The potential applications, including mobile apps and wearable technology, could be used by clinicians and family caregivers to improve terminally ill patients’ quality of life.
“By monitoring key symptom changes in real time using machine learning, we can enable early interventions to pre-empt and prevent the patient’s deterioration through the phases of illness,” says the lead author of the study Dr Margaret Sandham at Auckland University of Technology (AUT).
In New Zealand, palliative care is provided across different environments by any number of people, including hospital and hospice staff, district nurses, general practitioners, and whānau.
“Integrating patient-reported outcome measures into routine care, in an ethically responsible way, would contribute to common knowledge of patient symptoms and experiences, and a shared language between clinicians and family caregivers,” says Sandham.
AI and machine learning have been underutilised in palliative care.
Machine learning uses algorithms, a sequence of well-defined computer instructions, to identify patterns in data. It is a branch of AI that enables self-learning from data and applies that learning without the need for human intervention.
Sandham, Professor Ajit Narayanan and Professor Richard Siegert at AUT, together with co-authors from Waitematā DHB and Kings College London, collaborated on the study which aimed to identify whether machine learning could predict changes in patients’ clinical status using the Integrated Palliative Care Outcomes Scale (IPOS).
Anonymised self-reported symptoms from 800 adults, who were enrolled in palliative care services in New Zealand, was analysed through a combination of statistical tools, machine learning, and network visualisation.
Sandham and colleagues identified the variables for predicting transitions between phases of illness using six machine learning techniques. Network analysis of these variables revealed that poor appetite and loss of energy are the most critical symptoms.
Poor appetite was linked to nausea, vomiting, constipation and sore and dry mouth. While loss of energy was linked to drowsiness, shortness of breath, and lack of mobility.
“This is the first time palliative-specific data has been used in machine learning, as far as we are aware,” says Sandham.
Previous research in this area used general patient data and focused on mortality rates and survival prediction, rather than identifying possible changes in clinical status and the type of response needed to improve the quality of life for terminally ill patients.
This study is important in part because it challenges us to think of ways that technology can improve our care.
The findings of the study, Intelligent palliative care based on patient-reported outcome measures, were published in the Journal of Pain and Symptom Management.
These preliminary results indicate that future digital therapeutics in palliative care, based on mobile apps and wearable devices, could focus on sensors dealing with pain, nausea, mobility, weakness, and shortness of breath – generating a data feed linked to a patient profile, to be used by specialists and non-specialists alike.
“Automatic hourly readings from different sensors can be a non-intrusive method for detecting changes in clinical status, often in advance of periodic clinical assessments or patient-reported outcomes,” says Sandham.
“Such is the progress in these systems and technologies that the question for palliative care is not whether they will be used, but when and how.”
Sandham is a senior lecturer in Nursing at the AUT School of Clinical Sciences. Her research draws from dual careers in intensive care nursing and clinical psychology. And specialist knowledge of psychometrics (measuring mental capacities and processes) and Patient-Reported Outcome Measures (PROMs) has led to research collaborations in palliative case and mental health.
The role of PROMs in palliative care has become increasing important in line with rapidly ageing populations. Yet research into the use of PROMs in routine clinical care of palliative patients is limited.
“Another complexity has been converting any understanding of reasons for key symptom changes into a practical monitoring solution without requiring constant and possibly intrusive measurement,” says Sandham.