 Good afternoon. We are going to have a second very rich and interesting session. All healthcare systems face cost containment quality access to healthcare quality trade-off. This trade-off may involve static and dynamic issues. The first and third presentation by Thierry Maniac and Julian Reeve that we shall have the pleasure to listen, the trade-off in a dynamic setting. By dynamic, I mean the basket of technologies has to be defined. The second presentation by Carol Popper on the other hand will analyze the team healthcare production by focusing on the multiple sources of savings for national health insurance system. So please, the first presentation is Thierry Maniac from Professor at Toulouse School of Economics. So this is joint work with Pierre Dubois. I think that there are some slides. The green one, okay. Oh, the green one there. Sorry. So we want to talk about something that has been talked about this morning is the optimal way of spending expenses on drugs when there are dynamic elements. So instead of having a bargaining game between health authorities and pharmaceutical firms, annually we could have a mechanism where there are long-run commitments between these companies and the health authorities. In the literature, there have been a series of papers showing that there could be gains from trading of the expenses of a time in a better way than it's done now. But these gains depends on diseases and drugs and I'm going to take the case study of hepatitis C in order to show how we could think about the trade-offs that these intertemporal smoothing of expenses of a time as... So let me just remind you and I know that you all know this but the structural characteristics of hepatitis C so it's a non-explicit disease, at least in Western Europe since the 2000. There were ineffective drug treatments until the 2010 and there are a lot of asymptomatic cases of this disease and the effective drug treatments were introduced in 2014, also not for all virus genotypes but one characteristic is that the new treatment was more costly than the traditional one and there were also at the end of the stories or not the end but after that other treatments introduced between 2014 and 2020 adapting to different genotypes and cocktails of drugs and all this kind of thing. So the objective of this paper is to try to look at the optimal assignment of expenses of a time in order to deal with this epidemic. So in a sense what we are trying to do is controlling an epidemic here and the important thing in this modelling is the trade-offs between current expenses and future expenses and we are going to try to look at this problem in a framework in a model which is the susceptible infected recovered model, the standard epidemiological model with this kind of epidemic. So it's also a way of re-questioning the measurement of cost-effectiveness because you have dynamic externalities due to the infection. So we are going to look to have these sort of models where you have susceptibles ST infected IT and unchecked or undetected infected UT and it's going to be an important part of what we are going to say about hepatitis C later. We have a natural remission rate but I won't talk about that and the main assumption that we are going to have and that's typical of these models where you have intertemporal trade-offs is that you are going to have decreasing returns to treatment by the new drugs so the curing rate is going to be an effective and non-time-varying increasing and concave function of expenses per patient. So there are various... If you don't have this sort of assumption and I'm going to give you some arguments behind this assumption, the best way to control the epidemic is to spend all the money you have now and kill the epidemic and then everybody is better off in the next period. But it's not going to be the case because of these decreasing returns to scale assumptions. So why do we have this assumption? So there are medical justifications for deaths. The first one is heterogeneous treatment effects and it depends on virus genotype in the case of hepatitis C. There are also justifications due to the organizational capacity to identify the patients who benefit the most from the new treatments and in particular in hepatitis C, the undetected have a role of patients who could be cured but who don't know they are infected and so it's difficult to identify them. You will have to check to test everybody in the population. There are also economic justifications. I'm not going to spend too much time on that. We have another joint paper where we have a bargaining pricing game between health authorities and drug companies. There is also an argument where the health authorities might think that there are more effective innovations in the future which is going to give another type of trade-off that you have between spending now and spending in the future. So the dynamics of the epidemies is very simple because it's a non-expansive disease. So the long-run stable stationary equilibrium is disease-free and what we can do is just to accelerate the rhythm at which the disease disappears. So health authorities are supposed to be given a certain endowment, A1. That's the difference between the current situation and what we imagine here is that they are given an endowment that they can spend over time but in a way they decide for themselves optimally, maybe. And so we're going to say simply that there are certain expenses every period, capital BT, and they are going to be spent and used in order to curing patients with the new drugs. So implicitly the cost of the traditional treatment which is not effective in the sense of not curing for good the disease is also implicitly included but I'm not going to talk about this. And one important assumption is also the social welfare. So the social welfare is going to be a convex function of the infected rates. So we care about infected people in the population and not, in a way, which is the more so they are more infected but also it doesn't depend on undetected and checked and it could be questioned, obviously, as an assumption. Okay, so that's typically what you get when you simulate. We simulate the model with French, calibrating with French data. That's typically what you get. So on the vertical axis, you have the rate of prevalence of undetected infection and 1-susceptibles, hopefully that's the rate of prevalence is quite low. And the blue curve is the rate of infected, the solid blue curve. So that's the natural rate of disappearance of the epidemic. And when you have a policy enacted at period 5, you see the dotted curve in blue and it's just accelerating the rhythm at which the disease disappears. So on the unchecked and the non-susceptibles, the effect is not that great, large. So the effect, the first effect of the policy is on the infected, obviously. So what we're going to do first is to try to see if a tool, which is the variational calculus, is useful in terms of deciding about the optimal policy. So the idea is to take the expenses of the health authorities as given. This is a flow of expenses, B1, etc., until infinity. And there is a constraint is that the discounted sum of these expenses should be equal to the endowment. You spend all the money you have until the end. And the variational calculus means that you're going to play with the fact of moving expenses from one period to the next or from the next period to the present period. So you increase dbt by a positive number and you decrease the expenses