We found a way to know who is going to die, when they are going to die, and how.
I distinctly remember that day as if it happened yesterday. It was in autumn almost five years ago that my brother called to inform me that the doctors had given their prognosis: our father was unlikely to survive the week. They said we should prepare for the eventuality. While my father had been in and out of hospitals a number of times over the prior year, there had been no indication that his health would deteriorate to this level so soon. Collectively everyone in my family expressed one wish that we had more time to prepare, that we had been forewarned, that we could have structured the last few months or even last few weeks of my father’s life differently. Death is an inevitable part of life. Yet how we react to and care for death, especially in old age, can be partially controlled by us. This forces the important question of why and how in an era of AI and machine learning can we help millions of families like mine to better cope with the potential loss of a loved one.
In the US a large percentage of elderly patients on Medicare pass away in acute care hospitals. For many, the last six months of life are characterized by complex medical procedures, repeated emergency visits, and frequent hospital stays. This corresponds to significant costs to the taxpayers with marginal benefits to the patients and their families. In most cases, one gets a very small window of time to prepare for and care for the patient about to expire. Thus, many patients and their families do not have a sense of control over the last stages of their lives. In surveys, a majority of Americans, around 70 percent, expressed their wish to ideally die at home; living their normal way of life, surrounded by their loved ones. Palliative care services are designed to provide alternatives to hospital-based medicine for patients in their last months of life. The number of patients who utilize these services has been growing steadily so that it now comprises of almost half of people (48%) who die annually. However even among those who utilize hospice services, the average length of palliative care is a mere 26 days. Just 26 days.
We hope that by assisting physicians to predict the onset of mortality risk for elderly patients can increase their ability to have conversations with patients and families sooner and improve the overall quality of care towards the end of life. This week we will be presenting some key insights from our research on predicting mortality risk six months to one year out at the Thirtieth Conference on Innovative Applications of Artificial Intelligence held at annual AAAI (Association for the Advancement of Artificial Intelligence) conference. Using extracts of data from Electronic Medical Records and Patient Claims we developed various machine learning models to predict such risk of mortality. In many healthcare settings there is a reluctance to use machine learning models because a large number of such models are unfortunately what we term as ‘black-box’ models, i.e., it is not always clear to even a human expert in that domain, in our case physicians, why the model is making certain predictions. To address this problem, KenSci has invested significantly in developing models which not only have a high predictive power, but are also interpretable. This ensures that using the KenSci platform, accompanying each prediction the machine learning system also specifies why the prediction was made, and which factors might have been important to the system that made the prediction. Care providers can thus meaningfully consider the risk recommendation and objectively decide to accept or ignore such recommendations given their own expertise and understanding of the larger picture. This is one way to make the underlying machine learning system more assistive and trustworthy.
Interpretable models, particularly within healthcare, area growing area of interest for the AI community, and at KenSci Research our vision is to foster global research partnerships with health systems and academic research to integrate best-practices in healthcare with state of the art in AI leading to broad social impact. Our recent efforts on interpretable risk of mortality models is one such effort.
While we have no doubt that advances in AI will one day help patients experience a more fulfilling life, my thoughts often return to that day in autumn. I still feel the loss, but I also feel a great sense of hope that by contributing towards the advancement of what we at KenSci term Assistive Intelligence will one day enable physicians to make and also communicate their decisions much sooner giving families more time.
The Association for the Advancement of Artificial Intelligence (AAAI), announced that KenSci’s academic research paper has been recognized as EMERGING and will be presented at the 2018 AAAI conference’s track Innovative Applications of Artificial Intelligence 2018 (IAAI-18) in New Orleans, Louisiana. The paper dives into how machine learning techniques are used to predict the risk of mortality for patients from two large hospital systems in the Pacific Northwest along with the explanations for end of life predictions and insights that are derived from the predictions can then be used to improve the quality of patient care towards the end of life.
Download the paper here: http://bit.ly/Death-Vs-DataScience-Research-Paper
