Predicting Phlebitis with Machine Learning: A Revolutionary Approach to Improve Patient Care

Peripheral intravenous catheter-associated phlebitis is a common yet serious complication in healthcare, particularly for critically ill patients. Phlebitis, which refers to the inflammation of a vein, can lead to discomfort, infections, and increased hospital stays. To address this, researchers have developed a machine learning (ML)-based model to predict the incidence of phlebitis, offering a proactive solution to prevent this complication.

Study Overview
This study used the AMOR-VENUS database to develop several machine learning models that predict the occurrence of peripheral intravenous catheter (PIVC)-associated phlebitis. By analyzing data from 3,429 PIVCs inserted in critically ill patients in 23 intensive care units (ICUs) in Japan, the researchers aimed to create models that could predict the risk of phlebitis before it occurs.

Methodology
Researchers tested multiple machine learning models, including random survival forests, logistic regression with LASSO, random forests, and gradient boosting trees. These models analyzed 40 potential risk factors, including patient characteristics, catheter specifics, and the types of medications administered. By dividing the dataset into development and validation cohorts, the researchers ensured the accuracy and reliability of their predictive models.

Key Findings
Phlebitis Incidence: The incidence of phlebitis was 8.7% in the development cohort and 10.2% in the validation cohort.
Performance of Models: The random survival forest model showed the highest predictive performance for time-to-event phlebitis outcomes. Logistic regression and random forest models also performed well for binary outcomes.
Predictive Factors: Key predictors of phlebitis included the insertion site of the catheter, catheter material, patient age, and certain medications like nicardipine.

Conclusion
The study demonstrated that machine learning models, particularly random survival forests and random forests, can effectively predict the incidence of PIVC-associated phlebitis. These predictive tools can help healthcare providers take preventive measures, ensuring better patient outcomes by allowing early interventions for high-risk patients.

Implications for Patient Care
By integrating machine learning models into hospital settings, clinicians can receive timely alerts about patients at risk of phlebitis. This approach can lead to faster interventions, reduced complication rates, and better resource allocation, ultimately improving the overall quality of care.

Authors: Hideto Yasuda, Claire M. Rickard, Olivier Mimoz, Nicole Marsh, Jessica A. Schults, Bertrand Drugeon, Masahiro Kashiura, Yuki Kishihara, Yutaro Shinzato, Midori Koike, Takashi Moriya, Yuki Kotani, Natsuki Kondo, Kosuke Sekine, Nobuaki Shime, Keita Morikane, Takayuki Abe

Read More: https://sciendo.com/it/article/10.2478/jccm-2024-0028