 OK, so Jen Murphy is our first speaker, so Jen is a PhD student in the Centre for Data Analytics and Society based in the Social Statistics Department at Manchester University. Her research interest is in equality and specifically the impact of deprivation on health. OK, so I'll skip straight to the first slide because I've already been introduced. So I'm not going to go through all the background, you know, we all know about the pandemic and I've again got a child at home again homeschooling as of this morning. But the work I'm going to talk about today considers the non COVID impact of the pandemic and it uses the COVID data sets collected as part of understanding society. So we hypothesized that as a result of the pandemic and the accompanying lockdown that well being impacted and that there's likely to be widespread indirect effects which are important to policymakers and health professionals. As we as a population recover from what's been really quite an unprecedented experience. And further to that in this work we hypothesized that the impact's not been felt equally amongst the population and that there will be deprivation effects in the impact of COVID on non COVID related well being. So on this slide here I've reproduced the research questions that we've addressed as part of this work. There's been widespread reports in the press of a mental health crisis. So we looked at the reported decline in well being. We wanted to confirm this was true and also to look at whether or not there were deprivation effects in the extent of the decline. And likewise if well being was being more adversely affected for people who had underlying medical conditions. And we then wanted to explore whether or not this decline was static or mobile during the first wave and whether those changes were experienced equally. Well being is a bit of a loose term so I'm just going to specifically define what we mean in this study. We use standard measure of mental distress which is the GHQ12. There's 12 questions things like have you recently been feeling unhappy or depressed and the respondent give one of four answers. It's a fairly standard tool but I just want to draw your attention to the fact that there has been research that's shown there tends to be under reporting bias in this measure particularly for men. In our research we use the case in the score which collapses the responses to a binary for each question and then sums a score over 12 questions. So on this slide here I've included some of the factors affecting well being that may have been relevant in the pandemic. I'm just going to spend a couple of minutes now going through each one. If we start at the top with access to healthcare so we know the healthcare system has been impacted by the pandemic. The period of this study is restricted to what we're calling the first wave of Covid in England so that was kind of March through July. During this time the highly infectious nature of the virus meant there were significant organisational changes for the healthcare services. There was a study release that looked at diabetes, COPD and hypertension and it showed that because of the resource relocation from chronic disease to Covid there was quite a significant disruption in the continuity and the quality of care for these other diseases. In England elective surgeries and outpatient clinics were cancelled or postponed and there was a big move to teleconsulting and in the earlier stages demand for respiratory care caused a knock on effect in resourcing. So there were other restrictions of specialist services and there has been some reports that conditions like cancer we may see things like 3000 excess deaths in the next five years as a result of delayed diagnoses. The availability of healthcare also impacts on those with chronic or long term conditions so people with long term conditions may not have been able to access their normal healthcare setting or service and actually social engagement and access to information services such as support groups is part of the ongoing self management for people who live with these types of illnesses. Self management physical mental health they're all supported by social involvement with groups and in some cases these networks can even substitute for formalised healthcare but of course these social networks have been significantly disrupted by the pandemic and that leaves me on to the next factor which is social isolation. So just at the time when patients might have been unable to access healthcare and need to draw upon their social network for support access to groups socialising and even spending time with friends and family were all heavily restricted and it was worse for those with significant care because they were instructed to shield. So loneliness has been shown to be associated with well-being for people with and without disabilities. There may also have been an additional impact on the breakdown of any network care for those types of people. Having a risk factor for COVID-19 has been particularly stressful during the pandemic and this may have played into the other factors affecting well-being. It's really stressful to be aware that you are in a high risk group. Shielders were very isolated during the first wave and this also impacts on job security and finances and the range of other factors that were coming to. Next up on my diagram is age. We know that age is a risk for severe COVID disease. We know that older people were instructed to shield and we also know this older population is more likely to be faced with management of long term conditions and loneliness. There's also evidence that was gathered from the SARS outbreak in 2003 that there was a real increase in suicides in the 65 plus age group and the research there concluded that socialist engagement and mental stress, anxiety and fear of being a burden were all big factors in that decline in well-being. The next few items all relate to the change in the economic situation around the lockdown. So we've included financial stress either through a loss of work or reduction of paid hours or furlough. The increase in the unpaid work burden here I'm talking about homeschooling or caring, both of which have really impacted people's well-being either through having to work at the same time or having to stop working or just a general change in their work conditions and job security, whether that's working from home or having to conduct your normal way of working differently. Community and organisational resilience are the final two on this slide. There are organisations on which we depend, so this might be because of health needs, but there's also things that we use for health well-being such as social groups or sports clubs and how resilient those organisations have been to the pandemic can be impactful. And finally community, living in a community that's supportive may influence our people of cope with day to day life and in the face of the lockdown this might be more important than ever. We use the understanding society COVID survey at the time that we did this work, the later waves weren't published so we used the first four waves and these broadly correspond to the first wave of the pandemic in England. Data were collected April, May, June and July. By July there was some reopening of the economy although still subject to social distancing and mask wearing and so on. We selected only English cases because we wanted to look at deprivation and we used the IMD which is place-based so we can't compare between countries within the UK. And we included only respondents who had replied to all four waves of data collection and who were contained in the baseline data which was wave nine, the latest data available from the main study at the time of the work. We only looked at over 18 so this is just a study that looks at adults and we only considered online responses. Age, sex and the baseline score from wave nine were predictors of response so the missing data is not random here. The response rate was just over 20% when we looked at the longitudinal response across the four data waves that we used. You can see here just going to be a little bit more into who didn't