INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

DURING THE LAST DECADE, DEVELOPMENT of human decompression procedures has been significantly aided by the use of probabilistic models for predicting decompression sickness (DCS; Ref. 25). By using techniques such as maximum likelihood, the probability of DCS can be expressed as a continuous function that includes variables such as depth, gas composition, and decompression profile. Unfortunately, although DCS is assumed to result from an excess of inert gas somewhere in the body, little is known about the properties of gases under high pressure in tissues (26) or about how bubbles develop and presumably lead to DCS (5, 23). Consequently, probabilistic modeling has generally been done empirically by fitting functions to a set of dives that allow risk prediction but have little, if any, physiological basis. As a result, most human predictive models do not extrapolate reliably to dive profiles much different from the original data, particularly to those that have high risk.

Use of animals offers the potential to improve prediction of DCS in humans. Large numbers of animal dives can be performed relatively easily and inexpensively over a broad range of DCS incidence rates. These data are often more suitable for modeling decompression risk than human data containing relatively few cases of DCS. However, much animal research has focused on investigating the pathophysiology of DCS or addressing basic scientific questions such as the role of bubbles in DCS rather than on helping to estimate human risk of DCS. Exceptions to this include the early use of goats to help develop decompression procedures for humans (4). A next major step is to develop a formal procedure for using animal decompression outcome to help predict human DCS.

Modeling efforts in our laboratory with small animals have demonstrated that prediction of risk of DCS can be improved by accounting for differences among gases in terms of 1) "relative potency" (RP), which defines the level of decompression risk on a per-unit-pressure basis, and 2) gas exchange rates (12-15). These previous models have generally dealt with specific types of dives such as saturation, gas switching, or dives with variable decompression. This report describes the next step of combining the different data sets and developing one general mixed-gas model for prediction of DCS in rats over a wide range of dive profiles with various mixtures of He, N₂, Ar, and O₂.

We believe that this is the first attempt to construct a model of such scope in terms of gases and types of dives. The rationale behind our work was that such a model should provide additional information about the role of different gases in decompression, thus increasing our understanding of the mechanisms of DCS. The hypothesis for this study was that such a model could be produced that would allow reliable predictions without significant bias with regard to the exact dive profile. This work was not done in anticipation of an increased use of mixed-gas diving in the US Navy. However, our work does provide a global model for possible future evaluation of its usefulness for human prediction. Recently, a similar effort using animal decompression outcome to predict human DCS was begun with sheep (2).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Hill-equation dose-response models were fitted by the technique of maximum likelihood to a database of over 5,000 rat dives created by combining five sets of dives with very different profiles that were conducted over the last decade at the Naval Medical Research Institute (NMRI). The first four sets (data sets 1-4) had been analyzed separately earlier and previously reported (12-15); a fifth set (data set 5) represents new data that have not been modeled or described before. These dives, using various mixtures of He, N₂, Ar, and O₂, had times at depth up to 2 h and varied decompression profiles. A general mixed-gas model for prediction of decompression risk in rats was produced.

Animal Use

Animal numbers were one of many details carefully reviewed by the NMRI committee, and statistical justification was required to support all animal requests. In the case of decompression modeling, large numbers of dives are often required to produce parameter estimates with the precision needed to resolve differences in risk or gas exchange. Discussion regarding selection of the specific number of animals used is contained in some of the past reports (12-15). Similar studies involving large numbers of rodents continue to be approved at NMRI, as evidenced by the most recently conducted experiments comparing the decompression risk of He and H₂ (16).

Data Used for Modeling

Experimental Protocols

Male albino rats (Rattus norvegicus, Sprague-Dawley strain) weighing ~200-300 g were obtained from a local supplier and housed in the Institute's animal care facility for at least 1 wk before use. The animals were allowed access to food and water up until immediately before the dive began.

General diving procedures. Five animals were placed in a cylindrical cage, which was then loaded into a small-animal hyperbaric chamber. The chamber was compressed at a rate of 1.8 atmospheres absolute (ATA)/min to depth. Gas mixtures were made by serially adding air, O₂, or one of the inert gases (N₂, He, or Ar) during compression, with the precise amounts depending on the intended composition. For dives in which N₂ was to be absent in the dive mixture, the chamber was flushed with an appropriate amount of O₂ over a 5-min period before compression to remove all N₂ (13).

