For patients with advanced cancer, the median overall survival following case follow-up start date for lung cancer was 5 months; for colorectal cancer, 8 months; for pancreatic cancer, 4 months; for gastroesophageal cancer, 4 months; and for breast cancer, 16 months. The overall survival at 5 years for each cancer was 3%, 5%, 3%, 2%, and 16%, respectively. Each cancer-free control was matched to a case with the specific cancer diagnosis by age, sex, race/ethnicity, and Surveillance, Epidemiology, and End Results tumor registry. For the cancer-free controls, the median overall survival was not reached and the overall survival at 5 years was 82% for lung cancer, 80% for colorectal cancer, 81% for pancreatic cancer, 81% for gastroesophageal cancer, and 85% for breast cancer.
PSA, prostate-specific antigen; GI, gastrointestinal. P values were based on the Gray test (applicable to cumulative incidence curves). For each of the 4 screening tests, P < .001.
Sima CS, Panageas KS, Schrag D. Cancer Screening Among Patients With Advanced Cancer. JAMA. 2010;304(14):1584-1591. doi:10.1001/jama.2010.1449
Author Affiliations: Department of Epidemiology and Biostatistics, Memorial Sloan-Kettering Cancer Center, New York, New York (Drs Sima and Panageas); and Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts (Dr Schrag).
Context Cancer screening has been integrated into routine primary care but does not benefit patients with limited life expectancy.
Objective To evaluate the extent to which patients with advanced cancer continue to be screened for new cancers.
Design, Setting, and Participants Utilization of cancer screening procedures (mammography, Papanicolaou test, prostate-specific antigen [PSA], and lower gastrointestinal [GI] endoscopy) was assessed in 87 736 fee-for-service Medicare enrollees aged 65 years or older diagnosed with advanced lung, colorectal, pancreatic, gastroesophageal, or breast cancer between 1998 and 2005, and reported to one of the Surveillance, Epidemiology, and End Results (SEER) tumor registries. Participants were followed up until death or December 31, 2007, whichever came first. A group of 87 307 Medicare enrollees without cancer were individually matched by age, sex, race, and SEER registry to patients with cancer and observed over the same period to evaluate screening rates in context. Demographic and clinical characteristics associated with screening were also investigated.
Main Outcome Measure For each cancer screening test, utilization rates were defined as the percentage of patients who were screened following the diagnosis of an incurable cancer.
Results Among women following advanced cancer diagnosis compared with controls, at least 1 screening mammogram was received by 8.9% (95% confidence interval [CI], 8.6%-9.1%) vs 22.0% (95% CI, 21.7%-22.5%); Papanicolaou test screening was received by 5.8% (95% CI, 5.6%-6.1%) vs 12.5% (95% CI, 12.2%-12.8%). Among men following advanced cancer diagnosis compared with controls, PSA test was received by 15.0% (95% CI, 14.7%-15.3%) vs 27.2% (95% CI, 26.8%-27.6%). For all patients following advanced diagnosis compared with controls, lower GI endoscopy was received by 1.7% (95% CI, 1.6%-1.8%) vs 4.7% (95% CI, 4.6%-4.9%). Screening was more frequent among patients with a recent history of screening (16.2% [95% CI, 15.4%-16.9%] of these patients had mammography, 14.7% [95% CI, 13.7%-15.6%] had a Papanicolaou test, 23.3% [95% CI, 22.6%-24.0%] had a PSA test, and 6.1% [95% CI, 5.2%-7.0%] had lower GI endoscopy).
Conclusion A sizeable proportion of patients with advanced cancer continue to undergo cancer screening tests that do not have a meaningful likelihood of providing benefit.
Cancer screening programs evaluate asymptomatic patients for the detection of early forms of cancer and have contributed substantially to the decline in cancer mortality.1,2 The premise of these programs is that detection of an early-stage tumor can reduce a person's risk of dying from a screen-detectable disease. Utilization of routine screening procedures such as mammography, Papanicolaou test, prostate-specific antigen (PSA) and colonoscopy, and flexible sigmoidoscopy (lower gastrointestinal [GI] endoscopy) has become deeply ingrained in medical culture. Primary care practices have developed tracking and reminder systems to ensure that patients are adherent to screening guidelines. Programs to monitor and improve the quality of health care delivery, such as the Centers for Medicare & Medicaid Services Physician Quality Reporting Initiative,3 feature adherence to cancer screening. In addition, during the past decade a large body of literature has focused on underutilization of cancer screening and efforts to increase adherence have been the subject of numerous research grants and public awareness campaigns.4- 8
Although the benefits of cancer screening are compelling for most members of the population, its value is less certain for patients with concurrent illnesses that severely limit life expectancy. In the extreme situation of patients with advanced cancer, screening will lead to overdiagnosis (detection of a cancer which, if not found by active search, would not affect survival) in virtually all cases when a new malignancy is found. In addition, patients may be subject to unnecessary risk due to subsequent testing, biopsies, and psychological distress. Current guidelines9,10 acknowledge the potential for overuse and the lack of well established benefit for elderly patients for whom natural life expectancy limits the benefit of screening; however, they do not directly address the appropriateness of screening for individuals with terminal illness.
