Survival analysis 1 Multiple choice Questions



Statistics Advanced concepts 1 Multiple Choice Question testThis document should be completed during the course and submitted by the specified dateStudent name: Date submitted:Other information that may be relevant (optional):Important informationIn the table below please indicate the correct option(s) for each question. Some questions have more than one correct response if that is the case please ensure you indicate all the correct responses for example if a and c are the correct ones show this by a, c. MCQ numberSurvival analysisLogistic regressionMultiple regression123456789101112131415161718192021222324252627282930Survival analysis 1 Multiple choice QuestionsBurton P R, Walls J 1987 Selection-adjusted comparison of life-expectancy of patients on continuous ambulatory peritoneal dialysis, haemodialysis, and renal transplantationVariables that significantly influenced probability of survivalVariableExponential coefficient (risk multiplying factor)Statistical significanceAdverseAge (each additional decade)1.68p<0.0001Amyloidosis8.26p<0.0001Acute or acute-on-chronic presentation2.73p<0.005Ischaemic heart disease1.65p<0.025Convulsions3.17p<0.03Beneficial:Male sex0.48p<0.001Parenthood0.45p<0.001Pyelonephritis0.48p<0.02Residence in Leicestershire0.64p<0.051. Burton and Walls 1987 investigated the survival of patients on one of three types of renal replacement therapy, peritoneal dialysis, heamodialysis and transplantation details given opposite. What is the usual name for the exponential coefficient column? (one correct choice) Hazard Rate (HR)Hazard Ratio (HR)Hazard probabilityHazard proportionHazard logarithm2. Considering the results from Burton and Walls 1987 given opposite. Which is the most appropriate way of interpreting the values in the exponential coefficient column (one correct choice)OddsProbabilityTime to eventProportion failing Odds ratio3. Considering the results from Burton and Walls 1987 given above. Which variable represents the greatest hazard (one correct choice)Age (in decades)AmyloidosisConvulsionsIschaemic heart diseaseAcute or acute on chronic presentation4. Considering the results from Burton and Walls 1987 given above. Which variable represents the greatest benefit (one correct choice)Male sexParenthoodPyenonephritisResidence in LeicestershireAbsence of Ischaemic heart disease5. Considering the results from Burton and Walls 1987 given above. If anyone were considering dropping a variable from the model which one would it most likely be? (one correct choice)Male sexParenthoodPyenonephritisResidence in LeicestershireAbsence of Ischaemic heart disease6. Considering the results from Burton and Walls 1987 given above. What is the Exponential coefficient value likely going to be for the female sex? (one correct choice)0.511- 0.481+ 0.483562350552457. Considering the results from Rait et al 2010 given opposite. What is the more usual term for the Y axis? (one correct choice)Survival function S(t)LogitInverse hazardActuarial survivalProportion censored8. Considering the results from Rait et al 2010 given opposite. The cohort detail below the x axis are? (one correct choice) Irrelevant and should not be shownConfuse the issuesMore important than the graphProvide useful additional information Can be calculated from the graph9. When gathering the failure times to calculate the Kaplan Meier plot which of the following statements is correct? (one correct choice) Its accurate measurement is of minimal importance Can be grouped into equal intervalsCan be calculated from other measuresIts accurate measurement is of major importance It is best to collect then at the end of the study period only10. Which of the following are not included in the Censored observations . . .? (one correct choice) Those who experience the event during the followup period of the studyThose that are lost to followupThose that fail to provide event dataThose subjects whose survival time is less than the followup period of the studyThose who experience the event after the followup period of the study11. Censored observations are . . .? (one correct choice) More important than non-censored ones in survival analysis Are assumed to be normally distributed over timeAre assumed to have the same survival chances as uncensored observations Are essential to allow calculation of the Kaplan Meier plotAre allocated to the baseline survival curve12. A Cox regression analysis . . .