How Good is Gestalt?

If you are a listener to the Journal Club podcast with Justin, then you will know that we talk a lot about decision-rules and diagnostic tests. The subject of “clinician gestalt” comes up a lot. Although the original meaning of gestalt has been lost – we all know what it is that we are talking about…

We all think we have some degree of “spider sense” when it comes to picking the big, bad diseases in the Emergency Department. But do we really? Can we actually rely on that little voice inside our heads when it comes to making the right calls? Which patient should we run through the CT-of-truth? Can we quantify the power of our gut feelings? Fascinating questions!

So lets do a review of a paper that attempts to answer these questions. Specifically, the question of clinical gestalt for the diagnosis of pulmonary embolism. Here is the paper:

Accuracy of physicians’ intuitive risk estimation in the diagnostic management of pulmonary embolism: an individual patient data meta-analysis. by Dr Roseanne van Maanen et al published in the Journal of Thrombosis and Hemostasis, June 2023. ( https://doi.org/10.1016/j.jtha.2023.05.023 )

As the title suggests – this is a meta-analysis of individual level patient data. The authors have looked at nearly 4000 individual trials and come down to just 16 studies that provided the individual patient data and clinical “gestalt” information. Overall this represents 20,770 unique patients suspected of having a possible PE. The patients were not just Emergency patients – they also included primary care, self-referred, nursing home and ward inpatients.

The trials had to include either an objective, positive diagnosis of pulmonary embolism or at least 30 days of PE-free follow up to count as a ‘negative’.

It is important to note that the majority of these studies were completed in Western Europe and two studies from Canada – there were no Australasian or U.S. trials in the mix. This is important as we know that the pre-test probability and diagnostic yield of PE work up does vary widely across the globe.

The clinicians’ gestalt (intuitive risk estimation) was dichotomised by the question: ” Is PE the most likely diagnosis?” prior to diagnostic testing. For example, if the Wells’s score was applied – then the gestalt was deemed positive if the “PE most likely diagnosis” was answered “yes” and the +3 points given. Different studies used other scoring systems to ‘document’ clinicians’ gestalt eg. Geneva. This diagnostic impression was recorded prior to D-dimer testing.

So in summary – we have 20,000+ patients with symptoms that could be a PE, we have the clinician’s best guess as to whether PE was the ‘most likely diagnosis’ and we have the actual diagnosis post work-up or follow up period. Yes, this is a big data dredge that is full of potential biases, errors and all the usual diagnostic (radiological) uncertainty that plagues PE research. However, it is a big data set and worth a look under the bonnet to try and quantify our brain’s best bets!

So what did they find? How potent is gestalt in the diagnosis of pulmonary embolism?

Bottom line: – the overall rate of pulmonary embolism in this data set was 20 % [range 7.4% to 40.9% ]

In the patients in which the clinicians had a POSITIVE Gestalt the rate of PE was 29% [22.1% to 62.1%]

In the Gestalt NEGATIVE group the rate of PE was 9%.

So if you do the math on relative risk it turns out that overall clinical gestalt roughly triples the chance that a patient will have a PE.

Strangely, the D-dimer levels were doubled (on average) in cases where the gestalt was deemed positive… although only a true cynic would attribute any causality to this observation… we all have days when it seems to be that way! ^{{not science}}

If you look at the individual trials then the prize for most accurate gestalt / spider sense must go to the Swiss. The two papers by Perrier et al (2004, 2005) were closest to the pin when it came to picking PEs but not overcalling it! The majority of trials showed a that “PE most likely” was 2-4 times higher than the actual rule-in rate for PE. Only a single trial from Spain (Galipienzo et al, 2012) went the other way… their gestalt under-called the PE rate by about 4%! I am not really sure if this says more about the patient group (secondary referral) or the clinical mindset in Spain?

Probably the most interesting data to be dredged up in this paper is the remarkable consistency of the effect of gestalt over a number of countries, time, clinical settings and patient sub-groups. Kabrhel et al published in 2005 a 583 patient trial that suggested gestalt-accuracy increased slightly with clinical experience. This new paper would refute that effect, although the actual experience levels of the clinicians are widely implied in this current study.

One caveat to all of this was the effect of patient age on the clinical gestalt. The confidence intervals on the relative-risk get very wide, almost exponentially so, after the age of 60. This is probably a result of smaller numbers and therefore less data points… but it is good to know. Maybe our gestalt does get watered down once we start guesstimating in the presence of lots of co-morbid diseases and poly-pharmacy etc?

For me this is really interesting stuff. Fascinating research that tries to answer a question that is really tough tot tease out of the data. Of course, this is not going to change your practice… because we cannot change our gut-feelings and if you believe the authors it may not even be something that you get better at with time and experience.

Love to hear your thoughts – comments below or via the socials.

Casey