Background: Better understanding of migraine treatment in US clinical practice could be facilitated by availability of Migraine Disability Assessment (MIDAS) questionnaire results collected in routine care. We present results for migraine patients with MIDAS collected in routine clinical practice through an electronic medical record (EMR) system that presented the MIDAS questionnaire as an electronic form during the patient office encounter. The purpose of this retrospective observational study was to gain better understanding of migraine disability and migraine treatment patterns in US real-world clinical practice. Methods: In this EMR database study, patients were required to have 12 months baseline time for review of patient and clinical characteristics. Adult patients with documentation of migraine with subsequent MIDAS questionnaire data collected between March 2017 and September 2018 were included. Based on MIDAS responses patients were categorized into grade I—little or no disability, grade II—mild disability, grade III—moderate disability, and grade IV—severe disability. Results: This study included 2731 migraine patients with MIDAS results. Overall, 2309 (84.5%) were female with an average age of 46.7 years. Distribution by disability grade was 1161 (42.5%) little or no disability, 424 (15.5%) mild disability, 477 (17.5%) moderate disability, and 669 (24.5%) severe disability. Compared to overall, a larger proportion of patients with severe disability had baseline treatment with acute (71.3% vs. 67.6%) or preventive medications (70.4% vs. 62.0%) and to be on 3+ acute (9.4% vs. 7.0%) or 3+ preventive therapies (17.0% vs. 14.5%). Conclusion: Availability of MIDAS results in usual care provides additional insight into migraine care.

To describe treatment patterns with migraine preventive therapy in the US administrative claims database. This retrospective study identified a migraine patient cohort in the Truven Health MarketScan® Research Databases initiating a preventive migraine medication (index date) between 2006-2012. Patients had to be continuously enrolled for at least 12 months pre- and 24 months post-index and were followed for up to 5 years. Treatment patterns were evaluated within one year of starting the first, second, or third agent until a 60-day index agent gap or agent switch. Restarts or switches were evaluated in the year following a therapy gap. A total of 147,832 patients initiated preventive migraine treatment (mean age: 43.6 years, 83.0% female); 79,922 (54.1%) went on to second-line, and 40,853 (27.6%) to third-line treatment within one year. Persistence at one year was similar for first (23.3%), second (23.2%) and third lines of therapy (21.8%). Switches without a gap were similar across therapy lines (9.2%-9.9%). Following a 60-day gap after first-line therapy, 59.4% of patients did not use any preventive therapy within 12 months of the gap start; 23.6% restarted the same agent and 16.8% switched to a new agent. Similarly, 48.3% of second-line and 40.5% of third-line patients were not on preventive therapy for at least one year after the treatment gap, 32.5% of second-line and 38.0% of third-line patients restarted the same agent and 19.2% of a second-line and 21.5% of third-line patients switched to another agent. Persistence with current migraine preventive therapies are low, regardless of line of therapy or index agent and this leads to frequent preventive cycling. Less than one-quarter of patients were persistent with their index preventive migraine medication at one year.

To estimate the size of the migraine population currently treated with preventive therapy in the US, and describe patterns of use for migraine preventive medication. We identified continuously enrolled migraine patients at least 18 years of age in the Truven Health MarketScan® Research Databases between January 1, 2012 and December 31, 2014 and described their prophylactic treatment for migraine in 2014. People with migraine (PWM) were identified using a claims-based algorithm consisting of the migraine ICD-9 code 346.xx and one or more acute migraine-specific therapies (triptan or ergot-derivative). Those with evidence of epilepsy were excluded. Use of migraine preventive therapies (antiepileptic agents, beta-blockers, calcium channel blockers, anti-depressants, and other medications) was assessed using prescription claims. Indication for preventive medications was assigned based on proximity of diagnosis codes to the prescription fill, and only fills with an indication of migraine were included in the analysis. Patients were classified based on the number of distinct prescription preventive treatments for migraine they had received in the look-back period from January 1, 2012 until the most recent preventive therapy fill date. Approximately 359,000 PWM were identified in the MarketScan® database. Eighty-two percent were women, and 73% were between the ages 30-59 years. Overall, 34% (n=122,128) of PWM were currently using a preventive migraine therapy. Women were more likely than men to receive prevention. Among PWM on prevention 55% were on their initial therapy, 27% were on their 2nd therapy, 11% were on 3rd-therapy and 7% were on 4th preventive or higher. Using diagnostic and treatment data to identify migraine, just over one third of PWM were also prescribed a preventive treatment. Many individuals who receive preventive medications for migraine cycle through multiple therapies. These estimates are higher than previous studies because of restriction to persons with medically diagnosed migraine.

