Why are hospitals not interested in AI?

Mondo Health Updated on 2024-01-30

The industry is looking to clearer regulatory standards to guide the way.

textXin YingedWang Xiao

If there is a medical AI with complete and exhaustive knowledge of all the medical literature, with billions of hours of clinical experience, it is always valuable and extremely cheap.

How long does it take for a company to develop?Will the hospital pay for it?

In an October 2023 presentation, Ilya Sutskever, Chief Scientist at Open AI, said, "AI will have a huge and incredible impact on the medical field." All parties are looking forward to the application of AI in the medical fieldAccording to incomplete statistics, domestic medical treatmentaiThe number of large models has been exceededPiece. According to data released by CCID Research Institute on December 14, the market size of China's generative artificial intelligence (AIGC) is expected to exceed 10 trillion yuan in 2023.

However, at this stage, the implementation and commercialization of large medical models are still in the early stage of exploration. In China's most important medical service scene, public hospitalsEven if it's provided for freeaiLarge-scale model productsNot necessarilyYesEnterHigh wallsWithin, this is also Chinese medical careaiIt has been at a low point for a long timeThe real side

According to "Finance and Health", the industry authorities have been working on the statistical research on the application of AI in public hospitals. On December 9th, it was hosted in "Finance and Health".aiHelping the Healthcare Industry to Transform Forward".The guests at the meeting conducted in-depth discussions on topics such as policy support, landing applications, and commercialization paths for AI medical care.

In 2023, newly released medical AI models will enter hospitals one after another, providing pre-diagnosis consultations for patients, helping doctors write medical records, and even providing advice to doctors when they encounter incurable diseases.

After the upgrade, doctors are more willing to accept it. Taking the most common record of blood pressure as an example, in the previous intelligent voice entry products, if the doctor dictated the blood pressure, most of the systems would only record it so bluntly, and the medical measurement unit had to be manually completed. Some doctors believe that it is not as convenient as typing directly.

The large language model is an artificial intelligence model based on deep learning Xi natural language processing, so the tasks that natural language processing can complete are the fields that large language models are good at, but they should be supplemented by sufficient medical data training.

Zhang Xu, deputy director of the Information Department of Beijing Friendship Hospital affiliated to Capital Medical University, introduced that the large language model product pre-trained and fine-tuned through the hospital's data will automatically generate standard medical written language from the oral expression of the doctor's consultation, such as the above situation, it will be expressed as "systolic blood pressure 120mmHg, diastolic blood pressure 80mmHg", and doctors can give higher quality cases in less time.

New tools allow clinicians who feel the benefits to move from passive users to supporters.

InFully integratedMedical greatmodelInternet Hospital, the effect of "burden reduction" is clearerVisibleZhang Songyue, senior R&D manager of the Intelligent Algorithm Department of JD Health Technology Product Department, introduced that since the launch of the "Jingyi Qianxun" large language model in July 2023, the time for doctors to write medical records has been reduced by 50%, the efficiency of pharmacists reviewing prescriptions has been increased by 200%, and the real-time prompt function during diagnosis has also reduced the core consultation questions of patients by 10%.

Improving the efficiency of doctors and increasing the supply of medical services through technology is more urgent in the context of the accelerated aging of the population. Expectedyears or so,The elderly population aged and over will break throughbillion, and the proportion of the total population will exceed, entering the stage of severe aging.

Chen Hu, deputy director of the Medical Management Center of the National Health Commission, pointed out that under the aging trend, in addition to continuing to expand high-quality medical resources, it is also necessary to consider optimizing and adjusting the existing medical model to adapt to the medical increment and active health service demand generated under the new population structureaiThe application will provide valuable solutions to further improve labor productivity in the healthcare field, enabling society to obtain better health care at a lower cost.

countriesMedicineChen Jinfu, former deputy director of the Insurance BureauIt is also pointed outDriven by the technological revolutionThreeMedicineThe key to linkage high-quality development is to let technology get rid of the shackles of human skills, get out of the walls of traditional medical service hospitals, and serve patients in a seamless and homogeneous way.

Although public hospitals have begun to introduce large medical models, and even cooperate with enterprises to participate in research and development, largeMany hospitals are only trying out a certain feature in some departmentsMajoraiMakerforCovering the whole process of medical consultation and scientific researchDesign outproducts, a lot useless.

What kind of "martial arts" are needed to enter the hospital smoothly and use the medical model for it?

After a period of exploration, Zhang Xu summed up four key points of hospital demand"The first thing that needs to be privatized is to ensure the privacy and security of hospital data;Secondly, there are application scenarios and corresponding special fields;In addition, high-quality medical knowledge data, clinical diagnosis and treatment data, etc.;In the end, it has to be low-cost, high-quality, and high-output. ”

All AI products are under the banner of "truly solving clinical needs", but some seemingly perfect products will not be effective when they are applied. This is a "pit" that many hospitals have walked through in the process of informatization construction.

The system bought with money will be upgraded as soon as it goes online, and some of them cannot be used at all after going online, and a set of systems does not meet expectations, and many hospitals have spent money on buying several sets of systems to "patch" later.

After several tosses, even though they have tens of millions of funds for informatization construction to be used every year, hospital directors are becoming more and more cautious in spending money.

aiIn the future, the application scenarios in hospitals must be very wide, but what kind of needs can be solved in hospitals, whether it is compliantHospital's castinput-output ratio, neededPartiesmoreDepthto consider. ”Li Tianqing, deputy secretary of the Party Committee and secretary of the Discipline Inspection Commission of Fuwai Hospital of the Chinese Academy of Medical Sciences, said.

