Foreign um okay family foreign fair and what scares me. Is that making those decisions and making those wrong decisions and the realization that that decision of that decision in three four weeks. Time where you end up having a patient who in effect can be crippled by life-sustaining therapy um which may not have been in their best interests or their family’s best interests and that still scares me and i think it scares a lot of colleagues but um intense care is much more multidisciplinary now and i think that has helped our decision making and has reduced that fear factor honesty and humility. Go a long way. And most patients and relatives understand that we’re also human and mistakes can happen and should be openly discussed together with my colleagues in particular the nursing colleagues. We arranged a wedding for a patient who was at the end of his life and had decided to marry his long-term partner before he later died and it was very moving to be part of this very special event in the intensive care unit. It’s a real privilege to be able to support a patient and their family through that. Um what i hope is the worst thing that ever happens to them in their lives to to be able to support them through that hopefully to recovery or to support them through managing a dignified death foreign intensive care medicine allows me to learn something new every day so every day i need colleagues and i hear of their new technologies new advances and progress in their particular specialty but most importantly every day is a different day with new challenges and i enjoy the interaction with my colleagues and most importantly with patients and their relatives so it is a provides a great variety. It’s different every day and it’s a specialty where progress is made on a on a regular basis and water guys. We will repeat the marathon at the end of may again a marathon of eight hours with experts from different parts of the world to share what we’ve learned from covet. How we can repeat the future and how we can come out of this situation together.
I’m an anesthesiologist and i’m proud to be an intensivist. We intensivists are working together to fight kobe 19. Together we are intensive care medicine. Hi i’m proud of being an intensivist. We are fighting kobe 19. Together we are intensive care medicine. I am an anesthetist and i’m proud to be an intensivist. We are working together to fight covet 19. Together we are intensive care medicine. I am an intensivist and proud of it. Intensivists with various specialty backgrounds work together as a multi-disciplinary team to prevent and treat temporary risk of death due to covet 19. My name is john de valle. And i’m proud to be a surgery trained intensivist and work in intensive care. We work together to beat covet 19. Together we are intensive care medicine. Hello everybody and welcome to this webinar on how to conduct research how to do research, good research is my name is john laffey. I’m an intensivist and anesthesiologist from national university of ireland. Galway and galway university hospitals as chair of the translational biology section. I’m delighted to welcome you to this webinar. On how to conduct research how to do research, good research is our first speaker is dr mariangela pellegrini and she will speak on the challenges for an early career investigator in setting up a laboratory research how to do research, good research is program. Our second speaker in the webinar is dr harmian degrooth and he will speak on the pitfalls and opportunities for icu related observational research how to do research, good research is. Both speakers will speak for 20 minutes. And if you have any questions and we strongly encourage you to ask them we will take them and ask them at the end of the two talks so we will collate them and ask them at that point so now i am delighted to uh welcome our first speaker dr marie angela pellegrini marie. Angela is a young critical care specialist at the academic university hospital of uppsala in sweden. She’s a postdoctoral research how to do research, good research iser at the academic university hospital and works at the hedensterna laboratory in uppsala.
She’s a member of the next committee. And of the esicm research how to do research, good research is committee dr pellegrini’s research how to do research, good research is is focused on respiratory mechanics and her main research how to do research, good research is. Focus is on diaphragmatic activity during spontaneous breathing and mechanical ventilation in patients with aords. Dr pellegrini uses advanced signal analysis and image processing to perform real-time synchronized analyses of lung imaging and respiratory mechanics. Today marie angela will speak about the available tools for planning and conducting good experimental research how to do research, good research is and she will focus on the perspective of a young early career investigator. Dr pellegrini thank you very much. I’m really looking forward to your talk. Thank you john hi. I’m very glad to be here to be virtually connected with you with you this afternoon. In this short talk i’ve been invited to describe the challenges. An early career investigator has to face through his or her path to become an independent research how to do research, good research iser. You would expect this kind of talk by an experienced and successful research how to do research, good research iser. Well i’m not but i’m actually leaving the early career phase of my life and i’m deeply in all those challenges that i’m going to speak about. We can easily describe concepts by the help of metaphors. And if we want to describe the life of an early career investigator someone one will describe it as in this figure few doors few alternatives a door opening with new perspectives. This is a quite simple image heterostatic metaphor. I would rather represent our life through this second image a life full of challenges new obstacles to that have to be overcome all the time. Several hymns has to be accomplished at the same time. During some particular circumstances a early career investigator can feel life as this image during our paths. Through towards our scientific independency we must cope with several aspects of our life. First of all we need to be resilient and motivated this is the most important aspect then. We need to adapt to changes to mobility be always available grasp opportunities at the right time and in the right place confront ourselves with other people learn and be challenged by our mentors and be able to build our network with new collaborators all around the world and then we will have need to have time.