at period t plus 1 by the discounted value in terms of just keeping the expenses of a time constant. So we did that and it's not a very helpful tool in this case. You can see that infection decreases in period t and increases in period t plus 1 if you reallocate all of your budget from t plus 1 to t. You can compute the indirect benefits, etc., but there are complicated dynamic effects. It's almost impossible to do the analytics of this and no clear-cut predictions. And so we have to resort to simulations in order to try to understand how this tool can be used. And the type of thing that you get is this graph. So the scales of the different curves are not comparable. So on the vertical axis is the impact. For example on the red curve this is the impact infected at a reallocation at period 9 from period 9 to period 8. And you see that in this case you get a strong decrease of the infected so you increase a lot the budget in period t but you have a rebound effect because you don't spend anything in period t plus 1. And so you have this and you see that the long run effect it's not clear on the infected. It's very close to zero. So I'm not going to comment the other curves but they are also interesting if I add more time. So this is a tool which can be used but we can ask ourselves what is the optimal budget policy in this case. So the idea is to try to compute the optimal timing the timing of optimal expenses by the health authorities and so the dynamic of infection is given by the epidemiological model and you have an optimal control of the infection. So technically it's using simulations it depends on various hypothesis in particular preferences for the present of the health authorities and we are going to compare the optimal policy with a very conservative policy of consuming interest of the endowment only. A constant policy every period which is the current situation let's say and the optimal policy. So this is what we get so the vertical axis is the asset depletion so that's the asset is equal to 0.005 at the beginning. This is roughly 3 times the cost with the traditional drugs until the disease disappears and then the optimal policy is in solid black the constant policy constant policy every period is the red dotted curve and the conservative policy is the little dots. Respect to the constant policy there is a lot of front loading with the optimal policy you spend a lot more at the beginning in order to try to control the infection. Much more than the constant policy constant policy is just quite conservative and the interest revenues is even more conservative. So how does it impact the rate of control of the infection? So these are the same curves so the solid black is the optimal policy the constant policy is the red dotted curve and the interest revenues are going to is given by the small dots. So you see that the optimal policy impacts the infection much more than the constant policy so you spend more money but the gain from the optimal policy is broadly speaking the area between the black curve and the red curve so that's the gain in social welfare that you gain from this optimal policy. An interesting aside is the rebound obviously at period 10 we have seen in the previous how do I get back so you see that after period 10 you don't have more money to spend so you are out of money there and so there is a rebound of the infection from period 10 and that's built up in the social in the optimal policy construction it's also the preference for the present which is playing a role here while with the red curve you spend so in fact the net effect is in a sense the discounted areas of the difference between the red curve and the black curve. Looking at the unchecked there is not a lot of difference so controlling the epidemic through this is not very effective on the susceptibles that's the same thing it's the non-susceptibles and so in terms of conclusions so the first the first effect is that a disappearing epidemic like like hepatitis C an equal budget policy certainly dominated by a front-loaded policy because you try to control the epidemic at the same time as you cure patients this result is really calibrated on the hepatitis C case study so it's really difficult to prove any external validity of the results that I've described to you and it depends very much on the main trade-offs so spending more today implies not only less infection tomorrow and some dynamic anxiety but also less effective cures but there are strong arguments for this and there are other things to do for example no feedback on the innovation process for the moment but we're still working on this thank you it was an inter-temporal timing ah yeah yeah you mentioned that maybe we may pay some attention to you about regarding new treatment entries so I wanted to know what would be your recommendation policies to achieve this optimal pass ok so starting with you you you you you you you ok so starting with the second question Jean-Tierrault was speaking about the soft budget constraint this morning and this is exactly what you see in the graph the rebound of the infection is due in the model to the hard budget constraint which is in the model do I believe in this? No I think there are going to be renegotiation what we do here is really normative analysis of what could health authorities with all commitment possible could do but it's certainly not the case in reality so that's an important element that should be taken on board after a while so one way of doing this is that to put constraints on the amount that should be left after a certain period of time and in particular using the information about the undetected so you can have guesses about the rate of undetected in the population which is really fueling the infection that's the channel through which the infection is going through obviously there are other ways of so you might say that there are migrations between different countries and it's the case of a single country so your first question is about how do you deal with the optimal tradeoff in fact we don't do it very rigorously at the moment so we just assume that it's a fix non time varying function of decreasing returns to scale but probably it is not because the population is changing the rate at which the undetected can become detected affect this function because they are newcomers in terms they are not people who we know that they are with hepatitis C since many for many years so we don't do it very well but the problem is also the data it's extremely difficult to calibrate this function so calibrating a time varying function with the quality of the data that we have is really I don't think it's possible at the moment Questions, Katarina So I'm curious about the prices that you use to deal these predictions the prices we don't have prices it's just an issue of the health authorities announce a certain health certain level of health expenses and that's what is going to be the budget to be spent and prices are implicitly constant it's really a pure welfare analysis of the technical analysis I guess that a payoff of delaying is that expecting prices to decrease by half or sometimes four times So you mean we didn't compute the supporting prices of this kind of inter-temporal equilibrium level of expenses but we could that's a good point but we didn't So you mentioned that if I understand that you're judging the optimality of these policies by the number of infections that are being reduced if we also care to reduce the number of asymptomatic or