Literature Review: March 2026 (Part 2 of 2)

NTSP Literature Review Podcast: March 2026 (S7, Ep3.2) Part 2

This months papers cover:

• Next Generation Sequencing of Tracheal Aspirates
• Robotic PDT
• Bronchoscopy and PDT
• Staff roles and view of PDT
• Machine learning limitations for predicting PDT

Brendan is joined by Dr James Orr to explain all about sequencing techniques to identify bacterial fragments and what that means for clinical care.

Link to supporting PDF: https://tracheostomy.org.uk/Podcast-Resources

The UK National Tracheostomy Safety Project (NTSP) is committed to providing education, information and resources to improve patient safety and the patient experience for those with tracheostomies and laryngectomies. All of our resources are housed on our website www.tracheostomy.org.uk, accessed by over 30,000 visitors each month from around the world.

This is the only podcast to bring you literature reviews, hot topic discussions and interviews with healthcare staff, patients and families.

Our goal is to improve the safety and quality of care for patients with tracheostomies and laryngectomies through education. We work closely with patients, families and healthcare professionals to develop new resources to improve care. We’ve collaborated with key stakeholders in tracheostomy care since 2009, and developed freely accessible resources, supported by online learning developed with the UK Department of Health. We’ve worked with the Global Tracheostomy Collaborative since 2012 to improve care for patients and their families everywhere.

We are funded by grants, donations and in partnership with medical device companies through unrestricted awards. This podcast series is supported by unrestricted education funding from the Atos Learning Institute. The funding supports the professional production of the podcasts and videos, and the medical device companies that support us do not have any creative influence over the content that we record. All of our work is undertaken by volunteer healthcare staff, patients and their families. 

Most of our content is supported by videos. You can access our training videos and resources for Basic Care, Emergency Care and Vocalisation & Swallowing. Download and print bedhead signs and emergency algorithms from our resources. 

You can support our work by watching or clicking any of the advertising links that appear via the NTSP YouTube Channel. You can also donate directly to the NTSP through the NTSP website, or by clicking the Buzzsprout podcast hosting "support" links. You can support our work by watching or clicking any of the advertising links that appear via the NTSP YouTube Channel.

