Author | Utada
I sat in Beijing Tongzhou Liyuan Community Hospital for 4 hours.
Even on Saturday, the number of people who came to register began to show an endless trend after 10 am. Most of them were elderly people living nearby. After paying the fees, they had to wait for the called number in front of the two departments that opened the second.
In most of the waiting time, they are silent. You can only hear the sound of footsteps and the friction of the clothes. This quiet to a bit suffocating atmosphere can simply squeeze out the unique syrup of the hospital.
The legs and feet are still good and they are squatting in the hallway, and the pace is slightly slow, so they sit silent on the plastic bench. Their eyes will stare at a certain point on the wall for a long time, or they will continue to look at the information and read the information over and over again.
Everyone who comes to see a doctor seems to be very worried. Few people are willing to read carefully the so-called health tips hanging on the wall, not to mention the large chunks of orange on the corridor wall. “Advertising board.
This is one of the few grassroots centers that the National Standardized Metabolic Disease Management Project (MMC) has set up in Beijing to make online reservations.
As a national medical project initiated by Shanghai Ruijin Hospital (the top three hospitals), it has been labeled as big data, Internet of Things and artificial intelligence, and its online beauty is endless. Ali Health, AI Technology’s fourth paradigm, AI medical imaging company voxel technology have appeared in the introduction text of this project.
However, this management project, which had the privilege of being in the medical insurance system and trying to eliminate the medical gap between urban and rural areas, formed a huge contrast between the original intention and the landing.
indifferent group reaction
In the prominent part of the wall of Beijing Tongzhou Liyuan Community Hospital, orange MThe MC project introduction board is almost unattended. The editors are careful to erase the artificial intelligence or big data uncommon words that are not easily accepted by the public, and are condensed into five words – “standardization management.”
But even then, the above-mentioned row of “test items” and the winding process of the treatments make people read a little hard.
“What? Can? Diabetes management?” A grandfather muttered a few words: “I am a sore throat to take medicine.”
Every morning, I was trying to talk to the old people waiting to see the diabetes management project deliberately, and their reaction was almost exactly the same as the uncle.
An aunt mistakenly heard that I wanted to “see” diabetes. I suddenly came to the spirit: “Girl, hurry to the big hospital, come here to do something! Just take two sentences and take a medicine, what is there? Good doctor.”
The clinic with the MMC brand is at the end of the corridor, but it may be the weekend, the room is empty, but fortunately no one has found it.
The community doctor finally shouted my name. But when I pretended that the diabetics had any special features about the MMC project, it took him two or three seconds to react to what I said.
“Well… Diabetes project? In fact, it is to supervise and supervise you not to mess around with drugs. Experts will regularly send you a report of the condition and give some professional advice.”
This explanation of nothing has seemed to confirm the reasons for the distrust of the patients who came here.
Since we can’t give enough trust to small hospitals or community health centers, why do we trust the testing machines that are stationed in regional hospitals?
Even in the eyes of many old people, the definition of “advanced instrumentation” or “big data” is almost equivalent to an APP.
“Download the app? Download the app again! This is a white download.chant? Although an uncle did not dispel the enthusiasm of explaining the technique to me (I seem to be really a salesman), while listening to me “唠叨”, I held the MMC blood glucose management advertisement book in front of me for a long time. He finally only came up with such a sentence.
Serious differences in cognition make it easy to confuse and misunderstand new technologies, including AI, which in turn exacerbates patient distrust of technology.
Most people are basically like my father when they hear about diabetes AI management:
“Do you think this is more than a tumor? Your mother’s old girlfriend is not very successful in controlling sugar for 10 years. I am almost an expert. I also control my diet and I need to see a doctor.
“It doesn’t make any sense anyway. If you spend more than 1,000 yuan to manage, you might as well hire a nanny while doing housework and beside you.”
Don’t understand, don’t care, it’s easy, not just needed, afraid to spend money.
This is to let the national endocrinologists smash their chests, so that the country has to print the “diabetes outpatient medication guidance” reality.
the terrible situation that has to be concerned
In the corner of the waiting area of the Department of Metabolism in Ruijin Hospital, Zhao Zhiyun, director of the Clinical Medicine AI Laboratory of Ruijin Hospital, frowned, and said:
“More than 100 million diabetic patients in China, only 2.7% actually controlled effectively.”