We anticipated that a small proportion of patients diagnosed with advanced cancer continue to be screened for cancers other than their primary tumor, despite the fact that they have essentially no chance of benefiting from these procedures. We determined the extent to which patients are tested in this very specific circumstance and identified characteristics associated with unnecessary screening. Our goal was to identify a potential opportunity to simultaneously improve care and eliminate wasteful spending in the Medicare program.11
Data for our study was obtained from the Surveillance, Epidemiology, and End Results (SEER) cancer registry linked with Medicare claims. SEER is a consortium of 16 population-based cancer registries that collect data on incident cancer cases, with information on the date of diagnosis, site and extent of disease, sociodemographic characteristics (including race/ethnicity, which was provided by the Enrollment Database File maintained by the Centers for Medicare & Medicaid Services), and vital status.12,13 Medicare is the primary health insurer for US individuals aged 65 years or older, its coverage including but not limited to inpatient hospital services (Part A), physician services, and outpatient care (Part B). SEER-Medicare linkage allows for claims-based identification of medical care received by patients with cancer aged 65 years or older.
The institutional review board at Memorial Sloan-Kettering Cancer Center reviewed and approved the research team's procedures for using SEER-Medicare data. The National Cancer Institute reviews all manuscripts that use SEER-Medicare data to protect the confidentiality of patients and professionals represented in these data. The manuscript was approved on April 23, 2010.
We used SEER data to identify all patients diagnosed between 1998 and 2005 with the following cancers: stage IIIB-IV lung, stage IV colorectal, breast, and gastroesophageal, and advanced stage pancreatic. Patients with lung, colorectal, and breast cancers were classified according to the American Joint Committee on Cancer Staging14; SEER historic stage algorithm15 was used for pancreatic and gastroesophageal cancers. We selected these tumors because they are common and have 5-year overall survival rates of less than 20%, which are associated with very low likelihood of benefiting from early detection.
We restricted our cohorts to patients who were 65 years or older at the time of diagnosis, had ante-mortem diagnosis, known diagnosis month, and survived at least 2 months following diagnosis with advanced cancer. To ensure complete capture of claims, patients were continuously enrolled in Medicare Parts A and B, but not in a health maintenance organization between diagnosis and death or the censoring date of December 31, 2007.
We identified utilization of the following screening tests: (1) mammography among women in the lung, colorectal, pancreatic, and gastroesophageal cohorts; (2) Papanicolaou test among women in all cohorts; (3) PSA testing among men in all cohorts; (4) lower GI endoscopy testing (colonoscopy and flexible sigmoidoscopy) among men and women in the lung, breast, pancreatic, and gastroesophageal cohorts; and (5) cholesterol testing among men and women in all cohorts. Screening tests were identified using the International Classification of Disease, Ninth Revision diagnosis and procedure codes16 and Healthcare Common Procedure Coding System codes17 that health care practitioners must submit to Medicare in order to obtain reimbursement.
Cancer screening procedures are assigned distinct codes for screening and for diagnosis. We restricted our analysis to codes that specifically identify screening tests (eTable). Furthermore, we minimized the potential for mislabeling procedures ordered as part of establishing a cancer diagnosis and staging workup by excluding claims billed during the first 2 calendar months subsequent to cancer diagnosis. Consequently, screening rates for patients with cancer reflect the tests performed between the third calendar month following cancer diagnosis (case follow-up start date) and death or December 31, 2007, whichever came first (case follow-up end date).
To understand the rates of cancer screening for patients with advanced cancer in context, we compared them with the rates observed in a matched cohort of fee-for-service Medicare enrollees without cancer, identified from the random 5% sample of Medicare beneficiaries residing in the SEER areas. Each case was individually matched to a cancer-free control by year of birth, sex, race, and SEER registry. Similar to the cancer cases, controls were required to be alive and continuously enrolled in Medicare Part A and B, but not in a health maintenance organization between the date of the corresponding case diagnosis and case follow-up end date.
