醫生八成工作將由科技代勞

????今天的醫療保健往往靠的還是“醫術”，而不是“醫學”。

????拿治療發燒為例。150年來，醫生一般都是開布洛芬等退燒藥幫助退燒。但在2005年，邁阿密大學（the University of Miami）的研究人員對82名病?；颊哌M行了一項研究。82名病人被隨機分成兩組，其中一組的病人在體溫超過攝氏38.5度時便服用退燒藥（即“標準療法”），另一組只有在體溫達到攝氏40度時才服用退燒藥。結果隨著試驗的開展，有7名接受標準療法的病人死亡了，而沒有服用退燒藥的那一組中，只有一名病人死亡。但就在這時，試驗由于倫理問題被終止了，因為研究團隊認為再讓更多病人接受標準療法是不道德的。

????像退燒這樣基本的病癥的治療都難以避免地帶有“醫術”的痕跡，而且這種情況持續了100多年都沒有改觀，我們不禁要問：還有哪些療法是依托于傳統，而不是依托于科學？

????今天的診斷方式部分根據患者的病史，部分根據患者的癥狀（不過患者一般都不擅長描述癥狀）。甚至大多數時候，醫生是根據醫藥廣告，以及多年前在醫學院學到的知識做出診斷。且不說醫學院講授的內容好多已經過時，而且時間一久，醫生自己也忘了許多課本上的知識，何況光是從醫學院學來的知識也難免存在認識偏差、經驗編差以及其它人為錯誤。許多時候，三個醫生診斷同一個病癥，可能會得出三種不同的診斷和三種不同的療法。

????結果是患者不僅診療效果不理想，而且還花了大筆冤枉錢。約翰霍普金斯大學（Johns Hopkins?University）的一項研究發現，美國每年都有40,500余名患者因為誤診而死亡，幾乎與死于乳腺癌的人數相當。另一份研究則發現，有65%的誤診病例都與處理不當、團隊合作不協調、溝通不暢等所謂的“系統相關因素”有關。而75%的誤診病例都存在所謂的“認識因素”，而其中最主要的原因，是由于醫生堅持一開始的錯誤診斷，忽視了其它合理的可能。這種誤診也增加了醫療支出，每次誤診索賠的平均金額為30萬美元。

????醫療診斷應該更多運用數據演繹的方法，減少“摸著石頭過河”的成分。而這個目標離開現代科技則很難實現，因為如今可使用的數據和研究方法越來越多。新一代的醫療技術將用到更多、更復雜的生理模型，并且使用更多的傳感器數據，從而提供個性化的診療，這些數據可能不是光靠一個人類醫生就能理解得了的。每次診斷都將依托于成千上萬個基線數據和多個組學數據點，以及更加全面的病史和患者行為等要素。

????日益完善的對話管理系統將有助于對病人進行更為有效和更為全面的數據捕捉及探測。其中的關鍵就是數據科學。最終，它將有助于降低醫療成本，減少醫生的工作量，提高患者的醫療水平。

????Healthcare today is often really the "practice of medicine" rather than the "science of medicine."

????Take fever as an example. For 150 years, doctors have routinely prescribed antipyretics like ibuprofen to help reduce fever. But in 2005, researchers at the University of Miami, Florida, ran a?studyof 82 intensive care patients. The patients were randomly assigned to receive antipyretics either if their temperature rose beyond 101.3°F ("standard treatment") or only if their temperature reached 104°F. As the trial progressed, seven people getting the standard treatment died, while there was only one death in the group of patients allowed to have a higher fever. At this point, the trial was stopped because the team felt it would be unethical to allow any more patients to get the?standard treatment.

????So when something as basic as fever reduction is a hallmark of the "practice of medicine" and hasn't been challenged for 100+ years, we have to ask: What else might be practiced due to tradition rather than science?

????Today's diagnoses are partially informed by patients' medical histories and partially by symptoms (but patients are bad at communicating what's really going on). They are mostly informed by advertising and the doctor's half-remembered and potentially obsolete lessons from medical school (which are laden with cognitive biases, recency biases, and other human errors). Many times, if you ask three doctors to look at the same problem, you'll get three different diagnoses and three different treatment plans.

????The net effect is patient outcomes that are inferior to and more expensive than what they should be. A Johns Hopkins?study?found that as many as 40,500 patients die in an ICU in the U.S. each year due to misdiagnosis, rivaling the number of deaths from breast cancer. Yet another?studyfound that 'system-related factors', e.g. poor processes, teamwork, and communication, were involved in 65% of studied diagnostic error cases. 'Cognitive factors' were involved in 75%, with 'premature closure' (sticking with the initial diagnosis and ignoring reasonable alternatives) as the most common cause. These types of diagnostic errors also add to rising healthcare expenditures, costing?$300,000?per malpractice claim.

????Healthcare should become more about data-driven deduction and less about trial-and-error. That's hard to pull off without technology, because of the increasing amount of data and research available. Next-generation medicine will utilize more complex models of physiology, and more sensor data than a human MD could comprehend, to suggest personalized diagnosis. Thousands of baseline and multi-omic data points, more integrative history, and demeanor will inform each diagnosis.

????Ever-improving dialog manager systems will help make data capture and exploration from patients more accurate and comprehensive. Data science will be key to this. In the end, it will reduce costs, reduce physician workloads, and improve patient care.