SCCM Resource Library-Medical Journalism in the Age of Social Media, Big Data, and Artificial Intelligence

Video Transcription

I particularly really appreciate the critical care societies, ESICM, SCCM, and ATS. You've been very welcoming over the last seven or eight years since I arrived at JAMA,  so I really want to appreciate the ability to come and speak today. I'm very appreciative. I am the editor of JAMA, but I really represent quite a few people,  and I just really want to mention that before I start. JAMA and the other journals, the new journals, Oncology, Cardiology,  and now our new offering in the digital space, JAMA Health Forum, there's about 250 people who work on these journals.  They're very well-resourced, and so some of my comments really reflect what represents a well-resourced and extraordinarily well-run collection of journals.  In addition, particularly in critical care medicine, I think people know that Derek Angus has been our associate editor for many, many years,  and I think that's really helped the journal in terms of its relationship, both with individual authors as well as with the various societies.  My other comment is I have nothing to declare. I have a contractual arrangement with the American Medical Association, which, one, guarantees editorial independence,  and secondly, it does not permit me any honorarium, and I hold no stock in any pharmaceutical house. So I'm going to start with some comments about social media.  Joy mentioned that when I started, we had about 13,000 followers. This morning, we have well over 900,000 followers on Facebook and Twitter.  Recently, we've added LinkedIn, YouTube, and Instagram. JAMA last year had 55 million views of its content, more than a million views a week,  and 20% of those views are generated through social media. And so for a modern communication center, social media, obviously, along with PubMed, New York Times, Google,  have really become an incredibly important source of contact between us and our readers, many of whom are authors outside the United States.  At JAMA, about 51% of our readers are from outside of the United States. But a number of issues have begun to develop in social media.  So I think people recognize over the last three or four years in the United States and around the room, rather, around the world,  there's been this notion of near facts, facts that aren't facts, and facts that seem to be elusive. I am tremendously concerned that that's beginning to dominate within healthcare.  Over the last year, I've seen forms of social bullying by very prominent academic physicians that have gone on in social media.  And I worry that in five years, we actually will not recognize communication in science. How we deal with that is very, very uncertain. We've published a number of articles.  Rainer Martian, University of Pennsylvania, Protecting the Value of Medical Science in the Age of Social Media and Fake News,  Addressing Health-Related Misinformation on Social Media by Rob Califf. So this has really become an increasingly important issue.  We're often asked if we can correct errors that occur in social media after we publish a paper. It simply isn't possible.  I'm often asked, well, could you call up the Wall Street Journal and the New York Times to get them to correct the mistake? That doesn't work very well for us.  We're very careful about what we say about articles, what authors say about articles, the press releases we use. But on the other hand, there's a complexity to social media  that I don't think we anticipated 20 years ago with the birth of the Internet and then Twitter and Facebook.  And I think it's obviously been exacerbated over the last couple of years because of the tenor of the conversation within politics, both in the United States and around the world.  But there are some truly amazing success stories for social media. This is the largest case series published to date on COVID-19, obviously the coronavirus epidemic.  We published it a week ago, one week ago. As of this morning, it has 492,000 views from around the world. It's an extraordinary number.  In social media, it's been tweeted 4,454 times. Those 4,500 tweeters have 13 million followers.  This was unimaginable, unimaginable a decade ago, the way in which you could use or move information around the world so rapidly.  And for obvious reasons, there's a thrust, a thirst of information or a thirst of interest in information about coronavirus. I was mentioning to people over the last week,  we've gotten more than 50 articles about it. Now, when journals publish rapidly, we make mistakes. And that is one of the major issues that all of the journals have faced  in the last week or ten days. This paper was held up for one week, rather for one or two days, because there's a very important statement that was picked up  by all of the media, that of the 138 cases, 40 or so were healthcare personnel that were infected by the index case within the hospital. We didn't understand how they defined it,  but we knew that that was an incredibly important message. And so we held it up for one day until we were quite sure that we understood that definition.  But when we begin to publish very quickly, all of the journals are subject to some concern about the validity of the results.  And I've said publicly and in print that all of the papers on coronavirus should really be seen as preliminary reports. They're largely case series. They're very selective patients.  