The real world: CLL

In this issue of Blood, Danese and colleagues present a large database analysis of patterns of care and outcomes with respect to chemoimmunotherapy for older patients with chronic lymphocytic leukemia (CLL).¹  Their analysis demonstrates some of the important benefits and limitations of health services research as it relates to hematology.

The purpose of a large retrospective data analysis such as the one performed by Danese and colleagues is not to supplant data from controlled clinical trials, but to provide information about the effects of treatment interventions in the “real world” populations in which such treatments are actually used. For CLL, this likely means an older population with more extensive comorbidity than subjects enrolled in clinical trials. Indeed, the median age of patients in the German CLL Study Group's CLL8 study—which found that rituximab plus chemotherapy improved overall survival compared with chemotherapy alone—was 61 years, with ∼ 57% of the patients enrolled having an Eastern Cooperative Oncology Group (ECOG) performance status of 0.² 

At 72 years,³  the median age of ordinary CLL patients is significantly older than the population enrolled in most investigational trials, and such real-world patients have, on average, poorer performance status and more comorbid conditions. Data beyond existing clinical trials are clearly needed to determine the comparative effectiveness of treatment strategies for such patients. One possible solution would be another trial specifically enrolling older and sicker patients; however, from a practical standpoint, such studies can be very difficult to fund and to conduct. Fortunately, when performed and interpreted with sophistication, health services analyses can provide an important additional source of data.

Danese and colleagues used the publically available SEER-Medicare database, which is composed of the National Cancer Institute's (NCI) Surveillance, Epidemiology, and End Results (SEER) cancer registry linked to Medicare enrollment and claims data. This dataset offers a large population-based cohort that is extremely useful for longitudinally tracking patients 65 years and older over the course of a cancer diagnosis, including procedures, treatments, and follow-up. Although there are some differences between those residing in the SEER catchment areas and the general United States (eg, the SEER population is more urban and has a higher proportion of foreign-born), the population covered includes a rich race-ethnic mix and is comparable to the general US population with respect to education and prevalence of poverty.⁴  As a result, SEER-Medicare has long been a standard tool for health services researchers studying patterns of care and outcomes for solid tumors in the elderly.⁵ 

Perhaps reflecting the rare incidence of hematologic malignancies compared with other cancers, and the resulting concern about the quality of SEER and SEER-Medicare data as it relates to associated diagnoses and treatments, Danese et al's report is the first SEER-Medicare study to be published in Blood. The authors report the appropriately guarded conclusion that use of rituximab plus chemotherapy likely improves survival in older patients with CLL compared with treatment with chemotherapy alone (see table). Given the retrospective nature of the SEER-Medicare data, and the fact that there is no way to determine with certainty if it was the combination of rituximab with chemotherapy or patient characteristics associated with being eligible for such treatment that was responsible for the improved survival compared with patients who did not receive chemoimmunotherapy, the authors analyzed the survival data in several different ways. First, they fit a traditional adjusted, multivariable survival model that allowed them to explore independent predictors of overall mortality. Next, they analyzed their data with propensity score analysis, a technique that allows for more fine control of selection bias through use of a propensity score, but can be more limited in that it focuses more specifically on the effects of treatment. Finally, the authors performed several sensitivity analyses, restricting their main analyses to those patients aged 70 years or older (no change in effect) and to those with evidence of more advanced disease (the benefit of adding rituximab was lost).

As with all large retrospective database analyses, the study has important limitations. First, SEER-Medicare provides no data with respect to patient experiences, provider documentation, or test results, and the authors thus had to rely entirely on claims data, which can be incomplete. Moreover, treatment side effects are not measured in SEER-Medicare, but must be inferred from claims.⁶  Second, SEER and SEER-Medicare are only able to capture CLL diagnoses that are reported to member registries, and diagnosis made in a private physician's office may be less likely to be reported. This issue has been a possible source of under-ascertainment for CLL in other large registries,⁷  reflecting what might ironically represent these registries' own lack of real-world applicability.

On the other hand, one benefit of such large database analyses is that in addition to providing insight regarding the generalizability of clinical trial data, they can also help policymakers understand current patterns of care. For example, the authors found that the use of rituximab for CLL treatment increased from ∼ 11% of patients in 2000 to 43% in 2005—a finding with obvious implications for health care utilization and funding. The authors also found that the likelihood of initial infused therapy for CLL was decreased in certain race-ethnic groups and in females. The first finding is of interest given a recent report of lower treatment rates for blacks versus whites for diffuse large B-cell lymphoma,⁸  and suggests that race-ethnic disparities observed in the treatment of patients with solid tumors⁹  may exist for hematologic malignancies as well. In addition, although there are gender differences in the incidence and outcomes of CLL,¹⁰  it is unclear why, when matched for other covariates, women would be less likely to be treated with infused therapy. This finding is especially interesting given recent data demonstrating that women with CLL are likely to experience longer diagnostic delays.¹¹ 

A final and significant feature of Danese and colleagues' work is the measured, careful way in which the authors report their results, even more important given that 2 of them are affiliated with the company that makes rituximab. Just as data from clinical trials imperfectly inform the best treatment strategy for an individual patient, data from heath services analyses must also be interpreted with caution. In this case, beyond the limitations of the SEER-Medicare dataset itself, the authors appropriately underscore the possibility of biases because of unobserved differences in patients who received different treatments, even after attempts to control for these differences. Despite these limitations, in a time when a drive toward truly understanding real-world comparative effectiveness has captured the nation and its policymakers, it is exciting to see such an analysis published in Blood. There will surely be more to come.