Effect sizes for symptomatic and cognitive improvements in traumatic brain injury following hyperbaric oxygen therapy

Abstract

Hyperbaric oxygen therapy has been proposed as a method to treat traumatic brain injuries. The combination of pressure and increased oxygen concentration produces a higher content of dissolved oxygen in the bloodstream, which could generate a therapeutic benefit for brain injuries. This dissolved oxygen penetrates deeper into damaged brain tissue than otherwise possible and promotes healing. The result includes improved cognitive functioning and an alleviation of symptoms. However, randomized controlled trials have failed to produce consistent conclusions across multiple studies. There are numerous explanations that might account for the mixed evidence, although one possibility is that prior evidence focuses primarily on statistical significance. The current analyses explored existing evidence by calculating an effect size from each active treatment group and each control group among previous studies. An effect size measure offers several advantages when comparing across studies, as it can be used to directly contrast evidence from different scales, and it provides a proximal measure of clinical significance. When exploring the therapeutic benefit through effect sizes, there was a robust and consistent benefit to individuals who underwent hyperbaric oxygen therapy. Placebo effects from the control condition could account for approximately one-third of the observed benefits, but there appeared to be a clinically significant benefit to using hyperbaric oxygen therapy as a treatment intervention for traumatic brain injuries. This evidence highlights the need for design improvements when exploring interventions for traumatic brain injury and the importance of focusing on clinical significance in addition to statistical significance.

INTRODUCTION

Hyperbaric oxygen therapy (HBO₂) is a technique designed to administer breathing air with an above average oxygen concentration inside a pressurized environment. The combination produces a substantial increase in blood oxygenation levels that facilitates healing conditions for numerous potential health issues. For example, the Food and Drug Administration (FDA) cleared HBO₂ to treat a limited range of conditions, including decompression sickness and carbon monoxide poisoning, even though various organizations and advertising materials have promoted HBO₂ as a solution for conditions ranging from Alzheimer’s disease to stroke (1). These marketing oversteps prompted regulatory warning from the FDA, yet scientific reservation for new uses does not undercut the established reliability of an existing treatment.

One promising alternative use has been the potential to treat mild traumatic brain injuries (mTBIs) or traumatic brain injuries (TBIs). This possibility has received attention from multiple communities where concussive incidents are pervasive problems, especially for the military and athletes in contact sports (2–6). Multiple reviews have strongly suggested that HBO₂ should be widely used to treat mTBI/TBI (7–10). However, randomized clinical trials with placebo controls have failed to consistently produce significant improvements for active HBO₂ treatments relative to the control group (11, 12). This inconsistency has led official government reports to offer far more tepid enthusiasm about HBO₂ treatments relative to some advocacy organizations (13, 14). Moreover, there are lingering questions about the efficacy, durability, and validity of HBO₂ in treating traumatic brain injuries. This review will explore many of the larger HBO₂ interventions from recent years with a particular focus on effect sizes. Specifically, does the potential for sham placebo effects discount the clinical significance of observed improvements?

In the case of HBO₂ applications for brain injuries, there is a potential mechanism of action to explain why the treatment might be effective. Blast-related injuries can create changes in cerebral vasculature that lead to chronic disruptions of microvasculature long after the initial exposure event (15). Hyperoxygenated blood with a high concentration of dissolved oxygen in the bloodstream—as made possible by the increased pressure—would allow oxygen to penetrate deeper into these damaged brain tissues. This oxygenated physiological condition may allow for a neuroprotective or regenerative environment that promotes angiogenesis and neurogenesis (16–20). HBO₂ would then help concussion-related injuries through increased dissolved blood oxygen levels penetrating deeper than otherwise possible without the combination of higher oxygen concentrations within a hyperbaric chamber. This potential explanation has been demonstrated in rat models, but the possibility of a causal relationship between the treatment and medical benefit is an important step in supporting alternative uses of the procedure. A causal mechanism also provides a pathway for future research to test and validate the ideas in other animal models before refining treatment protocols with humans.

Unfortunately, this mechanistic success has been supported more conclusively by the animal studies. Human studies have yielded an inconsistent set of conclusions. We should note that this preceding language is chosen very carefully—that is, emphasis is placed on inconsistent conclusions, not inconsistent evidence. Conclusions can be complicated by design-related issues and overreliance on statistical probabilities rather than clinical significance. Both problems occurring in concert among a rapidly expanding literature make it difficult to draw reliable inferences from the evidence, which requires greater attention to how omnibus conclusions could be misled.

Four arguments currently exist to explain why the evidence may be inconsistent (21). First, there is the pervasive concern of placebo effects in HBO₂ protocols. The simplified account would be that control groups exhibit such a large placebo effect that the active treatment value becomes masked. The problem with controlling this placebo effect is that the source may be multifaceted. A placebo could be induced by natural healing among the control group over the treatment period, a Hawthorne effect in answering questions (22), or simply the best intentions and attentions of committed care advocate providing treatment to wounded veterans and injured athletes. Inconsistent placebo effects could likewise expand the standard deviations of any observed outcomes and thereby make it more difficult to statistically differentiate among active treatment and control conditions. Any such influence could prompt control conditions to exhibit a placebo-like effect and diminish perceived benefits.

Second, sham controls may actually induce a therapeutic benefit (23). Pressurized conditions can be detected by individuals at relatively low atmospheres of pressure (24), and so, sham conditions often involve some partially pressurized environment to prevent participants from knowing their active or control designation. The problem is that the partial pressure could induce higher blood oxygenation levels and promote the same—albeit smaller—healing environment intended for the active treatment condition. So-called active shams would then complicate any statistical conclusion because the comparison would be against an elevated baseline due to a partial version of the treatment, not a true control.

