编程知识 cdmana.com

Stanford's use of algorithm to distribute new coronal vaccine is under fire: medical staff only get 0.14 percent

Thepaper.cn reporter Wang Xinxin

Earlier this month. , Pfizer's new crown vaccine was awarded by the U.S. Food and drug administration (FDA) After emergency approval of , The federal government began distributing vaccines to health systems across the country . According to trump's plan ,2020 By the end of the year ,2000 Ten thousand people in the United States will get the first vaccine .

 Stanford's use of algorithms to distribute new coronal vaccines is under fire : The medical staff were only assigned 0.14%

Local time 2020 year 12 month 16 Japan , Arlington, Virginia, USA , Medical staff at the Virginia Hospital Center received Pfizer's new crown vaccine . People's vision Data figure

Over population in this area 3 Billion countries , The distribution of vaccines must give priority to vulnerable people 、 Transportation and other factors . Because of the large number 、 Time is running out , The task of how to distribute vaccines falls on artificial intelligence algorithms . But using algorithms to make decisions , There is another problem : The algorithm lacks transparency , The decision-making process cannot be specifically known , In view of previous cases, it has been shown that the algorithm can be applied to some population groups , In particular, ethnic minorities and low-income people are biased , Using algorithms to solve the problem of vaccine distribution deserves more attention .

actually , This kind of prejudice has happened , And it happened at Stanford, one of the top universities in the United States . recently , Stanford University got the first 5000 Vaccine , But in the list of vaccinated people , The number of medical staff is only 0.14%. say concretely ,5000 There is nothing but 7 Famous doctors and nurses got the vaccine . But at Stanford University , Yes 1300 More than one paramedic 、 researcher ,7 People on the vaccination list , It means only about 0.5% Of the medical staff assigned to the vaccine . contrary , Some work from home 、 Instead of facing new crown patients every day, senior teachers are selected in large numbers .

Front line medical staff are heroes in the fight against the epidemic , But they're not being cared for in vaccinations . The list caused discontent , The medical staff also protested directly . And it all started with a vaccine distribution algorithm .

After seeing the protest from the medical staff , Stanford medical spokesperson Lisa Kim Express , We are fully responsible for the mistakes in the implementation of the vaccine distribution plan . Our original intention was to create a morality 、 Fair vaccine distribution procedures .

Tiberius system

Stanford's vaccine distribution strategy is just a microcosm . At the federal level , In order to achieve 2021 The first quarter of the year is 1 The goal of 100 million Americans being vaccinated , U.S. Department of health and human services and data mining companies Palantir We've worked together on a project called Tiberius (Tiberius) Software platform .

Tiberius's data cover the United States 50 Two states ,8 A region , State and federal agencies are allowed to view VTrckS Website and American disease control and Prevention (CDC) A vaccine tracking system for . They can also use Tiberius to decide where to send the vaccine in their jurisdiction .

The current development of new coronal vaccine in the United States follows the following process : Thursday every week , Vaccine manufacturers will inform the U.S. public private partnership for the development of new crown therapies Warp Speed, How much vaccine is available in the next week . Friday , Tiberius system , The algorithm accounts for variables including the adult population and vaccine intake , And allocate the amount of vaccine available in each state . next , Every Saturday , Each state has to complete its order , And complete the delivery on Monday .

According to previously published documents , The system will allow health authorities to “ Integrating a broad population 、 Employment and public health data sets , To identify priority groups ”, and “ Support allocation decisions ”. Health officials can use data systems to create analytical maps to plan the distribution and distribution of vaccine doses , And follow the hospital 、 Delivery of vaccines at clinics and other injection sites .

Warp Speed The person in charge of the action plan Deacon Maddox The Colonel will Tiberius The algorithm is described as “ Simple ”,“ fair ”. But the crux of the problem is that the system relies heavily on the U.S. Census Bureau for input data 2018 Population estimates from the American community survey in . Due to time constraints 、 Lack of manpower and other confounding factors , The Census Bureau often underestimates the population of some parts of the United States . for example , There are studies that show that , Black men ( In especial 30 to 49 A year old man ) The underestimation rate is much higher than the net underestimation rate of the total black population or the total male population .

“ This census data has problems of under reporting and over reporting . If you don't include the vulnerable in a large way , So you're underestimating the number of infected people .” From the Department of design and engineering, human center, University of Washington AI researcher Os Keyes In an interview with foreign media, he said , And he thinks it may increase social inequality .

The impact of relying on such data input is enormous . According to John · Hopki University (Johns Hopkins University) The data of , Black Americans are almost three times as fast as white Americans , And the risk of dying from the virus is more than twice that of white Americans . Part of the reason is that black people are more likely to get sick , No health insurance , And it's more likely to be a job that's not suitable for home office .

Inequality continues , How to solve ?

In addition to the federal government's allocation algorithm , In some states of the United States , Including Ohio , Wisconsin , South Carolina , New Hampshire , Delaware , Tennessee and Arizona included , Both are using algorithms developed in-house to determine how vaccines should be distributed locally . meanwhile , In, for example, George Washington University Hospital (GWUH) Such local institutions , Also running this algorithm system , According to age , Basic medical conditions , Prevention of transmission and the ability of hospitals to continue their operations to select the population .

Stanford University's vaccine distribution has highlighted the limitations of the algorithm , From the federal government to the state government to the various agencies , In the algorithm layer by layer calculation , Many experts in the United States believe that this practice will inevitably lead to more social inequality , But it may not be so easy to improve the algorithm .

Associate professor at Georgia tech Mark O. Riedl Express , If Stanford has a well funded Center 、 Specialized research AI The morality of algorithms , fair 、 Just use it fairly .

After the protest , President and CEO, Stanford Health Bureau David Entwhistle Indicates that the algorithm will be corrected , New vaccine distribution plans . And for this algorithm “ error ”, He explained. :“ Vaccine distribution algorithms follow federal guidelines , That is to give priority to healthcare workers and older employees .”

But the outside world doesn't seem to buy it . An email sent by a Stanford resident to his colleagues showed that , They're at a disadvantage in the whole system , Because there is no specific “ Location ” It can be inserted into the system for calculation , And they're very young .

According to the technology website Venturebeat reports , As the number of vaccines increases next year , Federal and state agencies and the health care system are likely to improve the algorithm to deal with the allocation problem . For all that , The less vaccinated phase still needs to worry . Officials in a dozen states complained that , They received less than they promised . Pfizer claims that , The federal government has yet to release shipping instructions for the millions of doses of vaccine currently in storage . If we don't solve the problem of long transportation time , Lack of transparency , Plus the imbalance of data and methods , It has the potential to exacerbate health-care inequality in the entire American community .

editor : Li Yuequn

proofreading : Shi Zhen

本文为[Never give up]所创,转载请带上原文链接,感谢

Scroll to Top