By Robert L. Moore, MD, MPH, MBA, Chief Medical Officer
“We didn’t stratify the data by American Indian/Alaska Native, because the numbers were too small.”
– DHCS webinar, reviewing quality data stratified by race
Two years ago, Partnership first stratified Quality Outcome data based on the race/ethnicity we received from DHCS. As noted in prior newsletters, this data showed that outcomes were much worse for the self-identified American Indian/Alaska Native (AI/AN) population than for any other racial group. This prompted Partnership to launch a Tribal Engagement Strategy to build relationships with the 21 Tribal Health Centers and their associated 51 individual tribes, so that we can work together on improving health and wellness for our Tribal communities.
Two months ago, while preparing a presentation for the Medi-Cal Managed Care Advisory Group about Partnership’s Tribal Health Liaison Yolanda Latham, I was looking through race/ethnicity data on our members, and comparing it to the official California Census data, and discovered something very concerning: The number of AI/AN members enrolled in Partnership seemed very low. After a little digging (details below), I discovered that the magnitude of the undercounting is somewhere between 213% and 900%, and maybe even higher.
The reason for this is the way DHCS takes the race/ethnicity/tribal affiliation data from the official Medi-Cal application and uses an algorithm to assign a single race. The Medi-Cal application encourages individuals to choose all races that apply, in accordance with federal recommendations going back to 2000.
Page 4 of the Medi-Cal application:Page 20 of the Medi-Cal application:
The mechanism that DHCS uses to convey membership information to Partnership and other Medi[1]Cal managed care plans is a file called the 834 file or membership file. This file lists just one single race-ethnicity category per enrollee. DHCS uses an algorithm to translate the application race and ethnicity responses to this single category.
While the exact algorithm is not publicly posted, it seems likely that if an AI/AN member also identifies as Hispanic or Latino, this trumped their AI/AN status, and they were assigned a Latino ethnicity. Additionally, if an enrollee identified as both AI/AN and any other racial status, they were classified as “other” or “mixed race,” a category with poor outcomes similar to the AI/AN population, but as it is mixed with all other mixed-race individuals is completely non-actionable.
Here are three mechanisms used to estimate the scope of this undercounting:
- Census Data
One way to estimate the scope of undercounting is to compare the proportion the Medi-Cal enrolled population identified as AI/AN compared to California census data on AI/AN ethnicity.
Official Medi-Cal statistics show a total of 55,302 (only 0.4% of all beneficiaries) AI/AN individuals enrolled in Medi-Cal as of July 2023. (Medi-Cal Fast Facts).
In contrast, in the 2020 census, 1.6% of the California population identified as American Indian and Alaska Native race alone, and an additional 2% of the population identified as American Indian or Alaska Native in combination with some other race, for a total of 3.6% of the population categorized at AI/AN alone or in combination. Even if we assume that the proportion of the AI/AN population of California with Medi-Cal is the same as the non-Medi-Cal population (a highly unlikely assumption), Medi-Cal is undercounting the AI/AN population by as much as nine-fold. Put another way, the true number is 900% higher.
Extrapolating the scope of the undercounting based on census data, as many as 495,000 Medi-Cal beneficiaries would be categorized as AI/AN alone or in combination, instead of just 55,302.
- American Community Survey
An analysis of the 2018 American Community Survey conducted by the National Indian Health Board estimated the California Medi-Cal population to be 242,813. An updated estimate from 2021 put the number at 330,959, or 600% higher than the official state data.
- Tribal Health Centers
Confirmatory evidence of racial mis-categorization comes from the subset of Tribal health centers, which only allow enrolled Tribally-affiliated members to be served. Of those Medi-Cal members served at these Tribal health centers, 53% were categorized by Medi-Cal 834 data as not being AI/AN. Meaning that the true number is 213% greater than the identified AI/AN at Native-run health centers.
Extrapolating this underestimate would mean that the actual number of AI/AN members receiving Medi-Cal is about 118,000 individuals.
Why such a broad range?
The range of undercounting (from 213% to 900%) is so large, partly because the U.S. Census groups together indigenous populations from Central America (such as the Maya and Aztec), South America and Canada into its totals. Of these groups, those who identify as indigenous from Central America are large and growing, resulting in a shift from the Latino category to the indigenous/AI/AN category. In contrast, Indigenous persons from outside of the United States are not generally eligible to receive care at Tribal health centers that are limited to Tribal members.
The American Community Survey assesses race and ethnicity differently, in a way that likely does not include Indigenous individuals from Central America in the AI/AN count, which lowers that count relative to the census estimate.
Impact of Undercounting
Official methods of categorizing race have a centuries-long history of being built on racist assumptions and bias. While I would like to think that the algorithm decisions that led to the undercounting of the AI/AN population in Medi-Cal were not intended to harm the AI/AN population, such large-scale undercounting has several important impacts.
First, it reinforces the perception that American Indians are no longer present in California; “erasure” is the term used by American Indian scholars and activists. In fact, in the past century, erasure was an official U.S. government policy, as tribes were “terminated” in the 1950s and 1960s, children kidnapped and taken away to boarding schools to indoctrinate them into American culture. The residual evidence of erasure reflects a lack of acknowledgment and sensitivity of this historical trauma.