And I think we don't really know how well the information from who they really is going to generalize around the rest of the world. My other comment about this particular outbreak,  I believe that the success around the world in reducing spread has been remarkable. There are only about 16 or 18 cases in the United States.  More than 2,000 individuals traveled back from Hubei to China over the previous six weeks prior to the acknowledgement or the recognition of the outbreak.  And so it's a huge success story. And I think our public health facilities and community and the ID community are to be congratulated. As an aside, 12,000 individuals died from flu  in the United States in the last six months. And I think it's just beginning to emerge kind of the fascination with coronavirus versus the lack  of recognition of the incredible mortality that flu takes or carries each year. So JAMA's been in the business of big data and artificial intelligence for many years.  Actually, I'm pleased that I think of the major journals, we really have led the way. And I think when we published this in 2014,  Zach Kohane from Children's in Boston is the lead author. We should have recognized what was going to occur over the next five years.  So Zach sent this in, and like many other papers that we had been getting, he said, well, can I write something about big data? And I said, can you define big data?  And he said, no, I can't define it, but I can draw it. So he hand drew this graph, and then we perfected it. This viewpoint, which is only two pages, has 150 citations.  It represents what has become to be known as liquid gold, which is data. Hence the interest of Google, Apple, Microsoft, and Amazon in healthcare.  They own and create an extensive amount of data. And what's unique about this is that data can be structured within the healthcare system or unstructured.  Blogs, tweets are on my right, your left. And I think this really represented what we should have understood was going to be the forerunner of artificial intelligence  or machine learning in medicine. And when I went to meetings over the last couple years and people were talking about AI or ML or big data, this slide almost always appeared.  And I think it really represents the richness of what's available to us. Whether we can make good use of it is a very, very different question.  Now, I know that the plenary speaker yesterday spoke about artificial intelligence. I think my views of it may be a bit different than others. The paper at the top,  Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs, was published in 2016.  As of this morning, it has 925 citations and close to 300,000 views. It is unequivocally the landmark paper in terms of recognizing what the promise is of machine learning and AI.  As I said, it was published in 2016. It comes from the Google folks. They have since taken this algorithm and are using it to diagnose diabetic retinopathy in India.  And there's particular reasons why they're doing it in India. And ultimately, they hope to report out patient outcomes from the use of this algorithm.  Since then, we've published five other research reports and probably 20 opinion pieces about artificial intelligence and machine learning. I want to point out the one at the bottom.  What this computer needs is a physician, humanism, and artificial intelligence. That's by Abraham Verghese, who I think for years and years and years  have talked about the need to touch patients, to feel patients. That's the humanism. That's the humanism that patients want. I'm actually impressed that critical care physicians  are superb at doing this, despite the wizardry. Despite the wizardry, you see and touch your patients. I was very struck by the late breaker yesterday  on non-sedation, seeing all those pictures of people up and around, walking around in the ICU was very, very impressive. But Abraham has been very concerned  about artificial intelligence, electronic health records, and whether or not that's going to strip away what is one of the most important aspects  of the physician-patient relationship. Artificial intelligence in healthcare, will the value match the hype? This is by Zeke Emanuel. Zeke purposely tries to be provocative.  Some people are uncomfortable with that, but it's always, always an interesting read. We're about to publish another viewpoint from him  about what medical education should look like in five years. It, too, is quite provocative, but it's an important read. And then Jeff Hinton, who's the father,  the father of deep learning, wrote for us about two years ago. I have to say, when the viewpoint first came in, I didn't understand a single word of it. And we had this exchange.  I had to keep telling Jeff he was writing for people who aren't bioengineers and know this world. And it must have taken five back and forth with him  because he just struggled to make it simpler so that you could understand it. And then the one on your right, potential biases in machine learning algorithms  using electronic health record data. This was actually published in JAMA-IM two years ago. I think people are aware that there was a Nature report  earlier this year that got quite a bit of attention and press. I'm pleased that, once again, we anticipated the future, but we published it almost two years ago.  