CAT ANALYSIS EPIDEM
CAT ANALYSIS EPIDEM EPI 536
Popular in Course
Popular in Epidemiology
This 5 page Class Notes was uploaded by Garry Marvin on Wednesday September 9, 2015. The Class Notes belongs to EPI 536 at University of Washington taught by Staff in Fall. Since its upload, it has received 31 views. For similar materials see /class/191973/epi-536-university-of-washington in Epidemiology at University of Washington.
Reviews for CAT ANALYSIS EPIDEM
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 09/09/15
EXAMPLE MATCHING IN CASECONTROL STUDIES BiostatEpi 536 Discussion Session November 18 2008 The following is based on a homework assignment from Dr Norm Breslow s Autumn 2006 BiostatEpi 536 course A case control study of esophageal cancer was conducted in Singapore to study the association between cancer occurrence and the consumption of cigarettes and alcoholic beverages and the temperature at which various non alcoholic hot beverages were consumed The study involved 80 male cases of Chinese background who were individually matched to four controls on the basis of age date of cancer diagnosis and other factors Subjects were also classified according to their dialect groupquot an indication of their ancestral origin in China and hence of various dietary customs The first two controls for each case were drawn from the same hospital ward whereas the second two controls were from a general orthopedic hospital We begin by restricting our analysis to matched pairs consisting of the case and the first control in each matched set We fit a model with dialect group as the only covariate three ways using ordinary logistic regression ignoring the matching MODEL 1 using ordinary logistic regression with the matched set treated as a factor with 80 levels MODEL 2 and using conditional logistic regression MODEL 3 QUESTION Write the equations for MODELS 1 2 and 3 Here dialect group is represented by the 01 variable DIAL with 1 indicating the HohkienTeochew dialect group and 0 indicating the CantoneseOther dialect group Let pi probability of cancer for subject i p probability of cancer for subject i in groupj MODEL 1 10gitpi 60 61DIALi MODEL 2 10gitpi 60 6 SETX 2 3SETi 3 sOYSETi 80 fDIALZ MODEL 3 10gitp 11 a fDIALi We fit each of MODELS 1 2 and 3 using STATA and obtain the following results OUTPUT 1 L3g 13g lmalihn 123E765 5 s l Did 7 mi l 33539143 42 165 Luciuquot 1EEES 5 quotaeaesa mparing eople of risk 0 conditional logistic regression are u QUESTION Comment on the relationship between the odds ratio estimates co f L 1 and p a 7 7 sed What do you conclude from this Using the ordinary logistic regression model odds of cancer are 386 times higher in people or 39 fa t man in 39 quot 39 However remember that the cases and controls were matched at time of data collection so some confounding has been accounted for y t e frequency matching Using the conditional model odds are 344 times higher 777ese estimates are not too different about 9 d as quot 39 39 However t 39 39 tl it a O di grgnt it is the on 39L39 mew hair iti Mi t A 0 any 5M r m 5 r A or unmeasured that are matched on INCEEEEE 1mm 385 In 1185 or about the quara at What Izshaud be m dearya Wham the number at matched 21 Increase a the ampe Ize maraaxax QUESTION WHICH ISIhe preIerred modeI In WIS c3397 th mrchmg wlthun mrmdncmg the bum 114a Harm m Mudel 2 numbers oI paIrs wIIh boIh case and conIroI In me HothenTeochew dIaIecI group we number oI paIrs wIIh case In IhIs group and serum In Ihe CanIoneseOlher aIaIecI group em I as IoHows I Controls Exposed Unexposea I Total Cases Expused unexpesea I Total I may be obIaIned Irom the 2x2 IabIe we can useme ones wIIh conIrasIIng exposure sIaIusIo esIImaIeIhe OR 73 See sIIde I 4 0I LecIure I2 Ior deIaIIs In WIS sImpIe matched paIrs sIIuaIIon we are Suppose now we return to the original dataset with 4 controls per case We t a model u u for u 101ml u Y 1 r i 1 r Gigs Samsu Wine consumption 01 samsu and beverage temperature continuous bev using conditional logistic regression and obtain the followmg xef must QUESTION How would you interpret each of the coef cients in the final model DIAL The log odds ratio of esophageal cancer for an individual With HohkienTeochew dialect compared to an individual in the same matched set With CantoneseOther dialect given cigarette Samsu Wine and hot everage consum ion are he same or these two individuals We are assuming that this log OFl comparing two individuals in the same set 3 CIGS The log odds ratio of esophageal cancer for an individual With one unit higher cigarette consumption compared to an indiw39dual in the same matched set With one unit lower cigarette consumption given dia ect Samsu Wine consumption and hot beverage consumption are the same for these two individuals SAMSU The log odds ratio of esophageal cancer for an individual who consumed amsu Wine compared to an individual in the same ma c ed set Who did not given dialect cigarette use and hot beverage consumption are the same for these two individuals BEV The log odds ratio of esophageal cancer for an individual Who consumed matched set With one unit lower temperature given dialect cigarette use and Samsu Wine consumption are the same for these two individuals Finally suppose that we lit a model with DIAL CIGS SAMSU and BEV treated as grouped linear variables to the 1 4 matched data We plot the deltarPearson and del a39ueld 39 439 39 n iquot 39 the leverages 39 L L 39 ior cases and controls and obtain the iollowing new Peamnvs average We identiiy the matched sets 30 48 and 49 as having the largest deltar Pearson and Cook s distance values By set the data ior the case and controls are as iollows Set 30 Set 48 Set 49 QUESTION Based on the data above why might these three sets be so iniluential Each has a control with high risk factor levels 148171 30 23 7 in 48 and243 in 48 quot L 39 revel 4 quot 39 39 39
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'