Third, there is a substantial amount of treatment heterogeneity among various protocols. Different participants undergo different oxygen/pressure combinations for different individual session lengths and different total treatment lengths. This heterogeneity could easily contribute to different effect sizes, especially given that pooled analyses have indicated the potential for a dose-response curve in treatments (25). Even if the treatment itself were applied consistently, mTBI/TBI can present with heterogenous symptom profiles of differing severity. The same blast exposure might not produce the same symptoms in two different individuals, which often requires a larger participant sample to overcome any issues in symptom severity. These symptoms are also not measured by the same scales, as some studies use cognitive tests, other studies use self-report symptom profiles, and still others use different self-report measures. Heterogeneity among all these sources can complicate any conclusion, as apples become compared with oranges and bananas.

Finally, there is the facilitation argument, which pertains more to alternative uses associated with general quality of life (QOL) or psychological issues. The premise is that alleviating cognitive difficulties will facilitate improvements in general quality of life. In this sense, there is no direct benefit of HBO₂ to a psychological disorder—it merely improves impaired cognitive abilities, and the changed mental status affords the patient more opportunity to focus on other aspects of life. The result is that HBO₂ would facilitate a general recovery without applying to any causal relationship in improvement. From a medical perspective, the comparison is akin to cancer treatments being less complicated when the patient has no other health issues that might compromise the immune system.

Of these four concerns, the placebo effect is the most problematic for clinical investigations with randomized controlled trials. Placebo effects further complicate the sham-is-not-a-sham argument that proposes partial pressure conditions promote a healing environment for the patient. Treatment heterogeneity could be addressed by more tightly controlled designs and dose-response curves, whereas the facilitation argument becomes less relevant for issues that are not purely psychological. Placebo effects, meanwhile, contribute to challenges in statistically detecting a difference between the treatment and control conditions. Combined with the tendency for natural improvement over time, placebo effects could mask the active benefit of any HBO₂ treatments.

There are several reasons to consider why these placebo effects might be masking a real therapeutic benefit for patients with mTBI/TBI who undergo the treatment. Foremost, the trend of improvement remains constant. HBO₂ conditions may not be statistically larger than control conditions within individual studies, but the treatment effect appears to be consistently larger than placebo-controlled improvements across multiple studies. This notion can be difficult to identify since most studies use different scales to measure outcomes and not all studies report effect size improvements in their investigations, which would help clarify the clinical significance of the treatment effect. However, effect sizes could be determined from published studies based on means and standard deviations, and this possibility can be tested statistically in review rather than offered only as a presumption. In addition, although study sample sizes tend to be quite small, observed improvements tend to be quite large. Individuals often report substantial improvements following HBO₂ treatments that should be considered when evaluating clinical significance. This issue can also be addressed by comparing the effect size of any treatment across multiple studies rather than trying to limit general inferences to statistically significant differences between groups within a given study. Finally, studies often use cognitive tests, symptomatic self-reports, or a combination of both. Cognitive studies are not immune to placebo effects (26, 27), but self-reported measures are certainly vulnerable to placebos. If anything, the definition of placebo is reporting improvement because the individual believed there should be a benefit. So, cognitive changes compared with self-reported symptomatic changes could help identify whether the observed improvements could be attributed, in part or entirely, to patients succumbing to a placebo effect without any real underlying benefit.

The current analysis took a different approach than previous reviews to exploring the results across multiple studies. Specifically, although not every study reported effect sizes and different studies use different scales, HBO₂ interventions do regularly report means and standard deviations (or standard errors) along with their statistical comparisons. These data allow for a Cohen’s d to be calculated from simple pretest and posttest differences regardless as to whether the individual was assigned to the control group or the active group or the measurement being used. In this way, effect size comparisons allow analyses across studies that can account for placebo effect sizes among control groups and treatment effect sizes among active treatment groups. Moreover, observed improvements can be broadly categorized across cognitive improvements to reaction times and memory versus symptomatic improvements to issues such as headaches and disorientation.

Several hypotheses offer insight into the efficacy of HBO₂ treatments when viewing the data this way. If there is a robust and reliable treatment effect, then it should be consistent across studies, and larger effect sizes should be observed among active treatment conditions than placebo or control conditions. If placebo effects better explain the observed improvements, then effect sizes should be comparable between control conditions and active treatment conditions. For the type of improvement, it is possible that placebo effects are larger among self-reported symptomatic improvements relative to cognitive assessments. Each option provides a different potential outcome when exploring the effect sizes associated with control conditions versus treatment conditions, which has implications for clinical significance in addition to statistical significance.

METHODS AND INCLUSION CRITERIA

Based on these ideas, we reviewed HBO₂ studies explicitly pertaining to mTBI or TBI. Our search involved entering the terms hyperbaric oxygen therapy and traumatic brain injury into PubMed for the years between 2000 and 2020. The search returned 193 results. These studies were further searched for investigations with the following: 1) at least 10 participants to exclude case studies and controlled series of case studies, 2) only studies involving human participants, and 3) only studies with either a cognitive or a symptomatic component. For example, one study primarily derived its evidence from eye tracking (28), and although there were cognitive elements to these tasks, eye tracking as a primary source of evidence was not included. These limitations left 12 studies meeting the inclusionary criteria (Table 1).

Table 1.

Studies analyzed for effect size differences before and after control or treatment conditions

Study (reference)	Cognitive		Symptomatic
	Control	Treatment	Control	Treatment
Lin et al. (29)	–	–	0.60	0.91
Harch et al. (30)	–	0.53	–	1.27
Wolf et al. (31)	0.98	1.18	1.25	1.42
Boussi-Gross et al. (32)	0.11	0.68	0.53	1.31
Cifu et al. (11)	–	–	0.06	0.38
Miller et al. (12)	−0.34	0.23	0.12	0.41
Harch et al. (33)	–	0.54	–	1.10
Tal et al. (34)	–	0.99	–	–
Weaver et al. (35)	−0.02	0.30	−0.43	0.14
Hadanny et al. (36)	–	0.41	–	–
Mozayeni et al. (37)	–	0.65	–	0.72
Harch et al. (38)	0.42	0.81	0.18	1.34
Average	0.23	0.63	0.33	0.90

Effect sizes are divided into either cognitive or symptomatic assessments and whether the assessment involved the control condition or the treatment condition. Scores indicate an effect size as calculated by Cohen’s d, which can be broadly interpreted as a small effect (d = 0.2), a medium effect (d = 0.5), or a large effect (d = 0.8). Dashes indicate the condition was not present in that study.