Inert-gas decompression dives (data set 5). These dives involved various inert-gas mixtures (20.9% O₂) using N₂, He, and Ar. Most animals were held at depth for 1 h before being decompressed slowly (1.8 ATA/min) to a prescribed stopping depth and held there for up to 40 min. Animals were then rapidly (<10 s) decompressed to the surface. The exception to these profiles was one group of animals, dived on Ar-O₂, in which the time at depth was varied from 10 to 120 min before rapid decompression to the surface. These latter dives provided data from a type of profile that had been done previously with air and He-O₂ (data set 2) but not with Ar-O₂. The exact dive profiles that were performed are given in the leftmost column of Table 2.

Data Analysis

Parameter values of the model were adjusted to maximize the log likelihood (LL) of the model by use of a modified Marquart nonlinear estimation algorithm (17). The likelihood ratio (LR) test was used to evaluate the significance of estimated parameters on the basis of improvement in fit (11). The shape of the likelihood surface near the converged parameters was used to estimate the precision of the parameter values. All DCS cases were treated equally, with no allowance in the model for grades in severity. The only accommodation for differences in response was to model all DCS cases (including fatalities) and then to repeat the modeling with only cases of death.

The dose-response model used for this analysis was the Hill equation

(1)

where P₅₀ represents the dose at which there is a probability of 50% for the occurrence of DCS (or death), and the exponent (n) is the order of the Hill equation that controls the steepness of the central portion of the sigmoidal curve.

The dose in Eq. 1 represents a measure of decompression stress and was defined in a manner similar to previous reports (12-15) on the basis of the traditional idea of total gas supersaturation. In the present experiments, the gases contributing to the decompression response are He, N₂, Ar, and O₂; CO₂ is ignored. Here, O₂ is treated as an inert gas, adding to the risk of DCS. Although other ways of modeling a potential effect of O₂ exist, no one method appears to offer a clear advantage (19, 22). This issue is discussed further later in the text.

To allow for possible differences among gases, each partial pressure is multiplied by a RP value that weights each gas according to risk on a per-unit-pressure basis. The resulting equation for dose (in ATA) is

(2)

where Pti_He, Pti_N2, Pti_Ar, and Pti_O2 are the tissue partial pressures (Pti) of He, N₂, Ar, and O₂, respectively (in ATA), in the animal immediately on reaching the surface. The parameters RP_He, RP_N2, RP_Ar, and RP_O2 are the RP values for He, N₂, Ar, and O₂, respectively. The subtraction of 1 ATA defines the amount of supersaturation after decompression to the surface.

The Pti of He, N₂, Ar, and O₂ in the animal were calculated on the basis of single exponential kinetics, assuming a single compartment (i.e., whole rat) even though real tissues may need a more complex model (10, 18). After a step change in partial pressure of one gas at time 0 (t₀), the Pti of the gas at time t can be described by the following equation for t > t₀

(3)

where Pam_i is the ambient partial pressure at time t, Pti₀ is the initial Pti, TC_i is the time constant (TC) affecting the rate of gas uptake or washout, and i is an index referring to each gas.

The calculation for Pti becomes more complicated during actual dives in which the ambient gas partial pressures change in a nonstepwise manner (e.g., during compression or decompression). The solution, as previously described in Ref. 16, was to treat each dive as a series of partial pressure ramps that describe the pressure history of the dive. Partial pressures of individual gases at the beginning and ending of each ramp were estimated by multiplying the chamber pressure by the percentage of each gas as measured. For calculation purposes, each ramp was subdivided into a number of smaller time intervals, and Eq. 3 was used to successively compute Pti values at each interval along each ramp until decompression was completed. This procedures was performed for each of the gases that was being considered, in this case He, N₂, Ar, and O₂.

Separate TCs for the uptake and washout of each gas allowed for the possibility of asymmetry in gas kinetics. This was implemented by comparing Pti to Pam_i for each gas and adding the parameter TC_i to TC_i when Pti₀ > Pam_i in Eq. 3 (i.e., during washout). This situation generally occurred some time after decompression began, although this was not the case for those instances in which the dive gas was switched to a different mixture before decompression.

