27  Results

Using the Generator I created a dataset 1000 times larger than the initial one.

10000 Times didn’t work…

The Generator is not perfect so some of its output contains only 0. Since that is clearly impossible I decided to remove those rows. In total they were less than 2% of the generated dataset.

I then calculated the mean number of cases for each events I would get if I generated a dataset of the same size as the initial one

Varnames	0_real	0_synthetic	diff_perc
B_COAGDIS_AESI	73.0	73.715309	-0.98
C_ARRH_AESI	65.0	69.045921	-6.22
C_CAD_AESI	31.0	31.479591	-1.55
C_MYOCARD_AESI	20.0	17.986734	10.07
C_VALVULAR_AESI	50.0	51.130611	-2.26
DEATH	258.0	253.684692	1.67
D_PANCRACUTE_AESI	13.0	9.452041	27.29
E_DM1_AESI	38.0	44.208164	-16.34
E_GOUT_AESI	27.0	25.344898	6.13
G_KIACUTE_AESI	15.0	14.710204	1.93
G_UTI_AESI	23.0	23.219387	-0.95
I_INFLUENZA_AESI	16.0	14.138776	11.63
M_FRACTURES_AESI	60.0	59.522449	0.80
M_OSTEOARTHRITIS_AESI	28.0	27.116327	3.16
N_STROKEHEMO_AESI	10.0	9.992857	0.07
SO_OTITISEXT_AESI	36.0	37.114285	-3.10
V_THROMBOSISARTERIALALGOR_AESI	45.0	45.730614	-1.62
V_VTEALGORITHM_AESI	19.0	19.263266	-1.39

The script reproduce well most of the variables. However D_PANCRACUTE_AESI, C_MYOCARD_AESI and I_INFLUENZA_AESI are slightly underrepresented in the generated data, while E_DM1_AESI on the other hand is over represented.

The Mean Square Error is 5.63 which is good enough in this case.

Finally I wantef to see the number of events each persons has:

Number of events	0_real	0_synthetic	diff_perc
1.0	639	639.180612	-0.03
2.0	52	52.792857	-1.52
3.0	28	27.562245	1.56

The script does a very good job in this case. I removed the impossible cases when the person has 0 events but even with that taken into account the performance does not drop much.

The Mean Square Error is only 0.28 which is very good.