Data from each study were explored to identify means and standard deviations among treatment and control conditions present in the study, if any. Several studies included an intermediate condition between the full HBO₂ treatment and placebo/sham conditions. These intermediate conditions were not included in the effect size analyses. Full details regarding how effect sizes were calculated from each study are available in appendix. Measures were divided among cognitive assessments and symptomatic assessments. From each reported measure, a Cohen’s d was calculated based on the pretest and posttest difference. The current analysis will use the typical guidance for small effects (d = 0.2), medium effects, (d = 0.5), and large effects (d = 0.8) when comparing effect sizes across these studies (39). The analyses yielded 32 effect sizes from the collected studies: five results from cognitive control assessments, 10 results from cognitive HBO₂ assessments, seven results from symptomatic control assessments, and 10 results from symptomatic HBO₂ assessments.

RESULTS

All effect sizes were entered into a univariate analysis of variance (ANOVA) with the dependent variable of effect size (Cohen’s d) and the independent variable of assessment condition (cognitive-control, cognitive-treatment, symptomatic-control, symptomatic-treatment). There was a significant effect of the treatment condition, F(3,28) = 3.62, P = 0.03, η_p² = 0.28, which justifies additional comparisons. Symptomatic differences between treatment and control assessments were significantly different [mean difference = 0.57, SE = 0.24, t(15) = 2.36, P = 0.03, 95% confidence interval (CI): 0.06–1.09], indicating that effect sizes of improvements for self-reported symptoms were significantly larger among the treatment group than the control group. Cognitive differences between treatment and control assessments were close to being significantly different [mean difference = 0.40, SE = 0.20, t(13) = 1.91, P = 0.07, 95% CI: −0.04 to 0.84]. Across all studies, there does appear to be a robust and significantly different effect size in the treatment condition compared with the control condition.

The initial analyses compared all observed results from across studies, but some of the studies included both a control and treatment group. Additional comparisons were made about effect size differences specifically limiting the input to studies that included both control and treatment groups for paired samples testing. For symptomatic effect sizes, there was a significant difference between treatment and control results [mean difference = 0.51, SE = 0.13, t(6) = 3.88, P < 0.01, 95% CI: 0.19–0.84]. Individuals who underwent HBO₂ reported larger improvements than individuals in the control condition. For cognitive effect sizes, there was a significant difference between treatment and control results [mean difference = 0.41, SE = 0.07, t(4) = 5.69, P < 0.01, 95% CI: 0.21–0.61]. Individuals who underwent HBO₂ exhibited larger improvements in cognitive assessments than in the control condition. Despite the small sample size, the consistency and robust effect sizes produced these significant differences when limiting the analyses only to studies that included both the treatment and control conditions.

A final consideration involved measuring placebo effects. Specifically, what portion of the results could be attributed to placebo effects rather than treatment effects? If treatment results represent the combination of true therapeutic benefit and placebo effects, then the control results could be used to estimate what portion of the treatment results might be attributable to placebo alone. This approach can also help answer whether placebo effects were larger among self-reported symptomatic measures compared with cognitive measures (Fig. 1). Overall, the self-reported symptomatic effect sizes (d = 0.90) were larger than the cognitive effects (d = 0.63). However, using the control results to make estimations, placebo effects could account for a comparable proportion of the treatment outcome in both symptomatic and cognitive measures. This outcome suggests that placebo effects, while accounting for a sizeable contribution to the overall treatment benefit, cannot explain the entirely of the improvement. At least a moderate effect size remains to explain the therapeutic benefit even after accounting for placebo effects.

Figure 1.

Effect size comparisons between cognitive measures and symptomatic measures after accounting for placebo effects. The relative contribution of placebo was determined by taking the observed treatment benefit as the maximum improvement and using improvements observed in the control condition to estimate the placebo-induced benefits.

DISCUSSION

Hyperbaric oxygen therapy (HBO₂) has been proposed as a treatment method for mTBI and TBI. In theory, concussive injuries disrupt cerebral vasculature, and HBO₂, by increasing the content of dissolved oxygen in the blood above physiological levels, allows this oxygen to penetrate deeper into brain tissues than would otherwise be possible with either normobaric or oxygenated conditions alone. This hyperoxygenated environment promotes angiogenesis and neurogenesis that ultimately lead to patient improvement (18, 34). Although there is a potential causal mechanism to explain the benefit, randomized controlled trials have produced mixed evidence where the treatment groups fail to statistically outperform the control groups. Despite statistically nonsignificant differences among individual studies, there remains a consistent pattern of large effect sizes where the treatment group outperforms the control group. The current analyses explored the effect sizes across studies to help address whether placebo effects could account for the bulk of observed therapeutic benefits.

When exploring the changes through effect size metrics, there are consistent benefits of treatment conditions above and beyond the control conditions. These improvements occur for both symptomatic measures and cognitive measures, although symptomatic benefits tend to be larger and more consistent than cognitive effects. It is also worth noting that the proportion of placebo effects from control conditions is comparable across symptomatic and cognitive measures—accounting for about one-third of the treatment benefit. This outcome underscores the pervasive nature of placebo effects in these studies, but it does not discount the sizeable therapeutic benefit that appears to be reliable. Much more work exists to be done in this area, and there are design-related issues that new studies could improve upon (40). Notably, pooled analyses have indicated a possible dose-response trend with benefits increasing alongside increased oxygen partial pressure (25). Establishing dose-response curves and conditions for HBO₂ interventions will be another critical step in establishing a clinically effective and reliable treatment regimen.

