TEDMED 2015
Russ Altman: What really happens when you mix medications?
洛斯．阿特曼: 當你混合用藥時會發生什麼事？
Filmed:
Readability: 4.2
1,766,922 views
如果你因為不同的病因得同時服用兩種不同的藥，這裡有一個值得深思的忠告：你的醫生也許完全不知道，當他們同時開這兩種藥給你時，會產生什麼後果，因為藥物之間的相互作用真的很難研究。在這場精彩又容易理解的演講中，洛斯．阿特曼將展示醫生如何透過驚人的搜尋引擎資料，來研究意想不到的藥物作用。
Russ Altman - Big data techno-optimist and internist
Russ Altman uses machine learning to better understand adverse effects of medication. Full bio
Russ Altman uses machine learning to better understand adverse effects of medication. Full bio
Double-click the English transcript below to play the video.
00:12
So you go to the doctor
and get some tests.
and get some tests.
0
811
3321
你去看醫生，接受了一些檢查。
醫生診斷出你的膽固醇過高，
00:16
The doctor determines
that you have high cholesterol
that you have high cholesterol
1
4674
2620
00:19
and you would benefit
from medication to treat it.
from medication to treat it.
2
7318
3171
建議你服藥治療可能有幫助。
所以，你拿到了藥罐子。
00:22
So you get a pillbox.
3
10981
1556
00:25
You have some confidence,
4
13505
1199
你有點信心，
00:26
your physician has some confidence
that this is going to work.
that this is going to work.
5
14728
2937
你的醫師也有信心，認為這藥會有效。
00:29
The company that invented it did
a lot of studies, submitted it to the FDA.
a lot of studies, submitted it to the FDA.
6
17689
3553
發明這個藥的公司做了很多的研究，
然後呈送給食品藥物管理局。
然後呈送給食品藥物管理局。
00:33
They studied it very carefully,
skeptically, they approved it.
skeptically, they approved it.
7
21266
3107
他們很仔細、審慎地研究，
並核准了這藥物上市。
並核准了這藥物上市。
00:36
They have a rough idea of how it works,
8
24397
1889
他們大概知道這藥物如何運作，
00:38
they have a rough idea
of what the side effects are.
of what the side effects are.
9
26310
2453
也大略知道會有什麼副作用，
00:40
It should be OK.
10
28787
1150
應該沒問題。
00:42
You have a little more
of a conversation with your physician
of a conversation with your physician
11
30864
2818
你跟醫師又多聊了一會，
00:45
and the physician is a little worried
because you've been blue,
because you've been blue,
12
33706
2963
而醫師有點擔心，因為你很憂鬱，
精神欠佳。
00:48
haven't felt like yourself,
13
36693
1293
無法像平常一樣盡情享受生活點滴。
00:50
you haven't been able to enjoy things
in life quite as much as you usually do.
in life quite as much as you usually do.
14
38010
3731
00:53
Your physician says, "You know,
I think you have some depression.
I think you have some depression.
15
41765
3186
你的醫師說：
「我認為你有一點精神憂鬱，
「我認為你有一點精神憂鬱，
00:57
I'm going to have to give
you another pill."
you another pill."
16
45792
2315
我再開個藥給你。」
01:00
So now we're talking
about two medications.
about two medications.
17
48934
2483
所以，我們現在有兩種藥了。
01:03
This pill also -- millions
of people have taken it,
of people have taken it,
18
51441
3104
這個藥也有好幾百萬人服用過，
01:06
the company did studies,
the FDA looked at it -- all good.
the FDA looked at it -- all good.
19
54569
3631
公司做了研究，食品藥物管理局
也檢查過，全部都沒問題。
也檢查過，全部都沒問題。
01:10
Think things should go OK.
20
58823
2057
想一下，這東西沒問題，OK的。
01:12
Think things should go OK.
21
60904
2197
想一下，這東西沒問題，OK的。
01:15
Well, wait a minute.
22
63125
1439
但，請等一下。
01:16
How much have we studied
these two together?
these two together?
23
64588
3517
我們對這兩種藥混在一起吃
做了多少研究？
做了多少研究？
01:20
Well, it's very hard to do that.
24
68630
2300
其實，這很難評估。
01:22
In fact, it's not traditionally done.
25
70954
2130
事實上，傳統上都不會做。
01:25
We totally depend on what we call
"post-marketing surveillance,"
"post-marketing surveillance,"
26
73108
5518
在藥物上市後，我們完全倚賴一種
叫做「上市後監察系統」的機制，
01:30
after the drugs hit the market.
27
78650
1880
01:32
How can we figure out
if bad things are happening
if bad things are happening
28
80996
2848
我們要如何確認，兩種藥之間
01:35
between two medications?
29
83868
1357
是否有什麼不好的事會發生？
三種？五種？七種呢？
01:37
Three? Five? Seven?
30
85249
2030
01:39
Ask your favorite person
who has several diagnoses
who has several diagnoses
31
87708
2415
問你身邊有各種疾病在身的人，
01:42
how many medications they're on.
32
90147
1834
他們正在吃多少藥。
01:44
Why do I care about this problem?
33
92530
1580
為什麼我在乎這個問題？
01:46
I care about it deeply.
34
94134
1157
我非常在乎。
01:47
I'm an informatics and data science guy
and really, in my opinion,
and really, in my opinion,
35
95315
4304
我是念資訊和數據科學的人，
真的，在我看來，
真的，在我看來，
01:51
the only hope -- only hope --
to understand these interactions
to understand these interactions
36
99643
3745
了解藥物彼此間的交互影響
唯一的希望只有
唯一的希望只有
01:55
is to leverage lots
of different sources of data
of different sources of data
37
103412
3056
運用不同來源的龐大資料，
01:58
in order to figure out
when drugs can be used together safely
when drugs can be used together safely
38
106492
3556
才能找出這些藥
何時可以安全地一起服用，
何時可以安全地一起服用，
02:02
and when it's not so safe.