In this over 100 million people, most of them do not know that they have got sick, especially in cities below the third and fourth line and in remote and impoverished areas. They are all severely ill and will be sent to the hospital.
An aunt who brought a domestic help to his family asked him to read a report of his rural relatives. “After reading the report, I said that the patients were all neuropathy and amputated. But they are still jealous. Don’t know!”
Only 30% of people with diabetes in China know that they are sick, which means 70% are not known; only 30% of these 30% go to the doctor or inquire about it. I have consciously done the treatment; even those who have done 30% of the treatment, the real control is only 30%. In the end, only 2.7% of Dr. Zhao’s mouth was controlled.
Many people with diabetes know that the terrible thing about this disease is not this.The disease itself, but the various complications caused by it. If you can’t control your blood sugar for a long time, it will cause lesions in your heart, brain, kidney, eyes, feet, etc., and even cause disability and even death.
To put it bluntly, we have always felt that controlling diabetes is not as “just needed” to treat cancer, but the role played by this disease is more like the last straw that crushed the camel.
A netizen once said that the blood sugar control is not obvious, and you will not become strong because of the blood sugar control. On the contrary, for many people, controlling blood sugar is a very tiring thing.
Without control, it’s a very cool thing.
When early control is possible, many people don’t know how to control, or even know if they don’t want to control it.
This is the status quo of Chinese diabetic patients.
The MMC Center on the second floor of Ruijin Hospital has a lot of people to consult
But another problem is that even if you see a doctor, you may not get the best sugar control and medication options.
According to China’s 114 million people with diabetes, even if it is calculated according to the needs of a doctor who manages 1,000 people, 100,000 doctors are needed.
But domestic endocrinologists are far from this magnitude. Especially in the fringe of towns and villages, the serious shortage of medical resources is a problem that has always existed for decades.
The Ministry of Health has issued regulations requiring village doctors to take an assistant doctor’s license. Although 60 points is the passing line, the village doctor can pass only 30 points. But even so, many hospitals in the country do not have doctors with assistant medical licenses.
For example, if patients with diabetes for the first time do not have a high level of blood glucose according to international standards, it is preferred to use drugs such as dioxin.
But Dr. Zhao is in a few yearsIn the previous sample survey, doctors in many areas rarely chose “first-line drugs” such as dioxin in the first place (referring to drugs that can be selected first or according to the patient’s condition), but instead injected insulin at the beginning.
“You will find that many township hospitals and barefoot doctors are not exposed to the latest research results and medication trends, and will only give patients a completely outdated medication regimen, which may not do anything at all.
The difference in the speed of access to the latest medical information will actually increase the gap between the medical level of the city and the township.
So, we’ve seen such a ubiquitous dilemma:
But patients with certain conditions are desperately trying to squeeze into a large hospital. Even if the team is in a dozen hours, it is not too late to be called “next” in the face of a doctor for less than 5 minutes. After they came out, they usually regretted that some questions were not asked.
The outside of the MMC Center of Ruijin Hospital is the waiting area. This time is already noon, and many people have been waiting for a few hours
If you stand at the perspective of a secretory doctor, they are suffering from a “torture” that is the opposite of the patient – most diabetics ask almost the same question.
In fact, sometimes we are thinking about it. If these problems are directly responded to by the machine, or if they are in the queue, someone will give them a basic science about diabetes control. Dr. Zhao felt that after a day of sitting, he basically lost his desire to speak, and he was more willing to sit alone in the corner and silence.
Therefore, time contradictions are actually true contradictions.
In order to resolve this contradiction, the state has been sparing no effort to promote “division”, and many top three hospitals are willing to work with technology companies to study how to use big data and artificial intelligence to “copy” the brains and medications of excellent doctors.
“This project actually plays its value even if it is based on a guaranteed number of patients.”
Is there a way to control diabetes?
Medical and artificial intelligence are destined to come together.
