Methodological approaches to population based research of screening procedures in the presence of selection bias and exposure measurement error: colonoscopy and colorectal cancer outcomes in Ontario

Table 6 Risk reductions in CRC mortality § associated with colonoscopy -comparison of different methods

Model	Predicted probability (95% Cl)		Absolute risk reduction (95% Cl)
	No colonoscopy	Any colonoscopy
Unadjusted Model	0.23 (0.19, 0.27)	0.15 (0.12, 0.17)	-0.08 (-0.13, -0.03)
Multivariate Logistic Regression Model†	0.24 (0.18, 0.29)	0.16 (0.13, 0.34)	-0.08 (-0.19, -0.04)
Propensity Score Weighted Model‡	0.25 (0.20, 0.29)	0.17 (0.11, 0.23)	-0.08 (-0.15, -0.02)
Propensity Score Matched Model∞	0.27 (0.05, 0.49)	0.17 (0.01, 0.41)	-0.10 (-1.24, -1.04)
IV Adjusted Model∂	0.21 (0.17, 0.26)	0.04 (0.01, 0.12)	-0.17 (-0.21, -0.14)

Data are presented as percentages.
CRC, colorectal cancer; CI, confidence interval.
† Multivariate logistic model adjusted subject characteristics such as age, sex, charlson comorbidity index, rural area of residence and income quintile.
‡ Propensity score weight calculated as 1/ps for colonoscopy users and 1/(1-ps) for the non users.
∞ PS matched model on 86,592 pairs.
∂ IV adjusted probabilities were estimated using 2-stage probit models. In the first stage, the probability of receiving colonoscopy was modeled as a function of the PCP rate of colonoscopy. In the second stage, the probability of CRC incidence and mortality was modeled using the predicted probability of colonoscopy as an independent variable, adjusted for subject characteristics.
§ Mortality up to 5 years.
**Negative values indicate a reduction in the absolute risk of CRC mortality.

ISSN: 1471-2288