Prediction of Recurrence in Thyroid Cancer Patients Using Plotkin’s Least General Generalisation LGG as a Method of Inductive Logic Programming ILP

  • Background:
    Although thyroid cancers are known to have a relative low risk of recurrence, there are factors associated with a higher risk of recurrence. However, predicting disease recurrence and prognosis in patients undergoing thyroidectomy is clinically difficult. To detect new algorithms to predict recurrence, inductive logic programming was used.

    Methods:
    From January 2009 to June 2010, 797 cases of thyroid cancer patients who underwent bilateral total thyroidectomy with following radio-iodine treatment from our database were studied. 638 (80%) cases were used to create algorithms to detect recurrence. 159 (20%) cases were analyzed for validation of created rules. Least Generalized Generalization LGG was chosen as a method of inductive logic programming to extract rules which represents algorithms to predict recurrence. Delmia PRD was used for analysis.

    Results:
    Of the 638 cases, there were 46 (7.2%) cases with recurrence. There were 4 rules detected which could predict all of the cases with recurrence. Postoperative thyroglobulin was the most powerful variable which correlated with recurrence. When the 4 recognized rules were applied to 159 cases for validation, it was possible to predict 72.7% (8 cases among 11 of the recurrences). When factors known for high and intermediate risk were selected for creating rules, the most optimal combination could only predict 66.7% of the recurrence cases.

    Discussion & Conclusion:
    From our database, rules to predict recurrence were identified which were able to predict recurrence more precise than already known high and intermediate risk factors.

 

Leave a Reply