Dimensionality Reduction for capturing the multifaceted nature of Oncogeriatric Patients: Enhancing ML Interpretability

In resource-constrained healthcare environments, the development of accurate and interpretable predictive models is crucial for optimizing treatment decisions, particularly for geriatric cancer patients. A significant challenge lies in balancing model complexity with explainability, especially when incorporating a large number of variables with limited patient cases. This study addresses this gap by exploring the application of dimensionality reduction techniques to enhance the interpretability of machine learning models predicting treatment eligibility for geriatric cancer patients, as determined by the Oncological Treatment Board. The primary objective was to simplify model complexity while preserving predictive performance, thereby providing greater insight into the factors influencing treatment decisions. Key findings indicate that the OTB tends to prioritize geriatric assessment indicators over the traditional ECOG score when evaluating treatment eligibility. Dimensionality reduction enabled the identification of salient features, resulting in a more transparent model without compromising predictive accuracy. In conclusion, this approach offers a balanced trade-of between model accuracy and explainability, which is valuable for informed decision-making in resource-constrained healthcare settings.