Ralph H. Raasch, Jonathan S. Serody, David A. Rizzieri, Anthony S. Stein, Michael E. Trigg, G. Thomas Ray, Patricia Saddier, Neil Hayes, Laurel A. Habel, Trung Nam Tran, Yan Li, and David J. Weber
Abstract 2484 Poster Board II-461 Background: Hematologic (HM) and solid tumor malignancy (STM) patients may be immunocompromised (IC) due to their underlying diseases and the immunosuppressive therapies they received. Availability of a practical and robust algorithm to classify HM and STM patients into IC levels using data from healthcare databases could be valuable for a variety of epidemiologic, health service or outcome research in the field of oncology. Classification of the IC status of patients also permits accurate prediction of the types of supportive care that such patients may need to prevent infectious complications that often accompany treatments for the underlying malignancy. Methods: An expert panel mainly comprised of hematologists and oncologists developed an algorithm to classify HM and STM patients' level of IC into either none/low, medium, or high, using data available in electronic databases. The algorithm was based on the following factors: (1) the type of chemotherapy agents and/or corticosteroids, (2) time since last chemotherapy/corticosteroid treatment, and (3) specific type of HM (for HM patients). Chemotherapy agents were classified into levels of IC, irrespective of dose used. Corticosteroid therapy was classified into levels of IC based on dose and duration of treatment. IC levels were allowed to change monthly to reflect what chemotherapy agents were used, dose and duration of corticosteroid if any, and time since the last IC treatment. In the base case scenario, the patient's IC level (based on treatment) stayed at an assigned level for 6 months after the last treatment and then moved to the next lower level for an additional 6 months. In alternative scenarios, sensitivity analyses were also performed using the 1, 3, 9, and 12 month cutoffs. If the patient received multiple chemotherapy agents/corticosteroid regimens, the most immunocompromising agent determined the IC level during that time period. We applied and tested this algorithm in a study of HM and STM patients diagnosed in 2001-2005 at Kaiser Permanente Northern California (KPNC). Herpes zoster (HZ), a viral disease caused by the reactivation of varicella zoster virus, is associated with impairment of cell-mediated immunity. Therefore, we used incidence of HZ as a proxy for true IC status. The KPNC cancer registry was used to identify cancer diagnoses and the type, stage and grade of the underlying HM. Data on specific chemotherapy agents and/or dose and duration of corticosteroids as well as time since last IC treatment were obtained from KPNC pharmacy databases. Potential episodes of HZ in 2001-2006 were identified from HZ diagnosis codes and antiviral use in various KPNC databases. HZ diagnosis was confirmed by clinical review of patient's medical records. We measured HZ incidence rates in HM and STM patients and examined whether they were correlated with IC level based on our algorithm. Results: In the base case scenario, among the 4,465 patient-years (py) of follow-up in HM patients, 25.3%, 34.4%, and 40.3% of follow-up time was categorized as none/low, medium, or high IC, respectively. The corresponding rates of HZ were 13, 25, and 48/1000 py. Among the 23,072 py of follow-up in STM patients, 74.9%, 8.0%, and 17.1% of follow-up time was categorized as none/low, medium, or high IC, respectively. The corresponding rates of HZ were 10, 20, and 19/1000 py, respectively. The algorithm was not sensitive to changes from 3 to 12 months, but was sensitive to the 1 month cutoff, in the assumption of duration of IC since the last IC treatment. Conclusions: It is feasible and practical to categorize cancer patients into IC levels using electronic pharmacy and cancer registry databases. The correlation between incidence of HZ and levels of IC in both HM and STM patients suggested that the proposed algorithm may appropriately assign IC levels in these patients. Additional testing in other cancer populations may be needed to further validate this algorithm. Disclosures: Tran: Merck & Co., Inc.: Employment. Ray:Merck & Co., Inc.: Investigative. Saddier:Merck & Co., Inc.: Employment. Trigg:Merck & Co., Inc.: Employment. Hayes:Merck & Co., Inc.: Consultancy. Li:Merck & Co., Inc.: Investigative. Rizzieri:Merck & Co., Inc.: Consultancy. Stein:Merck & Co., Inc.: Consultancy. Weber:Merck & Co., Inc.: Consultancy. Serody:Merck & Co., Inc.: Consultancy. Raasch:Merck & Co., Inc.: Consultancy. Habel:Merck & Co., Inc.: Investigative.