unchecked individuals does that affect the conclusions yes it does what is the qualitative effect we didn't do that so we have to we have to investigate the issue but certainly it does the difficulty is also that we have to test in the population for this and this is not included in the model so an optimal strategy would also be a testing strategy I think and we should develop more this line but we didn't Just a clarification on your question so we don't have price but we have the cost of treatment as we observed in France so in the way we calibrate the effectiveness of the budget you know this increasing concave function is calibrated using the cost of treatment that was used in France that was observed in France which depend on the price of all these drugs but we have no as Thierry said we have no prediction on the future price we just use this to calibrate we assume it's constant yeah that's the that was the question of the curing function is constant we have even issues with calibrating this constant curve so we would have more issues with calibrating time varying kind of curves so if I understand the correctly you're using a fixed budget that the government can use to treat this disease what would happen if there comes another disease that would make fewer budget and more budget on that new emerging disease worthwhile from a social point of view because then in your optimal scenario all the budget is spent and nothing is left so the thing is that what is decided is the endowment at the beginning because to allow a smoothing consumption of the time means that you need to delegate to the health authorities the spending the smoothing of the spending of the spending of the time so this is an endowment given to the health authorities so obviously any event in the future which make this the importance of this endowment socially cannot change it's a given so yeah that's a restriction but yeah that sets the question of optimally deciding on on the level of the endowment at the beginning and maybe revision but it's difficult to it's difficult to yeah we can talk about this but it's difficult to do Last question maybe Thank you very much Julien from LEM Thank you very much for this presentation I wanted to come back to the question of the price of the time understand that you assume that it was constant of the time to treat the single patients from hepatitis C as we discussed this morning we know that there is a decreasing curve of price treatment in France so what's interesting is that for the same price you may get marginally better product in the future that will actually enable you to treat better those patients is this something that could be tested in the model or is it too specific already to the treatments available No it should be tested absolutely that's related to this curve of efficiency of the treatment in fact we make it constant of the time but it's probably not we don't think it is in this curve there are trade-offs well I've been talking about these trade-offs and there are many things in this curve so it would be interesting and that's what we try to do in a companion paper is to try to model the bargaining game between health authorities and firms and so prices will appear at that stage but at the moment we are very reduced form we have very good quality data in order to fit this curve so we maintain the assumption that it's constant but it's probably not so one conclusion of this work is also that there are many parameters we have to fit many parameters in order to try to understand this in the number of trade-offs so any general conclusion about this should be taken with a lot of questions I guess my point was that the rebound effect that we see might be less marked if you take into account the fact that the price is decreasing maybe maybe no more questions ok thank you very much now we have the pleasure to listen to Carol Popper I'm not going to do a brief summary of Carol Popper's CV she's a professor at the imperial college London she's or she was a member of president's Macron excellent expert commission so Carol thank you very much so this is a kind of complete switch of topic and I'm not actually going to talk about savings at all but I'm going to be talking about the production which you could then think about in terms of whether it engage savings so what I'm going to talk about is team production in healthcare and the reason this is joint work with two colleagues from the institute for fiscal studies and the kind of full title is we're interested in team composition and evidence from nursing in the English NHS and I'm going to hope that I kind of convince you why this is a kind of interesting topic to study so team production is interesting I think in itself they have a big role in many forms of production teams have advantages they allow specialisation they allow knowledge sharing they allow complementarities but interestingly despite its pervasiveness team production has been relatively understudied in economics from except in two aspects and these two aspects are that most studies focus on the impact of team organisation on the behaviour of individuals so there's a huge literature obviously on peer effects and you can think about a classroom or you can think about peers at work and peer effects and then there's a large literature on financial incentives how giving financial incentives to team members affects the output of a team and a classic sort of set of examples of this are the kind of studies of strawberry prickles by Ariana Bandieri and her colleagues a lot of those studies for example are actually studies of individual workers who are relatively low income and low skilled but in many cases the output is genuinely collective I guess a kind of classic example of that would be sport team sport and understanding how the team operates and who the key players in that team are important and healthcare is a very important example of team production there are many teams in healthcare and it's a growing belief that team production is important in things like the treatment of cancer that previously were thought of as you know you had some great doctor who did something and then other people who did something after him but there's been a shift in many areas to thinking about team production because of this issue of specialisation and knowledge sharing and complementarity of skills our focus as I say is on team composition in inpatient hospital care so we're restricting ourselves to a particular setting which is inpatient hospital care and nursing teams the reason is that healthcare workers are around 10% of the labour force in many economies and nurses are the largest group within that in the UK which is what I'm going to study there are over half a million nurses and nursing support staff and also a worldwide shortage of nurses that's been going on for some time and so it's important to make the best use of the range of skills that there are within the nursing labour force and one then on the other side one key feature of hospital nursing is the organisation of work patient care the work being patient care into teams output is collective and individuals rotate between teams in an inpatient setting due to the need to provide 24-hour care and individuals don't work 24-7 so teams are subject to both unplanned and planned changes and that allows us essentially we can exploit the changes in teams between what the optimal team ought to look like and what the actual team on the day in our case 3 months later does look