Deep Stops Increases DCS

[leadduck]

Please see the heat map below. It provides a comparison of the supersaturation patterns of 4 profiles -- the 2 NEDU study profiles (A2 and A1) and two models we're all more familiar with, VPM-B and GF. Both VPM-B and GF were dialed in to have the same runtime (209 minutes).

(.....)
The heat map shows visually what Dr. Doolette's "deep stop skew" chart showed conceptually. A2 and VPM skew stops deeper and GF and A1 skew stops shallower. The impact of this skew is seen in the integral supersaturation (ISS) chart below. ISS is an index of time exposure to supersaturation. The only credible theory for the higher DCS rate in A2 is that it's higher integral supersaturation (i.e. diver's exposure time to supersaturation) caused the additional DCS risk. For reference, VPM's ISS is 95% of A2's, GF's is 84% and A1's is 80%.

The nice thing about these charts is WE KNOW the risk of A2 (~5%) and A1 (~1.6%). And you can see the relative similarities of VPM and GF. Many of us have concluded based on these NEDU results that VPM's mechanism of allocating stop time simply allocates too much time deep. And GF's algorithm more naturally skews time in better proportions to shallower depths and, therefore, better reflects the NEDU study.

As much as I like the heat maps and ISS concept, I see how difficult it may be for some to grasp the whole thing. The argument of similarity in ISS will always be countered that the actual profiles (fsw,min.) do not "look" similar whatever that means. That's why I brought up the table above, to let the models (VPM, ZHL+GF, ...) decide if a profile is good or bad. In this way we can correlate each model response directly with the experimental result and need not argue about a metric for similarity of different profiles.

Would you mind checking with your VPM software, whether A2 and A1, when entered as multilevel dives, hit the ceiling? Is there any conservatism setting (ICR) that get's the right answer (A2 hits the ceiling but A1 doesn't)?

 

[rossh]

I don't know why you reply this to my post. I haven't talked about "deep stops". I don't like these terms with no clear definition. There are no "deep stop models" or "shallow stop models". The meaning of "deep stop" is vague (unless we talk about Pyle stops).

The profiles of the NEDU study are indeed different from the optimal profiles of a VPM or ZHL+GF model. So what. Let's just call them A1 and A2 and move on. The more interesting question is, how do A1 and A2 perform when evaluated by ZHL, VPM, RGBM, ...?

.

Sorry, I was getting a bit excited, with all the fakery going on.

"The more interesting question is, how do A1 and A2 perform when evaluated by ZHL, VPM, RGBM, ...?"


As I showed above (#600), its impossible to make VPM produce an A1 or A2 profile. It cannot be done. GF will get a bit closer, but still a long way from a replication.

The reasons are, the A1 is a gross exaggeration of actual required deco . Then A2 is a purely hand crafted profile that has non-sequential stop times. No finished deco model can ever make the same plans. The reason it was done this way, was because they wanted to artificially force the end run times to the same.

So the question, cannot be answered in a simple way. Hence all the convoluted comparison methods.

I use supersaturation charts, as it is independent of all models, and reveals what the model is thinking and its attributes - there is no better insight method than this.

In the VPM-B Fortran code however, I don't see temperature considered, and I'm not aware that anybody ever ran controlled experiments to see how VPM profiles need to be adjusted for cold.

No, nothing like that. It is impossible to calibrate this. Staying longer, getting colder, just makes the problem worse, with diminishing returns. How would the diver ever know how to judge this level of cold? There is only one solution - avoid getting cold.

 

[leadduck]

Hi, yes I found it in the meantime. It show that the so called "Deep stop" profile from NEDU trial is very different from profiles generated with normal bubble models, such is RGBM. Sure there is much more in that paper and not in favor of the NEDU trial.

Sure, the authors felt they need to defend RGBM here. I referred to this paper because they confirm that the "deep stop" profile A2 is worse than A1 when evaluated in RGBM for DCS risk. They do not claim that A1 is better or equal than A2 if it weren't for thermal stress, or that thermal stress distorted the results. Their line of argumentation is that RGBM would have predicted the NEDU study results (A2 worse than A1), and hence the results do not invalidate RGBM. Their Table 6 brought me to the idea to apply the same concept to ZHL+GF and VPM, to see which model and which setting would have ordered A2 and A1 correctly, not needing any concepts for similarity of profiles.

