Receipt of General Medical Care by Colorectal Cancer Patients: A Longitudinal Study

Data Sources

This study used the Surveillance, Epidemiology, and End Results (SEER) cancer registry data linked with Medicare claims to identify cancer patients, and Medicare claims from the annual 5% random sample of Medicare beneficiaries without cancer who resided in SEER registry areas to identify controls.¹⁵ SEER data include patient demographics and cancer type and stage. Medicare data include enrollment dates, health maintenance organization membership, patient demographics and, for fee-for-service beneficiaries, new billed claims that include diagnoses, tests, and procedures provided in hospitals, physician offices, and clinics. The American Medical Association's (AMA) Physician Masterfile data linked to the Medicare claims provided physician specialty.

Study Population and Care Phases

Cancer patients had CRC in stages 0 to 3 reported between 1993 and 1999 and were 67 to 89 years old at diagnosis (n = 64, 410). Sequentially excluded were individuals missing diagnosis month (n = 114), those with a simultaneous stage 4 cancer (n = 244), those with atypical histology (n = 284), and those diagnosed at autopsy (n = 108). Noncancer controls, made up of the complete 5% random sample of all Medicare beneficiaries who lived in SEER registry areas between 1993 and 1999, were randomly assigned a pseudodiagnosis (from here on called “diagnosis”) month and year between January 1, 1993, and December 31, 1999, so that, like cancer patients, they had an anchor date from which to measure study variables and care phases. The 161, 115 noncancer controls who were 67 to 89 years old at the assigned diagnosis date were included. Because we wanted to examine the association between a primary CRC diagnosis and receipt of general medical care, and because additional cancer diagnoses might influence receipt of general medical care, we excluded patients with additional SEER cancer diagnoses through 1999 (n = 13, 989).

We defined 5 care phases for the study (Figure 1): before diagnosis (13 months before 1 month prior to the diagnosis month); washout (one month before through the fifth month after the diagnosis month); after initial treatment (the year after washout); surveillance (2 to 4 years after washout); and survival (5 to 7 years after washout). We also gathered data in the year before the phase before diagnosis to define pre-existing comorbidity, the number of hospitalizations, and diabetes diagnosis. We required continuous enrollment in fee-for-service parts A and B Medicare from the year before the phase before diagnosis through the phase after initial treatment (resulting in exclusion of 27,510 cases and 79,446 controls). Based on SEER/ Medicare data availability, cases (n = 22,161) and controls (n = 81,669) with diagnoses starting in February 1993 were included.

• Download figure
• Open in new tab

Figure 1.

Study care phases.

Study individuals contributed at least 2 data years (before diagnosis and after initial treatment) and up to 8 data years (Table 1⇓⇓). Observation years with evidence of metastatic disease (eg, ablation of liver lesion; see Appendix for codes) or recurrence treatment (ie, chemotherapy and radiation administration; see Appendix for codes) were censored, as were all subsequent years, because we surmised that recurrence or metastasis would decrease both screening and monitoring of chronic disease. The average number of study years varied significantly by cancer diagnosis and stage: 4.7 for controls, 4.4 for patients with stage 0 to 1 CRC, and 4.0 for patients with stage 2 to 3 CRC.

Outcome Variables

We measured 2 types of noncancer care: (1) preventive care (see Appendix for codes), specifically mammography for women ages 67 to 74 years (n = 23, 982) and influenza vaccination; and (2) diabetes care as an example of chronic disease care, evaluating glycosylated hemoglobin (HgbA1c) and lipid test measurements (see Appendix for codes) among diabetics (n = 11, 858). These measures were standard, guideline-recommended care for older adults at the time of this study.^16,17 Individuals with diabetes had one inpatient or 2 outpatient International Classification of Diseases, Ninth Revision, Clinical Modification, diagnosis codes indicating diabetes in the year before prediagnosis (250.XX). Because influenza vaccination was not a Medicare-covered service until May 1993, we allowed a 7-month ramp-up and included only individuals with a phase before diagnosis year starting in January 1994 (n = 73, 421) in those analyses.

Independent Variable of Interest

Our independent variable of interest was CRC diagnosis status. Initial analyses identified substantial differences in noncancer care between early-stage (stages 0 and 1) and later-stage (stages 2 and 3) cancers, so we separated these groups.

Covariates

Time-independent covariates included patient sex; race (white, black, Hispanic, Asian, or other); ZIP code–based median annual household income in the diagnosis year, ZIP code–based percent of the population ≥25 years old who were high school graduates in the diagnosis year; and reason for initial Medicare entitlement (disability or end-stage renal disease vs age ≥65). The 2000 census provided the ZIP code–based variables. Time-dependent variables included age at the observation year's start (67 to 69 years, 70 to 74 years, 75 to 79 years, and 80 to 89 years); the number of hospitalizations during the year before observation (0, 1, 2, 3, 4, ≥5); the number of visits to noncancer care physicians during the observation year; the number of visits to cancer care physicians (general or colorectal surgeon, medical oncologist, radiation oncologist) during the observation year; and comorbidity during the year before observation (using the National Cancer Institute's combined comorbidity index specific to CRC patients¹⁸). We identified cancer and noncancer care physicians using linked data about primary specialty, secondary specialty, and board certification from the 1993, 1997, and 2002 AMA Physician Masterfiles. When AMA Physician Masterfile data were missing, we used the specialty code from the Medicare claims file.