(one correct choice) Is used to analyse survival data when individuals in the study are followed for varying lengths of time.Can only be used when there are censored dataAlways assumes that the relative hazard for a particular variable is constant at all times Uses the logrank statistic to compare two survival curvesRelies on the assumption that the explanatory variables (covariates) in the model are Normally distributed. Personal note: I added (taken from p. 210) but can’t find the reference!Logistic Regression Multiple choice Questions1. In Simple Logistic regression the predictor (independent variable) . . .? (one correct choice)is always interval/ratio datamust undergo a logarithmic transformation before undergoing logistic regression be in the range of 0 to 1represent ranked scoresmust be a binary variable2. A logistic regression model was used to assess the association between CVD and obesity. P is defined to be the probability that the people have CVD, obesity was coded as 0=non obese, 1=obese. log(P/(1-P)) = -2 + 0.7(obesity) What is the log odds ratio for CVD in persons who are obese as compared to not obese? (one correct choice)0.7-22.7Exp(0.7)Exp(2)Personal note: in the equation above log(P/(1-P)) = -2 + 0.7(obesity). Log(P/1-p)) is the logit function. -2+0.7(obesity) raised to e is called the Linear predictor. Equating with a + bx, a and b are log odds ratios for each of the variables. Therefore 0.7 is the log odds ratio for obesity. REMEMBER right hand side odds, left hand side separate odds ratios for each b 3. Which of the following formula produces the correct value for the probability of having CVD (Cardio Vascular Disease) from the logistic regression equation log(P/(1-P)) = -2 + 0.7(obesity) where Pi=1/1+exp(-zi) where zi is the linear Predictor (LP) (one correct choice)Pcvd= exp(-2+.7)/1- exp(-(-2+.7))Pcvd= exp(-2+.7)/ 1+ exp(-2+.7)Pcvd= exp(-2+.7)/ 1+ exp(-(-2 x .7))Pcvd= exp(-2 x.7)/ 1+ exp(-(-2+.7))Pcvd= exp(-2+.7)/ 1+ exp(-(-2+.7))4. In logistic regression the logit is . . . : (one correct choice) the natural logarithm of the odds . an instruction to record the data. a logarithm of a digit. the cube root of the sample size. 5. In binomial logistic regression the dependent (or criterion) variable: (one correct choice)is a random variable is like the median and is split the data into two equal halves. consists of two categories. is expressed in bits. 6. A model in binomial logistic regression is: (one correct choice)a set of predictors which classify cases on the dependent or criterion variable. another name for a contingency tablea miniature version of the analysis based on a small number of participants. the most common score7. Like linear regression logistic regression . . .: (one correct choice)has one or more independent variables. provides a value directly from an equation for the dependent variable Uses the same method to estimate b weights. has a dependent variable. 8. A classification table: (one correct choice)helps the researcher assess statistical significance. indicates how well a model has predicted group membership. indicates how well the independent variable(s) correlate with the dependent variable. provides a basis for calculating the exp(b) value 9. In simple logistic regression the traditional goodness of fit measure, -2(log likelihood of current model – log likelihood of previous model) is : (one correct choice)a statistic that does not follow a Chi square PDF. indicates the spread of answers to a question. an index of how closely the analysis reaches statistical significance. how close the predicted findings are to actual findings. 10. Step 0 in simple logistic regression is: (one correct choice)when there is no correlation between the predictors and the outcome variables. when there is 0 spread around the regression line. when there are no predictors only a constant term in the model. when all the predictor variables are included in the model 11. In simple logistic regression the pseudo R square values . . . (one correct choice)Provide a greater degree of accuracy than those provided in linear regressionShould not be thought of a affect size measuresAre not based on the -2Log Likelihood valuesYou should only consider the rough magnitude of them Provide the most appropriate method a assessing parameters12. Likelihood (In the statistical sense) . . (one correct choice)Is the same as a p valueIs the probability of observing a particular parameter value given a set of dataattempts to find the parameter value which is the most likely given the observed data. minimises the difference between the model and the datais another name for the probability13. A Maximun Likelihood Estimator (in the statistical sense) . . (one correct choice)Is the same as a p valueIs the probability of observing a particular parameter value given a set of dataattempts to find the parameter value which is the most likely given the observed data. Is the same as R Squareis another name for the probability14. In simple logistic regression analysis in both SPSS and R which of the following is produced in a standard output (one correct choice):Likelihoods (rather than -2log likelihoods)F statisticB (natural log odds ratio) for each parameterT statistic and associated P valueHazard functionThe table below is from a simple logistic regression analysis, for