p********a
发帖数: 5352 | 1 ☆─────────────────────────────────────☆
yalier (丫梨儿) 于 (Sat Dec 1 12:30:37 2007) 提到:
各位大侠好，我现在做一个PROJECT，有一个DATASET里面一个VARIABLE叫作RESULT
,是用来记录MEASURE OF EFFICACY的，
36
39
40
38
32
。
。
现在想对这个变量做LOCF分析，并且每个PATIENT有一个OBSERVATION,DATASET里还有一个
变量是PATIENT ID。请问如何实现呢？我觉得该用MERGE,RETAIN，但是具体不知怎么做。
我是第一次做这种分析，SAS初学者。麻烦大家不吝赐教，多谢了！！
☆─────────────────────────────────────☆
tosi (我的名字叫/tu'zi:/) 于 (Sat Dec 1 15:15:16 2007) 提到:
What is LOCF? I mean, do you want to sum up within each patient or what? |
|
|
|
k*******a
发帖数: 772 | 4 data a;
set a;
by id;
retain locf;
if first.id then locf=a;
else a=locf;
drop locf;
run; |
|
R*********i
发帖数: 7643 | 5 It souds like a typical LOCF question:
proc sort data=a;
by id week;
data locf;
set a;
by id week;
where value1>.;
if last.id;
data out;
merge a locf (keep=id value1 rename=(value1=value2));
by id ;
proc print;
run; |
|
e****t
发帖数: 766 | 6 也报个offer，临床统计， PhD, experienced. 工作从7月份开始找到现在， 主
要是当前工作的公司在裁人，安全起见， 所以开始骑驴找马。 Recruiter 电话很多，
但是到HM 只有5个， 三个onsite， 1 个offer。 也算圆满了。工资少了些， 主要是
离LD 从250 mile 变成 90 mile， 钱不是很重要了。 有点遗憾的是， 绿卡又得从新
开始了。
还记得一些面试题， 回馈本版， 谢谢其他贴面经的朋友。
Linear regression: basic assumptions, how to diagonistic,
if violate any assumption, how to handle
Bootstrap basic idea.
Bayesian theory/Frequetist difference. How to explain to non statistician.
What is empirical Bayesian?
What is difference between logrank test and wilco... 阅读全帖 |
|
s*********r
发帖数: 909 | 7 发信人: papertigra (长工胖头猪), 信区: Statistics
标 题: CRO SAS Interview questions
发信站: BBS 未名空间站 (Fri Feb 26 21:12:00 2010, 美东)
http://www.sas9.blogspot.com/
SAS Programer Position
1. What kind of AE tables are there?
2. What difference between proc means and freq?
3. What does run statement mean?
4. What is ITT? What assessment in ITT definition is?
5. Which procedure can produce standard deviation of a variable?
6. What do put and input functions do?
7. How to validate your program?
8. How to identify... 阅读全帖 |
|
|
A*********u
发帖数: 8976 | 9 **先sort by id date time;
data locf;
set olddata;
by id date time;
retain lastnmis; ** means last no-missing value;
if first.date then lastnmis=.;
if v>.z then lastnmis=v;
else v=lastnmis;
run;
之后如果还有missing, 删掉就是了
can you tell us how you solve the problem? |
|
A*********u
发帖数: 8976 | 10 不行
首先lag记录的是上一次你用lag的时候那个var_1的值
不一定是上一个obs里var_1的值
其次,这段code不能解决有多个连续missing的情况
这样
data locf;
set olddata;
retain lastnmis; ** means last no-missing value;
if var_1>.z then lastnmis=var_1;
else var_1=lastnmis;
run;
try the lag function:
data mydata;
set mydata;
if var_1 = . then var_1 = lag1( var_1 );
run; |
|
|
A*******r
发帖数: 194 | 12 Great. Lots of practically useful questions.
I think if you don't have clinical trial experience, you don't need to know
the ansewrs to TEAE, LOCF, ITT, MedDRA, Lab datasets, lab tables, lab shift
tables. |
|
t**i
发帖数: 688 | 13 I think I knew LOCF and ITT :( |
|
s**c
发帖数: 1247 | 14 itt=intent to treatment?
what's locf? |
|
l****a
发帖数: 352 | 15 many to many merge is the most common log statement that you need to check
in the log window
other things that you need to notice include error,warning and uninitiate
MOST industry use noncopartmental method for early phase study and there is
almost no stat there.Data/model are more complex for phase iii/iv studies.
You need to pay attention to the study procedure by visits, the endpoints
defination (i.e.,LOCF) and the data/model sepec.
Also as a sas programmer, I think the key point is to build |
|
c**********e
发帖数: 2007 | 16 I am surprised to see that so many people think they know statistics, but
they can not do a statistics test in a real world problem.
