Table A.

Article Quality Metrics

1) Sufficient sample size. A rough estimate of 100 individuals per group was used as a cutoff for a sample size necessary to attain a reasonable level of precision for inference to population-level estimates of ED use. Smaller sample sizes may be sufficient for testing hypothesis of differences between groups (e.g. people with and without dementia) depending on the effect size. It was noted whether a study included an analysis to determine whether there was sufficient power to detect a difference between groups based on hypothesized or observed effect size.

2) Robust sampling methodology. Probability-based samples were categorized as robust. Multi-site convenience samples were also considered robust. Convenience samples from a single clinic were considered to have a high risk of selection bias.

3) Valid and reliable measure of dementia. Identifying PWD in a study sample poses significant challenges due to the insidious onset of dementia and the lack of reliable physical measures and biomarkers. Studies were considered to have a valid and reliable measure of dementia if they employed some type of participant evaluation, such as neuropsychological testing or at the minimum a brief cognitive screening instrument, and applied a diagnosis of consistent with commonly accepted criteria such as the Diagnostic and Statistical Manual of Mental Disorders (DSM). Identifying PWD through International Classification of Disease (ICD) diagnostic codes in Medicare or other health insurance provider administrative claims has been shown to be subject to misclassification and selection bias. However, this method may have acceptable specificity and sensitivity if certain procedures are followed (Taylor, Fillenbaum, & Ezell, 2002; Taylor, Ostbye, Langa, Weir, & Plassman, 2009). These include a search period of 3 consecutive years’ worth of claims, and the inclusion of both inpatient and outpatient claims. Also, sensitivity and specificity are improved if a broader set of ICD codes is used, in comparison to codes for specific subtypes of dementia.

4) Valid and reliable measure of ED use. Studies that used an objective measure of ED use, such as a health insurance claim or documentation in the medical record, were categorized as having a valid and reliable measure of ED use. The use of self-report to measure ED use was deemed to be at high risk for recall bias. This determination is based on research studies that demonstrate that people both with and without dementia are inaccurate reporters of their own healthcare use, even for relatively infrequent events such as an ED visit (Callahan et al., 2015).

5) Consideration of confounding variables in study design and analysis. Isolating the effect of dementia or particular characteristics of PWD on ED use in observational studies can be enhanced with appropriate study design and statistical techniques, such as propensity score matching or regression, to account for potentially confounding variables. Key confounding variables include, but are not limited to, age, gender, education status, race, comorbidities, and survival or death within the study period.

References

Callahan, C. M., Tu, W., Stump, T. E., Clark, D. O., Unroe, K. T., & Hendrie, H. C. (2015). Errors in self-reports of health services use: impact on alzheimer disease clinical trial designs. Alzheimer Disease and Associated Disorders, 29(1), 75–81. doi: 10.1097/wad.0000000000000048

Taylor, D. H., Jr., Fillenbaum, G. G., & Ezell, M. E. (2002). The accuracy of medicare claims data in identifying Alzheimer’s disease. Journal of Clinical Epidemiology, 55(9), 929–937.

Taylor, D. H., Jr., Ostbye, T., Langa, K. M., Weir, D., & Plassman, B. L. (2009). The accuracy of Medicare claims as an epidemiological tool: the case of dementia revisited. Journal of Alzheimer’s Disease, 17(4), 807–815. doi: 10.3233/jad-2009-1099