Relation Between Medicare Screening Reimbursement and Stage at Diagnosis for Older Patients With Colon Cancer FREE

Regular screening is the most effective mechanism for detecting colon cancer at an early, curable stage.^{1 - 6} Although widespread use of colorectal cancer screening has been advocated by professional organizations, dissemination and adoption at the population level has been suboptimal.^{7 - 9} The 2000 National Health Interview Survey demonstrated that 42.5% of respondents 50 years of age and older were up to date with colon cancer screening with any of the recommended modalities; among persons 65 years of age and older, 48.7% were up to date.¹⁰ Studies focusing on endoscopic screening have reported even lower adherence rates, with only 23% to 43% of eligible patients reporting endoscopic screening at the recommended time intervals.^{11 - 13}

There are several important barriers to colon cancer screening including patient and physician awareness, preferences, and access to care.¹⁴ Several studies have suggested that lack of insurance coverage may be one of the most important barriers to colon cancer screening.^{12 ,15} This may have directly affected the use of colon cancer screening among older persons because prior to 1998, Medicare did not routinely reimburse for colon cancer screening. In response to these concerns, Congress included a provision to ensure Medicare reimbursement for colon cancer screening procedures in the Balanced Budget Act of 1997. As of January 1, 1998, Medicare was required to pay for colon cancer screening tests including fecal occult blood testing, barium enema, and sigmoidoscopy. Screening colonoscopy was also covered in individuals who faced an increased risk of colorectal cancer, such as a family history. The Consolidated Appropriations Act of 2001 extended the screening colonoscopy benefit to all individuals, regardless of risk, beginning on July 1, 2001.¹⁶

Prior analyses of the effect of the 1998 reimbursement change on the use of colon cancer screening tests have yielded mixed results. An evaluation of Behavioral Risk Factor Surveillance System data from approximately 58 000 Medicare beneficiaries found no significant increase in colorectal cancer screening with either sigmoidoscopy or fecal occult blood testing between 1997 and 1999.¹⁷ Another review of 1.3 million Medicare beneficiaries' claims in the state of Washington analyzed changes in colorectal cancer screening between 1994 and 1998. Although the overall rate of screening did not change significantly, there was a small but significant increase in the proportion of beneficiaries who had received screening colonoscopy (1.04% in 1994 vs 1.12% in 1998; P = .003).¹⁸ Following the 2001 reimbursement change, a longitudinal study of 580 gastroenterologists noted a trend toward increased use of colonoscopy for screening purposes.¹⁹

It is unclear whether the expanded colorectal cancer screening reimbursement policies have translated into clinical benefit for Medicare beneficiaries. We therefore evaluated whether the implementation of these policies was associated with changes in the use of colonoscopy among Medicare beneficiaries without cancer, as well as changes in the proportion of colon cancer patients who were diagnosed at an early stage. We also determined whether the new policies had a stronger effect on stage of presentation for patients with lesion of the proximal colon. Not only are proximal tumors reliant on colonoscopy (rather than other modalities such as sigmoidoscopy) for early detection, but they are increasingly common among older colon cancer patients.^{20 - 21} Proximal colon cancers, being less likely to present with gastrointestinal symptoms such as changes in bowel habits than distal lesions, may be more dependent on regular screening in order to be diagnosed at an early stage. Hence, we hypothesized that facilitating access to screening colonoscopy would have a stronger effect on early detection of proximal lesions than distal lesions.

Our objective was to assess whether there was an independent relation between the reimbursement policies and patterns of care with regard to both screening and diagnosis of colorectal cancer in Medicare beneficiaries. We used 3 approaches to account for secular trends in colorectal cancer screening and diagnosis. First, to ascertain if the Medicare policies were associated with an increased use of colonoscopy, we identified a cohort of Medicare beneficiaries without cancer who lived in one of the original Surveillance, Epidemiology, and End Results (SEER) areas during the study period. We then compared changes in the rate of colonoscopy with that of flexible sigmoidoscopy. Because it is not feasible to reliably distinguish screening from evaluative (diagnostic) endoscopies using administrative claims data, we included all colonoscopies and flexible sigmoidoscopies that were included in outpatient and physician billing files in this analysis. Assuming that the burden of disease that would require an evaluative colonoscopy has not changed during the short time interval of the study period, our assumption was that changes in the use of endoscopy during the study period were related to screening.

