Professor Andrea Cipriani Thank you.  Thank you very much for the invitation   and the kind introduction. Can you hear me well? Professor Bernadka Dubicka Yes. Professor Andrea Cipriani Yeah, okay. So, I  will present something that we have done over   the past few years and trying to contextualise  in the child and adolescent population, also,   the experience we had in the adult population,   and this is my main conflict of interest. I’m  an Adult Psychiatrist, so I’ll try to bring   as much as I can to the party today, but I’m  here more to learn than to actually present. This is my conflict of interest and  the acknowledgement. As you said,   I’m funded mainly by the  NIHR and the Wellcome Trust. So, I’ll crack on with one important aspect. I’m  a Clinician and I see patients every week in the   clinic, and it was very important for me to read,  about ten years ago now, this paper published in   Journal Internal Medicine. It’s a systematic  review and they collected all the information   about what are the most important question for  patients? It’s interesting to see that the top one   is, “What is the drug of choice for” that specific  “condition?” So, identifying the best intervention   for each patient is actually one of the most  important thing that patients ask us to do. The second point is, okay, in order to make  this decision, we need to be evidence-based,   and what is the support that evidence synthesis  can give us? This is one typical hierarchy   of evidence, the pyramid, these made by the  Centre for Evidence-Based Medicine in Oxford,   but there’s plenty of examples  and they’re mainly identical. Level one is, of course, a randomised controlled  trial, and I think the nice thing of this   hierarchy is that they – within each level,  they break down different types of evidence,   depending on the question. For what  we are interested in today, which is   identifying the best treatments, the first one  is therapy, is what we are going to focus on. So, level one is randomised controlled trial, and  they divide a and b, because 1a is a systematic   review of RCT, so a collection of trials, 1b  is an individual study. Level two is cohort   studies or observational studies, again divided  into a and b, depending on the – whether it’s just   one study or a systematic review of studies.  And this is what we teach medical students,   but actually, as a Clinician, I don’t entirely  agree with this, because the problem is, nowadays,   we have a huge amount of information  on the internet, in the literature,   so how can we make sense of all this large body  of evidence? We need something more innovative. And this is a real example from a systematic  review published, again, 15 years ago, but the   point here is this is a very good review published  in a good journal, by a very good team of   Researchers. The point is not the methodology here  in – per se. Is more how we can use this data, and   the example outside, if you want, mental health,  is on purpose, so it can apply – can be applied   to everything. So, “Safety and Effectiveness  of Different Methods of Earwax Removal.” And   if we go straight into the ‘Results’ section of  the abstract and we summarise the findings, this   is what we find. “On measures of wax clearance,  Cerumol, sodium bicarbonate, olive oil and water   are all more effective than no treatment.” Second  finding, “TP is better than olive oil.” Third,   “Wet irrigation is better than dry irrigation.”  Fourth, “Sodium bicarbonate drops followed by   irrigation by Nurse is more effective than sodium  bicarbonate drops followed by self-irrigation.” So, we have this long list of results, but they  don’t really help the clinical decision-making,   because what we need to know,  and patients want to know, is   how big – what is the ranking of the different  options and how big are these differences? So,   standard or pairwise meta-analysis don’t  really help, because this is an example   of one systematic review and meta-analysis.  On the left you have a pooled estimate of a   group of studies comparing Treatment A  versus placebo, for instance, and then,   you have a completely different set  of studies comparing B versus placebo,   another systematic review and meta-analysis. So,  what if my question is comparing A versus B? In   this case, there’s no evidence. I can’t  compare directly Treatment A versus B. So, a relatively new methodology called  network meta-analysis, or multiple treatments   meta-analysis, allows to compare interventions  that have not been compared directly. How? Well,   because group one of studies has placebo, which  is incumbent with the second group of studies. So,   if we use this common treatment, we can compare  indirectly A versus B via placebo. And, of course,   it’s crucial to understand this point – apology,  it’s a bit of methodological section at the   beginning, but I always do these talks, mainly  because we need to be very critical about the   evidence which is presented. So, the key point of  the placebo in Group 1 and placebo in Group 2 is   they have to be similar enough to be combined.  