each of the boxes pointing to a place in the table select the option that correctly names and explains the column.3 4 Variables in the EquationBS.E.WalddfSig.Exp(B)95% C.I.for EXP(B)LowerUpperStep 1atime-0.0156830.0072564.6714301.0000000.0306680.9844400.9705390.998540Constant12.7273635.8031084.8101161.0000000.028293336839.852549a. Variable(s) entered on step 1: time.2 1Natural log odds ratio 15. Box one is pointing to . . . (one correct choice):-2log likelihoodsAkaike's information criterionThe value of the natural log odds ratio for the parameter estimateThe value of the odds ratio for the parameter estimateThe standard deviation of the estimated B sampling distribution16. Box two is pointing to . . . (one correct choice):-2log likelihoodsThe standard error of the Logistic functionThe value of the natural log odds ratio for the parameter estimateThe mean of the estimated B sampling distributionThe standard deviation of the estimated B sampling distribution17. Box three is pointing to . . . (one correct choice):P value associated with the linear predictor (LP)P value associated with the Wald statistic for a specific parameter estimateP value associated with the Wald statistic for all parameters combinedP value associated with the Wald statistic for the confidence interval for the specific parameter estimateP value associated with none of the above18. Box four is pointing to . . . (one correct choice):-2log likelihoodsAkaike's information criterionThe value of the natural log odds ratio for the parameter estimateThe value of the odds ratio for the parameter estimateThe standard deviation of the estimated B sampling distribution19. A logistic regression reports a Exp(B) value of .9844 for a time variable along with a p value of <.03 You would interpret the p value, remembering that the p value is a conditional probability, where the null parameter value is zero . .. . . (one correct choice):You would obtain a result like this or one more extreme three times in a hundred given that there was a strong relationship between time and the predictor variable You would obtain a result like this or one more extreme three times in a hundred given that there was not relationship between time and the predictor variable You would obtain a result like this or one less extreme three times in a hundred given that there was not relationship between time and the predictor variable You would obtain a result like this or one less extreme three times in a hundred given that there was a strong relationship between time and the predictor variable You would obtain a result like this at least 97 times (i.e. 1-.03) in a hundred given that the alternative hypothesis is true20. This question has been taken from Bland 1996 p.327 (adapted). The following table shows the logistic regression of vein graft failure on some potential explanatory variables.Logistic regression of graft failure after 6 months (Thomas et al. 1993)VariableCoef. (log odd)Std. Err.z=coef/sep95% Conf. Interval (Coef)Odds ratio exp(coef)white_cell count1.2380.2734.539<0.0010.6951.7813.448Graft type 10.1750.8760.2000.842-1.5701.9201.191Graft type 20.9731.0300.9440.348-1.0803.0252.645Graft type 30.0381.5180.0250.980-2.9863.0611.039female-0.2890.767-0.3770.708-1.8161.239.7486age0.0220.0350.6330.528-0.0480.0921.022smoker0.9980.7541.3230.190-0.5042.5012.712diabetic1.0230.7091.4430.153-0.3892.4352.7815constant-13.7263.836-3.5780.001-21.369-6.0830.00001Number of observations = 84, chi squared = 38.05, d.f.= 8, P < 0.0001From this analysis, which one of the following statements is FALSE (one correct choice):patients with high white cell counts had over 3 times the odds of having graft failurethe log odds of graft failure for a diabetic is between 0.389 less and 2.435 greater than that for a non-diabetic ignoring the statistical significancegrafts were more likely to fail in female subjects, though this is not statistically significantthere were four types of graft (hint: think reference groups)The relationship between white cell count and graft failure may be due to smokers having higher white cell counts.Following MCQs are adapted from Statistics at a glance by Petrie & Sabin 3rd edition website.21. In logistic regression . . . (one correct choice): The Wald statistic may be used to determine whether the b coefficient for a single explanatory (independent) variable is statistically significant, where the null hypothesis is that it is equal to zero.The Wald statistic may be used to determine the overall fit of the model.The Wald statistic may be used to determine whether the b coefficient for each explanatory (independent) variable is statistically significant, where the null hypothesis is that it is equal to zero.The Wald statistic may be used to determine whether the b coefficient for each explanatory (independent) variable is statistically significant, where the null hypothesis is that it is