After identifying a sample of patients with colon cancer, we used 2 control groups for our analyses. First, because colonoscopy is particularly well suited to identify cancerous lesions of the proximal colon, we a priori elected to perform analyses of temporal trends in stage at diagnosis for patients with colon cancer, stratifying by tumor location. For example, patients with distal lesions served as an internal control group, in order to compare trends in stage at diagnosis with patients with proximal colon lesions.

Second, because the reimbursement policies applied to beneficiaries with fee-for-service insurance, we also constructed a control group of cancer patients who were members of Medicare health maintenance organizations (HMOs) during the study period. Prior work has demonstrated that Medicare HMO patients had greater access to preventive services and were more likely to have cancers diagnosed at an earlier stage in the 1980s and early 1990s.^{22 - 24}

We hypothesized that the new reimbursement policy, which was directed toward fee-for-service beneficiaries, would have a greater effect on stage at diagnosis for fee-for-service beneficiaries than those with HMO coverage. We therefore compared trends in stage at diagnosis between cancer patients with fee-for-service and those with Medicare HMO coverage. Institutional review board approval of this study was granted by the Yale University Human Investigations Committee.

The SEER-Medicare database is a collaborative effort of the National Cancer Institute, the SEER registries, and the Centers for Medicare & Medicaid Services.^{25 - 26} Patient-level information includes sociodemographic characteristics, cancer type, site, and stage.²⁷ All incident cancer patients reported to the SEER registries are cross-matched with a master file of Medicare enrollment.²⁸ In addition to incident cancer patients, the SEER-Medicare database includes a 5% sample of Medicare beneficiaries without cancer who reside in SEER areas. Thus, the Medicare data enabled us to investigate endoscopic procedure use in the general (noncancer) population, to determine which patients were enrolled in Medicare HMO (vs fee for service), and to identify additional covariates that could be associated with early stage at diagnosis, such as burden of comorbidity or frequency of physician visits prior to cancer diagnosis.

The noncancer sample was restricted to patients who were 66 years of age or older, resided in the pre-2000 expansion SEER registries, and had fee-for-service and part B coverage for at least 12 consecutive months during the study period. For our primary analysis of trends in early stage cancer diagnosis, we identified a cohort of patients with a primary diagnosis of colon cancer during the years 1992-2002. We only included patients who were 67 years of age and older at the time of their cancer diagnosis and had fee-for-service or part B coverage for the entire 2 years prior to their cancer diagnosis, to ensure that these patients' claims were included in the SEER-Medicare files (Figure 1). Patients who were diagnosed with stage 0 (noninvasive) tumors or who did not have adenocarcinomas were also excluded from the study sample. Patients with missing race or stage data or whose data came from one of the Year 2000 expansion registries (Greater California, Kentucky, Louisiana, and New Jersey) were also excluded.

We created a categorical time variable to correspond with Medicare coverage policy changes. The categories included period 1 (no screening coverage, 1993-1997), period 2 (limited coverage, January 1998-June 2001), and period 3 (universal coverage, July 2001-December 2002). We identified patient sociodemographic and clinical factors that were potentially related to stage at diagnosis. For instance, race was classified as white, black, or other. Because prior work has suggested that the frequency of ambulatory care visits was related to stage of cancer diagnosis, we also estimated the total number of physician outpatient visits during the time period between 24 and 2 months prior to cancer diagnosis.²⁹ The total number of comorbid conditions that comprise the Charlson comorbidity index were identified by searching inpatient, outpatient, and physician claims during the 2-year interval prior to cancer diagnosis for each patient.³⁰ Median annual household income and percentage of the adult population who completed college or high school at the ZIP code level was used as a proxy for socioeconomic status. Stage was classified as early (stage I) vs all other (stages II-IV). Because tumors located at the splenic flexure or in a site more proximal are beyond the reach of a sigmoidoscope, all such lesions were classified as proximal.³¹

Colonoscopy claims were identified by searching outpatient and physician/supplier claims files, using Health Care Common Procedure Coding System codes G0105, G0121, 44388-44389,45378, 45380, and 45382-45385 and International Classification of Diseases, Ninth Revision, Clinical Modification codes 45.23, 45.25, 45.41-45.43, and 48.36; flexible sigmoidoscopy claims were identified with Health Care Common Procedure Coding System codes G0104, 45330-45331, 45333, and 45338-45339, and International Classification of Diseases, Ninth Revision, Clinical Modification codes 45.22, 45.24, 48.22, and 48.24, in a manner consistent with prior analysis.^{32 - 33} Duplicate procedures were removed if a patient had more than 1 claim for a procedure within the same calendar month. The results were not substantively altered when we also included procedure claims for hospitalized patients (from the Medicare Provider Analysis and Review file), and are therefore not reported.

