintentional childhood poisoning of children in Tehran

Backgrounds and Aims: Identifying risk factors affecting on incidence poisoning is a fundamental measure to prevent poisoning in the community, In the meantime, find an equation for determination effects of factors and severity association of them has great importance. Purpose of this study compare the performance of two models ordinary logistic regression and conditional logistic regression in determination risk factors for unintentional childhood poisoning. Materials and Methods: In the present study was used of data a case-control study that to determination risk factors for unintentional childhood poisoning in Tehran was carried out. After collecting the relevant data into the software stata and to determine the risk factors on incidence poisoning of two conditional logistic regression model and ordinary logistic regression was used. Odds ratios (OR) with confidence interval (CI) 95% for crude and adjusted with sensitivity, specificity, and area under the ROC curve were estimated for each of the proposed models, then they were compared. Results: In multiple conditional logistic regression model addiction in the family, previous poisoning, maternal occupation and inaccessibility of poisoning products have shown a significant impact on incidence of poisoning. But In multiple logistic regression model, addition to the items listed, mental -mental illness in the family showed a significant relationship with outcome (poisoning). The AUC of ROC for conditional logistic regression model %95.82 (CI: 95% 0.94 ? 0.97) with sensitivity and specificity 92.9% and 84.2% respectively and for logistic regression model %89.75 (CI: 95% 0.86 ? 0.93) with sensitivity and specificity 83.6% and 83.1% respectively was obtained. Conclusion: Although confidence interval associated with conditional logistic regression model compared with logistic regression is narrow and on the other hand, higher under curve area of ROC is a good indication of the overall precision of this model in the diagnosis of poisoning. But should be noted that use of each of these models depends on the feature of data (matched or unmatched).