respond. You can see here from this plot there's a skew in the age of respondents so the mean respondent age is 55 years compared with 49.1 for the main survey. And you can see also that there is a skew in the deprivation desile so we used the LSOA of each respondent to impute the desile. You can see we've got many more people in the higher desiles which is the least poor backgrounds. Females were more likely to respond so the data was again skewed towards female sex and there were only 786 non-white valid English cases within the dataset which is 9.4%. And that compares with the underlying rate of 20% in the main survey so there's some underrepresentation here but it's not something we've been able to investigate as part of this work. The method is fairly straightforward. We fitted a number of models using an ordinary least square regression in Python using the stats models library. We noticed a feature in the data that there was a decline in wellbeing that is to say the score went up and then a corresponding recovery. So we decomposed this into two separate phases, decline in the recovery and model them separately. And we also ran separate models for men and women based on the literature around different reporting behaviours. And this approach has been confirmed by a couple of papers that have been published on the same data recently in Lancet Psychiatry by another team. As I said on the method slide we ran two models one for decline in recovery. We included the baseline case and score in both models and for the recovery we also included the decline in the case and score for each individual. And we included sex and age, a binary variable to capture whether or not the person had a long term health condition and responses to the loneliness module of the COVID data collection. We used food bank use as a proxy for the experience of an acute financial crisis and we coded income change based on equalised household income to indicate worsening or improving financial situation over the period of the decline on recovery. And community cohesion was coded from questions in the survey. Here you can see the trajectory of the wellbeing score. So just remembering that a higher score is poorer wellbeing. So we can see that males and females followed the same trajectory through the lockdown but that women fell further from a higher baseline score and then recovered by a similar amount. Whereas the men had a less steep decline in recovery but from a lower baseline. So the next few slides show the models for the decline in the recovery. You can see here I've indicated statistical significance by an asterix and we've got that across a number of variables. And here in this model you can see we've got just around about 38% of variants explained. But actually when I further investigated it we discovered that for both the decline and the recovery. The only things that really mattered were the baseline and loneliness variables. So the next slide shows the model for decline with the extra variables taken out. And you can see we've still got almost the same R squared value. So just to explain it a little bit more what we're seeing here is the change in the GHQ caseness score between wave nine and the April measurement. And this model shows there's an association between the baseline score and loneliness. So if you had a higher baseline score to start with your decline was less. But if you experienced loneliness then your decline was much greater. If we look at the recovery and we can see we've got really similar effects. Again no effect seen here for deprivation and no effect seen here for having a long term condition. And that was the same across all of the models. And moving on to this side you can see yet again most of the variants is explained by the baselines and loneliness. So the baseline score is important. The amount of the decline is important for the recovery. And again loneliness is important. And this loneliness variable was slightly different to the one in the decline because we summed it over the waves to try and capture some difference between people who experienced loneliness. Sometimes people who experienced persistent loneliness throughout the course of the pandemic. So well being declined at the beginning of the lockdown and it recovered as the restrictions were lifted almost back to the baseline. The differences we expected to see between groups who were more deprived or people who had long term health conditions. We didn't see those and deprivation showed no association with well being. This has confirmed other work in the field. So what does this actually mean for people and how they've experienced the pandemic and how their well being has responded to that? Well, experiencing loneliness was predictive of decreasing well being and experiencing continued loneliness seemed to have resulted in people recovering less well as the pandemic progressed. The difference between people who are and aren't lonely may indicate differences in the way that some people experienced the loosening of the lockdown. So some people chose to remain isolated after concern for their health or because of shielding advice whereas others made the most of the new freedoms. One of the most interesting findings here was that the social gradient which is well known to exist in health inequality doesn't seem to be detectable here. It's really interesting because it means that our initial hypothesis about deprivation effects is not supported by the study. And it may be due to an overriding community effect present at the national level during the first wave of the pandemic. So more affluent people were perhaps less resilient to the impact of the pandemic as we might expect or possibly poorer communities were more resilient to the lockdowns. We did look at community cohesion, but we found very little evidence to support it being important for the well being for this sample. It showed a weak effect in women, but only for the decline and no effect in the recovery. And this might be because it's either it's either not important or it might be that actually in the first lockdown there was such a strong national narrative and rhetoric that the cultural environment of resilience may have been protective at a national level. And that might have overridden any more localised effects. Just one more minute left, John. The results here, you know, they're interesting and slightly counterintuitive. We found no evidence of a social gradient, although mental health certainly suffered during the first lockdown. For this sample, well being was shown to be highly elastic and it indicates a national level of resilience which is cut across the usually observed health inequalities. Really we need to have more research to target the groups who are underrepresented, but the study does suggest that national efforts to raise the spirits may well in this context have been useful. And really actually the best thing that we can do in terms of policy for raising people's well being is to mitigate against the pandemic and ensure that we can allow people to return to their normal social life. I would be really interested to have seen the same data collection from the second lockdowns because certainly anecdotally it seems to feel a lot worse. So I have rattled through that in the interest of time, but thank you very much for listening. Okay, thanks, Jen. So the next speaker is Sarah Sieger. So this is about tapping into the world's largest observational research network with the OHDSI community. So Sarah has 20 years, over 20 years experience across the UK health sector specialising in healthcare analytics, statistics and informatics. She conducted large-scale public health analytics within the NHS and designed and implemented data lakes and the creation of a new data science function for one of the largest UK private medical insurance companies. She joined the observational health data science community in 2018 and is an active member of various OHDSI working groups as well as the OHDSI steering committee. Your main areas of expertise are statistical data analysis, geospatial analysis, data visualisation, data science methodologist, lots of others, and she's