BNP Level and Thrombotic Events in Older Patients With HF
BNP Level and Thrombotic Events in Older Patients With HF
We obtained hospitalization data from the ADHERE Core registry of patients hospitalized with acute decompensated HF. More than 300 community and academic centers in the United States participated, and >185,000 patients were enrolled between January 2001 and March 2006. The data include demographic characteristics, comorbid conditions, medications, hospital course, laboratory values, procedures, and discharge disposition.
To obtain long-term follow-up data, we linked the ADHERE data to the 100% Medicare inpatient and denominator files. The inpatient files contain hospital claims for reimbursement under Medicare Part A. For beneficiaries in fee-for-service Medicare, these files include service dates and diagnosis and procedure codes. The denominator files contain beneficiary demographic characteristics, enrollment information, and death dates. The files contain an encrypted identifier unique to each beneficiary to allow for longitudinal follow-up. The latest date of Medicare data availability for this study was December 31, 2007.
We linked ADHERE hospitalizations to Medicare claims using several indirect identifiers—hospital identifier, admission date, discharge date, patient sex, and either birth date or month and year of birth, as available. Combinations of these identifiers are almost completely unique, enabling identification of registry hospitals and registry hospitalizations in the Medicare claims data. The ADHERE records used for linking included hospitalizations of patients aged ≥65 years with complete data on the identifiers listed above. Medicare inpatient records used for linking included all hospitalizations of patients aged ≥65 years with an associated HF diagnosis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] diagnosis code 402.x1, 404.x1, 404.x3, or 428.x) in any position on the inpatient claim.
We included patients aged ≥65 years who had an ADHERE hospitalization linked to fee-for-service Medicare claims data. If multiple hospitalizations were linked for a single patient, we used the earliest. Eligible patients lived in the United States and were discharged alive to home. We excluded patients who left the hospital against medical advice, had a history of AF, or were receiving long-term warfarin therapy at admission. We further restricted the population to patients discharged between January 2003 and December 2006 who were treated in hospitals that collected BNP level at admission for at least 60% of patients enrolled in the registry. We excluded 2,366 patients with missing BNP values.
The study variable of interest was BNP level at admission, which was measured in each site's clinical laboratory. Assay and quality control methodology were at the discretion of the site. We analyzed BNP level as a continuous variable (range 0–5,000 pg/mL) and in quartiles (0–416 pg/mL, 417–816 pg/mL, 817–1,510 pg/mL, and 1,511–5,000 pg/mL).
We followed patients for up to 1 year after discharge. The outcomes of interest were all-cause mortality, thromboembolic events, MI, and ischemic stroke. We determined all-cause mortality based on death dates in the Medicare denominator files. We identified new MI (ICD-9-CM code 410.x1) based on a primary diagnosis on an inpatient claim after hospital discharge. Likewise, we identified thromboembolic events based on a primary diagnosis on a subsequent Medicare claim. These events included cerebral occlusion, nonhemorrhagic stroke, or transient ischemic attack (TIA) (433.x-437.x); arterial embolism or thrombosis (444.x and 445.x); or deep vein thrombosis, pulmonary embolism, or other venous thrombosis (415.1x, 451.1x, 451.2, 451.81, 451.9, 452.x, and 453.x). Finally, we identified ischemic stroke (433.x1 and 434.x1) based on a primary diagnosis on a subsequent inpatient Medicare claim. Data for patients who enrolled in Medicare managed care during follow-up were censored from the date of managed care enrollment.
Baseline characteristics from ADHERE included demographic characteristics, medical history, findings from the initial clinical evaluation, initial vital signs, laboratory test results, discharge symptoms, and discharge medications. For variables with low rates of missingness (ie, <5% of records), we imputed continuous variables to the overall median value and dichotomous variables to "no." We created categorical variables that included a category for missing values for variables with higher missing rates, including evaluation of left ventricular function (12.7%), admission weight (6.3%), and discharge symptoms (9.5%).
