%PDF-1.3
%
123 0 obj
<>
endobj
182 0 obj
<>stream
Repeated measurements are commonly collected in research settings. While the correlation coefficient is often used to characterize the relationship between two continuous variables, it can produce unreliable estimates in the repeated measure setting. Alternative correlation measures have been proposed, but a comprehensive evaluation of the estimators and confidence intervals is not available. We provide a comparison of correlation estimators for two continuous variables in repeated measures data. We consider five methods using SAS/STAT® software procedures, including a nave Pearson correlation coefficient (PROC CORR), correlation of subject means (PROC CORR), partial correlation adjusting for patient ID (PROC GLM), partial correlation coefficient (PROC MIXED), and a mixed model (PROC MIXED) approach. Confidence intervals were calculated using the normal approximation, cluster bootstrap, and multistage bootstrap. The performance of the five correlation methods and confidence intervals were compared through the analysis of pharmacokinetics data collected on 18 subjects, measured over a total of 76 visits. Although the nave estimate does not account for subject-level variability, the method produced a point estimate similar to the mixed model approach under the conditions of this example (complete data). The mixed model approach and corresponding confidence interval was the most appropriate measure of correlation as the method fully specifies the correlation structure. <br/><br/>Katherine Irimata, Arizona State University <br/><br/>Paul Wakim, National Institutes of Health <br/><br/>Xiaobai Li, National Institutes of Health
Session 2424
en
jeff.foxx@sas.com
Repeated measurements are commonly collected in research settings. While the correlation coefficient is often used to characterize the relationship between two continuous variables, it can produce unreliable estimates in the repeated measure setting. Alternative correlation measures have been proposed, but a comprehensive evaluation of the estimators and confidence intervals is not available. We provide a comparison of correlation estimators for two continuous variables in repeated measures data. We consider five methods using SAS/STAT® software procedures, including a nave Pearson correlation coefficient (PROC CORR), correlation of subject means (PROC CORR), partial correlation adjusting for patient ID (PROC GLM), partial correlation coefficient (PROC MIXED), and a mixed model (PROC MIXED) approach. Confidence intervals were calculated using the normal approximation, cluster bootstrap, and multistage bootstrap. The performance of the five correlation methods and confidence intervals were compared through the analysis of pharmacokinetics data collected on 18 subjects, measured over a total of 76 visits. Although the nave estimate does not account for subject-level variability, the method produced a point estimate similar to the mixed model approach under the conditions of this example (complete data). The mixed model approach and corresponding confidence interval was the most appropriate measure of correlation as the method fully specifies the correlation structure. <br/><br/>Katherine Irimata, Arizona State University <br/><br/>Paul Wakim, National Institutes of Health <br/><br/>Xiaobai Li, National Institutes of Health
2018-04-03T15:46:20.628-04:00
0bedacdd036837f3ac8af57cb62f1da4989a8539
382046
application/pdf
2018-04-03T15:46:20.732-04:00
Estimation of Correlation Coefficient in Data with Repeated Measures
thirdparty
support:sgf-papers
year:2018
support:customer-roles/institutional-researcher
software:STAT
support:sgf-papers/session-type/breakout
support:sgf-papers/topic/analytics/statistics
support:customer-roles/statistician
endstream
endobj
122 0 obj
<>
endobj
3 0 obj
<>
endobj
67 0 obj
<>
endobj
110 0 obj
<>
endobj
109 0 obj
<>
endobj
114 0 obj
<>
endobj
118 0 obj
<>
endobj
119 0 obj
<>stream
xZKs6Wi-Mi=NiFKIE~dp$HQ`;>D}]#jKlK<<MHnlI.'d乞d2#~|PdX9%L{drCWz;1Fpӡ
v~v3{e ɍŁ_a&ƿ` }FU`(XEwW$J~;'W맇l~ {n6U6krYn USn.2.sY)f}!bA
z_fwrڭї,GSJI-EА9RҎ,U(Ojb-'0|v%}~Z䔶|߇]~ h5tF~ZM#nۡ :!s=Z`c4A4G]wRz!s;%VBQ;<M%|ݩױ<:AAZc~?y*%5&YeK[agS-QR_[^jɞXPkN\ 'ULg t,xXԍV}ɋkz=KT#sG!54=05xlmw؟>Mn:t0;NE8e0p;Tbg>[9P;0fnf'=ٕ!97*qRvn{ka^ev=#8J$fGl8Qwwy8ʗr67LJsl%#Esig Su|GQ̺eyi+yFplxVEՁ$
j~)b_nC<"Vr$!FLRP|d:È LJ驽