Predicting Incipient Symptom Deterioration from Serial Patient-Reported Data During Cancer Chemotherapy Course Using LSTM Modeling

Joseph Finkelstein, Aref Smiley, Christina Echeverria, Kathi Mooney

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

A dataset was constructed from 12 specific symptoms of 339 cancer patients who reported daily symptoms during chemotherapy treatment. Two values were recorded for each symptom (except fever which had only one value): the severity and the distress, which were evaluated from 1 to 10. By adding all the feature scores, the "total score" response variable was calculated for each day, ranging from 0 to 230. A "cycle change" feature was also added to indicate when a patient's chemotherapy cycle changed. To predict the total symptom scores, we employed Long Short-Term Memory (LSTM) regression models. A total of seven LSTM models were used to predict the total scores. Each model was trained with a combination of the features from previous days for up to a week to predict the total score of the following day. In addition, the models were trained on the prior days' symptoms without combining the data from the older days to predict the total score up to a week later. Including a "cycle change" feature was also evaluated for all the models. The LSTM models were trained on 80% of the dataset and tested on the remaining 20%. The results showed good predictive accuracy, with R-squared values more than 0.52 for most models with combining features from prior days. For these models, the Root Mean Square Error (RMSE) values were generally low, indicating the effectiveness of the LSTM models in forecasting symptom scores. For those models without combining the features from prior days and without including the cycle change feature, the R-squared values decreased from 0.40 to 0.23, predicting the total scores based on 2 days to one week ago, respectively. Adding the "cycle change" feature did not improve the predictive performance of the models. Overall, our results showed the potential of the LSTM models in predicting cancer symptoms for up to a week, offering valuable insights for personalized symptom management in cancer patients.

Original languageEnglish
Title of host publicationProceedings - 2024 IEEE 1st International Conference on Artificial Intelligence for Medicine, Health and Care, AIMHC 2024
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages177-180
Number of pages4
ISBN (Electronic)9798350371987
DOIs
StatePublished - 2024
Externally publishedYes
Event1st IEEE International Conference on Artificial Intelligence for Medicine, Health and Care, AIMHC 2024 - Hybrid, Laguna Hills, United States
Duration: 5 Feb 20247 Feb 2024

Publication series

NameProceedings - 2024 IEEE 1st International Conference on Artificial Intelligence for Medicine, Health and Care, AIMHC 2024

Conference

Conference1st IEEE International Conference on Artificial Intelligence for Medicine, Health and Care, AIMHC 2024
Country/TerritoryUnited States
CityHybrid, Laguna Hills
Period5/02/247/02/24

Keywords

  • LSTM
  • cancer symptoms
  • chemotherapy
  • prediction

Fingerprint

Dive into the research topics of 'Predicting Incipient Symptom Deterioration from Serial Patient-Reported Data During Cancer Chemotherapy Course Using LSTM Modeling'. Together they form a unique fingerprint.

Cite this