For baseline characteristics, we present categorical variables as frequencies and continuous variables as means with SDs. We grouped patients by quartile of BNP level at admission and tested for differences in baseline variables using χ tests for dichotomous and unordered categorical variables, the Cochran-Mantel-Haenszel row mean score for ordered categorical variables, and Kruskal-Wallis tests for continuous variables. In a preliminary analysis, we tested differences in baseline variables among patients with and without missing values for BNP level at admission.
We report unadjusted outcome rates stratified by quartile of BNP level. We used Kaplan-Meier methods to estimate mortality at 30, 60, 90, and 180 days and 1 year after discharge and used log-rank tests to assess differences in mortality by quartile. For other outcomes, we used the cumulative incidence function, which accounts for the competing risk of death, to calculate cumulative incidence estimates at 30, 60, 90, and 180 days and 1 year after discharge. We used Gray tests to assess differences in outcomes between groups.
We used Cox proportional hazards models to examine associations between BNP level at admission and mortality, MI, thromboembolic events, and ischemic stroke at 1 year after discharge. In multivariable analyses, we modeled each 1-year outcome as a function of log BNP level, age, sex, race, medical history, findings from the initial clinical evaluation, initial vital signs, laboratory test results, and length of stay >7 days. We used robust SEs to account for clustering of patients within hospitals. We censored follow-up data from the date of death in the models of MI, thromboembolic events, and ischemic stroke. In secondary analyses, we added interaction terms to test whether the effect of BNP level differed by serum creatinine level at admission, weight, or ejection fraction. We performed 2 sensitivity analyses. First, because BNP level was missing for approximately 17% of patients, we used multiple imputation in the Cox regression models and included patients with missing values for BNP level at admission. Second, we included discharge medications in the Cox regression models to exclude the possibility that differences in treatment at discharge influenced outcomes.
We used a significance level of .05 and 2-sided tests for all hypotheses. We used SAS version 9.2 (SAS Institute Inc, Cary, NC) for all analyses. The institutional review board of the Duke University Health System approved the study.
This work was supported by a research agreement between Duke University and Johnson & Johnson.
Methods
We obtained hospitalization data from the ADHERE Core registry of patients hospitalized with acute decompensated HF. More than 300 community and academic centers in the United States participated, and >185,000 patients were enrolled between January 2001 and March 2006. The data include demographic characteristics, comorbid conditions, medications, hospital course, laboratory values, procedures, and discharge disposition.
To obtain long-term follow-up data, we linked the ADHERE data to the 100% Medicare inpatient and denominator files. The inpatient files contain hospital claims for reimbursement under Medicare Part A. For beneficiaries in fee-for-service Medicare, these files include service dates and diagnosis and procedure codes. The denominator files contain beneficiary demographic characteristics, enrollment information, and death dates. The files contain an encrypted identifier unique to each beneficiary to allow for longitudinal follow-up. The latest date of Medicare data availability for this study was December 31, 2007.
We linked ADHERE hospitalizations to Medicare claims using several indirect identifiers—hospital identifier, admission date, discharge date, patient sex, and either birth date or month and year of birth, as available. Combinations of these identifiers are almost completely unique, enabling identification of registry hospitals and registry hospitalizations in the Medicare claims data. The ADHERE records used for linking included hospitalizations of patients aged ≥65 years with complete data on the identifiers listed above. Medicare inpatient records used for linking included all hospitalizations of patients aged ≥65 years with an associated HF diagnosis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] diagnosis code 402.x1, 404.x1, 404.x3, or 428.x) in any position on the inpatient claim.
We included patients aged ≥65 years who had an ADHERE hospitalization linked to fee-for-service Medicare claims data. If multiple hospitalizations were linked for a single patient, we used the earliest. Eligible patients lived in the United States and were discharged alive to home. We excluded patients who left the hospital against medical advice, had a history of AF, or were receiving long-term warfarin therapy at admission. We further restricted the population to patients discharged between January 2003 and December 2006 who were treated in hospitals that collected BNP level at admission for at least 60% of patients enrolled in the registry. We excluded 2,366 patients with missing BNP values.