Each control was followed up during the same calendar months as his or her matched patient with cancer to evaluate the utilization of cancer screening tests. Each control had exactly the same amount of follow-up time as his or her corresponding cancer case, allowing for a meaningful comparison of screening rates for patients with and without advanced cancer.
For each test, the rate of cancer screening among both patients with cancer and cancer-free controls was calculated as the percentage of patients who had at least 1 screening procedure in the interval between case follow-up start date and case follow-up end date. Overall survival was calculated using Kaplan-Meier method, with time origin at case follow-up start date. Time from case follow-up start date to the receipt of screening was examined using the cumulative incidence function.
To identify patient and disease characteristics associated with continued screening, we calculated screening rates within subgroups categorized by cancer site, screening history (for patients with cancer only), age at cancer diagnosis (65-69 years, 70-74 years, 75-79 years, and ≥80 years), race/ethnicity (non-Hispanic white, black, Hispanic, Asian, and other [North American Native, other, and unknown]), marital status (married, not married—information available for cancer cases only), year of diagnosis (1998-1999, 2000-2001, 2002-2003, and 2004-2005), and income (quartiles of median household income at the zip code level). Screening history was defined as utilization of the screening test during the 1-year interval between 24 and 12 months before advanced cancer diagnosis. Patients with cancer who were younger than 65 years (therefore not yet enrolled in Medicare) at the beginning of this interval were excluded from the evaluation of screening history. Race/ethnicity was assessed in light of the extensive literature indicating the presence of racial disparities in health care delivery in general18,19 and in cancer screening in particular.20 For patients with cancer only, screening rates across levels of these variables were formally compared by fitting 5 multivariable competing risks models (for each cancer screening test and for cholesterol screening).
Statistical analysis was conducted using SAS version 9.2 (SAS Institute Inc, Cary, North Carolina) and the cmprsk package in R (http://www.r-project.org/). All significance tests were 2-sided and used a 5% level of significance.
Baseline characteristics of the 87 736 patients with advanced cancer included in our analysis are shown in Table 1. Sixty-one percent of the patients had stage IIIB-IV lung cancer, 14% had stage IV colorectal cancer, 14% had advanced stage pancreatic cancer, and the remaining patients were diagnosed with advanced gastroesophageal cancer or stage IV breast cancer. Median age at diagnosis was 75 years (interquartile range, 70-80 years) and did not vary remarkably by cancer site; 81.4% (95% confidence interval [CI], 81.1%-81.7%) of the patients were white and 52.4% (95% CI, 52.0%-52.7%) were married. By study design, demographic characteristics of the controls mirrored those of the cancer cases.
Among patients evaluated for each test, 26.4% (95% CI, 25.9%-26.9%) had a history of mammography, 13.2% (95% CI, 12.9%-13.6%) had a history of Papanicolaou testing, 37.1% (95% CI, 36.6%-37.5%) had a history of PSA testing, and 4.0% (95% CI, 3.9%-4.2%) had a history of lower GI endoscopy testing.
Median survival varied between 4.3 months (95% CI, 4.2-4.5 months) for the pancreatic cancer cohort and 16.2 months (95% CI, 15.2-17.5 months) for the breast cancer cohort. Overall survival at 5 years was 15.5% (95% CI, 14.3%-16.9%) for the breast cancer cohort and 5% or less for the remaining cohorts (Figure 1). Between 80% and 85% of the matched cancer-free controls were alive 5 years after the corresponding case follow-up start date. Median survival was not reached in this group during the follow-up period of the study (Figure 1).
Among women with advanced cancer, 8.9% (95% CI, 8.6%-9.1%) received at least 1 screening mammography and 5.8% (95% CI, 5.6%-6.1%) received at least 1 Papanicolaou test compared with 22.0% (95% CI, 21.7%-22.5%) and 12.5% (95% CI, 12.2%-12.8%) of controls, respectively. Rates were higher in the group of women who had a history of the same screening test (16.2% [95% CI, 15.4%-16.9%] of these women had mammography; 14.7% [95% CI, 13.7%-15.6%] had a Papanicolaou test) and among women aged 65 to 69 years (12.9% [95% CI, 12.2%-13.7%] had mammography; 9.4% [95% CI, 8.9%-10.0%] had a Papanicolaou test).
Among men with advanced cancer, 15.0% (95% CI, 14.7%-15.3%) received PSA testing compared with 27.2% (95% CI, 26.8%-27.6%) of controls. Among men with a PSA screening history, 23.3% (95% CI, 22.6%-24.0%) were also screened postdiagnosis.