Analysis

We compared the characteristics of the 3 study groups from which our study samples were drawn (patients with stage 0 to 1 CRC, patients with stage 2 to 3 CRC cases, and controls) in the phase before diagnosis. We calculated unadjusted annual rates of receipt of preventive and diabetes services by group in each phase before diagnosis, after initial treatment, surveillance, and survival phase year; we then tested for trends in receipt of these services within each of the 3 study groups over the 8 observation years and for differences in trends between our 3 study groups using the general estimating equation (GEE) approach with a logit link, using year, study group, and their interaction terms in the models. We used GEE with an unstructured correlation matrix to account for clustering by individual because each could contribute up to 8 data years. We excluded the washout period in our analyses to account for the likelihood that noncancer care would be logically deferred because of pressing cancer diagnosis and treatment needs during the time period immediately before and after cancer diagnosis.

Next, we conducted GEE analysis (logit link, unstructured correlation matrix) with cases or controls in different care phases (eg, stage 0 to 1 CRC cases in the phase before diagnosis, stage 2 to 3 CRC cases in the phase after initial treatment) as the independent variables and receipt of preventive and diabetes services in each observation year as outcomes. We adjusted for patient age and sex (when applicable) in all the models and included all other covariates described above if they were significant predictors of our outcomes at the P ≤ .05 level. The diabetes care models also adjusted for the presence of a diabetes diagnosis (one inpatient or 2 outpatient International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes) during the year before the observation year to account for variation in diabetes disease activity among individuals. Adjusted odds ratios were converted to relative risks using published methods.¹⁹ We used SAS 9.2 software (SAS Institute, Inc., Cary, NC) to conduct all analyses.

Study Group Characteristics

Controls were the youngest of the 3 study groups and patients with stage 2 to 3 CRC were the oldest (Table 2). The majority of all 3 groups were women, although controls were most likely (62.8%) and patients with stage 0 to 1 CRC least likely (53.9%) to be women. Before diagnosis, controls had the least comorbidity (76.1%; comorbidity score, ≤0) and patients with stage 0 to 1 CRC had the most comorbidity (71.7%; comorbidity score, ≤0). Before diagnosis, patients with stage 0 to 1 CRC had the greatest number of visits to noncancer care physicians. These findings were generally consistent across the influenza, mammography, and diabetes study subsamples, though the mammography subsample included only women, and the comorbidity index was not significantly different across groups in the diabetes subsample.

Noncancer Care Receipt

Influenza Vaccination

Before diagnosis, patients with stage 0 to 1 CRC had significantly higher annual influenza vaccination rates (unadjusted annual rates with confidence intervals (CI), 50.9% [49.6% to 52.1%]) than both patients with stage 2 to 3 CRC (unadjusted annual rates with CI, 46.4% [45.4%, 47.4%]) and controls (unadjusted annual rates with CI, 46.1% [45.7%–46.5%]) in unadjusted analysis (Table 3); they also had higher likelihood of influenza vaccination in adjusted analyses (adjusted relative risk [aRR], 1.09; 95% CI, 1.06–1.12); aRR, 0.99; 95% CI, 0.97–1.01; aRR, 1.0 [reference group], respectively) (Figure 2). In adjusted analysis, patients with stage 0 to 1 CRC had a higher likelihood of influenza vaccination than controls throughout the study phases and had a higher likelihood of influenza vaccination than patients with stage 2 to 3 CRC patients until the survival phase (Figure 2). Patients with stage 2 to 3 CRC significantly increased their likelihood of influenza vaccination between the phases before diagnosis and after initial treatment, and they largely sustained these increases (Figure 2). Visits with noncancer and cancer providers explained a substantial proportion of the higher likelihood of influenza vaccination among cancer patients after diagnosis, though during the survival phase significant differences in the likelihood of vaccination between both patients with stage 0 to 1 CRC and those with stage 2 to 3 CRC and controls remained (results not shown).

• Download figure
• Open in new tab

Figure 2.

Adjusted relative risk of influenza vaccination, mammography, and diabetes care measures (hemoglobin A1c and lipid testing) by cancer/control status and phase. Adjusted for the following variables: age, sex (when applicable), race ethnicity, median annual household income in ZIP code percent of those ≥25 years old who graduated from high school in zip code, reason for Medicare, National Cancer Institue combined comorbidity index, and number of hospitalizations. Note that relative risk scales differ for each measure.