Unless you need to compare your results with old data there is no real reason to choose an old genome build over a newer. An exception may be that the annotation you want to use is only available in the older build, which may be the case for datasets in hg19 not transferred to hg38. But there is also Liftover (which is imperfect).

thank you for the reply. I am just looking for a logical answer because despite that we know there are some problem in using Liftover, but it supposed that we should exploit it instead of using an old genome build. I saw it in an credible article and it made me curious that is it possible to be a limitation in Affymetrix OncoScan FFPE Express approach or Nexus application to use hg19?