We often assume that diagnostic procedures will help patients. A lot of training goes into learning how to do these procedures. Procedures are dramatic. We like performing them. Patients are impressed, perceiving that we are “doing” something for them. Everything is awesome.
However, when strict evidence-based medicine is applied to procedures, they are often less impressive. One example in critical care is the Swan-Ganz catheter, which has generally failed to demonstrate meaningful improvements in patient outcome (despite valiant attempts to prove benefit). Likewise, the usefulness of bronchoscopy for pneumonia has been disappointing in patients without substantial immunocompromise.1
This post describes a systematic way of thinking about the benefit of any diagnostic procedure. This isn’t anything groundbreaking (experienced clinicians make these calculations implicitly). However, rendering this explicit may help us understand why diagnostic procedures often disappoint.
The chain-of-benefit for a diagnostic procedure
In order for a diagnostic procedure to benefit a patient, every single one of the above chain links must remain intact. If any link fails, the chain is broken and the procedure won’t help the patient.
To illustrate this chain, let’s walk through the links individually. We will use bronchoscopy and Swan-Ganz catheterization as illustrations, because these are classic examples of critical care procedures. However, these principles apply to any diagnostic procedure.
To begin with, there should be a reasonable likelihood that the patient has an illness which is potentially diagnosable by the procedure. For example:
- Bronchoscopy: This is useful for infections and selected inflammatory conditions (e.g. eosinophilic pneumonia). The yield is higher if these conditions are more likely, and if treatment hasn’t already been initiated.
- Swan-Ganz catheterization: How likely is it that the Swan will reveal a diagnosis which isn’t already clear? In some situations (e.g. excellent echocardiographic windows), there is little doubt regarding hemodynamics and therefore it’s unlikely that the Swan will yield any new information. In other situations, hemodynamics are murky so the potential yield is higher.
If a diagnostic procedure incurs severe harm, it’s unlikely to yield any net benefit. For example:
- In severe hypoxemia, bronchoscopy may run the risk of respiratory deterioration and intubation.
- Swan-Ganz catheter can precipitate ventricular tachycardia or heart block, leading to cardiac arrest. This isn’t purely academic – I’ve seen both complications.
The procedure must be technically successful and must yield accurate information. For example:
- Swan-Ganz catheterization: Accurate tracings and pressures must be obtained, with careful attention to placement of the catheter and proper zeroing.
- Bronchoscopy: Adequate specimens must be obtained and processed properly. Results must return promptly enough to guide care.
The results are interpreted correctly, yielding a correct diagnosis. For example:
- Swan-Ganz catheterization: Waveforms must be interpreted correctly and integrated with other hemodynamic and clinical data to reach an accurate diagnosis. Studies have shown that physicians were notoriously poor at interpreting this data, even in the heyday of the Swan.
- Bronchoscopy: Many organisms will often grow from a bronchoalveolar lavage – some are likely to be contaminants (e.g. candida), whereas others are more likely to be pathogenic. Judgement and integration with the clinical scenario is required to interpret bronchoscopic data correctly to reach a diagnosis.
The diagnosis must be treatable, and correct treatment must be utilized. For example:
- Swan-Ganz catheterization: The Swan will yield a lot of hemodynamic data. There may be an urge to try to “normalize” all the numbers, which may cause harm. Ideally, a sound therapeutic strategy should be based on reaching the correct diagnosis and applying evidence-based therapies for that specific diagnosis.
- Bronchoscopy: If the bronchoscopy yields an untreatable diagnosis (e.g. lymphangitic carcinomatosis), then this won’t lead to clinical improvement.
Furthermore, in order to be beneficial, the treatment should be superior to empiric management without any diagnostic procedure. For example:
- Swan-Ganz catheterization: A patient with heart failure may be likely to receive dobutamine and diuretics with or without a Swan, so the Swan may not lead to substantial change in therapy.
- Bronchoscopy: If a patient with pneumonia is diagnosed with a run-of-the-mill bacteria pathogen (e.g. Haemophilus influenzae), then this is unlikely to help the patient because that bacteria would have been covered adequately by empiric broad-spectrum therapy. However, if the bronchoscopy reveals a fungal pneumonia which wouldn’t be covered by standard empiric therapy, then treatment for the fungal pneumonia is likely to lead to meaningful clinical benefit.
Unfortunately, correct therapy may not always lead to clinical improvement. In order to gain true patient-oriented benefit, the therapy needs to be effective. For example:
- Swan-Ganz catheterization: If the Swan shows that the patient is in profound heart failure and correct treatment is initiated, the patient still needs to respond to this treatment to derive benefit. Unfortunately, end-stage disease may not necessarily respond to the best therapy – so there may be no clinical improvement despite having done everything right.