39
110072
1777
以及何時不行。
所以，讓我來告訴各位
一個數據科學的故事。
一個數據科學的故事。
02:04
So let me tell you a data science story.
40
112615
2051
02:06
And it begins with my student Nick.
41
114690
2154
這要從我的學生尼克開始講起。
02:08
Let's call him "Nick,"
because that's his name.
because that's his name.
42
116868
2380
我們就稱呼他為尼克吧，
因為那就是他的本名。
因為那就是他的本名。
02:11
(Laughter)
43
119272
1592
（笑聲）
02:12
Nick was a young student.
44
120888
1201
尼克很年輕，
02:14
I said, "You know, Nick, we have
to understand how drugs work
to understand how drugs work
45
122113
3079
我說：「尼克，
我們必須了解藥物如何運作，
我們必須了解藥物如何運作，
02:17
and how they work together
and how they work separately,
and how they work separately,
46
125216
2626
以及藥物在一起會如何運作、
分開會如何運作，
分開會如何運作，
02:19
and we don't have a great understanding.
47
127866
1922
而我們並沒有了解很深。」
02:21
But the FDA has made available
an amazing database.
an amazing database.
48
129812
2405
但食品藥物管理局已經
有一個很驚人的資料庫，
有一個很驚人的資料庫，
02:24
It's a database of adverse events.
49
132241
1699
是一個藥物不良反應通報資料庫。
02:26
They literally put on the web --
50
134321
1642
資料真的直接放在網路上
02:27
publicly available, you could all
download it right now --
download it right now --
51
135987
3119
供大眾查詢，你現在就可以全部下載，
02:31
hundreds of thousands
of adverse event reports
of adverse event reports
52
139130
3627
從病人、醫生、公司、藥劑師通報上來
02:34
from patients, doctors,
companies, pharmacists.
companies, pharmacists.
53
142781
3760
好幾百萬個的藥物不良反應通報。
02:38
And these reports are pretty simple:
54
146565
1749
這些報告都相當簡單：
02:40
it has all the diseases
that the patient has,
that the patient has,
55
148338
2658
上面有病人所有疾病
02:43
all the drugs that they're on,
56
151020
1767
及所有藥物的使用狀況，
02:44
and all the adverse events,
or side effects, that they experience.
or side effects, that they experience.
57
152811
3818
還有他們經歷過的
所有不良反應事件或副作用。
所有不良反應事件或副作用。
02:48
It is not all of the adverse events
that are occurring in America today,
that are occurring in America today,
58
156653
3436
雖然沒有現今在美國
發生的所有不良反應事件，
發生的所有不良反應事件，
02:52
but it's hundreds and hundreds
of thousands of drugs.
of thousands of drugs.
59
160113
2578
但卻有上百萬種藥物資科。
02:54
So I said to Nick,
60
162715
1299
所以，我跟尼克說：
02:56
"Let's think about glucose.
61
164038
1826
「我們來想一想葡萄糖。
02:57
Glucose is very important,
and we know it's involved with diabetes.
and we know it's involved with diabetes.
62
165888
3567
葡萄糖非常重要，而且
大家都知道它與糖尿病有關。
大家都知道它與糖尿病有關。
03:01
Let's see if we can understand
glucose response.
glucose response.
63
169479
3970
讓我們來看看是否可以
了解葡萄糖的反應。」
了解葡萄糖的反應。」
03:05
I sent Nick off. Nick came back.
64
173473
2458
我請尼克去找資料，
03:08
"Russ," he said,
65
176248
1786
他回來後說：「洛斯，
03:10
"I've created a classifier that can
look at the side effects of a drug
look at the side effects of a drug
66
178351
5112
我已經建造了一個分辨器，
可以透過這個資料庫
可以透過這個資料庫
03:15
based on looking at this database,
67
183487
2051
來檢視一種藥物的副作用，
03:17
and can tell you whether that drug
is likely to change glucose or not."
is likely to change glucose or not."
68
185562
4271
而且還可以告訴你，這個藥
會否改變病人血糖狀況。」
會否改變病人血糖狀況。」
03:21
He did it. It was very simple, in a way.
69
189857
2016
他用一個方法做到了，很簡單。
03:23
He took all the drugs
that were known to change glucose
that were known to change glucose
70
191897
2635
他把所有已知會改變葡萄糖的藥物
及所有不會改變的藥物拿出來做比較，
03:26
and a bunch of drugs
that don't change glucose,
that don't change glucose,
71
194556
2389
03:28
and said, "What's the difference
in their side effects?
in their side effects?
72
196969
2888
「它們之間的副作用有什麼分別？
03:31
Differences in fatigue? In appetite?
In urination habits?"
In urination habits?"
73
199881
4852
疲勞狀況上的差異？食慾上的差異？
排尿習慣上的差異？」
排尿習慣上的差異？」
03:36
All those things conspired
to give him a really good predictor.
to give him a really good predictor.
74
204757
2960
所有這些事情都可以協助他
做出一個很棒的預測器。
做出一個很棒的預測器。
03:39
He said, "Russ, I can predict
with 93 percent accuracy
with 93 percent accuracy
75
207741
2548
他說：「洛斯，我能預測
哪種藥可改變血糖，
哪種藥可改變血糖，
03:42
when a drug will change glucose."
76
210313
1572
準確率可以高達93%。」
03:43
I said, "Nick, that's great."
