Gosh, you must be just sick about this IVC Crap
We have discussed it a bunch here on the podcast:
Then I debated Stone at Castlefest 2013
Then the French Canadians did an incredible podcast on the Ultrasound Podcast.
These folks were even brave enough to integrate Guyton curves and still keep it interesting.
A Review Article in Critical Care
- Bodson and Vieillard-Baron Critical Care 2012, 16:181
Mayo Reveals New Evidence is Pending
My friend Paul Mayo dropped an enticing hint of things to come at a recent sepsis meeting. He told us that a French gent is about to release a study showing IVC ultrasound does indeed assess fluid status quite well. More to come…
SMACC Back from Justin Bowra
SMACC Back #2 was on Justin Bowra’s excellent SMACC 2013 lecture on the myth of the use of IVC for fluid assessment. If you haven’t heard Justin’s talk or the SMACC Back, none of the below will make sense so go listen to both of them.
Thanks very much to Scott for his enjoyable and even painful ‘SMACC back’. Consider me SMACCed!
But seriously folks… I’m not sure his arguments make that much difference overall.
Here’s a summary of the issues so far:
- Scott says we agree on 90% of the topic, so in that context our differences (outlined below) are probably not that important.
- All you can really say is that IVC is probably only useful at extremes (fat and full versus flat and collapsing). I don’t think there’s any point in trying to be more accurate using this tool… pending the result of better studies (Scott’s suggestion, which I agree with).
- Scott believes that IVC collapsibility index (IVCCI) is a great test of fluid responsiveness. [Editor: Never said this. Said it is a great test of fluid tolerance, maybe a decent test for fluid responsiveness based on the evidence we have so far] However, the studies he quotes (by Lanspa and Muller) suggest that IVCCI is promising at best.
- Two other studies he quoted (by Blehar & Miller) are actually of patients with heart failure, who were sitting up at 45 degrees. It’s not a great idea to use them as surrogate evidence to support a 15% IVCCI cutoff in supine patients (a different population).
- He wrapped up by mentioning LVOT VTI assessment of cardiac output as a gold standard. This is a complex issue, which is worth discussing in depth (see below).
Let’s look at the issues one by one:
Q. In the spontaneously breathing patient, is IVC useful as a marker of fluid status, fluid responsiveness & fluid tolerance? Are there any numbers we can use?
My original SMACC answer was that based on the available evidence, the best one can say is that IVC is probably only useful at extremes (fat and full versus flat and collapsing). The best numbers I could find were:
- Maximum IVC diameter (IVCD) <0.9cm = emptyish. IVCD >2cm maybe fullish.
- IVC collapsibility index (IVCCI) >40% = emptyish (72% of the time). IVCCI <15% = fullish.
- Serial IVC measurements seem useful.
Scott’s SMACC back went as follows (I’ll put his comments in italics):
‘A quick look at size and collapsibility gives huge amounts of information’
JB: I AGREE. As I noted in my talk, IVC assessment helps with the diagnosis of obstructive shock, and helps at the extremes. But that’s all.
Scott then noted that the receiver ROC curves in the Lanspa & Muller studies both had points on which the IVCCI didn’t miss any fluid responders.
JB: I DISAGREE. There are 2 studies here: one by Lanspa et al, and the other by Muller et al. Both looked at IVCCI as a marker of fluid responsiveness. Both used LVOT VTI [obtained in the A5C view] as a gold standard. Both found similar results.
Yet the authors reached completely opposite conclusions.
- Lanspa et al (14 patients) asserted a ‘good diagnostic accuracy’ for IVCCI, based on an area under the curve [AUC] of 0.83 (95% confidence interval, 0.58 – 1.00).
- Muller et al (40 patients) asserted the opposite, based on an AUC for IVCCI of 0.77 (95% CI 0.60 – 0.88)
- Scott says that AUC is the wrong thing to look at, and in fact each curve had a point (about 20-30% IVCCI) where no fluid responders were missed.
