Simple and Multivariable Logistic Regression
When controlling for other covariates the adjusted model resulted in a
decreased likelihood of a discordant referral for those with high-grade
lesions (ASCH/AGC/HSIL) (OR: 0.03; CI: 0.006, 0.166; p<0.001)
compared to those with low grade or no lesions (ASCUS/LSIL/NILM)
results. These results suggest that those with low-grade lesions are 97
times more likely to be incorrectly referred for a colposcopy. The
adjusted models’ odds of a discordant referral with high-grade lesions
was much lower than the univariate regression for the odds of a
discordant referral with high-grade lesions (OR: 0.75; CI: 0.58, 0.96;
p<0.05). The likelihood of a discordant referral from a family
medicine provider (OR:7.72; CI: 2.23, 26.69; p<0.01)and
internal medicine provider (OR:10.94; CI: 2.56, 46.79; p<0.01)
compared to an OBGYN is also much higher in the adjusted model compared
to the likelihood of a discordant referral from a family medicine
provider (OR:4.12; CI:1.58,10.74; p<0.01) and internal
medicine (OR:5.42; CI:1.63,18.03; p<0.01) physician in the
unadjusted model. The likelihood of a discordant referral in patients
under 24 years of age (OR: 19.55; CI: 7.62,50.15; p<0.001)
compared to women between the ages of 25 and 29 is much higher in the
adjusted model compared to women under the age of 24 (OR:7.80;
CI:4.09,14.86; p<0.001) in the unadjusted model. The
multivariate logistic model strongly predicted similar rates of correct
referrals to those observed in the sample (c-statistic 0.873).