Yes because we are never going to say no to anything and then we have a privatized life much much more the first and most important aspect to conduct a good quality. Research how to do research, good research is is your motivation. We need to be inspired and you get inspired by clinical experience by confronting yourself with colleagues by attending conferences and by reading reading a lot a lot of science especially in your own field of research how to do research, good research is you will need to build the basis of your scientific knowledge. Even if motivation is an essential start. This is not enough to grow as research how to do research, good research iser. We need a stimulating environment around us that allow us to grow if we uh if we are lucky enough well the the research how to do research, good research is we are interesting in are already performed in our institution otherwise we basically can migrate everywhere and for what concern. Basic research how to do research, good research is is sometimes tricky to know where the laboratories are. It’s not only simple to find information about the different centers that conduct that kind of research how to do research, good research is that we are interested in one advice is to look at the affiliation for in those people that are closer to that kind of research how to do research, good research is that you would like to conduct in my case. My scientific interests brought me from south of italy bari. A very warm and sunny place to the north of europe to uppsala in sweden uppsala is a beautiful nordic city. But well this is one of the first pictures in uppsala 2014. At that time i was challenged by low temperature and small in my case the laboratory from where everything started was and still is the head heinen laboratory. The eden hernan laboratory is among the most prejudiced prestigious research how to do research, good research is platform in europe for experimental intensive care several groups of research how to do research, good research isers from uppsala but also from all around the world working in experimental research how to do research, good research is at intensive care come to uppsala to perform the most complicated and demanding experiments the header.
Hana laboratory is made of a comp of a competent team of people that help you and let you always feel at home as a whole big research how to do research, good research is laboratory all around the world it offers you a very open environment in which you can always confront yourself and your ideas with other research how to do research, good research isers at our laboratory. An annual international symposium is organized and is an amazing experience of brainstorming and confrontation. Then you will need at least one mentor. There are several mentors in our life and also in research how to do research, good research is. You need someone that is more experienced than you. That can motivate you that can show you the direction and coaching you and challenging you. Sometimes in my case i have been extremely lucky having two supervisors during my during my phd studies to high ranked scientists with focus on mechanical ventilation professor heather. Hernan and professor larson. I have learned a lot from them. Concerning mobility and mentorship. It’s important to cite next. The young part of acsm next constantly. Try to help you young intensivists and investigators through the early phase of their professional career. So one important advice is always follow next activity by regularly. Visiting our webpage back to the lab. A research how to do research, good research is laboratory is not only the people the lab is also the structures and the equipment that it offer you to perform your research how to do research, good research is. This is important to perform high quality research how to do research, good research is in my case. The headache laboratory includes several theaters equipped with all is needed for performing experimental intensive care. The lab has the same instruments as in a normal ico as well as a direct access to ctm and pet. Then you have always to consider that a part of your equipment of your material materials you will need to buy by uh your research how to do research, good research is money and for what concerned the money.
This is always a challenge and for all kind of research how to do research, good research isers experienced or not scientific ideas need money to become real in my in my case. I’ve started quite early training looking for money. Yes because to apply for research how to do research, good research is. Funding is something that you learn as you learn how to perform an experiment. Often you will get a negative answer and your research how to do research, good research is will not be founded but at the same time you become better for each new grant you will apply for. There are grants that will provide you. Part of your salary holder grants that will provide you money for equipment. I would like to take this opportunity to remember that. The acscm is always promoting research how to do research, good research is all around europe offering an annual base grants to which you can apply. Several of them are for young investigators. I have applied for the acsm awards. Some years ago have never been successful after some rejections at the same time you will need to train to become a good research how to do research, good research iser. There are a lot of important courses. You will try to attend to be better. In my case. During the phd course in uppsala i have attended several of those courses. But he’s never enough. Moreover if you are aiming to an academic career you will need to know how your university and our your institution work. One advice is in this case is to always ask uh never be shy. Always ask and try to get all the answers you need one more challenge. Come from the fact that we are early career investigator but at the same time we are also young intensivists. In my case i became an intensive care doctor. At the same time i became a phd student and these give you even more challenges because at the same time you have to try to become a good research how to do research, good research iser and a good intensivist you have to train. You have to acquire clinical skills and experience and you have to become a fundamental part of your icu team. Then what is important is that the clinical sp experience will be important for your research how to do research, good research is because only in this way you will find scientific hypotheses and exactly on the opposite uh you will need to test your experimental findings in your clinical practice to feel that they are being meaningful.
