Adults with cardiovascular disease who help others: a prospective study of health outcomes

“[T]he generous will prosper, and those who refresh others will themselves be refreshed.” (Proverbs, 11:25).

Abstract

Little is known about the health impact of helping behaviors among individuals with high-risk chronic diseases such as cardiovascular disease (CVD). Using a nationally representative, longitudinal survey, we examined the subsequent health of adults with CVD (n = 4,491) who spent time providing non-paid assistance to family and friends outside of their households compared with those who had provided no assistance. After both adjusting for baseline characteristics and using propensity score matching methods, spending up to 200 h over the prior 12 months helping others was associated with lower odds of experiencing a new CVD event or dying in the subsequent 2 years. Providing up to 100 h of assistance was associated with reporting fewer depressive symptoms. This threshold effect raises the question of whether assistance beyond a certain number of hours may impose a burden that mitigates health benefits from helping others. Health care providers could play an important role exploring ways that patients with CVD can provide beneficial levels of assistance to others in their own social networks or communities, thereby possibly also improving their own health.

Appendix

Propensity Score Matching (PSM) Methods

We implemented propensity score matching method in selecting respondents with CVD who have similar propensity of helping non-coresident friends and relatives over the prior 12 months in order to address a reviewer’s concern and suggestions corresponding to an earlier version of the study manuscript:

.. it’s also likely that such healthy people are more likely to take on the task of helping others. The longitudinal analysis seems to control for as many of these things as possible, but (as the authors acknowledge) there are likely unaccounted for correlates of these things (e.g., adherence to medication, adherence to other medical recommendations, access to quality health care, blood pressure, etc.) that it is not possible to account for because they were not measured. Normally controlling for covariates is sufficient, but in this case the differences between the helper and non-helper groups at baseline are so dramatic, it doesn’t lead to a very compelling case. I’m unsure what the best way would be to attempt to parse this out, but given the large sample size, there may be ways to provide more compelling evidence for the health benefits of helping others. One possibility would be to attempt a matched control design or sub-analyses within socio-economic status (SES) categories. As it stands, the data suggest that healthier, happier, better educated, and more affluent people are more likely to help others and retain their health later (relative to less healthy, happy, educated, and affluent people).

Using all baseline covariates (gender, age, race, education, poverty ratio, living arrangement in 2002, comorbidity, self-rated-health status, activities of daily living (ADLs), instrumental activities of daily living (IADLs), depressive symptoms (CESD score), smoking, physical activity and drinking behavior in 2002), we obtained propensity scores of providing assistance to non-coresident friends, neighbors and relatives over the prior 12 months in the 2002 survey. Figure 3 below plots a kernel density estimate for propensity scores by helping status in 2002. Kernel density estimation is a non-parametric way of estimating the probability density function of a variable, which approximates the density from observations similar to histograms. A detailed description on kernel density estimation employed in this paper can be found elsewhere (STATA 12 reference manual: kdensity, twoway kdensity).

Fig. 3

Distribution of propensity score by helping non-coresident friends and relatives over the prior 12 months: before matching (Analysis sample 4,490 respondents)

We then implemented one-to-one match based on the nearest neighbor match. Out of 4,490 respondents, we have 1,352 respondents matched (676 not helped in 2002 and 676 helped in 2002). Figure 4 shows the kernel density of propensity score after match.

Fig. 4

Distribution of propensity score by helping non-coresident friends and relatives over the prior 12 months: after matching (Analysis sample 1,352 respondents)