So who’s correct?
Well, it’s tricky. I’m not too experienced with ROC, so I asked my friendly hospital statistician and she replied ‘The confidence interval on the (Lanspa) study is too wide to be drawing definite conclusions.’
I’ve also been doing some reading of my own. See my analysis at the end of this document for the details if you’re into that sort of thing, but the summary is:
- It’s true that Lanspa’s AUC of 0.83 is pretty good.
- However, with only 14 patients, Lanspa’s results include a very wide 95% confidence interval (CI) of 0.58 – 1.00. This means that the test may have an AUC as low as 0.58- and there’s also a 5% chance that the true AUC may be even lower.
- Muller’s larger study (40 patients) achieved an AUC for IVCCI was 0.77 (95% CI 0.60, 0.88). Recognizing that the true AUC may have been as low as 0.60, the authors concluded that IVCCI ‘cannot reliably predict fluid responsiveness in spontaneously breathing patients with ACF.’
- Therefore the Lanspa and Muller studies suggest that IVCCI is promising at best, but needs larger studies to determine its accuracy. To quote Lanspa: It appears that either very large or very small values of VCCI may have some utility in predicting response to VE, although VCCI may have a wide range where it is clinically indeterminate in the spontaneously breathing patient.
So let’s get back to Scott’s SMACC Back.
Scott states that ‘IVCCI 15% had good accuracy (92% sens/84% spec) for CHF (Blehar et al. The American Journal of Emergency Medicine 2009;27(1):71)… and similar results in the 2012 Miller study.’
JB: I DISAGREE. Although it’s true that both these studies had good accuracy for the diagnosis of acute decompensated heart failure, it’s quite another thing to say that they can be applied to supine shocked patients. Both these studies measured IVCCI in breathless patients, to answer the Q ‘CCF or not?’ (Which certainly is not a bad surrogate for fluid tolerance.)
HOWEVER, Blehar 2009 measured with the patients up at 45 degrees. If lying flat, it’s likely they would have had a lower IVCCI, so the 15% cutoff in this study probably doesn’t apply to supine patients. Put simply, they are two different populations: lying down very probably makes the IVC ‘fill up’.
Same goes for Miller 2012: this study measured patients at 30 degrees. Using an IVCCI index of15% as a cutoff yielded a 37% sensitivity (ie you’ll only ID 37% of the Ps who are probably fluid intolerant, or in fact 22-55% when you look at the 95% CI) but 96% specificity (the ones you ID had better not get any fluid loading from you).
To me, this means that:
- 15% cutoff only works if sitting up- this figure likely lower in supine
- Even in those sitting up, Miller’s study suggests that this cutoff will incorrectly reassure you that a fluid challenge is ‘safe’ in 63% of patients in whom it’s likely to be dangerous.
Scott then said: ‘Go for gold: look for a 15% increase in SVi with a REAL cardiac output monitor or skilled evaluation of LV VTI.’
JB: LVOT VTI estimation by US can be bloody tricky. It’s worth digging into this issue, because LVOT VTI has been used as the gold standard in a number of the above studies, and is increasingly popular in critical care.
So… Just how valid is LVOT VTI (using TTE in spontaneously ventilated patients) as a marker of cardiac output?
Here’s my summary:
- LVOT VTI is fiddly, time consuming & inaccurate except in the hands of experts.
- There are also plenty of patients you can’t really use it on: eg those with AF, valvular disease or marked respiratory distress.
- It’s probably too inaccurate to use as a standalone measure of absolute cardiac output.
- But it is probably OK if you limit its use to what Scott describes: when you’re looking for a change in cardiac output after an intervention (e.g. a fluid bolus).
See below (A LOOK AT LVOT VTI) for details.
Final question: Just how accurate do I need to be when I measure the IVC?
This is important. Do I need to measure the IVC down to the nearest millimetre, or can I be 10-20% out?
