Methodological Issues

As with other areas of empirical study, researchers conducting studies on caregiving need to take into account a series of methodological concerns specific to caregiving. Researchers should take careful note of the need to:

• Review the literature carefully to prevent repetition of areas which have already been well-studied
• Understand methodological trade-offs that affect the feasibility of conducting certain kinds of studies
• Consider key considerations that often compel caregiving researchers to make tough choices about the questions they wish to study

A sufficiently thorough literature review should be undertaken because many areas of caregiving research are already very well developed. It may be difficult to add new and important information to a given topic. For example, there are already many studies documenting cross-sectional associations of caregiver distress with indicators of caregiving stress, personality, and social support among care recipients with dementia, stroke, cancer, and other prevalent chronic conditions.

There are methodological trade-offs that will affect the feasibility of conducting certain kinds of studies. For instance, a project that attempts to identify very homogenous groups of caregivers may have difficulty accruing significant sample sizes. Examples:

• A project that attempts to recruit African American husbands caring for wives in the aftermath of stroke might be difficult to accomplish because of the relatively low life expectancies of African American men, the likelihood that daughters become the family caregiver for African American women, and the relatively low percentage of older African American women who are married.
• A sample of caregivers for persons who have recently incurred a spinal cord injury—a condition of relatively low incidence--might also be difficult to recruit within a brief period of time.
• A sample of parents of children with a rare hereditary cancer syndrome, like Li-Fraumeni Syndrome (about 1000 known cases) would be difficult to accomplish at any single site.

Consequently, researchers must often compromise to recruit sufficient samples (unless they have extensive financial support or can link to a larger project such as the Nurses’ Health Study).

Propensity matching approaches are a useful way to isolate the effect of a potential causal variable when random assignment is not possible. Propensity matching methods are used to be able to compare individuals exposed to a treatment or risk factor (e.g., caregiving) to unexposed individuals that are similar in terms of potential confounding factors. For example, a study might use propensity matching to compare depressive symptoms among caregivers and non-caregivers balanced on age, sex, and marital status. During matching, a propensity score would be calculated for each participant to reflect the probability of the participant being a caregiver based on the set of chosen factors (i.e., age, sex, and marital status). This is done through logistic regression, where group membership is the dependent variable and possible confounds are the predictor variables. Next, caregivers’ and non-caregivers’ propensity scores would be compared, and each caregiver would be paired with one or more non-caregivers with a propensity score closest to them. This process would repeat until each caregiver has been paired with one or more matched non-caregivers, resulting in a sample of caregivers and propensity-matched non-caregivers.

Propensity matching has several benefits. After controlling for systematic differences between comparison groups, we can be more confident that any remaining differences are attributable to the variable of interest. Therefore, propensity matched analyses may yield more precise estimates of the effect of a potential causal variable (e.g., caregiving status). However, there are several important considerations and potential limitations of this method.

One important consideration is selecting one of the various methods for matching based on propensity scores: Optimal matching, greedy matching, etc. An overview of these methods is available onlineopens in new window (see also Austin, 2014). A limitation of the propensity score method is that it is does not accommodate missing data. For example, past research has utilized propensity score matching to examine mortality risk among CGs and non-CGs (Roth et al., 2013) using 15 variables: sex, race, age, geographic location, education, income, marital status, possession of medical insurance, smoking status, alcohol use, results of cognitive screener, self-rated health, and presence of hypertension, diabetes, and cardiovascular disease. Of potential participants, 6.2% were excluded from analysis due to missing data on the outcome of interest or at least one of the 15 propensity-matching covariates.

Another potential concern is that the matching procedure might not yield exact matches in samples with small group samples. Further, there must be enough variability in caregiver and non-caregiver groups to warrant the use of propensity matching (Fredman, Lyons, Cauley, Hochberg, & Applebaum, 2015). Finally, potential confounds that should be addressed through propensity matching may not be represented in the data. For example, when calculating propensity scores (i.e., the probability of being a CG), it is likely important to look at the health status of close others, for whom non-caregivers are potential caregivers. In other words, a non-caregiver should have a family member or friend that requires care for an illness or disability. If this crucial covariate is left out, then the comparison groups would remain unbalanced and results may be biased.

References

Austin, P. C. (2014). A comparison of 12 algorithms for matching on the propensity score. Statistics in Medicine, 33(6), 1057-1069. doi: 10.1002/sim.6004

Fredman, L., Lyons, J. G., Cauley, J. A., Hochberg, M., & Applebaum, K. M. (2015). The relationship between caregiving and mortality after accounting for time-carying caregiver status and addressing the healthy caregiver hypothesis. Journals of Gerontology. Series A: Biological Sciences and Medical Sciences, 70(9), 1163-1168.

Roth, D. L., Haley, W. E., Hovater, M., Perkins, M., Wadley, V. G., & Judd, S. (2013). Family caregiving and all-cause mortality: Findings from a population-based propensity-matched analysis. American Journal of Epidemiology, 178(10), 1571-1578.

Key considerations that often compel caregiving researchers to make tough choices about the questions they wish to study

In an article highly recommended by this APA Task Force, “A future for family care and dementia intervention research? Challenges and strategies,” Zarit and Femia (2008) discuss a number of important research issues which they urge researchers to address in caregiver intervention research.

Here is list of key considerations that often compel caregiving researchers to make tough choices about the questions they wish to study:

• Population-based samples versus clinical or convenience samples
• Defining caregiving
• Relationship of the caregiver to the care-recipient
• Methods of evaluating interventions
• Next best designs
• Inclusion and exclusion criteria
• Measurement challenges specific to caregiving
• The kind of change being studied
• Difficulties in generalizing
• Weighing the potential of a potent intervention package against real-world limitations
• Challenges in studies of caregiving in pediatrics
• Innovative methods

Reference

Zarit, S. H., & Femia, E. E. (2008). A future for family care and dementia intervention research? Challenges and strategies. Aging & Mental Health, 12, 5-13.