And this is a big issue in evidence synthesis,   because often, the Researchers combine things that  are – should – cannot be combined, and this is a   violation of an assumption called transitivity,  which has to hold in these kind of study designs. So, how can we use these technique network  meta-analysis? This is an example from the   adult literature, The Lancet paper we  published in 2009, where you can see   that we have this network of treatment. And  the size of each node is proportional to the   number of patients randomised to each drug,  and the lines represent the direct trial,   comparing two interventions, and the width of the  line is proportional to the number of trials. So,   many studies comparing sertraline with  fluoxetine and many patients randomised   to fluoxetine. However, there are no studies  comparing reboxetine with mirtazapine. So,   what we can do is to calculate the indirect  evidence reboxetine-mirtazapine via a common   comparator, like fluoxetine. So,  we fill the gaps in the network. The other important thing is when we  have an existing direct comparison,   like sertraline-fluoxetine, we can use the  network to compare this direct evidence with   the indirect evidence, sertraline-fluoxetine,  via a common comparator like venlafaxine. And   I can do this with all the loops in the network,  via citalopram, via fluvoxamine and so on. So,   it’s important because this is what we call  consistency, the direct and the indirect evidence.  Let me give you an example. If I  tell you that sertraline is better   than fluoxetine and fluoxetine is better than  paroxetine, what you expect is that sertraline   is also better than paroxetine. If this is the  case, the loop is consistent or coherent. Now,   sometimes in the evidence, we have the  direct evidence of sertraline-paroxetine   actually says that sertraline is  worse than paroxetine. In this case,   the loop is inconsistent or incoherent. And  that’s another key information to check in the   network meta-analysis to understand  whether the results are reliable. So, what – these are the advantages of network  meta-analysis. Comparing interventions which   haven’t been compared directly. Use  all available data. Improve precision,   because you increase the number of  patients in each comparison, but also,   we can rank treatment from the top to the  bottom. How do we do it? Using the results   of the direct and indirect comparison, we can  calculate, we can estimate for each treatment,   which is the probability of being the best?  In this fictional network, we have only four   treatments, for simplicity, and that probability  is for Treatment A – sorry, to be the best,   is for Treatment A 25%, 50% B, 25% C, zero  D. So, it’s easy to say that B is number one. And we do this for each position. So,  for position two, again, we have 25,   25, 50% for C and zero for D, so C is clearly  number two. Third position, 25, 25, 25, 25. So,   it’s impossible to say what is the best if we  look at the actual probability of being the third,   but if we look at the cumulative probabilities  to be among the best three, we have 75% for A,   100% for B, 100% for C, only 25 for D. For this  reason, A is the third and D is the fourth,   and this is confirmed by the 70% probability  for D to be the fourth treatment. So, we can rank treatment and we can do  nice things like this. This is, again,   the 2009 Lancet paper, and what we wanted to  do, we wanted to plot the – for each drug,   what was the probability of being – sorry,  we don’t have the numbers here, like, first,   second, third. We have 12 drugs, and this is  the probability. So, you can visually see for   efficacy the solid red line in acceptability,  dropout rate, discontinuation rate. The dotted   blue line, that some drugs, this is in adults, for  depression in adults, some drugs like paroxetine   are clearly not the top treatment, the best  treatment. Sertraline is much better. Oh,   citalopram is interesting, because it’s well  tolerated but not very efficacious, the opposite   as opposed to venlafaxine. So, you can really get  more information if you use network meta-analysis. Okay, so, that’s the end of the methodological  section. I promised, in my abstract, to look   at the most recent network meta-analysis  in the field comparing antidepressants for   depression in kids – in children/adolescent.  And this is the latest network meta-analysis,   published a couple of years ago. It’s in Chinese.  