For Medicare beneficiaries without colorectal cancer, colonoscopy and sigmoidoscopy rates were estimated for each calendar year quarter. During each quarter, the numerator was the number of patients with a claim consistent with a specific procedure, and the denominator was the number of eligible beneficiaries during that entire quarter. We used a Poisson model to determine whether endoscopy rates changed across time periods (1, 2, and 3) using procedure counts as the outcome variable and treating the number of eligible beneficiaries as the offset variable. In addition, we accounted for secular trends (calendar year quarter) and changes in reimbursement policy by including time period and calendar year quarter in the model.

In our analysis of cancer patients, we performed a bivariate analysis for each of the candidate covariates using the proportion of patients diagnosed at stage I (vs all other stages) as the outcome and χ² and t tests as appropriate. We then constructed a multiple logistic regression model in which time period was the independent variable and stage I was the outcome. Candidate covariates included comorbidity, age, race, socioeconomic status, site, sex, marital status, prior physician visits, and SEER registry. Variables that were significant in bivariate analyses were included in the initial model and retained if they remained significant. Because administrative claims data were only available for these patients beginning in 1991, we were only able to derive comorbidity and physician visit variables for patients diagnosed beginning in 1993. Hence, multivariate analysis incorporated patients diagnosed from 1993-2002. We used Zhang's³⁴ correction to convert the odds ratios to relative risks. To account for secular trend, we also included year as a linear variable (1993-2002) in preliminary models, but this was dropped because it was not a statistically significant term. We assessed the interaction between time period and tumor site (proximal vs distal colon). We calculated the adjusted percentage of patients diagnosed with early stage cancer according to tumor location and time period using the estimate option of PROC GENMOD in SAS 9.1 (SAS Institute Inc, Cary, NC) with covariate weights corresponding to the observed margins.³⁵ For our final analysis, we compared the proportion of patients diagnosed at an early stage in each time period among HMO and fee-for-service patients using χ² tests.

Among the sample of Medicare beneficiaries who did not have cancer, there was an increase in colonoscopy use during the study period (Figure 2). During period 1, the mean colonoscopy rate was 284.6/100 000 beneficiaries per quarter (95% confidence interval [CI], 255.2-313.9). This increased to a mean rate of 888.8/100 000 per quarter in period 2 (95% CI, 706.8-1070.7); crude incidence rate ratio vs period 1, 3.10 (95% CI, 3.01-3.20). The rate increased further in period 3 to 1918.9 (95% CI, 1720.9-2116.9; incidence rate ratio vs period 1, 6.54; 95% CI, 6.35-6.75). In the Poisson regression analysis, there was a 5% increase in colonoscopy rate with each calendar year quarter during the study period (incidence rate ratio, 1.05; 95% CI, 1.04-1.05). After adjusting for this temporal trend, there were additional increases in rate in periods 2 and 3 (incidence rate ratio vs period 1, 1.21; 95% CI, 1.14-1.27 and 1.59; 95% CI, 1.48-1.72), respectively.

The mean rate of flexible sigmoidoscopy in period 1 was 570.6 (95% CI, 551.2-589.9). In period 2, the rate increased to 691.9 (95% CI, 631.4-752.5), and then in period 3 it decreased to 267.5 (95% CI, 203.3-331.7). After adjusting for temporal trends, the sigmoidoscopy rate was 18% higher in period 2 than in period 1 (incidence rate ratio vs period 1, 1.18; 95% CI, 1.13-1.25), while in period 3, the rate was approximately 54% lower than in period 1 (incidence rate ratio, 0.46; 95% CI, 0.42-0.50).

The final sample of patients with colorectal cancer consisted of 44 924 patients with a mean age of 77.4 years (SD, 7.0); 56% were women and 8% were black (Figure 1). In 1992, 23% of patients were diagnosed with stage I. This proportion remained stable at approximately 23% through 1997, then increased to about 26% in the latter part of the study period (Table 1). Other factors that were associated with early stage at diagnosis included tumor site (20.1% of proximal vs 31.5% of distal lesions were diagnosed at stage I), sex, marital status, educational attainment in patient's ZIP code, and physician visits prior to cancer diagnosis (Table 1).