My gut says that it’s OK to be 10-20% out because probably none of these measurements (not even LVOT VTI) are any better than that. This is one reason that I prefer a ‘Goldilocks Algorithm’ of ‘obviously full, obviously empty, or somewhere in between’.
BUT if we are talking about applying ‘hard’ numbers such as a cutoff of 15% IVC collapse for ‘wet versus dry’, then it becomes very important indeed. If I measure IVCCI as 10% but someone else measures it as 20% in the same patient, then we might draw very different conclusions about the same patient.
The problem is that until all the big studies are done, no-one knows the answer to this question. IVCCI might well be accurate enough to use hard numbers… but it might not.
In the spontaneously breathing patient:
Q.Is IVC useful? Undoubtedly. It helps with the diagnosis of obstructive shock and fluid status, taken as part of the overall clinical context.
- What IVC numbers predict ‘empty patient’?
- IVCD <0.9cm, probably.
- IVC collapsibility index >40%, probably.
Q. What IVC numbers predict ‘full patient’?
- First make sure you’ve excluded all the other stuff that causes a large IVC (obstructive shock, right heart disease). That leaves…
- IVCD >2cm probably = ‘full’ (Certainly 2.5cm predicts it in mechanically ventilated patients, and they have bigger IVCs overall.)
- IVC collapsibility index for ‘fullness’ is tricky.
- <35% is suggested by some, but as a predictor of high RA pressure (by Brennan) rather than ‘won’t tolerate fluids’
- Scott’s quoted studies of seated patients use 15% as a cutoff for ‘fullish’, which suggests that <10% = probably fullish if supine
- In other words, IVC is probably better at predicting ‘empty’ than ‘full’ (there are too many other conditions that lead to a big IVC)
- None of these numbers have been validated in good studies of supine patients. For example, some come from patients sitting up.
- Serial IVC measurements seem useful.
- Go for gold by all means… but for most of us (weekend warriors and part-time cardiac sonologists who certainly don’t use LVOT VTI every day), ‘gold’ is clinical assessment plus whatever tests you have to hand, including bedside US.
…But as they say, that’s just one man’s opinion. What do others think?
Cheers from Justin Bowra in Sunny Sydney
NOW FOR THE ANALYSIS OF THE ROC CURVES IN THE LANSPA & MULLER STUDIES
I had a big long think about these two studies and why they seemed to come up with opposing conclusions despite similar results. Using the references above and with sage advice from Jill the hospital statistician, here’s what I’ve come up with:
As Scott pointed out in his SMACC Back, it looks like there’s a spot on Lanspa’s curve where not a single fluid responder was missed. And it’s also true that Lanspa’s AUC of 0.83 is pretty good. Roughly speaking, looking at the AUC is a way of assessing how good a test is at predicting the outcome of a binary variable. A handy guideline is the following:
- AUC 0.5 = no better than a random guess (flipping a coin)
- 0.6– 0.7 = not great
- 0.7-0.8 = fair (still not that good)
- 0.8-0.9 = good
- >0.9 = excellent
- 1.00 = perfect
Therefore a result of 0.83 is not bad at all.
However, with only 14 patients (which is reflected in the shape of their ROC) Lanspa’s results include a very wide 95% confidence interval (CI) of 0.58 – 1.00. This means that the test may not be that good at all- in fact it may be have an AUC as low as 0.58- and there’s also a 5% chance that the true AUC may be even lower. It also means that the numbers simply aren’t high enough to draw any firm conclusions. Lanspa didn’t miss any fluid responders at a 20-30% IVCCI cutoff… but that might’ve been just a chance finding after all.
Muller’s study was larger, with 40 patients, reflected in its narrower CI and smoother ROC. Their AUC for IVCCI was 0.77 (95% CI 0.60, 0.88). Recognizing that 0.77 is fair rather than great, and that there’s a 95% chance the true AUC was as low as 0.60, the authors concluded that IVCCI ‘cannot reliably predict fluid responsiveness in spontaneously breathing patients with ACF.’
