Bubble model vs. Gradient Factors redux

[KenGordon]

While I can see the argument for the same run times from one scientific perspective, are identical run times a worthwhile comparison from a diving perspective?

The run times are one of the outcomes of the particular algorithm. A valid diving comparison is one which compares different algorithms for the identical diving mission: i.e. Each diver dives to X depth for Y minutes and performs Z work. The 1. run times** taken to decompress, 2. the resulting DCS, 3. residual nitrogen, 4. blood/tissue micro-bubbles, and 5. anything else -- are all outcomes of each algorithm.

To eliminate individual subjective physiological factors, each diver should repeat the same set of dives with each algorithm, with sufficiently long surface intervals to completely recover between sets of dives. Eg performing the dives a week apart.

EDIT: ** By run times I mean to encompass the entire decompression stops depths/times ladder.

Only if you want a beauty contest for the algorithm.

For example, suppose plan A gets me out at a run time of 90 minutes and plan B at 100 minutes and plan a is slightly more bubbly, which is better? What if I stay an extra 10 minutes at 6m on plan A and I am now less bubbly than on plan B, now A padded is better than B, so is A better than B?

So the question is, why do we care how long it takes to get out of the water? Well, there are constraints to the possible time on the bottom due to ascent time and thus gas used, so we want the deco to be as fast as possible without being too fast.

On CCR the main limiting factor is deco gas use on bailout.

 

[ToneNQ]

Well, if you want to compare models, you can only change one parameter at the time. Otherwise, you can't know where the difference is. That's research 101......................

You'd need to be really careful with this, otherwise your findings have no reflection on the models.

To understand the variance within a model, you can only change one parameter at a time. But, to understand the variance between models, you must dive the same profile with the same divers. If you adjust them for a common factor, you are no longer comparing the algorithm. So what is the goal?

Spinsi compared RD to a reasonably deep GF for ZH-L16. It looks like the DAN study is going to compare to different sets of gradient factors within ZH-L16. If the plans each set of GFs produce and then hacked for the same run time, what are you actually studying? The GFs become less relevant because the GFhi is longer valid if you overstay your shallow stop.

Unless I am missing something. I must confess to enjoying these threads recently, its stretching my thinking in an area where I should be a bit more up-to-date.

 

[BenjaminF]

If you can sit through to the twenty minute mark of the Powell interview, there is a striking theoretical heat map comparing VPM on/offgassing with GF's.
It's not a study, but the heat map concept has been borne out by the bubble data, IMO.

Given this and @jvogt 's data, it seems that VPM becomes far less conservative as one does deeper dives, AND it causes one to surface with more loaded tissues when compared to GF. As such, (leaving aside significant investments - finanial, time and reputation) how could anyone with more than a single brain cell in their skull recommend VPM?

 

[Storker]

To understand the variance within a model, you can only change one parameter at a time. But, to understand the variance between models, you must dive the same profile with the same divers. If you adjust them for a common factor, you are no longer comparing the algorithm.

Well, that was what the NEDU study did. They compared two algorithms with the same run time. One emphasized deeper stops, the other emphasized shallower stops. Same bottom time, same decompression time, different ascent schedules.

Multivariate DOE is a powerful technique, but you need some indications about which variables have an effect on the outcome. That is usually obtained from simple OVAT studies, and then you can proceed to a multivariate DOE to study interactions between the variables. The research question in the NEDU study was pretty simple, and some of the leading hyperbaric scientists in the world chose to do an OVAD study. I certainly won't try to tell them they were wrong, because I'm not a hyperbaric scientist.

If the plans each set of GFs produce and then hacked for the same run time, what are you actually studying?

The effect of the ascent schedule. Given a certain amount of time spent on deco, would you be better off by doing deep stops and shaving the shallow stops, or would you be better off using what has been called "bend and mend"?

The GFs become less relevant because the GFhi is longer valid if you overstay your shallow stop.

Please elaborate.

 

[ToneNQ]

.........
The effect of the ascent schedule. Given a certain amount of time spent on deco, would you be better off by doing deep stops and shaving the shallow stops, or would you be better off using what has been called "bend and mend"?

In the interests of full disclosure, I dropped out before DOE, the interest in statistics just wasn't there. The concern I would have is that you are comparing ascent profiles, but not the factors within the algorithm. Correlation, causation, etc, I don't think you need to push into 'bend and mend' to find that. The Spinsi study got statistically relevant data sticking with a dive profile, rather than a runtime/TTS. The DAN study appears to have done the same. Is it not possible to test deep stops as a 10/85 against a 50/85, and the TTS is simply what it needs to be?

