RT Journal
A1 Braillon A
T1 PRediction and statistical analysis
JF Archives of Surgery
JO Archives of Surgery
YR 2011
FD May 1
VO 146
IS 5
SP 633
OP 633
DO 10.1001/archsurg.2011.82
UL http://dx.doi.org/10.1001/archsurg.2011.82
AB 
In a short series of 67 patients, Oussoultzoglou et al1 studied 26 variables for the prediction of mortality after major hepatectomy in patients without cirrhosis. This deserves several comments. First, for statistical analysis, the case-to-variable ratio is at least as important as the absolute number of cases. Second, the use of cut points to derive subgroups is not appropriate when there is a continuous distribution of the values with no obvious modal values. Third, the P values must be adjusted for multiple testing (n = 52?). Oussoultzoglou et al1 at least performed a second set of analyses in a subgroup of 47 patients. Last but not least, the significance of post hoc (retrospective) analysis must always be challenged because it gives no more than an hypothesis that must be tested in a prospective series.