Then you need also to certify your clinic and knowledge and your competencies in critical care taking the headache exam and also this is a big challenge. Well now we have discussed a lot of challenges how to find a valid research how to do research, good research is hypothesis. A lab the equipment the money how to try to be a better intensivist and investigator at the same time. But then when you have all these as bases well you know uh you need to know the method uh how to plan your study and then you have several tools that we are going to go through. The first thing is that you will need to write your ethical application and get it approved legislation. Rules can vary from a country to another country. In sweden is europe groups work at the authority that regulates the evaluation and the approval of experimental research how to do research, good research is. You will need to have a training in laboratory animal science and in europe the standard is set by philasa fellas also organize all over europe. Lots of courses about this then you will need to know the concept of the three errors. This is a very important concept that you will need to keep in mind while you are performing your experimental research how to do research, good research is. The three errors are replacement reduction and refinements where replacement is try to avoid or replace animal use in your studies for example using cells or humans. Reduction is when the animal is necessary. So you will keep the number at the minimum a refinement is to be sure to minimize pain and suffering and improve welfare. Then there are some guidelines that can really uh help you through this preparation and planning your study. One of those guidelines is the prepare guideline where prepare is an acronym for planning research how to do research, good research is and the experimental procedure on animal. This guideline is actually a checklist for the preparation of study and is composed by three different parts.
One is the formulation of the study. The second is the the diagolog between the scientists. You and the animal facility the laboratory and the third one is the control of the quality of your study number is the mobile guideline. It’s the arrival guideline way. Also in this case is an arrival is an acronym that is animal research how to do research, good research is reporting of in vivo experiments and the arrived guidelines are developed to helping you reporting your studies also in this case. You have a checklist and uh this is very helpful to do helping you to not miss important. Uh parts uh in your study not only by in writing but also planning your study and conducted them it will be also useful for reviewer by looking and evaluating experimental studies. Then there is a recent tool that has been um published by the nc3. Uh hair is a free in its online tool that is called experimental design assistant and that will help you give you advice in how to plan your study. How to perform statistical analysis power calculation and so on at the end you get a flow chart of your protocol in all different space and you will analyze all the points that are important and well that where you can miss something. Our duty as young research how to do research, good research iser is to take advantages of all those tools and be sure that we perform high quality science for the future. One more step vermont. Challenge how to get your pa. Your work published. These can be felt extremely exciting but at the same time can be also frustrating. Because often it happened that you get your work rejected or some reviewers ask you to completely transform your study in this location. It’s very easy to get upset and give up. Never do that. Everyone ever even god as you can read on the right side of the slide can be challenged by a peer reviewer. So put aside the hunger and work on yourself and on your study if you have been lucky enough to get a good peer review you will. This will help you improving your scientific skills one more aspect to which you will be exposed will be the need of promoting your research how to do research, good research is and these sometimes can be felt as a challenge produce and promote science are two parts of the same medal and if the first come from the work of a single person or a team the second needs socialization science imply the need implies the need of sharing your findings and get them challenged by other research how to do research, good research isers in a confrontation of ideas.
Which is the basis of a scientific process and promote your science at conferences can bring several challenges. You have to know how to present scientific data you have to fight your shyness and overcome your fears for example speaking. What is not your native language. Normally what works most is to think that no one knows your science better than you. The major expression of your scientific presentation skills is going to be tested during the defense of your phd thesis in that occasion at least in sweden with some differences among countries. The phd student will literally defend his or her cases against all the open criticism and comments of an international opponent and of a committee that at the end are going to judge you in front of all people. You care the most your colleagues your boss your family your friends and it’s a lot of pressure but at the end no one knows your science better than you and at the end you will enjoy a wonderful dinner with hundreds of guests and everyone will celebrate your scientific efforts. Well we come now to the end of this. Uh short talk through this. I hope i succeed in sharing with you. Some experiences and good advices directly from the battlefield. This last slide is to say that everything i discussed. Today it’s actually only the beginning of a long path towards a high quality research how to do research, good research is. So thank you and good luck. Thank you very much dr pellegrini. That was a very interesting talk. We’ve already received some questions. Uh through the sicm tv and so i strongly encourage others who are thinking of questions to submit them and we will put them to marie angela at the end of our next talk.
