Using machine learning to predict patient-reported symptom clusters in prostate cancer patients receiving radiotherapy

  • Elke Rammant*
  • , Emile Deman
  • , Valerie Fonteyne
  • , Lindsay Poppe
  • , Renee Bultijnck
  • , Piet Dirix
  • , Gert De Meerleer
  • , Karin Haustermans
  • , Ann Van Hecke
  • , Miguel E. Aguado-Barrera
  • , Barbara Avuzzi
  • , David Azria
  • , Jenny Chang-Claude
  • , Barbara N. Chiorda
  • , Ananya Choudhury
  • , Patricia Calvo-Crespo
  • , Dirk De Ruysscher
  • , Antonio Gomez-Caamano
  • , Philipp Heumann
  • , Ashley M. Hopkins
  • Kerstie Johnson, Maarten Lambrecht, Alan Mcwilliam, Bradley D. Menz, Filip Poelaert, Tiziana Rancati, Kato Rans, Tim Rattay, Barry S. Rosenstein, Petra Seibold, Jane Shortall, Elena Sperk, Nora Sundahl, Christopher J. Talbot, Ana Vega, Peter Vermeulen, Adam Webb, Catharine M. L. West, Liv Veldeman, Sofie Van Hoecke, REQUITE consortium
*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

Abstract

Purpose/objective Prostate cancer (PC) survivors frequently experience multiple co-occurring symptoms that adversely affect health-related quality of life (HRQoL). Identifying symptom clusters (SCs) may help to improve symptom management and patient care. The aim of this study is to investigate (1) SCs in PC patients, (2) associations of SCs with HRQoL, and (3) predictors of SCs. Material/methods We used data from an international, multi-centre, prospective cohort study (REQUITE). SCs were identified from patient-reported outcomes collected with the EORTC Core Quality of Life questionnaire (EORTC QLQ-C30) and pelvic symptom questionnaires. Machine learning techniques identified SCs, associations with HRQoL and SCs predictors. The dataset was divided into training (80%) and validation (20%) cohorts. Results Data were analysed from 1538 (before radiotherapy (T0)), 1490 (end of radiotherapy (T1)), 1322 (12-months (T2)), and 1219 (24-months (T3)) patients. SCs identified at T0: SC1 (gastro-intestinal), SC2 (fatigue, urinary, emotional and cognitive functioning), and SC3 (pain, physical, role, and social functioning). SCs changed at T1: SC1 (gastro-intestinal symptoms), SC2 (fatigue, urinary problems, insomnia), SC3 (social and role functioning), and SC4 (pain, bowel problems, physical, emotional, and cognitive functioning). At T2, symptoms returned to baseline clusters. SCs including 'fatigue' or 'urinary symptoms' were most frequent across time-points. At T0, T2 and T3, HRQoL was best predicted by clusters 2 and 3 (35-45% explained variance). At T1, cluster 4 was the best predictor (52% explained variance). Planned radiotherapy target volume, prostate specific antigen (PSA) at pre-diagnostic biopsy, age and alcohol consumption were the best predictors of SC2 at T1 and SC3 and fatigue-dyspnoea at T3. Conclusion Although SCs including fatigue and urinary symptoms were most common, the 'pain, bowel problems, physical, emotional and cognitive functioning' SC at T1 was associated most strongly with HRQoL. The predictors can help to identify men at risk for specific SCs.
Original languageEnglish
Article number3
Number of pages11
JournalHealth and Quality of Life Outcomes
Volume24
Issue number1
DOIs
Publication statusPublished - 29 Nov 2025

Keywords

  • Prostate cancer
  • Radiotherapy
  • Symptom clusters
  • Patient-reported outcomes
  • Machine learning
  • QUALITY-OF-LIFE
  • QLQ-C30

Fingerprint

Dive into the research topics of 'Using machine learning to predict patient-reported symptom clusters in prostate cancer patients receiving radiotherapy'. Together they form a unique fingerprint.

Cite this