And I would make just a couple comments about machine learning and potential biases. I was in a meeting in China, and Deb Schroeder is our associate editor in oncology,  and she got a call from a patient. She took the call. She was with me in China. And a paper had just come out in another journal. And the father, or rather the husband, was saying,  well, Dr. Schroeder, why can't my wife get this treatment? And Deb looked up the paper. It was a randomized controlled trial. Incredibly well done. His wife was 90.  The table one in the randomized clinical trial were 60-year-old women and men who were healthy. So this issue of non-representation in artificial intelligence is very similar  to the concerns about clinical trials. Critical care is a bit different, but in cardiology and oncology, there's a long history of those patients  that get enrolled in clinical trials not being particularly representative of the patients who are being seen by those physicians.  And so this notion about machine learning algorithms being non-representative is very, very similar to the same issues that have arisen around randomized clinical trials.  There's some other hidden issues around algorithms and AI that people worry about. And just a couple other comments. AI isn't a monolith. There's many, many different aspects of it.  And so some of the concerns that are raised are true for certain types of AI, but not others. So across the landscape of the articles that we see, we see four types of AI articles.  Those that are recognizing novel patterns in CT scans or in cancer diagnosis. Reading images has, without doubt, been the most prominent. Mammography, diabetic retinopathy,  x-rays of many different types. Making predictions. Now, in making predictions, some of the algorithms have been very good. But like polygenic risk scores, the question is,  is it any better than the more simpler approach? and sometimes we get seduced by all of this new wizardry, but in the end it doesn't offer very much more.  So there's now a genetic-based risk score for cardiovascular disease. It is no better than Framingham, and Framingham is a lot easier to understand than to do.  And so I think we need to worry about or be concerned about being seduced by AI. And then the fourth bullet, improving clinician decisions.  And I am extraordinarily pleased that this afternoon, at this meeting, we believe we will publish, along with the author who's presenting it, the first machine learning  paper that represents an improvement in patient outcomes based upon the use of machine learning. It's a randomized clinical trial from Europe. It will be published this afternoon.  It's a late breaker at this session. It's a process measure, not a true patient outcome, but we believe it represents the  first, if not one of the first papers that really indicates that AI can improve patient outcomes.  So I think original investigations really need now to move away from that the machine  is as good as physicians or better than physicians or as good as clinicians or better than clinicians, and really begin to focus on whether or not it's improving patient outcomes.  We're often asked, how do we review AI papers? You know, it's kind of this black box, and people have been concerned about that.  Rob Golub is a director of scientific review at JAMA. He's a methodologic and statistical whiz.  And in general, I would say the extraordinary principles that dominate in high-quality research  are the same in AI. And so depending upon what the article is focused on, we generally use the same type of checklist guidelines that have been developed for other article types.  So randomized clinical trials, diagnostic test interpretation. And those are all based upon the Equator website. The last is, what does AI's future mean to a journal editor?  I can envision in the next five to seven years that every paper that comes into JAMA, we get 4,500 papers a year, about 1,500 go out for peer review, that every paper coming in  to JAMA or the JAMA network journals will go through some form of AI evaluation that will provide a checklist for peer reviewers to focus on.  That is imaginable to me, in part because there's so many checklists that are now available for the article type.  So if the article type indicates it's an observational cohort study, or a randomized clinical trial,  or a meta-analysis, then CONSORT, or STROBE, or PRISMA could be applied through the AI or the machine learning to that article and generate a checklist of whether or not those  authors followed the recommended items for each of that type of design. And so I can envision the use of AI in peer review. These are really complicated issues.  I'm going to spend quite a bit of time on this slide. I only have three more slides. I really have tried not to use many.  I want to start with the first bullet, which is data sharing versus reproducibility. So these are very, very different issues, and people have conflated them, and it's created  a huge problem. Three years ago, the International Committee of Medical Journal Editors, ICMJE, JAMA and the other leading journals founded this in 1978.  They're the group of individuals that have brought you conflict of interest forms that you all love to complete, author criteria, and then the need for trial registration.  Three years ago, ICMJE put out a statement because there had been a lot of pressure about sharing data that underpinned randomized clinical trials.  I want to be very careful that ICMJE has been focused on randomized clinical trial data. We put it out as a test case, and I really appreciate my colleagues from Canada.  They really are extraordinarily nice people. We got the angriest letter from 300 trialists from Canada, and they're such nice people.  