AI and Machine Learning May Improve Cancer Pain Prediction and Management

Artificial intelligence (AI) and machine learning techniques show promise for improving cancer pain prediction and management, according to study results published in the Journal of Pain and Symptom Management.

Researchers aimed to systematically review the use of AI/machine learning algorithms in cancer pain research, assess their applications and performance, and evaluate adherence to reporting guidelines by systematically searching 3 databases. They included English-language studies on AI/machine learning applications in cancer pain medicine. The review included 44 studies published between 2006 and 2023.

The review demonstrated a rise in publications related to AI/machine learning cancer pain research, with 1 article published between 2006 and 2009 and 26 published between 2020 and 2023. Most studies (17) utilized a prospective uni-institutional cohort, with a median sample size of 320 (range 21-46,104; IQR 140-1000), though 5 studies did not specify population size. The cohort size in 52% of studies ranged from 100-1000 patients, while 20% involved more than 1000 patients. Random Forest and Logistic Regression were the most common AI/machine learning algorithms used, particularly in studies exploring multiple models (55%).

Although the AI/machine learning models showed potential in predicting cancer pain and understanding patient satisfaction, only 3 out of 18 studies performed external validation, and 7 discussed clinical application. The median area under the receiver operating curve (AUC) across studies was 0.77%, with the highest performance seen in cancer-related pain studies (median AUC 0.86), followed by studies on cancer treatment-related pain (median AUC 0.71) and cancer pain management (median AUC 0.80). However, many studies lacked details on model calibration and external validation, leading to high risks of bias in 34 out of 44 studies. Compliance with TRIPOD guidelines was 70.7%, with significant shortcomings in areas like blinding and handling missing data.

Study limitations include heterogeneity among the identified studies, lack of some AI/machine learning performance metrics, and other symptoms or cancer therapy toxicities that may have relevance to pain were not explored.

 “AI/[machine learning] models demonstrated strong performance in predicting cancer pain, risk stratification, and enabling personalized pain management,” the authors concluded.

This article originally appeared on Clinical Pain Advisor