TY - JOUR
T1 - Regression Analysis of Mixed Panel-Count Data with Application to Cancer Studies
AU - Li, Yimei
AU - Zhu, Liang
AU - Liu, Lei
AU - Robison, Leslie L.
N1 - Funding Information:
The work was partly supported by NIH [R03CA219450] to Zhu.
Publisher Copyright:
© 2020, International Chinese Statistical Association.
PY - 2021/4
Y1 - 2021/4
N2 - Both panel-count data and panel-binary data are common data types in recurrent event studies. Because of inconsistent questionnaires or missing data during the follow-ups, mixed data types need to be addressed frequently. A recently proposed semiparametric approach uses a proportional means model to facilitate regression analyses of mixed panel-count and panel-binary data. This method can use all available information regardless of the record type and provide unbiased estimates. However, the large number of nuisance parameters in the nonparametric baseline hazard function makes the estimating procedure very complicated and time-consuming. We approximated the baseline hazard function to simplify the estimating procedure. Simulation studies showed that our method performed similarly to that of the previous semiparametric likelihood-based method, but with much faster speed. Approximating the baseline hazard not only reduced the computational burden but also made it possible to implement the estimating procedure in a standard software, such as SAS.
AB - Both panel-count data and panel-binary data are common data types in recurrent event studies. Because of inconsistent questionnaires or missing data during the follow-ups, mixed data types need to be addressed frequently. A recently proposed semiparametric approach uses a proportional means model to facilitate regression analyses of mixed panel-count and panel-binary data. This method can use all available information regardless of the record type and provide unbiased estimates. However, the large number of nuisance parameters in the nonparametric baseline hazard function makes the estimating procedure very complicated and time-consuming. We approximated the baseline hazard function to simplify the estimating procedure. Simulation studies showed that our method performed similarly to that of the previous semiparametric likelihood-based method, but with much faster speed. Approximating the baseline hazard not only reduced the computational burden but also made it possible to implement the estimating procedure in a standard software, such as SAS.
KW - Cancer studies
KW - Longitudinal studies
KW - Mixed penal-count data
KW - Regression
UR - http://www.scopus.com/inward/record.url?scp=85089513368&partnerID=8YFLogxK
U2 - 10.1007/s12561-020-09291-2
DO - 10.1007/s12561-020-09291-2
M3 - Article
C2 - 33747242
AN - SCOPUS:85089513368
SN - 1867-1764
VL - 13
SP - 178
EP - 195
JO - Statistics in Biosciences
JF - Statistics in Biosciences
IS - 1
ER -