They were not very happy with what we were recommending, so we floated the idea that after six months, a trialist would have to make those data available to other individuals.  It would be those data that underpin the report of a randomized clinical trial. Why is this necessary? Because those data don't belong to the investigator.  They belong to the patients, and patients beginning about five or six years ago have begun to complain that very valuable information is being lost because trialists have been  holding on to those data. This has become an important mandate through the NIH and clinicaltrials.gov, although it's slightly different.  We heard the criticisms, and also many other issues evolved over the subsequent 18 months, which were, if you share data, who pays for it?  If there's one trialist, he shares the data, or she shares the data, and then someone else calls that person up to get help.  What's the responsibility of the initial investigator to help that individual? Was the goal to say whether or not the investigator had analyzed the data correctly, or was the  goal hypothesis generation? And so numerous issues arose after our first ICMJE statement, and so we released a second  one a year ago, which said, beginning on July 1 of 2019, if you submit a randomized clinical trial to one of the ICMJE 2000 plus journals, all we're going to do is request  that you tell us if you will share the data, how you will share the data, and when you will share the data. We have not mandated data sharing.  Some journals, plus Medicine, BMJ, mandate data sharing, but the journals that publish the lion's share of high-quality randomized clinical trials do not mandate data sharing,  and there remains this tension about, if you share data, why are you sharing it? Who are you sharing it with?  What are the administrative and financial underpinnings of data sharing? And many of those issues continue to need to be worked out.  This is different than the issue of reproducibility in lab-based science. So about five years ago, there was a number of reports that indicated that industry could  not reproduce many of the underpinning scientific experiments that were being conducted in various laboratories around the world.  Now in hindsight, it's possible that they actually weren't reproducing them exactly. So instead of using a 25-mL beaker, they used a 50-mL beaker.  Well, that creates a problem around reproducibility, but this has become a central issue, and the National Academy of Medicine has weighed in with a very long monograph about it.  But people have confused data sharing of clinical trials with reproducibility issues around preclinical research, and they are very, very different issues.  Moving to preprint servers. So what is a preprint server? Actually there are 55 preprint servers in the world, actually more than 50.  BioRxiv is the best known, SSRN is another well-known one. A preprint server emerged in clinical medicine about three years ago when BioRxiv began to accept clinical papers.  But actually, preprint servers have been alive and well in economics, mathematics, and physics for over two decades.  It allows an investigator to put an article up, prior to submitting it to a journal, and then it's available to the world for commenting.  Theoretically, people will comment on it, people will read it, they will modify the manuscript and then submit that manuscript to a journal.  And that has been going on for more than two decades in an enormous number of non-biomedical sciences.  As I said, this came to clinical medicine beginning about three years ago, and there has been some concern that clinical medicine is different than other types of investigations  and whether or not the clinical community will be well-served by preprint servers. There's been tension about this. It's unresolved.  Some of the concerns are that if there's fundamental errors in terms of the paper that goes up on a preprint server, are they ever corrected?  Some people have said, well, preprint servers will not be consumed by the media. We need not worry. Other people have said, well, that simply isn't the case.  I have said, I am a pediatrician, any article that indicated that immunizations likely cause  autism should never go up on a preprint server, that that paper really needs to be peer-reviewed  because based upon a Bayesian analysis, the likelihood of that being true is very small. Harlan Krumholtz from Yale, along with Joe Ross, has started something called MedArchive  about six months ago, and it's an outgrowth of BioArchive, and it's hosting clinical medical papers. I give Harlan and Joe a lot of credit.  They heard the concerns about clinical medicine being very, very different. And so about 20 or 25% of the papers have not been posted following pre-review.  There were either ethical concerns or concerns about what the conclusions were. There's already been a number of examples of papers that have been placed on preprint  servers that are wrong and may influence clinical care. And I think the story about preprint servers remains uncertain.  Most journals, including the JAMA Network journals, are permitting authors to post articles on preprint servers, but we want to know that you've posted it on a preprint server,  and it will be part of the deliberations when we decide to publish or not to publish a paper. Open access. Open access debuted in the early 2000s.  