Similarly, Muller’s ROC curve missed a couple more fluid responders than Lanspa at the 20-30% IVCCI cutoff, and this might be because IVCCI just isn’t as good as we’d like it to be, and the higher numbers in this study made this more apparent.
So how many numbers do we need before we can say with confidence that an IVCCI cutoff of 20-30% will pick up all the fluid responders? Well, I don’t know. But we probably need more than the numbers in these two studies.
I’ve also come across a couple of sites that question the utility of ROC analysis, but that level of analysis is beyond me.
Therefore the Lanspa and Muller studies suggest that IVCCI is promising at best, but needs larger studies to determine its accuracy. To quote Lanspa: It appears that either very large or very small values of VCCI may have some utility in predicting response to VE, although VCCI may have a wide range where it is clinically indeterminate in the spontaneously breathing patient.
…but we knew that already.
NOW A LOOK AT LVOT VTI:
- My 1st point here is a mea culpa: I’m no expert. I’ve done a search (helped greatly by ace cardiac sonographer Sharon Kay) and I’ve picked the brains of Drs Chris Choong, Jason Sharp and Greg Nelson (cardiologists at my hospitals who perform LVOT VTI regularly) and Dr Adrian Goudie FACEM DDU (who certainly does more cardiac US, and at a higher level, than yours truly).
- But I may have missed something important. So please let me know if that’s the case.
- LVOT VTI is used as a noninvasive surrogate marker for cardiac output. It looks like a great idea: using a view such as the PLAX, measure the LVOT diameter at the AV annulus immediately proximal to valve leaflets (the best site seems to differ among the studies and guidelines). Use this diameter to derive the cross-sectional area at that spot (this assumes the annulus is round). Then using A5C view, measure the flow through that same spot. Derive the velocity time integral (VTI) and multiply by the cross-sectional area and you’ll get the stroke volume (SV) (i.e. how far that column of blood travelled through that LVOT). Multiply the SV by the heart rate and you’ll get the cardiac output. Nice.
- There are potential problems with this technique, however. To summarise what I’ve found from the literature:
- The first problem is that the technique may not be reliable enough as a standalone measure of cardiac output. Coats’ review suggests that this technique ‘has a 95% chance of lying within ± 28% or or 1 -41 min~’ of the standard.’ Note that this is not the same thing as simply looking for a change in CO- see below.
- The next problem is the exclusions: arrhythmia eg AF, aortic valve disease or prosthetic aortic valve. Huntsman et al estimated that Doppler estimation of CO could be performed in only 85% of patients in their study. Coats also listed COPD patients as difficult.
- It’s fiddly and time-consuming in practice: in PLAX, Lewis et al averaged AV diameter measured across 5-10 cardiac cycles; in A5C, Lewis averaged VTI across average of 5-10 cycles. Bouchard averaged across 3 cycles in NSR patients, and across 10 cycles if ‘rhythm or haemodynamic disturbance’. Sharon Kay tells me you need to average across at least 3 beats.
- Inter-observer variability can be a issue:
- i. In Lewis et al’s study, this wasn’t so bad: only 6.8% +5%
- ii. In Bouchard’s study, slightly worse: 9-11%
- Patient position: patient needs to be in left lateral position- Lewis- or in ‘steep left lateral’- Bouchard. But also the patient position should remain constant, to make it easier to obtain the same angle of probe position / insonation. This isn’t so hard in ventilated patients in ICU, but a good deal harder in the critically ill patient who’s spontaneously breathing.
- Patient respiratory variation can also be an issue: eg a large respiratory effort (such as in our unstable patients pre ventilation) can also alter beat-to-beat measurements, both real (by altering the actual stroke volume) and by altering angles. (Thanks to Adrian Goudie for pointing this out.)