a certified ethical hacker as well. Fire away then, you've got 15 minutes, I'll give you a 2-minute warning. And you can see my slideshow, yeah? Perfect. Perfect, fantastic. Yeah, thanks everyone for joining. So I'm going to just sort of briefly walk through sort of what is OHDSI as it's pronounced. So OHDSI is the observational health data sciences and informatics programme. It's a complete open network. So although I work for IQVIA, we are also part of this open community which contains multiple stakeholders, interdisciplinary, collaborative, and essentially the idea is to sort of create open source solutions that bring out the value of observational health data through large-scale analytics. So you can see that there's a number of different collaborators from across the world and they range from academia through government, pharma, hospitals payer, provider. Everyone, anyone can join this open community for the sake of doing global research and really the premise of this is around standardisation. As I mentioned, this is public, it's open, it's not funded by pharma, it's completely international and anyone can be part of this. Even if you don't have data and you're wishing just to partake in research, you can be part of this community and get involved. So what we call the OHDSI network, we've got a number of collaborators as I mentioned. We've got lots of different organisations involved, just to give you an idea of who we have, who are active at the moment. We've got lots of emerging collaborators who are coming on board and there's also obviously a lot who are intending to come and join. So as I said, it's completely open, it's a great community, very global. But OHMOP, so OHDSI and OHMOP, they're two terms that are used together in synchronous. OHDSI is a community and OHMOP, which stands for Observational Medical Outcomes Partnership, is essentially the common data model. So what OHDSI, the community have done is that they have actually created a standardised and harmonised common data model. And everyone who is part of this OHDSI community either has data that they have converted themselves to OHMOP or they may not even have data. They may just wish to get involved in the research. But what does that mean? So essentially for those that do have data, so for myself I work for a large organisation which is very, very data rich and has lots of different data sources and data types. But we have lots of raw data. So we, for example, even in my own role, I have access to US claims data, to UK primary care data, to European hospital data. But all of it, as you can imagine, is very, very different. They have their different systems, their different sources. And if you think about plugs, this is one of our favourite slides, whether you like it or not. You know, you go to a different country, there's different sockets. You know, how do you know you've got the right converter? And then when thinking about the analysis, it's like, OK, so as a researcher, how can I do global research if all of the data is very, very different? Let's forget about the access for just now. But all of the data is different. So, you know, how do I go about even thinking about adherence to a particular drug? Or how do I go about doing a health needs assessment around a particular condition or therapeutic area? Well, what OMOP does is that because it takes the raw data and converts and standardises that, you then have this global set of what we call OMOPT data. And so no matter what the data in its raw native source looks like or where it comes from or in what language or what country, it will technically look exactly the same and that therefore means we can then come along and actually do global research. And so just sort of briefly just to go around sort of the common data model. I won't go into too much of technical detail because I want to get to the COVID stuff. But really the CDM common data model, it's a patient centric model. It's based on sort of a relational database construct. And so within this, all of the tables, all of the fields for no matter what data that you have are converted. Now, as part of the community, you have people who can help you do this. Or they are on the Odyssey website, anyone can go and do this themselves. So from the data aspects, the tables are in this sort of standardised way. And no matter whether it's a health survey or electronic health records, it's still able to sort of form this construct. Now, it's not just about the data. It's all about the vocabularies or clinical coding or ontologies or however you want to describe it. You can tell that I work for global teams. There's lots of different terminologies for the same thing. Essentially, they're all different, various across all the different systems and all clinical events in the OMOP common data model expressed as concepts. So we've got drugs, conditions, procedures, measurements. And essentially, all of these are then standardised themselves as well. So as you can imagine, this is a huge task in itself. I mean, clinical coding on its own is huge. But when looking at global data, this is also something that's taken into consideration as well. Now with OMOP, it's not just the data, it's not just the vocabularies. Also the analytics, the data science as well that is standardised. So the open community, Odyssey community, they've even designed specific R libraries that allow researchers to conduct common types of analytics, which for the COVID examples, which I will come to very shortly, meant that we were able to do or at least start working on these COVID studies very, very quickly and very, very efficiently. Because a lot of the sort of pre-canned analytics and codings were there already available for anyone to actually obtain themselves and put within their own environment. Now what Odyssey does and what I do on a day-to-day basis is we work in this sort of research network environment so that I wanted to use this one to give you an example of how that would work. So I'm done the bottom in IQVIA. We have lots of data and we act as a coordinating centre at my teeth. And we have lots of data all converted to OMOP. But as part of OMOP, the brilliant thing is that to do global research, you do not need to access anyone's data. No one will ever access your data. All that is done is the analytical code is created by the coordinating centre or whoever. And it's just that analytical code that is then shared to anyone within that network, to anyone that's external, to anyone that's in the Odyssey group. They then take that code, run that against their data, assuming they have data, and then it's just aggregated results are sent back. And then the coordinating centre would just then compile all of these insights. And before you know it, you've had or you've completed a global research study. So as you can imagine, it sort of mitigates the need for data governance and security and privacy because no data is ever being shared. You don't have to worry about firewalls. It is just the actual analytical code, the cohorts that you are creating as part of your research. And that's what's being shared. So I wanted to give you that context because then when I go on to sort of the COVID, it gives you an idea of how it was actually done. So I'll get to the exciting bit. So as part of the, well, because of the pandemic, for us at the Odyssey community, it brought the opportunity to do an awful lot of COVID research. And so back in March of last year, there was a huge studyathon, as we call them, and it was open up to the entire Odyssey community. So to all 2,500, 3,000 collaborators, everyone was able, should they wish, to join this studyathon. And the idea was that there were sort of three research focuses. So the main one was really around characterisation and this project was called Criptus. And that stands for characterising health associated risks and your baseline disease in SARS-CoV-2. And essentially, that was around sort of analysing, it was an in-depth study of COVID-19 and other viral disease populations. So, for example, we looked at adults that were hospitalised with influenza back in 2009 to 2010. We looked at flu seasons as well