Patients, Family Caregivers, and Quality of Chronic Care
Patients, Family Caregivers, and Quality of Chronic Care
We conducted cross-sectional analysis of baseline data for patient–caregiver dyads enrolled in a cluster-randomized controlled trial of a model of primary care (Guided Care).
The Guided Care intervention was designed to enhance the quality of health care for high-risk older adults by integrating a specially trained registered nurse into primary care practices. The patient eligibility criteria for the Guided Care study were community-dwelling, aged 65 or above, seen by a participating physician within the previous year and identified as at-risk of incurring high health care costs. Patients were identified by analyzing health insurance claims with the Hierarchical Condition Category predictive model. Eligible patients were in the highest quartile of risk for incurring high health care costs in the next year. Exclusions were cognitive impairment with no legal representative, inability to participate in the baseline or follow-up interviews and not speaking English. Among eligible patients, 38% consented to participate in the baseline survey and be randomized to the Guided Care intervention or control group for three years (N = 904). Following informed consent, participating patients were administered an in-person baseline interview. All baseline interviews were conducted in the patients' homes by trained, closely supervised professional interviewers who used computer-assisted interviewing technology and underwent 10% reliability testing (see Boult et al. for more details on the parent study).
During the baseline interview, patients enrolled in Guided Care were asked whether they received assistance with activities of daily living (ADL), instrumental activities of daily living (IADL) or health care tasks from a family member or unpaid friend. Patients who reported receiving assistance were then asked to identify the person who helped the most (their 'caregiver'). These identified caregivers were then invited to participate in the Guided Care study and baseline interview. Patients identified 353 eligible caregivers; 86% consented to participate (N = 308). Following consent, baseline interviews were conducted with caregivers in the caregivers' homes by the trained interviewers. If the patient and caregiver lived together, efforts were made to conduct the baseline interview with the caregiver and patient separately. The time between the patient and caregiver interviews averaged 26 days (see Wolff et al. for more detail).
Quality of care was assessed using the Patient Assessment of Chronic Illness Care (PACIC), a validated measure of patients' perceptions of the quality of chronic illness care. It consists of 20 Likert-scaled items that ask about the frequency with which specific care processes occurred during the past 6 months (response categories: almost never = 1, generally not = 2, sometimes = 3, most of the time = 4 and almost always = 5). For caregivers, the PACIC questions were modified to inquire about the caregiver's perceptions of the quality of chronic illness care (e.g. 'Over the past six months, when (PATIENT) received care for his/her chronic illness, he/she was…'). Caregivers were also given the opportunity to respond 'don't know' on any PACIC item. Items on the PACIC that were answered by the patient or caregiver with responses of 'don't know' or 'refuse' were counted as missing.
Previous validation studies using confirmatory factor analysis have shown that the PACIC measures a one-dimensional construct. Therefore, a median PACIC item score (median response across all cases to a single item) and a median respondent score (median response across all 20 PACIC items for a single respondent) was constructed. We hypothesized a priori that agreement would differ across objective and subjective items, based on previous researches. To distinguish objective PACIC items from subjective items, we convened a panel of experts including both geriatricians and health services researchers. Through a Delphi process, questions were categorized into those that could be answered objectively, those that inquired about communication about health care and those that were subjective in nature.
The baseline interviews with patients also collected information regarding sociodemographic characteristics (age, gender, race, education and financial situation), health (SF-36 and numbers of medications), self-reported chronic conditions (hypertension, angina, congestive heart failure, heart attack, stroke, asthma, chronic obstructive pulmonary disease, arthritis, sciatica, diabetes, cancer, osteoporosis, hip fracture and dementia) and the nature of assistance received from caregivers (number of ADL, IADL and health care tasks). The baseline interviews with caregivers also collected information about sociodemographic characteristics, including age, gender, education, relationship to patient, co-residence with the patient and average hours of care provided in a typical week. Caregivers were also administered the Caregiver Strain Index (CSI), a 13-item index (range: 0–24) originally developed to screen for caregiver strain after hospital discharge of an elderly family member. Patients and caregivers were administered the Health Care Task Difficulty (HCTD) scale, which assessed their difficulty in performing (or assisting with) health care management tasks such as taking medication, visiting health care providers and managing medical bills. Using this scale, caregivers and patients were categorized into no (HCTD = 0), low (HCTD = 1), medium (HCTD = 2) or high (HCTD = 3+) difficulty groups.