There were a number of statements from Berlin and other countries indicating that we needed to move to a different model of publishing.  So for 150 plus years, the traditional model is you would send an article into a journal. The journal would decide to accept or not accept that publication.  If they accepted that publication, the journal would own copyright. We would monetize it in some way, and then at some point it would be published.  And then beginning about 15 years ago, original investigations after a year would be made publicly available for anyone who wanted to read it.  But that initially, that article would be unavailable unless you had a subscription to that journal.  And beginning in the 2000s, people felt like that, in terms of the world of open science, was not the best model going forward. And so open access debuted about 2005.  It continues to evolve. I think people are aware that the next evolution of open access is the so-called Plan S model  coming out of Europe, in which a collection of funders, including Wellcome and Gates, NIH equivalents in Germany and the UK, have said, if you take our money, then the article  must be published in an open access format. That is, where the author owns the license, the CCBY license, and the article is immediately  available, immediately available to the public. Now there's often consequences when you develop new models.  And so I think there's been two or three major consequences for open access that people did not anticipate. The first is the so-called predatory journals.  You open up your email every morning, and you're asked to submit an article to a journal, and for $50, they will publish it. Those are the so-called predatory publishers.  That is not PLOS. That is not eLife. But it's this whole cottage industry in very rapid publication.  And I think people have come to recognize that it's unfair to conflate all of open access publishers with so-called predatory journals.  Now there's about two or two and a half million articles published a year in SMT journals, science, medicine, and technology.  The estimate is that about 15% to 20% of those are published in an open access format. Many of those are published in so-called hybrid journals.  So PLOS or Biomedcentral are fully open access. Everything they publish is under a CCBY license. Hybrid journals mix.  So the JAMA network journals, JAMA IM, JAMA Oncology, about 10% or 15% of the authors, depending upon the journal, choose a CCBY license.  So some of the articles will be immediately open access, others won't. Now this story continues to evolve.  And one of the central underpinnings of open access that I think people didn't recognize,  it's very much like in the United States, value versus volume. The business model for open access is entirely based upon how much you publish.  If you publish more articles, you have more income. And I think some people have begun to articulate that that brings with it its own ethical dilemmas.  And so open access continues to evolve, but it's a very, very important issue. I just spoke to someone from the NIH.  The HEAL initiative, the opioid dollars, which is about a billion dollars that the Congress  has released to the NIH to fund opioid-related research, as well as the cancer initiative.  In those two initiatives, if you take those dollars you are obligated to publish in an open-access format. And so  that's become increasingly important to the NIH and those dollars were released to the NIH with the agreement that the articles would be published in an  open-access format. JAMA, New England Journal, do not have open-access models at the moment. All of the JAMA Network journals do, except for JAMA, and then we  have a fully open-access journal, JAMA Network Open. Conflict of interest. I have almost always said, unless you win the Nobel Prize, you do not want to be on the  front page of the New York Times, and you definitely don't want to be on the front page of the New York Times above the fold. Last year, a very distinguished  investigator from Memorial Sloan-Kettering, Jose Bezegla, was on the front page of the New York Times two or three times. And it was around, yet again,  the complexity of conflict of interest, both at the individual level as well as at the institutional level. It is something that we continue to struggle  with in medicine. The picture at the bottom is from a series of articles that we published almost three years ago. I think we recognize that it was  continuing to evolve as a complicated issue. More and more, the investigators of articles that we were processing clearly had conflicts of interest. They consulted  to companies. They had their own startup companies. Of interest, I would say, is that I'm much less concerned about our review of original research. We see the  science. We see the data. So even if the investigators are from industry or from the NIH and have conflicts of interest, those are data-driven pieces. And I'm  much less concerned about conflicts of interest in those pieces. Plus, for people who have submitted to JAMA, they know that we work with authors quite  carefully about the language. I'm more concerned about conflict of interest in opinion pieces and in clinical reviews. And we do struggle more and more with  what is conflict of interest that would eliminate someone from being an author. Just two other evolving issues around conflict of  interest. The first is many of you have completed the form and one of them says, please declare everything that you think is relevant. And then you quickly hear  from a journal, well, someone else thinks something else is relevant that you didn't declare. So this relevant statement has really made it difficult  for investigators. And so I think over the next year or two, we may move away from the so-called relevant statement and just say, tell us all of your  potential conflicts of interest, asterisk the ones that you think are relevant, and that's what we will publish. And I think this is gaining more and more traction.  