back in 2014, compared to adults who were hospitalised and tested positive or who were diagnosed with SARS. We also did a prediction model. So this study is aimed at flattening the curve by understanding individual risk for COVID-19 outcomes across the international network. We also validated existing models, really to try and understand who will utilise intensive care or intensive services, depending on what part of the world you come from. And then the third one was what we call an effect estimation. So this is a pharma co-epi study testing the effectiveness and safety for treatment of prophylaptic use. So, for example, Hydroxychloroquine took me a while to learn how to say that, either alone or in conjunction with azithromycin. So, just to give you an idea of what this studyathon looked like. So it was run across a period of 88 hours so that it could cover all time zones. So it was almost like an 88 hour machine. So we had everyone from Europe and then they were joined by those guys in the US. And then as they went to sleep, the guys in Asia Pacific joined. So it was a real great global event. Over, I think, 30 countries were involved. There were lots of what we call global huddles, so little work streams. So one team would go off and work on how to design the phenotype. Others would go and work on, say, with the data validation of those that had data and were contributing to the studyathon. There was a whole range of different work streams, even to the point of literature reviews and what else was available. It also resulted in a number of publications, which there are quite a few links within this PowerPoint so you can actually catch those later on. And there were some actual study packages, so 13 different study packages. And what that means is that on the Odyssey website, you can go and actually utilise those study packages right now and start to see what the concepts were used, how the cohorts were built, what populations they were looking at. It was just a truly fantastic studyathon, but it's still continuing to this day because obviously data is refreshed and constant. And so it's, I think, was one of our more prestigious events that we had. And also, yes, okay, cool. And then I'm thinking about some of the results. So I won't go through all of them because I added quite a few slides at the end, but I thought you guys can do a little bit of bedtime reading. And they also link to the publications themselves, so you can always do a bit of research there. But essentially, so Charybdis, you will see at the bottom here there is a link to a shiny app. You can go to that link right now, and it's an interactive dashboard where you can see all of the results from this study from the different providers that took part in it. But essentially some of the highlights, I'll go real quick. So for this particular publication, the main highlight from this was around how obesity was more common amongst COVID-19 patients compared to the influenza patients, and that obese patients presented more severe forms of COVID-19 with higher levels of hospitalisation. And then there was another publication which looked at multiple medicines that were used in the first months of the COVID-19 pandemic. So bearing in mind, this is global so that you can just understand and get a sense of the variety of the different geographies that were involved. We also have, I think there's probably about another 10 or so slides, but as I said, you can go and really get to grips into the detail a little bit more. But again, we looked at comorbidities and the prevalence of those with those patients who were hospitalised. Multiple medications used to treat pregnant women and whether they had any benefits from that. The results go on. As I say, the study-a-thon was really quite huge. Charybdis on his own, which is the characterisation, we're just looking into the disease factors of COVID-19. There are so many different angles, so many different nuances that there's quite a bit of reading and some results to get your head around, which is great. How am I doing for time? Am I nearly there? You're just about over, but we've got a bit of leeway, so you can carry on. I'll just flick through them quickly. I'll just go through the highlight section. Again, this is still Charybdis. This was looking at the different testing practices led to the baseline characteristics outcomes. It just really showed the importance of large-scale characterisation of something like COVID-19, which can really help with informing planning and resource allocation across multiple countries, multiple health systems. Here, we looked at HIV and COVID-19 co-infected patients. Essentially, we found that across the different care settings, co-infected patients who received intensive services were more likely to have more serious underlying disease or a history of more serious events compared to those who were diagnosed with COVID-19. Then, we have a few others. I still have some more results around Charybdis. I do have some other results. Let me move to the population-level effect of estimation. Here, we looked at the short-term use of hydroxychloroquine because that was a huge thing in the news at one point. Then, there's also some results around, if I get to this one, the patient-level prediction. We looked at seven predictors, whether the patient's had a history of cancer, COPD, diabetes, heart disease, hypertension, hyperlipidemia and kidney disease. Then, combined that with an age and sex discrimination, which then allowed just to experience really the different outcomes. Again, this is another shiny app. You can go and play with this. It's rather a fantastic one, to be honest, from a geeky point of view. You can change the different variables and really get to grips with the data without even having to own or access the data or even ask for permission. Again, OMOP Odyssey is about aggregated results. It means we can do an awful lot of sharing and not have to worry too much about patient sensitivities and things like that. I'll leave that there for now, but please do go and read this PowerPoint a bit more. There's more links in there, and I've gone sort of super quick, super high level, but there's lots of, as I said, applications to go and play with, as well as read. Johnny's head of data within the ONS health analysis team. He has expertise in health data sharing across government departments, and his team are responsible for the processing and curation of the civil registration data that ONS received from the general register office for statistical purposes. Previously, he's worked in several government departments in a range of analytical roles, including Ministry of Defence, DEFRA, and the Cabinet Office. And Vahee, who is going to be speaking, is a principal statistician at the Office for National Statistics leading the Health Modelling Hub in the Health Analysis Division. He's also a research fellow at the London School of Hygiene and Tropical Medicine, working in the Population Health Innovation Lab, where he moved after obtaining a PhD in economics at King's College London. And his current research interests include estimating the health effects of social policies and studying health inequalities. So, over to you, Vahee. OK, so I'm going to talk you through an innovative data set that we've built during the pandemic to help with a policy response called a public health data asset. So, clearly the pandemic increased the need for timely evidence to monitor at the beginning the differential impact of COVID-19, the inequalities in mortality, and still does to an example in monitoring the inequality in vaccine uptake and infection. And this highlighted really the importance of having population level data that you can use to look at inequality. Because, for instance, mortality is a relatively rare outcome. And if you want to look at mortality in small groups, then what you really need to do is to get this population level data set. Because even a large survey would be difficult to use to look at small groups. The limitation of existing data sources, so that on one hand you had the electronic health records that are really good because they are population