Gas kinetics in Eq. 3 assume that at equilibrium the partial pressures in the animal become equal to those in the chamber. This assumption avoids the difficult task of choosing a more complicated model on the basis of inadequate knowledge of tissue-blood relationships. Although Eq. 3 models gas kinetics in the rat, it is important to emphasize that gas uptake and washout were not actually measured. Consequently, what is being modeled is the animal response to decompression and the effect of the estimated gas partial pressures on decompression risk. The TCs that are being estimated define the rates of change in processes that affect the probability of DCS or death, although interpretation in terms of gas kinetics is made in this report.

The partial pressures of the gases are reported in atmospheres absolute. To estimate the RP values, one of the potencies had to be fixed so that the other potencies could be calculated in relation to it. The RP_{N 2} value was arbitrarily set to 1.0 so that the P₅₀ is expressed in terms of the partial pressure of N₂ in atmospheres absolute. The effect of this weighting calculation was to convert exposures of He, Ar, and O₂ into equivalent N₂ exposures.

Because rat weight has been shown to have a significant effect on the decompression outcome, a weight correction term for dose was included in the model, as previously done, with a power function (13). Animal weight (Wt), normalized to a weight (260 g) close to the average weight of all animals (252 g), was raised to an exponent denoted as the weight factor (WtF). Consequently, the final expression for dose was

(4)

The modeling effort began with the individual data sets, which were then combined in all possible ways until the model was finally fitted to the combined group of all five sets, data sets 1-5. Parameter estimation began with the simplest possible model: P₅₀, n, and symmetrical gas kinetics with a single TC for uptake and washout for each gas. Each subsequent level of complexity of the model, with all possible combinations of estimated parameters, was evaluated by at least five different sets of starting parameter values to ensure that the maximum LL found was a global and not a local maximum. Parameter combinations that showed promising improvement to the fit were explored with many more starting values so that, overall, several thousand separate starting parameter sets were evaluated.

The next step was to perform an LR test to determine whether the data sets were statistically combinable

(5)

where LL₁ and LL₂ are the LLs of the same model fitted to data sets 1 and 2 separately, and LL₁₊₂ is the LL fitted to the combined data sets. Data sets found by this criterion not to be combinable are thought to generally produce better predictions of decompression outcome when modeled separately. This issue will be discussed in more detail later.

In summary, this model 1) predicts the probability of DCS in rats subjected to dives with mixtures of He, N₂, Ar, and O₂; 2) assumes that the decompression response is dependent on the degree of supersaturation of these gases in the animal; and 3) is used to estimate the parameters governing the location and shape of the dose-response curve, the values of RP, TC, and TC for the individual gases, and the exponent correcting for animal weight.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Data Set 5

Summary of All Data (5 Data Sets)

Model Parameters

Reported results include those with the model fitted to all of the data combined as well as fitted separately to each of the five individual data sets. Because of the unique aspects of the gas-switching experiments (data set 3, Ref. 15), modeling results are also given for the combined data after exclusion of data set 3 (noted as data sets 1, 2, 4, 5 in Tables 4 and 5). Current parameter estimates agree closely with previously reported results for the individual data sets (12-15), with the exception of data sets 1 and 2, which had been fitted earlier with a somewhat different model. This would be expected but represents a good check of the new computer code and database structure that were written to allow analysis of the combined data. We note that the previous models included only one TC for both uptake and washout of each gas, thus forcing symmetrical gas exchange.

P₅₀ values and exponents. For all the individual data sets except data set 2, P₅₀ values were greater for death than for DCS, reflecting the higher dose of decompression stress required to cause death. With the exception of data set 3 (gas switching), exponents were also greater for death vs. DCS, indicating a steeper response slope for death. For the model fitted to the combined data, with and without data set 3, the P₅₀ was similar for DCS and death, although the exponent was greater for death.

Potencies. As was previously reported (12-15), large differences in potency for causing DCS or death were seen here among gases, although the magnitude of the differences varied somewhat with the specific data set. Helium potency was estimated for the individual data sets at ~10-20% less than that of N₂ for both DCS and death, although it was somewhat lower for data set 3. Ar potency for both responses was estimated to be up to ~55% greater than that of N₂, and O₂ was shown to introduce significant decompression risk with a potency up to 90% (DCS) or 50% (death) of that of N₂. Potencies for the combined data, with and without data set 3, agreed with the values estimated from individual data sets.