The study variable of interest was BNP level at admission, which was measured in each site's clinical laboratory. Assay and quality control methodology were at the discretion of the site. We analyzed BNP level as a continuous variable (range 0–5,000 pg/mL) and in quartiles (0–416 pg/mL, 417–816 pg/mL, 817–1,510 pg/mL, and 1,511–5,000 pg/mL).
We followed patients for up to 1 year after discharge. The outcomes of interest were all-cause mortality, thromboembolic events, MI, and ischemic stroke. We determined all-cause mortality based on death dates in the Medicare denominator files. We identified new MI (ICD-9-CM code 410.x1) based on a primary diagnosis on an inpatient claim after hospital discharge. Likewise, we identified thromboembolic events based on a primary diagnosis on a subsequent Medicare claim. These events included cerebral occlusion, nonhemorrhagic stroke, or transient ischemic attack (TIA) (433.x-437.x); arterial embolism or thrombosis (444.x and 445.x); or deep vein thrombosis, pulmonary embolism, or other venous thrombosis (415.1x, 451.1x, 451.2, 451.81, 451.9, 452.x, and 453.x). Finally, we identified ischemic stroke (433.x1 and 434.x1) based on a primary diagnosis on a subsequent inpatient Medicare claim. Data for patients who enrolled in Medicare managed care during follow-up were censored from the date of managed care enrollment.
Baseline characteristics from ADHERE included demographic characteristics, medical history, findings from the initial clinical evaluation, initial vital signs, laboratory test results, discharge symptoms, and discharge medications. For variables with low rates of missingness (ie, <5% of records), we imputed continuous variables to the overall median value and dichotomous variables to "no." We created categorical variables that included a category for missing values for variables with higher missing rates, including evaluation of left ventricular function (12.7%), admission weight (6.3%), and discharge symptoms (9.5%).
For baseline characteristics, we present categorical variables as frequencies and continuous variables as means with SDs. We grouped patients by quartile of BNP level at admission and tested for differences in baseline variables using χ tests for dichotomous and unordered categorical variables, the Cochran-Mantel-Haenszel row mean score for ordered categorical variables, and Kruskal-Wallis tests for continuous variables. In a preliminary analysis, we tested differences in baseline variables among patients with and without missing values for BNP level at admission.
We report unadjusted outcome rates stratified by quartile of BNP level. We used Kaplan-Meier methods to estimate mortality at 30, 60, 90, and 180 days and 1 year after discharge and used log-rank tests to assess differences in mortality by quartile. For other outcomes, we used the cumulative incidence function, which accounts for the competing risk of death, to calculate cumulative incidence estimates at 30, 60, 90, and 180 days and 1 year after discharge. We used Gray tests to assess differences in outcomes between groups.
We used Cox proportional hazards models to examine associations between BNP level at admission and mortality, MI, thromboembolic events, and ischemic stroke at 1 year after discharge. In multivariable analyses, we modeled each 1-year outcome as a function of log BNP level, age, sex, race, medical history, findings from the initial clinical evaluation, initial vital signs, laboratory test results, and length of stay >7 days. We used robust SEs to account for clustering of patients within hospitals. We censored follow-up data from the date of death in the models of MI, thromboembolic events, and ischemic stroke. In secondary analyses, we added interaction terms to test whether the effect of BNP level differed by serum creatinine level at admission, weight, or ejection fraction. We performed 2 sensitivity analyses. First, because BNP level was missing for approximately 17% of patients, we used multiple imputation in the Cox regression models and included patients with missing values for BNP level at admission. Second, we included discharge medications in the Cox regression models to exclude the possibility that differences in treatment at discharge influenced outcomes.
We used a significance level of .05 and 2-sided tests for all hypotheses. We used SAS version 9.2 (SAS Institute Inc, Cary, NC) for all analyses. The institutional review board of the Duke University Health System approved the study.
This work was supported by a research agreement between Duke University and Johnson & Johnson.
Source...