Journal of Clinical Oncology & Advanced Therapy

Abstract

Background: Most studies on thyroid cancer have been aimed at prediction of recurrence, survival analysis, and machine-learning classification models that are often binary or unordered, thus ignoring the ordering of disease severity. The purpose of this study was to explore factors associated with the severity of thyroid cancer in an ordinal cumulative logit model that maintains the natural order of disease progression. Also, to estimate stage-transition probabilities and test the ability of an ordinal severity model to predict disease severity.

Methods: This study was carried out by a retrospective analysis of data from 383 thyroid cancer patients that were retrieved from the UCI Machine Learning Repository. The likelihood-based model selection and ordinal cumulative logit regression methods were used. Models were compared for adequacy using threshold parameters, information criteria, likelihood ratio tests, predicted probabilities, and classification performance.

Results: The findings revealed that age significantly increased the odds of progression to a higher disease stage (OR = 1.132, 95% CI: 1.090-1.175, p < 0.001), while uni-focal tumours substantially reduced the odds of severe disease (OR = 0.106, 95% CI: 0.039-0.294, p < 0.001). The final ordinal model had a markedly better fit than the null model (AIC: 157.613, LogLik: −72.807) as compared to the null model (AIC: 265.048, LogLik: −130.524). The predicted probabilities indicated that the chances of having Stage I disease were 94.6% for low-risk patients and 53.7% for patients > 60 years of age, with advanced stages of disease having significantly higher probabilities as well.

Conclusion: This study proposes an ordinal modelling method based on severity index, which overcomes the information loss problem of binary and multinomial models, and succeeded in giving clinically interpretable estimates of the progression of thyroid cancer. To facilitate early intervention and risk-stratified management, enhanced age-centred surveillance and intensified monitoring of multifocal tumours are recommended.