There is an additional possible use for hyperbaric oxygen therapy that has not been directly addressed here, although it should be mentioned given the potential comorbidity with mTBI/TBI. Specifically, several researchers have made claims that HBO₂ can help with posttraumatic stress disorder (PTSD) in addition to alleviating problems associated with concussive injuries (5, 41, 42). The review here primarily focused on concussive-related trauma given the clearer link to a potential causal mechanism in explaining the therapeutic benefit (16–20). High comorbidity between TBI and PTSD (43) creates a challenge in disentangling a causal benefit for one condition and any changes relative to the additional condition. Indeed, one hypothesis for improvement in PTSD due to HBO₂ is the facilitation argument, which suggests that HBO₂ might benefit PTSD simply by alleviating cognitive challenges that impeded general quality of life (21). Especially for military personnel, the combination creates a significant challenge, but without a clear causal link to explain how HBO₂ would produce a placebo-unrelated improvement in psychological condition without a physiological change, the current analysis focused primarily on the concussive injury implications with a clearer potential causal mechanism to explain the improvement.

CONCLUSIONS

Clinical significance should always be a consideration for any medical intervention. Placebo effects complicate any assessment of clinical significance because they suggest perceived improvements that may lack durable therapeutic benefits. HBO₂ may provide another method to treat brain injuries, and effect size analyses suggest that at least a sizeable portion of the findings can be attributed to treatment-related improvements rather than placebo effects. Still, additional evidence will need to be collected with tighter experimental designs. Among these future data collections, increased focus on potential dose-response curves would indicate a reliable clinical improvement and provide much needed guidance for designing active protocols to treat brain injuries.

APPENDIX

This appendix provides additional detail on the calculations pertaining to each effect size as presented in the main text. This context is necessary given the classification of some effects and calculations of different subscales. Each reference is provided in full before the effect size calculation information. The number represents the reference within the main text. References are ordered chronologically in the appendix.

29. Lin JW, Tsai JT, Lee LM, Lin CM, Hung CC, Hung KS, Chen WY, Wei L, Ko CP, Su YK, Chiu WT. Effect of hyperbaric oxygen on patients with traumatic brain injury. Acta Neurochir Suppl 101: 145–149, 2008.

Primary data observed in Table 2, p. 146. There were means for the Glasgow Coma Scale (GCS) for each group but not standard deviations. So, a standard deviation was created using the three scores within each group (GCS mean upon arrival, GCS mean before, and GCS mean after) to create a standard deviation. This approach yielded a standard deviation of 2.75 for the HBO₂ group and a standard deviation of 1.84 for the control group. A t test was conducted to ensure that this assessment could sufficiently replicate the P < 0.05 finding reported in the Results section. These standard deviations were significantly different, t(42) = 1.98, P = 0.05, which suggests that the standard deviations constructed from the available information provide a reasonably accurate estimation. The end result produces a mean difference of 1.1 for the control group and 2.5 for the treatment group with effect sizes of d = 0.60 and d = 0.91, respectively. No cognitive data were available.

30. Harch PG, Andrews SR, Fogarty EF, Amen D, Pezzullo JC, Lucarini J, Aubrey C, Taylor DV, Staab PK, Van Meter KW. A phase I study of low-pressure hyperbaric oxygen therapy for blast-induced post-concussion syndrome and post-traumatic stress disorder. J Neurotraum 29: 168–185, 2012.

Cognitive outcome variables were reported with means and standard deviations in Table 5. Relative effect size improvements were determined for each individual neuropsychological task (IQ: d = 1.57; delayed memory: d = 0.64; Rivermead paragraph: d = 0.65; working memory: d = 0.74; Stroop interference: d = 0.88; TOVA inattention: d = 0.09; TOVA impulsivity: d = 0.37; TOVA RT: d = 0.32; TOVA variability: d = 0.41; finger tap dominant hand: d = 0.46; finger tap nondominant: d = 0.17; grooved pegboard dominant hand: d = 0.41; grooved pegboard nondominant hand: d =0.15). These values produce an average Cohen’s d of 0.53 for cognitive outcomes.

Additional analyses were conducted for symptomatic outcomes in psychological conditions. Psychological outcome variables were reported with means and standard deviations in Table 6. Relative effect size improvements were determined for each psychological outcome variable (Rivermead PCS: d = 1.58; PCL-M: d = 1.50; PHQ-9: d = 1.75; GAD-7: d = 0.86; perceived QOL: d = 0.90; % back to normal cognitive: d = 1.03; % back to normal physical: d = 1.02; % back to normal emotional: d = 1.53). These scores averaged together for a psychological outcome improvement of d = 1.27. Psychological outcomes were included within the symptomatic or self-report measures.

31. Wolf G, Cifu D, Baugh L, Carne W, Profenna L. The effect of hyperbaric oxygen on symptoms after mild traumatic brain injury. J Neurotrauma 29: 2606–2612, 2012.

There were two primary dependent variables: PCL-M composite score and ImPACT total scores. Precise values were not available and had to be interpreted from the available information. For PCL-M scores, means were taken from Table 1 and provided precise values, but standard deviations had to be interpreted from standard error bars as shown in Fig. 1. These combined data points yielded the following information for the control group at pretest (mean = 48.9, SE = 1.4, N = 24) versus posttest mean = 40.6, SE = 1.3, N = 24), which produces a Cohen’s d of 1.25. A similar procedure yielded the following information for the investigational group at pretest (mean = 50.0, SE = 1.3, N = 24) versus posttest (mean = 41.6, SE = 1.1, N = 24), which produces a Cohen’s d of 1.42.