The reason why I will not cover this is because,   actually, the paper we did in 2016,  published in The Lancet, is still, I think,   the reference. There’s another Cochrane Review  that I’ll mention at the end, but I want to start   the presentation starting from the 2016 Lancet  paper, which is this one that we did comparing   all antidepressant in a network meta-analysis  that were trialled in children and adolescents. This is the search – the PRISMA flow chart,  to summarise the search strategy and what we,   in the end, found. We started with about  6,000 records and we also checked about 2,500   of unpublished data from drug registries.  And at the end of the process, we managed   to include 31 publication, corresponding to  34 randomised controlled trial and 40 pairs   of investigations. And what is important  here, as a first take home message for you,   is that we have very few trials for the  great, great majority of comparisons. So,   we only have ten studies for fluoxetine, but  for the others, except paroxetine with five,   it ranges between one, two and three.  So, it’s a very small amount of evidence. The second point is the comparisons are mainly  with placebo. So, the majority of the drugs in the   network do not have head-to-head comparisons.  So, I expect the shape of the network,   that I’ll show you in a second, to be less  connected. Is more like a star-shaped network,   where we have the placebo in the middle, with  all the drugs connected directly with trials,   and less, or very few, head-to-head  comparisons. This is the list   of the studies that we identified, just to  show that these are the doses and that also,   the sample size of the in vivo studies  not great in terms of large sample. And this is the network. So, we have placebo, very  big, because it’s the most frequent comparator,   and all the drugs are compared with placebo, but  only a few are compared head-to-head. Duloxetine   is compared with fluoxetine. That’s  one of the few head-to-head comparison.   Fluoxetine is compared with nortriptyline and  venlafaxine. Sorry, no, we have Clomipramine,   which is the drug without a placebo controlled  trial. So, there’s no line between Clomipramine.   It’s connected directly only with paroxetine. So,  all this to say that the results of this network   meta-analysis should be taken with great caution,  because the lack of a connected network means that   we are not able to check the consistency  of the network and the findings. Hence,   we cannot check direct versus indirect evidence  and the results are less robust, if you want. This is a bit complicated, a busy slide,  but it’s an important piece of analysis   standard in any meta-analysis, because this  shows all the comparisons for, in this case,   two outcomes in one go. On the bottom left  triangle, in blue, we have efficacy measure   as change in symptoms, using a standardised  mean difference, so combining different scales   into an effect size. While for that right, top  right triangle, we have discontinuation rate,   what we called tolerability, because it’s  discontinuation due to adverse events. In the middle, we have all the drugs and  placebo in order, like a ranked base,   on – from the best to the worst, which means that  in this case, at the top left, we have fluoxetine,   it is because it’s the drug with the best  ranking, as opposed to all the others,   because we are comparing everything  with everything. How to read it is   column versus row. So, if we take fluoxetine and  we go down three cells from the bottom, we have   -0.51. If you go to the right, that’s the estimate  fluoxetine versus placebo. And the estimate is   highlighted in bold because it’s statistically  significant. It ranges between -99 to -03. Two important consideration. If you take  this line, this, sorry, this row, the only   highlighted and bold result is for fluoxetine.  So, based on the 2016 network meta-analysis,   only fluoxetine was better than placebo.  And the other examples are here,   if you look in terms of tolerability, there are  a few differences, but there are not that many   differences between antidepressant in terms  of efficacy and tolerability, and fluoxetine   is clearly the best antidepressant in this case.  And nortriptyline is worse than everything else,   it’s at the bottom. We don’t have data  about tolerability for nortriptyline. The other important consideration to contextualise  the result is the confidence interval. So,   it’s true that fluoxetine is statistically  significantly better than placebo, but the   true value is between 0.03 to 0.99. So, it’s  a large uncertainty around the estimate. The   best bet is .51, which is a relatively  moderate effect, what we say. However,   the true value ranges between, basically,  no effect to a very substantial effect.   And this is important to understand in  terms of interpretation because this may   be related to the heterogeneity of the  studies, but may be also linked to the   heterogeneity of the individual patient who  may respond or not to that specific drug. And this is also raising an issue  that we should always remember,   which is that it makes sense to combine many  treatments in the same network meta-analysis,   especially are we combining pharmacological  with non-pharmacological treatment together?   