Indian Journal of Public Health

: 2015  |  Volume : 59  |  Issue : 3  |  Page : 230--231

Reporting error in the use of multivariable logistic regression

Rajeev Kumar 
 Department of Biostatistics and Medical Informatics, University College of Medical Sciences, Delhi, India

Correspondence Address:
Rajeev Kumar
Room No. 401, Department of Biostatistics and Medical Informatics, University College of Medical Sciences, Dilshad Garden, Delhi - 110 095

How to cite this article:
Kumar R. Reporting error in the use of multivariable logistic regression.Indian J Public Health 2015;59:230-231

How to cite this URL:
Kumar R. Reporting error in the use of multivariable logistic regression. Indian J Public Health [serial online] 2015 [cited 2021 Oct 25 ];59:230-231
Available from:

Full Text


Gururaj et al. published a study on headache disorder and its associated factors in the population-based survey from Bengaluru, Karnataka, India. [1] It was a good attempt to measure the burden caused due to headache. The author applied multivariable logistic regression (MLR) to find factors associated with headache disorder. I would like to present the reporting errors in the use of MLR that can be prevented; prevention of these types of errors not only improves the reliability of the results but also increases the confidence of the readers on the study. In addition, other errors were presented.

In MLR, negative log odds represent corresponding odds ratios (ORs) that should be less than one. In [Table 3](a), gender had negative log odds while the OR was more than one. This may be topographical but sometimes it is difficult for the reader to decide which one is correct. The frequency of headache was very high (about 64%) and OR is a poor estimate of the relative risk. The author reported that "being female or living in a rural area increased the risk of headache by two-fold." Interpreting the OR for common outcome as the relative risk overestimates the effect and overestimation increases as prevalence increases when OR is greater than one. [2] It is unclear from the MLR tables whether age is considered as continuous or dichotomous or categorized as described in Table 2 and included as continuous. It should have been explicitly mentioned in the text or in the MLR result table while using different scales for different statistical methods within an article. In addition, in urban habitations, the relation between age and prevalence of headache seems to be quadratic Figure 1 and this figure depicted 30% prevalence for the urban males aged 56-65 years but in table it was 26.5%.Initially, marital status had four categories Table 1 but in MLR Table 3a, b, c, the author just wrote marital status. The coding and reference category of the independent variable should be properly reported in either the table or the material and methods section, along with the categories to be collapsed for MLR. In univariable and multivariable analyses, the same categories should be used because sometimes collapsing may change the significance. The percentage of overall classification gives wrong information when the outcome is skewed, especially in cross-sectional studies. Overall classification provides correct information in case-control studies with an equal number of subjects in both the groups. Thus, in addition to the overall classification, the author should report correct classification for both the positive and negative outcomes of interest.Reporting guidelines and instructions to the authors of medical journals advocate the exact P value [3] unless it is very small (P < 0.001). The author did not report the exact P value of chi-square tests applied in Table 1. In addition, instead of chi-square symbol just 2 was written. Education and income are the ordinal scale variables; it would be better if the author also applied and reported the chi-square for this trend.There are various methods to determine the confidence interval (CI) for binomial distribution. In the absence of any method of description, the reader usually assumes Wald method for 95% CI. Description of the method used helps the reader to verify the 95% CI in case of any typographical errors. In addition, there are totaling errors for males and females in rural habitations Table 2.

The correct and accurate reporting of MLR results helps the readers to understand the model results, their interpretation, and the model reproducibility for their data. [4] Our recent study on the evaluation of the MLR quality based on well-established, 10-point criteria revealed that the reporting quality of MLR is poor in Indian medical journals. [5] A joint effort of statisticians, peer reviewers, editors, and authors is required to improve the standard of statistical reporting in Indian medical journals.

Financial support and sponsorship


Conflicts of interest

There are no conflicts of interest.


1Gururaj G, Kulkarni GB, Rao GN, Subbakrishna DK, Stovner LJ, Steiner TJ. Prevalence and sociodemographic correlates of primary headache disorders: Results of a population-based survey from Bangalore, India. Indian J Public Health 2014;58:241-8.
2Holcomb WL Jr., Chaiworapongsa T, Luke DA, Burgdorf KD. An odd measure of risk: Use and misuse of the odds ratio. Obstet Gynaecol 2001;98:685-8.
3Lang T, Altman D. Basic statistical reporting for article published in clinical medical journals: The SAMPL guidelines. In: Smart P, Maisonneuve H, Paolerman A, editors. Science Editors Handbook. United Kingdom: European Association of Science Editors; 2013. p. 180.
4Kumar R, Chhabra P. Cautions required during planning, analysis and reporting of multivariable logistic regression. Current Medicine Research Practice (CMRP) 2014;4:31-9.
5Kumar R, Indrayan A, Chhabra P. Reporting quality of multivariable logistic regression in selected Indian medical journals. J Postgrad Med 2012;58:123-6.