Logistic regression models in obstetrics and gynecology literature.

OBJECTIVE To evaluate the reporting of multivariable logistic regression analyses and assess variations in quality over time in the obstetrics and gynecology literature. METHODS Methodologic criteria for reporting logistic regression analyses were developed to identify problems affecting accuracy, precision, and interpretation of this approach to multivariable statistical analysis. These criteria were applied to 193 articles that reported multivariable logistic regression in the issues of four generic obstetrics and gynecology journals in 1985, 1990, and 1995. Rates of compliance with the methodologic criteria and their time trends were analyzed. RESULTS The proportion of articles using logistic regression analysis increased over time: 1.7% in 1985, 2.8% in 1990, and 6.5% in 1995 (P < .001 for trend). Violations and omissions of methodologic criteria for reporting logistic models were common. The research question, in terms of dependent and independent variables, was not clearly reported in 32.1%. The process of variable selection was inadequately described in 51.8% of the articles. Among articles with ranked independent variables, 85.1% did not report assessment of conformity to linear gradient. Tests for goodness of fit were not given in 93.2% of articles. The contribution of the independent variables could not be evaluated in 36.2% of the articles because of a lack of coding of the variables. Interactions between variables were not assessed in 86.4% of articles. Analysis of variations in the quality of logistic regression analyses over time showed no increase in reporting of the criteria concerning variable selection and goodness of fit. However, the proportion of articles reporting one quality criterion concerning interpretation of the substantive significance of independent variables showed a trend toward improvement: 42.3% in 1985, 73.6% in 1990, and 75.4% in 1995 (P = .004 for trend). CONCLUSION The reporting of multivariable logistic regression models in the obstetrics and gynecology literature is poor, and the time trends of improvement in quality of reporting are not particularly encouraging.

[1]  J. Marks,et al.  Mortality and infectious disease associated with infant-feeding practices in developing countries. , 1984, Pediatrics.

[2]  T. Treasure,et al.  Minimisation: the platinum standard for trials? , 1998, BMJ.

[3]  E. Funai OBSTETRICS & GYNECOLOGY IN 1996: MARKING THE PROGRESS TOWARD EVIDENCE‐BASED MEDICINE BY CLASSIFYING STUDIES BASED ON METHODOLOGY , 1997, Obstetrics and gynecology.

[4]  M. Shipley,et al.  Confounding of occupation and smoking: its magnitude and consequences. , 1991, Social science & medicine.

[5]  F. Harrell,et al.  Regression models for prognostic prediction: advantages, problems, and suggested solutions. , 1985, Cancer treatment reports.

[6]  D. Kleinbaum,et al.  Applied Regression Analysis and Other Multivariate Methods , 1978 .

[7]  B. Sibbald,et al.  Understanding controlled trials: Why are randomised controlled trials important? , 1998 .

[8]  M. Bracken Reporting observational studies , 1989, British journal of obstetrics and gynaecology.

[9]  D. Hosmer,et al.  Applied Logistic Regression , 1991 .

[10]  S Lemeshow,et al.  The importance of assessing the fit of logistic regression models: a case study. , 1991, American journal of public health.

[11]  P. Brennan,et al.  Interpreting the results of observational research: chance is not such a fine thing , 1994, BMJ.

[12]  B. Sibbald,et al.  Why are randomised controlled trials important , 1998 .

[13]  H. Bauchner,et al.  Studies of breast-feeding and infections. How good is the evidence? , 1986, JAMA.

[14]  K E Warner,et al.  Smoking and lung cancer: an overview. , 1984, Cancer research.

[15]  Richard K. Riegelman,et al.  Statistical first aid : interpretation of health research data , 1994 .

[16]  J. Concato,et al.  The Risk of Determining Risk with Multivariable Models , 1993, Annals of Internal Medicine.