2023年lab5.pdf

《2023年lab5.pdf》由会员分享，可在线阅读，更多相关《2023年lab5.pdf（7页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、 l a b 5 精品好文档，推荐学习交流 仅供学习与交流，如有侵权请联系网站删除 谢谢6 Topics:proc reg,proc logistic,proc genmod Text reference:Online documentation Lab exercise due Apr 24,2007 by 5:00 p.m.Today we will look at some modeling procedures in SAS.The materials you will need are in lab5.zip.Download and extract all the datasets

2、.Turn in the requested material in bolded italics.Create a libname to point to the location where you stored the data.For example:libname lab4“I:sascourselab4”;Part A:The REG procedure.Treadmill stress tests were administered to patients with angina pectoris before and 4 weeks after once-daily dosin

3、g with an experimental anti-angina medication.The investigator wanted to know if the improvement in exercise duration is related to the patients disease histor y.Disease duration since initial diagnoses(in years)is stored in the variable duration and percent-improvement in treadmill walking times is

4、 stored in the variable improvement.Both variables are in the dataset angina.We wish to discover if there is a significant linear relationship between improvement on medication and disease duration.a.First print the angina dataset to see what the data looks like,and then performing a simple linear r

5、egression using improvement as the outcome(dependent)variable and duration as the regressor(independent)variable.proc print data=_;run;proc reg data=_;model _;run;Next let s add a plot statement to the regression showing the outcome,predicted outcome,the lower 95%confidence level of mean predicted o

6、utcome and the upper 95%confidence of mean predicted outcome on the y-axis and duration on the x-axis.plot(_)_/overlay;run;b.This plot does not look very good.It would be nice to have some more control over the plot.In order to do this we will request proc reg to produce an output dataset containing

7、 the information we need and then make the appropriate plot using the proc gplot 荐学习交流仅供学习与交流如有侵权请联系网站删除谢谢精品好文档推荐学习交流仅供学习与交流如有侵权请联系网精品好文档，推荐学习交流 仅供学习与交流，如有侵权请联系网站删除 谢谢6 procedure.Lets call the output dataset RegResults.output out=_ p=p uclm=uclm lclm=_;run;quit;And now for the plot:goptions reset=al

8、l;proc gplot data=_;symbol1 v=dot c=blue;symbol2 v=none i=join c=red;symbol3 v=none i=join c=green line=2 repeat=2;plot(_)_/overlay haxis=0 to 7;run;quit;From the output,would you say there was a statistically significant linear relationship between outcome and duration at the 1%level?For each year

9、increase in duration,by how much would you expect the outcome to decrease?Label your answer“Part A”;Part B:More on proc reg.A study was conducted to determine the degree of recovery that takes place 90 minutes following treatment of 36 children diagnosed at a clinic with moderate to severe dehydrati

10、on.Upon diagnosis and study entry,patients were treated with an electrolytic solution at various dose levels(0,0.5,1.0,1.5,2.0,2.5,3.0 mEq/l)in a frozen,flavored,ice popsicle.The degree of rehydration was determined by using a subjective scale based on physical examination and parental input.These s

11、cores were converted to a 0 to 100 point scale,representing the percent of recovery.These data,together with the childs age and weig ht are given in the dataset dehyd.First print the dataset to take a look at the variables.The recovery variable is going to serve as our outcome variable here.Run a si

12、ngle regression model using dose,age and wt as regressors.What is the probability associated with the test that the parameters associated with the three regressors(dose,age and wt)are simultaneously zero?Are all three variables statistically significant?Add the keyword“corr”to the proc reg statement

13、 to see the correlation str ucture between the variables and add the options“SS1”and“SS2”to the model statement to request the type-1 and type-2 sum of squares.What do you notice about the correlations?荐学习交流仅供学习与交流如有侵权请联系网站删除谢谢精品好文档推荐学习交流仅供学习与交流如有侵权请联系网精品好文档，推荐学习交流 仅供学习与交流，如有侵权请联系网站删除 谢谢6 Along with

14、 the model you have run above,run three more models on the data.1.using dose only as a regressor 2.using dose and age as a regressor 3.using dose and wt as a regressor.If you had to choose a best model among the four,which one would you choose and why?Label your answers from this section“Part B”.Par

15、t C:Logistic Regression One hundred and two patients with acute myelogenous leukemia(AML)in remission were enrolled in a study of a new antisense oligonucleotide(asODN).The patients were randomly assigned to receive a 10-day infusion of asODN or no treatment(control),and the subjects were followed f

16、or 90 days.In the AML dataset you will find the response variable called outcome,with outcome=1 indicating relapse,death,or major intervention,such as bone marrow transplant before 90 days.The treatment variable,trt,is 1 if the patient received asODN and 0 if the patient was a control.a.Create a for

17、mat for the outcome variable and the treatment variable(trt).proc format;value outcome _ _;value trt _ _;run;Using a data step attach the appropriate format to the variables and create labels for the two variables.data aml;set _;label trt=_ outcome=_;format _;run;Now lets do a logistic regression wi

18、th outcome as the response variable and trt as the independent variable.One way to do this is to use proc freq.In the 2 by 2 table created by proc freq you want the treatment variable to go down the side and outcome to go across the top.proc freq data=_;tables _*_/chisq relrisk;run;荐学习交流仅供学习与交流如有侵权请

