Deep Learning Prediction Model for Patient Survival Outcomes in Palliative Care Using Actigraphy Data and Clinical...

Area of Science:

• Palliative Care Medicine
• Biomedical Engineering
• Data Science

Background:

• Accurate survival prediction in end-of-life care is vital but challenged by subjective traditional methods.
• Wearable technology offers continuous monitoring for more objective survival outcome prediction in palliative care.
• Deep learning (DL) models show promise for enhancing prognostic accuracy in end-stage cancer patients.

Purpose of the Study:

• To explore deep learning (DL) models for predicting survival outcomes in end-stage cancer patients.
• To compare the accuracy of an activity monitoring and survival prediction model using DL against traditional prognostic tools.
• To evaluate the efficacy of wearable sensor data, alone and combined with clinical information, for survival prediction.

Main Methods:

• Recruited 78 patients from a palliative care unit; 66 were included in the DL model.
• Utilized wearable sensor data (actigraphy) and clinical information for model development.
• Compared DL model performance against the Karnofsky Performance Scale (KPS) and Palliative Performance Index (PPI).

Main Results:

• The Karnofsky Performance Scale (KPS) and Palliative Performance Index (PPI) had accuracies of 0.833 and 0.615.
• Actigraphy data alone achieved an accuracy of 0.893.
• Wearable data combined with clinical information yielded the highest accuracy at 0.924.

Conclusions:

• Integrating clinical data with wearable sensors significantly improves prognostic accuracy.
• A 48-hour data collection period from wearable sensors is sufficient for accurate survival predictions.
• This approach can enhance clinical decision-making, patient support, and personalized end-of-life care planning.