Lower GI endoscopy was performed in 1.7% (95% CI, 1.6%-1.8%) of all patients with cancer and in 6.1% (95% CI, 5.2%-7.0%) of patients with a precancer screening history. In comparison, 4.7% (95% CI, 4.6%-4.9%) of controls received lower GI endoscopy screening. Cholesterol was tested in 19.5% (95% CI, 19.3%-19.8%) of patients with advanced cancer and in 37.4% (95% CI, 37.0%-37.7%) of controls (Table 2).
Half of the patients who received mammographies, Papanicolaou tests, or PSA tests did so within 10 months of the cancer diagnosis, and half of the patients with a lower GI endoscopy screening received it within 18 months of diagnosis. Most patients screened had testing within 36 months of diagnosis (Figure 2).
Higher socioeconomic status and married status were significantly associated with a higher probability of screening for each test evaluated (Table 2). For example, 9.8% (95% CI, 9.2%-10.4%) of women in the highest income quartile received mammography compared with 7.4% (95% CI, 6.9%-8.0%) in the lowest quartile (P for trend < .001), and 8.0% (95% CI, 7.6%-8.4%) of married women received Papanicolaou test screening compared with 4.6% (95% CI, 4.4%-4.9%) of nonmarried women (P < .001). Screening rates for all tests except PSA showed an inverse association with age; the rate of PSA testing peaked in the 70- to 74-year-old category. Lower GI endoscopy and cholesterol tests, the only tests applicable to both sexes, were more likely to be performed in women (lower GI endoscopy, 2.2%; 95% CI, 2.0%-2.3%) vs in men (1.3%; 95% CI, 1.2%-1.4%; P < .001). White patients tended to have higher screening rates compared with nonwhite patients; however, in multivariable models, these differences reached statistical significance only for mammography.
We evaluated the utilization of common cancer screening tests by fee-for-service Medicare beneficiaries diagnosed with advanced cancers associated with median survival of less than 2 years. Notwithstanding their limited life expectancy, a meaningful proportion of patients with advanced cancer continue to undergo routine cancer screening. Specifically, 9% of women with advanced cancer received screening mammography and 6% received Papanicolaou testing. Among men, 15% underwent PSA testing. Lower GI endoscopy screening was obtained by only 2% of patients. It is especially striking that these rates represented 35% to 55% of the rates observed in matched cancer-free controls followed up over the exact same period. Consistent with screening guidelines, screening rates declined precipitously among persons older than 75 years. Our analysis was limited to the population aged 65 years or older, and the rates we report are likely to be higher among younger patients who are commercially insured.
The strongest predictor of screening in the setting of advanced cancer was the receipt of a screening test before diagnosis. The most plausible interpretation of our data is that efforts to foster adherence to screening have led to deeply ingrained habits. Patients and their health care practitioners accustomed to obtaining screening tests at regular intervals continue to do so even when the benefits have been rendered futile in the face of competing risk from advanced cancer. Other investigators have identified examples of the culture of screening on “autopilot.” The study by Sirovich and Welch21 identified continued utilization of Papanicolaou tests among women lacking a cervix due to prior hysterectomy. Furthermore, we hypothesize that neither primary care physicians nor oncologists routinely engage in the difficult discussions that require explanation of why continuation of procedures to which patients have become accustomed to is no longer necessary.22 There is substantial evidence that even when physicians recognize that life expectancy is limited, they do not consistently communicate prognosis,23 and patients may use denial as a coping strategy to face impending loss.24 Our findings represent one manifestation of this communication deficit.
Our analysis should be considered in the context of the limitations of our data set. We cannot determine for any individual whether a particular test was inappropriate. Women with advanced breast cancer may sometimes live a number of years, making it somewhat more sensible for them to undergo screening. To eliminate tests that could have been obtained for diagnosis, we limited our analyses to codes designated for screening. Nevertheless, it is possible that some claims for diagnostic tests in fact represent screening. In addition, we excluded screening claims billed within the first 2 calendar months following diagnosis. It is possible that a small proportion of Medicare enrollees with additional private coverage had screening tests that were not captured by Medicare claims. Overall, our results are conservative estimates of screening test use by patients with advanced cancer. A key limitation of the available data is that we cannot determine whether screening is driven by oncologists, primary care physicians, health care settings, or patients themselves. An evaluation of the ordering health care practitioners on specific claims would not distinguish between patient and physician driven test use.