 

[leadduck]

As I showed above (#600), its impossible to make VPM produce an A1 or A2 profile. It cannot be done. GF will get a bit closer, but still a long way from a replication.

Agreed, I also think there's no setting of conservatism in VPM that makes it produce A1 or A2 as the optimal deco profile. But reading the optimal profile from a model is not the only way to use a model. Think about using a dive computer. You can follow the ceiling exactly, but don't have to. As long as you stay under the ceiling, the model says you're fine. Following the ceiling is just the fastest way to get out, but not the only way.

Some models allow to calculate DCS risk as a function of a profile entered. I believe that's what Bruce Wienke does in the paper above in Table 6.

So I'm not looking for VPM parameters that produces A1 or A2 as the optimal profile ; I know it doesn't exist. I'm looking for a model and setting for which A2 hits the ceiling, but A1 doesn't. This setting would've been the right choice for the NEDU test subjects to use on their dive computers. Seems like ZHL-16B + GF 90/90 fulfills this requirement, but what about VPM?

 

[igor p]

As much as I like the heat maps and ISS concept, I see how difficult it may be for some to grasp the whole thing. The argument of similarity in ISS will always be countered that the actual profiles (fsw,min.) do not "look" similar whatever that means. That's why I brought up the table above, to let the models (VPM, ZHL+GF, ...) decide if a profile is good or bad. In this way we can correlate each model response directly with the experimental result and need not argue about a metric for similarity of different profiles.

Would you mind checking with your VPM software, whether A2 and A1, when entered as multilevel dives, hit the ceiling? Is there any conservatism setting (ICR) that get's the right answer (A2 hits the ceiling but A1 doesn't)?

A1 and A2 hit ceiling under all settings of VPM. Just checked with multideco. VPM asks for stops before first stops of both A1 and A2 profiles. So if I understood your tesis correct it rejects both. How does Subsurface implement VPM do not know. But know original fortran VPM-B program rejects A1 and A2 sure with properly implemented multilevel - original program does not check for hitting ceiling within multilevel.

 

[rossh]

Agreed, I also think there's no setting of conservatism in VPM that makes it produce A1 or A2 as the optimal deco profile. But reading the optimal profile from a model is not the only way to use a model. Think about using a dive computer. You can follow the ceiling exactly, but don't have to. As long as you stay under the ceiling, the model says you're fine. Following the ceiling is just the fastest way to get out, but not the only way.

Some models allow to calculate DCS risk as a function of a profile entered. I believe that's what Bruce Wienke does in the paper above in Table 6.

So I'm not looking for VPM parameters that produces A1 or A2 as the optimal profile ; I know it doesn't exist. I'm looking for a model and setting for which A2 hits the ceiling, but A1 doesn't. This setting would've been the right choice for the NEDU test subjects to use on their dive computers. Seems like ZHL-16B + GF 90/90 fulfills this requirement, but what about VPM?

The dive computer analogy doesn't really hold there. A good one (like mine) is continuously recomputing the ascent home, using its existing tissue state. The computer will not know or care how straight or wobbly the path has been to reach this current point.


As for forcing the model to comply: Well, you can pound the model with a hammer so its a square peg in a round hole. But that becomes garbage in - garbage out. You cannot tell the model how to make deco - it doesn't care for your version of stops and levels. It only knows how to make deco on its own terms.

But if you insists, the first stop has to be a -10 (cr0.25) which is a violation of 12m/40ft, so its already broken the process. Next stop is a +30 (cr3.0++)..... makes no sense, and demonstrates clearly that VPM-B cannot make the A2 profile. The inverse is true also - the A2 does not represent a VPM-B.


No profile will ever match the A2. It does not have a natural off gas pattern.

.