To quantify agreement, a weighted kappa statistic was calculated. The weighted kappa (Kw) statistic describes the level of agreement, corrected for agreement by chance and adjusted for the ordinal nature of the response categories. Weights are based on the degree of difference in response. For example, a patient and caregiver that differed by one category (almost never vs. generally not) would be weighted 0.25, where as a patient and caregiver that differed by four categories (almost never vs. almost always) would be weighted 1.00. A higher Kw indicates higher agreement (Kw > 0.40 is considered 'moderate agreement'). Analyses were conducted using STATA v.11. Confidence intervals were estimated using bootstrapping techniques with the kapci command.
Initially, we calculated a patient–caregiver Kw for each PACIC item to examine the variability in agreement across items. A lack of significant variability across individual items allowed us to calculate a median patient–caregiver Kw that reflected overall agreement between patients and caregivers across all 20 items. Association between caregiver/patient characteristics and agreement was measured by comparing Kw and the corresponding 95% CI across characteristics. A significant difference was determined by non-overlapping 95% CI. To allow for comparison, all continuous variables (age, SF-36, number of chronic conditions, hours of care and CSI) were converted to binary variables above and below the sample medians.
A weighted kappa is limited by the prevalence of the trait being measured (e.g. a measure of a rare trait is likely to have a low kappa). To account for the possible bias inherent in the kappa statistic, we conducted a sensitivity analysis with polychoric correlations. The polychoric correlation is used when examining correlation across ordinal categories.
Methods
Study Design
We conducted cross-sectional analysis of baseline data for patient–caregiver dyads enrolled in a cluster-randomized controlled trial of a model of primary care (Guided Care).
Parent Study Description (Guided Care)
The Guided Care intervention was designed to enhance the quality of health care for high-risk older adults by integrating a specially trained registered nurse into primary care practices. The patient eligibility criteria for the Guided Care study were community-dwelling, aged 65 or above, seen by a participating physician within the previous year and identified as at-risk of incurring high health care costs. Patients were identified by analyzing health insurance claims with the Hierarchical Condition Category predictive model. Eligible patients were in the highest quartile of risk for incurring high health care costs in the next year. Exclusions were cognitive impairment with no legal representative, inability to participate in the baseline or follow-up interviews and not speaking English. Among eligible patients, 38% consented to participate in the baseline survey and be randomized to the Guided Care intervention or control group for three years (N = 904). Following informed consent, participating patients were administered an in-person baseline interview. All baseline interviews were conducted in the patients' homes by trained, closely supervised professional interviewers who used computer-assisted interviewing technology and underwent 10% reliability testing (see Boult et al. for more details on the parent study).
Study Sample
During the baseline interview, patients enrolled in Guided Care were asked whether they received assistance with activities of daily living (ADL), instrumental activities of daily living (IADL) or health care tasks from a family member or unpaid friend. Patients who reported receiving assistance were then asked to identify the person who helped the most (their 'caregiver'). These identified caregivers were then invited to participate in the Guided Care study and baseline interview. Patients identified 353 eligible caregivers; 86% consented to participate (N = 308). Following consent, baseline interviews were conducted with caregivers in the caregivers' homes by the trained interviewers. If the patient and caregiver lived together, efforts were made to conduct the baseline interview with the caregiver and patient separately. The time between the patient and caregiver interviews averaged 26 days (see Wolff et al. for more detail).
