End-of-life care can be stressful for patients and their loved ones, but a new algorithm could help provide better care to people during their final months.
A paper published in arXiv by researchers from Stanford describes a deep neural network that can look at a patient’s records and estimate the chance of mortality in the next three to 12 months. The team found that this serves as a good way to identify patients who could benefit from palliative care. Importantly, the algorithm also creates reports to explain its predictions to doctors.
Palliative care is a growing trend in the U.S. It can make the end of someone’s life much less painful, and it can usually be done at home. Even as such care becomes more widespread, though, the researchers note that although 80 percent of Americans say they would like to die at home, only 20 percent end up getting to do so.
The paper points out that a shortage of palliative-care professionals means patients face delays in being examined for services, so using an algorithm could help overstretched doctors focus on patients in the greatest need.
The system works by training on several years’ worth of electronic health records and then analyzing a patient’s own records. It generates a prediction about the patient’s mortality, as well as a report for doctors to review about how it came to its conclusion. This includes details on how much certain factors—like the number of days someone has been in the hospital, the medications prescribed, and the severity of the diganosis—played into its prediction. The results have so far been positive, and the algorithm is being used in a pilot program at a university hospital, though the team didn’t say where.
As we have noted before, doctors are much more likely to trust and accept an automated system if they understand its reasoning.
Andrew Ng, a coauthor of the paper and the former head of AI research at Baidu, has worked on previous automated systems that have been shown to outperform doctors in diagnosing lung diseases and spotting heart arrhythmias. But the addition of a clear way to explain the machines’ superhuman abilities may be the most valuable advance yet.
Source : https://www.technologyreview.com/the-download/609574/a-new-algorithm-identifies-candidates-for-palliative-care-by-predicting-when/
Improving the quality of end-of-life care for hos- pitalized patients is a priority for healthcare organizations. Studies have shown that physicians tend to over-estimate prog- noses, which in combination with treatment inertia results in a mismatch between patients wishes and actual care at the end of life . We describe a method to address this problem using Deep Learning and Electronic Health Record (EHR) data, which is currently being piloted, with Institutional Review Board approval, at an academic medical center. The EHR data of admitted patients are automatically evaluated by an algorithm, which brings patients who are likely to benefit from palliative care services to the attention of the Palliative Care team.
The algorithm is a Deep Neural Network trained on the EHR data from previous years, to predict all-cause 3-12 month mortality of patients as a proxy for patients that could benefit from palliative care. Our predictions enable the Palliative Care team to take a proactive approach in reaching out to such patients, rather than relying on referrals from treating physicians, or conduct time consuming chart reviews of all patients. We also present a novel interpretation technique which we use to provide explanations of the model’s predictions.
We demonstrate that routinely collected EHR data can be used to create a system that prioritizes patients for follow up for palliative care . In our preliminary analysis we find that it is possible to create a model for all-cause mortality prediction and use that outcome as a proxy for the need of a palliative care consultation. The resulting model is currently being piloted for daily, proactive outreach to newly admitted patients. We will collect objective outcome data (such as rates of palliative care consults, and rates of goals of care documentation) resulting from the use of our model . We also demonstrate a novel method of generating explanations from complex deep learning models that helps build confidence of practitioners to act on the recommendations of the system.
We thank the Stanford Research IT team for their support and help in this project. Research IT, and the Stanford Clinical Data Warehouse (CDW) are supported by the National Center for Research Resources and the National Center for Advancing Translational Sciences, National Institutes of Health, through grant UL1 TR001085. The content of studies done using the CDW is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
Source : https://arxiv.org/pdf/1711.06402.pdf