This transcript was automatically generated using speech recognition technology and may differ from the original audio. In citing or otherwise referring to the contents of this podcast, please ensure that you are quoting the recorded audio rather than this transcript.
Gavin: Hello, welcome to a special episode of The Lancet Voice with me, Gavin Cleaver, and with my co host, Jessamy Bagnall. Over the last year, there have been many reports on the United Kingdom's striking disparities in COVID 19 infection, hospitalisation, and mortality rates among different minority ethnic groups.
OpenSafely, a data platform which has access to many millions of patients full pseudonymised NHS health records, has a research paper out today in The Lancet. It analyses more than 17 million adults in England, and it demonstrates the poorer outcomes for minority ethnic groups over the course of the pandemic.
To find out the story behind the results, first, Jessamy spoke with lead author Dr Rohini Maver of the London School of Hygiene and Tropical Medicine, and then I spoke with OpenSafely's director Dr Ben Goldacre to find out more about the platform and what it's being used for.
Jessamy: So Rohini, perhaps you might just walk us through some of the kind of basic methodology and the main highlighted results for this study that looks at ethnic disparities and COVID 19 in the UK.
Rohini: Sure, so this was an observational cohort study using the OpenSafely data platform. And at the time the study was conducted, we had access to the primary care health records for 40 percent of the English population. And we linked these records to data on SARS CoV 2 infections. COVID 19 related hospital admissions, intensive care admissions, and deaths.
And in primary care records ethnicity data are captured at the time of registration or consultation, and it's self reported by individuals. And for analysis purposes, we aggregate these individually reported ethnicity data into the standard categories of the census. And in our analysis, we looked at ethnic disparities in infection, hospitalization, and death over two time periods corresponding to, I would say, the two waves of the pandemic.
So we looked firstly from February to August 2020 for wave one, and then secondly from September to December 2020 for wave two. And we looked at the ethnic differences after accounting for a number of other potential explanatory factors. So these included Sociodemographic characteristics, health related factors, region, and household size.
And we found that in the first wave, the risks of all outcomes, so testing positive, hospitalization, ICU admission, and death, were increased across all ethnic minority groups in comparison to white groups. And this persisted even after accounting for all of the other known explanatory factors. And in the second wave, we found that the disparities were increased in size between South Asian and white groups, but then reduced slightly between black and white groups.
So we saw a different patterning of ethnic disparities across the two waves. And because of the large sample size that we had in OpenSafely, we were also able to look at differences between the more detailed ethnic group categories, which revealed important heterogeneity. So for example, in the first wave in the black ethnic group, the risk of being admitted to ICU was twice as high for Caribbean groups, but four times as high for African groups compared to the white British population.
So that was also very useful to be able to describe and delineate important. population differences in that way as well.
Jessamy: And Rohini, just to go off topic slightly, because we had Angela Saini with us recently, and she was talking, we had a discussion about race and, how these sort of groups are very much social constructs.
And what's interesting about your study is that it doesn't just look at the main one, it looks at kind of 16, right, different, much more, from your perspective, you're obviously a researcher in race and inequality, equity. How much meaning can we take from these different categories, which we've put onto society ourselves?
Do you know what I mean? I'm just interested to hear your perspective.
Rohini: Yeah, absolutely. So I would say there's nothing inherently true, about these categories. These are just tools we have to help us identify disparities and identify population groups who have different kinds of experiences and outcomes.
And, what it means to be a South Asian British woman now, like I am is probably very different to what it was 20 years ago, 50 years ago. So these categories are very transient and they really need to be interpreted in context. And really they're just a useful lens one tool in our toolkit with which to identify disparities.
Jessamy: And do we know that the people who are inputting the data into the databases are normally accurate in documenting people's race, like the difference between Caribbean and African could be misinterpreted or not understood by, a physician or a healthcare worker that was putting it in.
What's the sort of, what are the issues around that?
Rohini: Yeah, so it's a really good question. So quite often when people are asked to report their ethnicity, and you may have gone through this yourself, you may be given a form with the standard categories and you just. tick the one you think applies to you.
And, on one hand people might find that, people might feel that their ethnic identity changes over the life course as they accumulate different experiences. There's nothing to say that your ethnicity is necessarily stable over time. But then there also may be issues with how people interpret those categories.