level data sets. But they tend to be relatively limited in terms of social demographic characteristics that are included. On the other hand of the spectrum, you've got surveys which are very detailed, cover a lot of different social demographic characteristics, but are of relatively limited sample size and are also quite expensive to set up. So the UNS has done a good job at setting up the COVID infection survey, which is about 500,000 people are followed monthly. And this has been crucial, but this is very expensive and relatively took quite a while to set up, even though it was done at an incredible pace. So what we did at the beginning of the pandemic is to try to create a new data source linking the census to electronic health records and mortality data. So that's what we call the UNS Public Health Data Assets. So it's a new population level data set designed to enable analysts and researchers to conduct public health research using data for multiple sources. So it's still an evolving product and it's been going through several phases, but at the core of it is a 2011 census which was linked to the National Health Service Patient Register to obtain an NHS number. With a linkage rate of about 95%. And then this linkage to the patient register has allowed us to link to different data assets from the UNS, the DEZ data, but also from other government departments including NHS digital. So we've got the linkage most importantly to the GP data that was done by NHS digital. So that's all primary care records for patients that were alive and active and living in England in November 2019. And also to a wide range of data sets, so the hospital statistics, the test interest data, the vaccination data, so on and so forth. So that's been made possible by getting the NHS number attached to the census. So what it covers, it's basically what the NHS Public Health Data Assets cover is basically the intersection between the census and the GPS data. So it's everybody who were enumerated at the 2011 census and were still residing in England in 2019 and alive. So it does exclude quite a few people, so it exclude recent migrants, new birth as well by reconstruction. And also it exclude people who are not registered within NHS or people of course who have moved away and that's a good thing that excluded. So in terms of coverage, what's in this data set, so we've got data on about 40 million people aged nine plus that were alive at the beginning of the pandemic. And as you can see on the bottom right, we've got a coverage of nine plus of about 79%. And what is interesting is that the coverage improves is higher for older age groups that reflect that the linkage rate between the census and the patient registered was higher for older people. So that we cover about 90% of 65 plus, which is for this pandemic is important because of people that were more severely affected were the elderly. And then in terms of what we have in the data sets, we've got a wide range of social demographic characteristics. So we've got obviously ethnicity, which is self-reported, unlike in electronic health records. We also have religion, main language and a wide range of cultural factors. We also have some social demographic characteristics, some socioeconomic characteristics, so the House of Deprivation, the tenure and the House of Composition. We also have some occupational exposure, so obviously the main limitation is data that was collected nine years ago. But for the geographical factor of when we could link this to more recent patient registers to update all the geographical information. And we also derived a medical history following the QCovid risk model, the medical history that was used to build that model. So we cover a wide range of severe conditions and less severe that were all associated with COVID-19 mortality. In terms of outcome, I should have updated this a bit now. We've got the mortality hospitalisation for COVID-19, but also we've just acquired the test and trace data and to look at infection positivity. And also we have linked vaccination data, so now we can look at vaccination as an outcome. And so the data can be accessed now, most of it not the vaccination data yet, via the secure research service. And I've put some links here for you to use after, if you're interested. We have had our first research accessing the data a few weeks ago, so it's now open running and all set up for researchers to use. So in terms of work, so we've done quite a lot of work on, so here's a summary of some of the work we've done and I'm going to go into some more detail in this presentation. We've done a lot of work on inequalities in COVID-19 mortality, so by ethnicity, by religion, also to do some work on household composition. And we've done a validation of a prediction model, the Q-Covid risk model, which is a model that was used as a result of this validation for the population risk assessment and was used to update the shielding list that was used for informing the prioritisation of a vaccination campaign. And we have a work programme on long COVID and looking at the post COVID syndrome. And we've done also some, we've got an emerging work programme on vaccination, so on uptake and also on effectiveness. So I'm going to mainly focus on the work we've done on ethnicity, because that was the first piece of work we did, and also on the post COVID syndrome work that we conducted. So at the beginning of the pandemic, there was some early signals of an unique impact of the pandemic on several ethnic minority groups. But there was no way to measure this directly using just mortality records because ethnicity is not collected on the death certificate. So what we did was exploiting the new linked census to mortality records that we had developed just before the pandemic to estimate the difference in COVID-19 mortality between ethnic groups. So what we did was first just adjusting for age to look at the role type of inequality and then trying to exploit the characteristic from the census to try to understand these inequalities and understand why do we observe these differences. And finally, we also did some work trying to look at how the lockdown affected these differences. So did the lockdown reduce or increase the inequalities? That's the question we answered. So very briefly, so that's our kind of main result. So that's for wave one of death. So that was just for the first wave. We've seen that updated it, but that was our first publication. So it looks at the relative difference in COVID-19 mortality between several ethnic minority groups and the white ethnic groups. And it shows a result that has had ratios for different models. So on green, adjusted for age, and then sequentially adjusted for several different factors to look at how the differences change when you adjust for different factors. So what we can see is that the differences were massive, up to three times more likely to die for black men were three times more likely to die than white men. And it was about 2.54 people from Bangladesh in Pakistan in the background and very raised risk as well for female and for all the groups. And here, I mean, we usually when we look at an equality in health when hazard ratio of 1.5 is dim as really large here, it's far beyond, far above and beyond what previously be documented for other disease. And that reflects to an extent that it is an infectious disease and that's very different from a non-communicable disease. Because once we found that adjusting for population density and local authority districts for geography factors, that really decreased the estimated differences in mortality. That suggests that a lot of the inequalities were driven by geography. So the pandemic hit some larger cities, larger urban areas, more densely populated areas more quickly in the first wave. So that accounted for about 50% to 60% of the differences were accounted by geography. And then adjusting further for different measures of social demographic characteristics did reduce a bit further the hazard ratios. And for instance, if you look for the hazard ratio for Bangladesh and Pakistani female, once you've adjusted for all the characteristics in the census, then