- However, there are advantages as well:
- Repeating the same measurement, at the same site, in the same patient (who’s in the same position) from the same window should eliminate variables due to measurement of LVOT area and hopefully angle of insonation
- If the same operator performs pre- and post-scans, this should also eliminate inter-observer variability.
- If the two measurements are performed just before and after an intervention such as a fluid bolus, then this should also improve accuracy.
- Coats’ review article states that LVOT VTI assessment from the A5C window ‘probably’ has a 90% chance of picking up a 7% difference in CO… therefore it should be even more likely to pick up a 15% change.
- Here are the validation studies I’ve been able to find:
- Coats AJ. (1990) According to this review article, Doppler methods of cardiac output estimation show accuracies varying from 10 to 22%. The A5C estimation of LVOT VTI is the best, but even that ‘has a 95% chance of lying within ± 28% or or 1 -41 min~’ of the standard.’
- Huntsman (1983): used CW Doppler at the suprasternal notch compared to thermodilution, rather than A5C. So it may not be relevant, unless plenty of people are using suprasternal window instead of A5C.
- Ihlen H (1984): This study showed very good correlation between Doppler VTI estimation of CO and invasive estimation of CO. However, it looks like they obtained Doppler measurements from suprasternal window, not apical. So I’m not sure if this study applies to the technique of using the A5C window.
- Lewis et al (1984): a great study: the gold standard was thermodilution performed within minutes of the TTE. Good result with interobserver variability only 6.8% +5%
- Bouchard (1987): another good result using thermodilution. However, used CW Doppler, not PW Doppler. Also, used 3 different locations including suprasternal notch. And in 28 of the 43 patients, the gold standard was performed within 24h.
- There are plenty of papers on the USCOM device, but I don’t think this is being used much so I didn’t include them.
- From my reading, it looks like LVOT VTI is fiddly, time consuming & requires skills beyond the reach of non-experts. It doesn’t look like it has the accuracy to measure absolute cardiac output.
- But that’s quite a different question from asking if it can be used to measure a 15% change in cardiac output. If the same operator performs the measurement at the same site and angle in the same patient who’s lying in the same position, and if that operator is an expert and not just a weekend warrior, then yes.
NOW FOR THE REFERENCES (a couple of others weren’t listed here but appear at the end of my SMACC talk)
- Lanspa MJ, Grissom CK et al. Applying dynamic parameters to predict hemodynamic response to volume expansion in spontaneously breathing patients with septic shock. Shock 2013. 39(2). pp. 155-160
- Muller L et al. Respiratory variations of inferior vena cava diameter to predict fluid responsiveness in spontaneously breathing patients with acute circulatory failure: need for a cautious use. Critical Care 2012, 16:R188
- Blehar DJ, Dickman E, Gaspari R. Identification of congestive heart failure via respiratory variation of inferior vena cava. Am J Em Med 2009;27:71–5.
- Miller et al. Inferior vena cava assessment in the bedside diagnosis of acute heart failure. Am J Emerg Med 2012;30:778-83
- Coats AJ. Doppler ultrasonic measurement of cardiac output: reproducibility and validation. Eur Heart J (1990) 11 (suppl I): 49-61
- Huntsman LL, Stewart DK, Barnes SR, Franklin SB, Colocousis JS, Hessel EA. Noninvasive Doppler determination of cardiac output in man: clinical validation. Circulation 1983;67:593-602.
- Ihlen H et al. Determination of cardiac output by Doppler echocardiography. Br Heart J 1984; 51: 54-60
- Lewis JF, Kuo LC, Nelson JG, Limacher MC, Quiñones MA. Pulsed Doppler echocardiographic determination of stroke volume and cardiac output: clinical validation of two new methods using the apical window. Circulation 1984;70:425-31.
- Bouchard A et al. Measurement of left ventricular stroke volume using continuous wave Doppler echocardiography of the ascending aorta and M-mode echocardiography of the aortic valve. JACC 1987;9:75-83