Please elaborate.

The GFhi defines the length of your shallow stops. If you overstay those stops for the purpose of achieving a TTS, then the GFhi is not a defining your ascent profile. Setting 50/85, then overstaying, you may have been diving 50/75. Or am I misunderstanding? Just seems to me like you are introducing special cause data, to reproduce the results using the dive profile and the GFs, you also need to include the TTS adjustment that was specific to the comparative study.

 

[Umuntu]

For example, suppose plan A gets me out at a run time of 90 minutes and plan B at 100 minutes and plan a is slightly more bubbly, which is better? What if I stay an extra 10 minutes at 6m on plan A and I am now less bubbly than on plan B, now A padded is better than B, so is A better than B?

These would all be valuable and valid outcomes.

An experiment that requires the same run times is not really testing the GF profile or Algo.

Where do you shave off the minutes to make a GF Profile's longer run times fit into your required run time ?

And are you then actually testing that GF Profile because you have done "negative padding" ?

You end up "padding" either way: A fixed run time requires negative padding.

 

[scubadada]

As was discussed in the video, there's not much money to support decompression research these days, not the Navy, not the oil industry. I'm glad to have whatever new data we can get.

That being said, we need to make sure we take full advantage of any opportunity we have to obtain more information regarding decompression strategy. In addition to detection of bubbles at several time points following the dive, perhaps following the levels of inflammatory markers, like the chemokines CCL2 and CCL5, as was done in the Spisni study, published in 2017, would be a good idea. Personally, I do not know if these markers are accepted and validated, but they were the basis for the difference in the decompression strategies in that study.

There was no mention of these markers being obtained in the DAN study mentioned by Mark Powell in the video. Even if funding for additional tests like this cannot be obtained at the time the study is performed, appropriate specimens can be obtained for future use, serum or plasma, for instance. These samples could be used for additional testing when sponsorship or funding is obtained. Perhaps more importantly, these specimens would be available for testing in the future, should additional markers of decompression stress be identified.

We have to make the most of what we have.

 

[Storker]

The concern I would have is that you are comparing ascent profiles, but not the factors within the algorithm

The algorithms are just mathematical representations of models. Dissolved gas models like ZHL-16 are based on well-known physics and physiology and have been verified¹ in vivo. Bubble models are based on well-known physics (the Young-Laplace equation), but has afaik never been verified in vivo, only in vitro.

¹ Note: if a model is verified, that only means "it can be used for predictions within the investigated boundaries", not that it's a correct representation of reality.

Setting 50/85, then overstaying, you may have been diving 50/75. Or am I misunderstanding?

I believe so. I'm not talking about overstaying. At least I believe I'm not

Take a look at post #37. There, @jvogt has basically done what I'm suggesting. Two different algorithms, same depth, same run time. I'd probably try to find a GF combo with a GFlo around 50, which is what the smart guys currently choose for their own dives, and perhaps play around with the conservatism level in VPM-B to get reasonable numbers for all three parameters (conservatism, GFlo and GFhi). If that puts my GFhi only close to e.g. 85, so be it. I'd be comparing VPM-B+x to, say GF 50/82 (or whatever the GFhi number might be), but so what? They'd still have different ascent profiles, one emphasizing deeper stops, the other emphasizing shallower stops.

Or I might decide to take it all the way and deliberately choose a higher GFlo to emphasize the differences in ascent schedules. That might put me past the peak of the Laffer curve of decompression, so I might want to run three ascent schedules, say VPM-B+4, GF 82/82 and GF 50/xx.

Tl,dr: whatever GFs I'd need to reach the same total run time as the bubble model's run time is those GFs I'd want my research subjects to dive. No overstaying. And then I'd be comparing VPM-B+x to GF yy/zz properly.

 

[J-Vo]

At GF 108 Herr Dr Buhlmann says you should get bent. This kind of experiment would be unethical unless you're NEDU.

Ya, but if your still a VPM user, thats +4 conservative, but we could up it to +5 and probably get that number down.

 

[Storker]

Ya, but if your still a VPM user, thats +4 conservative, but we could up it to +5 and probably get that number down.

Careful now. One particular self-professed deco science guru (/programmer) without academic credentials might come along telling you that if you up the VPM conservatism level too far, that doesn't make sense and isn't a proper algorithm. Even if there's no such limitations in the original algorithm.