In the electronic health records that we have, I think we have over 120 codes. for ethnicity, and we group them into the census categories ourselves. And, the people who gave us that data may not necessarily agree with where we've put their ethnicity, and we do the best that we can, but there is always some kind of pragmatic consideration that we group people together in a way that maybe does not reflect the complexity of their identity.
So it's a real problem. And different research groups do this in different ways. And one of the nice things about OpenSafely is that all of our methods are open and reusable for other research groups. So they can see what we've done. They can replicate it if they agree with it, they can build on it or change it.
And we can document all of those changes. So there's like a clear history of how these categories have been made and changed over time.
Jessamy: Exactly. And I suppose Broadly though, the message is that there are these huge disparities in COVID 19 outcomes in this country. What do we know about data elsewhere in other countries, and what are we really seeing here?
Rohini: Something that was very interesting to me when I first started my research career looking at ethnicity is that it's actually illegal to capture ethnicity data in a number of countries, so particularly in Europe, it's illegal to ask people about their ethnicity or their origins in Germany and France, so in a lot of Other high income countries, there aren't actually good quality data on ethnic disparities, which means there's no way to measure differences in social inequalities and racism because these data just don't exist.
And there's a stigma around capturing them. So a lot of the comparative data we have are from countries like the USA or Canada, where these data are part of regular national data collection. And we see really strikingly similar findings across these settings that ethnic minority groups have higher rates of infection, severe disease, and mortality than the majority population and that these disparities are related very strongly to differences in social determinants of health, such as income, housing, and occupation.
So a very consistent story across settings where these data are available.
Jessamy: And so what we're seeing there is a sort of the structural biases and the structural problems within our societies that are then being translated to COVID 19 outcomes. And we're seeing it more starkly with COVID 19 because it's such a huge problem.
Rohini: Absolutely. So these the drivers of ethnic inequalities and COVID 19 are by no means unique. And I think these are drivers of all sorts of different health and social outcomes. And so these wider social and structural determinants are very well documented, well known. It's just that the pandemic has brought them into such sharp relief that there is this heightened awareness.
of the disproportionate impact of these social structural factors on ethnic inequalities. And by, I think, by addressing These social determinants, you will inadvertently, and usefully reduce ethnic disparities because they're so related.
Jessamy: And it's speculation. You, and we can only speculate, but the difference between the first wave and the second wave and in some groups things getting worse.
What might be the causes for that sort of just. Hypothesizing, we obviously can't have the evidence behind it, and it can't really be backed up, but
Rohini: Yeah electronic health records can only take us so far in terms of the data they capture. But the ONS did some useful investigation into this, because they've got a much broader data set.
And one of the hypotheses that they put out is that in the first wave the risk of infection outcomes was largely driven by geography, so hotspots of infection. And it was that ethnic minority groups tended to live in these hotspots, which if it's understandable if you think that these were urban, highly dense densely populated city areas.
And in the second wave, what we found is that the the higher rate of infection outcomes in South Asian groups relative to white groups may be more explained by differences in exposure to infection. So how does infection come into the household? And is it related to your occupation? Are, are some ethnic groups more likely to work in occupations where they're unable to work from home or unable to properly socially distance?
We know that South Asian groups in particular tend to live in larger multi generational, highly connected household. And while this is excellent for kind of social community support and cohesion, and particularly now with vaccine rollout it does increase your proximity to other people physically.
And now, we know that COVID 19 is airborne. And so actually not being able to maintain distance in the household and the community and in the workplace are probably what's driving the higher rates in South Asian groups because of the unequal distribution of. deprivation housing and employment related factors.
Jessamy: Now we've been able to see these. Disparities which have always been there, but perhaps, we haven't been talking about them to the same degree. As a sort of inequality equity researcher, what are the types of policies that you would like to see now being thought about, being instituted, so that we can disentangle some of this data, understand what the really, the crucial driving forces are, and how we can overcome them.
Rohini: Absolutely. I think it's a really important question and ethnicity is such a, an aggregate of all these other kinds of social exposures and determinants. And so I think, particularly in terms of the pandemic, but also more broadly, having policies which really address things like income inequality, housing quality and workplace protection and rights.