Background: Chemotherapy-induced febrile neutropenia (FN) is a clinically important complication that affects patient outcome by delaying chemotherapy doses or reducing dose intensity. Risk of FN depends on chemotherapy-and patient-level factors. We sought to determine the effects of chronic comorbidities on risk of FN.Design: We conducted a cohort study to examine the association between a variety of chronic comorbidities and risk of FN in patients diagnosed with six types of cancer (non-Hodgkin lymphoma and breast, colorectal, lung, ovary, and gastric cancer) from 2000 to 2009 who were treated with chemotherapy at Kaiser Permanente Southern California, a large managed care organization. We excluded those patients who received primary prophylactic granulocyte colony-stimulating factor. History of comorbidities and FN events were identified using electronic medical records. Cox models adjusting for propensity score, stratified by cancer type, were used to determine the association between comorbid conditions and FN. Models that additionally adjusted for cancer stage, baseline neutrophil count, chemotherapy regimen, and dose reduction were also evaluated.Results: A total of 19 160 patients with mean age of 60 years were included; 963 (5.0%) developed FN in the first chemotherapy cycle. Chronic obstructive pulmonary disease [hazard ratio (HR) = 1.30 (1.07-1.57)], congestive heart failure [HR = 1.43 (1.00-1.98)], HIV infection [HR = 3.40 (1.90-5.63)], autoimmune disease [HR = 2.01 (1.10-3.33)], peptic ulcer disease [HR = 1.57 (1.05-2.26)], renal disease [HR = 1.60 (1.21-2.09)], and thyroid disorder [HR = 1.32 (1.06-1.64)] were all associated with a significantly increased FN risk.Conclusions: These results provide evidence that history of several chronic comorbidities increases risk of FN, which should be considered when managing patients during chemotherapy.

High-dimensional propensity score (PS) methods have been used in health care claims data to improve control of confounding by adjusting for a large number of covariates that may be proxies for unobserved factors. We have previously shown that PS models are biased for non-linear link functions when confounders were included. We conducted a simulation study to understand whether inclusion of covariates that are not confounders may also bias the association by estimating Monte Carlo mean bias, relative efficiency (RE) and coverage probability (CP) of log odds ratios when covariates only related to the exposure or only related to the outcome were included. We conducted 1000 Monte Carlo simulations, and estimated effect of exposure using logistic regression models. The propensity score was included in the logistic model as a linear predictor or as a smoothed covariate using restricted cubic splines. Simulations were conducted for scenarios including 5, 15, and 25 covariates. Using the PS with 25 covariates related only to the binary exposure, Monte Carlo bias, standard error (SE), RE and CP were -0.002, 0.015, 1.34, and 0.94 when the PS was included as a smoothed covariate, and –0.002, 0.015, 1.31, and 0.94 when the PS was included as a linear covariate. The bias, SE, RE and CP for 25 covariates related to the binary outcome were 0.307, 0.096, 21.6, and 0 when the PS was included as a linear covariate. Bias tended to increase with more covariates. We observed minimal bias when using PS models where covariates were related only to the exposure, but substantial bias when the covariates were related to the outcome. PS models may not be appropriate for logistic models because these models do not adequately deal with errors in the outcome due to the covariate.

High-dimensional propensity score (HDPS) methods have been used in health care claims data in an attempt to control confounding by adjusting for a large number of covariates that may be proxies for unobserved factors. We have previously shown that PS models are biased with non-linear link functions. We conducted a Monte Carlo simulation study to understand whether inclusion of covariates that are children of unmeasured confounders and other unmeasured parents of the outcome variable (colliders) may bias the relationship between exposure and outcome by estimating mean bias, and standard errors. We used directed acyclic graphs to replicate the causal network of plausible confounding scenarios. We simulated a scenario where the outcome variable Y is a function of a confounder, C, and another parent, U, but not of exposure X (function of C). Covariate Z is a function of parents C and U. All variables had normally distributed random errors. We conducted Monte Carlo simulations of the causal network, with varying strengths of each of the causal relations, and estimated the effect of X on Y, using linear regression models, while adjusting for covariate Z. Correctly specified models were unbiased. Bias was large in models with X only (bias 1.5 with variable standard errors). There was some reduction in bias in some situations where Z was highly correlated with the confounder, C, but increased bias when much of the variance in Y was determined by U. Adjustment for colliders that are children of unmeasured determinants of the outcome variable, but not of the exposure, may also increase bias. This is of great importance in observational studies, particularly when using HDPS to adjust for large numbers of variables that are not true confounders. Researchers should always use causal knowledge when using data to make causal inference.

