Using Operative Reports to Predict Heart Transplantation Survival

Area of Science:

• Medical Informatics
• Computational Biology
• Surgical Outcomes Research

Background:

• Heart transplantation outcomes face uncertainty due to factors like late rejection and mortality.
• Current predictive models for heart transplant success, based on donor/recipient data, have limitations in accuracy.
• Operative reports contain rich textual data often underutilized in outcome prediction.

Purpose of the Study:

• To introduce a novel method for predicting heart transplant outcomes using textual data from operative reports.
• To evaluate the efficacy of this text-based prediction method against traditional data-driven approaches.
• To assess the potential of operative reports as a supplementary data source for improving transplant outcome predictions.

Main Methods:

• Utilized a dataset of 300 heart transplantation surgical reports.
• Applied truncated TF-IDF vectorization to extract features from the textual data.
• Employed logistic regression for outcome prediction, validated using five-fold cross-validation.

Main Results:

• Achieved a macro F1-score of 59.1% for one-year survival prediction.
• Achieved a macro F1-score of 54.9% for five-year survival prediction.
• Demonstrated that textual operative data can significantly discriminate transplant outcomes.

Conclusions:

• Textual information from surgical reports can effectively predict heart transplant outcomes.
• This text-based approach offers a valuable addition to existing heart transplant prediction systems.
• Further research with larger datasets could enhance the predictive power of this method.