On a population level we'll really improve things across the board and also highlight whether these things can also then reduce ethnic disparities. And I hypothesize that if we address these things broadly, we will see ethnic disparities in health related outcomes and other areas reduced.
And if they don't, then that will give us a really good pointer as to what are we missing, where are the gaps that we need to focus our resources on. But I think, yeah, more widely. Things about poverty reduction and employment protections and things like that will go a long way to Reducing a lot of these problems
Jessamy: indeed and I mean on the news this morning even there's that report about ethnic minorities being disproportionately impacted by job losses which again feeds into all of these gaps that are being widened.
And it's possible to think that things might get worse before they get better. As we see the fallout of the pandemic in this country where we have the NHS and people can access healthcare fairly. easily and well. Is equity to health care access such an issue? I know that you've worked in Canada before.
And, we talked about the states and Canada in terms of the data there. What do you expect to see with those two sort of comparisons in terms of equitable access to health care and then the impact of COVID 19?
Rohini: Yeah. In the UK, we're so lucky that healthcare is free at the point of access, which is completely in contrast to the USA, where it's so much tied to your income and appointment and your health insurance.
And most of the work that I've done before COVID is looking at ethnic inequalities in other health related outcomes, particularly diabetes. And so part of that work has been looking at all the ethnic differences in kind of identification and prevention of disease, which is. related to access, and we find that there aren't massive differences in who accesses the GP.
So there aren't disparities in when people are necessarily diagnosed or identified, and I think healthcare practitioners are quite proactive. With looking at health, identifying health conditions in ethnic minority groups because it's part of the national guidelines. So there's a good evidence base already that ethnic minority groups have higher burden of, particularly cardiovascular and long term conditions.
But I think beyond that, there are difficulties in access when it comes to things such as language or having high quality consultations. And we know that, for the same health condition, someone who doesn't speak English or is from an ethnic minority background may require more consultations to get to the same diagnosis or the same level of care.
So I think that's where the inequality comes in. It's that interaction and the the healthcare infrastructure to adequately identify and tackle what we know are potentially higher risks in these groups.
Jessamy: That's very interesting, Rohini. Thank you. And I know before the interview you said that you might want to just touch on vaccine rollout and the ethnic issues around there and it would be great to hear your views on that.
Rohini: Yeah, so this is related, to your question about kind of the policies and next actions, and I think I just want to say that there's already been a tremendous amount of mobilization and work around reaching out to ethnic minority communities around improving uptake of the vaccine and overcoming, concerns and vaccine hesitancy.
So I think, over the course of the pandemic, there has been such raising of awareness that we are now hopefully reaping the benefits of this targeted action. And I think it's going a long way to improving vaccine uptake rates across the country, which I'm sure Ben can speak more about.
Ben: Yeah we've actually been tracking vaccine uptake using OpenSafely. Because one of the things that's quite unusual about the platform that we've built is we've got access to very rich detailed information about patients in near real time. And so we've been producing graphs and tables on a weekly basis showing exactly who is and isn't getting vaccinated in England.
And not just small demographic subgroups. We can look at different groups of people with different diseases. Looking at that, we've been able to see first of all, among the over 80s, a very substantial disparity between different ethnic groups. So black people are almost half as likely to have been vaccinated as white people in the over 80s group.
Or at least they were initially when the vaccine program was first being rolled out in December, January, February, and they've caught up now. So they're now perhaps about a third less likely. to have been vaccinated. And that's of course consistent with what we know from previous vaccination programs for previous diseases, that there are very striking disparities between different ethnic groups.
Interestingly, usually Bangladeshi people have quite good vaccination rates that wasn't seen during COVID, but just. Very recently over the course of the last month, we've seen vaccination coverage amongst Bangladeshi people absolutely rocket up overtaking nearly everybody else. So there's clearly some targeted action happening there.
And all of those findings around ethnic disparities are consistent with some of the survey data that we've seen about vaccine hesitancy during COVID. And as I said, we've also been able to look at. things like vaccination rates amongst learning disabilities people vaccination rates amongst people with learning disabilities and among people with severe mental illness, where again, we've seen lower rates of vaccine coverage, but amongst people with chest problems, cardiovascular problems broadly speaking, similar vaccination rates to general population.