like and what we do in this paper is essentially exploit those changes in order to determine which members contribute most to a team so what we're going to do is our focus is we're interested in both quantity and quality of team members so our focus is how different levels of skills contribute to team output teams consist of nursing staff with more less training that's common worldwide nurses are at different levels of seniority but there are also kind of two big groups of nurses there are nurses who are essentially nursing assistants who do not hold in the main degree level qualifications and nurses who essentially hold degree level qualifications and have much more medical training one of the trends worldwide is because of shortages particularly in the first type of nursing is to substitute degree trained nurses at the margin with non degree trained support staff that's and we're going to be looking explicitly at that but the other thing about these teams is they consist of individuals with more or less experience just thinking about how long you've worked as a nurse but also more or less familiarity with the firm that you work in the hospital and with other team members because of this issue of rotation and because of the issue that a nurse might be have worked 20 years in her life but worked it at 3 or 4 hospitals so what we're going to do is we're going to use data at the 24 hour level so we define a team as 24 hours of patient care and we're going to link those patients that are under the care of that team to the nursing teams responsible for their care we're going to take as our case study one large English hospital group that consists of essentially three separate sites and we have about 44,000 patients about 3.5 million years hours of patient care we look at about 4,500 uniques staff working 300,000 shifts in about 59 wards in 3 hospitals so we have very detailed data on who cares for which patients and what we're going to do and I'll kind of flesh this out a little bit but is to exploit exogenous and we think plausibly random variation in the size and composition of the team on one important outcome of the team which is patient mortality there are lots of reasons that people in healthcare when they're looking at production focus on mortality one is it's not a good outcome patients care whether they exit the hospital alive or dead but the second is that it's actually quite difficult to fake and when you're worried about things you can't you don't want things that can be manipulated by the team the third in our context is because it is what's called a never event it's not meant to happen though it does it's recorded to the exact time at which it happens so it allows us to link it to the appropriate team so what we're going to do is we're going to examine as I say two things the effect of team size so teams vary in size not because they're planned to vary in size but because people are unexpectedly or expectedly absent and we're going to ask the question does it matter if the team's not fully staffed and secondly does it matter if which staff member is missing now clearly at a kind of extensive margin if you had nobody in the team it would matter but what we're asking is essentially at the intensive margin at the marginal change in either numbers or a marginal change in the type of staff who are there for unexpected reasons what impact does that have who matters most and as I say it's important because teams contain a mixture of skills and one response to shortages is to replace highly trained nurses at the margin with less skilled nursing assistants the second set our data allows us to ask is how much does familiarity with the hospital, the ward and other team members matter so in the broader language of firm familiarity plant familiarity and individual familiarity is kind of if you want to link it to the broader issues in kind of thinking about team production and that's important because one of the responses to shortages in hospitals across the world and particularly the developed world is to use agency nurses nurses who are not employed by the hospital and therefore not familiar with either the team or the hospital in which they're working so a kind of quick look I'm not going to kind of there's lots of gory details behind this but I'm not going to walk you through those but just to kind of point out that we have three hospitals about a third of a two thirds of individuals in the year we look at, we only look at 2017 work in one hospital but the remaining two thirds rotate and then we look at the number of units which is essentially the number of wards not teams but wards worked in and we find that band six work in about six point five units the lower band nurses are here on the left the lower band nurses work in more but there's quite a lot of rotation so individuals don't always go to their place of work so it's not like economics departments where you always go to the economics department in this case you're sent to the politics department okay so we have as I say 53 wards in three hospitals teams on average, that's wrong teams on average have about 20 nurses 75% of whom are degree qualified the average experience in hospitals about six years the average number of patients is about the same as the number of nurses about 20 and on average a nurse has worked for 30 days out of the 90 on a particular ward so kind of what do we find first of all these changes are at the margin and we're not meant to have any shortages the system sets an optimal number of nurses who are rostered three months in advance to these wards but because of unplanned shortages somebody's sick, somebody's child is sick, someone sent on a course when the actual date comes you have shortages and we do find that unplanned shortages of nurses are associated with increased patient deaths however this is only driven by a shortage of nurses with degree level training about two thirds of the nurses on a unit have degree level training one third don't an unexpected absence of a support staff worker makes no difference to a patient dying or not secondly there are strong returns to seniority one degree nurse absent for a 12 hours the average shift is 12 hours from an average team of 20 nurses results in a 10% higher probability of death but if the most senior staff are absent these are nurses with considerable amount of both extra qualifications and longer service they their impact is about twice that of a newly qualified nurse and then finally we find that experience in the hospital matters experience with your team members matters less and experience on the ward matters so people can move from say a hematology ward to another kind of ward but what really matters is experience in the hospital or in the firm the longer a nurse has worked in a hospital the more her absence matters but again the length of employment which is about the same nursing assistants as nurses qualified nurses does not matter for nursing assistants in other words nursing assistants at the margin are pretty fungible so kind of policy implications this is I understand a policy conflict so I thought I'd have some policy conflict so the increased use of less qualified nurses is likely to continue given nursing shortages and tight budgets on margin our results indicate that that is probably not a good policy less qualified nursing assistants are not effective substitutes for degree qualified nurses and that means that if you did push that more a change in skill mix could reduce quality