The relationship between PCP visits and previous CRC screening was evaluated using multivariable logistic regression. Likewise, we used multivariable logistic regression models to examine the relationship between PCP visits and early-stage diagnosis, excluding persons with unknown stage (n=8049). Odds ratios of early-stage (0 or I) compared with late-stage (II, III, or IV) diagnosis and corresponding 95% CIs were calculated for each category of PCP visits compared with the reference group (0 or 1 visit). To determine whether PCP visits were associated with early-stage diagnosis beyond receipt of previous CRC screening, we fitted logistic models with and without previous CRC screening and examined changes in the estimated odds ratio for PCP visits.
We considered the following potential confounders in multivariable models: number of non-PCP visits, age at diagnosis, sex, race/ethnicity, marital status at diagnosis, census-derived measures of median household income (approximate quin-tiles within each registry), educational levels (approximate quintiles within each registry), metropolitan statistical area, SEER geographic registry (with indicator variables for each registry), Charlson comorbidity index29 (link) (determined from both inpatient and outpatient physician claims), anatomic site (proximal vs distal lesions), and histologic cancer type (adenocarcinomas, including all subtypes, carcinoid tumors, and other). Medicare reimbursement of CRC screening occurred incre mentally, with no coverage from 1994 to 1997, limited coverage between 1998 and 2000 (FOBT yearly, flexible sigmoidos-copy every 4 years, and colonoscopy only for high-risk persons), and full coverage thereafter (colonoscopy every 10 years for all persons). We therefore created indicator variables corresponding to these 3 periods and included them in the models. To account for healthy behaviors in patients seeking more primary care,30 (link) we added receipt of an influenza vacciation in the preceding 3 to 27 months as a proxy for healthy behavior.
We examined CRC-specific mortality and all-cause mortality among persons having invasive CRC (excluding 7732 in situ cancers).8 (link),28 (link) The association between PCP visits and CRC mortality was analyzed using Cox proportional regression models, adjusting for potential confounding factors described in the preceding paragraph in addition to tumor characteristics to control for residual confounding within each stage. To determine whether associations between PCP visits and CRC mortality were primarily the result of previous CRC screening and an earlier stage at diagnosis, models were first performed without previous CRC screening, stage at diagnosis, and tumor characteristics and then repeated including previous CRC screening, stage at diagnosis, and tumor characteristics. Similar analyses were performed to examine the relationship between PCP visits and all-cause mortality. All analyses were performed using commercial software (SAS version 9.2; SAS Institute, Inc, Cary, North Carolina). Data are given as mean (SD) unless otherwise indicated.
We considered the following potential confounders in multivariable models: number of non-PCP visits, age at diagnosis, sex, race/ethnicity, marital status at diagnosis, census-derived measures of median household income (approximate quin-tiles within each registry), educational levels (approximate quintiles within each registry), metropolitan statistical area, SEER geographic registry (with indicator variables for each registry), Charlson comorbidity index29 (link) (determined from both inpatient and outpatient physician claims), anatomic site (proximal vs distal lesions), and histologic cancer type (adenocarcinomas, including all subtypes, carcinoid tumors, and other). Medicare reimbursement of CRC screening occurred incre mentally, with no coverage from 1994 to 1997, limited coverage between 1998 and 2000 (FOBT yearly, flexible sigmoidos-copy every 4 years, and colonoscopy only for high-risk persons), and full coverage thereafter (colonoscopy every 10 years for all persons). We therefore created indicator variables corresponding to these 3 periods and included them in the models. To account for healthy behaviors in patients seeking more primary care,30 (link) we added receipt of an influenza vacciation in the preceding 3 to 27 months as a proxy for healthy behavior.
We examined CRC-specific mortality and all-cause mortality among persons having invasive CRC (excluding 7732 in situ cancers).8 (link),28 (link) The association between PCP visits and CRC mortality was analyzed using Cox proportional regression models, adjusting for potential confounding factors described in the preceding paragraph in addition to tumor characteristics to control for residual confounding within each stage. To determine whether associations between PCP visits and CRC mortality were primarily the result of previous CRC screening and an earlier stage at diagnosis, models were first performed without previous CRC screening, stage at diagnosis, and tumor characteristics and then repeated including previous CRC screening, stage at diagnosis, and tumor characteristics. Similar analyses were performed to examine the relationship between PCP visits and all-cause mortality. All analyses were performed using commercial software (SAS version 9.2; SAS Institute, Inc, Cary, North Carolina). Data are given as mean (SD) unless otherwise indicated.