For example. When you have a cold, the doctor may tell you a lot of reasons, but these reasons are actually derived from clinical research, never directed at you.
The “clinical findings” here are often the result of finding a group of moderately sized patients to perform clinical trials by changing certain variables in a given environment.
So, if an algorithm model based on sufficient patient data is used as a measure of our condition, is it also true?
This is the most intuitive use of artificial intelligence in the judgment and prediction of diabetes. The data that determines this level of competence is of course data.
Dr. Zhao, who has dealt with or is dealing with many technology companies, knows that no one knows more about the deadly “hunger” of data support than algorithm engineers.
Many AI-related technology companies have been trying to incite the medical market in recent years. To make a structured map of the electronic medical records, and to do image recognition to do CT image-assisted screening for doctors… The first element to make these projects is the data.
At 10:00 pm, I saw Tu Weiwei in the Fourth Paradigm Office. The system they worked with the MMC project led by Ning Guang Academician was embedded in A variety of hardware.
The fourth paradigm of the chief scientist Tu Weiwei, who saw Dr. Zhao 14 times in more than two years, contributed tens of thousands of lines of artificial intelligence code to the MMC project.
“Don’t look at artificial intelligenceIt’s too magical, it just does what it should do, and it can’t do it. The typical code farmer who has been working overtime for 30 hours is still very cautious and restrained in answering each of my questions.
In the beginning, what they really wanted to do with Ruijin was not the AI prediction software. Instead, they wanted to use a collection of hundreds of thousands of chronic disease metabolic data to simulate a person’s metabolic system.
“But it’s actually too difficult, because the amount of data is not enough.” Tu Weiwei discovered the difficulty of simulating the whole body’s metabolism after two months of project cooperation,
“The metabolic system is definitely associated with other systems in the body. You can’t separate the functioning of the metabolic system alone. This is one.
Second, dietary habits vary widely from region to region. You may eat more sweets in Shanghai, you may eat a bit more salty in Beijing, and then go north to the west may eat a little more. We say three meals a day, but it may be normal to eat six meals a day in Guangzhou. These must be counted in.
It sounds like this is not a question of billions of data, because the cost and time cost of sorting and labeling data alone is hard to imagine.
So, they quickly adjusted their direction. In other words, how to use this hundreds of thousands of data to do a valuable thing, but also to use this data to cover different people in different regions.
This involves a method of machine learning – migration learning.
As long as you build a common chronic disease judgment model based on some key and similar data dimensions, then based on such a model, enter your own key body indicators to generate a relatively personalized assessment. result.
Doctors may feel ridiculous, but algorithmic engineers believe that artificial intelligence can do it.
For example, in the diabetes test, there is an indicator called “2 hours of blood sugar after breakfast.”
This is a very symbolic indicator that has a strong indication of predicting diabetes, high blood pressure or cardiovascular and cerebrovascular diseases, stroke, and myocardial infarction.
Rugin collected the 10,000-level data for this indicator, which means that the data for this indicator corresponds to tens of thousands of diabetics who have measured this indicator.
The body’s various indicators are all driven by the whole body. If you predict this indicator based on some other physiological indicators of these measured people, you can also construct a holistic prediction model andThe models “migrate” to those who have not tested.
After the completion of the prediction system, the Ruijin MMC Metabolic Disease Center conducted a comparative test.
The doctors used the data on the actual development of a part of diabetics that were continuously monitored for 3 years as the standard. The relevant data from 3 years ago were input into the model to see if they could be consistent with the actual data after 3 years.
The final result shows that the correct rate can reach more than 80%.
“In fact, we did a lot of tests by changing the number of measurements. A simple version of a system with only six dimensions (weight, age, blood pressure, value, etc.) can achieve an accuracy of 80%. With more metrics like blood lipids, blood sugar, etc., the accuracy will be higher.
We didn’t think the results were good at first, but the results surprised many doctors.
With a slight agreement, I copied a real patient report that I tried this diabetes prediction system.
In this 14-page report, I saw a detailed analysis of blood glucose, blood pressure and other indicators, as well as the risk level of related diseases such as cerebrovascular disease, and for the next 3 years, 6 years and 9 years of risk of diabetes risk.