- Bronchoscopy: Unfortunately, diagnosing a severe infection via bronchoscopy is no guarantee that the patient will respond to antibiotics. In many cases, patients may succumb despite perfect diagnosis and ideal management.
The chain-of-benefit for a therapeutic procedure
It’s illustrative to compare this to a therapeutic procedure (that is, a procedure performed to treat an established diagnosis). As shown above, the chain-of-benefit is much shorter for a therapeutic procedure. This makes sense, because before a therapeutic procedure is initiated it’s already known that the patient has a treatable disease which may be alleviated by the procedure.
The distinction between a diagnostic procedure and a therapeutic procedure is an important one. Therapeutic procedures generally offer greater potential benefit, so they may warrant taking greater risk.
Diagnostic procedures: RCTs versus individual patient management
I asked the Twittersphere whether they knew of any solid evidence for benefit from diagnostic procedures. There didn’t really seem to be any (confirming my suspicion that RCT-level evidence is generally absent for diagnostic critical care procedures).
Given the numerous links in the chain-of-benefit for diagnostic procedures, it should come as little surprise that most RCTs of diagnostic studies show no benefit. For most patients, there is a break in the chain. For the average patient, the procedure yields little or no benefit.
How should we use evidence-based medicine to evaluate a diagnostic procedure? This is a tough question. There are roughly two schools of thought on this.
- Strict school of thought: Design RCTs which link a procedure to a patient-oriented endpoint. If the procedure cannot be shown to improve any patient-oriented endpoint, then it’s not useful.
- Lenient school of thought: A diagnostic procedure isn’t a therapeutic intervention, so it’s unfair to expect a diagnostic procedure to improve patient-oriented endpoints. The diagnostic procedure may therefore be evaluated using softer endpoints (e.g. change in diagnosis or change in therapy).
Ultimately, clinical judgement is required here. A procedure which is more invasive, riskier, and more expensive will require a stronger evidence basis to justify performing it (e.g. Swan-Ganz catheterization). Alternatively, a procedure which is safer and less invasive may be justifiable despite less robust evidence (e.g. arterial catheter).
Lack of RCT-proven benefit obviously doesn’t indicate that a diagnostic procedure won’t benefit any patient. For every individual patient, the risks and benefits of the procedure should be carefully weighed. In some situations, it’s quite obvious that patients do benefit from diagnostic procedures (e.g. the patient diagnosed with a fungal pneumonia). Thus, this post shouldn’t be misinterpreted to mean that we should never perform diagnostic procedures – but rather that we should perform these procedures thoughtfully and with brutally realistic recognition of their limited benefit for most patients.
The decline of ICU diagnostic procedures: Has imaging killed the procedural star?
There is another reason for the declining utility of diagnostic procedures: improved imaging and laboratory capabilities. In particular, ultrasonography and CT scanning are continually improving in terms of speed, accuracy, and safety (e.g. safer contrast dyes). Laboratory techniques to identify pathogens are continually evolving (e.g. using PCR). Meanwhile, basic ICU diagnostic procedures have made little progress over time (e.g. Swan-Ganz catheterization is probably less reliable now than in the 1990s, due to reduced procedural volume which decreases proficiency with the procedure).
The natural evolution of this process is that over time we will obtain more data from noninvasive studies, with less need to perform diagnostic procedures. And when we do perform diagnostic procedures, we will be armed with better information going into the procedure (thereby increasing the likelihood that the patient will genuinely benefit from the procedure).
Diagnostic procedures can be glorious, so this evolution makes me a bit sad. However, this is what progress looks like. So pour yourself a cup of coffee, look at the high-resolution CT scan, review the patient’s history, and figure out the right cocktail of antibiotics and labs. Don’t just do something – sit there.
- The infamous Bronchy Donky post by Scott Aberegg on his Status Iatrogenicus blog. This post is brutally honest, hilarious, and overall a must-read for anyone in critical care.
- Forebearance with Bronchoscopy: A review of gratuitous indications. Mehta AC et al, CHEST 2019 (electronically published last week).
- Why we fail at hemodynamics: The swan's curse (PulmCrit)
- Alice in Intensiveland (Easily one of the most epic articles ever written about hemodynamic monitoring, if you haven't read this stop what you're doing and read it immediately)
Image credit: Yellow snake
- IBCC chapter:Guide to APRV for COVID-19 - April 8, 2020
- PulmCrit Theoretical Post – The COVID Severity Index (CSI 1.0) - April 2, 2020
- PulmCrit wee – Why the SCCM/AARC/ASA/APSF/AACN/CHEST joint statement on split ventilators is wrong. - March 29, 2020