Time period was significantly related to stage at diagnosis. Approximately 22.5% of patients were diagnosed at an early stage in period 1 (1992-1997) compared with 25.5% in period 2 (P<.001) and 26.3% in period 3 (P vs period 1 <.001; P vs period 2 = .25). In the multivariate analysis, patients diagnosed in periods 2 and 3 were significantly more likely to have early stage illness than patients diagnosed in period 1 (adjusted relative risk [RR] vs period 2, 1.13; 95% CI, 1.09-1.17; vs period 3, 1.17; 95% CI, 1.12-1.23; Table 2). However, patients diagnosed in period 3 were not significantly more likely to have early stage at diagnosis than patients diagnosed in period 2 (RR, 1.04; 95% CI, 0.99-1.10).

Among the subgroup of patients who had proximal lesions, 18.1% of those who were diagnosed in period 1 were early stage, compared with 22.1% of such patients in period 2 and 24.1% in period 3 (P value for comparison vs period 1<.001 for both periods). The adjusted risk of early stage diagnosis in period 2 was 1.19 times the risk of early diagnosis in period 1 (95% CI, 1.13-1.26; Table 2) and in period 3, it was 1.1 times the risk of early diagnosis in period 2 (95% CI, 1.02-1.17). The adjusted proportion of patients diagnosed at stage I increased from 17.9% in period 1 to 21.5% in period 2 and 23.5% in period 3. P<.001 for period 2 vs period 1 and for period 3 vs period 1.

Among patients with distal lesions, there was a statistically significant increase in the probability of early stage at diagnosis from 30.5% in period 1 to 33.2% in period 2 (Table 1; P = .001 for pairwise comparison). However, the proportion of patients diagnosed at early stage in period 3 was not significantly different from period 1 or 2 (P values for pairwise comparisons, .13 and .49, respectively). Similarly, after adjusting for patient clinical and sociodemographic characteristics, the probability of early stage cancer for patients diagnosed during period 2 vs 1 was RR, 1.07; 95% CI, 1.01-1.13; Table 2), while the probability of being diagnosed at early stage was not significantly higher in period 3 than in period 2 (Table 2). The adjusted percentage of patients diagnosed with stage I disease was 31.0% in period 1, 32.8% in period 2, and 32.0% in period 3. P = .02 for period 2 vs period 1 and P = .30 for period 3 vs period 1.

Although the proportion of fee-for-service colorectal cancer patients who were diagnosed at an early stage increased with each successive time period, there was no significant change in early stage diagnosis among HMO beneficiaries (results not shown). During period 1, there was no significant difference in early stage diagnosis between fee-for-service (22.5%) and HMO (23.4%) patients (P = .12). During period 2, fee-for-service patients were significantly more likely to be diagnosed at an early stage than were HMO patients (25.5% vs 23.3%, respectively; P = .002), and similar findings were also noted in period 3 (26.3% vs 23.1%, respectively; P = .004).

We found that the proportion of older colon cancer patients who were diagnosed with early stage disease increased significantly after the initiation of new Medicare screening reimbursement policies. These findings were robust, in that they were unchanged even after accounting for important patient and geographic factors. However, the stage shift was primarily limited to proximal lesions, because patients with cancer of the distal colon were only slightly more likely to present with early stage disease after the new policy was enacted. The selective effect of the coverage change on proximal colon lesions suggests that whole-colon screening modalities such as colonoscopy may have played a pivotal role. This hypothesis is strongly supported by our finding that there was a significant increase in colonoscopy usage following each of the reimbursement changes, whereas sigmoidoscopy use actually decreased dramatically after the second policy change.

As Medicare continues to evolve into a purchaser with an increased level of scrutiny on the quality of care, it is important to determine whether policy changes have the intended effect on the health status of beneficiaries.³⁶ The benefit of legislative remedies that have been studied has been mixed. One recent study found that state-mandated coverage for in vitro fertilization was associated with increased use of these services, although the success rates were lower.³⁷ Another study reported that a state mammography reimbursement mandate did not yield a substantial increase in mammography use.³⁸ Our findings are consistent with a prior analysis of the impact of a Medicare mammography reimbursement policy. That analysis suggested that the policy may have been associated with increased use, but overall use remained low even after the new benefit.³⁹ When our findings of a shift toward earlier stage at diagnosis are considered in the context of other work showing persistent underuse of colorectal cancer screening, it is clear that there is still significant work to be done. Increasing the use of screening tests further has the potential to diagnose many more beneficiaries at an early stage.