Although I still have doubts about the accuracy of the results of the analysis, I can be sure that the doctor will never say as detailed as this report.
“The past diagnosis and diagnosis mode will be completely subverted with the introduction of this new technology and the improvement of data collection methods. It only takes time and money.” The medical technology company that died in the past few years has let Zhao The doctor is awkward,
“So, startups may be afraid that the AI bubble will burst, but the medical industry is not afraid. The combination of medical and AI is an inevitable trend. We will not do it, others will do it.”
landing “the deserted board”
Theories and experiments are always satisfactory.
But based on the indifference and indifference of the people we have seen above, there is a general lack of payment incentives for diabetes management, and few people will take the initiative to find and join the diabetes management program.
According to Dr. Zhao, MMC probably managed 200,000 patients, of which more than 50,000 were APP users, but there was no data feedback on user activity. The MMC grassroots center of Tongzhou Liyuan Community Hospital joined about 200 patients in one year.
These figures are negligible for the country’s 114 million diabetic patients.
In fact, a large number of diabetes management startups have appeared in the market five years ago. “Everyone was doing APP at the time, but now they are living in a few companies like Zhiyun Health.” Ding Junxing of Heli Investment believes that it is difficult for people with diabetes to take the initiative to manage chronic diseases, especially for online management.
There is not enough attention, or the market has not established sufficient knowledge of medical forecasting products, so it is difficult for such testers to pay for it, not to mention those who do not understand the technology, have time and Thrifty elderly patients.
Another reason why the diabetes management entrepreneurs of that year could not support it was that at that time, medical insurance could not pay for diabetes management, and the Chinese commercial insurance market system was relatively immature.
Although the management project led by a formal Chinese medical institution has finally entered the medical insurance system, since medical insurance can only be used locally in the insurance, community doctors do not recommend that patients in different places join the management program.
This has, to a certain extent, kept some of the off-site patients out of the door, because the area where the project is laid across the country is certainly limited. “If you don’t have local health insurance, I don’t recommend you to join MMC. The cost of nearly 2,000 yuan may be more expensive if you don’t take medical insurance.”
Of course, this has to involve another problem that will always exist – the geographical interest barrier of the medical industry.
“The top three hospitals in the first-tier cities, between different levels of hospitals in the same city, and even between different departments of hospitals, will involve barriers to interest. This also explains why the medical industry has always had poor data circulation. The data islands have a lot to do with the administrative and interest barriers in the hospital.”
An industry source who asked not to be named said that management projects like Ruijin Hospital may be difficult to enter Beijing or Guangzhou’s top three hospitals because they may be doing similar projects.
In other words, in addition to being scalable locally, it is difficult to achieve dense distribution in other regions.
In addition, even the top-ranking projects of the top three hospitals, the marketing and sinking methods adopted by them are still the “old roads” that medical device companies, drug dealers and diabetes management companies have experienced.
For example, in order for a diabetic to take the initiative to enter the MMC, it is largely up to the doctors in the general department to “recommend the recommendation.” If you want to go to the 3rd, 4th and 5th tier cities for promotion, most of them are still doing a ‘free clinic’ in the top three hospitals, doing free inspections for the people, or asking professors to go to the countryside to give lectures.
“Traditional industries, especially the highly regulated ‘iron wall’ industry, such as medical treatment, the commercialization of new medical projects and new technologies is difficult to ‘sexy’. Just like medical representatives will always exist.” Entrepreneurs who are biomedical think that it is ridiculous to talk about business models or marketing innovations in the medical industry.
If a famous doctor endorses your product, the credibility of the project will be greater and the number of patients will be more. Many projects, including MMC, have expanded the market in the form of franchises. The “famous doctors” are the biggest driving force to attract community or private hospitals.
But even the reputation of a large hospital like Shanghai Ruijin has been proven to be far from expected.
Technology + medical all the way to the pits
Advanced technology such as AI has been controversial.
Some technology startups have tried to achieve commercialization through the “Centre of Medical Examinations”. The original intention is to package the obscure biotechnology or AI technology into a medical check-up package, which may make some of them economically capable and strong. Conscious person becomesThe earliest users of this advanced medical product.