77
211909
1416
我說：「尼克，這太棒了！」
他是個年輕的學生，
他是個年輕的學生，
03:45
He's a young student,
you have to build his confidence.
you have to build his confidence.
78
213349
2896
你必須建立他的信心。
03:48
"But Nick, there's a problem.
79
216269
1390
「但，尼克，有一個問題。
03:49
It's that every physician in the world
knows all the drugs that change glucose,
knows all the drugs that change glucose,
80
217683
3960
就是全世界的醫師都知道
這些藥會改變葡萄糖，
這些藥會改變葡萄糖，
03:53
because it's core to our practice.
81
221667
2038
因為這是我們實務上的核心。
03:55
So it's great, good job,
but not really that interesting,
but not really that interesting,
82
223729
3722
所以，你很棒，幹得好，
但並沒有人對這有興趣，
但並沒有人對這有興趣，
絕對還不適合公布你的研究結果。」
03:59
definitely not publishable."
83
227475
1531
04:01
(Laughter)
84
229030
1014
（笑聲 ）
04:02
He said, "I know, Russ.
I thought you might say that."
I thought you might say that."
85
230068
2550
他說：「我知道，洛斯。
我知道你可能會這麼說。」
我知道你可能會這麼說。」
尼克很聰明。
04:04
Nick is smart.
86
232642
1152
04:06
"I thought you might say that,
so I did one other experiment.
so I did one other experiment.
87
234149
2874
「我知道你會這麼說，
所以我多做了另一項實驗。
所以我多做了另一項實驗。
04:09
I looked at people in this database
who were on two drugs,
who were on two drugs,
88
237047
2928
我仔細觀察資料庫裡
同時服用兩種藥的人，
同時服用兩種藥的人，
04:11
and I looked for signals similar,
glucose-changing signals,
glucose-changing signals,
89
239999
4422
然後尋找他們之間
葡萄糖改變的相似訊號，
04:16
for people taking two drugs,
90
244445
1624
04:18
where each drug alone
did not change glucose,
did not change glucose,
91
246093
5569
但前提是，這些藥單獨服用
不會改變葡萄糖，
不會改變葡萄糖，
04:23
but together I saw a strong signal."
92
251686
2460
一起服用時，會有強烈訊號的藥物。」
04:26
And I said, "Oh! You're clever.
Good idea. Show me the list."
Good idea. Show me the list."
93
254170
3149
我說：「喔！你真聰明，
好主意，讓我看一下清單。」
好主意，讓我看一下清單。」
04:29
And there's a bunch of drugs,
not very exciting.
not very exciting.
94
257343
2254
有一大堆藥，並沒有令人非常興奮。
04:31
But what caught my eye
was, on the list there were two drugs:
was, on the list there were two drugs:
95
259621
3932
但引起我注意的是，清單上有兩種藥：
04:35
paroxetine, or Paxil, an antidepressant;
96
263577
3393
帕羅西汀或稱克憂果，
這是一種治療憂鬱症的藥，
這是一種治療憂鬱症的藥，
04:39
and pravastatin, or Pravachol,
a cholesterol medication.
a cholesterol medication.
97
267756
3570
還有普伐他汀或稱美百樂，
一種治療心臟疾病的藥。
一種治療心臟疾病的藥。
04:43
And I said, "Huh. There are millions
of Americans on those two drugs."
of Americans on those two drugs."
98
271936
4283
然後我說：「哈！有上百萬
美國人正在服用這兩種藥」。
美國人正在服用這兩種藥」。
04:48
In fact, we learned later,
99
276243
1246
事實上，我們之後才知道，
04:49
15 million Americans on paroxetine
at the time, 15 million on pravastatin,
at the time, 15 million on pravastatin,
100
277513
6032
當時有1500萬美國人正在服用帕羅西汀，
1500萬人正在服用普伐他汀，
1500萬人正在服用普伐他汀，
04:55
and a million, we estimated, on both.
101
283569
2817
而我們預估有100萬人，
同時服用這兩個藥。
同時服用這兩個藥。
所以，有100萬人
04:58
So that's a million people
102
286767
1254
05:00
who might be having some problems
with their glucose
with their glucose
103
288045
2453
可能有葡萄糖上的問題，
05:02
if this machine-learning mumbo jumbo
that he did in the FDA database
that he did in the FDA database
104
290522
3206
如果他用食品藥物管理局的資料庫
做的機械學習判讀器真的有用的話。
05:05
actually holds up.
105
293752
1254
05:07
But I said, "It's still not publishable,
106
295030
1927
但我說：「還是不能發表，
05:08
because I love what you did
with the mumbo jumbo,
with the mumbo jumbo,
107
296981
2296
因為我雖然喜歡你做的
05:11
with the machine learning,
108
299301
1246
機械學習判讀器，
05:12
but it's not really standard-of-proof
evidence that we have."
evidence that we have."
109
300571
3864
但我們沒有真正的證明標準
來證明我們是正確的。」
來證明我們是正確的。」
所以，我們來必須做些其他事來驗證。
05:17
So we have to do something else.
110
305618
1589
05:19
Let's go into the Stanford
electronic medical record.
electronic medical record.
111
307231
2876
我們去找史丹佛的電子病例紀錄。
05:22
We have a copy of it
that's OK for research,
that's OK for research,
112
310131
2064
我們有一個副本，可以用來研究，
05:24
we removed identifying information.
113
312219
2046
我們移除了病人個資。
05:26
And I said, "Let's see if people
on these two drugs
on these two drugs
114
314581
2503
我說：「讓我們來看看，
服用這兩種藥的人
服用這兩種藥的人
05:29
have problems with their glucose."