Whichever the impetus, screening utilization by patients with advanced cancer adds to the mounting concern about overdiagnosis. Other studies have emphasized the potential harms associated with such screening.25- 28 Studies estimate that 1 in 3 breast and prostate cancers detected in a population offered organized screening are overdiagnosed.29,30 The study by Walter and Covinsky31 estimated that a woman aged 70 years or older with limited life expectancy due to comorbidities has less than 2% risk of dying from breast cancer, less than 1% risk of dying from cervical cancer, and less than 0.1% risk of dying from colorectal cancer. The authors concluded that patients with life expectancy of less than 5 years are unlikely to derive any benefit from cancer screening. In an ideal health care system, health care practitioners would discontinue cancer screening for patients whose prognosis is too limited for the benefits of early detection to be realized.
It is conceivable that some of the screening tests that we observed were scheduled before cancer diagnosis, therefore reflecting the inertia of a system that does not have into place mechanisms to recognize when an otherwise routine test becomes unnecessary. An efficient health care system should have the capacity for learning and adaptation as indications for interventions evolve. Ideally, interoperable electronic health records should be able to flag patients with advanced cancer for reconsideration of the risk/benefit ratio of interventions such as screening.
Currently, in the United States, the focus is on controlling the skyrocketing costs of the health care system in general and of Medicare in particular. One strategy to limit costs in the Medicare program is to intensively investigate fraud, a recent focus of the Obama administration.32 An alternative strategy is to systematically identify unnecessary care that does not provide meaningful benefits. Identification of wasteful care is challenging because each patient's circumstances are unique, and it is difficult to reliably define episodes of overuse. Moreover, efforts to curb unnecessary care and thereby costs of Medicare prompt physician concerns about interference in the practice of medicine and patient fear about “rationing.”33- 36 We have identified a very specific circumstance in which the case for wasteful care is as clear and compelling as the unnecessary use of Papanicolaou tests described by Sirovich and Welch.21
Our results have several policy implications. First, greater awareness that screening in the face of limited life expectancy from advanced cancer is of dubious benefit may in and of itself limit use. Second, as electronic medical records and reminder systems are developed to foster screening adherence, they should also include program features that flag when conditions suggest reevaluation or cessation of screening based on competing comorbidities. Electronic medical records increasingly have the sophistication to track cancer stage at diagnosis and disease status and to link this to screening reminder systems. Alternatively, the Medicare program might not provide coverage for cancer screening procedures for patients with life expectancy of less than 2 years.
Although any attempt to place restrictions on care that patients can receive is routinely met with vocal opposition by physicians and patients alike, overuse of screening for patients with advanced cancer is likely to be relatively uncontroversial. Curbing cancer screening for patients with advanced cancer would have a small impact on Medicare as a whole given that it is performed for only a minority of all Medicare patients with cancer. However, iteration of this paradigm across other diseases and conditions could systematically improve the value of each Medicare dollar spent. Each medical specialty needs to engage in thoughtful self-scrutiny to identify episodes of unnecessary care. We suggest that the road to a high-performing, high-value health care system will be paved with small stones such as the example we have identified. Such a systematic approach to identification of wasteful care need not compromise the quality of care patients receive. Indeed, identification of episodes of unnecessary care and strategies to curb them has the potential to be a win-win for patients, health care practitioners, and the public.
Corresponding Author: Camelia S. Sima, MD, MS, Department of Epidemiology and Biostatistics, Memorial Sloan-Kettering Cancer Center, 307 E 63rd St, Second Floor, New York, NY 10065 (email@example.com).
Author Contributions: Dr Sima had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Sima, Panageas, Schrag.
Acquisition of data: Sima, Schrag.
Analysis and interpretation of data: Sima, Panageas, Schrag.
Drafting of the manuscript: Sima, Panageas, Schrag.
Critical revision of the manuscript for important intellectual content: Sima, Panageas, Schrag.
Statistical analysis: Sima, Panageas.
Obtained funding: Schrag.
Study supervision: Sima, Schrag.
Financial Disclosures: None reported.
Funding/Support: This research was supported by grant R01 1147503 from the National Cancer Institute (Dr Schrag).
Role of the Sponsor: The funding organization had no role in the design and conduct of the study, in the the collection, management, analysis, and interpretation of the data, or in the preparation, review, or approval of the manuscript.
Disclaimer: This study used the linked Surveillance, Epidemiology, and End Results (SEER)-Medicare database. The interpretation and reporting of these data are the sole responsibility of the authors.
Additional Contributions: We acknowledge the efforts of the Applied Research Program, National Cancer Institute; the Office of Research, Development, and Information, Centers for Medicare & Medicaid Services; Information Management Services; and the SEER Program tumor registries in the creation of the SEER-Medicare database.