Abstract
Purpose: Cancer-related fatigue (CRF) is experienced by 30% of the breast cancer survivors up to ten years after diagnosis. Many interventions exist to support patients with CRF, at the same time, there is not one gold-standard intervention that is effective for all. Therefore, we work towards a holistic decision support tool to give personalized intervention advice for CRF after breast cancer.
Methods: The decision support tool consists of several parts. First, patients at risk of (chronic) CRF should be identified to support them as early as possible in preventing (chronic) CRF. For this, we developed a prediction model with machine learning to predict individual risks for CRF. Second, to investigate personalization possibilities, we created an overview of eHealth interventions and their preference sensitive attributes using a systematic scoping review. Third, to find out how patient preferences regarding these attributes vary between individuals, we set up a pilot study in which we used Analytic Hierarchy Process (AHP) to rate the relative importance of intervention attributes. Fourth, using decision rules, an intervention advice based on preferences was developed by linking the relative importance of attributes to interventions.
Results: The CRF prediction model showed poor discrimination (AUC: 0.67) and cannot be used to identify high-risk patients. In our scoping review, we identified 35 eHealth interventions and confirmed variation within preference sensitive attributes, e.g. in duration and healthcare professional contact. The AHP pilot study (n=6) showed variation in patient preferences, but its complexity resulted in high inconsistency scores. Therefore, a simpler methodology (Best-Worst Scaling) was selected to weigh the importance of the attributes. Lastly, our decision rules resulted in matching percentages to rank CRF interventions based on patient preferences. The output does not only consist of the ranking of CRF interventions, but additionally shows for which attributes an intervention does or does match the indicated preferences.
Conclusions: This study used a comprehensive approach to work towards decision support for CRF after breast cancer. Despite that we are not able to target high-risk patients, our decision rules can help to come to CRF intervention recommendations based on individual preferences. In future studies, another method or other data is necessary to identify high-risk patients. Additionally, future work should focus on incorporating intervention effectiveness and holistic patient monitoring to optimize personal recommendations.
Methods: The decision support tool consists of several parts. First, patients at risk of (chronic) CRF should be identified to support them as early as possible in preventing (chronic) CRF. For this, we developed a prediction model with machine learning to predict individual risks for CRF. Second, to investigate personalization possibilities, we created an overview of eHealth interventions and their preference sensitive attributes using a systematic scoping review. Third, to find out how patient preferences regarding these attributes vary between individuals, we set up a pilot study in which we used Analytic Hierarchy Process (AHP) to rate the relative importance of intervention attributes. Fourth, using decision rules, an intervention advice based on preferences was developed by linking the relative importance of attributes to interventions.
Results: The CRF prediction model showed poor discrimination (AUC: 0.67) and cannot be used to identify high-risk patients. In our scoping review, we identified 35 eHealth interventions and confirmed variation within preference sensitive attributes, e.g. in duration and healthcare professional contact. The AHP pilot study (n=6) showed variation in patient preferences, but its complexity resulted in high inconsistency scores. Therefore, a simpler methodology (Best-Worst Scaling) was selected to weigh the importance of the attributes. Lastly, our decision rules resulted in matching percentages to rank CRF interventions based on patient preferences. The output does not only consist of the ranking of CRF interventions, but additionally shows for which attributes an intervention does or does match the indicated preferences.
Conclusions: This study used a comprehensive approach to work towards decision support for CRF after breast cancer. Despite that we are not able to target high-risk patients, our decision rules can help to come to CRF intervention recommendations based on individual preferences. In future studies, another method or other data is necessary to identify high-risk patients. Additionally, future work should focus on incorporating intervention effectiveness and holistic patient monitoring to optimize personal recommendations.
Original language | English |
---|---|
Pages (from-to) | NP182-NP182 |
Journal | Medical decision making |
Volume | 44 |
Issue number | 3 |
DOIs | |
Publication status | E-pub ahead of print/First online - 4 Mar 2024 |
Event | 45th Annual North American Meeting of the Society for Medical Decision Making, SMDM 2023: Innovation in Decision Making: Thinking to Tools - Philadelphia, United States Duration: 22 Oct 2023 → 25 Oct 2023 https://smdm.org/meeting/45th-annual-north-american-meeting |