Transcript

SPEAKER_00 0:07

So, welcome to part two of March 2026's tracky talk. We're talking about papers that have caught my eye in this particular month, and there's a lot to cover this month, and so this is part two. Okay, let's get straight into the papers. Next up is yet another paper tackling one of the long-running issues in PDT. Do we actually need routine bronchoscopy during the procedure? This paper was a multi-centre study called FibreTrack, conducted in four medical surgical ICUs in Spain and led by Jose Agnon. Now for many clinicians, bronchoscopy during tracheostomy is standard practice, and the rationale is fairly obvious. Bronchoscopy allows direct visualisation of the away during the procedure, and in theory this should help reduce complications like missing the trachea, so paratracheal puncture, posterior tracheal wall injury where you put the needle in too far, false passage creation where you miss the trachea altogether, or misplacement of the tube. But equally there's another side to the bait. So critics of routine bronchoscopy argue it increases costs, which it does, it prolongs the procedures, which it might, it requires additional staff procedures, uh maybe, and it could introduce physiological problems of its own making. And in particular, that is related to the partial obstruction of ventilation by the mere presence of the bronchoscopy within the breathing circuit, and that can lead to increased airway pressures and hypercapnia where the CO2 levels rise up. And despite years of debate, international guidelines still don't give particularly strong recommendations either way. We've looked at this recently on this podcast with papers examining PDT in resource-limited settings. And I think the conclusion is that what we're seeing is bronchoscopy is helpful, but it's not necessarily essential, and that means you can do it without a bronchoscopy. So one of the limitations is that the bronchoscopy itself doesn't solve all of your navigation problems with PDT. It confirms that you're in, but it offers no guidance as to passage of the needle between the skin and the entry point of the trachea because that is just sort of inferred from when you're watching the trachea, it gets squashed and deformed, and you're trying to work out whether you're off to one side or in the middle. You know, that can be quite tricky. And so what this study did was take this sort of prospective multi-sender randomized trial to try and look at the question. So the investigators enrolled 442 critically ill adult patients across four ICUs between 2019 and 2024, obviously with the pandemic in the way, which caused a few problems for recruitment. Patients were randomized to either endoscopic guided PDT, meaning they use bronchoscopy throughout the procedure, or non-endoscopic PDT, where they did it without bronchoscopy. Importantly, all the procedures were performed by experienced clinicians using the single dilatation technique. The study specifically excluded high-risk patients, so things like difficult anatomy, coagulopathy, raised intracranial pressure, difficult airways, or very high oxygen requirements. So I think you can reframe this and say, well, this paper is actually asking in relatively standard ICU patients, does routine bronchoscopy actually improve safety? And from the answer to this study is essentially no. There was no statistically significant difference in perioperative complication rates between the two groups. So complication rates were 11% in the bronchoscopy group and 13% in the non-bronchoscopy group. So although there's more complications without the bronch, the difference was not significantly different when they ran it through some stat tests, meaning that difference could just be down to chance. Major complications were uncommon overall, just 1% of all the patients in both groups. The biggest complication was minor bleeding, which reflects what we see in routine practice. And so you could summarize that in experienced hands, routine bronchoscopy did not appear to reduce overall complication rates in these low-risk ICU tracheostomy candidates. What the study did find was a clear physiological effect from the bronchoscopy presence itself. So the bronch PDT group developed significantly higher peak airway pressures during the procedure. Now I don't think that's a particular problem because we actually turn our airway pressures up anyway for anyone that we're doing a tracheostomy on. So I wasn't too worried about that. But what they found was that there's significantly higher carbon dioxide levels afterwards, which led to lower pH, which, because CO2 is an acid gas, it drops your pH level or you get more acid blood effectively. And you can classify that physiologically as a mild respiratory acidosis, which obviously makes sense physiologically because with a tube with a bronchoscope in it, you are reducing effective ventilation, you're increasing resistance, and that can lead to that overall effect of reduced ventilation efficiency. However, in most patients, I don't think that's a problem. However, there are some patients where tight COT CO2 control does matter. For example, the classic would be a significant hypoxic or traumatic or whatever brain injury if patients got pulmonary hypertension or they've got significant respiratory instability. It's a lot more relevant because changes in CO2 can suddenly create big problems for either your intracranial pressure, for your pulmonary blood flow, or your pH if it's already on the back foot due to poor gas exchange. So what the paper says, I think it moves the conversation away from absolute. It's not saying bronchoscopy is bad, and the authors are very careful to say that bronchoscopy should never be used, that they're supporting a more selective risk-based approach. So they also comment that bronchoscopy is very valuable during the learning curve in patients with difficult anatomy, obesity, previous neck surgery, distorted airways, or where there's uncertainty about landmarks or chew positioning. But the study asks the question whether routine bronchoscopy is necessary for every standard