It should also be noted that these data demonstrated a temporal relation between the Medicare policy change and earlier stage at diagnosis, which may not be causal in nature. However, it is unlikely that these changes in stage at diagnosis can be attributed only to temporal trends in patterns of care for several reasons. First, the changes in both colonoscopy use and stage at diagnosis occurred in a relatively short time frame. Second, our finding that a meaningful shift toward earlier stage at diagnosis was only noted in patients with proximal lesions implicates colonoscopic screening as a mediator, because this modality would be expected to preferentially impact the detection of right-sided lesions in a population with relatively stable sigmoidoscopy rates. Third, we identified a contemporaneous increase in colonoscopy usage in the noncancer Medicare population, in contrast to a decrease in flexible sigmoidoscopy use.

We also found that the policy changes did not affect HMO patients' probability of early stage diagnosis, in contrast to their favorable effect on fee-for-service patients. This suggests that the reimbursement change, which was directed toward fee-for-service beneficiaries, was directly associated with the change rather than national secular trends in clinical care. There is no reason to believe that physicians caring for HMO patients are immune from secular trends in clinical care. Our finding that fee-for-service beneficiaries were actually more likely to be diagnosed at an early stage than HMO beneficiaries in the latter time period therefore represents a reversal of earlier findings and reinforces the importance of ensuring reimbursement for effective screening modalities.^{22 ,24}

One consideration in the assessment of colonoscopy use is that it is difficult to differentiate reliably between screening procedures and those being done to evaluate specific symptoms. However, the colonoscopy rate increased nearly 5-fold during the study period. It is unlikely that the actual symptomatic presentations of Medicare beneficiaries, or approach to evaluating specific gastrointestinal complaints, changed so dramatically during the study period. Hence, the marked increase in colonoscopy use likely reflects screening rather than diagnostic colonoscopies.

Our findings are also consistent with prior studies. A national survey found that older persons were more likely to report ever having had a screening colonoscopy in 2003 than in 2000.⁴⁰ Similarly, a single-institution study found that more colonoscopies were being performed for screening after the 2001 reimbursement policy change.⁴¹

It is important to emphasize that colorectal cancer screening can yield clinical benefit in 2 ways, by reducing the incidence of disease (by removing precancerous lesions before they become invasive adenocarcinoma), and by detecting invasive cancer at an earlier stage. Prior work has demonstrated a decrease in incidence of colorectal cancer between the mid-1980s and the mid-1990s.⁴² SEER data demonstrate that the incidence rate of distal colon cancer has declined 30% from 1982 to 2002, while proximal colon cancer incidence has only decreased 10% over the same time period, suggesting that the decline in colon cancer incidence is associated with the use of flexible sigmoidoscopy.^{21 ,43} Because the adenoma to carcinoma sequence may take as long as 10 years, we hypothesized that any increase in colon cancer screening among the Medicare population would be associated with a shift in stage distribution long before any reduction in the cancer incidence rate was noted. Future work should explore the impact of enhanced endoscopy coverage and utilization on rate of polypectomy and, eventually, cancer incidence.

Our finding that new Medicare policies may have facilitated early diagnosis is encouraging and supports the institution and evaluation of other efforts to broaden the access to and use of screening tests in the older population. Given that there are approximately 60 000 cases of colorectal cancer diagnosed annually among patients 65 years of age and older in the United States, even a 4% increase in the percentage of patients whose cancer is diagnosed at an early stage can have a substantial impact at the population level.⁴⁴ However, the impact was modest, and is consistent with prior work showing overall underutilization of colon cancer screening.^{18 - 19} The majority of patients with colon cancer are still being diagnosed with later-stage disease.^{6 ,10 ,15} Finally, not all Medicare beneficiaries should undergo colorectal cancer screening. There is uncertainty about the upper age limit, or lower boundary for life expectancy, for which screening should be offered.^{1 ,45} Efforts to enhance use of screening should incorporate a thoughtful approach to patient preferences as well as to factors that can alter the balance of risks and benefits in older patients.