Gavin: Ben you first appeared on this podcast last year and you spoke with Richard Houghton at the time about your kind of commitment to open science, and you're looking forward to OpenSafely, and obviously it's been a kind of huge 12 months for you now, I see you on Twitter all the time, recruiting for OpenSafely, it seems like it's going very well, you're expanding rapidly.
Where do you hope to take it over the next few years? I personally
Ben: have no ambition to run an organization with 300 staff. We're here to we, we seek influence rather than direct administrative power. I hope that OpenSafely can be seen as a model for modern open collaborative computational data science.
I'm very optimistic that the NHS across all of the different settings, NHS E, X, D, and so on are embracing these new ways of working. As you may know, I'm doing a review for Matt Hancock, Secretary of State for Health and Social Care. Laura Wade Geary, Chair of NHS Digital, is also doing a review on how we can get better use of data in England, and I think over the course of the next couple of years, we will hopefully see a revolution around these modern, open ways of working.
And to be clear, I'm Matt Hancock. When we talk about openness and transparency with clinical trials, we're talking about people doing the right thing. We're talking about propriety. When we talk about open ways of working, when it comes to computational data science, it's really fundamentally about efficiency and getting the job done.
There's something very peculiar about electronic health records research in the UK. We haven't embraced modern open working methods that have become the norm in structural biology, in structural genomics, in physics. You would see as a matter of routine that people put all of their code on GitHub, that all of these intermediate knowledge objects get shared.
And for some reason with electronic health records research in the UK that community has somewhat been left behind for these modern open ways of working. But I think also I see an enormous amount of enthusiasm, especially amongst younger people working in those jobs. People take to tools like GitHub and Python and Docker like fish to water.
I feel very confident that we'll see a real explosion. And what we really want to get to is a collaborative ecosystem where for any given job in the sort of data management pathway up to a finished analysis, as far as is possible, it gets done once really well and then shared for other people to review it, kick the tires and reuse that code if they want to.
And I think managing ethnicity data is actually a really interesting concrete example of that. There is no such thing, as Rohini said, as a single ethnicity code in the electronic health record. Ethnicity is considered to be essentially a clinical event that can be recorded at many times throughout your life.
And I think that's probably reasonable, because as Rohini said, it cleaves nature at the joints. People may actually feel that their ethnic identity has changed over time. But the challenge with that is then, that you don't have something as simple as one ethnicity variable that you pull. from the record to use in your analysis.
Each person has between zero and a hundred ethnicity codes over the course of their lifetime. And so you have to start thinking how do I collapse these down? How do I map different ethnicity codes from different eras onto each other? But also, if it changes, what rules do I apply? And then also you get into quite interesting epidemiological issues, like you get about 70 percent coverage for ethnicity.
If you just use primary care data, you can bump it up to 95 percent complete recording if you add in ethnicity coding from secondary care data where you've matched it on. Superficially, that feels like a good idea, but it brings problems because it means that you've got much more complete recording.
much less missing data in the subset of your population that is more unwell, such that they have come into contact with hospital services. So that's a, that's an epidemiological issue. You're then thinking about, is it missing at random? What biases have I got in my data? Now, the tragedy is if all of that careful thought In managing all of those different ethnicity records into one variable.
If all of that careful thought has to be done separately, every time anybody does any analysis that uses an ethnicity variable, that makes no sense at all. And the OpenSafely model is Rohini, who is one of the great gods of ethnicity research in electronic health records. She puts a lot of thought into how we produce an ethnicity variable from the raw data once in OpenSafely.
And then everybody who wishes to, can reuse all of that code in future, so that you get a culture of people reusing stuff for efficiency. But also, I would trust Rohini's ethnicity variable more than I would trust many others, because she's thought about it, she's been around the block, but also it's been, it's had its tires kicked by lots of subsequent researchers.
Gavin: Have you been really excited to see the results of OpenSafely over the last 12 months? What are some of the, what are some of your favorite kind of, outcomes for, from dealing with this huge data set over the last year? I
Ben: think first up the fact that we were able to get our first end to end analysis completed six weeks from project commencement was extraordinary.
And Rohini was there at the start and there were tearful moments. It was And tears of joy, but it was I've never lived through any project like it. There was a huge number of people who've still never met. I don't think I've ever met Rahini in person. I haven't seen Liam Smith since 2019 and we've produced this extraordinary thing together.