At present, the penetration rate of Chinese medical examiners is as high as 40% (about 500 million people), which is also a prerequisite for its logical establishment.
However, before this is almost the only commercial entry that can be walked through, fierce competition has made the value of the technology itself start to go away.
Two months ago, Xinhua News Agency had a series of exposures to the newly introduced screening program at the Medical Center:
In order to obtain higher profits, some informal medical examination centers have introduced new technologies such as cancer screening and AI, deliberately exaggerating the effect and raising the price of medical examinations. However, in fact, many products have no effect of cancer screening at all. .
Advanced technology aimed at providing accurate medical care has gradually been distorted as a catalyst and accomplice to deliberately “selling fear”.
The fact that these have happened is to dig pits ahead of the road to artificial intelligence in diabetes. AI+Medical, which shoulders the mission of change, has been under the close supervision of the Food and Drug Administration.
For example, although artificial intelligence-assisted decision-making products can enter the cooperative testing of hospitals, they may not be able to register through the medical device approval of the Food and Drug Administration. If it is not approved, any commercial use is illegal.
At the official level of the National Drug Administration’s Center for Medical Device Evaluation (CMDE), I found something interesting: more and more artificial intelligence-related words appeared in various approval documents.
This aspect proves that artificial intelligence medical decision-making products are getting more and more attention and recognition from the medical industry, and surveillance is becoming more and more strict.
“The patient’s data of course involves privacy issues. In fact, this is not only our focus, but the country is also very strict with this control.” Even if the model is developed on a private cloud basis, Tu Weiwei is giving the hospital When doing a project, it is also required to set up more firewalls for the data acquisition channel from the technical dimension.
“Isn’t the so-called data desensitization not dangerous? No, there is a very common attack method called differential attack. It means that although your identity information is hidden, I know your gender. If you know some of your other indicators, and if there is the same data in the external pile of data, then I may be able to ‘crash’ this person.”
Therefore, they need to adopt a similar technical solution used with Apple’s mobile phone – add “noise” to these medical data. But this approach also raises a problem: the overall data quality will definitely get worse.
People’s cognitive limitations on technology, industry geography and barriers to interest, business models are severely constrained by industry specificity, being used as a profit tool to exaggerate the potential risks of functionality, and the enormous pressure exerted by regulators… Br>
Like the gradual weakening of typhoon registration from south to north, we can’t ignore the process of AI technology gradually moving from product to commodity to transaction.
Conclusion: Wait for our generation to grow older
In the interview, a friend who was very upset about his father’s diabetes control, after listening to what I said about sugar management and AI testing, took notes carefully because “her father came The season of eating dumplings and moon cakes will be sent to the hospital.”
This feedback gives me some new feelings:
Many young people, while acting as parents’ “sugar control”, are more concerned about diabetes than our previous generation.
“Why do many technology companies die, but the general trend of applying biotechnology and artificial intelligence technology will never change. Because waiting for the market to really shape, we need this generation of well-educated people to support it.” A technology industry investor believes that in the medical industry, the outbreak of new technologies has not yet arrived.
She shared a real story around a friend who studied nutrition:
The girl who lived in the rural area of Jiangxi suddenly fainted one day. She was diagnosed with severe diabetes and urinary infection. The county doctor said that they could not do anything. The general meaning is that “it’s almost enough, you will tell her something nice. I’m almost waiting to die.”
At that time, her aunt lived in the ICU. After doing “anti-infection” for a few days, she was taken home without any improvement. She went to consult a friend who was a medical doctor and got advice to fight infection and lower blood sugar.
She feels that her family has passed the ICU, and what is necessary to save it. However, because she felt that she had a culture, she listened to her words and drove it to the county-level city hospital for more than an hour. While doing blood sugar lowering, she did anti-infection.
After half a month, her aunt was discharged.
“In fact, this is equivalent to saving a life, the grassroots really have difficulties you can’t imagine.” He was not surprised at some of my knowledge. Because the facts are more cruel than what I have seen, but there is more room for change.
“Maybe when we get older, the situation will improve.”