you find no differences anymore. So this suggests that all the differences are explained by the factors included in our model. So this is go against the theory at the beginning that the differences were driven by differences in fatality rates. That the virus would be more affecting some ethnic minorities more severely is just that the main reason why they were more likely to die is because they were even more likely to be infected. That was confirmed later on when using studies of antibody prevalence as a reactive study that found that the case fatality ratio between groups were exactly the same. So our second analysis that we did on this was to look at the difference in COVID-19 mortality between ethnic groups pre and post lockdown. And here what we find is that the lockdown really attenuated the inequalities. So in green it's the hazard ratio before lockdown and they fully adjusted hazard ratio before lockdown and in yellow it's after lockdown. And here we can see it's clearly there is a massive reduction in inequalities after the lockdown. So that showed that lockdown policy not only helped reduce infection, but it also helped reduce inequalities in exposure. So again this is another indirect proof that the differences were probably driven by infection rather than by differences in severity. Because if it was just a severity of vitamin D or the other type of explanation, the lockdown shouldn't have made a difference or very little. So we had a wide impact in the wide media coverage in the press when we released this on the UNS website. And it's also had a big policy impact with the UK government ordering a review of the disproportionate impact of the pandemic on the Bain community. Also several community-led interventions tried to raise awareness of the risks. And so we provided regular updates and further analysis for SAGE and ISD subgroup. And we also had a relatively big scientific impact so we published a lot of our work, not only on the UNS website but also in academic peer review journals. That helped raise our profile and develop collaboration with some leading academics in the field and obtain further funding for our work. So now I'm going to complete this change topic and tell you a bit more about the work we've done on the epidemiology of the post-COVID syndrome. So this is not a single condition but it's more a wide range of sign and symptoms that, as I said, as defined by NICE, they developed during or following an infection consistent with COVID-19, which continue for more than 12 weeks. And I'll not explain by an alternative diagnosis, that's the official definition. And so what was known when we started the study that there was a high post discharge mortality and readmission amongst individuals that had been hospitalized with COVID-19. But there was less evidence about the damage, the consequences on morbidity. And what we did is to estimate the rates of post-digital diagnosis of respiratory, cardiovascular, metabolic, kidney and liver disease. And we used a match patients as a control group to try to identify the kind of causal effect of being hospitalized with COVID-19. So what our setup was to use, so we focused on our group of interest with patients hospitalized with COVID-19 that had been discharged by August 2020. And we did a one-to-one matching based on demographics and a 10-year history of comorbidities to get a control group. And then we looked at outcome in the hospital-episod statistics and in the primary care. The matching variables, we used age, sex, ethnicity, region, IMD, quintal, smoking status and several previously conditioned hypertension, major cardiovascular events and respiratory disease, etc. Hi, Farhi. You've got two minutes left. Perfect. Okay, so I'll go quickly to the main graph. So here he shows the rates per thousand patient years of adverse events in discharge patients in England compared to with match control. So in dark blue you've got the people who have been hospitalized with COVID-19 and in light blue that's the rate for the control group. So that's kind of the rates of events that you would have expected for that particular group if they had not been admitted to hospital. So we can see a raised risk of all events. We looked at the diabetes, major adverse cardiovascular events, chronic kidney disease and chronic liver disease. And we find a raised risk not only for diagnosis, but also for new diagnosis of people when previously not being diagnosed. That's why it's shown on the right panel. It's new diagnosis of people who in the last ten years have not been diagnosed with diabetes or have not experienced major adverse cardiovascular events or chronic kidney disease or chronic liver disease. So we find a large increase so that suggested that there would be quite a big burden on the NHS for all. So here it's really looking at the most severely affected patients because patients are in hospital. So we do have a work program looking at to range to replicate this in the population of just infected population, not just the people who have been hospitalized. And with our data we're also able to look at the differences in relative risks. Different groups of developing different conditions. So you can see that in absolute risk of course people aged 70 plus who had been admitted to hospital had a higher mortality rate than younger people, but in relative risk. The relative risk was much greater among younger people. And it's found consistently for all the disease that younger people who had been admitted to hospital had a much greater probability or risk of developing different conditions than older people. We also found interestingly very little difference between men and women, but we found a slightly raised risk for a non-white population, again in relative risk. So in terms of the impact this study was published in the BMJ and it was presented at several events to medical practitioners at the BMJ webinar. And also it's been presented as part of a long COVID training event. It's been used extensively in the academic lecture and as well as in the NIHR's review of international evidence for long COVID. And it's been used to brief Matt Hancock and Chris Whitty as well as the prime minister. Okay, I'm going to stop here and I'm happy to take any question. I had another slide on a vaccine update, but I'll let you, I'll show the slide afterwards anyway. Okay, thank you. If you just got one more slide for you, you could show that we do have extra time. Okay, good. So I mean it was just the study we did on vaccination coverage by social demographic characteristics with the NIMS data. So what we did was to link the vaccination data to the, to our public health data assets. And we achieved a linkage rate for the 50 plus of 86%. So it means that 86% of people who had been vaccinated, we managed to find them in our data set. And that's relatively consistent with the coverage figure after new at the beginning. And we did two things that just estimate the coverage rates by social demographic characteristics. So we looked at ethnicity, religion, a measure of household tenure, education, a wide range of measures that we had, a main language, et cetera. And we also produced some odd ratios of not being vaccinated to try to understand a bit better, a bit more the inequalities in vaccine coverage. So I've just got the results for the ethnicity. So what we highlighted was massive differences in terms of coverage in adults 50 plus. That was by I think mid May, but the picture hasn't changed much. So we could see that over 90% of people from age 50 plus were from the white British ethnic group had been vaccinated. And it was much lower in other groups, such as especially among Black Caribbean and Black African, as well as Pakistani and other groups as well. So yeah, massive differences. And these, interestingly, were not really explained to some extent by social demographic characteristics as shown on the right. It was a graph, but it remained massively raised risk of not being vaccinated. So for the Black Caribbean, they remain more than 5.5 times a greater odds of not being vaccinated after we adjusted for all the social demographic characteristics and