of care the other thing is that this is an industry in which people leave they leave after they're first trained they leave often when they don't want to work shifts anymore and they are conflicts with family responsibilities and so one of the things that you might take from policy is whilst there's a worldwide shortage of trained nurses and trained nurses one of the things that the hospital that we're looking at might do is try and invest in retention of degree trained nurses rather than substituting them with either agency nurses or nursing assistants so thank you for your attention thank you very much Carol for your presentation maybe I will start with a very short question would you be kind enough to give us some details regarding on what you call plausibly random variation because regarding the size of the teams it makes an option that makes sense but maybe regarding the composition one may suspect that there were some criterions when you adjust upward or downward the size and maybe these criterions are related to some a priori maybe connected to your results or maybe not well we spent so the short answer because if you're worried about endogeneity of patient severity we have about 27 checks that shows essentially that patient severity i.e. the probability that a patient might die is not related in any way to staff absences planned or not planned or to team size planned or not planned the kind of backing back from that so we spend a lot of time being economists worried about exactly that but there's check after check of that because that's key our key assumption that allows us this is that there is exogenous variation in the staff there and the types of staff there but just backing back a bit what happens in this is team rotas are set 3 months in advance by the senior managers in the hospital they're set together so each type of ward has a given staffing complement of both nursing assistants and registered nurses and then within registered nurses levels there are essentially 3 levels there are 4 levels of registered nurses that's set 3 months in advance so that staff know 2 months in advance there's a bit of kind of argy bargy around that that's set 2 months in advance for the staff so they know which shifts they're working if you work with European working directives you can't work all the time and as I showed you nurses work essentially a third of the time on award so then the issue is why are people absent well they're absent as I say because of short term sicknesses and you can't replace them they're absent because of things like maternity leave people suddenly find they need maternity leave want to take maternity leave earlier than they said they were going to take maternity leave or they might come because you have a trust wide training day for senior nurses and half your nurses have to go to that but as I say the main identification assumption is that that is exogenous to the severity of the patients that you have on the ward and that it's not caused for example a death the day before doesn't cause people to take the day off afterwards and the answer is they don't thank you very much and just a last point before sharing the questions about gender issue because in experimental economics they focus a lot on if there is more or less cooperation within teams according to the composition of teams I don't want if you have checked this we haven't looked at gender at all we've looked at team familiarity so whether you worked with a nurse in any setting in the hospital in the last 90 days we only have a years data in the last 90 days before you before the events that we look at we don't look at gender we could look at homophily and gender it's just not been a focus of what we've been interested in team familiarity because of this issue of handover now clearly men may never handover to women well or men bosses may be different so one of the issues we are going to look at because we found such returns to senior staff we're going to look at the value of bosses in a kind of AKM type setting but a very different with a second paper and there I think we will look at whether you've got male bosses or female bosses and how that affects team performance I don't want to take the mic from anyone else so that was terrific this is great I really like it I had a question I just looked it up I was just reading this paper last week that I didn't know about but it's over a decade old in management science they sort of in healthcare differentiate between the sort of learn what versus learn how of healthcare teams and it strikes me sort of in the spirit of second papers I think this is already the first paper but it seems to matter a lot for the managerial implications for the nurses if what they're doing is just is it the case that they have a better grasp on the current best practices for treating those patients or is it the case that they're just better managers and mentors of the younger staff because sort of if it's the former actually the policy prescription is just like making sure everyone has more information about guidance from medical professional societies if it's the latter that's like much harder to the sort of mentorship piece you can't just substitute with teaching or something I mean I think that's super interesting I'm not sure that we can actually get at that with our data but we have been playing with ideas as I say we've been going we've been wanting to look at the value of bosses so we might extract a boss fixed effect using AKM because genuinely these are kind of random variations and then kind of look at what that boss has done in terms of her time in the hospital how much management she has in other words perhaps how long her staff stay with her though that's not a choice she has because people tend to be are rotated by this system so we can try and get at some of that and I think it's super interesting what makes a good boss in this context and that has implications or for what you want to do about staffing indeed so and then we probably also want to do a kind of back of the envelope allocation that if you allocated individuals to bosses who were good compared to not what kind or the other way around what kind of gains or losses might you make we're pretty sure that there's going to be quite different differences whether we can get at exactly why there are those differences is another matter and I have just one sort of half baked comment and then I'll pass the mic one person to the left you talked a bit about this margin of substitution sort of at the bottom of the skill distribution and so I am completely I believe the results that you showed us it doesn't seem like adding sort of more administrative staff helps but what about this margin of substitution that we often talk about between things like very highly skilled nurses versus physicians are you able to see that in the data? So we assume this is just an assumption we can't actually test even at all is that the physician rotation is exogenous to the nurse rotation at the ward level and that I mean given that how junior doctors are allocated junior doctors essentially in the UK system work six months in a specialty senior consultants hardly ever come to the ward they do a ward round every morning when they're followed by all their little acolytes they have to follow them because they're the great god but in oncology you never get an oncologist on an inpatient ward so these essentially are wards run by nurses are nurses just a correction are nursing assistants are not