115
317108
1769
是否有葡萄糖上的疾病。」
05:31
Now there are thousands
and thousands of people
and thousands of people
116
319242
2207
在史丹佛病例紀錄中有成千上萬的人
05:33
in the Stanford medical records
that take paroxetine and pravastatin.
that take paroxetine and pravastatin.
117
321473
3459
同時服用這兩種藥。
05:36
But we needed special patients.
118
324956
1799
但我們需要特定病患。
05:38
We needed patients who were on one of them
and had a glucose measurement,
and had a glucose measurement,
119
326779
4597
我們需要已經做葡萄糖檢測
且服用其中一種藥的病人，
且服用其中一種藥的病人，
05:43
then got the second one and had
another glucose measurement,
another glucose measurement,
120
331400
3449
另外再找到另一個已經做過
另一個葡萄糖檢測的病人，
另一個葡萄糖檢測的病人，
05:46
all within a reasonable period of time --
something like two months.
something like two months.
121
334873
3615
全部都在合理期間做的，
例如兩個月內。
例如兩個月內。
05:50
And when we did that,
we found 10 patients.
we found 10 patients.
122
338512
3159
當我們開始著手進行時，
我們找到十個病人。
我們找到十個病人。
05:54
However, eight out of the 10
had a bump in their glucose
had a bump in their glucose
123
342592
4538
然而，十個人裡面
有八個葡萄糖異常增加現象，
有八個葡萄糖異常增加現象，
05:59
when they got the second P --
we call this P and P --
we call this P and P --
124
347154
2645
在他們服用第二個P時
─我們稱呼這個叫 P&P─
─我們稱呼這個叫 P&P─
06:01
when they got the second P.
125
349823
1310
當他們服用了第二個 P。
06:03
Either one could be first,
the second one comes up,
the second one comes up,
126
351157
2562
哪一個先服用都行，
當第二個藥服用後，
當第二個藥服用後，
06:05
glucose went up
20 milligrams per deciliter.
20 milligrams per deciliter.
127
353743
2847
葡萄糖濃度每公升會增加20毫克。
06:08
Just as a reminder,
128
356614
1158
提醒各位一下，
06:09
you walk around normally,
if you're not diabetic,
if you're not diabetic,
129
357796
2325
如果你能正常走動，沒有糖尿病，
06:12
with a glucose of around 90.
130
360145
1359
你的葡萄糖濃度約90毫克／公升。
06:13
And if it gets up to 120, 125,
131
361528
2076
如果上升到120、125，
06:15
your doctor begins to think
about a potential diagnosis of diabetes.
about a potential diagnosis of diabetes.
132
363628
3450
你的醫生會開始認為
你有潛在的糖尿病症狀。
你有潛在的糖尿病症狀。
06:19
So a 20 bump -- pretty significant.
133
367102
2991
所以，一下子增加20是相當明顯的。
06:22
I said, "Nick, this is very cool.
134
370601
1904
我說：「尼克，這很酷。
06:25
But, I'm sorry, we still
don't have a paper,
don't have a paper,
135
373616
2053
但，很抱歉，我們仍然沒辦法寫報告，
06:27
because this is 10 patients
and -- give me a break --
and -- give me a break --
136
375693
2579
因為只有十個病人，饒了我吧，
06:30
it's not enough patients."
137
378296
1245
病人樣本數根本不夠。」
06:31
So we said, what can we do?
138
379565
1306
所以，那怎麼辦？
06:32
And we said, let's call our friends
at Harvard and Vanderbilt,
at Harvard and Vanderbilt,
139
380895
2976
我們來打電話給哈佛
及范德堡大學的朋友，
及范德堡大學的朋友，
06:35
who also -- Harvard in Boston,
Vanderbilt in Nashville,
Vanderbilt in Nashville,
140
383895
2587
就是波士頓的哈佛
及納許維爾的范德堡，
及納許維爾的范德堡，
06:38
who also have electronic
medical records similar to ours.
medical records similar to ours.
141
386506
2821
他們都有跟我們很像的
電子病歷紀錄。
電子病歷紀錄。
06:41
Let's see if they can find
similar patients
similar patients
142
389351
2020
讓我們看看，他們是否
也可以找到相同的病人，
也可以找到相同的病人，
06:43
with the one P, the other P,
the glucose measurements
the glucose measurements
143
391395
3276
也有我們需要的已經服用這兩種藥，
06:46
in that range that we need.
144
394695
1600
並做過葡萄糖檢測的病人。
06:48
God bless them, Vanderbilt
in one week found 40 such patients,
in one week found 40 such patients,
145
396787
4955
上天保佑，范德堡一個星期內找到40個
06:53
same trend.
146
401766
1189
有同樣趨勢的病人。
06:55
Harvard found 100 patients, same trend.
147
403804
3620
哈佛找到100個有同樣趨勢的病人。
06:59
So at the end, we had 150 patients
from three diverse medical centers
from three diverse medical centers
148
407448
4281
所以，最後，我們從三個不同的
醫學中心找到150個病人
醫學中心找到150個病人
07:03
that were telling us that patients
getting these two drugs
getting these two drugs
149
411753
3297
服用過這兩種藥，
然後有葡萄糖異常增加現象。
07:07
were having their glucose bump
somewhat significantly.
somewhat significantly.
150
415074
2703
07:10
More interestingly,
we had left out diabetics,
we had left out diabetics,
151
418317
2810
有趣的是，我們沒有考慮糖尿病患者，
07:13
because diabetics already
have messed up glucose.
have messed up glucose.