ImPACT scores were used as cognitive data since that assessment includes multiple cognitive tasks. However, complete data were not available. Means and standard deviations had to be interpreted from standard error bars as shown in Fig. 1. These combined data points yielded the following information for the control group at pretest (mean = 38.5, SE = 2.5, N = 24) versus posttest (mean = 26.5, SE = 2.5, N = 24), which produces a Cohen’s d of 0.98. A similar procedure yielded the following information for the investigational group at pretest (mean = 37.0, SE = 0.80, N = 24) versus posttest (mean = 32.5, SE = 0.75, N = 24), which produces a Cohen’s d of 1.18.

32. Boussi-Gross R, Golan H, Fishlev G, Bechor Y, Volkov O, Bergan J, Friedman M, Hoofien D, Shlamkovitch N, Ben-Jacob E, Efrati S. Hyperbaric oxygen therapy can improve postconcussion syndrome years after mild traumatic brain injury-randomized prospective trial. PloS ONE 8: e79995, 2013.

The Boussi-Gross et al. (32) design was a randomized, controlled trial with a crossover control group; that is, the control group received no treatment while the primary active group received treatment, but the control group did receive treatment afterward following the posttreatment evaluation. The study recorded two primary types of evidence relevant for the current review: cognitive functions and quality of life.

Cognitive functions included information processing speed, attention, memory, and executive function. The treated group exhibited significant improvement on all scores (information processing speed: d = 0.74; attention: d = 0.57; memory: d = 0.73; and executive functions: d = 0.66). These effect size statistics were reported in the main text (p. 5), and the average cognitive improvement is evaluated as d = 0.68. The control group did not exhibit any significant improvement during the control period, although effect sizes were not given in the main text. Effect sizes for the control group during the control period were not included in the main text. Instead, these metrics were calculated from the reported t and P values as well as the data presented in Fig. 2. Using this information, the control group effect sizes were calculated to reflect the observed effect sizes for the nonsignificant statistics (information processing speed: d = 0.11; attention: d = 0.07; memory: d = 0.15; and executive functions: d = 0.11). The control group was thus evaluated as having an average effect size of d = 0.11 during the control period. It is also worth noting that the crossover control group did receive the treatment intervention after the initial control period. There was significant observed improvement following the treatment (information processing speed: d = 0.40; attention: d = 0.47; memory: d = 0.65; and executive functions: d = 0.46). The average improvement following the control period was thus d = 0.50.

The second relevant assessment was on quality of life (QOL). This assessment included both the EQ-5D that evaluated mobility, self-care, usual activities, pain/discomfort, and anxiety/depression as well as the EQ-VAS that evaluated respondent’s self-rated health on a visual analog scale. There were no effect sizes reported in the main text, and so, effect sizes were calculated for the EQ-5D from t values, P values, and sample sizes available in the main text. There was a large observed effect size for the treatment group during the control period (d = 1.31), a moderate effect size observed for the control group during the control period (d = 0.53), and a large effect size observed for the crossover control group following the secondary treatment period after the control period (d = 1.26).

11. Cifu DX, Hart BB, West SL, Walker W, Carne W. The effect of hyperbaric oxygen on persistent postconcussion symptoms. J Head Trauma Rehabil 29: 11–20, 2014.

Cifu et al. (11) utilized a three-condition design where all conditions involved 2.0 ATA but with different inspired oxygen levels to control for simulated exposure via dissolved oxygen content described as sham, 1.5 ATA, and 2.0 ATA. The primary scales used were the Rivermead Post-Concussive Symptom Questionnaire (RPQ) and PCL-M. Item means were provided for both via Table 1 and Table 2, although no standard deviations were provided. To calculate a standard deviation, the items were average rather than summed, and the standard deviation was taken according to the standard deviation associated with the average not the sum. Similar t tests were conducted as with the reported results to determine whether the results would be comparable using this alternative measure of calculating score via item average rather than item sum. The results were highly similar, and so, these numbers were used to approximate an effect size for the analyses. RPQ scores were separated first as both the RPQ-3 and RPQ-13.

For the RPQ-3, means and standard deviations were found for the pretest analyses of the sham group (mean = 1.74, SD = 1.08), 1.5 ATA equivalent group (mean = 1.68, SD = 1.21), and 2.0 ATA equivalent group (mean = 1.53, SD = 1.17). Means and standard deviations were found for the posttest analyses of the sham group (mean = 1.70, SD = 0.96), 1.5 ATA equivalent group (mean = 1.73, SD = 1.08), and 2.0 ATA equivalent group (mean = 1.33, SD = 1.03). These scores produced Cohen’s d pretest to posttest analyses for the sham group (d = 0.05), 1.5 ATA equivalent group (d = 0.04), and 2.0 ATA equivalent group (d = 0.18).

For the RPQ-13, means and standard deviations were found for the pretest analyses of the sham group (mean = 2.12, SD = 0.73), 1.5 ATA equivalent group (mean = 1.87, SD = 0.82), and 2.0 ATA equivalent group (mean = 1.99, SD = 0.95). Means and standard deviations were found for the posttest analyses of the sham group (mean = 2.14, SD = 0.64), 1.5 ATA equivalent group (mean = 1.95, SD = 0.79), and 2.0 ATA equivalent group (mean = 1.74, SD = 0.80). These scores produced Cohen’s d pretest to pos-test analyses for the sham group (d = −0.02), 1.5 ATA equivalent group (d = −0.11), and 2.0 ATA equivalent group (d = 0.28).