View larger version (20K): [in this window] [in a new window]   Fig. 1.   Saturation dives. Probability of decompression sickness (DCS; A) and death (B) increases with saturation depth for He (), N2 (), and Ar (); all single inert-gas mixtures with 20.9% O2. Differences in the plots of the 3 gases reflect the differences in potency (relative decompression risk on a per-unit-pressure basis): He < N2 < Ar. Decompression was rapid (<10 s) to the surface. Predictions were based on the model for the combined 5 data sets with rat weight set at 240 g, which was close to the mean weight (238 g) of the saturation dives. ATA, atmospheres absolute.

Gas exchange: Single data sets. For the individual data sets, gas uptake and washout rates were generally the same, although there were large differences among the three inert gases. The TC for He was estimated at ~3-6 min, although it was considerably greater for data set 3. If equilibration is considered to occur in four TCs, then He saturation or washout would occur in ~10-25 min. In contrast, gas exchange for N₂ and Ar was generally considerably slower, with saturation estimated at ~50-80 min. The major exception to these findings was data set 5, which included many long decompression dives that might be expected to improve the estimation of gas washout. Interestingly, for this data set, N₂ washout for DCS was estimated to be much faster than N₂ uptake. For prediction of death, N₂ uptake and washout for data set 5 were symmetrical and much faster than with the other data. For all five individual data sets, the O₂ TC was indistinguishable from zero and therefore was fixed at this value during model estimation (the TC for O₂ was actually fixed at 0.00001 to avoid the calculation problems associated with using zero). O₂ in the animal would thus be expected to equilibrate almost immediately with a change in chamber PO₂.

Gas exchange: Combined data. In the DCS model for the combined five data sets, gas uptake rates were indistinguishable from washout rates for He and N₂ but not for Ar. With the exception of Ar, these results agree closely with the results from the individual data sets in terms of both symmetrical gas exchange and magnitude of the TCs. However, for death, significant differences in rate of gas uptake and washout were predicted for all three inert gases on the basis of the five-data-set model. Although the uptake rates for death were similar to those for DCS, He washout for death was approximately twice as slow as uptake, and N₂ and Ar washout were approximately twice as fast as uptake. These asymmetrical gas kinetics for death differ from the symmetrical gas exchange estimated for the separate data sets and are also observed when the data is modeled after data set 3 is omitted.

View larger version (20K): [in this window] [in a new window]   Fig. 2.   Variable time at depth. Probability of DCS (A) and death (B) increases with time at depth for He (), N2 (), and Ar () for a 6.30-ATA dive; all single inert-gas mixtures with 20.9% O2. Differences in the maximum probability each gas reaches at saturation reflect differences in the potencies of the 3 gases (He < N2 < Ar). Helium reaches its asymptote faster than N2 or Ar because of its faster gas uptake time constant. Decompression was rapid (<10 s) to the surface. Predictions were based on the model for the combined 5 data sets with rat weight set at 260 g, which was close to the mean weight (258 g) of the variable time-at-depth dives.

View larger version (21K): [in this window] [in a new window]   Fig. 3.   Predicted gas pressures. Predicted partial pressure changes (thin lines) in rats during a 1-h dive at 9.3 ATA with 3 single inert-gas mixtures with 20.9% O2. Initial ascent was done at 1.82 ATA/min to a stop at 4.64 ATA for 20 min, followed by rapid (<10 s) ascent to the surface. Partial pressure differences reflect differences in uptake and washout time constants. Thick lines represent total chamber pressure. Predictions for DCS (A) and death (B) were based on the model for the combined 5 data sets.

Weight correction. Inclusion of a weight correction (the parameter WtF) in the model produced a significant improvement in fit for both DCS and death in all data sets except data set 5. However, the magnitude of WtF varied somewhat with the data set. A significant weight correction was also found with the combined data, with or without data set 3. This correction results in a greater probability of DCS and death in heavier animals.

Effectiveness of the Model

View larger version (23K): [in this window] [in a new window]   Fig. 4.   Model error for the 5 data sets. Maximum likelihood models predict DCS (A) or death (B) without significant bias and generally equally well regardless of total decompression time and the specific data set. This is shown by plotting the difference between model prediction and observed incidence vs. total decompression time. Each of the 5 data sets is plotted with a different symbol: data sets 1 (), 2 (), 3 (), 4 (), and 5 (). Predictions based on model for combined 5 data sets with rat weight set at 260 g. The observed values are the mean incidence rates of the specific dive profiles.

