JMIR Publications

ABSTRACT

Background:

People with cancer frequently experience severe and distressing symptoms associated with their cancer and its treatments. Clinicians and researchers continue to struggle with accurately predicting which symptoms will develop and in which individuals with cancer. Machine learning algorithms can improve the ability to predict symptoms prospectively.

Objective:

The aim of this systematic review is to synthesize the literature that has used machine learning algorithms to predict the development of cancer symptoms, as well as to identify the predictors of these symptoms.

Methods:

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses checklist. We conducted a systematic search of CINAHL, Embase, and PubMed for English records published from 1984 to August 11, 2023, using the following search terms: 'cancer', 'neoplasm', specific symptoms, 'neural networks', 'machine learning', specific algorithm names, and 'deep learning'. All records that met the eligibility criteria were individually reviewed by two co-authors, and key findings were extracted and synthesized.

Results:

A total of 42 studies were included, the majority were published after 2017. Most were conducted in North America (43%) and Asia (40%). The most prevalent category of algorithms was supervised machine learning, accounting for 93% of the studies. Each of the methods — deep learning, ensemble classifiers, and unsupervised machine learning — constituted less than 3% of all studies. The machine learning algorithms with the best performance were Logistic Regression (17%), Random Forest (13%), Artificial Neural Networks (9%), and Decision Tree (9%). The most commonly included primary cancer sites were head and neck (22%) and breast (19%). However, in 41% of the studies, the specific cancer site was not explicitly mentioned. The most frequently studied symptoms were xerostomia (14%), depression (13%), pain (13%), and fatigue (10%). Significant predictors across the studies included age, gender, type/number of prior treatments, cancer site, cancer stage, chemotherapy agents, radiotherapy dose and volume, chronic disease, comorbidity, and previous symptoms such as depression, anxiety, fatigue, pain, and sleep disturbance.

Conclusions:

This review outlines the algorithms employed for predicting symptoms in individuals with cancer. Given the diversity of symptoms people with cancer experience, analytic approaches that can handle complex and non-linear relationships is critical. This knowledge can pave the way for crafting algorithms tailored to specific symptom. Additionally, to improve prediction precision, future research should compare cutting-edge machine learning strategies such as deep learning and ensemble methods with traditional statistical models.