The other issue that I think the New York Times series highlighted was that many, many years ago, for people who are older, remember that we  said, just tell us if it's more than $10,000. If it's less than $10,000, you don't really need to tell us. And then we did away with the kind of dollar  amount about 15 years ago. And so we haven't really attended very much to the dollar amount. Well, the figures that were reported in New York Times were in six  figures. Some were in seven figures. And in my conversations with different ethicists, they believe there's a difference between $10 and $500,000.  And I think organized medicine has to grapple with whether or not the amount of dollars involved really need to resurface again in conflict of interest  statements. The fifth bullet, data sharing and politicalized science. At the very bottom, you'll see the challenges of sharing data in an era of  politicalized science. In the fall of this year, the Trump administration announced that all data that underpin the EPA policies need to be made  publicly available before they could be used, before they could be used in any changes to policy. This has sparked enormous amounts of concern about the  politicalization of sharing data. Now, this is not data that underpin clinical trials, but this is mandating sharing data in climate change or the environment.  Whenever we publish articles, for example, about air pollution, we hear from a particular group demanding that those data be made publicly available or that  the article be retracted immediately. And I just spoke to someone who's very involved in this issue and I said, you know, five years ago, who would have  thought that sharing data would end up becoming incredibly controversial? And I think the times have just changed. And so now I think we're beginning to  struggle with what does it mean to share data in an era when it could be manipulated in such a way so that the conclusion were to change. And so  this is an increasing struggle and it's unclear how it will evolve. Part of the struggle for the EPA, many of those data were collected in such a way that it's  possible that individuals could be identified, for example. And so I would say some people are, although they would champion sharing data, they are somewhat  concerned about why the Trump administration has mandated or proposed this type of sharing data. I want to finish with randomized clinical trials  and significant testing. So I think people are aware that in critical care medicine there was a really prominent article published last year in a New  England Journal. I was at this stage at ATS when it was presented. Jeff Drazen wasn't there, but he asked the editor who was there to read a comment. And it  was about the effectiveness of ECMO for ARDS. And the conclusion of the paper, New England Journal, indicated that it was not a successful trial. Reading from  Jeff's comment, he indicated although that was the conclusion of the trial, he actually thought it worked. Now here you have one of the most prominent editors  in the world criticizing the conclusion of a study published in his journal. And then I think people know that in the fall of this year we  published a special communication around a Bayesian analysis to the data that underpinned the E. coli trial. That study has been enormously well received, that  special communication. And I think it has revived, along with a number of other issues, about the appropriate interpretation of randomized clinical  trials. And I'm happy to carry on conversations with people after this ends. I can say unequivocally that deserting p-values and moving to  confidence intervals will not solve the problem of how to interpret a randomized clinical trial. Randomized clinical trials are complicated. Why are they so  important? Because they unequivocally stand at the top of the evidence pyramid. They stand at the top of the evidence pyramid either alone or as part  of meta-analyses. They more fundamentally change practice than any other article type. But interpreting clinical trials is enormously complicated. And I have said  on a number of occasions, both privately and publicly, you can't really summarize the value of a clinical trial in an abstract or in two sentences. Sometimes  you actually have to read the entire paper. And I know people don't always like reading the entire paper, but sometimes you simply have to. We don't  design clinical trials. We try to help authors be faithful to the way they design the clinical trial. So if they want to desert p-values in the  frequentist approach and go to a Bayesian approach, we're happy to support that. But what you can't do is design the trial with certain assumptions and then  at the end say, we'd really prefer to avoid those assumptions and reinterpret what we really meant would be a successful clinical outcome. That's what  drug companies did in the 1980s and that's why we have SAPs and protocols. There's a reason why SAPs and protocols now exist. So there's complexity to  interpreting clinical trials. Now where has JAMA tried to change? If you look at the trial that we just published earlier this week, it was presented in Belfast.  