I’m now delighted to welcome our second speaker. Dr harmian degroot a doctoral group will speak on pitfalls and opportunities for icu related observational research how to do research, good research is. Dr degrooth is a clinical fellow and a research how to do research, good research iser in the intensive care department at amsterdam umc in the netherlands. He has a background in statistics and a phd in research how to do research, good research is methods. His work is focused on the intersection. Between clinical and statistical validity dr gruth will discuss what sets clinical research how to do research, good research is apart in critical care from other clinical fields and how we can adapt our research how to do research, good research is methods to the core challenges of icu research how to do research, good research is. I’m really looking forward to this talk. I’m jan and i invite you to take the floor. Thank you very much john. It’s great great to join you and also a thanks to the translational biology section for inviting me to contribute something here. My name is harmian. Hodd i am a physician scientist from amsterdam umc in the view university in amsterdam and for my research how to do research, good research is work i specialize in methodological and statistical problems of clinical research how to do research, good research is. That are more or less specific to the icu. We’ll see some examples in the next few minutes. I have no conflicts of interest to declare now the essence of translational. Biology is of course a bench to bedside approach to research how to do research, good research is. Let’s say as a translational research how to do research, good research iser you have identified an interesting biological mechanism in a preclinical model at some point. You’re going to want to establish whether the hypothesized mechanism is also present in actual patients and a logical first step would not be to immediately start an intervention study an rct but rather to use observational data to find the hypothesized association. You’re looking for using either. An existing clinical database or even by designing your own prospective observational. Study but it can also be the other way around perhaps you come across an observational study that reports a strong independent association between some phenomenon and patient outcomes you.
You wonder whether this is something that you ought to investigate. In a preclinical model. In either case as a translational research how to do research, good research iser you’ll be interested in some of the important pitfalls of observational studies. Either to design a study yourself or to critically evaluate the validity of an existing study. Now i’m not going to tell you that. Correlation and causation are often confused in observational studies. We all know that yet. This question of causality deserves some more attention as a translational research how to do research, good research iser you are specifically interested in causal questions. So let’s take a look at the prohibition of cause on causal language in the reports of observational studies. We can take the following statement. Low plasma ascorbic acid was independently associated with increased mortality risk after adjusting for severity of illness. As you can imagine this statement could be from the conclusion of a real observational study. But what is the point here. Are we interested in the prognostic significance of ascorbic acid. Or do we want to use it. To calculate the patient’s risk of death why not state the causal goal explicitly because the argument goes an observational. Study can never prove causation. This is true so given that we can never prove a causal link the standard advice from most research how to do research, good research isers editors and reviewers is this don’t ever mention causality in an observational study. But i disagree and so do many statisticians cloaking the causal goal of an observational study leads to inappropriate statistical methods and misleading conclusions. So what value should we attach to statements like this. Let’s start at the beginning. A scenario we often encounter in critical care. Research how to do research, good research is is this. We want to estimate the effect of a treatment on mortality or any other outcome. The problem is that the initiation of the treatment is associated with a patient’s baseline risk of illness or a patient’s baseline severity of illness.
I’m sorry the second patient is the more likely he or she is to receive the treatment. We are investigating because severity of illness is also also influences death. It is a confounder by randomizing treatment. We make sure that the exposure is influenced by nothing but chance we can. Now arrive at an unbiased estimation of the effect of exposure on outcome but randomization is often not possible so we have to use statistical methods with for example propensity matching or propensity scoring investigators. Try to sever the relationship between the confounder and the treatment to arrive at a pseudo population that includes patients as if they were randomized with logistic regression. Investigators try to adjust the effect of the confounder on death and both method methods can lead in principle to an unbiased estimation of the effect of treatment on death. But there’s one obvious problem. We cannot observe severity of illness directly. We can only estimate it using a scoring system such as the apache score or a sofa score or simplified acute physiology score etc now these severity of illness scoring systems are not perfect and obviously a poor scoring system cannot make a good risk adjuster so we must ask how well does the severity of illness scoring system need to perform to lead to a truly unbiased estimation of the effect of treatment on outcome michael schuding and colleagues at the university of michigan. Try to find an answer to precisely this question when do con founding my indication and inadequate risk adjustment bias. Critical care studies. It’s a severity of illness. Scoring system. Should be able to discriminate between patients. Who are going to survive. And those who will not and the quality of this. Discriminating ability is expressed as the area under the roc curve now using simulation methods michael shouting shows that if we aim to adjust the relation of interest by a severity scoring system such a system must have an auroc of 0.
