Chapter 3 METHODS
3.8 Statistical analysis
To answer the research questions, statistical models were run to examine associations between the independent variables of interest and screening colonoscopy quality indicators (dependent variables). Indicators studied are defined earlier. For the protocol type, we compared the 1-person technique specialists group and the 2-person technique specialists group with 2-person technique PCPs group. As for the sedation type, we compared the propofol sedation procedures with midazolam-meperidine sedation procedures. In each model, we controlled for bowel preparation status because bowel preparation is a patient-dependent variable that greatly influences the quality indicators. GEE modeling was used to account for patients clustered within physician. SAS version 9.3 is used.
3.8.1 Model 1: Procedure time (continuous variable)
A linear GEE regression model was used to investigate protocol type/sedation type using “proc genmod” syntax in SAS.
Yprocedure time=β0 + β1*2-person technique/physician specialty + β2*patient age +
β3*patient gender + β4*patient race + β5* number of polyps found + β6* sedation type
+ β7* bowel preparation + β8*sedation type*bowel preparation + β9*training
procedure status +εerror
This linear GEE regression model tested the association between protocol type (1- person technique vs. 2-person technique) and the sedation type and the procedure
84
For all models we tested the interaction term of sedation type with bowel
preparation and because it was statistically significant, we compared procedure times for Midazolam-meperidine/Fair (bowel prep), Midazolam-meperidine/Excellent,
Propofol/Poor, Propofol/Fair, and Propofol/Excellent to Midazolam-meperidine/Poor. However on comparing the models with the above categories with the two variables modeled separately, the results were not substantially different, but readily interpretable. Hence the latter results were used for interpretation.
3.8.2 Model 2: Polyp detection likelihood (dichotomous variable)
A logistic GEE regression model was used to investigate protocol type/sedation type using “proc genmod” syntax in SAS.
Ypolyp detected =β0 + β1*2-person technique/physician specialty + β2*patient age +
β3*patient gender + β4*patient race + β5* sedation type + β6* bowel preparation +
β7*sedation type*bowel preparation + β8*training procedure status + εerror
This logistic GEE regression model tested the association between protocol type (1-person technique vs. 2-person technique) and the sedation type and the polyp detection controlling for the remaining variables.
3.8.3 Model 3: adenoma detection likelihood (dichotomous variable)
A logistic GEE regression model was used to investigate protocol type/sedation type using “proc genmod” syntax in SAS.
Yadenoma detected=β0 + β1*2-person technique/physician specialty + β2*patient age +
β3*patient gender + β4*patient race + β5*sedation type + β6* bowel preparation +
85
This logistic GEE regression model tested the association between protocol type (1-person technique vs. 2-person technique) and the sedation type and the adenoma detection controlling for the remaining variables.
3.8.4 Model 4: Advanced neoplasms detection likelihood (dichotomous variable)
A logistic GEE regression model was used to investigate protocol type/sedation type using “proc genmod” syntax in SAS.
Yadvanced neoplasms detected=β0 + β1*2-person technique/physician specialty + β2*patient
age + β3*patient gender + β4*patient race + β5*sedation type + β6* bowel preparation
+ β7*sedation type*bowel preparation + β8*training procedure status+εerror
This logistic GEE regression model tested the association between protocol type (1-person technique vs. 2-person technique)/sedation type and advanced neoplasm detection likelihood controlling for the remaining variables.
3.8.5 Model 5: likelihood of finding additional polyps (ordinal variable)
An ordered logistic GEE regression model to investigate protocol type/sedation type using “proc genmod” syntax with “dist=multinomial” option in SAS.
Ynumber of polyps found =β0 + β1*2-person technique/physician specialty + β2*patient age +
β3*patient gender + β4*patient race + β5* sedation type + β6* bowel preparation + β7*
sedation type*bowel preparation+ β8*training procedure status+ εerror
This ordered logistic GEE regression model tested the association between protocol type (1-person technique vs. 2-person technique) and the sedation type and likelihood of finding an additional polyp in a patient controlling for the remaining variables.
86
3.8.6 Model 6: likelihood of finding additional adenomas (ordinal variable)
An ordered logistic GEE regression model to investigate protocol type/sedation type using “proc genmod” syntax with “dist=multinomial” option in SAS.
Ynumber of adenomas found =β0 + β1*2-person technique/physician specialty + β2*patient age
+ β3*patient gender + β4*patient race + β5* sedation type + β6* bowel preparation +
β7* sedation type*bowel preparation+ β8*training procedure status+ εerror
This ordered logistic GEE regression model will test the association between protocol type (1-person technique vs. 2-person technique) and the sedation type and the ability of finding an additional adenoma in a patient, which controls for the remaining variables.
3.8.7 Model 7: right colon polyps (ordinal variable) likelihood
An ordered logistic GEE regression model was used to investigate protocol type/sedation type using “proc genmod” syntax with “dist=multinomial” option in SAS.
Yat least one right colon polyp detected=β0 + β1*2-person technique/physician specialty +
β2*patient age + β3*patient gender + β4*patient race + β5* sedation type + β6* bowel
preparation + β7* sedation type*bowel preparation + β8*training procedure status +
εerror
This ordered logistic GEE regression model tested the association between protocol type (1-person technique vs. 2-person technique PCPs and 2-person technique specialists) and sedation type vs. the likelihood of a right colon polyp detection, which controls for the remaining variables.
87
An ordinal logistic GEE regression model to investigate protocol type/sedation type using “proc genmod” syntax with “dist=multinomial” option in SAS.
Ypolyp size =β0 + β1*2-person technique/physician specialty + β2*patient age +
β3*patient gender + β4*patient race + β5* sedation type + β6* bowel preparation + β7*
sedation type*bowel preparation+ β8*training procedure status+ εerror
This ordered logistic GEE regression model tested the association between protocol type (1-person technique vs. 2-person technique) and the sedation type and the likelihood of finding a small adenoma vs. medium adenoma and large adenoma in a patient controlling for the remaining variables.