6027 Background: To estimate US rates of neutropenic complications per annum in inpatient settings over time, among discharges of all patients, patients with any cancer diagnosis, and patients with lung cancer, non-Hodgkin lymphoma (NHL), and female breast cancer. Methods: This descriptive, cross-sectional analysis used data from 1989-2007 from the Agency for Healthcare Research and Quality Healthcare Cost and Utilization Project Nationwide Inpatient Sample (NIS). Neutropenic complications were defined by a neutropenia diagnosis code (ICD-9- CM=288.0X) only, or the combination of neutropenia and infection diagnosis codes. Rates of neutropenic complications per 10,000 discharges were calculated for all discharges, cancer discharges (140.XX-208.XX), lung cancer (162.XX), NHL (NHL, 201.XX and 202.XX) and female breast cancer (174.XX). Discharges for patients < 18 years of age and for patients receiving therapeutic radiology (92.2X) or stereotactic radiosurgery (92.3X) were excluded. Results: The estimated annual number of U.S. hospital discharges ranged from 28 to 33 million from 1989-2007. Cancer discharges accounted for 2.3-2.7 million discharges per year. The rates of neutropenic complications per 10,000 cancer discharges increased through 1999 and stabilized or declined after 1999. The rate of neutropenia discharges per 10,000 cancer (female breast cancer, lung cancer, and NHL only) discharges was 288.9 (95% CI 265.7-312.1) in 1989. The rate increased to 560.4 (95% CI 533.3-581.4) by 1999, and then declined to 495.0 (95% CI 469.4-520.7) in 2007. Trends of neutropenic complications within cancer types were similar. Neutropenia complication rates in patients with lung cancer and NHL peaked in 1997 and then declined through 2007; however, neutropenia rates in female breast cancer did not decline as much as in NHL or lung cancer in the more recent years. Conclusions: After 1999, rates of neutropenic complications among breast cancer, lung cancer, and NHL patients stabilized or declined. Further research is needed to understand whether this is the result of policy changes influencing hospitalization decisions or changing treatment patterns for cancer and the management of treatment toxicity. Author Disclosure Employment or Leadership Position Consultant or Advisory Role Stock Ownership Honoraria Research Funding Expert Testimony Other Remuneration Amgen Amgen Amgen Amgen

Background: G-CSF minimizes severe neutropenia risk associated with chemotherapy, but its cost implications following chemotherapy are unknown. Objective: To examine the impact of G-CSF on medical costs after initial chemotherapy in ESBC (Stage I-III). Methods: Women diagnosed with ESBC in 1999-2005, who had an initial course of chemotherapy that began within 180 days of diagnosis and included ≥1 highly myelo-suppressive agent, were identified from the Surveillance, Epidemiology and End Results (SEER)-Medicare database. Medicare claims were used to describe the initial chemotherapy regimen according to the classes of agents used: anthracycline ([A]: doxorubicin or epirubicin); cyclophosphamide (C); taxane (T: paclitaxel or docetaxel); and fluorouracil (F). G-CSF use during initial chemotherapy was classified as follows: primary prophylaxis, if the first G-CSF claim was +/− 5 days of the start of the first chemotherapy cycle; secondary prophylaxis, if the first claim was +/− 5 days of the start of the second or subsequent cycles; G-CSF rescue, if the first claim occurred outside of prophylactic usage; and no G-CSF. Patients were described by age, race, year of diagnosis, stage, grade, ER/PR status, National Cancer Institute (NCI) Comorbidity Index, chemotherapy regimen, and type of G-CSF. Total direct medical costs from 4 weeks after the last chemotherapy administration up to 48 months were estimated. Medical costs included those for ESBC treatment and all other medical services received after chemotherapy. Least squares regression, using inverse probability weighting to account for censoring within the cohort, was used to evaluate associations between G-CSF use and costs, adjusted for patient demographic and clinical factors described above. Results: A total of 6,947 patients were identified with an average age of 72 years, of which 63% had Stage II disease, and 59% were ER and/or PR positive. Compared to no G-CSF, those receiving G-CSF primary prophylaxis were more likely to have Stage III disease (30% v. 16%), to be diagnosed in 2003-2005 (63% v. 26%), and to receive dose dense AC-T (25% versus 5%), while they were less likely to receive an F-based regimen (13% versus 34%). The estimated average direct medical cost over 48 months after initial chemotherapy was $81,045 in the cohort overall. In multivariate analysis, significantly greater costs were associated with stage II or III diagnosis (compared to Stage I), NCI Comorbidity Index score of 1 or ≥2 (compared to 0), or FAC or standard AC-T (each compared to AC). Adjusting for patient demographic and clinical factors, costs in the G-CSF primary and secondary prophylaxis groups were not significantly different from the no G-CSF group. In contrast, G-CSF rescue was associated with significantly greater costs (incremental cost = $9,930; 95% CI $2,234- $18,099) than no G-CSF. Conclusion: Direct medical costs after initial chemotherapy were similar between those receiving G-CSF primary or secondary prophylaxis and those receiving no G-CSF after adjusting for potential confounders. Citation Information: Cancer Res 2010;70(24 Suppl):Abstract nr P1-09-01.