Second, such profoundly faulty data leads to faulty analysis of health inequities. If the racial data used to calculate rates of quality indicators is biased and faulty, then the inferences drawn by stratifying data by race are hints of the underlying reality, but any sanctions or penalties tied to reducing such inequities by any specified quantity are statistically invalid.
Lastly, such significant undercounting impacts public health prioritization based on population affected, and thus potentially impacts funding allocated proportional to the AI/AN population affected.
What should be done?
Major Tribal organizations representing health and public health policy issues have raised the problematic nature of categorization of AI/AN persons in multiple settings and give input into the newly updated 2024 OMB standards.
National organizations, especially the National Indian Health Board, have raised the issue of data incompleteness and undercounting. Some shorthand terms for the lack of sharing of accurate data about the AI/AN population is “data sovereignty” and the need to “decolonize data systems.” The National Council on Urban Indian Health issued an analysis of undercounting among Urban Indians. Other organizations that have weighed in on undercounting of AI/AN population data include the 12 regional Tribal Epidemiology Centers, and the state Tribal health organizations like the California Rural Indian Health Board.
Major changes in the new U.S. Office of Management and Budget (OMB) Standards
The Updated 2024 OMB Standards for categorizing race/ethnicity move Latino/Hispanic to be a co-equal race/ethnicity category, instead of a carved-out ethnicity category. The Middle-eastern/north African population was carved out of the White category, so there will now be 7 major race/ethnicity categories. One of which is American Indian or Alaska Native, with a box to fill in details with the following language: “Enter, for example, Navajo Nation, Blackfeet Tribe of the Blackfeet Indian Reservation of Montana, Native Village of Barrow Inupiat Traditional Government, Nome Eskimo Community, Aztec, Maya etc.”
The most concerning aspect of the new OMB standard is the list of options for handling individuals who identify more than one race/ethnicity category. The three options identified are (see page 22195):
- The “alone or in combination” approach mentioned earlier related to census data. There is some complexity to using this approach, but it substantially resolves the undercounting of the AI/AN population and should be the starting point of data sharing and equity analysis. A key feature of this approach is that the total of all categories is greater than 100%, as one individual maybe two or more categories; this requires special statistical methods to avoid errors.
- The “most frequent multiple responses” approach, in which the top combined categories are each presented with individual data. For example, in addition to each race ethnicity category alone, each combination is listed with the number of individuals. Some may be simple two-race categories (like Black-Asian), but more complex combinations are possible (like Latino-Black[1]White). This allows the most granular data analysis, and the numbers can be folded into the “alone or in combination” category. The sum of all individuals in all categories will total 100%.
- The “multiracial” approach in which any individual who chooses more than one race/ethnicity category is categorized as either “other” or “mixed.” This grouped category is impossible to analyze, so the “pure” race/ethnicity categories end up being the only way to look for health disparities. This appears to be the method currently used by DHCS, and it should be abandoned as soon as possible.
What Can DHCS Do Now?
First and foremost, DCHS should share the current detailed enrollment race/ethnicity/tribal affiliation data with all Medi-Cal Managed Care plans so they can better analyze and understand the inequities faced by their members. This could be done with a separate monthly report from DHCS and it could also be integrated into the new Medi-Cal Connect platform that DHCS is building to feed assorted supplemental data to health plans. In addition, if DHCS has separate member-level internal flags indicating Tribal affiliation or AI/AN status, from other sources, this should also be conveyed to the plans with the more complete enrollment demographic data.
This granular race/ethnicity/Tribal affiliation data will allow managed care plans to re-run our disparity analyses and release an analysis of our findings. In addition, we can pass on this information to primary care practices to give them the complete and accurate data they need to identify and address health inequities.
As DHCS plans its implementation of the new OMB race-ethnicity standards, they should convene a workgroup with representatives from the California Tribal Epidemiology Center, the California Department of Public Health, the California Rural Indian Board, the California Consortium of Urban Indian Health, and Region IX of US HHS to review the options for categorization of data, strongly considering either the “alone or in combination” approach or the “most frequent multiple responses” approach, which can be combined to create “alone or in combination” groups. These two approaches would stop the undercounting of the AI/AN population.
Finally, to stop presenting incomplete and inaccurate data about the AI/AN population, DHCS should create an internal team to review all presentations of data that is stratified by race/ethnicity to identify, correct and/or put into context the data as it relates to American Indian population. This team should be empowered to raise concerns anonymously to the DHCS Chief Health Equity officer if their concerns are not addressed.
As unintentional as it may be, the DHCS racial categorization algorithm is an example of structural racism that deserves to be addressed. With the increased emphasis on Health Equity at DHCS and CDPH, there should be a heightened sense of urgency to definitively address this issue. DHCS alignment with the OMB’s updated race and ethnicity data standards creates an opportunity to correct an issue that obscures Tribal communities and other small populations from the data.