There are several tests can be done. One is to use 2-way anova, which is
equivalent to pairwised t-test of the difference. Proc mixed can be used as
well.
The variable can be an AUC, or a half-life, 30%-life, or a fixed time
measurements, such as observations at 12 hour.
Missing values will not cause big problems. You can use LOCF for fixed time
value, and linear i |
|
g**r
发帖数: 425 | 17 如果你用MIXED MODEL，你的假设是MAR，忽略就可以了。
如果你不能假设MAR，去找一下SELECTION MODEL 和PATTERN MIXTURE MODEL的资料。
LOCF一般是在用ANOVA的时候才用的，虽然简单，这个现在不时髦了：被人证明存在明
显误差。
另外，还有一个保守的做法是BASELINE CARRY FORWARD，这个就狠了， PENALTY很大。 |
|
s*****n
发帖数: 2174 | 18 关键要看missing mechanism是什么,
是 MCAR, MAR, 还是 NIM? NIM 里面又分2种, outcome-based missing 和 random-
effect-based missing
简单的方法(比如mean-substitution, LOCF)最多只对MAR适用, 对于 NIM, 恐怕得用
MCMC, EM. |
|
w******8
发帖数: 59 | 19 We are doing a clinical trial where there is significant missing data at
follow up. Some were even missing baselines. ITT is the proposed analysis.
We used LOCF to impute missing data but this would not help missing
baselines. Shall I use more fancy MI technique? I haven't touched MI before
though.
Also, since ITT is the proposed analysis, when I do a response profile (with
mean and SE over the time), shall I present the imputed data or the
available cases (not just completers)? Or both?
Thanks ... 阅读全帖 |
|
m******u
发帖数: 277 | 20 LOCF
propensity score
predictive mean matching
^_^ |
|
z**********i
发帖数: 12276 | 21 想做点disparity的分析。
race在最初2年，很多missing，后来，好了很多.
有2篇文章介绍，geocoding和surname来填补missing,可能以后会试试。
basic imputation methods: case complete,locf,mean.
后来，有hot deck imputation, multiple imputation.
我也没有经验，具体在实际中该用什么，不太清楚。 |
|
k*******a
发帖数: 772 | 22 这个可以看作是个 LOCF的问题
可以用 array 来做，如果变量多的话
tedious |
|
t********m
发帖数: 939 | 23 多谢建议。我去做做locf的research。 |
|
j********t
发帖数: 201 | 24 Many people like to talk about interview questions.
Nothing is wrong about it.
However, there is something lacking: the interest in the framework behind
the questions.
I therefore encourage job hunters to ask the following questions:
(1) why and how the interview questions were formed?
(2) how to make the interviewers happy with your questions?
(3) how to improve your answers in later interviews.
To answer your first question, you need to learn the environment
biostatisticians work in. You need... 阅读全帖 |
|
j********t
发帖数: 201 | 25 Many people like to talk about interview questions.
Nothing is wrong about it.
However, there is something lacking: the interest in the framework behind
the questions.
I therefore encourage job hunters to ask the following questions:
(1) why and how the interview questions were formed?
(2) how to make the interviewers happy with your questions?
(3) how to improve your answers in later interviews.
To answer your first question, you need to learn the environment
biostatisticians work in. You need... 阅读全帖 |
|
w**********y
发帖数: 1691 | 26 try to use xts. Much more ore convenient.
(x = as.ts(1:5))
x[3] = NA
## na.omit.ts(x)
(x = xts(x,Sys.Date()+1:5))
na.omit(x)
na.locf(x) |
|
C****n
发帖数: 58 | 27 我才看到你发到我收件箱里的信。clinical data的特点是1、有些变量比较乱，因为很
多都是医生或者护士用语言描述的内容，没有规范化，所以需要数据分析人员自己去
clean。2、如何处理missing data。很多CRO都在使用Last Observation Carried
Forward来impute missing values，但是FDA在2012年的一篇论文里阐述了LOCF的缺点
，所以你可以和面试官提一下。3、规范化。因为有些clinical data是需要提交给美国
政府的，所以变量的格式有严格的要求，这个不像其它行业。4、大量的医学术语。作
为统计分析师，还是需要了解一定的医学术语和一些基本知识。譬如看到BNP这样的
variable，你可以大概知道什么样的值是outlier。
当然肯定还有其它特点，我现在只能想到这么多。
我觉得你以前处理数据的类型不是clinical没有关系，不是所有现在在clinical行业里
的人一入行开始做clinical data，关键是你要向面试官证明，你可以用SAS处理非常复
杂的数据。如果你们公司已经有了SAS Programm... 阅读全帖 |
|