74 or higher. If we use the score with less discriminating ability this may lead to false negative results or even to results that are opposite to the true and they conclude that in studies with 10 000 patients even with low confounding by indication and the severity score with an aoroc of let’s say 0.68 to 0.72. A beneficial treatment will appear harmful in a majority of studies so we must ask do most severity of illness scoring systems. Meet this mark. What is the area under the roc curve for the commonly used severity scores in the original publications these systems score very very well with aucs in excess of 0.80 but in external validation studies the performance of these scores is often in the range of point 65 to 0.75 exactly in the danger zone established here so the solution may seem simple. We should work on improving the performance of the severity scoring systems. But unfortunately it’s not so easy and to see why let’s consider another scenario so far we’ve looked at treatments that are either given or withheld in reality though treatments are often titrated to effect and this titration is associated with severity of illness. Now can we find out in theory. Which dose of a treatment is best given any level of illness together with my colleagues in amsterdam. We set out to investigate this problem to understand what we wanted to know. Have a look at this statement. Which could again be from. A hypothetical observational study a positive fluid balance was independently associated with increased mortality risk after adjusting for severity of illness. What we wanted to know is under which circumstances are statements like this valid reflections of a true causal effect so we built a simulation study. We built a simulation architecture and we asked this question given that the treatment is titrated higher in more severely ill patients and given that. There is an optimal dose for each level of illness.
What is the estimated effect of those on outcome after adjusting for severity of illness. What we found was that common statistical methods such as logistic regression propensity matching or inverse probability weighting misleadingly demonstrated a significant association between higher treatment dose and death when the severity score is less than perfectly calibrated so this confirms the findings from the previous study. But we also found that even if we have a perfectly calibrated severity of illness score this bias. This biased estimation could not be overcome except under unrealistically ideal assumptions. Estimating the correct treatment effect required a perfect severity score which doesn’t exist and perfect symmetry in the applied doses and perfect symmetry in the dose effect relationship. Now all of this together just doesn’t happen in reality. Finally and disconcertingly. We found that. I’m sorry finally. In this uncertainty we found that for observational studies investigating a therapy that is titrated to effect and associated with severity of illness. Larger sample sizes lead to more precisely wrong estimates. This is because the bias is systematic rather than random and it means that a 50 000 patient study is more likely to find a false treatment effect than a 5000 patient study all of this led us to conclude that for observational studies again investigating a titrated intervention the most commonly used statistical methods give misleading results. Because it’s hard to believe because these methods are used so often you can try this for yourself together with the paper. We built an online interactive tool. You can first define how a treatment dose affects the probability of death and then you define the accuracy of a severity of illness score from very inaccurate to perfectly predictive. And then in the next step you are able to analyze the study and you can try to estimate the true treatment effect so the optimal treatment knows that you previously defined yourself.
And you’ll see that it is almost impossible to arrive at an accurate result. So what’s going on here. At the heart of the problem is a phenomenon. Statisticians call endogeneity which means in technical terms that the errors of a model are correlated with the explanatory variable this endogenity arises through measurement error in this case of the severity of illness core. And through what you can think of as reverse causation also called simultaneity between treatment dose and illness severity. If we look at our castle structure again we had hoped that propensity matching or logistic regression. Were able to lead to an unbiased estimation of exposure on outcome in theory. This is quite possible. But if we add to this theoretical framework some clinical realism by stating that severity of illness can only be measured. Inaccurately the estimation becomes muddied and often invalid if we add some more realism by stating that. The severity of illness influences exposure but exposure also influences the severity of illness estimating the correct association becomes even more problematic. If not impossible the bad news is that there is just no way of obtaining a reliable estimate of the effect of treatment on outcome without additional data. But if we have some additional data there is a trick if we have a so-called instrumental variable a factor that influences only treatment directly and not outcome then. We can arrive again. At an unbiased estimation such instrumental variables are difficult to find but not impossible. This paper is one example of an excellent analysis that uses an instrumental variable so these were biasing mechanisms that are endogenous to the analysis that is to say these biasing mechanisms occur even though we include all relevant variables in the analysis in the model but there are of course also biasing mechanisms that are exogenous to the analysis that we need to be aware of when designing or reading an observational study specifically relevant to the intensive care are the time-dependent biases.
