Internet Explorer is no longer supported by Microsoft. To browse the NIHR site please use a modern, secure browser like Google Chrome, Mozilla Firefox, or Microsoft Edge.

Algorithms as good as doctors at predicting death in advanced cancer

Published: 30 April 2021

New research funded by the NIHR shows that algorithms can be as accurate as clinicians’ judgements for predicting how long terminally ill people with advanced cancer have left to live.

Two studies published in the journal PLOS ONE directly compare, for the first time, commonly used algorithms for predicting survival length (prognosis) against clinicians’ predictions, and combine interviews with patients, carers and healthcare professionals to assess the perceived usefulness of prognosis tools.

The research papers – Prognostic tools or clinical predictions and The Prognosis in Palliative care Study II – compared the effectiveness of five prognosis tools with clinician judgement in 1833 adults with incurable cancer, who had recently been referred to community, hospital and hospice palliative care services across England and Wales. Interviews were held with 29 patients, 20 caregivers and 32 clinicians. 

The findings of the work, led by the Marie Curie Palliative Care Research Department at University College London (UCL), show that at least two of the studied algorithms were equally as good as the combined judgement of an experienced doctor and nurse in predicting whether patients would live for days, weeks or months. All of the patients in the study had advanced cancer, with no further curative treatment options, and an average survival of only 45 days.

The lead researcher explained that although the best algorithms were as accurate as expert clinician judgement, the overall accuracy was only just over 60%. The research concludes that while prognosis algorithms are currently no more accurate than clinicians’ judgments they may be helpful in other ways such as providing a supplementary second opinion or helping less experienced staff navigate difficult conversations with patients and their loved ones about their prognosis and end of life plans.

Lead researcher, Professor Paddy Stone, from the Marie Curie Palliative Care Research Department at UCL said: “Even when patients are terminally ill, having an accurate prognosis can be extremely valuable. Knowing whether one only has days, weeks or a few short months remaining can help prepare for the end of life and to focus on achieving the best possible quality of life. 

“Although clinicians can recognise when patients with advanced cancer have become terminally ill and will no longer benefit from life-prolonging treatment, they still cannot be sure whether the remaining time is to be measured in days, weeks or months. For clinicians, knowing an accurate prognosis can help with planning care and with facilitating access to services. For example, knowing when someone is likely to die can unlock faster access to welfare benefits and extra social care support.

“Despite its limitations, professional judgment is still the best way to predict how much time remains for people who are terminally ill. Algorithms may, however, have a role in supplementing the judgments that clinicians make, or maybe helpful in guiding less experienced professionals. I would not recommend that prognosis algorithms should be used by patients themselves, since the results still need to be interpreted in the light of clinical judgement. As is often the case, good communication skills are essential. Open and honest conversations can help people to live the best quality of life and clinicians should not shy away from recognising uncertainty.”

The project was funded by the NIHR Health Technology & Assessment Programme. 

Latest news