Leveraging Medical AI to Assist VTE Prevention and Treatment, Dr Mayson is Trying to Solve the Problem of Pulmonary Embolism Prevention and Treatment with New Technology


Since the late 1980s, surgeons such as Caprini have been working to design an extremely detailed individualized venous thromboembolism (VTE) risk assessment model and to continually optimize its accuracy and clinical availability. However, more than 40 years have passed, and pulmonary embolism remains an important cause of unintended death in hospitalized patients.

The high mortality rate is still low in patients receiving VTE prophylaxis, and the doctors ignore the subjective and objective aspects of the VTE risk assessment model. Under normal circumstances, if the medical staff can accurately assess the VTE risk of the patient, the high-risk patients with VTE can be found in time and appropriate preventive treatment can be given. But in reality, the implementation of this set of scoring system has encountered many problems.

“It’s too time-consuming.” When asked why the various VTE assessment scales did not play their proper preventive effects, the doctors at a top three hospital in Beijing replied without thinking, “The ward of a large hospital can not be an extra bed. Miracles, even if doctors and nurses have the heart to prevent, they can’t seriously assess the situation of each patient. What’s more, the problem of VTE is not just the omission or irregularity of medical staff. Sometimes, it may be that doctors, especially young doctors, are not enough guides. Familiar with, there is not enough evidence-based judgment.”

The problem of VTE prevention is not only in large hospitals. In fact, the problems faced by primary medical institutions may be more serious. Many doctors in primary medical institutions do not have VTE’s prevention awareness and discriminating ability.

There are many problems. Fortunately, those who are willing to actively seek solutions are not afraid of challenges.

VTE prevention, what are the key issues?

VTE is a preventable disease. We can’t wait until the thrombus is formed to solve the problem. If we can carry out the early screening and early prevention of VTE, the difficulty of prevention and treatment of VTE will be solved. Among them, the key is to integrate enough patient information, achieve automatic capture and give reasonable judgment, and simplify the clinical workflow.

In fact, automated hospitals that have access to patient information are available, but the primary information system can only capture basic information such as the patient’s name, age, and gender. For more important and more diversified descriptions of the condition, depending on the writing habits and formatting of different doctors, most of them need to be filled in manually by doctors. This does not solve the actual clinical problems. It does not help much to reduce the workload of doctors, nor can it solve the problem of mis-evaluation and missing assessments that may exist in the manual completion of the evaluation form.

Automated analysis of patient care process information is challenging. “For example, let the computer judge the patient’s VTE risk factor information for the occurrence of cerebral infarction within a month, not only to help the machine understand the time limit of cerebral infarction – the event should occur in the near future as given One month, the cerebral infarction event of three months or half a year is not counted. At the same time, it is necessary to judge the patient’s disease status, such as grading, and often need to synthesize information such as past medical history and nuclear magnetic report to draw conclusions. The traditional information system cannot be completed. Such complicated work.” Zhang Qi, CEO of Beijing Huiyun Technology Co., Ltd. (hereinafter referred to as Dr Mayson) explained to the reporter of the arterial network.

The emergence of deep learning has brought a turn for big data processing problems. Emerging natural language processing (NLP) technology can realize the semantic understanding of medical record data, structure and standardize descriptive natural language, and deep learning algorithm can also incorporate personalized medical record data into rule base, combined with clinical guidelines. Various scales establish VTE knowledge maps, improve the accuracy of the final machine model, and solve the two problems of “difficult data entry” and “difficult data analysis” faced by doctors.

However, NLP also has its limitations. If the developer is not familiar with the VTE diagnostic process, there is no good or complete data on the patient’s past medical history, medication history, or a reasonable knowledge map. The accuracy of the results will be Greatly affected. Building an effective NLP is actually not that simple.

Based on CDSS, Dr Mayson opens up the NLP battlefield

Applying NLP to VTE control is not easy. Take Dr Mayson, who owns VTE quality control products, as an example. The whole product is full of twists and turns.