This is because the moment of inclusion in an observational study is often the moment that the patient is admitted to the icu or admitted to the emergency department and often the exposure we’re interested in occurs at some time point after inclusion so there is some time that the patient is included in the study but not exposed to treatment if the patient dies during that time he or she cannot be exposed to treatment. But if you are not careful such a patient can be compared to a patient that was exposed. This sounds trivial but in fact it leads to a very important bias. The so-called immortal time bias and it happens quite a lot in observational studies. There’s an excellent paper just out in the blue journal that i can highly recommend on this topic. Another factor to consider is time variable confounding. Suppose that you’re interested in the effect of a trajectory of some biomarker over the first days of admission perhaps if the biomarker decreased over 48 hours this is associated with poor outcomes. What you have to consider of course. Is everything else about the patient that changes in the same time frame. Ignore this and the results may be biased. Through time variable confounding and finally it is important to consider the outcome of interest and whether this outcome is possible at all in every patient. Suppose you are interested in the relationship between plasma vitamin c levels and the risk of acute kidney injury you find to your astonishment that severe vitamin c deficiency is associated with less risk of developing acute kidney injury. How can this be well. Perhaps because those patients with the most severe deficiency are most critically ill and they die before they get the chance to develop full-fledged aki death in this case is a competing risk for acute kidney injury. Now all of these biases seem perhaps a little straightforward or not but a quick screening of the recent literature learns that at least ninety percent of observational coveted studies suffer from at least one of these time dependent biases.
So where do we stand a new look. At this statement exposure x was independently associated with increased mortality risk after adjusting for severity of illness. There is nothing intrinsically wrong with this statement. If the goal of the study was to find statistical associations it has succeeded but what many readers and guideline authors also think that this means is that there is some indication some indication of a causal relation between treatment and mortality risk. But we should be very careful what this actually means is nothing if we do not evaluate all the relevant biases that are at play depending on the magnitude of confounding by severity of illness and the accuracy of the prognostic score the true causal relation may very well be opposite or contrary to the reported association. Even with low confounding and with a well-performing severity score a truly beneficial treatment or exposure will be mistaken for harmful in more than 50 of observational studies. So this sounds like bad news and the question is are we completely lost and the answer is probably no to start looking for solutions. We have to acknowledge that. The methods like logistic aggression and propensity matching are techniques from the 70s and 80s of the last century. But we’ve now clearly identified the problems that lead to unreliable observational research how to do research, good research is. Those problems are a complex interaction between patient characteristics disease characteristics and treatments. Over time many latent characteristics are difficult to measure and the course of disease is variable between inclusion and exposure and finally treatments are not delivered in an old fashioned but are often titrated to effect in the icu. And it’s not illogical that if the problems are so multi-dimensional in nature we need to find solutions. That are multi-dimensional in nature.
We need to find multi-dimensional data now for a few years. The most famous example of high resolution data has been the mimic database high resolution granular data with lots of continuous data. Points per patient but mimic is american and perhaps not so relevant for the european practice so we now have amsterdam um cdb. High resolution granular data from thousands of patients in amsterdam and completely new is the covert predict database collaboration of more than 40 hospitals that now contains more than 500 million data points of thousands of covet patients using this kind of granular and longitudinal data. We can employ modern statistical techniques to really learn something about the complex interaction between patient characteristics disease characteristics and treatments over time. I’ve already mentioned this. Example of an instrumental variable analysis. Applied here very cleverly to answer the question whether early ico admission leads to better outcomes than later ico admission now. This is another great example. How can we quantify the complex interaction between disease severity. The trajectory of disease the onset of delirium and poor outcomes sounds very complex and rather than simplifying this problem. The authors chose to use a marginal structural model addressing explicitly. Time variable time variable confounding competing risks the evolution of disease before delirium onset and the baseline covariates. So how should we see observational research how to do research, good research is in the icu. We should recognize that. Finding an independent association after adjusting for severity of illness is meaningless without any valuation of other biasing mechanisms. Besides confounding yet. The icu has the most data rich patient population possible with ample opportunities for very valid. Causal inference studies so we need not relegate observational. Icu research how to do research, good research is with a causal goal to the junkyard of science as long as we try to recognize and deal with all biasing mechanisms.
