In recent years I’ve been doing a lot more prehospital care. To be honest I’ve found it very challenging, although also rewarding. In the prehospital setting we don’t have a lot of diagnostic tools and so are pretty reliant on the history and examination of our patients to decide what to suspect and what to do about it. This is because we simply don’t have the same capacity or kit in the prehospital service that we do in hospital. The primary survey really relies on the clinical skills of the clinician, and although we do now carry ultrasound to the scene, there is relatively little additional testing information available to us.
Similarly, in the ED I am often making quite difficult probabalistic decisions based on relatively simple clinical information and I know that I don’t get it right all the time. For example I’ve given blood to patients who subsequently probably did not need it, and I’ve been a bit late with the blood for patients who probably did! I suspect I am not alone in reflecting that it’s a tricky and uncertain world we practice in.
Is it just me though, or is it just tricky? Well this week we have a paper that might help us answer the question. Special thanks to Cliff Reid for spotting this one and posting it on twitter. The abstract is below, but as always please read the full paper yourself and make your own mind up.
What kind of paper is this?
This is a retrospective cohort study. Retrospective cohort studies are often useful in exploring hypotheses and looking for associations, but they do rely on data that was not collected for the intended purpose of the study. That does mean that they are subject to a number of biases that can affect the results. In this study the authors were reliant on whether clinicians recorded their predictions about which life threatening injuries were likely to be present.
Tell me about the patients.
The authors identified a cohort of trauma patients who had been admitted to a major trauma centre. The data was consecutive and only involved adult patients over roughly a two year period. The patients were all treated by a single air ambulance service (London Air Ambulance). There is a little confusion here as this was a single MTC, but not all major trauma patients arrive by that particular service, and indeed London HEMS transport patients to other trauma centres and trauma units. So the patient cohort is a subset of patients in this particular MTC and a subset of patients treated by this particular air ambulance service. These sort of restrictions are commonly found in retrospective cohort studies and they may introduce bias into the results.
Thermal injury patients and children were excluded.
What were they testing?
Essentially the study compares what the prehospital team thought the injuries were with the eventual diagnosis from hospital records. In this ambulance service clinicians are required to list what life threatening injuries are suspected so the data is at least collected contemporaneously and systematically. However, there were differences in how clinicians expressed their certainty with their assessment. The authors have tried to apply a scoring system based on adjectives used by clinicians e.g. likely vs. suspicious vs. possibly have different weightings. Whilst this seems sensible it is based on a US model and may not therefore work as well in a multinational cohort of clinicians who may use different language preferences.
The final diagnoses were gleaned from the hospital records. In the UK this is a pretty robust system as it feeds into hospital records and into the UK TARN database that is a well established system for determining the injury burden of patients.
Tell me about the analysis.
This is somewhat exploratory. The authors first sought to define what they thought were a range of ideal – unacceptable outcomes for sensitivity, specificity, NPV and PPV. They did this using a consensus methodology of expert clinicians, which is reasonable as I don’t think there is any externally validated measure in practice. However, it is does lack external validity and is also seemingly defined from within the service being examined in the study. It is likely that those developing the measures would have a-priori knowledge of clinician performance from past governance processes. That might introduce bias in where the bar is set (for good or bad).
The authors then performed univariate and multivariate analyses to look for factors that might be associated with missed diagnoses. This is a reasonable approach to explore the data further, although can only define association and not causation.
What did they find?
The authors looked at 947 patients. As with most major trauma studies the majority of patients were young and male.
Three main groups were looked at
- Life threatening injury
- Limb threatening injury
- Life threatening bleeding
The paper has a nice graphic shown below that shows that across a range of conditions the clinicians predictions were pretty poor to be honest. In terms of identifying life threatening injury (look at the sensitivity and PPV graphs) then performance is especially poor on this data.
The multivariate analyses showed that polytrauma, SBP, and expressed diagnostic uncertainty were linked to overprediction of life threatening injury.
The multivariate analyses showed that polytrauma, SBP, and heart rate were linked to overprediction of life threatening injury.
So should we stop clinically examining the patients?
Clearly not ,and that would not be the right message to take away from this paper. Clinical examination will always be a part of the initial assessment of patients, but this paper reinforces the fact that we cannot entirely rely on the clinical examination. It must be supplemented by more further and more definitive testing, which may of course include repeated clinical examination.
That said, I think this paper is hypothesis generating rather than definitive. Whilst I probably agree with the findings I think the detailed results are subject to a number of biases. The retrospective nature of the study, the definition of the standard being created from within the team analysed and the fact that this is a subset of service patients in a subset of the wider trauma centre will almost certainly influence the accuracy of the data, but it is unlikely to make such a difference that clinical examination would turn out to be highly accurate.
The authors conclusion that there is a need for further diagnostic adjuncts and decision support systems seems reasonable, although these too would need to be tested for their effectiveness in future studies.
The bottom line is that I won’t be throwing my stethoscope away just yet nor will I not bother to examine patients, but I will continue to be sceptical about my own and others clinical certainty about the presence of life threatening injury.
- Diagnostic accuracy of clinical examination to identify life- and limb-threatening injuries in trauma patients https://sjtrem.biomedcentral.com/articles/10.1186/s13049-023-01083-z