Family Caregiving and All-Cause Mortality
Family Caregiving and All-Cause Mortality
Table 1 summarizes the descriptive comparisons between the 3,503 caregivers and the noncaregivers in the REGARDS Study. Prior to matching, caregivers differed significantly from noncaregivers on 12 of the 15 covariates. Caregivers were younger, on average, and more likely to be women, African American, and married. Caregivers were less likely to have health insurance and to report a history of cardiovascular disease. Subtle but statistically significant differences were also observed for education, income, smoking status, and alcohol use. After propensity matching, the 3,503 caregivers did not differ significantly from their 3,503 matched noncaregivers on any of the 15 covariates, confirming the success of the binary logistic regression and greedy matching procedure for identifying balanced groups of caregivers and matched noncaregivers for further analysis.
Figure 1 displays the descriptive survival curves for the 3,503 caregivers, for all of the 24,863 noncaregivers, and for the 3,503 propensity-matched noncaregivers. Of the 3,503 caregivers, 264 (7.5%) died during the follow-up period, whereas 2,782 of the 24,863 noncaregivers (11.2%) died during this same period. After propensity matching, 315 of the 3,503 matched noncaregivers were deceased (9.0%), which was a significantly greater proportion than the 7.5% of caregivers according to a simple χ test (P = 0.0269). The Cox proportional hazards analysis revealed that caregivers died at approximately an 18% lower rate than their individually matched noncaregivers over this 6-year period (hazard ratio = 0.823, 95% confidence interval: 0.699, 0.969; P = 0.0196).
(Enlarge Image)
Figure 1.
Survival curves for caregivers (black line, n = 3,503), propensity-matched noncaregivers (gray line, n = 3,503), and all noncaregivers (dotted line, n = 24,863) from the REGARDS Study over the 8 years of follow-up after enrollment, 2004–2012.
The sample of 3,503 caregivers included many different subgroups identified by race, sex, caregiving relationship, perceived caregiving strain, and amount of caregiving involvement. Table 2 summarizes the results of the subgroup analyses that were conducted. In each analysis, specific caregivers were individually matched with qualified potential noncaregiving controls by using a new logistic regression and propensity score matching procedure. In all cases, the propensity matching procedure was effective for balancing the caregiver and noncaregiver groups on the relevant covariates. All P values were greater than 0.12, and 170 of the 174 possible covariate comparisons resulted in P values greater than 0.20.
The results of the Cox proportional hazards models identified 1 subgroup of caregivers with a significantly lower death rate. Adult child caregivers who were providing care to a parent were found to have a significantly lower rate of death compared to their propensity-matched noncaregivers (P = 0.0064). In addition, trends that approached conventional levels of statistical significance were observed for white caregivers (P = 0.0791), female caregivers (P = 0.0703), and caregivers who provided 14 or more hours of care per week (P = 0.0752). The hazard ratio for each of these subgroups was similar to the hazard ratio for all caregivers, but the hazard ratios for the subgroups were no longer statistically significant at the P < 0.05 level because of reduced sample sizes and power. No subgroup of caregivers showed a trend for increased risk of death compared with propensity-matched noncaregivers. The strained spouse caregiving subgroup, which included spouses who reported either moderate or high caregiving strain, was similar to the spouse caregivers found to have an elevated rate of death in the CHES.
Results
Propensity Matching
Table 1 summarizes the descriptive comparisons between the 3,503 caregivers and the noncaregivers in the REGARDS Study. Prior to matching, caregivers differed significantly from noncaregivers on 12 of the 15 covariates. Caregivers were younger, on average, and more likely to be women, African American, and married. Caregivers were less likely to have health insurance and to report a history of cardiovascular disease. Subtle but statistically significant differences were also observed for education, income, smoking status, and alcohol use. After propensity matching, the 3,503 caregivers did not differ significantly from their 3,503 matched noncaregivers on any of the 15 covariates, confirming the success of the binary logistic regression and greedy matching procedure for identifying balanced groups of caregivers and matched noncaregivers for further analysis.
Mortality Effects Across All Caregivers
Figure 1 displays the descriptive survival curves for the 3,503 caregivers, for all of the 24,863 noncaregivers, and for the 3,503 propensity-matched noncaregivers. Of the 3,503 caregivers, 264 (7.5%) died during the follow-up period, whereas 2,782 of the 24,863 noncaregivers (11.2%) died during this same period. After propensity matching, 315 of the 3,503 matched noncaregivers were deceased (9.0%), which was a significantly greater proportion than the 7.5% of caregivers according to a simple χ test (P = 0.0269). The Cox proportional hazards analysis revealed that caregivers died at approximately an 18% lower rate than their individually matched noncaregivers over this 6-year period (hazard ratio = 0.823, 95% confidence interval: 0.699, 0.969; P = 0.0196).
(Enlarge Image)
Figure 1.
Survival curves for caregivers (black line, n = 3,503), propensity-matched noncaregivers (gray line, n = 3,503), and all noncaregivers (dotted line, n = 24,863) from the REGARDS Study over the 8 years of follow-up after enrollment, 2004–2012.
Mortality and Caregiving Subgroups
The sample of 3,503 caregivers included many different subgroups identified by race, sex, caregiving relationship, perceived caregiving strain, and amount of caregiving involvement. Table 2 summarizes the results of the subgroup analyses that were conducted. In each analysis, specific caregivers were individually matched with qualified potential noncaregiving controls by using a new logistic regression and propensity score matching procedure. In all cases, the propensity matching procedure was effective for balancing the caregiver and noncaregiver groups on the relevant covariates. All P values were greater than 0.12, and 170 of the 174 possible covariate comparisons resulted in P values greater than 0.20.
The results of the Cox proportional hazards models identified 1 subgroup of caregivers with a significantly lower death rate. Adult child caregivers who were providing care to a parent were found to have a significantly lower rate of death compared to their propensity-matched noncaregivers (P = 0.0064). In addition, trends that approached conventional levels of statistical significance were observed for white caregivers (P = 0.0791), female caregivers (P = 0.0703), and caregivers who provided 14 or more hours of care per week (P = 0.0752). The hazard ratio for each of these subgroups was similar to the hazard ratio for all caregivers, but the hazard ratios for the subgroups were no longer statistically significant at the P < 0.05 level because of reduced sample sizes and power. No subgroup of caregivers showed a trend for increased risk of death compared with propensity-matched noncaregivers. The strained spouse caregiving subgroup, which included spouses who reported either moderate or high caregiving strain, was similar to the spouse caregivers found to have an elevated rate of death in the CHES.