152
421151
2317
因為糖尿病患者本身的
血糖濃度就已經很混亂。
血糖濃度就已經很混亂。
07:15
When we looked
at the glucose of diabetics,
at the glucose of diabetics,
153
423492
2238
當我們觀察糖尿病患者的血糖濃度時，
07:17
it was going up 60 milligrams
per deciliter, not just 20.
per deciliter, not just 20.
154
425754
3435
會上升到每公升60毫克，
不只20毫克。
不只20毫克。
07:21
This was a big deal, and we said,
"We've got to publish this."
"We've got to publish this."
155
429760
3452
這事情很重要，我們說：
「我們必須發佈這件事。」
「我們必須發佈這件事。」
07:25
We submitted the paper.
156
433236
1179
我們遞交報告，
07:26
It was all data evidence,
157
434439
2111
裡面全部都是資料證明，
07:28
data from the FDA, data from Stanford,
158
436574
2483
有來自食品藥物管理局、史丹佛的資料、
07:31
data from Vanderbilt, data from Harvard.
159
439081
1946
有來自范德堡、哈佛醫學院的資料，
07:33
We had not done a single real experiment.
160
441051
2396
我們完全沒有做任何實驗。
07:36
But we were nervous.
161
444495
1296
但我們很緊張。
07:38
So Nick, while the paper
was in review, went to the lab.
was in review, went to the lab.
162
446201
3730
所以，當報告送去審核時，
尼克就去了實驗室。
尼克就去了實驗室。
07:41
We found somebody
who knew about lab stuff.
who knew about lab stuff.
163
449955
2462
我們找到會做實驗的人。
07:44
I don't do that.
164
452441
1393
我不做實驗的。
07:45
I take care of patients,
but I don't do pipettes.
but I don't do pipettes.
165
453858
2417
我會看病人，但我不會做分量管。
07:49
They taught us how to feed mice drugs.
166
457420
3053
他們教我們如何餵老鼠吃藥。
07:52
We took mice and we gave them
one P, paroxetine.
one P, paroxetine.
167
460864
2414
我們給第一組老鼠餵食帕羅西汀，
07:55
We gave some other mice pravastatin.
168
463302
2508
給第二組老鼠餵食普伐他汀。
07:57
And we gave a third group
of mice both of them.
of mice both of them.
169
465834
3595
第三組的老鼠兩種藥都餵食。
08:01
And lo and behold, glucose went up
20 to 60 milligrams per deciliter
20 to 60 milligrams per deciliter
170
469893
3946
驚奇的是，葡萄糖每公升上升20到60毫克，
老鼠也有相同的反應。
08:05
in the mice.
171
473863
1171
08:07
So the paper was accepted
based on the informatics evidence alone,
based on the informatics evidence alone,
172
475058
3158
所以，只有資料證據的報告被接受了，
但我們在最後加了註記說，
08:10
but we added a little note at the end,
173
478240
1894
08:12
saying, oh by the way,
if you give these to mice, it goes up.
if you give these to mice, it goes up.
174
480158
2899
如果把藥物給老鼠，葡萄糖也會上升。
08:15
That was great, and the story
could have ended there.
could have ended there.
175
483081
2508
太棒了，故事其實就到這裡結束。
08:17
But I still have six and a half minutes.
176
485613
1997
但，我還有六分半鐘。
08:19
(Laughter)
177
487634
2807
（笑聲）
08:22
So we were sitting around
thinking about all of this,
thinking about all of this,
178
490465
2815
所以，我們坐下來想一下所有的事，
08:25
and I don't remember who thought
of it, but somebody said,
of it, but somebody said,
179
493304
2735
我忘記誰曾經說過，但有人說：
08:28
"I wonder if patients
who are taking these two drugs
who are taking these two drugs
180
496063
3201
「不曉得同時服用這兩種藥的病人，
08:31
are noticing side effects
of hyperglycemia.
of hyperglycemia.
181
499288
3553
是否有注意到高血糖症的副作用。
08:34
They could and they should.
182
502865
1496
他們可能知道，也必須知道。
08:36
How would we ever determine that?"
183
504761
1877
我們要如何確定？」
08:39
We said, well, what do you do?
184
507551
1443
我們說，好吧，你會怎麼做？
08:41
You're taking a medication,
one new medication or two,
one new medication or two,
185
509018
2580
你服用了一種藥，一個或兩個新藥，
08:43
and you get a funny feeling.
186
511622
1538
然後你感覺怪怪的。
08:45
What do you do?
187
513184
1151
你會怎麼做？
08:46
You go to Google
188
514359
1151
你會去問 Google，
08:47
and type in the two drugs you're taking
or the one drug you're taking,
or the one drug you're taking,
189
515534
3349
然後搜尋你在服用的一或兩個藥名，
08:50
and you type in "side effects."
190
518907
1603
然後加上「副作用」。
08:52
What are you experiencing?
191
520534
1356
你會找到什麼？
08:54
So we said OK,
192
522239
1151
所以，我們說，好，
08:55
let's ask Google if they will share
their search logs with us,
their search logs with us,
193
523414
3056
我們來問 Google 能否
跟我們分享搜尋紀錄，
跟我們分享搜尋紀錄，
08:58
so that we can look at the search logs
194
526494
1833
讓我們可以觀察搜尋紀錄，
09:00
and see if patients are doing
these kinds of searches.
these kinds of searches.
195
528351
2565
看是否有病人也在做同樣的搜尋。
09:02
Google, I am sorry to say,
denied our request.
denied our request.
196
530940
3275
很抱歉我得這麼說，
但 Google 拒絕了我們的請求。
但 Google 拒絕了我們的請求。
所以，我很煩惱。
09:06
So I was bummed.
