Oral Presentation Cancer Survivorship Conference 2023

Machine learning insights into Victoria’s Cancer Council 131120 information and support service (#7)

Nikki McCaffrey 1 2 , Sajani Ranasinghe 3 , Gihan Gamage 3 , Harsha Moraliyage 3 , Nishan Mills 3 , Jessica Bucholc 1 2 , Katherine Lane 2 , Angela Cahill 2 , Victoria White 4 , Daswin De Silva 3
  1. Deakin Health Economics, Deakin University, Burwood, Victoria, Australia
  2. Cancer Council Victoria, Melbourne, Victoria, Australia
  3. Centre for Data Analytics & Cognition, La Trobe University, Melbourne, Victoria, Australia
  4. School of Psychology, Faculty of Health, Deakin University, Melbourne, Victoria, Australia

Aims: Charities such as Cancer Council Australia provide telephone information and support services to address unmet information and psychological support needs of people affected by cancer, including cancer survivors. Machine learning techniques, i.e. novel Artificial Intelligence (AI) algorithms and Natural Language Processing/Understanding (NLP/U), can analyse large volumes of call recordings to provide insights into the multifactorial context and rationale for contacting the service, interactions and outcomes. This analysis explores the key themes arising during interactions with the 131120 service, the relationship between themes and how these themes vary by cancer stage.

Methods: A custom AI framework composed of five modules was formulated to analyse the call recordings: speech-to-text pipeline for call recording transcription; acoustics pipeline to identify nurse and caller speech; unified data model to integrate un/structured data; AI-generated emotion profiles and transitions; and NLP/U for lexicon and learning-based insights (thematic analysis). The framework was applied to a sample of 18,336 call recordings (Jan18-Dec21) to generate key themes arising during interactions with the nurse.

Results: Most calls were from people living with or surviving cancer (42%), carers (30%), and the general public (27%). Overall, 44% calls lasted more than 15mins. Key themes included: treatment, mental health, family, COVID-19, conditions, donations, legal issues, frauds and scams, education, employment, household, insurance, side-effects and finances. Treatment, family matters, and mental health were discussed most frequently, particularly in the earlier stages of the cancer journey compared with remission or stable disease. In 2021, COVID-19 was discussed in 26% of calls (n=3,291), most often alongside side effects and finances.

Conclusions:The findings suggest most callers contact Victoria’s Cancer Council 131120 about treatment, mental health and family issues. This study illustrates the technical capability and practical value of machine learning techniques to provide insights into the psychological and information support needs of people living with cancer.