M. Rabilloud et al., ANALYSIS OF MEDICAL-PRACTICE VARIATIONS W ITH A 2-LEVEL LOGISTIC MODEL - EXAMPLE OF THE PROPHYLACTIC CESAREAN-SECTION, Revue d'epidemiologie et de sante publique, 45(3), 1997, pp. 237-247
We show the use of a hierarchical logistic model to study the variatio
ns of the prophylactic cesarean section rate between the maternity hos
pitals of the Rhone-Alpes region. These variations are analyzed accord
ing to the women characteristics at first level, and the maternity hos
pital characteristics at second level. We present the two-level hierar
chical logistic model and the method of estimation of the fixed and ra
ndom parameters. Then, we compare and discuss the results obtained wit
h those of the usual logistic model. The usual logistic model underest
imates the standard error of the regression parameters. In our example
however, the results obtained with the hierarchical model do not modi
fy the conclusions concerning the effect of the women characteristics.
All the women characteristics increase significantly the probability
for a woman to have a prophylactic cesarean section. Nevertheless, the
hierarchical model reveals the effect of the maternity hospital chara
cteristics and shows that the maternity hospitals which receive many '
'at risk'' women tend to perform fewer prophylactic cesarean sections
than the others, in women with the same characteristics. It permits to
estimate the residual variance of second level linked to the unobserv
ed characteristics of the maternity hospitals. It permits to show that
the effect of the main characteristics of the women (previous cesarea
n section, dystocia, chronic fetal distress) vary between maternity ho
spitals.