Evaluating the safety and use of erythropoietin among cancer patients in the VA medical system Continue work with the VA to refine our knowledge of and identify factors associated with the high risk of adverse effects related to erythropoiesis stimulating agent (ESA) use in cancer. Staff from the VA will provide expertise on the VA medical system in terms of both logistics of conducting the research (e.g. access to the data, expertise with using the data) and the study design. They will abstract the required data from the VA national database and construct the analytic files, which will be made available to the study team. They will also participate in investigator meetings and co-authoring study manuscripts. Study Design: In this study, we will develop a cohort of cancer patients who have received cancer care at the VA medical center. A historic cohort will be created from health care service, medical record, and laboratory results data from the national VA database. Patients diagnosed with cancer within the study period, from July 1, 2003 to June 1, 2008 will be included in the cohort. The study will be further limited to patients who consistently use the VA for medical care. This will be defined as those patients with at least two medical visits in the six months prior to the index date (first cancer diagnosis code) and receiving their cancer treatment (surgery, chemotherapy, radiation) at the VA. Patients will be followed from the initial diagnosis of cancer during the study period until the end of the study period, death, or stopped receiving care at the VA. Exclusions: Previous tumor diagnosis in the year prior to the beginning of the study period (July 1, 2003), age <18 years. Patients with End Stage Renal Disease (ESRD) and those with a history of ESA use in the year prior to tumor diagnosis will be excluded. Study variables: The first record of a cancer diagnosis and the type of cancer will be determined from the diagnosis codes available in the medical record. We will use the VA cancer registry to confirm the cancer diagnosis and determine cancer characteristics. If access to the national registry is not feasible or possible, then registry data will be collected for a subset of patients from one or more of the 21 regional centers or Veterans Integrated Services Networks (VISN). Cancer treatments will be characterized by type, date received, and order. Chemotherapy will be characterized by the therapeutic regimens received, the start and stop date for each regimen, and the total duration of chemotherapy. The dates of each ESA exposure will be abstracted from medical service (CPT codes) and pharmaceutical records. Patients will be classified by use of ESA (yes/no), duration of use, total number of prescriptions received, whether billed within the clinic or as an outpatient prescription, and timing of use in relation to other cancer treatments. Hemoglobin/hematocrit levels and the date of each assessment will be abstracted from laboratory results database. Clinical evaluation of anemia will be characterized by evidence of laboratory testing prior to and while on treatment (yes/no), testing frequency, baseline hemoglobin/hematocrit, last test prior to discontinuation, percent change in hemoglobin/hematocrit, and maximum hemoglobin achieved. Patient demographic and behavioral characteristics will be obtained from electronic medical records. Additionally, receipt of blood transfusions for anemia will be determined from medical service record patients classified by receipt of transfusion (yes/no), number of transfusions, and the timing of the transfusion(s). Major co-morbidities including hypertension, diabetes, chronic kidney disease, congestive heart failure, and other cardiovascular diseases will be identified by searching for these diagnoses in the medical record and services received (inpatient and outpatient) and sentinel prescriptions, and the burden of (non-cancer) chronic disease will be calculated. Evidence of venous thromoboembolism (VTE) will be identified through diagnostic codes in in-patient and outpatient services. However, the validity of using a VTE diagnostic code to identify true cases in the VA system is unknown. We propose to conduct a validation study, to evaluate the validity of a VTE diagnostic code among patients in this cohort who had a diagnosis of VTE. Records of up to 100 patients with a VTE code will be abstracted for evidence. Additionally, a random sample of 50 patients who received objective testing for VTE based on CPT codes but have no diagnosis of VTE will be selected (sample size to be determined). The medical record and test results will be evaluated independently by two clinical experts/physicians. They will then meet to review and determine the final status of cases where there was disagreement. All potential VTE cases will be classified as definite (based on objective test results), possible (based on clinical evaluation and examination), and no evidence (diagnosis ruled out by testing). The sensitivity and positive predictive value of diagnostic codes for identification of VTE from diagnostic codes will be calculated. Statistical analysis: The analysis will be conducted for the entire population, and then stratified by cancer site. Descriptive statistics will be calculated using SAS or STATA. The incidence of having at least one ESA prescription will be calculated monthly and annually. The overall rates of VTE and blood transfusions will be calculated by month and annually, with the index date (March 2007) as the date of initial FDA regulatory action. This was the month in which a health professional information sheet with the new safety information was disseminated. There was a change in labeling and public health alert in November 2007. The relative risk of these three key outcomes and 95% confidence intervals (Pre-post March 2007 and November 2007) will be calculated, using EpiInfo (CDC, Atlanta, GA). Potential confounders and effect modifiers will be identified by comparison of the crude, stratified, and Mantel-Hanzel pooled relative risks. The risk of outcomes (all-cause mortality and cancer-mortality, VTE, and transfusions) will be calculated using Cox proportional hazards models to control for confounders. The time period (pre/post regulatory action) will be added to the model as an independent variable and tested for effect modification.