medical conditions. So this highlighted the cultural factors played a big role, things that we don't measure in our data and behaviours. So that was quite impactful in trying to set up new initiatives to increase the uptake in these groups. Okay, so put the link to the publication as well. Okay, that's all for me. Thank you very much. The next speaker is Catherine or Katie Saunders. She's going to be talking about inequalities in access to primary care experienced by people with multiple morbidities during the COVID pandemic. So Katie is a statistician working in the primary care unit in the Department of Public Health and Primary Care at the University of Cambridge. This work was carried out as part of a project carried out by the Birmingham Rund and Cambridge Evaluation or Brace Centre, which is funded by the National Institute for Health Research to conduct rapid evaluations of new services and innovations in health and social care. Hi, my name is Katie Saunders. I'm from the University of Cambridge. I'm talking about some work I did as part of Brace, looking at the, which is a rapid evaluations into evaluating service innovations within the NHS. I'm going to talk a little bit about the work of them. We've been doing looking at the impact of the introduction of telephone triage in primary care both before and during COVID-19. With particular focus of inequalities in access experienced by people with multiple long term conditions. I'll give you a little bit of context, talk a little bit about the methods, the results discussion and some reflections, but please do jump in with thoughts. So telephone triage, telephone triage is when you call your GP and you either speak from straight away on the telephone or you wait for them to call you back. Then the issue is either dealt with on the phone or you're invited into the surgery for an appointment. So telephone triage, pre-2020, pre-pandemic was designed within the context of increasing primary care workload with the shift in the management of long term health conditions having moved mainly from secondary care to mainly primary care. And the ageing populations are an increased need for management of long term health conditions kind of is the context in which this increasing workload for GPs is occurring. And telephone triage was designed as a tool to kind of the demand management to manage the workload. It's been evaluated for the early trials and early evaluation using routine healthcare data and it found in general when telephone triage is implemented and introduced into GP practices, there's evidence of changes in process measures. People, patients speak to and speak to and see a GP more quickly than if a telephone triage approach isn't introduced. It kind of makes sense because the whole point of telephone triage is the day you need to see or speak to a GP. You call up and you speak to them on the phone. But it also found that there were more contacts with primary care. So a value institution has actually found that telephone triage doesn't reduce primary care workload. It's not, nor does it kind of save, nor does it lead to cost savings either in primary care or in secondary care. And the final kind of big thing, the earlier evaluation of telephone triage found was that there's heterogeneity between practices. Actually it's not the same for GP practices. For some it works very well but it's not kind of a magic bullet that's going to solve the problem of demanding, demanding services in primary care. So then we get COVID-19. So we kind of have this context of telephone triage as a service innovation in the NHS before COVID-19. And then with COVID-19 in March 2020 there was sudden shift from mainly in-person consultations in primary care to mainly remote consultations. Here the motivation was very homely not demand management kind of the purpose for which telephone triage had mainly been introduced but health protection to reduce practice but full to reduce risk of COVID-19. It was also a bit messy, there was lots going on in kind of health services and at the start of COVID-19 lots of different models being introduced, lots of other changes going on at the same time. Complex number four, this is ongoing recession, it's really political but I stopped trying to update the kind of political context of this research. NHS England and the BMA are still discussing in press and in public letters to each other about the return to face-to-face consulting and how this is going to happen, how it's going to be implemented. So multimobility, it's a policy priority for NHS England. Multi-mobility being how to manage and plan health services for people living with more than one long-term health condition. A lot of the kind of improvements in care and in clinical practice that we've seen in within the NHS over the last 20 years have come from kind of very guideline driven approaches kind of from evidence that's been really kind of designed for patients with a single long-term condition. But many people are living with multiple long-term conditions and so how to provide effective and efficient care for people living with multiple long-term conditions is kind of a priority. So this was the kind of pre-COVID-19 context and then the post, the COVID-19 context for this is actually multimobility remains a priority for health services delivery and understanding how to plan health services because of the really strong variation in COVID-19 outcomes among people living with multiple long-term health conditions. And then there's kind of some more theoretical context as well. The selfie framework is the theoretical framework. We've been thinking of this within the patient and their environmental kind of with the core of the framework. But then service innovations for people living with multiple long-term conditions are considered at the micro levels of the patient level, the MISA level, which might be the practice level or kind of the national level, which might be the sudden implementation of telephone triage in 2020. Last bit of context, I'm nearly getting the data, which is a good bit. Understanding the inequalities impact of service innovations from the NHS is actually fairly really hard, particularly if you want to do kind of a nice quantitative inequalities analysis. I say here on my slide, the intervention needs to work in the first place. Often service innovations are introduced and they have kind of qualitative or kind of research evidence about the intervention can find that really they've been doing a huge amount that a lot's been going on. The intervention has been making a difference in how services delivered. But actually, when it comes to measuring quantitative outcomes, often there isn't evidence of a really big change in the outcome or in the measures we're interested in. The sample size and the data set that you're using to evaluate inequalities impact has to be large enough. It's kind of a statistician. It has to be at least four times. You're testing for an interaction, not for a main effect. So you need to sample size about four times as large as you would for measuring a main effect. The data needs to be available often if you really want to measure the inequalities impact of a change. The outcome isn't measured or the groups you want to look at aren't measured in your data. They need to measure the characteristics of interest. Finally, there was a recent systematic review that was published at the start of 2021 and actually highlighted that there's not a lot of evidence on the inequalities impact for the start of telephone triage and roommate consulting. So this research comes into the context of wanting to understand how service innovations work for people living with multiple long term conditions and kind of a data context where it was possible to do this inequalities analysis. So the research project comes into the context of where it's possible to address these methodological challenges where often actually it's not possible to look at the inequalities impact of change. We looked at the time taken CO speak to a GP. I mentioned this at the beginning with telephone triage. This is the thing that really changes when a GP practice switches to a