View larger version (20K): [in this window] [in a new window]   Fig. 5.   Model error for the 3 inert gases. Similar to Fig. 4, maximum likelihood models predict DCS (A) or death (B) without significant bias and generally equally well regardless of total decompression time and the specific inert gas. This is shown by plotting the difference between model prediction and observed incidence vs. total decompression time. Only single inert-gas mixtures (20.9% O2) are plotted: He (), N2 (), and Ar (). Predictions were based on the model for all 5 data sets with rat weight set at 260 g. Observed values are the mean incidence rates of the specific dive profiles.

We emphasize that actual incidence data also have their own associated errors (not shown on graphs), which are based on the binomial distribution and are affected by the sample size. Although this data error will limit interpretation of the residuals, we point out that nearly all of the model predictions are within 40% of the observed incidence for DCS and within 30% for death. The ability of the model to predict that well over such a range of gas mixtures and dive profiles is, in our opinion, quite remarkable and agrees with past results when only two gases (He and H₂) were being modeled (16). This degree of error is all the more surprising given the nearly 150-g range in rat weight and its relatively large effect on risk of DCS (~1% increase in DCS with each additional gram of rat weight; Ref. 13). More importantly, the scatter of points around zero illustrates that the model predicts DCS or death without significant bias and generally equally well regardless of total decompression time. The five symbols for the data sets and the three symbols for the inert gases also appear to be randomly distributed about zero, suggesting the absence of systematic model distortion with respect to data set or inert gas. Also important is the fact that the mean value of the residuals, for each of the individual data sets and gases, is ~0, with the actual means ranging from 0.05 to 0.01.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The mixed-gas model developed here for prediction of DCS in rats is the first such model for any animal species that covers such a broad range of gas mixtures and dive profiles. The success of this model is demonstrated by 1) the relatively small errors, in view of the variety of the data, between model predictions and actual incidences and 2) the random distribution of these errors relative to total decompression time, gas mixture, and data set. Our results, based on over 5,000 dives, agree with many of the findings from past modeling of four subsets of these data (12-15), including prediction of 1) differences among the gases in potency for causing DCS and exchange rate, 2) significant decompression risk of O₂, and 3) increased risk of DCS with heavier animals. New results include the prediction of asymmetrical gas exchange with gas washout often unexpectedly faster than gas uptake.

As discussed earlier, gas kinetics were extrapolated here, as well as in our previous models, from rate constants estimated from changes in DCS risk rather than determined from direct measurement of gas uptake and washout. The simplistic approach we have taken here and in previous studies (12-15) avoids using more complicated models based on unproven assumptions about blood-tissue exchange. Experience has demonstrated that very different models can often predict the response to decompression equally well. For example, in the case of rats, little effect on predictability has been observed when 1) using alternate ways for correcting for rat weight or 2) omitting the 1-ATA subtraction in the dose (13). A term for the small constant contribution of the metabolic gases (venous PO₂ and PCO₂ and water vapor pressure) to decompression risk has sometimes been included in past modeling efforts by others. However, its importance to predictive ability has not been clearly demonstrated. For instance, a recent probabilistic modeling exercise with human decompression data (21) has shown that, when the specific values chosen for the metabolic gases were varied, there was no significant effect on predictability of DCS.

Our approach of modeling the rat response and treating all gases, including O₂, as a potential risk factor without using a physiologically based model is consistent with our past work. Again as before, we did not incorporate any link to bubble formation or growth into our overpressure model. Although a dose-response relationship has been reported for human decompression and Doppler bubble detection (7), symptoms of DCS in humans have never been shown to be well correlated with Doppler measurements (3). Watt and Lin (24) agree to the poor correlation of intravascular bubbles and DCS in animals.

A major difference with the rat model developed here, compared with the earlier models, is that this model allows for asymmetrical gas kinetics, which were predicted for all three inert gases for death but only for Ar for DCS. Until about 20 years ago, inert-gas elimination during decompression had been assumed to simply be the reverse of inert-gas uptake, both of which could be described by the same exponential function. However, a number of studies have since suggested that gas washout may be slower with decompression, owing in part, at least, to cardiovascular changes or bubble development (6, 9). Indeed, we have estimated that washout of He and H₂ is over an order of magnitude slower than uptake in rats after very deep dives (20-50 ATA), although the explanation for this extreme slowing is unclear (16). The need for kinetics that reflect slower tissue washout has recently led to use of decompression models with a combination of both linear and exponential rate functions (21). Consequently, in the present model, for the combined data, the slower washout of He predicted for rat death, relative to uptake, was not surprising. However, the faster washout predicted for N₂ and Ar was unexpected and, to the best of our knowledge, unreported before.