PDT performed by experienced operators. And I think practically lots of ICUs around the world are already somewhere on that spectrum, depending on local expertise, local staffing and equipment availability. So the paper sort of goes on to discuss probably a wider issue in procedural medicine more broadly, that as operator experience increases, that balance between procedural assistance and procedural efficiency changes. So what might be essential for training or complex cases might become optional in these selective routine scenarios. So the study does have some limitations. Complications, very subtle ones like posterior wall injury, might have actually only been detectable with the bronchoscopy itself, and that means in the non-bronch group they might have missed some of those problems, which would be a bit more obvious if you got a camera down there. The study is unblinded, and obviously the secondary outcomes always carry some risk of statistical overinterpretation. But I think overall this is a pretty well balanced pragmatic trial with large sample sizes, multiple centres, and experienced operators. And I think the conclusion is very reasonable that in critically ill patients without major anatomical or physiological risk factors, routine bronchoscopic guidance does not reduce perioperative complications compared to doing the procedure without endoscopy. And that's down to those raised CO2 levels and the higher airway pressures. But I think the practical takeaway is that bronchoscopy is an extremely useful tool during PDT, but maybe not for every patient, every time. I have to say, I think it adds a degree of safety and I don't worry too much about the potential problems that the authors identified in this study because I think we can mitigate for them just by making sure you set the ventilator appropriately. And if the patient's got really borderline intracranial uh hypertension or or or they're brittle, we shouldn't really be doing a tracheostomy. So I took this with a pinch of salt. Uh it's probably going to be useful if you don't want to use a bronchoscope, but uh for me I I felt it just sort of added to the uh the general status quo where we are that I think bronchoscopy is is helpful and I'm going to carry on using it. Okay, next up in this cluster of PDT papers is a quality study again from our group in Manchester. This time it looks at staff experiences of PDT in ICU. So this includes challenging problems that clinicians face, the complications clinicians worry about the most, and whether new guidance technologies might help to improve safety. So this paper is very different from the others that we've discussed this month. It isn't a randomised trial or a device validation study. Instead, they wanted to understand how ICU staff actually experienced tracheostric procedures in the real world. So PDT is often described as a routine procedure, but anyone who performs them regularly knows that some cases are anything but routine. So the study involved interviews and focus groups with 32 multidisciplinary ICU staff across three UK intensive care units. That included consultants, trainees, nurses, advanced practitioners and support staff. And the discussions explored training, procedural safety, teamwork, complications, and reactions to a prototype needle guidance device designed to help track your ostomy insertion accuracy. So several really thought important themes I think emerged. The first was around training and skill development. So participants consistently describe PDT as a procedure that requires supervised, experiential learning. In other words, you don't really learn the procedure from textbooks, you learn by doing them with expert supervision. But staff also described the difficult balance between maintaining training opportunities and ensuring patient safety. And that tension will probably sound familiar to anyone who's been involved in some of these high-stakes procedural tasks. We all want less experienced colleagues to develop procedural competence, but difficult anatomy, obesity, time pressures, and fear of complications can naturally push operators towards more senior-led practice. The study also highlights how variable formal tracheostomy training still is between different units and different countries. Some departments had very highly structured teaching approaches, while others relied much more heavily on sort of opportunistic experiential learning. Another major theme to emerge was around human factors and teamwork. So participants repeatedly described PDT as a highly team-dependent procedure. Consultants usually led decision making, but nursing staff, operating department practitioners, respiratory therapists, and assistants all played key safety roles. Interestingly, the paper highlights how ICU tracheostomy differs from operating theatre workflows. In ICU, responsibilities are often redistributed, with nursing staff frequently acting as an additional layer of procedural oversight and situational awareness. And I thought that was a really important observation, mainly because tracheostomy safety is rarely just about the operator. It's about communication, planning, preparation, equipment setup, and shared team awareness. The study also explored complications and procedural anxiety. Difficult anatomy, especially obesity, short necks, previous surgery, or poorly palpable landmarks, were repeatedly identified as one of the biggest challenges during PDT. And importantly, staff described the actual needle puncture phase as one of the most uncertain parts of the procedure. Now that's interesting because despite all the advances in bronchoscopy, ultrasound, and airway equipment, we still don't really have dedicated tools that actively guide the needle through the neck tissues into the trachea, as we've discussed in the previous papers this month. A lot of PDT still relies very heavily on operator judgement, anatomical interpretation and tactile feedback. And that uncertainty is really what motivated the discussion around proposed guidance development to help guide the precise placements of the needle. Overall, staff were surprisingly