Outcome Measure
Quality of care was assessed using the Patient Assessment of Chronic Illness Care (PACIC), a validated measure of patients' perceptions of the quality of chronic illness care. It consists of 20 Likert-scaled items that ask about the frequency with which specific care processes occurred during the past 6 months (response categories: almost never = 1, generally not = 2, sometimes = 3, most of the time = 4 and almost always = 5). For caregivers, the PACIC questions were modified to inquire about the caregiver's perceptions of the quality of chronic illness care (e.g. 'Over the past six months, when (PATIENT) received care for his/her chronic illness, he/she was…'). Caregivers were also given the opportunity to respond 'don't know' on any PACIC item. Items on the PACIC that were answered by the patient or caregiver with responses of 'don't know' or 'refuse' were counted as missing.
Previous validation studies using confirmatory factor analysis have shown that the PACIC measures a one-dimensional construct. Therefore, a median PACIC item score (median response across all cases to a single item) and a median respondent score (median response across all 20 PACIC items for a single respondent) was constructed. We hypothesized a priori that agreement would differ across objective and subjective items, based on previous researches. To distinguish objective PACIC items from subjective items, we convened a panel of experts including both geriatricians and health services researchers. Through a Delphi process, questions were categorized into those that could be answered objectively, those that inquired about communication about health care and those that were subjective in nature.
Exploratory Variables
The baseline interviews with patients also collected information regarding sociodemographic characteristics (age, gender, race, education and financial situation), health (SF-36 and numbers of medications), self-reported chronic conditions (hypertension, angina, congestive heart failure, heart attack, stroke, asthma, chronic obstructive pulmonary disease, arthritis, sciatica, diabetes, cancer, osteoporosis, hip fracture and dementia) and the nature of assistance received from caregivers (number of ADL, IADL and health care tasks). The baseline interviews with caregivers also collected information about sociodemographic characteristics, including age, gender, education, relationship to patient, co-residence with the patient and average hours of care provided in a typical week. Caregivers were also administered the Caregiver Strain Index (CSI), a 13-item index (range: 0–24) originally developed to screen for caregiver strain after hospital discharge of an elderly family member. Patients and caregivers were administered the Health Care Task Difficulty (HCTD) scale, which assessed their difficulty in performing (or assisting with) health care management tasks such as taking medication, visiting health care providers and managing medical bills. Using this scale, caregivers and patients were categorized into no (HCTD = 0), low (HCTD = 1), medium (HCTD = 2) or high (HCTD = 3+) difficulty groups.
Analysis
To quantify agreement, a weighted kappa statistic was calculated. The weighted kappa (Kw) statistic describes the level of agreement, corrected for agreement by chance and adjusted for the ordinal nature of the response categories. Weights are based on the degree of difference in response. For example, a patient and caregiver that differed by one category (almost never vs. generally not) would be weighted 0.25, where as a patient and caregiver that differed by four categories (almost never vs. almost always) would be weighted 1.00. A higher Kw indicates higher agreement (Kw > 0.40 is considered 'moderate agreement'). Analyses were conducted using STATA v.11. Confidence intervals were estimated using bootstrapping techniques with the kapci command.
Initially, we calculated a patient–caregiver Kw for each PACIC item to examine the variability in agreement across items. A lack of significant variability across individual items allowed us to calculate a median patient–caregiver Kw that reflected overall agreement between patients and caregivers across all 20 items. Association between caregiver/patient characteristics and agreement was measured by comparing Kw and the corresponding 95% CI across characteristics. A significant difference was determined by non-overlapping 95% CI. To allow for comparison, all continuous variables (age, SF-36, number of chronic conditions, hours of care and CSI) were converted to binary variables above and below the sample medians.
A weighted kappa is limited by the prevalence of the trait being measured (e.g. a measure of a rare trait is likely to have a low kappa). To account for the possible bias inherent in the kappa statistic, we conducted a sensitivity analysis with polychoric correlations. The polychoric correlation is used when examining correlation across ordinal categories.
Source...