equal to 1.The Wald statistic may be used to determine whether overall any of the b coefficients are statistically significant, where the null hypothesis is that they are all equal to zero.22. In logistic regression . . . (one correct choice):The -2log(likelihood) is a measure of lack of fit for the logistic model, the smaller the value the poorer the fit between the observed data and the model.The -2log(likelihood) is a measure of lack of fit of a single b coefficient, the smaller the value the poorer the fit between the observed data and the model.The -2log(likelihood) is a measure of goodness of fit of the logistic model, the smaller the value the closer the fit between the observed data and the model.The -2log(likelihood) is a measure of goodness of fit of a single b coefficient, the smaller the value the closer the fit between the observed data and the model.The -2log(likelihood) is only used in linear regression.23. In logistic regression . . . (one correct choice):The model chi square (traditional fit measure/likelihood ratio test) provides information on overall model fit where a significant p value indicates that the current model is a better fit than the previous one.The model chi square (traditional fit measure/likelihood ratio test) provides information on overall model fit where a significant p value indicates that the current model is a worse fit than the previous one.The model chi square (traditional fit measure/likelihood ratio test) provides information on model fit for a single b coefficient, where a significant p value indicates that the current model is a better fit than the previous one.The model chi square (traditional fit measure/likelihood ratio test) provides information on model fit for a single b coefficient, where a insignificant p value indicates that the current model is a better fit than the previous one.The model chi square (traditional fit measure/likelihood ratio test) provides information on overall model fit where a insignificant p value indicates that the current model is a better fit than the previous one.The following table shows the results of a multivariable logistic regression analysis on data from the Framingham study (1951) in which there were 5209 participants on whom 9 covariates were measured at baseline. The dependent variable was whether coronary heart disease (CHD) was present (coded as ‘one’) or absent (coded as ‘zero’) after 10 years.VariableDefinitionbiP valueRRiCI for RRiSexM=0,F=1-1.588<0.0010.200.14 to 0.29Ageyears0.081<0.0011.081.07 to 1.10Heightinches-0.053<0.050.950.95 to 1.00SBPmm Hg0.009<0.021.011.00 to 1.02DBPmm Hg0.006>0.051.011.01 to 1.02Cholesterolmg/ml0.007<0.0011.011.00 to 1.01ECG abnormalY=1N=00.854<0.0012.351.67 to 3.31Relative weight100wt/median wt)%1.359<0.0013.891.89 to 8.00Alcohol consumptionoz/month-0.059>0.050.940.88 to 1.01Constant terma=-5.37024. The odds of CHD in females compared to males is what % lower (one correct choice):20406080exp(1.08)25. Which of the following is true (one correct choice):Individuals with an abnormal ECG at baseline were more than three times the odds of suffering from CHD than those with a normal ECG, after adjusting for other factors.Individuals with an abnormal ECG at baseline were more than two times the odds of suffering from CHD than those with a normal ECG, after adjusting for other factors.Individuals with an abnormal ECG at baseline were more than two times as probable to suffer from CHD than those with a normal ECG, after adjusting for other factors.Individuals with an abnormal ECG at baseline were more than three times as probable to suffer from CHD than those with a normal ECG, after adjusting for other factors.Individuals with an abnormal ECG at baseline were more than two times likely to suffer from CHD if they also had the other risk factors than those with a normal ECG..26. Which of the following is true:Cholesterol level shows a statistically insignificant result and a large effect sizeCholesterol level shows a statistically significant result and a large effect sizeCholesterol level shows a statistically insignificant result and a small effect sizeCholesterol level shows a statistically significant result and a small effect sizeCholesterol level shows a statistically significant result with no reported effect size27. Which of the following might be dropped from a future model:Diastolic blood pressure (DBP), Alcohol consumption, heightDiastolic blood pressure (DBP), Alcohol consumption Systolic blood pressure (SBP), Age, Sex, Relative weight Diastolic blood pressure (DBP), Systolic blood pressure, CholesterolNone of the variablesMultiple Regression1. A partial correlation . . . . .(one correct answer)Controls for influence on the first of the variables being correlatedControls for influence on the second of the variables being correlatedControls for influence on both of the variables being correlatedDivides the influence of the specified suppressor variable(s), equally across the X and Y variablesSuppresses the influence of the specified suppressor variable(s), equally across the X and Y variables2. A part correlation . . . . .