positive about the idea of assistive guidance technology in helping them do the procedure. And participants felt it could potentially improve confidence, reduce uncertainty, and help difficult anatomical cases, and maybe even reduce referrals for surgical tracheostomy. There's also a really strong message about practicality. Staff emphasized that any new device would need to integrate smoothly into current ICU workflows, it had to remain simple to use and actually solve a genuine clinical problem. Because ICUs are full of examples of technologies that look promising initially but failed because they were cumbersome, complicated, difficult to clean or sterilise, or disrupted workflow. We've got a whole cupboard full of devices that don't have very good user interfaces as well, and so staff very quickly will make a judgment on a device based on what it looks like, how it feeds information back to you, and that's really important to whether you can actually implement and adopt new technologies. This paper gave some very useful examples of technologies that have been implemented, like ultrasound or video laryngoscopy for central line placement or airway management in sort of anesthetic and intensive care practice. And both of these technologies have become widely accepted because they actually improve safety without making the procedures excessively difficult. Another interesting finding was around safety governance. So most units used checklists, procedural briefings, and structured preparation systems, but participants also felt that incident reporting around tracheostomy complications was inconsistent and likely to be underreported. And that's important because without accurate reporting it becomes very difficult to truly understand the complication rates or identify opportunities for improvement. The study also highlighted variation in departmental culture. Some ICUs appeared more willing to accept procedural risks and bleeding risks in order to avoid delays associated with surgical tracheostomy referrals and access to theatre, which we've discussed before, can be problematic depending on where you work. Others were a bit more cautious and referred sort of difficult anatomical cases earlier to surgical colleagues. And again, I think this reflects real-world practice very accurately. So ultimately, tracheostomy decision making is often a balance between procedural risk, the risks of delay, resource availability, and your local setup, which includes the expertise and experience of the operators. Now there's clearly some limitations with this paper. This was a qualitative study, doesn't quantify complication rates or objectively prove that guidance improves outcomes. And as with any interview-based research, there's always a possibility of interviewer influence or selective participant sort of openness and what they want to discuss. But despite that, I think this is a useful paper. It explores something we don't really study very well in airway research, the human experience of performing tracheosomy procedures. The authors conclude that PDT practice is heavily shaped by training, teamwork, human factors, anatomy-related uncertainty, and local ICU culture, and staff appear very receptive to evidence-based assistive technologies that could improve procedural confidence and safety. So a bit like TrachePen we discussed earlier, even in highly technical procedures like PDT, safety is rarely determined by technology alone. It depends just as much on training, teamwork, communication, workflow design, and human factors. That brings us nicely onto what we're doing about it at the NTSP. If you look at the website, you'll see we've got two programmes running. The first is our immersive PDT course, specifically focused on the multidisciplinary team and the training required to do PDT safely in the intensive care unit. This course is aimed at senior residents or early career consultants, but also it's aimed at all the staff involved in the procedure, which is typically nursing staff if you're on the ICU. We run this in our immersive training suite at Lancaster University, and the separate podcast discussions about this if you have a quick look in the podcast listings. The second thing we are developing is that guided insertion for tracheostomy or gift system. Again, there's plenty of information on the website if you want more information, but this is essentially an electromagnetic guided system that can precisely guide the needle to the best place you need it for PDT insertion. We'll be starting our first in human trials in 2026 and we'll aim to keep you updated through the podcast channels. There's a lot of discussion about the design and the engineering and the background rationale to the GIF project in other pods, so check out the website or the podcast index if you want to find out a bit more. Right, we're going to stop talking about PDT and finally this month we're going to look at a paper which focuses on something we're hearing more and more about in healthcare research: machine learning prediction models. This time a model focused on predicting which cardiac surgery patients might eventually require tracheostomy. This paper comes from Felix Weissmüller at the University of Greifsfald in Germany. Now on paper, this sounds like exactly the sort of problem that artificial intelligence should be good at solving. After cardiac surgery, a proportion of patients require prolonged mechanical ventilation, and we know that earlier trachea inappropriate patients may reduce sedation exposure, facilitate weaning, shorten ICU, say, and potentially improve outcomes. So naturally, clinicians would love to know much earlier which patients are likely to require prolonged ventilation and therefore eventually tracheostomy. The challenge, of course, is that predicting tracheostomy accurately turns out to be much harder than it sounds. There have been plenty of studies that we've reviewed in the last 12 to 18 months examining the predictive power of retrospective datasets in trying to identify who will get a tracheostomy or not. So this German study retrospectively analysed