Nested case-control analysis with general additive-multiplicative hazard models.pdfVIP

App1．M ath．J．ChineseUniv 2012，27(2)：159—168 Nested case—controlanalysiswith general additive—multiplicativehazardm odels ZHENG Ming LIN Ren—xin SUN Yi YU Wen Abstract．This paper discusses the nested case—controlanalysisunder a class ofgeneral additive-multiplicativehazard modelswhich includestheCox modeland theadditivehazard modelasspecialcases．A pseudo—scoreisconstructedtoestimatetheregression parameters． Theresultingestimatorisshowntobeconsistentandasymptoticallynormallydistributed．The limiting varianee~covariancematrix can be consistently estimated by the nested case—control data．A simulationstudyisconductedtoassessthefinitesampleperformanceoftheproposed estimatorand arealexampleisprovidedforillustration． §1 Introduction Inlargeepidemiologicalstudiesandpreventiontrials，theassemblingofcovariatehistories CaB beextremelyexpensiveifitisdoneonthefullcohort．Severaleconomicalstudydesigns havebeen proposed to reducecostwhentheregressionmodelbetween covariatesand event timeisofmaininterest．Thenestedcase—control(NCC)designsuggestedbyThomas[]is probablyoneofthemostcommonlyusedonesamongthesecost—effectivedesigns．Insuchkind ofdesign，controlsaresampledfrom therisksetateventtimes．Meanwhile，subsequentcontrol samplingsareindependent．Thecovariatesarerecordedonlyforthesubjectsexperiencingthe eventandthesampledcontrols． TheNCC designhasreceivedmuch attention sinceitwasproposed．Theanalysisofsuch designhasbeenmainlyfocusedonsem iparametricestimationprocedureunderthemostpopular Coxmodel[4]which assumesthatthehazardfunction ofthesurvivaltimeassociatedwith a multidimensionalandpossiblytime—dependentcovariatex(t)follows A(tl2(t))=A0(t)exp{稍。())， (1) wherex(t)= { ()，0≤ ≤￡}，M(t)isanunspecifiedbaselinehazardfunctionand，y0 isanunknownparametervectorofinterest．S