For the PCL-M, means and standard deviations were found for the pretest analyses of the sham group (mean = 2.65, SD = 0.59), 1.5 ATA equivalent group (mean = 2.63, SD = 0.61), and 2.0 ATA equivalent group (mean = 2.90, SD = 0.57). Means and standard deviations were found for the posttest analyses of the sham group (mean = 2.58, SD = 0.57), 1.5 ATA equivalent group (mean = 2.55, SD = 0.53), and 2.0 ATA equivalent group (mean = 2.51, SD = 0.57). These scores produced Cohen’s d pretest to posttest analyses for the sham group (d = 0.13), 1.5 ATA equivalent group (d = 0.14), and 2.0 ATA equivalent group (d = 0.69).

Taken together, these scores were averaged into final symptom improvement effect sizes for the HBO₂ group (d = 0.38), the sham group (d = 0.05), and the 1.5 ATA equivalent group (d = 0.00).

12. Miller RS, Weaver LK, Bahraini N, Churchill S, Price RC, Skiba V, Caviness J, Mooney S, Hetzell B, Liu J, Deru K. Effects of hyperbaric oxygen on symptoms and quality of life among service members with persistent postconcussion symptoms: a randomized clinical trial. JAMA Intern Med175: 43–52, 2015.

Primary data included here come from the RPQ scores as reported in Table 2. Although the data are separated into intent-to-treat population and per-protocol population, only the intent-to-treat numbers are included in the analyses. The per-protocol data reflect a subset of the total data, and the authors report the intent-to-treat data as their primary finding in the abstract. Cognitive data are taken from the Automated Neuropsychological Assessment Metrics (ANAM) data as reported in Table 4.

For the RPQ data, means and standard deviations were found for the pretest analyses of the standard care group (mean = 32.5, SD = 14.4), active treatment group (mean = 33.0, SD = 15.8), and sham treatment group (mean = 30.2, SD = 14.2). Means and standard deviations were found for the posttest analyses of the standard care group (mean = 30.6, SD = 16.1), active treatment group (mean = 26.7, SD = 14.8), and sham treatment group (mean = 24.2, SD = 15.4). These scores produced Cohen’s d pretest to posttest analyses for the standard care group (d = 0.12), active treatment group (d = 0.41), and sham treatment group (d = 0.41).

For the ANAM data, data reported in Table 4 represented throughput scores, and an effect size improvement was calculated for each subscale based on the mean change score and associated standard deviation.

In the standard care group from pretest to posttest, the simple reaction time test showed a small decline (mean change: −15.0, SD = 25.2, Cohen’s d = −0.60), procedural reaction time showed a small decline (mean change: −4.50, SD = 20.20, Cohen’s d = −0.22), code substitutional learning showed a small decline (mean change: −5.40, SD = 13.20, Cohen’s d = −0.41), mathematical processing showed a small decline (mean change: −2.60, SD = 7.60, Cohen’s d = −0.34), and matching to sample showed a small decline (mean change: −1.30, SD = 8.70, Cohen’s d = −0.15).

In the active treatment group from pretest to posttest, the simple reaction time test showed a small decline (mean change: 1.70, SD = 27.80, Cohen’s d = 0.06), procedural reaction time showed a small decline (mean change: 7.20, SD = 24.10, Cohen’s d = 0.30), code substitutional learning showed a small decline (mean change: 2.90, SD = 20.30, Cohen’s d = 0.14), mathematical processing showed a small decline (mean change: 3.90, SD = 21.40, Cohen’s d = 0.18), and matching to sample showed a small decline (mean change: 6.50, SD = 13.60, Cohen’s d = 0.48).

In the sham treatment group from pretest to posttest, the simple reaction time test showed a small decline (mean change: 1.70, SD = 35.20, Cohen’s d = 0.05), procedural reaction time showed a small decline (mean change: −8.70, SD = 23.60, Cohen’s d = −0.37), code substitutional learning showed a small decline (mean change: -1.00, SD = 19.10, Cohen’s d = -0.05), mathematical processing showed a small decline (mean change: 8.10, SD = 25.40, Cohen’s d = 0.32), and matching to sample showed a small decline (mean change: 0.40, SD = 22.60, Cohen’s d = 0.02).

These scores produced average Cohen’s d pretest to posttest analyses for the standard care group (d = −0.34), active treatment group (d = 0.23), and sham treatment group (d = −0.01).

33. Harch PG, Andrews SR, Fogarty EF, Lucarini J, Van Meter KW. Case control study: hyperbaric oxygen treatment of mild traumatic brain injury persistent post-concussion syndrome and post-traumatic stress disorder. Med Gas Res 7: 156, 2017.

This study included both an active treatment group and a control group for comparison. However, control participants did not complete the neuropsychological and self-report symptoms test batteries, as their tests were focused on single photo emission computed tomography (SPECT) brain imaging. Given the parameters of this analysis, we examined the neuropsychological changes reported in Table 4 of their manuscript and divided the outcomes into two categories: cognitive and symptomatic. Table 4 provided means and standard deviations for these values. The change score was used to create a d value for each test.

Cognitive tests included the following assessments with an effect size calculated for each: Wechsler Adult Intelligence (WAIS)-IV Full Scale IQ test (d = 1.38), WAIS-IV delayed memory(d = 1.09), WAIS-IV working memory (d =1.08), Stroop C/W interference (d = 0.89), Test of Variables of Attention (TOVA; inattention, d = 0.29; impulsivity, d = 0.59; reaction time, d = 0.19; variability of RT, d = 0.34), finger tap (dominant, d = 0.40; nondominant, d = 0.31), and grooved peg board (dominant, d = 0.81; and nondominant, d = 0.31). These combined metrics were averaged together for a cognitive improvement of d = 0.54.

Symptomatic tests included the Generalized Anxiety Disorder Scale-7 (d = 1.38), Personal Health Questionnaire-9 (d = 1.38), PCL-M (d = 1.38), and Rivermead Post-Concussion Symptom Questionnaire (d = 1.38). These combined metrics were averaged together for a symptomatic improvement of d = 1.10.