So the scale of the teamwork and the scale of the speed and the efficiency, I think first up was extraordinary. Secondly, being able to harness it for near real time analytics, I think, has also been really exciting. So we were able to give insights back to the NHS on who was and wasn't being vaccinated before Christmas.
So within just a a couple of weeks of the vaccine programme commencing, that's also really exciting, I think. And then, lastly, and I'm sure you want single paper findings, but actually, lastly, the thing that's been really interesting and exciting has been bringing together people with electronic health records expertise, like Rahini, Liam Smith, and everybody in that group, and the software developers.
from my group and seeing people fold in those skills and work collaboratively has been extraordinary not in a dewy eyed way. It's always nice when people work collaboratively, but actually people pulling technical skills and knowledge and beginning the process of implementing some of the tacit skills and knowledge of electronic health records research into code libraries.
You do matching once for a case control study. But you don't do it in Stata. You take some really great Stata code for matching in a case control study, you implement it in Python, and you turn that into a package so that the next person who does that can do it with a single line command rather than hundreds of lines of Stata.
And just that process of being able to build a collaborative data science ecosystem, I think, has been really exciting. And actually, if I can have one wish, it would be that the Lancet would accept a paper. I haven't submitted one, you haven't rejected it, I haven't given you the chance, but if you take a paper, frankly, on how we did it, how OpenSafely works, rather than just the clinical findings, because I think one of the reasons why, one of the reasons why we haven't had really great software and tools and thinking around clinical informatics is that it has been regarded as being this sort of backroom.
Oh, it's just implementation in quotes. And actually having done a bit of epi and a bit of computational data science, I can tell you that they're both equally intellectually challenging. They both require that you have domain knowledge from clinical work, from epidemiology and the bear traps of data science around patient data but also software development and how you get things built and done.
So I would love to see organizations like the Lancet putting a real premium on this. And I think that will come, there are always shifts in what journals regard as legitimate and interesting activity. Used to get audits in the Lancet. I'd like to see those again as well. Oh, you invited me for an interview and now I'm lobbying you.
But I think these kinds of practical data science innovations, I think, are every bit as interesting as any single individual finding. In fact, in some respects, they can be more powerful.
Gavin: In terms of lobbying to get manuscripts in the Lancet, you've got the wrong person here, I'm afraid.
Finally, when when you know, touch wood, we start to move past Covid, at least it becomes less of an overriding concern for everyone. What are some of the immediate priorities for Open Safely past Covid?
Ben: The number on priority is service restoration. There's a backlog. We've got to understand it and we've gotta prioritize our way outta it.
Sometimes health economists talk about the parallelogram of doom. If you stop having an activity and then you try and catch up, you've still got this huge unmet need that especially for things where you've got an ageing population and so demand was rising already. I think the real challenge is to find the activities that were deferred with consequence and work out how best to prioritize our way out of that.
I would say, however, that from all of the work we've done looking at changes in health service activity, We've seen a lot of evidence of a really adaptive ecosystem of clinicians and commissioners and managers just in primary care alone. For example, we can see that cholesterol testing during the peaks of COVID waves disappears almost to zero, whereas INR tests for people on warfarin exhibit almost no change at all.
And so you can see that GPs canny about continuing to do things that needed to continue and deferring things that can be deferred. But nonetheless, there will be activities that were deferred with consequence. And then separately as part of that, and it is only a small part of that, there may also be activities that were deferred during COVID without consequence.
And I think you wouldn't look just at the data here and you'd be very cautious, but it may be that in the natural experiment of COVID, we can identify discretionary health care activity, which in actual fact in reality was low value, and they may be opportunities for rational disinvestment. But as I said, that's a very That's a very cautious analytic journey, and you wouldn't do something like that.
Gavin: That's absolutely fascinating. We'd look forward to having you back on in about a year's time where we can talk about that kind of stuff and how it's going. But for now, Rohini, Ben, thank you so much for joining us on The Lancet Voice. It's been a real pleasure having you both on.
Thank you for listening to this special episode of The Lancet Voice. Remember, you can subscribe to The Lancet Voice, where you usually get your podcasts, for the story behind the most important research right now. We'll see you again next time.