So now let's explore an example of mediation. So one example of a mediation analysis is a study we conducted in 2012 entitled Racial and Ethnic Differences and Determinants of Live Donor Kidney Transplantation in the United States. The primary objective of this study was to quantify the extent to which clinical versus socioeconomic factors contribute to racial and ethnic differences in access to live-donor kidney transplantation in the US. Our study question was motivated by a prior systematic review that we completed to identify published evidence of barriers to live-donor kidney transplantation among US racial and ethnic minorities. Based upon the results of the prior systematic review, we develop an evidence-based model, which is shown here on the slide that contextualizes reported barriers that racial and ethnic minorities may face when trying to get a live-donor kidney transplant. Now this model summarizes multilevel factors that may be encountered by potential transplant recipients and donors, such as unmet concerns, economic constraints, or lack of clinical suitability, or barriers at the healthcare provider level, such as communication, provider training, and knowledge, or beliefs and perceptions about patient's suitability for a transplant. There are barriers at the health system level, which include insurance coverage or information quality within healthcare systems. Then also at the population or community level, such as high chronic disease burden in certain neighborhoods or communities, and then geographic constraints that all may work together to contribute to racial and ethnic differences in completion of the live donor kidney transplant process. Additionally, minorities may experience these barriers doing one or more of the four primary steps of the live-donor kidney transplantation process, including during donor identification, transplant evaluation, kidney transplant surgery, or post-transplant recovery. We wanted to know which of these factors should be targeted, in future interventions to make the biggest difference in addressing these disparities. Therefore, we performed a retrospective cohort study of all US adults age 18-70 years old who were registered as having started treatment for end-stage renal disease between June 30th, 2005 and September 23rd, 2008, using patient data link together from the United States Renal Data System Registry, the 2005 centers for Medicare and Medicaid services medical evidence form, and the United Network for Organ Sharing, kidney transplant file. We excluded patients who were missing race or ethnicity or who identified as having cancer at the time of end-stage renal disease. We then obtain socioeconomic data from the 2000 US census by linking the patient files at the five digit zip code level with the US Census data. A total of 208,736 subjects were included in our study. Our primary study outcome was time from end-stage renal disease onset to receipt of a first live-donor kidney transplant by the patient's race or ethnicity. We also examined the contribution of clinical factors such as health insurance, predialysis nephrology care, chronic conditions such as diabetes or obesity, and active substance use, such as drug or alcohol abuse, and socioeconomic factors, including whether or not the patients lived in an area that was defined as a high-poverty area by the US Census, or in other words, an area where at least 20 percent of the residents were poor. Or in addition to these, patients could be living in an area where at least 20 percent of the households are linguistically isolated. Defined by the US Census as all household members age 14 and older who primarily speak another language and who speak English less well. We analyze the associations of race and ethnicity with time to a live-donor kidney transplant, using cox proportional hazards models. We censored patients at the time of death, receipt of a deceased-donor kidney transplant, or the end-of-the-study period. We then calculated the proportion of the reduced rate of live donor kidney transplantation due to clinical versus SES factors by adding beings factors to a base model of fixed demographic variables, including age, sex, and race. In calculating the difference and the log hazard ratios by race and ethnicity, we also performed bootstrap analysis to calculate the bias corrected confidence intervals around these estimates. As a result, we found that the degree to which reduce rates of live donor kidney transplantation among minorities was due to adjustment for measured factors vary by their race and ethnicity. As shown on this bar graph, the largest proportions of the disparity among Blacks, Pacific Islanders, and American Indian Alaskan Natives were due to differences in predialysis Nephrology tier. While the largest proportion of the disparity among Hispanics was due to differences in health insurance. Zip code level neighborhood poverty also contributed to a substantial proportion of the disparity among Blacks and American Indian, Alaskan natives. Unfortunately, few of these factors explain the disparities that we found among Asian. Another analytical method that we use in social epidemiology research is testing for effect modification. Effect modifiers are different from confounders and mediators because they are not necessarily direct or indirect causes of the outcome. They are also not caused by the exposure. We typically test for effect modifiers by using interaction terms or stratification. An example study question, would be, is the relationship between race, ethnicity, and treatment outcomes different for people living in wealthy neighborhoods versus those living in poor neighborhoods? So in other words, it's a disparity between black and white patients living in wealthy neighborhoods the same as the disparity between black and white patients living in poor neighborhoods. In this example, the effect modifier would be neighborhood level poverty, wealthy versus poor if we found that there are differences in that racial and ethnic disparity. So collectively, all of these methods can be used to inform health equity interventions. But let's be real, this is not new information. As shown on this timeline, disparities have been documented since at least the 1970s, and many existing interventions have been designed to achieve health care and health equity. However, many of these fall short because of gaps in knowledge and translation. Understanding these gaps could guide the development of feature interventions and policies to achieve health equity. Finally, this figure is an ecological model that we adapted from the work of Fisher and colleagues. Here, we highlight patient, family, organization and provider, and then also policy and community influences on health care disparities. In the second column, we include potential intervention targets at each of these levels. For example, organizational climate at Level 3 versus targeting patient education at Level 1. In the third and fourth columns, we represent key stakeholder interactions in outcomes that are affected by these multilevel influences, interactions, and intervention. In other words, we haven't made more progress because all of this is very complex. In a paper that we published in Health Affairs in 2016, we identify 15 critical knowledge in translation gaps, which are highlighted on the figure. We organize these gaps by their target intervention levels, which also allied with the four levels in our ecological model. Four gaps exist at all levels, and the remaining 11 gaps exist at one or more levels of the model. These are examples of the types of gaps that should be addressed by feature health equity intervention. Several of these gaps are related to addressing social determinants of health. Many disparities in health are rooted in an equities and their opportunities and resources needed to be healthier. These determinants of health include living and working conditions, education, income, neighborhood characteristics, social inclusion, and medical care, and these and other examples are included on the figure.