34. Tal S, Hadanny A, Sasson E, Suzin G, Efrati S. Hyperbaric oxygen therapy can induce angiogenesis and regeneration of nerve fibers in traumatic brain injury patients. Front Hum Neurosci 11: 508.

This study included an active treatment group with no control group. Participants completed a cognitive assessment and MRI scans. Due to the unique nature of the MRI scans relative to the rest of the HBO₂ literature, effect size analyses focused solely upon the cognitive tests. Participants completed NeuroTrax computerized cognitive tests. Data from these cognitive tests were presented in Table 2 of the manuscript. From this data, the following effect sizes were determined for each cognitive test: global (d = 1.41), memory (d = 1.13), executive functions (d = 1.08), attention (d = 0.53), IPS (information processing speed; d = 1.25), VSP (visual spatial processing; d = 0.75), and motor skills (d = 0.75). These combined scales indicated an average improvement on the cognitive tests of d = 0.99.

35. Weaver LK, Wilson SH, Lindblad AS, Churchill S, Deru K, Price RC, et al. Hyperbaric oxygen for post-concussive symptoms in United States military service members: a randomized clinical trial. Undersea Hyperb Med 45: 129–156, 2018.

Participants completed numerous test batteries as part of the BIMA study. Although participants completed multiple assessments after 13 wk and again 6 mo later, the focus of these analyses will be upon the 13-wk, posttreatment assessment to better compare with the other results.

Cognitive tests included the neuropsychological assessments as reported in Table 5. These tests included the Weschler Test of Adult Reading (N = 1; HBO₂: d = 0.05, sham: d = 0.03), Test of Memory Malingering (N = 1; HBO₂: d = 0.33, sham: d = 0.33), ANAM scales (N = 7; HBO₂: d = 0.40, sham: d = 0.10), California Verbal Learning Test-II (N = 7; HBO₂: d =0.30, sham: d = −0.11), and the Brief Visuospatial Memory Test-Revised (N = 6; HBO₂: d = 0.20, sham: d = −0.11). When averaged across the individual subscales as reported in Table 5, the average effect size of improvement was d = 0.30 for the HBO₂ group and d = −0.02 for the sham group.

Symptomatic results were taken from Table 3 data that included the Post-traumatic Stress Disorder Checklist Civilian Version (PCL-C and Rivermead Post-Concussive Symptom Questionnaire (RPQ). Other symptomatic inventories were available, but these two were best represented among the other studies being reviewed. Note that effect size scores are aligned with cognitive scores such that a positive value indicates an improvement, which for symptoms would be a decline from baseline to posttreatment. As such, a positive d value is a beneficial effect.

PCL-C scores improved for the treatment group relative to the sham group for both PTSD (HBO₂: d = 0.60, sham: d = −0.32) and no-PTSD subgroups (HBO₂: d = −0.06, sham: d = −0.41). RPQ-3 scores improved for the treatment group relative to the sham group for both PTSD (HBO₂: d = 0.06, sham: d = −0.61) and no-PTSD subgroups (HBO₂: d = 0.17, sham: d = −0.41). RPQ-13 scores improved for the treatment group relative to the sham group for both PTSD (HBO₂: d = 0.06, sham: d = −0.39) and no-PTSD subgroups (HBO₂: d = −0.02, sham: d = −0.43). These combined scores averaged to a d = 0.14 for the HBO₂ treatment group and d = −0.43 for the sham group.

36. Hadanny A, Abbott S, Suzin G, Bechor Y, Efrati S. Effect of hyperbaric oxygen therapy on chronic neurocognitive deficits of post-traumatic brain injury patients: retrospective analysis. BMJ Open. 8: e023387, 2018.

Cognitive functioning was assessed using the NeuroTrax system, and the raw data were presented in Table 2. The paper did not assign the reported values as means with standard deviations or standard errors. However, given the sample size and the reported P values, it was assumed that the information in Table 2 represented standard deviations; otherwise, the analyses would not have been significant.

Cognitive data were reported on seven domains: general (mean = 4.6, SD = 8.5, d = 0.54), memory (mean = 8.1, SD = 16.9, d = 0.48), executive functions (mean = 5.9, SD = 12.0, d = 0.49), attention, (mean = 6.8, SD = 16.5, d = 0.41), information processing speed (mean = 4.9, SD = 13.1, d = 0.37), visual spatial (mean = 3.4, SD = 14.6, d = 0.23), and motor skills (mean = 3.9, SD = 11.7, d = 0.33). These metrics averaged together for an effect size of d = 0.41. No symptomatic data were reported.

37. Mozayeni BR, Duncan W, Zant E, Love TL, Beckman RL, Stoller KP. The National Brain Injury, Rescue and Rehabilitation Study—a multicenter observational study of hyperbaric oxygen for mild traumatic brain injury with post-concussive symptoms. Med Gas Res 9: 1–12, 2019.

Participants completed both the Automated Neuropsychological Assessment Metrics (ANAM4) and the Central Nervous System Vital Signs (CNSVS) batteries as neurocognitive tests. Effect sizes could not be determined from the CNSVS, as the means were not presented with standard deviations or standard errors. As such, the focus was on the ANAM reporting, which included both mood scores and cognitive tests. Symptomatic scores were determined based on ANAM4 mood scale changes, whereas cognitive scores were determined based on ANAM4 cognitive assessments.

For symptomatic scores, the ANAM4 data included scales with different directional improvement. For example, a positive outcome would be an increase in happiness but a decrease in anger. Scores were recoded to be unidirectional so that a positive number indicated improvement. These scores included the following scales: sleepiness (mean = 1.13, SD = 1.34, d = 0.84), vigor (mean = 20.8, SD = 24.2, d = 0.86), restlessness (mean = 15.2, SD = 21.9, d = 0.69), depression (mean = 12.5, SD = 21.5, d = 0.58), anger (mean = 9.8, SD = 21.7, d = 0.45), happiness (mean = 19.3, SD = 23.2, d = 0.83), fatigue (mean = 23.8, SD = 25.1, d = 0.95), and anxiety (mean = 10.6, SD = 20.5, d = 0.52). These combined scales averaged together for an improvement of d = 0.72.