telephone triage approach. We looked in GPPS. It's the general practice patient survey. It's an annual cross sectional survey run by NHS England. It's designed to measure care quality and primary care. It's a large science device with about 125 people from every GP practice in England. The data are available. We applied for access from NHS England and it measures multimorbidity. We also looked at post COVID-19 in understanding society gendered this really nice introduction to that COVID-19 understanding society. Again, looking at data on multimorbidity and primary care utilization. We asked, does introducing telephone triage mean people with multimorbidity see or speak to a GP sooner? The analysis of GPPS. We looked at 150 practices that switched telephone triage between 2011 and 2017. In understanding society, we used the pre COVID-19 data for some of the characteristics of participants and we looked at monthly surveys, the six surveys from April to November 2020. Both have information on long-term conditions. We calculated weighted estimates using the understanding society survey weights. I kissed my colleague on the understanding society course to make sure we were using the right weights and used the GPPS weights as well. In the adjusted analysis, adjusting for age, sex, ethnicity, deprivation and GPPS because it's English based on IMD and quintiles of household income from, equivalised quintiles of household income from understanding society using pre pandemic data. We also adjusted for survey wave and for GP practice and GPPS analysis. PPI at the bottom says we had patient and public involvement in this research. We met with our PPI panel before the research started and said what you really need to look at are people with hearing problems. So we looked at this group separately and we also met with them again with our results and talk through and to see what they kind of thought about. Measuring multi-mobility and both understanding society and GPPS. So it's a survey measure of long-term health conditions. In GPPS it's a list of 15 conditions including a long-term mental health problem understanding society and the COVID waves. It's a 26 measure list. We created an ordinal measure of multi-mobility so we counted up the number of long-term conditions peoples that they were living with, 0, 1, 2, 3 or 4 or more. For understanding society we used the question which was asked in all six of the COVID waves, thinking about your situation now. Have you been able to access the NHS services, GP or primary care? You need to help managing your condition over the last four weeks. The response options to this question were in person? Yes, online or by phone only? No, I wasn't able to access. No, we decided not to seek help at this time or not required. It was only asked to people with long-term health conditions or ongoing treatment from April to September and in November it was asked for everyone. So we adjusted the way of analysis and we did various analyses looking at what the impact of this change was. Decided it was okay and kept all the later in. For this question we recoded the responses into four binary outcome variables to look at whether somebody had needed to access their GP, whether they tried to contact their GP if they needed to, whether they were able to access their GP and whether this access was online or by telephone or face-to-face. So results will start with GPPS results and then I'll go back to the understanding society results. So pre-COVID-19 we looked at, we found overall that there was a small difference in the time until somebody could see or speak to a GP, but among people with and without multimorbidity, people with zero or four or more conditions have a slightly better experience, about one to two percentage points better. But among everyone with and without multimorbidity there's a large 20 percentage point improvement after practice changed over telephone triage approach. So everyone saw respect to a GP faster on average after a practice changes to telephone triage with no differential impact for people with multimorbidity. So there's a big change when telephone triage comes in, but it doesn't seem to increase inequalities with people with multimorbidity. This is data, new 2021 data are available for GPPS there today, I got them off the internet. And actually this shows the same thing. This shows that actually during COVID-19 in GP practices, so we're looking at the dark blue bars for here. And actually there's an increase in 2021. People do see or speak to their GP more quickly. And here for a week or more later, there's a big reduction from 2020 in the number of people waiting a week or more to see or speak to their GP. So again this kind of tells us that probably during 2021 during COVID-19 we're seeing the same impact in our GPPS data as well. In understanding society, understanding society. It actually tells us really interesting things about what was happening in primary care during the start of the pandemic. Anecdotally, GPPS weren't very busy in April 2020. It was the start of the pandemic and GP colleagues were saying it was a bit worth the patience. And then because there are survey waves we could look at each way separately over time, see what's happening. So actually each month about 50% of the respondents reported they had a problem for which they needed to see or speak to their GP. And over 90% of people who did try to make an appointment with their GP were able to do so. And these were fairly constant across each way. But only about 20% of people saw a GP face-to-face in April 2020 of those who had an appointment. And this addressed them to about 40% by November. So even in November 2020 the majority of the appointments were still online or by telephone. Katie, two minutes. Two minutes. So one of our findings was that actually in April 2020 only, I apologise for the access on this graph, only about 80% of people who had a problem tried to contact their GP for this problem. And by July 2020 actually over 90% of people who had a problem for which they'd normally tried to see their GP were making an appointment. So it was quite a short-term reduction of people trying to make an appointment. And again, we see that the number of appointments that were face-to-face increased to about 40% by November 2020. For people living with multimobility we find that they were more likely to need to see a GP, but there was no evidence of any difference for people with multimobility whether they tried to access GP if they did have a problem, whether they were able to access a GP, whether the appointment was face-to-face or by telephone or online. So some very quick reflections. Telefon triage is introduced at the practice level or during COVID-19, it's kind of introduced nationally. It had a very big impact on some measures of access to primary care when it started and it can have the heterogeneous impact between different practices. We found people with multimobility are more likely to need to see a GP, but actually we found no evidence pre-2020 or post-COVID-19 that the impact of switching to telefon triage has a differential impact on primary care access to people with multimobility despite this differential level of need. Kind of thought, well actually on reflection this isn't hugely surprising, but what it kind of tells us is the impact of change to telefon triage is large compared with existing inequalities. And my final point is that actually these understanding society data on primary care and primary care and other healthcare acts has been really important. While preparing these presentations, there are three other analyses that have looked at various dimensions of kind of access to healthcare and inequality and access to healthcare during COVID-19, so it seemed worth highlighting them there. And then some interesting methods things. For example, we found that 20% of appointments were face-to-face in April 2020 and actually if you look at the electronic healthcare records, they're saying about 50%. So there's some very interesting kind of things about what survey measures and electronic healthcare record measures of access to healthcare during the pandemic. Thanks.