admin they're people who change grips, catheters, beds take patients to the toilet they're basically workers who would otherwise work often in the social care sector the nursing home sector so they're patient focused they do patient focused things but they're not sort of correctly treatment changing bedpans is sort of important to do correctly but it's not they wouldn't be fitting catheters and other things but they would be they would be making sure that patients don't fall out of beds don't fall on the way to the toilet I mean that sounds kind of trivial but actually patient falls are really important again another never event in a hospital so there's just a quick follow up do you ever see in the data a less skilled nurse being replaced by a nurse with higher skills we don't, what we see is the team composition so the team yeah so we've got essentially you never see a fill rate so the planned is known as the planned and the fill rate in this hospital is the ratio of actual to planned there are never more that fill rate is never more than one I see but of course you could have I mean that's what we are exploiting here a day in which your missing nurse was a band six band five and above the degree qualifier and you didn't have you had a 100% fill rate for your nursing assistants and in fact the fill rates are orthogonal to each other they're not correlated so it's not like you have a run on that ward where everybody runs away madly which is probably why we find that when we do all these tests of exogeneity it holds but yeah you never get more than one there's such a shortage it's hard enough to find so the average fill rate I mean you're usually about 4% down on average I was trying to understand if it was purely about quality or more about team production you mentioned team familiarity I probably missed it what is the effect of team familiarity we do find some effect of team familiarity for again only skilled nurses team familiarity is Nurse J worked with Nurse I in the previous 90 days to the ward she's working on anywhere in the hospital that team familiarity does not matter at all for less skilled nursing assistants it does matter for band the degree qualified folk but it's kind of if you I'm not I didn't present our estimates but there's quite a lot of overlap in the confidence interval I think even with you know 44,000 patients and all these shifts we probably nearly need another year of data but since it took me three and a half years to actually get this data it's probably going to take another three and a half to get that 2018 so we can't say anything but there are certainly hints that team familiarity does matter ward familiarity less and hospital familiarity clearly does and that makes sense because in a sense it's the top hospital management that kind of decides often on policies and how things work and that's implemented by the teams thanks thank you very much for this study it's fascinating have you had a look at team variability to the outcome of patients meaning when I look at the French system the team that will be on shifts and if you have 16 nurses in a team will constantly change from day to day people will put into shifts with other people but the the same team will very likely not be together fully again until several weeks later have a look at how this affects patients outcome that's exactly what we exploit we exploit the fact that a team today won't contain the same people as the team tomorrow and in fact it may be several months before the exact same team is operating so that's exactly what we're looking at ok so the the event of the same team being together is not occurring frequently enough to see if a single patient no we haven't looked at whether the exact composition it doesn't happen enough as I said a nurse works on average only a third of the time in 90 days and we've only got a year so the exact team I'm not sure how often the exact exact team occurs particularly because there are team absences thank you and for the last presentation we have the pleasure to listen Reif from a professor at University of Illinois who had a good idea to spend one year to do the school of economics no regrets well thank you so much for inviting me to talk today about my research on the value of life so this is based on work that I've co-authored with Danny Bauer and Darius Lachtowala on a talked today a little bit about what I think the implications of our research are for some important questions and healthcare policy so many of you are probably familiar with the notion of a value of statistical life economists generally define it as the amount of money that a large group of individuals is willing to pay to reduce a health risk that's expected to kill one of them so this value is used widely in a bunch of different areas ranging from environmental hazards to medical care to public safety and so on so for example the US Environmental Protection Agency assumes a VSL of 10.7 million dollars or 10.2 million euros when deciding whether or not a particular environmental regulation passes a cost benefit test and VSL was also used to inform the cost effectiveness analyses that are performed by a large number of different health agencies around the world so for example a recent study by Terrad and co-authors used VSL to estimate that the French value of equality that's a quality adjusted life year lies somewhere between 150 and 200,000 euros so in other words this implies that the French healthcare system should be willing up to this amount for medical treatment that extends life by one year unfortunately the model that underlines these calculations has some shortcomings and in particular in this conventional model individuals can only be in one of two different health states they're either alive or they're dead and so this has at least two important consequences first it means the model has nothing to say about whether VSL varies with underlying health can't tell you if it's higher for a sicker individual versus a healthier individual in addition the model can't distinguish between preventive care and medical treatment and that's unfortunate because it means that it can't answer a number of important policy questions so for example researchers have long observed that societies and indeed even individuals appear to invest less in preventive care than in medical treatment now is this something that is undesirable or inefficient the conventional model can't answer this question because it can't distinguish between preventive care and medical treatment in addition the model has nothing to say regarding whether or not reimbursements should be more generous for medical treatments of more severe diseases so in countries that practice strict cost effectiveness analysis they generally take the view that a health gain of one year should be worth the same to a mildly ill individual as it is to a very sick individual but if you ask people on surveys they reply very often that one should prioritize extending the life of a very sick patient ok, now which of these perspectives is more consistent with a standard economic model so that's something we can begin thinking about with this framework that I'll be describing today so at a high level what we do in our study is extend the conventional model to accommodate multiple health states so in other words rather than just being alive and dead living individuals in this model can reside in an arbitrarily large number of health states ok, so we denote that arbitrarily large number as N and so