Using artificial intelligence to predict when patients will die sounds like a science fiction movie plot, but researchers at Stanford University (USA) see this use of AI as a benign opportunity to help doctors and patients in this last stage.
Because many doctors often offer overly optimistic estimates of when their patients will die and delay the toughest or most difficult conversations. That human tendency, understandably on the other hand, can lead to patients receiving unwanted, expensive and aggressive treatment in a hospital, rather than being allowed to die peacefully in relative comfort. The alternative, which is being tested by a team at Stanford University, would use Artificial Intelligence to help physicians detect newly admitted patients who could benefit from palliative care options.
Previous studies have shown that most people would prefer to spend their last days at home if possible, but the truth is that more than half end up dying in a hospital while receiving aggressive medical treatment , according to research cited by experts.
Palliative care experts typically expect the medical team caring for a given patient to request their services, but Stephanie Harman, founding medical director of Palliative Care Services at Stanford Health Care, saw an opportunity to turn this routine around by giving Palliative care physicians the ability to proactively identify and communicate with patients.
The Stanford team’s AI algorithms are based on deep learning, the popular machine learning technique that uses neural networks to filter and learn from large amounts of data.
Researchers trained a deep learning algorithm on Electronic Health Records of approximately 2 million adult and child patients admitted to Stanford Hospital or Lucile Packard Children’s Hospital to predict the mortality of a given patient in the next 3-12 months. (less would provide very little time for the preparations needed in palliative care).
“We could build a predictive model using operational data routinely collected in the healthcare setting, as opposed to a carefully designed experimental study. The scale of available data allowed us to build a prediction model for mortality from all causes, rather than be a disease or a specific demographic factor “, explains Anand Avati, co-author of the work.
The pilot study’s use of an algorithm to predict patient mortality, which was approved by an institutional review board, turns out to be less reckless than one might think. From an ethical and healthcare standpoint, deep learning model assistance to help human physicians screen patients for palliative care generally has great benefits and few disadvantages.
“ We believe that keeping a doctor up to date and thinking of this as ‘machine learning + the doctor’ is the way to go rather than doing blind algorithm-based medical interventions, putting us on firmer ground both ethically and effectively. level of security “, clarifies Kenneth Jung, researcher at Stanford University.
The palliative care team is primarily concerned with accurately identifying patients who might benefit from their care, rather than having to know exactly why the algorithm predicts that a given patient might die within a year.
“That is why in this particular case we feel more comfortable with having a black box model. The palliative care intervention is not linked to why someone is getting sick,” Jung clarifies. Still, it can be helpful to know why the deep learning model made its predictions for research purposes. In this case, the Stanford group used a common error analysis technique, called ablation analysis, to provide insight into the decision-making of the deep learning model. Their method involved adjusting the model bit by bit by adjusting individual parameters to determine what impact those parameters had on model decisions.
In the end, the deep learning model’s approach to predicting death is far from sinister . Mortality simply turns out to be a useful measure in determining the best time for patients to receive a visit from the palliative care team.
“We want to make sure that the sickest patients and their families have a chance to talk about what they want to happen before they become seriously ill and end up in the ICU,” Jung says.
Referencia: Improving Palliative Care with Deep Learning. IEEE International Conference on Bioinformatics and Biomedicine 2017 Anand Avati, Kenneth Jung, Stephanie Harman, Lance Downing, Andrew Ng, Nigam H. Shah arXiv:1711.06402