With four years of algorithm development and knowledge mapping construction, clinical expert intervention, and policy-driven efforts, Dr Mayson’s VTE quality control system development has advanced rapidly. Zhang Qi told the arterial network reporter: “The whole system was built in May 2019. In just 4 months, Dr Mayson created a VTE intelligent control system with an accuracy rate of over 97%. This success is inseparable from Dr Mayson. The hard work is also inseparable from the support of the cooperative hospital.”

When talking about the development project for four months, Zhang Qi couldn’t help but feel a bit: “When we made the first generation in August, its accuracy was less than 80%. At the time we also faced many problems.”

The combination of CDSS system and clinical data is the first difficulty that every enterprise will encounter. Fortunately, Dr Mayson has been very skilled in handling CDSS access steps. The real trouble is to standardize medical record information. Because the medical record data is mostly unstructured text description, different doctors have their own writing habits, the terminology used is standardized, the degree of identity is low, and the multi-source heterogeneity is outstanding. Dr Mayson uses a common form to “interpret” the different information to the machine so that the computer accurately understands its semantics. Zhang Qi said: “Even if the team has been rooted in clinical for several years, and has incorporated hundreds of thousands of top three hospitals with real medical record training AI model and achieved good recognition results, we again iterate and validate the model for the real patient medical records of the cooperative hospital, the whole process It took up to three months to stabilize the recognition accuracy at over 97%.”

It is worth mentioning that Dr Mayson’s platform already has a fairly strong generalization ability. Even if the doctor’s report does not strictly follow the standard, every AI can accurately capture patient information and generate accurate results.

Remind the doctor to implement VTE early screening and early prevention

In practice, even with the help of VTE quality control system, doctors will inevitably miss some patients’ VTE risk assessment during busy work. In this case, the smart reminder of each VTE quality control system comes in handy.

“Our goal now is to advance the screening time. When the patient is just admitted to the hospital, he will help the doctor to solve the relevant problems and move the quality control point forward. The first 24 hours of admission is an important evaluation period. If the doctor has a omission, then within 24 hours, we will use the CDSS in the electronic medical record to remind the doctor that the patient has not completed his VTE risk assessment for admission.”

In order to ensure timely data statistics and feedback, Dr Mayson also set up a statistical management platform in the hospital to record the VTE prevention behavior of each patient, so that the patient information can be clearly presented to the responsible doctor, and the situation can be clearly seen at a glance. .

Taking the China-Japan Friendship Hospital as an example, Dr Mayson’s post-mortem big data statistical analysis platform is projected on the large screen of the nurse station. The nurse can visually understand the patient’s bed information and VTE assessment and preventive implementation at a glance.

In addition, Dr Mayson is also trying to get VTE to prevent and cure the grassroots hospitals.

Unlike the top three hospitals, the level of informationization in the primary medical institutions is not high, and doctors lack the ability to judge VTE. In this case, Dr Mayson’s VTE quality control system is mainly used for standardization and doctor education. In this process, the doctor’s ability to prevent and control will gradually increase.

The promotion of VTE prevention comes from the combination of multiple forces

On October 13, 2018, the “National Pulmonary Embolism and Deep Vein Thrombosis Prevention and Control Capacity Building Project” kick-off meeting officially approved by the National Health and Health Commission Medical Administration and Hospital Authority and the “China Guidelines for Prevention and Treatment of Thrombotic Diseases” was held in Beijing. The Japan Friendship Hospital held a strong push for VTE prevention. This is also an important opportunity for companies such as Dr Mayson to develop VTE quality control systems.

Many hospitals have also responded to policy requirements by establishing a prevention and management team, establishing a VTE prevention and management supervision mechanism, conducting discipline construction and health education, and establishing an in-house VTE prevention and control system.

However, whether it is the advancement of in-hospital informatization or the doctor’s awareness of VTE prevention and control, it will be a long battle. In this case, governments, enterprises, and hospitals all need to work together.


Please enter your comment!
Please enter your name here