Because the study design is different,  intervention is different, and the population   may be different. So, that’s a big question mark  that we need to assess every time we start in a   project like this. So, in this case, we  only included pharmacological treatment. This is for another outcome, which is suicidality,  and we only have one outcome. And you can see that   venlafaxine is clearly the worst in terms of  the effect. In this case, it is odds ratio,   so it’s dichotomous outcome, having  suicidal behaviour or ideation. So,   it’s not committed, completed suicide,  but again, the data are quite striking   supporting that venlafaxine is clearly  worse, as opposed to all other drugs. And then, I wanted, also, to show  you the absent number, because here,   we report the relative effects, but if  we look at the absolute numbers, we have   very few trial, reporting very few events. So, in  terms of how frequent is this event clinically,   of course, we are talking about  a rare event, but if we look at   the comparative data, it looks like there  are differences between antidepressants. Okay, I know we have only 30 minutes, so I’ll  move to the next publication, about the same   topic, which is this Cochrane Review done by  Sarah Hetrick and the team, the Cochrane team,   across Australia and the UK, and New Zealand,  of course. And they did something similar, but   not identical, and my take home message from this  review, oh, they didn’t include all the drugs that   we included in our paper, and they also looked at  the change in symptoms on the CDRS scale. So, they   didn’t look at standardised mean differences. They  really wanted to look at the mean difference, the   change in that – on that specific rating scale,  which means they use fewer data and fewer studies. This is the network for the first – for  the primary outcome, for the CDRS scale.   You see it’s very similar to what we had  for our 2016 network meta-analysis. There   are new drugs like desvenlafaxine, which was  not included ten years ago in The Lancet,   because we didn’t have data about this drug.  But all the other drugs are, basically,   the same, normal study as opposed to our  review. And even if we have fewer studies   and if we have a different network, in  the end, the results are very similar,   where fluoxetine is still the best antidepressant.  They also measure response rate and remission,   but you see the data are so sparse  that we don’t have anything relevant. Sorry, am I finished in terms of time  [pause]? Oh, good, because I wanted to   briefly mention the meta-review that we did more  recently, even though it is now five years ago,   where we wanted to look at the effect of  antidepressants across different population   of patients with child and – children and  adolescent with mental health disorders.   And we looked specifically at the effects of  the different antidepressants across different   disorders. This is the flow chart, because we  – rather than looking at individual studies,   we look at systematic reviews. The quality  of the individual studies is not great,   as we said at the beginning. The quality  of the systematic review is very similar. So, we included 11 systematic review and  meta-analy – and network meta-analysis,   of course, but the quality varies from  low or critically low, to moderate. So,   again, all this has to be interpreted  with a pinch of salt. And in the end,   even if the amount of evidence is very  low, we found something interesting.   We still found that fluoxetine is – was the best  treatment for, on average, a patient with major   depression. While we found some promising data for  anxiety about fluvoxamine and paroxetine, and in   terms of OCD, fluoxetine and sertraline  perform better. Antidepressants, basically,   didn’t really materially change symptoms in ADHD.  While we find – found something about some drugs,   like Clomipramine, in people with autism,  and nothing really for PTSD and anuresis. So,   inter-drug tolerability is, basically,  what we just said was about suicidal. So, in terms of this amount of data, I  think there are still reasons to think   that antidepressant can play a role, but the  effect of this medication in this population,   even if present, is limited. But why is this?  Is it because they really don’t work, or is   it because we are missing a trick? And I want to  tell you what we are trying to do in the adults,   and I hope that this will possibly open a line  of collaboration, because the big problem of   all this network meta-analysis and evidence  synthesis is we use averages. We use a, sort of,   median effect that is not really what we see in  patients. So, we have developed this project in   adult depression, but I hope to replicate  in other fields, and I know that Professor   Cortese is doing this for ADHD, so the field  is expanding, also, to adolescent psychiatry. PETRUSHKA, in his personalised antidepressant  treatment for unipolar depression combining   individual choices, risks and big data in  adults. And what we did was to use data from   the randomised trial at the individual patient  level. So, we identify the characteristics,   clinical and demographic characteristics.  