We want your dive profiles!

[Duke Dive Medicine]

OK thanks for the clarification, I think.

Dave,
I'll try to be more clear. We can access the PDE data, but what we're really interested in are profiles from people who've had DCS. In these cases, we are looking for more data points than the PDE data can provide. Ideally, we will get a few dive profiles from divers who have had decompression illness, put the profiles through the modeling software, come up with a measure of the divers' decompression stress, and compare that to specific treatment data (lab values, neurological examinations, number of chamber treatments, treatment outcome, etc). We'll then use these cases (de-identified, of course) in our monthly case conferences that we have with the other facilities that have hyperbaric fellowships.

 

[lamont]

i assume you know that will skew your responses as people who haven't had DCS decide its not worth their time to submit data?

 

[Duke Dive Medicine]

i assume you know that will skew your responses as people who haven't had DCS decide its not worth their time to submit data?

It's not a scientific study so selection bias is not a concern. Again, this is for training for hyperbaric fellows at Duke and elsewhere.
Thanks,
DDM

 

[Indian Valley Scuba]

Thank you for your detailed response!

 

[rossh]

Divers,

We are developing new training material for our Undersea and Hyperbaric Medicine fellowship and would like to start a database of diving computer depth/time profiles. We plan to run the profiles through probabilistic modeling software to determine the relative risk of decompression illness for each individual profile. If the dive resulted in some form of decompression illness, we'd like to know that as well, and if you're willing, we'll ask you for some more information about your treatment and outcome.

I have a data base of almost 40,000 profiles. The same data is linked to the PDE reporting at DAN. I was talking with another person at DAN recently about this idea.

There are problem is this study idea. The Navy probabilistic data base is built on square profiles with shallow stop methods. Dives that do not follow that profile, will compare badly. But real world rec / tech divers almost never do square profiles, and in our database only 2% of dives come close to being described as square. Almost all divers today use a deep stop model or system of some kind.

How is the comparison going to normalize a multilevel or ascending profile to a square one? How is the comparison going to account for different deco stop methods? If the comparison is just based on the max depth and end run time, then its not a valid comparison.

Regards

 

[vladimir]

There are problem is this study idea.

It's not a scientific study

I hope that was helpful.

 

[Duke Dive Medicine]

I have a data base of almost 40,000 profiles. The same data is linked to the PDE reporting at DAN. I was talking with another person at DAN recently about this idea.

There are problem is this study idea. The Navy probabilistic data base is built on square profiles with shallow stop methods. Dives that do not follow that profile, will compare badly. But real world rec / tech divers almost never do square profiles, and in our database only 2% of dives come close to being described as square. Almost all divers today use a deep stop model or system of some kind.

How is the comparison going to normalize a multilevel or ascending profile to a square one? How is the comparison going to account for different deco stop methods? If the comparison is just based on the max depth and end run time, then its not a valid comparison.

Regards

Hi Ross,
First, thanks for your all your work on V-Planner and for offering up your data to DAN. I'm familiar (at least in general) with your conversation with Petar and have just been talking with Dr. Vann about it. Let's keep the dialogue up, not necessarily in this forum, but certainly among those already involved. We all have the same goal, and that's to benefit the dive community.

Regarding your other point: the Navy data is the largest, most detailed, best controlled database we have and so at the moment, it's what lends itself best to probabilistic modeling. Work is being done to incorporate other data sets right now. I'll PM the deco modeling team at the EDU, hopefully someone can jump in here and provide more detail.

Right now, I'm hoping to have some divers respond to this thread who would like to contribute to the education of our hyperbaric fellows. If I get enough profiles, though, especially from people who've had DCS, I can tease out data points and detailed medical information on an individual basis, which I can pass on to Dr. Vann et al and enable the guys on the science end to further tune the model.
Best regards,
DDM

 

[TSandM]

I was reading the paper on comparisons that you linked to on FB, and I was really wondering how you can use the profiles generated by the various programs and estimate their DCS risk. Even after your explanation here, I'm still unsure. Is it that the risk is KNOWN for the Navy models, because of the amount of data gathered, and you have some mathematical method of extrapolating the stop profiles from other programs into the Navy algorithm? It seems as though Ross's concerns would be quite valid in that case.

 

[Duke Dive Medicine]

I was reading the paper on comparisons that you linked to on FB, and I was really wondering how you can use the profiles generated by the various programs and estimate their DCS risk. Even after your explanation here, I'm still unsure. Is it that the risk is KNOWN for the Navy models, because of the amount of data gathered, and you have some mathematical method of extrapolating the stop profiles from other programs into the Navy algorithm? It seems as though Ross's concerns would be quite valid in that case.

Lynne,
The Navy probabilistic modeling program is based on about 150,000 or so carefully controlled experimental dives, for which there are a known number of well-documented bends cases. I can't explain the statistics behind it as well as the EDU folks and Dick Vann can, but it's similar to approximating a curve. The more data points you have, the closer you can come to a realistic statistical model, and the fact that the dive profiles are square profiles becomes less and less relevant. It's not perfect, but then again no model is. That's why we're working with DAN and a number of other organizations to obtain more dives from a wider variety of sources. Of note, the dives have to have outcomes associated with them, i.e. we need to know if the diver attached to the profile had decompression sickness, otherwise the data are of limited utility.

Once again, my primary purpose in requesting dive profiles, especially from divers who've had DCS, is to provide some useful, real-world training for our fellows. We want to have the ability to analyze a case from start to finish, including the relative risk score of the dive and any treatment that the diver received. If enough divers are willing to share their profiles, we may be able to use the raw data for the purpose mentioned above.

So far, I've got zip. Hopefully by now the readers of the thread/potential participants realize that I really do work for Duke, I'm not a hacker after their identities :cool2: I'm not going to use names, locations, or any other identifying information other than that which I requested in the initial post, nor is anyone going to pass judgment. Many people still think that decompression illness is a mark of having done something incorrectly, and in our experience, that's usually not the case. We look at DCI as an athletic injury, just like an ankle sprain or ACL tear.

Anyone who wants more information can contact me either via PM or via the email address listed in the initial post, hexda001@mc.duke.edu.

 

[bvanant]

I am quite sure that no one really believes you are a hacker, your spelling is far too good. The question is what is the utility to your study of dives that did not result in any type of DCS. If you want the data, I can send a few thousand dives but none of them resulted (he says knocking on his head) in any type of DCS. From what I know of statistical outcome modeling you need lots of DCS cases and then lots of dives that are very close to the DCS dives that didn't end up in DCS but that is extrapolation from some other modeling work. In any case, if you need dives that didn't lead to any type of DCS I will be happy to send you a bunch.

Bill