Invited Commentary: Treatment Drop-in-Making the Case for Causal Prediction

Area of Science:

• Epidemiology
• Biostatistics
• Health Informatics

Background:

• Clinical prediction models (CPMs) guide treatment initiation by identifying high-risk individuals.
• Accurate CPMs require estimating the risk of adverse outcomes in the absence of treatment.
• Treatment drop-in, where individuals start treatment post-prediction, complicates accurate risk estimation.

Purpose of the Study:

• To address the challenge of treatment drop-in in clinical prediction models.
• To explore the integration of causal inference methods into the prediction pipeline.
• To enhance the accuracy, transparency, and generalizability of CPMs.

Main Methods:

• Utilizing causal estimates from external data sources, such as clinical trials.
• Adjusting CPMs to account for treatment drop-in using causal inference techniques.
• Applying causal inference to build prediction models with improved estimands.

Main Results:

• Demonstrated a pragmatic approach to adjust CPMs for treatment drop-in.
• Showcased the value of causal inference in improving prediction models.
• Highlighted the potential for causal inference to enhance model explainability and generalizability.

Conclusions:

• Integrating causal inference into CPMs offers a promising strategy to overcome challenges like treatment drop-in.
• Causal inference can lead to more accurate, transparent, and generalizable prediction models.
• This approach holds significant potential to enhance the role of prediction in healthcare decision-making.