Karthik Sivashanker, MD: [00:25] We're going to start just very basic, in a way. And I think that's actually an important and good thing to do partly because we have folks who are across the spectrum in terms of data, knowledge, and experience, and also because it's a really complicated topic, and there's a lot of different places we can go. So we want to start with the essential stuff first, and then work our way from there.
[00:49] So I just want to start off with some of the challenges that we're going to be facing. And to be clear, there are many, many more challenges; I'm just highlighting a couple. But one of them is that there's a lack of a clear and sufficient federal standard around this in terms of race, ethnicity, language, and other demographic or social identity data collection. And that's also complicated by the fact that race and ethnicity, as you may know, are social constructs. And put another way, they're completely made up, they don't have a strong association to genetic or biologic ancestry, and should not really be used as proxies for biologic or genetic ancestry at all. And they have been misused in that way. For example, in the case of EGFR, or other race corrections. And so you see, on the right side is an article, this is by Michelle Morris and colleagues, I would really recommend taking a look at that, because this is a nuanced and complicated subject and basically, this article talks about how we should and should not be using race/ethnicity data as a risk factor, because it's really important that we're collecting the data and using the data, that we're using it responsibly, and that we're not using it to perpetuate any harms.
[02:06] So just to make this as simple as possible, race and ethnicity, as social constructs are useful proxies for racism, meaning, how you look, the advantages you've been given in life, largely determine the type of treatment you're going to get or are definitely shaped the type of treatment you're going to get, which also shapes the outcomes you're going to get. And we should not be using them as proxies for biologic or genetic ancestry, as in the example of race corrections that have been really problematic. And as we're going through this, you know, I realized, and I'm going to say this, again, that this is a really complicated topic, and there's just so much to cover. We're not going to cover everything today. We're going to cover some basic stuff.
[02:50] The second big challenge is going to be misconceptions and fears. So this is a broader issue than just the work we're doing. So this is, you know, staff, whether it's providers, or more often registration staff who are reluctant to ask patients in collecting this information, which can impact the quality of the data that we are using. And then also, patients and communities may be reluctant. For example, we just talked about how race and ethnicity have been misused in, especially race has been misused, in medical algorithms to determine who does or does not get resources or services. And so patients if they know about that might be reluctant. They also may just have a general mistrust of the health care system. We have a long history in health care of exploitation of historically marginalized communities, and I'm not going to go into all the different examples but COVID-19 really highlighted how much mistrust there is. So we have a lot of work to do to earn the trust of patients and communities, and to earn the trust that we're going to use their data appropriately and safely and to improve their care, not to hurt them.
[03:57] And the other challenge is that there's no federal limitations around this or really guidance around this at the moment. So if we go to the next slide. A good first step is going to be just to assess the quality of your data. So I'm offering three prompts here to think about. First: is the data complete? And we know pretty well that across most health systems, when we're talking about race, ethnicity, language, data, etc. it's pretty incomplete. It's also not very accurate. And depending on who we're asking, you know, the population, it may or may not be more or less accurate. And then finally, is it self reported? So there's potentially challenges with all of this on average, and we've included a couple of references here, a lot of systems are hovering somewhere between 50 to 70% completeness and accuracy, and there's oftentimes problems in terms of is it self reported or not? Sometimes providers will assign, or staff will assign, race or ethnicity to patients. And it's always going to be better if it's self reported. So a very simple basic step is going to just be looking at those variables that we're going to cover in the next couple of slides, and asking yourself, how complete is it? Is it accurate? Is it self reported? And we're going to need to do this work at an enterprise or system level, to oftentimes, so we'll talk more about that. But let's go to the next slide.
[05:38] Another really simple next step is just going to be understanding your process for prospective data extraction. So we want you to start by looking at your harm event data. So we're talking about your safety events, we're talking about your patient complaints. And we want you to start looking at events from now moving forward. And what we want to do is think about: are we going to pull this data into our trackers manually? So this is going to be where the program manager on your team will just literally look in the chart, look in your electronic health record, to just pull that data or wherever else that may be located, versus some sort of automated process. So for example, for us, RL Solutions can provide some of this information in an automated way, versus some hybrid approach. But the key thing here is we don't want you to wait for the perfect process to begin to start the work, because it's very easy when we're talking about data to get into a trap of, we need to automate everything before we do this, or we need to build the perfect dashboard before we start doing the work. And actually, a lot of this work can be done very low tech, you know, we're talking Excel spreadsheet and just having a person look in the chart and pull the information into your trackers. It does not take a lot of time, and is a good place to start. Next slide.
[07:01] So this question is going to come up almost immediately. It's which variables should we be tracking? And I'm going to point you back to the original question of how complete is your data? How accurate is your data? And then also, how are you pulling the data in? The variables that we landed on at Brigham and Women's was race, ethnicity, language, age, sex—I'm being explicit and using that differently than for example, gender or gender identity—insurance type, disability, and then you see some in italics: sexual orientation, gender identity, and then as optional other variables you may want to collect. So why did we pick the ones that are not in italics from race down to disability, and it gets back to completeness and accuracy, for the most part. We just had enough data, enough usable data, to include those. Whereas in the case of sexual orientation and gender identity, the data was so incomplete, that we need to first do more work at an enterprise level or a full system level, to actually improve that data collection and train up our registration staff to do that, before we can reasonably use it. And when I say, you know, complete and accurate, what I'm talking about is, you know, once again, most systems are hovering around 60 to 70%. So we're not saying perfect, we understand that this data is far from perfect. But we also don't want to, once again, wait for perfection to begin doing the work. And even, you know, with that 60, 70% accuracy, and completeness, it's still quite usable, oftentimes. and helpful. Next slide.
[08:48] So very simple, once again, low tech, you're just going to start incorporating this into your trackers. So wherever you're tracking your safety data, or your patient complaints, all of those harm events, we're just going to start to add in the variables that we're going to track: gender, race, ethnicity, insurance, language, disability. We can see all of that here. How you decide to categorize some of these variables is a decision, once again. So when you say insurance, you know, you might decide that private versus Medicare versus Medicaid versus self-pay, you know, you can think about that. We can certainly share more information in one of our upcoming sessions about how we collapse some of these categories. And you're just going to embed that into your tracker and start prospectively pulling that data for your cases. Next slide.
[09:45] In parallel—and this was really important that we initiate this request sooner than later, because this can take some time, at least our experience has been that way—you want to initiate a request to your data team to extract and update your prior 12 months of safety data. And so we're talking once again about our safety events and patient complaint data, and we want to take that retrospective data and stratify it by the same variables.
Credit Designation Statement: The American Medical Association designates this enduring material activity for a maximum of 0.25 AMA PRA Category 1 Credit(s)™. Physicians should claim only the credit commensurate with the extent of their participation in the activity.
CME Disclosure Statement: Unless noted, all individuals in control of content reported no relevant financial relationships.
If applicable, all relevant financial relationships have been mitigated.