That may be at play. Thank you for your attention. And i think it’s time for the discussion in the meanwhile here are some interesting pointers to material that you may or may not be interested in thanks again. Thank you very much dr degroot i i really enjoyed both presentations and i think that the future of intensive care medicine is in very very good hands when i hear uh just the level of expertise that we’ve heard today so we have a number of questions uh at the moment they’re mainly for mariangela because yours are just coming in so i was going to start with questions for mariangela so the first one that we got here is you mentioned. Uh the negative. The concerns around negative results and maybe a pressure for positive results in the laboratory. And so a question was around you know. How do we guard against concerns about data falsification and you know is you know how can we have systems to protect against that well interesting question. At the moment i think that the most of the work is do is done by the peer reviewers and by the editor of the journal. Uh when you you send your paper in well your your data are not yours anymore. So they are free to to be scrutinized by everyone and actually our system is based on the well how to say the reliability of all the authors and uh it. It’s a it’s a that can be a problem and can be included in the future. Um by probably uh sharing all the raw datas and the analysis of each study and experiments and make it free for the use of everyone that wants to come. What’s your idea yes i i think that’s a very good point. I mean the more data you provide you know the raw data does help and and i think there are advanced techniques. Statistically to show you know that data is is is is true or or falsified a another question. Uh mariangela is asking about funding. And so they’re asking about you. Know commercial sources of funding so pharma or foundations or or institutions. And how does one stay independent.
Uh of of maybe influences. Uh that come with funding. So how do you separate. Uh you know the funding you get say from a company from the work that you’re doing. Oh well i’m probably too young to answer to this question. My i have tonight. My my founding came from public uh institutions or i have never been founded by a company. Um well john. That’s probably a question for you so yes i think that’s a fair point right angela. So i mean really i suppose the only tool we have is the declaration of funding. So you know an honest declaration of conflict of interest and so to allow readers then to know that right but of course the difficulty is maybe we don’t always know if there’s a full declaration or not. It’s certainly a concern. So um the last question for you marie angela is about uh your your animal models and so the question is how you know. What animals do you work with and you know. Why did you choose those animals. And how well do you think those models reflect the human condition where the immune system and comorbidities and so forth are more complex. Oh well each research how to do research, good research is has its own perfect error. More more suitable model to use there are well. There is a part of the research how to do research, good research is. Study the different kind of model that you can. You can use in my case. Uh in in mechanical ventilation or respiratory uh physiology. Uh the the model that is most useless swin model so we speak because of anatomically and well physiologically very close to the human respiratory system for what concern the studies about sepsis and immuno immunity. And so on i know well more different models can be more suitable and i know well. That’s for example mirin’s model are not uh how well can committed wrong but well from which kind of research how to do research, good research is we are you are performing. Which which model is the best suitable for your purpose. Yes thank you very much marie angela. There have been uh other questions as well but i think in the interest of time we’ll move to to uh some questions for uh harmion so uh honey and you mentioned uh that illness.
Severity scores function really quite poorly. In terms of how well calibrated. They are and how well they predict outcome. What is the solution to that. Do we need a better score a different score or you know. Could we look at illness. Severity over time. Perhaps as a way of doing this so i i think first the most important point is that uh. If you’re doing an analysis like this so you say i found an association between some vector of interest and and um and outcomes and i’m adjusting for severity of illness at baseline and i’m using as a severity score a well-validated score say the apache score. Then we need to realize that this these scores have been shown to perform very differently in in different studies and in different populations so of course they perform best in the in the original publications right but then there are various sub-populations for which the scores perform either not so well or just poorly so what i would recommend is the first step is just if you do this then report how. Well the score is calibrated and how well it discriminates in the population that you’re investigating which is quite easy. It’s it’s a few lines of code in a statistical program or a few clicks and you can put it in the appendix so that people interested in this like me can see all right. So how did the score perform in in this study was it was it accidentally or was it poorly if it was excellent then then that is another another checkbox that has been checked in the validity of the study but if it if it performs very poorly then then you need to find other solutions and then like exactly like you said secondarity if if if the analysis is more or requires more complexity over time then. I think it’s important to find this scoring system that tracks severity over time. So of course the most the most famous is uh. Is the sofa score. And that’s actually been that’s actually been employed a few times successfully in in these kinds of studies.