For cognitive scores, the ANAM4 data included seven scales: simple reaction time (mean = 26, SD = 32, d = 0.81), code substitution learning (mean = 21.2, SD = 27.2, d = 0.78), procedural reaction time (mean = 30, SD = 38, d = 0.79), mathematical processing (mean = 23.5, SD = 25.8, d = 0.91), matching to sample (mean = 13, SD = 31, d = 0.42), code substitution delayed (mean = 20, SD = 34, d = 0.59), and simple reaction time repeated (mean = 8, SD = 36, d = 0.22). These seven scales combined together for an average improvement of d = 0.65.

38. Harch PG, Andrews SR, Rowe CJ, Lischka JR, Townsend MH, Yu Q, Mercante DE. Hyperbaric oxygen therapy for mild traumatic brain injury persistent postconcussion syndrome: a randomized controlled trial. Med Gas Res 10: 8–20, 2020.

The authors conducted a thorough battery of tests on participants. Means and standard deviations were assessed for the primary treatment group and control group during the primary treatment period as presented in Table 3. Standard deviations were only reported on mean differences for the change scores. To calculate an effect size for the change within the control group and treatment group, respectively, the mean score standard deviation was used, as the authors did not describe an expected difference in standard deviations between changes scores for each group. Control group participants did not receive a sham treatment, nor did they receive the active treatment while the primary treatment group was being treated. The study also included a follow-up treatment condition for the control group. However, these analyses used the treatment versus control as presented in Table 3 without analyzing the control follow-up treatment of the dropout groups. For this analysis, the tests were divided into either a symptomatic or a cognitive category. Symptomatic scores were recoded so that a positive difference indicated improvement.

For the control group, the symptomatic scores included the following scales: Neurobehavioral Symptom Inventory (mean = 2.5, SD = 9.22, d = 0.27), Hamilton Depression Scale (mean = 1.6, SD = 6.85, d = 0.23), Hamilton Anxiety Scale (mean = 1.1, SD = 7.48, d =0.15), Quality of Life after Brain Injury (mean = 2.0, SD = 14.9, d = 0.13), Pittsburgh Sleep Quality Index (mean = 0.4, SD = 3.64, d = 0.11), and Post-traumatic Stress Disorder Check List (mean = 2.2, SD = 11.2, d = 0.20). These six scales combined together for an average improvement of d = 0.18 in the control group on symptomatic evaluations. Cognitive scores for the control group included the following scales: working memory index (mean = 6.0, SD = 6.5, d = 0.92), memory index (mean = 4.7, SD = 8.6, d = 0.55), information processing speed index (mean = 5.3, SD = 9.4, d = 0.56), executive function index (mean = −0.3, SD = 5.8, d = −0.05), Wechsler Adult Intelligence Scale Full Intelligence Quotient (mean = 4.5, SD = 5.76, d = 0.78), ANAM composite score (mean = 0.30, SD = 0.64, d = 0.47), Benton Visual Retention Test (mean = 0.3, SD = 1.72, d = 0.17), and Rey Auditory Verbal Learning Test Delay Recall (mean = −0.1, SD = 11.9, d = −0.01). These eight scales combined together for an average improvement of d = 0.42 in the control group for cognitive evaluations.

For the treatment group, the symptomatic scores included the following scales: Neurobehavioral Symptom Inventory (mean = 26.3, SD = 9.22, d = 2.85), Hamilton Depression Scale (mean = 7.7, SD = 6.85, d = 1.12), Hamilton Anxiety Scale (mean = 7.2, SD = 7.48, d = 0.96), Quality of Life after Brain Injury (mean = 18.2, SD = 14.9, d = 1.22), Pittsburgh Sleep Quality Index (mean = 2.9, SD = 3.64, d = 0.80), and Post-traumatic Stress Disorder Check List (mean = 11.9, SD = 11.2, d = 1.06). These six scales combined together for an average improvement of d = 1.34 in the treatment group on symptomatic evaluations. Cognitive scores for the control group included the following scales: working memory index (mean = 7.5, SD = 6.5, d = 1.15), memory index (mean = 11.6, SD = 8.6, d = 1.35), information processing speed index (mean = 8.5, SD = 9.4, d =0.90), executive function index (mean = −1.7, SD = 5.8, d =0.29), Wechsler Adult Intelligence Scale Full Intelligence Quotient (mean = 6.6, SD = 5.76, d =1.15), ANAM composite score (mean = 0.82, SD = 0.64, d = 1.28), Benton Visual Retention Test (mean = 0.0, SD = 1.72, d = 0.0), and Rey Auditory Verbal Learning Test Delay Recall (mean = 4.5, SD = 11.9, d = 0.38). These eight scales combined together for an average improvement of d = 0.81 in the treatment group for cognitive evaluations.

DISCLAIMERS

The views expressed in this article are those of the authors and do not necessarily reflect the official policy or position of the Department of the Navy, Department of Defense, or the US Government. The authors are military service members or employees of the US Government. This work was prepared as part of their official duties. Title 17 U.S.C. §105 provides that “Copyright protection under this title is not available for any work of the United States Government.” Title 17 U.S.C. §101 defines a US Government work as a work prepared by a military service member or employee of the US Government as part of that person’s official duties.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.T.B., H.M.D., and L.F.L. conceived and designed research; A.T.B. analyzed data; A.T.B. prepared figures; A.T.B., H.M.D., and L.F.L. drafted manuscript; A.T.B., H.M.D., and L.F.L. edited and revised manuscript; A.T.B., H.M.D., and L.F.L. approved final version of manuscript.