Evaluation on presentation quality of multivariable logistic regression results published in Chinese clinical medicine journals

Objective To evaluate presentation quality of multivariable logistic regression(MLR) results published in major Chinese clinical medicine journals and to provide references for correct presentation of the MLR results to authors, reviewers, and editors.Methods Totally 281 articles with MLR result presentations published from 2010 through 2014 in 5 Chinese clinical medicine journals(Chinese Journal of Cardiology, Chinese Journal of Oncology, Chinese Journal of Neuromedicine, Chinese Journal of Pediatrics, and Chinese Journal of Digestive Surgery) were searched from the Chinese Biomedical Data-Base, VIP Chinese Science and Technology Periodical Database, China National Knowledge Infrastructure, and Wanfang database with computer retrieval combined with manual searching.All the articles were evaluated with established 15 criteria for appropriate usage and results reporting of MLR analysis.Results The score of presentation quality of MLR results of all the articles ranged from 1 to 9, with a median of 5, and 75.44%(212) of the articles had the score of less than 6.The items properly presented by more than 50% of the articles were name and version of statistics software used(97.2%), the value of the odds ratio and its 95% confidence interval(85.4%), the number of required samples and independent variables(77.6%), the basis of independent variables selection(68.0%), and fitting procedures(54.8%);while the items properly presented by less than 10%of the articles were the participation of statisticians and epidemiologists(0.36%), sample size estimation method(0.71%), verification of collinearity among the independent variables(1.78%), verification of outliers(3.56%), verification of interactions(3.91%), and validation and goodness-of-fit(9.96%).The proportions of the articles presenting the data-entry-form, coding of variables, and ethics approval and informed consent were 14.6%, 16.7%, and 29.9%, respectively.For the 173 articles involved with continuous variable or ordinal categorical variable, only 2 presented the verification of the linearity between the statistics of the variables and the value of logit(P).Conclusion The quality MLR results presentation is poor for articles published in five Chinese clinical journals.Editors should develop statistical presentation guidelines concerning MLR and encourage researchers to cooperate with statisticians and epidemiologists to improve the quality of presentation.