Rough set theory based prognostic classification models for hospice referral.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• This study evaluates classical and dominance-based rough set theory (RST) for developing data-driven prognostic classification models for hospice referral.
• The research compares RST-based models against other data-driven methods focusing on clinical credibility factors: accuracy and accessibility (traceable, interpretable results using simple data).

Purpose of the Study:

• To explore and evaluate the application of RST for data-driven prognostic classification models for hospice referral.
• To compare the clinical credibility (accuracy and accessibility) of RST-based models with other data-driven methods.

Main Methods:

• Retrospective data from 9,103 terminally ill patients were used to design and implement RST-based models for identifying hospice candidates.
• Classical Rough Set Approach (CRSA) decision rules for six-month patient survival were induced using the MODLEM algorithm.
• Dominance-based Rough Set Approach (DRSA) decision rules were extracted using the VC-DomLEM rule induction algorithm.

Main Results:

• RST-based classifiers achieved average AUCs of 69.74% (MODLEM) and 71.73% (VC-DomLEM).
• Compared methods (logistic regression, SVM, random forests, C4.5) achieved higher average AUCs ranging from 70.88% to 74.59%.
• All evaluated classification methods demonstrated substandard performance for prognostic models for hospice referrals.

Conclusions:

• RST and its extensions offer enhanced accessibility for clinical decision support models.
• While non-rule-based methods showed higher accuracy (AUC), the interpretability benefits of rule-based methods like VC-DomLEM may be advantageous.
• Developing accurate prognostic models for hospice referrals remains a significant challenge across all evaluated methods.