19、联系网站删除谢谢精品好文档推荐学习交流仅供学习与交流如有侵权请联系网精品好文档，推荐学习交流 仅供学习与交流，如有侵权请联系网站删除 谢谢6 From the output,what is the likelihood ratio chi-square statistic for the test that there is no association between table rows,and what is the probability associated with this test?What is the relative risk for the outcome given

20、you were given asODN versus you were not given asODN?b.In order to perform a logistic regression using proc freq the data have to be particularly simple.A more general way is to model the data using proc logistic.Lets repeat the model you ran with the proc freq using proc logistic instead proc logis

21、tic data=_ _;model _;run;Make sure that you are modeling the correct outcome.At the top of the output from this procedure SAS will tell you which value you are modeling.If it is not correct what do you need to do?Make sure you get the same likelihood ratio chi-square statistic and probability as bef

22、ore.What is the odds ratio associated with the treatment effect?Is the effect significant at the 5%level?c.The time to remission from diagnosis or prior relapse at study enrollment was considered an important covariate in predicting response.This variable is included in the AML dataset and is called

23、 Rtime measured in months.Repeat the logistic regression in b.but this time add the Rtime variable to the model.What now is the odds-ratio and p-value associated with the treatment effect?Label your response to this section as“Part C”.Part D:Ordinal logistic regression Proc logistic can be used to f

24、it ordinal logistic regression(and multinomial logistic regression).For example we may score a patients progress using four categories:“worse”,“no change”,“improved”and“cured”and may treat this as an ordered outcome with“worse”being less than“no change”which in turn is less than“improved”and that in

25、 turn being less than“cured”.By default proc logistic will fit an ordinal logistic regression when you supply it with an outcome variable with more than two categories.Patients with severe symptoms of gastroparesis where randomized to receive an experimental therapy(A)or dietary changes with watchfu

26、l waiting(B).Response was measured after 7 days,using the following scale:1=no response,2=some response,3=marked response,or 4=complete response,based on the degree of symptom relief.A patients history o f severe gastorparesis,thought to be an important covariate,was also recorded as no prior episod

27、es,(0),one prior episode(1),or“more than one prior episode(2).The data is given in the dataset gi,with,response being the outcome variable,hist containing the patients history,and rx being the variable containing the treatment group.荐学习交流仅供学习与交流如有侵权请联系网站删除谢谢精品好文档推荐学习交流仅供学习与交流如有侵权请联系网精品好文档，推荐学习交流 仅供学

28、习与交流，如有侵权请联系网站删除 谢谢6 Perform a logistic regression using hist and rx as independent variables.Make sure you model the probability of increasing response and that rx is included as a categorical variable.In the SAS/Help look at contents SAS Products SAS/STAT the logistic procedure syntax class to det

29、ermine how to give a reference parameterization to the rx variable.The ordinal logistic regression invokes a“proportional odds assumption”,which in this case includes the assumption that the odds ratio associated with rx and with hist are the same across different comparisons in the ordinal response

30、.The logistic output shows a test for this assumption.Find this in the output.Does the“proportional odds assumption”hold in this case?From the output,do you see any evidence that there is a difference in response between patients who received the experimental therapy and the untreated patients?Cut a

31、nd paste your proc logistic program and include it in this labs email.Label this Part D.Part E:Proc Genmod/Logistic Regression For this section of the lab,we will continue to work with subset of the 1987 National Medical Expenditure Survey.Today we will be interested in modeling the prevalence of ma

32、jor smoking related diseases as a function of age,gender,and smoking status after adjusting for the additional covariates:marital status and education level.i.Open the lab5a.sas program.(It is a SAS program and should be opened in the editor window.)ii.Change the libname statement so that it will cr

33、eate the appropriate library for where you stored the data,i.e.change the directory.iii.Notice that below the libname statement there is a proc format provided to create formats for the nmes data.(This proc format is the same as the one you created last week.)iv.Perform a logistic regression to mode

34、l the prevalence of lung cancer,laryngeal cancer or COPD as a function of age,gender and smoking status,after adjusting for marital status and education level proc genmod data=lab5.nmes;class marital educate smoke male;model lc5=lastage male lastage*male smoke smoke*male marital educate/dist=binomia

35、l link=logit;format smoke smklb.male genlb.marital marlb.educate edulb.;run;v.As it stands this regression is modeling the probability of not having lung cancer,荐学习交流仅供学习与交流如有侵权请联系网站删除谢谢精品好文档推荐学习交流仅供学习与交流如有侵权请联系网精品好文档，推荐学习交流 仅供学习与交流，如有侵权请联系网站删除 谢谢6 laryngeal cancer or COPD.Modify the code so that it

36、 models the probability of having one of those outcomes.Had we not included the dist=binomial and link=logit options in the model statement,proc genmod would have assumed that the response variable,lc5,comes from what distribution?vi.Using the code provided above,perform a logistic regression to mod

37、el the prevalence of CHD given that the person does not have lung cancer,laryngeal cancer or COPD as the same function of age,gender,smoking status,marital status and education used in part iv.Copy and paste the section of your log file that contains the proc genmod commands to complete this problem

38、.vii.We described during the lecture that proc genmod can be used for response variables that follow a normal distribution,binomial(proportion)and Poisson.Using SAS Help,list the remaining distributions that SAS allows you to specify in the DIST option.viii.We described in class the identity,logit and log link functions that can be used in proc genmod.Using SAS Help,list the remaining link functions that SAS allows you to specify in the LINK option.Label your response to this section as“Part E”.荐学习交流仅供学习与交流如有侵权请联系网站删除谢谢精品好文档推荐学习交流仅供学习与交流如有侵权请联系网