(one correct answer)Controls for influence on the first of the variables being correlatedControls for influence on the second of the variables being correlatedControls for influence on both of the variables being correlatedControls for influence on either the first or second of the variables being correlatedSuppresses the influence of the specified suppressor variable(s), equally across the X and Y variables3. Linear multiple regression which only involves nominal predictors (inputs/independent) variables is traditionally analysed using . . . . .(one correct answer)Analysis of variance (ANOVA)Analysis of covariance (ANCOVA)Survival analysisGeneralised Estimating Equations (gee)Logistic regression 4. Linear multiple regression which involves both nominal and interval/ratio (continuous) predictors (inputs/independent) variables is traditionally analysed using . . . . .(one correct answer)Analysis of variance (ANOVA)Analysis of covariance (ANCOVA)Survival analysisGeneralised Estimating Equations (gee)Logistic regression 5. The linear multiple regression approach over ANOVA provides the following advantage. .(one correct answer)Allows analysis of non normally distributed samplesParameter estimation (B's and β's).Missing data analysisCopes better with smaller sample sizesReduced error estimates6. The squared part correlation for each of the parameter estimates in multiple linear regression represents . . . . .(one correct answer)increase in accuracy for that particular variable Percentage of error attributed to the parameterincrease in R2 for that particular parameter (unique contribution to the model)Level of normality exhibited by the parameter estimateAlternative to the R2 measure7. What does the term collinearity or multicollinearity mean with regard to multiple linear regression? . .(one correct answer)A desirable situation where there is a low correlation between one or more predictors (input variables) A undesirable situation where there is a low correlation between one or more predictors (input variables) A desirable situation where there is a high correlation between one or more predictors (input variables) A undesirable situation where there is a high correlation between one or more predictors (input variables) A undesirable situation where there is no correlation between the predictors (input variables) 8. Ross NA, Wolfson MC, Dunn JR, Berthelot J-M, Kaplan GA, Lynch JW. Relation between income inequality and mortality in Canada and in the United States: cross-sectional assessment using census data and vital statistics. BMJ 2000; 320: 898–902Ross et al. regressed mortality in working aged men against median share of income (i.e. the proportion of total income accruing to the less well off 50% of households) in 282 USA metropolitan areas and 53 Canadian metropolitan areas. The median income for the areas was included as an explanatory variable. They found the difference in slopes significant (p < 0.01), R2 = 0.51. Example courtesy of Micheal Campbell. (please select the THREE correct/true answers)The model is yi = a + b1X1i + b2X2i + b3X3i + b4X4iyi is the mortality per 100 000 for metropolitan area i, i = 1…335X1i takes the value 1 for the USA and 0 for CanadaX2i is median share of income for area i (defined above)X3i = Xi1.X2i (the product of X1i and X2i)X4i is median income for area iMortality is assumed to have a Normal distribution.The test to compare slopes is a t test with 330 degrees of freedom.The relationship between mortality and median income is assumed to be different for the USA and Canada.The relationship between mortality and median share of income is assumed linear.The variability of the residuals is assumed the same for the USA and Canada.Personal commentsa – false it is the residuals having allowed for median share of income and country that is assumed normal.b = true df =330 = 285 +53-1-4 [ b1,b2, b3, b4]c = false it is the relationship between median share that is assumed different9. Example courtesy of Micheal Campbell. In a multiple regression equation y = a + b1X1 + b2X2,(please select the TWO correct/true answers)The independent’ variables X1 and X2 must be continuousThe leverage depends on the values of y.The slope b2 is unaffected by values of X1.If X2 is a categorical variable with three categories, it is modelled by two dummy variables.If there are 100 points in the data set, then there are 97 degrees of freedom for testing b1.Personal comments:a- False x1 and x2 can be discreteb- False Depends on X1 and X2c – False changing values of X1 will alter relationship with X2 and so effect B2d - truee - true[end of document] ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download