In this study, we assessed the ability of Medicaid claims data to identify incident cases of breast cancer in the Medicaid population, using OCISS as the gold standard. This study highlights two important findings: First, the sensitivity was significantly lower among women who were enrolled for only part of the study year, potentially because of incomplete claims history. Although other factors, such as age, and dual eligibility for the Medicare and Medicaid programs, are associated with lower sensitivity, we found through additional analysis that sensitivity was higher among women who had been enrolled in Medicaid for all 12 months of the study year than those who had been participating in Medicaid for only part of the year, even among dually eligible Medicare–Medicaid individuals. Relative to women not participating in Medicare, for example, the sensitivity was 74.2 percent among those with partial-year enrollment in Medicaid, and 83.3 percent among those with full-year enrollment in Medicaid—a rate comparable to that reported for 1992 Medicare claims (Cooper et al. 1999
; Warren et al. 1999
). Similarly, among dually eligible women, sensitivity was 43.6 percent among partial-year enrollees and 75.6 percent among those enrolled in Medicaid for 12 months of the study year (p
<0.001 for above comparisons). That sensitivity remained lower among dually eligible Medicare–Medicaid enrollees than among others, even if they were continuously enrolled in Medicaid for 12 months of the study year, remains to be explored. Like other claims, crossover claims are believed to carry diagnostic and procedure codes, and to provide an account of the services received by the beneficiary. It is possible, however, that similar to those enrolled for only part of the study year, Medicare-eligible women with low incomes would be likely to enroll in Medicaid after they are diagnosed with and treated for their cancer, just as they transition to the phase where treatment is rendered in outpatient settings and out-of-pocket expenditures for Part B services begin to constitute financial burden. Such issues would be best explored through a linked OCISS and Medicare–Medicaid database.
Second, the PPV, which reflects the ability of claims to identify true positive cases, varied greatly, depending on the combination of diagnosis and procedures codes used to identify incident cases. The PPV was highest in groups where cases were identified using a combination of procedures for mastectomy, lumpectomy with chemotherapy/radiation therapy, in addition to breast cancer diagnosis—combinations of diagnostic and procedure codes that reflect the typical experience of a woman undergoing diagnosis and treatment of an incident case of breast cancer. These findings indicate that the presence of procedure codes that are specific to the treatment of breast cancer (mastectomy and/or lumpectomy with chemotherapy/radiation therapy), in addition to that of breast cancer diagnosis, greatly contributes to the correct identification of incident cases of breast cancer using claims data. Low PPV implies a high rate of false positives. For a large number of cases, claims data may have identified prevalent, rather than incident cases of breast cancer, especially (a) where breast cancer diagnosis was documented in the absence of any procedures specific to the disease, and (b) where breast cancer diagnosis was present along with codes for chemotherapy/radiation therapy, without the documentation of any surgical codes. In instances where breast cancer diagnosis was present with a lumpectomy code, but with no documentation for chemotherapy/radiation therapy, it was not possible to distinguish between cases that were coded as breast cancer in the presence of the disease, or to rule out the disease. Similarly, cases where only breast cancer diagnosis was documented in the absence of relevant procedure or treatment codes could well have been diagnostic rule-out cases, with the receipt of a screening or diagnostic mammography exam. In fact, nearly 30 percent of such cases presented with a mammography procedure code.
This study also enabled us to assess the utility of pharmacy data in identifying prevalent cases of breast cancer. Of the 3,052 prevalent cases identified across the two-year period through claims data, 250 (or 8.2 percent) had been identified through pharmacy claims only, based on the fact that they had been prescribed tamoxifen prior to the claims-based initial date of breast cancer diagnosis. Of these, 26 cases matched successfully with OCISS records. Given their prevalent status from the claims data, these cases were not accounted for in the sensitivity rates. Because of their small number, the inclusion of these cases in the numerator to measure the claims' sensitivity would not have resulted in a large increase in the rates. However, the decision of whether to account for these cases as prevalent based on pharmacy claims alone could be questioned, given that tamoxifen was being prescribed as a prophylactic therapy in recent years to prevent primary cancers among high-risk women or new primary cancers in the contralateral breast among women with history of breast cancer (Vogel 2000
The finding that sensitivity was somewhat higher among women with spend-down deserves to be explored further. While no study has documented that individuals participating in the spend-down program have incomplete claims history, that is believed to be the case, due to the fact that part of the services are covered by the patient through out-of-pocket payments. In this study, 73 percent of patients on spend-down were 65 years of age or older, and 89 percent participated in the Medicare program. Both factors are known to be associated with lower sensitivity rates. On the other hand, 61 percent of women on spend-down and 53 percent of their nonspend-down counterparts were enrolled full time in the Medicaid program during the study year (p<0.01), a factor favoring higher sensitivity rates. More studies are needed to gain a better understanding of the subgroups of the Medicaid population that are more likely than others to be represented in the claims data with incomplete history.
An important consideration in this study is that, despite its Silver Certification by the NAACCR, OCISS is not a reporting source that is as well established as SEER. Given the fact that this study is limited to the Ohio Medicaid program and the state's system of monitoring cancer incidence, some of the findings may not be generalizable to other states. Nevertheless, it is highly likely that the findings of this study, especially as they relate to the lower rates of sensitivity among individuals with potentially incomplete claims history, and the higher PPV in the presence of certain combinations of diagnostic and procedure codes, will hold true in other settings, including to other third-party payers.
To the authors' knowledge, this is the first study to evaluate the quality of Medicaid claims data to ascertain incident cases of breast cancer. An important strength of the study is the use of claims data originating from all categories of service to retrieve cancer-related diagnosis and procedure codes, as well as the use of pharmacy claims to identify incident and prevalent cases of breast cancer. This approach seems to have been successful, as more than 89 percent of cases that were successfully linked with OCISS records were incident, rather than prevalent cases of cancer.
Findings from this study highlight the need to exercise great caution in analyzing Medicaid data and interpreting the results, mainly because of the intricacies of the Medicaid program. Incomplete claims history is an important factor to account for in studying outcomes in the Medicaid population, mainly due to discontinuity in enrollment, and participation in spend-down and/or Medicare programs. Participation in spend-down results in out-of-pocket expenditures, and therefore incomplete documentation of services received, and participation in Medicare results in claims in the Medicare database. With Medicaid being the payer of last resort for the dually eligible Medicare–Medicaid population, it is possible that some of the services may have been undocumented in the Medicaid database, or documented in an incomplete fashion. Future studies should use linked OCISS and Medicare–Medicaid files to assess the incremental benefit of adding Medicare claims in the analysis of evaluating the ability of claims data to ascertain incident cases of cancer in the dually eligible Medicare–Medicaid population. It is also important to be mindful of the fact that Medicaid is a safety net program. As described above, this implies that individuals would be likely to join the Medicaid program after undergoing part of the diagnostic and therapeutic regimens while being uninsured, or part of a different health care delivery system, resulting again in incomplete claims history in the Medicaid database.
In conclusion, much remains to be explored in assessing the utility of Medicaid claims data in cancer-related outcomes. This study evaluated only the ability of Medicaid claims data to ascertain incident cases of breast cancer in the Medicaid population. In the process, it identified programmatic issues that could affect the completeness of claims history. Additional studies are needed to explore the utility of Medicaid claims data in analyzing other aspects of cancer-related outcomes, such as cancer stage, treatment, and follow-up care, and to identify with greater certainty the circumstances under which Medicaid claims can be considered useful in studying cancer-related outcomes.