197
534819
1151
09:07
I was at a dinner with a colleague
who works at Microsoft Research
who works at Microsoft Research
198
535994
3166
我跟一個在微軟研究室的同事吃晚餐時，
09:11
and I said, "We wanted to do this study,
199
539184
1941
我跟他說：「我們想做這個研究，
09:13
Google said no, it's kind of a bummer."
200
541149
1880
Google 說不行，我有點煩惱。」
09:15
He said, "Well, we have
the Bing searches."
the Bing searches."
201
543053
2086
他說：「我們有 Bing 搜尋引擎啊。」
09:18
(Laughter)
202
546195
3483
（笑聲）
09:22
Yeah.
203
550805
1267
是啊！
09:24
That's great.
204
552096
1151
太棒了。
09:25
Now I felt like I was --
205
553271
1151
現在，我感覺...
09:26
(Laughter)
206
554446
1000
（笑聲）
09:27
I felt like I was talking to Nick again.
207
555470
2412
我好像又在鼓勵尼克一樣。
09:30
He works for one of the largest
companies in the world,
companies in the world,
208
558437
2624
他在全世界數一數二的公司上班，
09:33
and I'm already trying
to make him feel better.
to make him feel better.
209
561085
2206
我已經開始要安慰他了。
09:35
But he said, "No, Russ --
you might not understand.
you might not understand.
210
563315
2445
但他說：「不，洛斯，你可能沒搞懂。
09:37
We not only have Bing searches,
211
565784
1500
我們不只有 Bing 啊，
09:39
but if you use Internet Explorer
to do searches at Google,
to do searches at Google,
212
567308
3340
如果你用 IE 在 Google、
09:42
Yahoo, Bing, any ...
213
570672
1891
雅虎、Bing 等任何搜尋引擎，
09:44
Then, for 18 months, we keep that data
for research purposes only."
for research purposes only."
214
572587
3643
之後18個月，我們保留這些數據
僅做研究目的使用。
僅做研究目的使用。
09:48
I said, "Now you're talking!"
215
576254
1936
我說：「這才像話嘛！」
09:50
This was Eric Horvitz,
my friend at Microsoft.
my friend at Microsoft.
216
578214
2198
這就是我的微軟朋友艾瑞克．霍維茲。
09:52
So we did a study
217
580436
1695
我們做了一項研究，
09:54
where we defined 50 words
that a regular person might type in
that a regular person might type in
218
582155
4619
我們定義出了50個
如果一般人有高血糖症時
會鍵入的關鍵字，
會鍵入的關鍵字，
09:58
if they're having hyperglycemia,
219
586798
1602
10:00
like "fatigue," "loss of appetite,"
"urinating a lot," "peeing a lot" --
"urinating a lot," "peeing a lot" --
220
588424
4762
像是疲勞、沒食慾、頻尿等。
10:05
forgive me, but that's one
of the things you might type in.
of the things you might type in.
221
593210
2767
請原諒我，但這些就是
你可能會鍵入的關鍵字。
你可能會鍵入的關鍵字。
10:08
So we had 50 phrases
that we called the "diabetes words."
that we called the "diabetes words."
222
596001
2790
所以，我們有了50個短語，
我們稱之為「糖尿病關鍵字」。
我們稱之為「糖尿病關鍵字」。
10:10
And we did first a baseline.
223
598815
2063
我們先設定了一條基準線。
10:12
And it turns out
that about .5 to one percent
that about .5 to one percent
224
600902
2704
原來，網路上有包含這些關鍵字的搜尋
10:15
of all searches on the Internet
involve one of those words.
involve one of those words.
225
603630
2982
占了0.5~1%的比例。
10:18
So that's our baseline rate.
226
606636
1742
所以，這就是我們的基準線率，
10:20
If people type in "paroxetine"
or "Paxil" -- those are synonyms --
or "Paxil" -- those are synonyms --
227
608402
4143
如果大家鍵入「帕羅西汀」或「克憂果」
──這些是同義字──
──這些是同義字──
以及剛剛其中一個關鍵字，
10:24
and one of those words,
228
612569
1215
10:25
the rate goes up to about two percent
of diabetes-type words,
of diabetes-type words,
229
613808
4890
那糖尿病類型的基準線率會上升到2%，
10:30
if you already know
that there's that "paroxetine" word.
that there's that "paroxetine" word.
230
618722
3044
如果你已經知道
「帕羅西汀」這個字的話。
「帕羅西汀」這個字的話。
10:34
If it's "pravastatin," the rate goes up
to about three percent from the baseline.
to about three percent from the baseline.
231
622191
4547
如果是「普伐他汀」，
那比率會從基準線率上升到3％。
那比率會從基準線率上升到3％。
10:39
If both "paroxetine" and "pravastatin"
are present in the query,
are present in the query,
232
627171
4390
如果「帕羅西汀」
和「普伐他汀」同時出現，
和「普伐他汀」同時出現，
10:43
it goes up to 10 percent,
233
631585
1669
那會上升到10%，
10:45
a huge three- to four-fold increase
234
633278
3461
有3到4倍的增加，
10:48
in those searches with the two drugs
that we were interested in,
that we were interested in,
235
636763
3389
用這兩種藥搜尋，會出現
我們感興趣的字在裡面，
我們感興趣的字在裡面，
10:52
and diabetes-type words
or hyperglycemia-type words.
or hyperglycemia-type words.
236
640176
3566
像是糖尿病類的字
或高血糖症類的字。
或高血糖症類的字。
我們發佈了這個研究，
10:56
We published this,
237
644216
1265
10:57
and it got some attention.