for convenience we'll call the state of death a state N plus 1 ok, so you can think of these health states as corresponding to different illnesses or something more finely grained and a health shock essentially in this setting corresponds to transitioning from one health state to another health state essentially an individual's quality of life and their probability of dying will depend on what health state they happen to be in currently so we use this setting to extend the idea of VSL to a more general concept that we term the value of statistical illness or VSI so in words VSI is the value of reducing the risk of transitioning from one health state I to another health state J so VSL will be identical to our measure of VSI when that state J is death but VSI is a lot more general so it allows us to begin posing some interesting questions so first of all VSI will allow us to compare risk reduction values across people who reside in different health states so for example is VSL higher for somebody in a very sick state as compared to somebody in a relatively healthy state and that's important for understanding whether there is a severity premium whereby it may make sense to pay more for treatments of very severe illnesses I'll come back to that in a couple slides in addition VSI will allow us to compute how much a healthy individual is willing to pay not just to prevent their risk of death but also their risk of contracting other illnesses such as Alzheimer's disease diabetes etc so that's the theory in the paper anyway and the final part of our study we applied this model to data to try to assess whether or not the insights from our model actually generates some empirically meaningful results so we work with some rich survey data from the United States that provides us with information about people's mortality probabilities their quality of life and crucially information on how these vary by age and by the number of comorbidities that individuals have and then to keep the model tractable we take all the individuals in our data set and divide them up into 20 different health states where a health state depends on the number of chronic conditions you have so for example diabetes cancer as well as the number of impairments you have so do you have difficulty walking difficulty bathing etc so this slide just gives you a brief snapshot of the data that we use so it presents some summary means for the 50 year olds in our data set so these means are broken down by what health state the individual is in so depending on that health state life expectancy for 50 year olds ranges from 31 years for the healthiest ones down to 9.1 years for people in the sickest health state we also measure their quality of life using a health index which ranges from 0 to 1 so 1 basically index is perfect health these are individuals that can run marathons they have no pain they are in perfect physical condition and it ranges down to 0 which is the equivalent of death and then we also have information on how much people spend on medical care as a function of their age and what health state they are in so before I show you our main results I want to just give you a hypothetical example of the kind of output our model produces to try to convey some of the intuition behind what we are doing so this slide plots via cell for a person who was initially healthy at age 50 we assume their VSL was 6.8 million dollars it's easy to use a different value if you prefer as is typical in these models VSL declines with age but then this individual suffers a health shock at age 60 and a severe health shock at age 70 and you'll notice that in both of these cases VSL deviates from this trend and actually increases now what's happening here is as individuals life expectancy was significantly reduced following these health shocks and this means that it is now optimal for her to spend down her wealth in her fewer remaining years that increases her willingness to pay for lots of things including health and longevity now in our paper what we do is we repeat this exercise for a population of individuals representative of the US population so this is the result of that exercise for that population the solid blue line is average VSL for individuals between the ages of 50 and 80 and then these light blue lines show you VSL for selected percentiles in this population so one takeaway here is that even though we set this up to look initially only at people who were healthy at age 50 there is already a lot of heterogeneity 20 years later all of that heterogeneity is coming from random health shocks that people suffer throughout the course of their life with these data in hand we can now calculate a bunch of statistics of interest and for the sake of time I will present just what for us these was the most striking result so this graph here plots the value of a quality adjusted life here for the 70 year olds in that data set as a function of what health state they're in so recall there were 20 health states in this model and I've arrayed them here in order of increasing life expectancy so for individuals who are age 70 and in the best health state so they have a life expectancy or a quality adjusted life expectancy of about 11 years their value of equality was $260,000 however this increases all the way up to $660,000 for individuals in the worst health state where their quality adjusted life expectancy is less than 3 years so in other words among the individuals who are in the worst health they're willing to pay more than 2.5 times more per quality than individuals who are in the healthiest state so they have a couple minutes left to just conclude with a few thoughts for discussion so first I want to return to this question about preventive care as I mentioned it remains a puzzle why so few people and systems appear to invest in prevention and I'll give you an example to make this concrete there's a lot of evidence that lifestyle modification programs for diabetes are in fact quite effective at preventing diabetes we had people at risk of diabetes appear uninterested in these programs in my own research on workplace wellness programs there was a lot of initial excitement and hope that these programs would increase the health and well-being of workers by making it easier for people to engage in preventive care at the work site but the best evidence we have so far to date suggest these programs do very little if anything and participation in these programs remains stubbornly low so our study provides a simple rational explanation for these previous findings and it's that individuals simply value preventive care less than medical treatment and thus we shouldn't be surprised to see under investment in preventive care especially when other potential explanations such as the fact that insurers may also not face the proper incentives to pay for preventive care so maybe the consumers are making mistakes on top of this those explanations are only going to reinforce this result that we find and then finally we think our study also has implications for the reimbursement of health care so as I mentioned before in standard cost effectiveness one assumes that there was only a single value for quality adjusted life here but this is very counter intuitive right if you talk to people they very frequently say it's important to prioritize the health of the severely ill and this notion that we should pay more for the treatments of severe diseases this is sometimes called having a severity premium has already found