Because asymmetry in gas exchange was generally not predicted for the individual data sets, the issue arises concerning data set combinability and its effect on parameter estimates of the model for the combined data, particularly gas exchange TCs. As was stated earlier, we had proceeded with modeling the combined five data sets, although LR testing had shown that most of the data was not combinable. Despite our opinion that the formalism of the LR test produces a very conservative test that is often very difficult to pass, an important question remains whether merging of the data caused model distortion. Such distortion may have incorrectly led to prediction of asymmetry, including faster washout of some of the gases. We had concluded above that the random distribution of errors with respect to data set and inert gas suggested the absence of model distortion. We also indicated that all of the parameter estimates for the combined model, with the exception of the TCs, agreed well with the estimates for the individual data sets. However, the modeling differences with data set 3, with its relatively long He TC, and data set 5, with its asymmetrical N₂ kinetics (DCS) or relatively short N₂ TC (death), make these two data sets suspect for causing problems when combined with other data. Nevertheless, results argue against this. First, as discussed earlier, parameters for the combined model, with and without data set 3, agreed closely except for the Ar TC. Second, a similar comparison of the combined model, with and without data set 5, also produced good agreement, although this was not presented in Tables 4 and 5.

The most likely explanation for the predicted asymmetry of the combined model is that the merged data provided the range in dive profiles, not available in the individual data sets, necessary to resolve uptake and washout rates. This reasoning is supported by the narrow spectrum of dives within each of the data sets. Clearly data sets that were primarily designed to look at differences in gas uptake (data set 2) and gas washout (data sets 4 and 5) might be expected to be limited in their ability to estimate both aspects of gas exchange. Data set 1 (saturation dives followed by rapid decompression) should provide little kinetic information, although data set 3 (gas switching) might be expected to contain data helpful to describe both gas uptake and washout. However, we again stress that we are modeling the rat response to decompression and not gas kinetics, and we propose that the risk of DCS may become uncoupled from the gas load under certain circumstances, as suggested previously with regard to rats (16). Indeed, the comparison of the time course of predicted bubble evolution with predicted DCS risk in humans has indicated that gas-phase dynamics may not fully explain persistent DCS risk (1).

Until recently, only inert gases have been included in estimation of risk of human decompression; O₂ has generally been ignored. O₂ was assumed to be metabolized so rapidly during decompression that it had little effect on bubble formation and, presumably, on DCS. The present results support our previous findings (12, 14) that O₂ can substantially add to the risk of DCS in rats, although this effect diminishes very quickly during decompression. Recently, two very different probabilistic models have predicted that elevated PO₂ levels contribute to DCS risk in humans, although less than the equivalent amount of N₂ (19). One model treated O₂ as we did, as an inert gas adding to the decompression load, although a threshold was included that limited the contribution of O₂ to pressures above a certain level. The other model allowed the PO₂ to alter the N₂ washin-washout kinetics, acknowledging the ability of PO₂ to affect central and peripheral circulation. Interestingly, both models predicted risk for various human profiles similarly, demonstrating the inability to distinguish these very different approaches to defining the role of O₂ in decompression. These results again emphasize the limitations of decompression modeling and raise the issue of often-unneeded model complexity. No further improvement in fit was seen in that study (19) when a combination of the two effects was employed. In another report (22), the effect of O₂ was implemented similarly into a bubble model for predicting human DCS after N₂ and/or He dives. The best fit in that case was obtained by including a combined O₂ effect via the direct contribution to the bubble as a gas and vasoconstriction induced by hyperoxia. Despite concerns raised in these and other studies that O₂ should not be used blindly without concern for potentially contributing to the risk of DCS, O₂ is still deemed very useful during decompression (19). The advantage of using high-O₂ mixtures was evident during the recent development of decompression procedures for the US Navy's upgraded 1.3-ATA PO₂ rebreather, in which significant reduction in decompression times was achieved (20).