238
645505
1466
並得到一些關注。
10:58
The reason it deserves attention
239
646995
1778
它值得被關注的原因是，
11:00
is that patients are telling us
their side effects indirectly
their side effects indirectly
240
648797
4312
病人會透過搜尋，
直接告訴我們藥物的副作用。
11:05
through their searches.
241
653133
1156
11:06
We brought this
to the attention of the FDA.
to the attention of the FDA.
242
654313
2138
我們得到了食品藥物管理局的關注。
11:08
They were interested.
243
656475
1269
他們很感興趣。
11:09
They have set up social media
surveillance programs
surveillance programs
244
657768
3606
他們已經成立社會媒體監測計畫，
11:13
to collaborate with Microsoft,
245
661398
1751
與微軟展開合作，
11:15
which had a nice infrastructure
for doing this, and others,
for doing this, and others,
246
663173
2794
他們有良好的設備來做這些事，
11:17
to look at Twitter feeds,
247
665991
1282
可以觀察推特的動態、
11:19
to look at Facebook feeds,
248
667297
1716
觀察臉書的動態、
11:21
to look at search logs,
249
669037
1311
觀察搜尋日誌、
11:22
to try to see early signs that drugs,
either individually or together,
either individually or together,
250
670372
4909
嘗試觀察引發問題的
無論單一藥物或混合藥物的早期症狀。
11:27
are causing problems.
251
675305
1589
11:28
What do I take from this?
Why tell this story?
Why tell this story?
252
676918
2174
我從這件事學到什麼？
為什麼要講這個故事？
為什麼要講這個故事？
11:31
Well, first of all,
253
679116
1207
首先，
11:32
we have now the promise
of big data and medium-sized data
of big data and medium-sized data
254
680347
4037
我們現在有大數據及中型數據稱腰，
11:36
to help us understand drug interactions
255
684408
2918
來幫助我們了解藥物的相互作用，
11:39
and really, fundamentally, drug actions.
256
687350
2420
以及真實、基本的藥物作用。
11:41
How do drugs work?
257
689794
1413
藥物是如何作用？
11:43
This will create and has created
a new ecosystem
a new ecosystem
258
691231
2836
這個將會創造一個新的生態系統，
11:46
for understanding how drugs work
and to optimize their use.
and to optimize their use.
259
694091
3267
來幫助我們了解藥物如何運作
以及有效使用它們。
以及有效使用它們。
11:50
Nick went on; he's a professor
at Columbia now.
at Columbia now.
260
698303
2659
尼克繼續往前走，
他現在是哥倫比亞的教授。
他現在是哥倫比亞的教授。
11:52
He did this in his PhD
for hundreds of pairs of drugs.
for hundreds of pairs of drugs.
261
700986
4072
他用好幾百對藥物做為博士研究。
11:57
He found several
very important interactions,
very important interactions,
262
705082
2517
他找到一些非常重要的藥物交互作用，
11:59
and so we replicated this
263
707623
1214
所以，我們複製這個模式，
12:00
and we showed that this
is a way that really works
is a way that really works
264
708861
2574
展示出利用這樣做
12:03
for finding drug-drug interactions.
265
711459
2339
來尋找藥與藥之間的作用真的有效。
12:06
However, there's a couple of things.
266
714282
1734
然而，還有一些事。
12:08
We don't just use pairs
of drugs at a time.
of drugs at a time.
267
716040
3046
我們不會同時一次只服用兩種藥。
12:11
As I said before, there are patients
on three, five, seven, nine drugs.
on three, five, seven, nine drugs.
268
719110
4469
就如我之前所說的，
有病人一次是服用三、五、七、九種藥。
有病人一次是服用三、五、七、九種藥。
12:15
Have they been studied with respect
to their nine-way interaction?
to their nine-way interaction?
269
723981
3642
他們有認真研究
這九種藥的相互作用嗎？
這九種藥的相互作用嗎？
12:19
Yes, we can do pair-wise,
A and B, A and C, A and D,
A and B, A and C, A and D,
270
727647
4208
沒錯，我們可以做成對的藥，
A+B、A+C、A+D，
A+B、A+C、A+D，
12:23
but what about A, B, C,
D, E, F, G all together,
D, E, F, G all together,
271
731879
4286
但如果同一個病人
同時服用ABCDEFG，
同時服用ABCDEFG，
12:28
being taken by the same patient,
272
736189
1762
那可能會互相產生那些作用？
12:29
perhaps interacting with each other
273
737975
2118
12:32
in ways that either makes them
more effective or less effective
more effective or less effective
274
740117
3778
藥效更好或更不好？
12:35
or causes side effects
that are unexpected?
that are unexpected?
275
743919
2332
或造成那些意想不到的副作用呢？
12:38
We really have no idea.
276
746275
1827
我們真的不知道。
12:40
It's a blue sky, open field
for us to use data
for us to use data
277
748126
3756
它是個開放式的藍天領域，
讓我們可以使用數據，
讓我們可以使用數據，
12:43
to try to understand
the interaction of drugs.
the interaction of drugs.
278
751906
2502
來嘗試了解藥物彼此間的作用。
12:46
Two more lessons:
279
754848
1370
另外兩件事：
12:48
I want you to think about the power
that we were able to generate
that we were able to generate
280
756242
4199
我想要各位去想想
我們所創造出來的力量，
我們所創造出來的力量，
12:52
with the data from people who had
volunteered their adverse reactions
volunteered their adverse reactions
281
760465
4711
就是我們已經可以透過藥劑師、
病人本身、病人的醫師，
病人本身、病人的醫師，
12:57
through their pharmacists,
through themselves, through their doctors,
through themselves, through their doctors,
282
765200
3269
來取得自願者身上
他們的藥物不良反應，
他們的藥物不良反應，
13:00
the people who allowed the databases
at Stanford, Harvard, Vanderbilt,
at Stanford, Harvard, Vanderbilt,
283
768493
3667
這些人同意他們的資料可以被
史丹佛、哈佛、范德堡醫學院
史丹佛、哈佛、范德堡醫學院
13:04
to be used for research.
