Validating the Follicular Lymphoma International Prognostic Index (FLIPI) Using SEER-Medicare Claims Data

Introduction: The FLIPI was published in 2004 and identified key risk factors for overall survival in patients with follicular lymphoma (FL).[1] A 2007 presentation showed that FLIPI score is also a good predictor of progression free survival. [2] While valuable for clinical decision-making, a FLIPI-like score may also be useful for comparing patients by severity levels in large claims databases. However, claims data do not contain sufficient information to construct such an index. Therefore, we explored the development of a simple FL severity index to facilitate analyses of claims data as well as to provide additional evidence of the importance of selected components of the original index.

Methods: The SEER-Medicare Follicular Index (SMFLI) was constructed using the National Cancer Institute’s SEER registry linked to Medicare claims. We identified 2,574 patients with FL (histology codes 9690, 9691, 9695, and 9698) as their first malignancy between January 1, 1998 and December 31, 2002. Patients were followed until the development of a second primary cancer, transition to an HMO, the end of their claims history (December 31, 2005), or death. Multivariate Cox proportional hazards analyses were performed to identify patient socio-demographic and clinical factors associated with survival. Potential risk factors in the models included age, gender, race, stage, comorbidity burden*, year of diagnosis, education, rural/urban status, poverty indicators, extra-nodal involvement, anemia*, neutropenia*, and thrombocytopenia* (*=based on ICD-9 diagnosis codes in claims data). Two FLIPI components, serum LDH and number of nodal sites, were not available in either SEER or Medicare data.

Results: Median age was 76 years, 42% were male, 10% were non-white, and 44% were diagnosed with Stage III or IV disease. There were 1,226 all-cause deaths in the follow up period. The strongest significant predictors of mortality, based on the Chi-square value from the multivariate model, included age 75–80 years (hazard ratio [HR]=1.91), age ≥80 years (HR=3.34), stage III (HR=1.60) or stage IV (HR=2.14), male gender (HR=1.30), and anemia (HR=1.35). Because the chi-square was similar for all factors except age ≥80 years, each was given equal weight and combined into a 5-level SMFLI scale. (Age≥80 years was given a weight of 2 to reflect its contribution.) A preliminary 3-level index was created by combining patients into approximately equal-sized groups in the following way: Stratum 1= SMFLI score 0–1, Stratum 2= SMFLI score 2, and Stratum 3=SMFLI score 3–5. Unadjusted survival among the 3 groups is compared in the figure below (p<0.001 for the log-rank test).

Discussion: Although re-constructing the FLIPI is not possible in claims data, a similar index can be created using available data. Key risk factors identified in both the FLIPI and SMFLI include age, stage, and hemoglobin (anemia in the SMFLI). Because the data are based on the Medicare population, age is a very strong predictor in the SMFLI, an effect that was present after adjustment for comorbidity burden. Using an index like this, it is now possible to compare patients across severity levels with regard to resource utilization and cost.

View largeDownload slide

Figure

View largeDownload slide

Figure

Close modal