Faculty of Public Health - Andalas University - OCS, 13th IEA SEA Meeting and ICPH - SDev

Font Size: 
Factors associated with nodal involvement among oral cancer patients: Comparative appraisal using conventional logistic and ordinal logistic regression models
Vishwajeet Singh, Sada Nand Dwivedi, SVS Deo, Maroof A. Khan

Last modified: 2018-09-16



Oral cancer is the most common cancer among Indian men. In oral cancer, nodal involvement strongly influences the five year survival rate and prognosis. Also, treatment of oral cancer can be tailored with the knowledge of nodal involvement. It is well known that dichotomization or ignoring the ordering does not fully utilize the available information. Accordingly, objective of this study was to compare the results under conventional logistic and ordinal logistic regressions models.

Material and Methods:

The data base on oral cancer patients available with the Department of Surgical Oncology, BRAIRCH, AIIMS, New Delhi, was used. Accordingly, 945 histopathologically proven oral squamous cell carcinoma (OSCC) patients who went under surgery including neck dissection during 1995-2013 were included. Keeping in view of the related assumptions, the logistic regression (node positive vs. negative) model and ordinal regression (number of involved nodes as 0, 1, 2-4 and >4) model (namely partial proportional odds model) was used for the analysis. To compare the results under both the methods, a common set of covariates was selected on its clinical relevance and/or maximum p value as 25% under univariable analysis. Further, the results in the form of odds ratio and corresponding 95% confidence interval (95% CI) were considered. The results were considered significant at 5% level of significance. Statistical software, STATA (version 14.2), was used for analysis.


Out of 945 patients, nodal involvement was highly prevalent, among 376 (39.8%) patients. Patients with age more than 60 years were 22.4% and male were 77.6%. Under each regression, pain at time of presentation, presence of clinical neck node, sub mucous fibrosis, degree of differentiation and tongue as compared to buccal mucosa emerged to be probable associated factors with nodal involvement. In addition, ordinal regression model gave one more predictor of nodal involvement, i.e., large tumor size. Further, comparatively ten covariates entered in stepwise logistic regression model where as only eight entered in to ordinal logistic regression model.


Pain at time of presentation, presence of clinical neck node, sub mucous fibrosis, degree of differentiation, oral site and tumor size are the most probable associated factors for nodal involvement. In case of ordinal outcome variable, one needs to preferably use ordinal logistic regression model. The current results might be helpful in clinical practice.