Kathy Ronan is the senior author. It's about a liberal versus a conservative approach to agonists in the critical care intensive care unit. And the p-value  is 0.07. And it says that if you let the mean arterial pressure come down a little bit, there's no difference in mortality. They designed it and  unfortunately the p-value was 0.07. They did not reach statistical significance. The confidence intervals, though, suggest that coming down on the mean arterial  pressure may have been helpful. So we insisted that based upon their design of the study, they conclude that the study did not meet statistical significance.  However, because of the confidence intervals, one should understand the entirety of the clinical scenario. And so we've tried to add context to the  interpretation of clinical trials. But they are complicated. So on the left, design and statistics, non-inferiority and equivalence hypotheses,  adaptive designs, Bayesian statistics, subgroup analyses, multiple endpoints and type 1 errors. Clinical trialists are so smart. When we began to hold them  accountable for secondary outcomes and that you couldn't report them unless you  had registered or had a SAP that indicated secondary outcomes, now the average clinical trial has 20 secondary outcomes, none of which are corrected for  type 2 error. So it creates just tremendous complexity in interpreting it. Many trials that we see, authors want to highlight two of the 18 secondary  outcomes but then neglect the other 16 that show no significant difference. Fidelity to the study protocol and the SAP, divergence and definitions of  primary, secondary and other outcomes. The H2 blocker PPI study from Paul Young that we just published, 25,000 patients. Part of the struggle was that 20 or 25  percent of the patients assigned to H2 blockers got PPIs. So you have crossover. It makes the interpretation of the trial far more difficult. And then  interpretation, clinical versus statistical significance, effect sizes smaller than planned, underpowered null study, primary versus secondary versus  exploratory versus post hoc analysis, and then the world of spin. So we will continue to work with the clinical community and investigators about  how to interpret clinical trials. But as I said, we don't design them. Investigators design them. We just try to help authors interpret them based upon  their initial assumptions. For the last slide, people have asked me, you know, what are the keys to publishing in high-impact journals? So the first  thing I would say, let's see, what's the first thing I do say? Read the journal. Okay? So if you read JAMA, we don't publish qualitative studies. Don't send  us qualitative studies. Okay? We're not likely to publish it. So you really have to read the journal that you're sending articles to. The cover letter may be  important. I don't have enough time. I love Richard Smith, but when I get a cover letter that says, Dear Richard, I assume that the paper didn't come to  JAMA first. It hurt my feelings. And so when you hurt my feelings, then I'm a little more likely to reject the manuscript. So when you write a cover  letter, know who you're writing it to, and I would suggest you don't write Dear Editors. You can find out the editor's name. And if you know that person, you  may want to address them by their first name. But if you don't know that person, you may want to address them in a more professional model. So the  cover letter may be important. This is the test of JAMA. Are the results new, true, and likely to change practice or public health? Now we have a high bar. 4,500  papers. We accepted 180 last year. The acceptance rate for non-randomized clinical trials is 2%. So we're a tough sell at JAMA. But this is true for most  journals. Even if it isn't necessarily new, what you need to do in most journals is at least say, what does it add to the medical literature? We  champion randomized clinical trials. When I arrived, we published 35 a year. Last year, we published 90. And people always say, do the major journals publish  trials that show no significant difference? Of those 90 studies, about 45 or 50 showed no significant difference between groups. So I think all of the  major journals have no problems publishing high-quality clinical trials because sometimes when the results show no difference, that changes practice or  influences practice as much as if it showed a significant difference, particularly for major clinical trials. The overall quality of the manuscript is  very, very important. The abstract is critical because many of the people who are deciding or voting on whether or not to go forward with that manuscript  haven't read the entire paper. They've just read the abstract. The other thing is if the numbers in the abstract don't match the numbers in the  paper, that doesn't make us feel very good about the way in which the study was conducted. 25% of published manuscripts have errors in numbers in  terms of confidence intervals not matching or frequency rates not matching. Our goal at JAMA is to be perfect in the abstract. We've had two  mistakes in 11 years of errors in our abstracts. Our goal is to be perfect. Then again, we have amongst the best manuscript editors in the world.  Okay, I'm done. And it says zero on the clock. It's a privilege to be here. Thank you so very, very much.