 

[igor p]

Sure, the authors felt they need to defend RGBM here. I referred to this paper because they confirm that the "deep stop" profile A2 is worse than A1 when evaluated in RGBM for DCS risk. They do not claim that A1 is better or equal than A2 if it weren't for thermal stress, or that thermal stress distorted the results. Their line of argumentation is that RGBM would have predicted the NEDU study results (A2 worse than A1), and hence the results do not invalidate RGBM. Their Table 6 brought me to the idea to apply the same concept to ZHL+GF and VPM, to see which model and which setting would have ordered A2 and A1 correctly, not needing any concepts for similarity of profiles.

Read the corelation paper I posted and I think you will find answer. Not direct but can be extracted from material.

 

[UWSojourner]

As much as I like the heat maps and ISS concept, I see how difficult it may be for some to grasp the whole thing. The argument of similarity in ISS will always be countered that the actual profiles (fsw,min.) do not "look" similar whatever that means. That's why I brought up the table above, to let the models (VPM, ZHL+GF, ...) decide if a profile is good or bad. In this way we can correlate each model response directly with the experimental result and need not argue about a metric for similarity of different profiles.

Would you mind checking with your VPM software, whether A2 and A1, when entered as multilevel dives, hit the ceiling? Is there any conservatism setting (ICR) that get's the right answer (A2 hits the ceiling but A1 doesn't)?

I guess I'm not really needing to do further analysis at this point. The deep stop debate convinced me, and surprised me, of the strong applicability of the NEDU study. My conclusion is that based on the best evidence we have currently, the de-emphasis of deep stops is a prudent step. But I can see if you're still on the fence that you might want to pursue other metrics. Good luck.

 

[UWSojourner]

B.R.Wienke and T.R. O'Leary: "Recent Deep Stop Data and Tests". Deep Stops and Decompression Workshop, Salt Lake City, June 2009
https://www.researchgate.net/profil...cf265693cef0185.pdf?origin=publication_detail
See Table 6.

The heat map below shows the supersaturation patterns for A2 (DCS~5%), LANL (from table 6), VPM-B, and A1 (DCS~1.6%). The LANL profile in Table 6 didn't seem to match the 209min run time so I padded the 10ft stop out to 209min.

In light of the fact that we know A2 has ~5% risk and A1 has ~1.6% risk, I don't see how you'd expect the LANL profile to be substantially better than A2.

The chart below shows the integral supersaturation by compartment for the 4 profiles.

LANL's was aggressive at protection of the fast tissues. You can see that the supersaturation exposure in compartments 9 and above map very closely for A2 and LANL. That would make it hard for me to think the LANL profile would perform substantially different than A2.

Finally, the total supersaturation exposure is shown below.

Absent some sort of solid dive trials (i.e. a closely controlled trial like the NEDU study), I look at those charts and wouldn't expect LANL to somehow perform better than A1 and really would expect something closer to A2.

 

[igor p]

The heat map below shows the supersaturation patterns for A2 (DCS~5%), LANL (from table 6), VPM-B, and A1 (DCS~1.6%). The LANL profile in Table 6 didn't seem to match the 209min run time so I padded the 10ft stop out to 209min.
View attachment 378447

In light of the fact that we know A2 has ~5% risk and A1 has ~1.6% risk, I don't see how you'd expect the LANL profile to be substantially better than A2.

The chart below shows the integral supersaturation by compartment for the 4 profiles.
View attachment 378448
LANL's was aggressive at protection of the fast tissues. You can see that the supersaturation exposure in compartments 9 and above map very closely for A2 and LANL. That would make it hard for me to think the LANL profile would perform substantially different than A2.

Finally, the total supersaturation exposure is shown below.
View attachment 378449

Absent some sort of solid dive trials (i.e. a closely controlled trial like the NEDU study), I look at those charts and wouldn't expect LANL to somehow perform better than A1 and really would expect something closer to A2.

Corelation papers show completely different story than your heat maps. It has a databank of a lot of dives with known outcome as a base. For more your heat maps have errors in them and you did not jet prove being correct.
That is all. Your conclusions are questionable and unreliable.