Is the prognosis favorable?

Google learns to predict the duration of illness and the risk of death

Anna Kerman, XX2 century, according to Bloomberg: Google Is Training Machines to Predict When a Patient Will Die

Google Company "taught" the neural network to predict how many days the patient will spend in the hospital, as well as to assess the likelihood of re-hospitalization and even death of the patient. In the work, the algorithm uses not only data processed by medical statisticians and presented in a form that is convenient for perception, but also information that was previously considered practically useless, for example, notes in PDF documents. Google's development not only provides forecasts, but also – if necessary – shows which data helped the neural network to come to such, and not other conclusions. Article by Rajkomar et al. Scalable and accurate deep learning with electronic health records is published in the journal Digital Medicine.

Doctors, medical institutions and healthcare organizers have been trying for many years to work with electronic medical records and other data related to the health of patients. The development of a tool that would ensure the appearance of certain information "at the right time and in the right place" could save the lives of patients or, at least, reduce the number of hours spent daily working with documentation. Doctors could spend less time filling out charts and reports and pay more attention to patients.

Google hopes that a new development of the company can become a "lifeline" for medical institutions drowning in electronic document management. The creators of the "medical AI" hope that soon the predictive system will be implemented in clinics, where it will do its direct work – it will predict with high accuracy the development of symptoms and the course of the disease, which will meet with hope or as an alarm signal.

However, the prospect of including the "medical brain" directly in the therapeutic and diagnostic process raises understandable concerns. Won't the system push doctors to make certain diagnoses and prescribe certain medications? Will AI be able to take into account all the factors that matter to a particular patient? In other words, won't the work of the "medical brain" do more harm than good?

One of the employees Google, which is involved in the development of AI, told the agency on condition of anonymity Bloomberg that the models used in the system ignore obviously significant events in the patient's life from a medical point of view. For example, the system does not take into account whether the patient has had surgical operations in the past. The employee called the existing models with manually developed code "an obvious, giant brick" in the treatment.

Although Google's management is optimistic about the potential of the new development, the preparation of an AI capable of improving treatment outcomes remains a colossal (and still unresolved) task. Relatively recently, another IT giant, IBM Corporation, tried to create an "electronic doctor". He received the name Watson and is currently really "working" in medicine. However, he is looking for new ways to reduce the cost of treatment.

To "train" the new AI, Google used information obtained from university hospitals of California, San Francisco and Chicago. After processing 46 billion anonymous case histories, the system was able to produce a predictive model for each hospital individually. Creating a model suitable for two institutions at the same time turned out to be a much more difficult task. It is obvious that the development of a universal solution, theoretically applicable in any clinic in the world, will cause even greater difficulties. But we should not forget that Google has accumulated a huge amount of information concerning both individuals and entire countries during its existence. And by combining data on traffic congestion with a personal medical history, Google will be able to assess the likelihood that the patient will get to the hospital safely in the future.

So, so far we do not have an affordable global solution that can take into account all factors related to health in one way or another, from search engine queries to last year's dentist's conclusion. But less ambitious (and more practical) projects have already been launched and are working. For example, you can estimate the probability of your death "according to" AI today: this will only require an iOS gadget and a fitness tracker.

Portal "Eternal youth" http://vechnayamolodost.ru