Video Summary

In this video, Dr. Howard Bauchner, editor of JAMA, discusses several important issues in scientific publishing. He first expresses his appreciation for the critical care societies and their support of JAMA. He also mentions the team of people who work on the JAMA journals and the resources available to them. Dr. Bauchner then discusses the increasing importance of social media in scientific communication. While it has the potential for rapid dissemination of information, there are concerns about misinformation and the impact on scientific communication. He also highlights the success of social media in the sharing of COVID-19 research. <br /><br />Dr. Bauchner then moves on to discuss artificial intelligence (AI) and its impact on healthcare. He mentions several important articles and viewpoints that have been published on the topic, including the use of AI in image recognition and the potential biases in AI algorithms. He believes that AI has the potential to improve patient outcomes and envisions a future where AI is used in the peer review process.<br /><br />Dr. Bauchner also talks about other evolving issues in scientific publishing, including data sharing, preprint servers, open access, conflict of interest, and the interpretation of randomized clinical trials. He emphasizes the complexity of interpreting clinical trials and the need for context and careful reading of the entire paper. He concludes by providing tips for publishing in high-impact journals, such as reading the journal, writing a strong cover letter, and ensuring the overall quality of the manuscript.

Asset Subtitle

Administration, Research, 2020

Asset Caption

The Society of Critical Care Medicine's Critical Care Congress features internationally renowned faculty and content sessions highlighting the most up-to-date, evidence-based developments in critical care medicine. This is a presentation from the 2020 Critical Care Congress held February 16-19, 2020, in Orlando, Florida, USA.

Medical Journalism in the Age of Social Media, Big Data, and Artificial Intelligence

Howard Bauchner

Learing Objectives:

Relate how social media has affected the source of readers for medical journals
Identify issues related to rapid publishing of medical articles on social media
Illustrate the impact of health related misinformation in social media

Meta Tag

Content Type Presentation

Knowledge Area Administration

Knowledge Area Research

Knowledge Level Foundational

Knowledge Level Intermediate

Knowledge Level Advanced

Membership Level Select

Membership Level Professional

Membership Level Associate

Tag Communication

Year 2020

Keywords

scientific publishing

JAMA

social media

COVID-19 research

artificial intelligence

healthcare

data sharing

peer review process

	Society of Critical Care Medicine 500 Midway Drive Mount Prospect, IL 60056 USA Phone: +1 847 827-6888 Fax: +1 847 439-7226 Email: support@sccm.org	Contact Us About SCCM Newsroom Advertising & Sponsorship DONATE	MySCCM LearnICU Patients & Families Surviving Sepsis Campaign Critical Care Societies Collaborative	GET OUR NEWSLETTER
© Society of Critical Care Medicine. All rights reserved. \| Privacy Statement \| Terms & Conditions The Society of Critical Care Medicine, SCCM, and Critical Care Congress are registered trademarks of the Society of Critical Care Medicine.