A third option is to develop your own. There’s nothing wrong with that. There’s nothing statistically wrong with developing or developing or taking a basket of variables that you think is especially relevant for the clinical question that you’re answering or especially irrelevant for the confounding that you think is happening. So if you’re doing a study in um in covet patients and you’re interested in some respiratory characteristic and you want to or some ventilator setting and you want to adjust the association for illness severity then is there. Are there good reasons to use the apache score or the substitute score. I think not. I think it’s much more reasonable to to take a basket of variables like the pf ratio like other uh prognostic variables and just use that in in a in a statistical model. And the what we. What would you what you would call the degrees of freedom that you lose by picking your own seven or eight variables if you have a thousand patients. That’s that’s more than reasonable so i think those are three important options. Okay thank you so a question. Another question uh from social media so it’s a question uh they ask. Am i. Correct that adjustments for only severity illness scores are flawed but that complex multivariable analysis are still viable so can we still do complex multivariable analysis. Are they still viable in the icu. So again there is the like the the statistical methods they work periods. They they do what they should do. If you recognize and acknowledge the assumptions that underlie the the analysis that you do so if i i can i honestly cannot think of a situation but if you can think of a situation that the confounding that is taking place between exposure and outcome is only through severity then perhaps using a severity score would work but in many cases it it. It’s just not the case so it doesn’t work and i think it’s difficult to to generalize like this what what does work. And what doesn’t work because everything depends on the on the type of question you’re interested in so even if you have the most complex and advanced statistical method multivariate using a marginal structural model and taking into account complex time relationship patterns.
Then everything you do can still be biased for example by a selection bias because some patients may be in the sample because they were treated further up the line. They were treated differently you know. They entered the sample while other patients didn’t so it’s i would say keep it as simple as possible but not simpler. Okay thank you so and this is just a general question. It’s the last question. Uh do is this particularly a problem in critical care um because of our syndrome diagnosis and multiple or you know our clinical syndromes. And the fact that you have multiple organs involved and that the pathway to death is often not clear um or is this. Is this a system-wide issue. It is to appoint a certain a system-wide issue but we have some problems in intensive care that make it specifically so especially the the complex time-varying the the time-varying characteristic of the disease of treatments. But we have to realize that if you believe that that smoking causes lung cancer then you believe that observational studies work and they do work because no rct has shown that smoking causes lung cancer. We know this. Purely from smart observational studies that have been performed in in in dozens of populations and each study has looked at another aspect and and and together the body of observational in that body of observational research how to do research, good research is. We’ve triangulated that word. 100 sure that smoking causes lung cancer so in that setting it was probably easier than what we’re trying to do in intensive care medicine. We have some. We have some specific problems that make it more difficult but we also have some specific advantages. We have tremendous amounts of longitudinal data. And i think if we if we can step back from the over simplistic approach of just saying this and this is associated with that and that and i’ve adjusted for severity of illness if we can step back from that then opportunities abound for observational research how to do research, good research is.
Thank you very much a really fascinating subject. Uh so i’m going to now close the webinar. We are just one minute to the hour and as a chair of the translation biology section. I am really delighted that we had uh such a great uh webinar. I really want to thank dr pellegrini and dr degrooth for two fascinating talks and i’ve no doubt that you’ll get plenty of questions through social media after this so thank you both and to our audience. Thank you for uh joining today and please do forward any questions that you have thank you very much. Thank you bye bye. We will repeat the marathon at the end of may again a marathon of eight hours with experts from different parts of the world to share what we’ve learned from covet. How we can repeat the future and how we can come out of this situation together you.
Where to find great research papers?
Various great research journals such as Global Research Letters are a great option and way to help you look up impactful research papers with a great format. Here, you will find a number of various research papers that are provided and made available to you in the journal, which will help you write your own paper.
You can very easily find papers on a variety of topics at Global Research Letters, which will help you with your own research work and understanding of writing and publishing research papers properly. With access to so many amazing research papers, you can practice and learn the process of writing research papers and their importance.