We combine them with real-world data. So,   we moved from relative effects to absolute  prediction, probabilities of responding or   having a specific adverse event, and then, we  also, incorporate patient preferences, which   is the third big pillar in order to personalise  treatment. We develop with the web-based algorithm   and now we are testing the algorithm in a trial,  which is in the UK, in Canada and also, in Brazil. This is the algorithm. It’s password protected  and patients and Clinicians go through this   list of adverse events and they choose the  best antidepressant for them, and thus, the   visualisation that we developed with patients at  the end, and for the trial, we have three options,   with a Tripadvisor like score to show how strong  we recommend this antidepressant. But then,   below, we have a breakdown of the specific  adverse events that matter the most to the   patient and we can tell then, predict which is the  absolute probability of having this side effect. The study won the award last year from  the NIHR and we are very proud of this,   and really hope that this can be  the direction of travel, also,   for the field in child and adolescent  psychiatry. Thank you very much. Professor Bernadka Dubicka Now I don’t know  who came up with that, but it’s certainly very   ingenious, ‘cause that can be a huge challenge.  So, lots of interesting things to pick up there.   We’ve got time for questions. There’s one  already here for you. Please – other people,   do please put questions in for us. So, the  first question is from Shermin Imran. Her   question is around “variability and response  to antidepressants and whether this could be   related to pharmacogenetic factors?”  Is that something you can comment on? Professor Andrea Cipriani Absolutely.  Very good question. Yes, in fact,   we just submitted a proposal with a funder  to do PETRUSHKA 2, incorporating exactly the   pharmacogenetic information. Yes, so we have data  – I mean, I’m talking about the adult literature,   but I think this applies, of course, to young  people, to some extent. So, we know that the   pharmacogenetic is really important, not only in  terms of the viability, for instance, metabolic,   fast metabolisers, low metaboliser, because  we know that we have very few people who are   fast metaboliser, very few people who are strong  metaboliser, and the majority find them in the   middle. But also, because we now have data about  the side effect profile of some medication. We   have data now, also, about the microbiome and  how this interact with the pharmacogenetics. So, the field is evolving and very recently,   a few months ago, there are new guidelines  from the Pharmacogenetic Assoc – Worldwide   Association which are very useful to build  this, sort of, matrix about the drugs and the   potential interaction. We also may find individual  deletion or genetic alterations that can be very,   very informative how not to prescribe this drug,  or the combination of different medications. But the other important thing I would like to  develop in the genetic field, and I’m work – I’m   not an expert, so I’m working with people across  the globe, is not only the pharmacogenetic   variations, but also, I’d like to use polygenic  risk scores in clinical practice. That’s my dream,   because when – especially when we see the patients  with depression, we don’t know whether they will   become bipolar, schizoaffective, so having this  tool will help the treatment plan, but also,   it’s a fantastic way to empower patients in the  shared decision-making process. Because when   we have the polygenic score, it is a score,  so it’s nothing causal. We are not certain,   but this forces the Clinicians to discuss  with the patient and involve them in the   shared decision-making process, which is the  real reason why we did PETRUSHKA initially. I know for kids, it’s more complicated because  you have the parents and everything else,   but I think this is in the direction of travel,   having evidence-based scores that can  inform the shared decision-making process. Professor Bernadka Dubicka Good, it’s good  to have a dream, Andrea. I hope we manage to   get towards your dream in the not too  distant future. The questions are all   coming in. I’m going to pick one here that I  was actually thinking of asking you, as well,   and it’s something that’s not talked about  enough in older adolescents and in particular,   in young people, but it’s something  that is talked about in the media and   amongst young people on social media, and that’s  regarding ‘sexual