Conducting Research with Vulnerable Populations: Cautions and Considerations in Interpreting Outliers in Disparities Research

Addressing the needs of understudied and vulnerable populations first and foremost necessitate correct application and interpretation of research that is designed to understand sources of disparities in healthcare or health systems outcomes. In this brief research report, we discuss some important concerns and considerations in handling “outliers” when conducting disparities-related research. To illustrate these concerns, we use data from our recently completed study that investigated sources of disparities in cancer pain outcomes between African Americans and Whites with cancer-related pain. A choice-based conjoint (CBC) study was conducted to compare preferences for analgesic treatment for cancer pain between African Americans and Whites. Compared to Whites, African Americans were both disproportionately more likely to make pain treatment decisions based on analgesic side-effects and were more likely to have extreme values for the CBC-elicited utilities for analgesic “side-effects.” Our findings raise conceptual and methodological consideration in handling extreme values when conducting disparities-related research. Extreme values or outliers can be caused by random variations, measurement errors, or true heterogeneity in a clinical phenomenon. The researchers should consider: 1) whether systematic patterns of extreme values exist and 2) if systematic patterns of extreme values are consistent with a clinical pattern (e.g., poor management of cancer pain and side-effects in racial/ethnic subgroups as documented by many previous studies). As may be evident, these considerations are particularly important in health disparities research where extreme values may actually represent a clinical reality, such as unequal treatment or disproportionate burden of symptoms in certain subgroups. Approaches to handling outliers, such as non-parametric analyses, log transforming clinically important extreme values, or removing outliers may represent a missed opportunity in understanding a potentially targetable area of intervention.