Hospitals have real needs, but many companies hold the "hammer" of technology, but they can't find the right "nail" in the hospital. Chen Weiheng, vice president of the Third Affiliated Hospital of Beijing University of Chinese Medicine, introduced that companies that provide 3D printing products are everywhere, but few companies can find an accurate location in the hospital, such as the simplest and most common "orthopedic splint", which is a product with a low cost, but there is none. "We are now exploring a personalized and intelligent upgrade of this product to provide better service products if it is acceptable to patients."

For businesses, there is another difficultyIn the actual landing process, even for a clear application scenarioHospitalsManagement, Information Section, Department Heads, Clinicians, patients,The need for a product is different at every levelFor example, patients focus on experience, medical staff focus on process, and managers focus on cost.

Managing the demands of a hospital can be challenging for product design. Some AI medical product practitioners said frankly.

Misalignment not only comes from within the hospital, but also from outside the hospital, the lack of patient confidence, the lack of standards for AI products, and legal and regulatory challenges will affect the implementation of products.

Zhen Xiantong, AI research director of Beijing United Imaging Intelligent Imaging Technology Research Institute, feels that in the process of exploring large models for different medical scenarios, technology and scene misalignment are often encountered. Since the development of large models requires a large amount of pre-training, which puts forward very high requirements for data and cost, the things that are technically easy to implement in the development process may not necessarily be the "pain points" in the medical scenario. On the contrary, the urgent clinical needs sometimes cannot be carried out smoothly and quickly due to the lack of various elements required for the development of large models, such as data annotation, etc., so it is necessary for the industry and hospitals to continue to run in and jointly explore products that are suitable for the actual situation and are truly used in clinical practice.

Liu Guoliang, chief physician of the Department of Respiratory and Critical Care Medicine of China-Japan Friendship Hospital and chairman of the Medical Artificial Intelligence Expert Committee of the National Center for Telemedicine and Internet Medicine, reminded that AI has certain advantages in disease diagnosis and screening with clear diagnosis and treatment criteria, and the purpose of current medical AI research and development is not to develop for the purpose of comprehensive outcomes or prognostic criteria for patientsFor patients with diseaseThe impact on outcomes and prognosis needs to be evaluated objectively, and there is currently an experienced physician in terms of reassuring and relieving distress in the process of communicating with patientsaiThe app can't reach it, andaiThe product has not yet shown a level of intelligence in this regard, and it often increases patient anxiety, which is worth paying attention to.

A number of experts at the meeting said that in the face of many resistances, the hospital's current choice is not to pursue the most cutting-edge AI products, and the need to maintain the operation of the hospital is the most important, because the high technology content also means higher adaptation costs, and the hospital may not want to take this risk.

With the accelerated iteration and evolution of various AI medical models, the commercialization prospect is expected to be further opened, and the cost of enterprises will be more controllable.

Taking large medical models as an example, products that originally required a small team to complete can now be trained by only half a person to train a high-quality large model. As a result, companies are more willing to invest more.

According to the research report of Huaxi **, the medical technology leader has elements such as technology and industry know-how, and the long-term cooperative relationship with the head medical institutions is conducive to completing the data, scenarios and other elements, and related products are expected to be the first to land.

The key is to find a clear entry point for commercialization.

"We still have to start with the pain points of hospital management. ”Jiang Mengxi, director of the Hospital Management and Reform Research Office of the Health Development Research Center of the National Health Commission, suggested that in the face of strong supervision and strict assessment of hospitals, one of the pain points of hospitals at present is the lack of an intelligent doctor evaluation system, which can make good doctors better and better, so that bad doctors can become better and better, and improve hospital management efficiency.

From the perspective of existing products, AI medical care mainly includes three levels: the basic layer, the technical layer and the application layer. The basic layer includes data, algorithms, computing power, etc., with many participants and an oligopolyThe technical layer includes CV, NLP, intelligent voice and other technologies, and the participants are relatively matureThe application layer is aimed at various applications such as AI medical imaging, CDSS, medical robots, and medical data intelligence platforms.

For large, technologically robust enterprisesIt will be approached from multiple levels at the same time

Wu Wenjing, deputy general manager of Ant Group's medical and health business department, said that with the gradual deepening of Alipay's cooperation with public hospitals and first-class units, and the gradual development with the help of Ant's self-built AI engine, Alipay has actively explored AI applications in many cities and cities to help patients and doctors experience medical treatment process scenarios, and has successively implemented practical cases such as medical insurance policy assistants, digital escorts, and doctor assistants, and will continue to explore innovative paths in the future.

In the face of the concerns of hospitals, it can also be approached from third-party cooperation. He Qing, senior architect of smart healthcare, said that it is trying to innovate in business models, such as cooperating with pharmaceutical companies to better enter the medical service market, in addition to consulting services for about 200 million patients per day on its own platform, it is also strengthening the services of offline hospitals.

There are also opportunities in subdivisions outside the top three. Kong Xiangpu, Senior Vice President of WeDoctor Group and Deputy Director of RuiYi Artificial Intelligence Research Center, introduced that it is providing intelligent examination and medication programs to the grassroots doctor market, introducing the experience standards of the top three into the grassroots to help better manage diabetes patients.

In order to continuously improve the level of AI products for assisting doctors, Lian Zeliang, head of AI medical business of Medical Union Group, said that it is expected that with the development of the industry, the boundaries of supervision will be clearer, and the product requirements will be more standardized, so that enterprises can clearly know the direction of exploration and research and development.

Related Pages