The National Heart, Lung, & Blood Institute-sponsored F
amily Cardiac Caregiver I
o Evaluate O
) Study was a prospective study designed to evaluate patterns of caregiving among cardiac patients and its association with clinical outcomes of patients hospitalized with CVD (1
). Consecutive patients admitted to the CVD service line at Columbia University Medical Center/New York Presbyterian Hospital (CUMC/NYPH) between November 2009 and June 2010 were asked to complete a standardized questionnaire regarding caregiving (93% enrollment rate). Medical cardiovascular service patients (N=3188) who had 1 year follow up by June 2011 were included in the primary analysis. The baseline characteristics and patterns of caregiving in this population have been published previously (1
). Briefly, hospital admission logs were reviewed daily to identify patients admitted to the CVD service, and trained bilingual research staff systematically distributed surveys in English and Spanish to potential participants to assess whether or not they had a caregiver within the past year and plans for one after discharge. Patients were excluded from survey administration if they were unable to read or understand English or Spanish, lived in a full-time nursing facility, mental status precluded participation, or they refused to complete the survey. Hospital logs were checked weekly to detect any uncollected surveys. If uncollected surveys were detected, research staff attempted to contact the patient prior to discharge, or in the event this was not feasible, the survey was mailed with a pre-stamped return envelope for the patient to complete and return. The study was approved by the Institutional Review Board of CUMC.
The definition of caregiving was adapted from the report of the National Alliance of Caregiving and AARP (2
). Methods for standardized assessment of caregiving have been described in our prior work (1
). A caregiver was defined as 1) a paid professional (e.g. nurse/home aide) or 2) an informal (non-paid) person, who assists the patient with medical and/or preventive care. Data on having a caregiver in the past year and plans for having a caregiver after discharge were evaluated and found to be similar, therefore the former was used. Patients who reported having both paid and
informal caregivers [n=120] were categorized as having a paid caregiver.
The extent of caregiving provided to each participant who reported having a caregiver was systematically assessed based on the specific tasks the caregiver performed. Tasks were defined using Basic Activities of Daily Living (e.g., assistance with dressing, bathing), and Instrumental Activities of Daily Living (e.g., assistance with meal preparation, transportation). Extent of caregiving provided was categorized as 1) Extensive (patient has a paid caregiver or an informal caregiver who provides assistance with Basic Activities of Daily Living only, or Basic Activities of Daily Living plus Instrumental Activities of Daily Living), or 2) Non-Extensive (informal caregiver provides assistance with Instrumental Activities of Daily Living or less, or patient has no caregiver).
Baseline characteristics, medical history, admission diagnoses, and prescription medications were documented by standardized electronic chart review. Patient medical records were accessed via a secure and comprehensive electronic clinical information system at CUMC/NYPH. Admission diagnosis (CVD vs. Non-CVD) was determined by ICD-9 billing code for admission or primary diagnosis, and were validated in a sub-study by an independent physician reviewer blinded to ICD-9 billing code and caregiver status (n=50; kappa=0.99). All research staff members were HIPAA trained and certified in the use of this clinical information system. Current and prior medical conditions including diabetes mellitus, renal disease, myocardial infarction, peripheral vascular disease, heart failure, and chronic obstructive pulmonary disease were determined using ICD-9 billing codes and physician or nurse practitioner notes. Prior medical history information was collected by a trained nurse research assistant and was available for 99% of this population. The number of different prescribed medications and names and/or types of medications were obtained from discharge summary notes and supplemented by the ambulatory electronic records if needed.
The primary outcome was all-cause rehospitalization or death within 1 year and secondary outcomes were CVD rehospitalization and all-cause mortality assessed individually. Methods used to collect outcomes data were similar to those previously tested in other studies of hospitalized CVD patients (3
). Rehospitalization was systematically obtained via CUMC/NYPH electronic clinical information systems which is updated daily. Patients’ admitting date, admitting diagnosis, and primary diagnose(s) for each hospitalization and rehospitalization were recorded. Readmission type was classified (CVD vs. Non-CVD) using ICD-9 billing codes. To supplement outcomes data obtained by CUMC/NYPH electronic medical records, all patients were systematically contacted via mail or telephone 1 year after the index hospitalization that corresponded with their baseline survey interview date and queried regarding rehospitalization in the prior year (80% response rate). Rehospitalization was defined as rehospitalization at CUMC/NYPH or elsewhere, for CVD or for other reasons; analyses using this definition were similar to analysis limited to readmission to CUMC/NYPH only. Vital status was obtained via the clinical information system, which is updated monthly with National Death Index data.
The Ghali Comorbidity Index (Ghali-CI) was calculated on all patients using past medical history data obtained through systematic electronic record review. The Ghali-CI was developed by assigning study-specific data-derived weights to the original, widely-used Charlson comorbidity variables (4
). Condition and study-specific comorbidity weights have been shown to be better predictors of adverse outcomes than standard scores used to summarize comorbidity (4
). The Ghali-CI has been shown to be superior to the Charlson for prediction of in-hospital mortality among cardiac patients (6
). The weighted conditions used to calculate the Ghali-CI are: myocardial infarction, heart failure, peripheral vascular disease, and moderate or severe renal disease. Total score range is 0–11, with patients scoring 0 at the lowest risk. For analysis the Ghali-CI was dichotomized at >1 versus 1 based on research indicating scores >1 are consistent with significant comorbidities. In a subset of participants with data available to calculate both scores the Ghali-CI was significantly correlated with New York State Department of Health Risk Scores for percutaneous coronary intervention (n=613; p<.001).
Surveys were created and processed using the intelligent character recognition software EzDataPro32™ (version 8.0.7, Creative ICR, Inc., Beaverton, OR) and ImageFomula (version Dr-2580C, Canon U.S.A., Inc., New York, NY). The data were double checked for errors and stored in a Microsoft Access database. Descriptive data are presented as frequencies and percentages. Caregiving was categorized as having paid, informal, or any (either paid or informal) caregivers. Chi-square tests were performed to determine the association between caregiving and baseline characteristics of hospitalized CVD patients using a Bonferroni correction for multiple comparisons (p<0.017). The independent association between caregiving and clinical outcomes was evaluated by logistic regression adjusted for confounders. A stratified analysis by baseline admission type (current or prior heart failure versus no history of heart failure) was also conducted.
To evaluate the potential role of exposure selection bias, propensity score weights were calculated and a propensity score-weighted logistic model was fitted (7
); model covariates included demographic variables (age, gender, race/ethnicity, health insurance), Ghali-CI, and comorbid conditions not accounted for by the Ghali-CI, but associated with death or rehospitalization at 1 year (diabetes mellitus, chronic obstructive pulmonary disease, number of prescription medications at discharge, and history of hypertension). Analyses were conducted using SAS software (version 9.2, SAS Institute, Cary, NC). Statistical significance for logistic regression models was set at p <0.05.