How did Surgisphere generate the publication data?
To simulate each observation (n=96,032) he'd have to know how to distribute all the characteristics and adjust them as needed.
To falsify tables, he'd have to account for counts carefully and generate confidence intervals and other statistics as a best guess from other papers while making sure things were fairly consistent.
And I don't think he would have published the first S3 table in the @TheLancet if this were simulated data.