nearly 4,800 cardiac surgery patients over a 10-year period, which included 118 patients who eventually underwent tracheostomy. The authors first evaluated an existing published prediction model, which had been developed by Wang and colleagues before trying to develop their own machine learning models using multiple different approaches. The Wang model performed pretty poorly when tested on this external German patient cohort. The area under the curve or AUC was 0.57, and that's only marginally better than chance. So what that means, AUC was developed by radar operators and an AUC of 0.5 essentially means that your model is no better than flipping a coin. So higher than 0.5 is slightly better, one is a perfect predictive model, and zero is the opposite. So 0.5 is pretty useless. The authors then try to develop several new machine learning models themselves, including things called random forests, neural networks, the nearest neighbour algorithms, and naive Bayesian models, which are all different statistical approaches to the same problem. They tested multiple different methods for selecting which variables that were routinely collected about these patients were the important predictive ones. Variables like pneumonia, sepsis, left ventricular function, the presence of mechanical devices like balloon pumps, how long there were bypass for, how old the patient was, and the type of bypass surgery all emerge as potentially important factors. But despite all that computational effort, none of those models achieve satisfactory predictive performance. I think that's one of the most interesting aspects of this paper because it highlights an important reality about AI in healthcare. The biggest limitation is not often the algorithm, it's the data that feeds the algorithm. The authors argue that retrospective ICU datasets may simply lack the depth, consistency, and standardisation needed to accurately predict something as complex as tracheostomy decision making. The decision to perform tracheostomy is not just physiological, it's influenced by clinician judgment, local culture, patient trajectory, sedation strategy, neurological recovery, family discussions, staffing, the weaning philosophy of the unit, and institutional practice patterns. So two very similar patients might receive completely different management decisions if they were managed in different ICUs. And we see that in our own institution where we've got different hospitals within our group and different ICUs within the same hospital that sometimes make different decisions. That variability is extreme and it's extremely difficult for retrospective models to capture. The paper also raises another really important point, I think, that tracheostomy prediction is not just a single time point problem. Most current models try to predict very early on using admission data or perioperative variables, but in reality the clinical picture evolves continuously over time, and most tracheostomies are performed at the sort of 12-14 day mark after that index admission. So patients develop infections, delirium, weakness, hemodynamic instability, renal failure, neurological complications, all manner of things that complicate their ICU state, which then influence the decision around tracheostomy. So the authors suggest that future systems might need to work on dynamic longitudinal models, so continually updating predictions as the new physiological and imaging data become available. I think that's probably where the field is heading. So rather than static risk scores, future AI systems might act like at sort of continuous learning clinical decision support tools. And the authors argue that incorporating advanced imaging analysis and physiological monitoring into these sort of multi-centre datasets is probably the way to go. And I think that's probably essential if these models are ever going to become genuinely clinically useful. One thing I appreciated about this paper is the authors were refreshingly honest about the negative findings. There's often a tendency in AI research to oversell this early modelling work, and we've discussed a few of these already on the podcast. But this paper essentially says: look, we've tried lots of multiple sophisticated approaches and they still don't work well enough. And that in itself is really valuable information because negative studies are just as important in helping define realistic expectations around AI in medicine. So this study does have some limitations. It was retrospective, single centre, and tracheosmenomas were relatively small compared to the overall cohort. But actually, many of these limitations are exactly the point that the authors are trying to make. Retrospective, routinely collected data in ICU might simply be insufficient for robust prediction models in this context. So the authors conclude, I think very reasonably, that neither existing clinical risk scores nor newly developed machine learning models currently provide reliable prediction of tracheostomy following cardiac surgery, and future progress is likely to require very large, prospective, multi-sender datasets with richer physiological and imaging information. I think the broader message here is really important. AI is a powerful tool, but it cannot magically compensate for inconsistent clinical decision making or poor quality data. In critical care, the complexity of human physiology and clinical judgment remains very difficult to reduce to a simple prediction score. So fortunately, clinicians are still worth their money. So that's it for this episode. We've cut a lot there, we've gone through tracheal aspirates, we've looked at PDT insertion, the power or lack of power for AI in predicting the need for PDT, and we've covered a few other things in between. Please follow us on our various platforms and social media and feel free to comment on the discussions. As ever, the views are mine and don't represent any of the organisations for which I work. So that was Tracky Talk for this month, edited and produced by Simon Williams. Thanks for listening and look forward to seeing you next time. Bye for now.