Multiple Imputation For The Non-monotone Missing Data And The Application Of Cardiac Rehabilition Comprehensive Intervention Effect Evaluation

ObjectiveCardio-cerebrovascular diseases have become the first major chronic diseases which arethe high prevalence,high morbidity,high mortality,high medical risk and high medical costs.Acute Coronary Syndromes are one of the most serious coronary heart diseases, study ACSpatients cardiac rehabilition comprehensive intervention effect evaluation method has importantpractical significance.But long follow-up time、different beginning time、differ follow-upintervals,collected data are missing,most of missingness are non-monotone missing pattern.Focusing on the per-operation、inter-operation、post operation、CCU supervise、cardiac rehabilition comprehensive intervention,we collect markers of myocardialinfarction(proBNP),provide new ideas for prognostic evaluation methods of the individualstudies.Content and MethodsIn this study,dropout can occur at any time of the survey, no rules to follow.We identifynon-monotone missing data.Using MI to fill monotone missing,continuous missing time,weexplore how to construct a longitudinal non-monotone missing data varying coefficient modelbased on multiple estimates.By using cardiac rehabilitation secondary prevention longitudinal data, we estimate themissing data by MI, to clarify the principles and methods of varying coefficient models forlongitudinal data analysis;A simulation study verifies the parameter estimation’s accuracy ofvarying coefficient models under various missing proportion and sample size.Then theproposed model is applied to longitudinal data of cardiac rehabilitation secondary prevention.ResultsThe mainly results are showed as follows:1、The sample size is between100-1000, missing proportion is between10%-60%, afterfilling appropriate number, variable coefficient model parameter estimates are more accurate.2、When sample size is fixed,with the increasing of missing proportion,the number of fill isincreasing,moreover,the standard error of parameter estimates will also increase; When themissing proportion is fixed,the number of fill is decresing with the incresing of samplesize.When the sample size is smaller then200,the missing proportion have more affect with theparameter estimates. When the missing proportion is more than50%,the parameter estimates are not good.When the number of fill is more than7, the model can always get the parameterestimates closed to the simulation truth values.Obviously,confirming the appropriate time of filling is important.3、This paper combines longitudinal monitoring data of cardiac rehabilitation secondaryprevention comprehensive intervention,further confirm the accuracy of the varying coefficientmodel parameter estimation of the different samples、different missing proportions and differentfilling times.The results showed varying coefficient models based on multiple estimates can bemore objectively explain the non-monotone missing of longitudinal data, the conclusion is inline with the actual rehabilitation medicine.The results of this study indicate that BNP of different age、 gender patients isdifferent.Women’s BNP are more than men’s.But BNP levels in the intervention group and thecontrol group of patients are not yet considered statistically significant, this may be due to thesmall sample size.In summary, this paper systematically describes the non-monotone missing datarecognition,the principle of multiple estimates,varying coefficient models、parameter estimationmethod, the computer programming and analysing of monitoring data of cardiac rehabilitationsecondary prevention comprehensive intervention.According to the relationship betweensample size and missing proportion, simulations confirm we should determine the optimalfilling time based on the actual missing data.The results of cardiac rehabilitation secondaryprevention indicate that the missing model conversion through multiple estimates,we can solvethe problem of missing data in any missing model.It also overcome the shortage that dropout iscontinuous time.So it is the optimum selection to deal with the longitudinal data of continuoustime with non-monotone dropout.