dysfunction’. And obviously,   that’s something that’s very important to young  people, but it isn’t discussed sufficiently,   and so, I’d just like any comments that you might  have on that. And I don’t think it’s brought   up as a adverse effect when you look at these  trials, either. Obviously, for younger children,   it’s not appropriate, necessarily, to  ask those questions, but it’s a really   important fact – you know, really important  aspect of young people’s lives, isn’t it? Professor Andrea Cipriani Absolutely, and yes,  again, I don’t have experience in treating kids,   but I – in my clinic, I see people from 14/15/16,  depending, because we are trying to combine not   having artificial barriers in terms of age. So,  yes, the sexual dysfunction is very important,   as you say, not just during the treatment,  but also, maybe post. It’s very difficult to   disentangle the mood aspects from the biological  aspects, but it’s always a discussion that we do   in the clinic. It’s not just about suicide, which  was initially a taboo. Now discussing sexual   dysfunction and – is especially important,  and not just for men, but also for women. In terms of sharing my clinical experience, I  raise this all the time, but I realise that there   are very important cultural issues around this,  because now that the – we – especially because we   are doing the trial now in Nigeria, in Pakistan,  so I was exposed to this cultural adaptation,   is one, how do you define it, so the words you  use to explain the side effect? And also, how   you discuss it, because maybe it’s not just about  religious or cultural failings, but for instance,   in Pakistan, all the shared decision-making is  done with – or in India, involving the family.   And, of course, it’s important to have this  discussion in an informative way, but also not   to put the person in a difficult position. So,  it’s a very complex thing, has to be mentioned,   also in terms of risk, but I would also be  careful in saying we don’t have enough data,   mainly because it’s a neglected area. So,  we really need better quality information. Professor Bernadka Dubicka No, and we really  do and thanks for, you know, sort of, starting   that discussion. So, I really urge Clinicians to  think about it in their consultations, as well,   and prescribing. I’m just going to just  expand that question a little bit further,   because one of my concerns is that, you know, we  see children and young people relatively early on,   compared to, you know, sort of, their adult  trajectory with depression and antidepressants,   and often, you know, we see young people  who maybe have had chronic anxiety for a   long time, particularly if they’re  autistic. You might start them on   an antidepressant/anti-anxiolytic, but then,  the issue of dependency concerns me greatly,   particularly when you’re starting these  antidepressants quite early on. And often,   there never seems to be a good time to stop  because they’ve got one milestone after another,   exams or transitioning to uni. And again,  I think, we don’t have much data, do we, in   terms of long-term impacts on young people, and  particularly around the issue of dependency,   which has been talked about a lot more in adults?  I don’t know if you wanted to comment on that? Professor Andrea Cipriani Yes, absolutely. As  you say, the big problem is the scarcity of data,   and this is why, again, what we try to do  with adults is – but it’s the same for kids,   is trying to use other sources of information,  like the observational data. Because RCT,   randomised data, are very good for  comparative – com – for comparisons,   but if you want to observe what happens over  time, now that we have a lot of registries,   a lot of population databases, population level  databases, that’s very, very important to use. In terms of the dependency and things, of course,  it’s a mixture of pharmacological effect and of   psychological. As you say, there’s – maybe  that some person is never the right moment   to try and stop. I tend to suggest, or to say at  the beginning, that “It’s likely – the benefit   is a – for a long-term treatment, but it  doesn’t necessarily mean for life.” And,   of course, the other important thing is “not  to take the decision based on how you feel now,   but also decide if things go well, over the next  few months, I will do this or that,” of course. The other important thing that now has  been emphasised by many, many people   is the tapering off of the medication. Of  course, it is a very personal experience,   but in my clinic, the longer people  have been taking the medication,   the longer is the tapering off. So – and it  can take months, because then, the person gets   familiar with the situation and they – it’s  not the main event, in one – in two weeks,   I taper off the medication. So, I think these,  kind of, practical things, but as you said,   it’s crucial to have more information, and also  of the risks of this medication, they’re known. Professor Bernadka Dubicka Yes. Professor Andrea Cipriani Yeah. Professor Bernadka Dubicka And we’ve got another  question, which also relates to something I was   thinking about, and that is how you make clinical  decisions when there’s comorbidity. And that   relates to a question I was thinking about, you  – I mean, you’re very well aware that the trials   are very limited and one of my concerns has been  about you talked about the difficulty of combining   some of these findings, but one of the issues  is combining the populations that were studied.   