Collaborative Filtering for the Imputation of Patient Reported Outcomes

Area of Science:

• Oncology
• Data Science
• Medical Informatics

Background:

• Missing data is a significant challenge in patient-reported outcome (PRO) datasets.
• Accurate data imputation is essential for machine learning and data mining in healthcare.
• Head and neck cancer symptom ratings are often incomplete in clinical datasets.

Purpose of the Study:

• To propose and evaluate collaborative filtering for imputing missing symptom ratings in head and neck cancer patients.
• To compare the efficacy of collaborative filtering against established imputation methods.
• To enhance the usability of patient-reported outcome data for further analysis.

Main Methods:

• Utilized two collaborative filtering approaches: patient-based and symptom-based imputation.
• Compared collaborative filtering performance against Multiple Imputation by Chained Equations, Nearest Neighbor Imputation, and Linear Interpolation.
• Evaluated imputation accuracy using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) metrics.

Main Results:

• Collaborative filtering demonstrated strong performance in imputing missing symptom ratings.
• Patient-based and symptom-based collaborative filtering were effective imputation strategies.
• Collaborative filtering outperformed traditional methods like Multiple Imputation by Chained Equations and Nearest Neighbor Imputation.

Conclusions:

• Collaborative filtering is a viable and superior method for imputing missing patient symptom data.
• This approach can significantly improve the quality of head and neck cancer PRO datasets.
• The findings support the integration of collaborative filtering in clinical data preprocessing pipelines.