its way into some policies so for example the UK several years ago established something called the new cancer drugs fund which pays, basically sets aside extra money to pay for cancer drugs over and above what the standard cost effectiveness threshold is this was very controversial so I'm showing here an editorial that was published in the Lancet arguing against this fund the Lancet editorial basically argues that this is not an efficient use of money right it's a waste of resources essentially from their perspective any ad hoc adjustment to the cost effectiveness threshold is necessarily inefficient and subjective but you know our framework suggests otherwise it suggests that in fact there may be a very good reason at least from an economic perspective to reimburse more generously for treatments of severe illnesses and you know there's still a lot of work to be done incorporating the insights that we've generated in our study into the standard practice of cost effectiveness we think already these preliminary results do suggest or provide some support for the idea of a severity premium and so I think I will end it there and open it up for questions and comments Thank you I was wondering how this fits this fits with the notion that actually you may want to save people who have a long expected life I mean think about Covid for example there was a shortage and then they prioritize the people who had a long lifetime expected lifetime over the people who are going to die anyway so it doesn't quite fit so the question is actually it doesn't reason it to me at least it doesn't seem like it's reason it to either the kind of preference for the severity premium as you call it it's kind of strange is that because there is an imperfect financial market so I have a lot of money and I hear that I'm going to live for only one year I'm willing to spend all that money which would not happen for example if I had a monthly pension and so on I don't know what's going on there it's kind of strange there are a lot of points you brought up there so first of all to be clear we're focused here on the value of a quality adjusted life year so this is per unit of health an individual who has 10 qualities left in their life that's going to be worth 10 times more and that's the basis of what we think of reimbursements you're correct that imperfect financial markets are very important for this result they have significant effects on individuals willingness to pay because it affects how much money and wealth and savings they have there are some alternative ways to think about that that we're thinking about for follow up pavers in terms of different welfare functions one might use here we're using just willingness to pay sort of set up in a real world setting but I agree that sort of bigger picture welfare questions are very complicated here and so we have not fully waved into that literature yet Luca Katarina I had a very similar question so I'm just going to expand on it a little bit maybe actually more of a clarification then it seems like there is maybe a fixed correlation between life expectancy in the states that you define and also the quality of life adjustment parameter does that correlation matter for your result you could think of those two things as moving completely separately I could have a disease that really affects me but doesn't affect my life expectancy vice versa does that matter the short answer is no basically if you assume quality of life does not change in this setting it has very little effect on our qualitative results at least using the typical parameterizations that are used in this literature a big challenge is that we don't really know the effect of health on someone's marginal utility of consumption that's an open question am I more willing to spend money if I'm very sick well maybe yes if I have to buy you know new I have to retrofit my home maybe no if I'm just bad written and have no use for money but that's not what's driving our results so in our paper we dig into that thanks Katarina and then so your research basically is is a preference I mean people reveal their preferences right but then the thing is how do you aggregate that on a social welfare fashion I mean because of course if I'm doing tomorrow I'm willing to spend all my money today perhaps not but I mean another thing is like from a societal point of view where do I want to allocate money because those are the preferences of people but then what do I want to do from a collective perspective yeah this is a good question it's related to what Jean brought up here we're just taking the stance of measuring this using willingness to pay which is what many people use it's not the only one and so we can look at alternative frameworks of something we're interested in what I will say this is also in response to Jean's question you can also use this model to for example look at a single healthy individual and just say what is my willingness to pay to reduce the risk of different illnesses and here you don't have this issue of their wealth is changing and you still find here that your willingness to pay to reduce the risk of more severe illnesses is higher per unit of health than it is for less severe illnesses although the magnitude is smaller so it does matter Julia I wanted to know if you able to disentangle when you are talking about prevention between primary and secondary prevention and on the other hand if you measure risk preferences especially risk aversion and prudence because we know from a theory that there is a convoluted relationship between prevention and risk aversion and especially with prudence because when you want to avoid the worst scenario maybe the worst scenario is when the prevention is costly and it doesn't work so if you are very risk averse and prudent maybe you don't want to be in this scenario so I wanted to know if you have some measures of risk preferences yeah so we use sort of just what the standard measures are in the literature without getting too technical prudence does matter for these results if you use whatever the standard estimates are you find that essentially you get us a factor in order to increase following a health shock but if you assume an extreme enough value for this it can reverse it that's actually one of the new things we show in the study prior to this people thought it could only go up it turns out that actually more generally it can go up or down sort of depending on that parameter but it's a fairly technical condition no more questions ok thank you very much Carol and now the last words Jean yeah I discovered with a lot of emotion that I was supposed to conclude this session so I have to call it a day before you move on to state n plus 1 there's been a very intense day and afternoon I would like to thank the speakers very much I learned a lot and this was a great afternoon but I just I'm not going to embark into a Fidel Castro speech that will be very brief I would just like to to thank all those who made this happen this conference happened so on scientific side Pierre for organizing everything Pierre and the team I mean you have seen a number of people in our team both faculty and students I want to thank Eve and Cecil for the TSC site for organizing that very efficiently I'm very grateful as well to Lem to Eric I don't remember Juliet and I don't know who is still around but Lem was very instrumental in this whole venue and so on so we are very very grateful and I think all those deserve a big a big round of applause