As you’re aware, a lot of them were done in the  States, they’re not necessarily applicable to NHS   clinical population. The lev – and in particular,  around a level of severity and the sorts of young   people we might decide to prescribe here in the UK  and maybe other countries, other than the States.   So, just any comments you have in terms of how do  Clinicians make decisions around who to prescribe   to in terms of complexity and comorbidity,  based on the limited data that we have? Professor Andrea Cipriani Yes, yeah,  exact – yes, that’s a very good point,   also, the what we call the external  validity and how generalisable are the   results to the population of people we see in  the clinic? I have to say that that’s true and   unfortunately, I think the response that  we, as Clinicians, tend to have is based   on our experience and we try to be on  the safe side. The example is, again,   it’s taken from the adult literature, but 85%  of the prescription of antidepressants are   about three drugs. So, if we could involve the  patients in the shared decision-making process,   automatically we move out of our comfort zone,  but at the same time, we share the responsibility. And we need also – well, this is what  I do in the clinic, I share with the   patients the uncertainty, because I like  to present numbers as much as possible,   but with also the uncertainty around the estimate.  So, you have a 30% probability of responding,   but the true can be ten to 50%, for instance.  This is why we need to move from these averages   to the individual patient, and we have  great examples outside mental health. Nowadays, all GPs use QRISK for the cardiac  risk, in order to decide whether to prescribe   or not a statin. So, it’s very often used outside  mental health. So, we need to bridge this gap. Professor Bernadka Dubicka Thank you,   and raising a really important point and making  – the joint decision-making and fully informing   young people in those decisions  based on the evidence available. Right, so we have a couple of – literally  two minutes left and for me to wrap up,   as well. So, two questions and very quick  bullet point answers. First question is,   “NICE focuses on three SSRIs for young  people. If two SSRIs have failed,   should we consider duloxetine  as the third next, yes or no?” Professor Andrea Cipriani I would, yes,  yes, definitely. I mean, the guidelines are   a tool to help the clinical decision, but  I mean, you can feel seeing the patients,   whether there’s any point insisting with an SSRI  or change it. So, I strongly defend the freedom   of the clinical judgment, as long as you share  the decision and you explain to patients, because   they may go and look at the NICE guidelines and  challenge your decision. But absolutely, yes. Professor Bernadka Dubicka And I don’t think  that’s something that we probably use very   often in practice, but something to consider.  And the very la – we’ve got one minute to go,   so a very last question. I  don’t quite understand it,   but it’s the issues around “teenagers with poor  diets and microbiomes,” and I don’t know how   that relates to antidepressants. I don’t know  if that’s something you can comment on or not. Professor Andrea Cipriani Yes, yes.  So, of course, it is related to diet.   It’s more related to drugs. So, for  instance, if you take an antibiotic,   this changes your microbiome. The good thing  is that we are now developing treatments,   also like to reset the microbiome and also, you  have a tablet with a healthy microbiome. So,   it is more complex than the diet, but  definitely the diet can have an influence. So,   at the moment, it’s still early days, because  at the moment, the microbiome can be a big hype   or can be something real. I’m more keen to think  the latter, so I still think there’s a lot that we   need to discover about the bacteria in our gut,  but it’s too early days to tell how to do it. Professor Bernadka Dubicka Thank you, and I think  on that note, we’re coming to an end, but I think   that would be an excellent topic for a future  conference around nutrition and mental health   in children and young people. So, thank you for  raising that question and all the other questions   that people have brought to us. And thank you,  again, Andrea, for a really interesting talk.