284
772184
1427
來做研究使用。
13:05
People are worried about data.
285
773929
1445
大家都擔心個資問題。
13:07
They're worried about their privacy
and security -- they should be.
and security -- they should be.
286
775398
3187
他們擔心自己的隱私及安全
──他們必須要擔心。
──他們必須要擔心。
我們需要保全系統。
13:10
We need secure systems.
287
778609
1151
13:11
But we can't have a system
that closes that data off,
that closes that data off,
288
779784
3406
但我們不能有一個
把資料關起來的系統，
把資料關起來的系統，
13:15
because it is too rich of a source
289
783214
2752
因為它的資源太豐盛了，
13:17
of inspiration, innovation and discovery
290
785990
3971
它對醫學界的鼓舞、
創新、發現新事物
創新、發現新事物
13:21
for new things in medicine.
291
789985
1578
實在太重要了。
13:24
And the final thing I want to say is,
292
792494
1794
最後，我想說的是，
13:26
in this case we found two drugs
and it was a little bit of a sad story.
and it was a little bit of a sad story.
293
794312
3357
我們發現這兩個藥的案例，
的確是令人難過的故事。
的確是令人難過的故事。
13:29
The two drugs actually caused problems.
294
797693
1921
這兩個藥一起服用真的會有問題。
13:31
They increased glucose.
295
799638
1475
同時服用會增加葡萄糖，
13:33
They could throw somebody into diabetes
296
801137
2446
會造成一個原本沒糖尿病的人
13:35
who would otherwise not be in diabetes,
297
803607
2294
發生糖尿病情形，
13:37
and so you would want to use
the two drugs very carefully together,
the two drugs very carefully together,
298
805925
3175
所以，各位如果想一起使用
這兩種藥，一定要非常小心，
這兩種藥，一定要非常小心，
最好不要一起服用，
13:41
perhaps not together,
299
809124
1151
13:42
make different choices
when you're prescribing.
when you're prescribing.
300
810299
2340
當你要開處方簽時，
看看有沒有不同的選擇。
看看有沒有不同的選擇。
13:44
But there was another possibility.
301
812663
1846
但，也有其他的可能。
13:46
We could have found
two drugs or three drugs
two drugs or three drugs
302
814533
2344
我們或許能找到兩或三種藥，
13:48
that were interacting in a beneficial way.
303
816901
2261
一起服用時也許可以更有效。
13:51
We could have found new effects of drugs
304
819616
2712
我們或許也可以找到
藥物本身沒有的作用，
13:54
that neither of them has alone,
305
822352
2160
13:56
but together, instead
of causing a side effect,
of causing a side effect,
306
824536
2493
但在一起服用時不但沒有產生副作用，
13:59
they could be a new and novel treatment
307
827053
2425
反而產生新作用，有可能變成最新的
14:01
for diseases that don't have treatments
308
829502
1882
絕症疾病治療方式，
14:03
or where the treatments are not effective.
309
831408
2007
或者原本的治療方式完全是無效的。
14:05
If we think about drug treatment today,
310
833439
2395
如果我們想想現今的藥物治療方式，
14:07
all the major breakthroughs --
311
835858
1752
所有的重大突破──
14:09
for HIV, for tuberculosis,
for depression, for diabetes --
for depression, for diabetes --
312
837634
4297
愛滋病、肺結核、
憂鬱症，糖尿病──
憂鬱症，糖尿病──
14:13
it's always a cocktail of drugs.
313
841955
2830
總像是藥物雞尾酒。
14:16
And so the upside here,
314
844809
1730
這件事的好處是，
14:18
and the subject for a different
TED Talk on a different day,
TED Talk on a different day,
315
846563
2849
也許哪一天不同的TED主題，
我們又會來到這裡分享，
我們又會來到這裡分享，
14:21
is how can we use the same data sources
316
849436
2593
我們要如何用同樣的資料來源
14:24
to find good effects
of drugs in combination
of drugs in combination
317
852053
3563
來找到藥物混用時產生的好效果，
14:27
that will provide us new treatments,
318
855640
2175
它將提供我們新的治療方式，
14:29
new insights into how drugs work
319
857839
1852
以及對藥物如何作用提供新的見解，
14:31
and enable us to take care
of our patients even better?
of our patients even better?
320
859715
3786
並且讓我們的病人得到更好的照顧。
14:35
Thank you very much.
321
863525
1166
非常謝謝各位。
14:36
(Applause)
322
864715
3499
（掌聲）
ABOUT THE SPEAKER
Russ Altman - Big data techno-optimist and internistRuss Altman uses machine learning to better understand adverse effects of medication.
Why you should listen
Professor of bioengineering, genetics, medicine and computer science at Stanford University, Russ Altman's primary research interests are in the application of computing and informatics technologies to problems relevant to medicine. He is particularly interested in methods for understanding drug actions at molecular, cellular, organism and population levels, including how genetic variation impacts drug response.
Altman received the U.S. Presidential Early Career Award for Scientists and Engineers, a National Science Foundation CAREER Award and Stanford Medical School's graduate teaching award. He has chaired the Science Board advising the FDA Commissioner and currently serves on the NIH Director’s Advisory Committee. He is a clinically active internist, the founder of the PharmGKB knowledge base, and advisor to pharmacogenomics companies.
Russ Altman | Speaker | TED.com