Effects of Acute Intermittent Hypoxia on Sensory Function in Healthy Adults: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Effects of Acute Intermittent Hypoxia on Sensory Function in Healthy Adults (NCT03283072), a randomized na study. The primary outcome is Sensory Function.

2. Clinical Question

3. Trial Design

This is a randomized, double-masked crossover study listed as completed. Its title identifies healthy adults, and the registry reports 22 participants enrolled. The sponsor is the University of Florida. The recorded start date is 2017-09-18 and the primary completion date is 2020-07-06. The phase is listed as NA.

A crossover design means that the same participants can receive more than one study condition, with the order organized into sequences. The registry lists four sequences. Each sequence includes Intervention 1, Intervention 2, Intervention 3, and the sham condition, but the order differs. This structure makes within-participant comparisons potentially relevant: measurements from the same person may be more alike than measurements from different people. The registry does not report sequence-specific results, period effects, carryover assessments, or a washout interval.

The three active conditions differ in the stated number and duration of hypoxia:hyperoxia bouts. The sham is described as bouts of normoxia. All four are identified as device interventions. The registry does not give further device details, exposure monitoring information, or a description of how participants and study staff were kept masked.

The primary outcome is described as thermal thresholds, measured in degrees celsius. Measurements were taken every 10 minutes for 60 minutes post intervention, with an average reported. The outcome record says a two-way analysis of covariance with repeated measures was conducted, with time and intervention as factors and gender as a covariate. More detail on the model specification, handling of repeated measurements, and analysis populations is not reported in the registry record.

The registry posts 20 observations for each intervention condition, while the enrollment count is 22. Because this is a crossover study, those condition counts should not be read as four separate groups of 20 people. The record does not state why the posted counts differ from total enrollment or whether every participant contributed data under every condition.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Overall Study

Analysis population for the primary outcome, as defined in the registry: A 2-way analysis of co-variance with repeated measures (time\*intervention) was conducted with gender as the covariate.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

6. Primary Endpoint

The registry lists the primary outcome as Sensory Function, measured over Every 10 minutes for 60 minutes post intervention, average reported.

The primary endpoint is sensory function assessed through thermal thresholds, expressed in degrees celsius. The registry says that the measure is an average and that measurements were taken every 10 minutes for 60 minutes after each intervention. The reported arm values are means, with standard deviations as the measure of spread. The registry does not provide additional detail about the specific threshold task, the meaning of a change in degrees celsius, or a clinically important difference.

A thermal threshold is an operational measure: it translates a sensory response to a thermal stimulus into a value that can be compared across conditions. Using a numerical measure allows investigators to describe the level and variability of responses and to estimate differences between intervention conditions. Averaging measurements across the stated post-intervention window gives a summary of sensory measurements during that period, rather than a separate reported value for every time point.

The outcome is relevant to the trial’s stated aim because the registry describes the study as examining sensory effects of acute intermittent hypoxia. The endpoint can therefore address whether the average reported thermal-threshold values differ after the listed exposure conditions and sham. It does not by itself explain the biological process underlying any difference, establish whether a difference matters to participants, or describe effects beyond the post-intervention period. Those interpretations require information and evidence not reported in the registry record.

The registry describes the outcome as continuous and reports a mean for each condition. A mean gives a central summary, while the standard deviation describes variation among the observations contributing to that summary. Neither statistic alone describes the shape of the individual responses or the degree to which measurements from the same participant are associated across the crossover conditions.

7. Results Posted in the Registry

• Intervention 1: mean 46.70 versus 45.90 on control; difference 0.80 (95% CI -0.33 to 1.93), standardized effect d = 0.44; z-test p = 0.166.
• Intervention 2: mean 46.40 versus 45.90 on control; difference 0.50 (95% CI -0.74 to 1.74), standardized effect d = 0.25; z-test p = 0.430.
• Intervention 3: mean 46.10 versus 45.90 on control; difference 0.20 (95% CI -1.04 to 1.44), standardized effect d = 0.10; z-test p = 0.752.

8. Statistical Interpretation

9. Statistical Methods Explained

The registry describes the primary outcome analysis as a two-way analysis of covariance with repeated measures. Time and intervention are named as the factors, and gender is named as a covariate. The sponsor’s posted analysis is labeled ANCOVA and reports p = 0.592 for the four groups. The record does not give enough detail to reconstruct the model or determine which model term the reported p-value tests.

What does an analysis of covariance do?

In general, analysis of covariance compares outcome levels while accounting for one or more covariates. A covariate is a variable included in the model to account statistically for variation associated with that variable. Here, gender is the covariate named in the registry description. Adjusting for a covariate can change an estimated contrast and its uncertainty compared with an unadjusted comparison. The registry does not report the adjusted means, adjusted differences, coefficients, or uncertainty estimates, so the magnitude and direction of any adjusted contrasts cannot be described from the posted analysis.

Why are time and repeated measures relevant?

Repeated-measures methods address observations that are not independent because multiple measurements can come from the same participant. In general, measurements from one person may share participant-specific features. A model that accounts for this dependence can estimate condition and time patterns more appropriately than a method that treats every measurement as if it came from a different person. The registry identifies time and intervention in a repeated-measures model, but does not state the covariance structure, the exact time coding, or whether an interaction between time and intervention was included and how it was tested. The primary outcome description names time and intervention, while the analysis-population text refers to time*intervention; fuller specification is not reported in the registry record.

In general, an interaction asks whether the relationship between one factor and the outcome varies according to another factor. For this study, a time-by-intervention interaction could address whether the pattern across the post-intervention measurements differs by condition. That is a general explanation of the term, not a claim about which specific result the reported p-value represents. The registry gives no separate interaction estimate or p-value.

What do the posted mean differences represent?

The recalculated comparisons subtract the posted sham mean from each active-condition mean and report unadjusted mean differences. They are based on the arm-level means and standard deviations. They do not use participant-level records, a repeated-measures covariance estimate, or the gender adjustment named for the sponsor’s analysis. Consequently, the estimates and intervals summarize simple, unadjusted comparisons and should not be interpreted as a reproduction of the ANCOVA.

In general, a mean difference is expressed in the outcome’s units and describes the contrast between two means. A positive estimate means that the first listed mean is higher than the comparator mean on the numerical scale. It does not establish that the difference is important or that the intervention caused a meaningful change. The outcome’s clinical interpretation is not defined in the registry.

Why can over-dispersion matter?

Over-dispersion is a general concern when observed variability is greater than a statistical model’s assumptions allow. It can make model-based uncertainty too small if the extra variation is ignored. It is especially relevant to outcomes such as counts, where variation may exceed what a simple count model expects. This primary outcome is continuous and is reported as a mean with a standard deviation; the registry does not report an over-dispersion assessment or identify over-dispersion as an issue. The concept is included as a general statistical consideration, not as a finding about this trial.

What can and cannot be inferred from the p-value?

In general, a p-value compares observed data with what a specified null model would predict, conditional on the model and its assumptions. It does not provide an effect size, a confidence interval, the probability that a finding will replicate, or the probability that the null hypothesis is true. The registry reports p = 0.592 but does not provide the test statistic, degrees of freedom, or a detailed model statement. The result can be reported as the registry’s overall ANCOVA p-value, but its exact estimand and test specification are not reported in the registry record.

The confidence intervals accompanying the recalculated comparisons describe uncertainty under their unadjusted summary-statistic approach. General interpretation of a confidence interval is that it expresses a range of values compatible with the data and the method used to construct it. It is not an interval for every individual participant’s response. The confidence level for the posted recalculated intervals is not reported in the registry record, and the intervals should not be attributed to the sponsor’s ANCOVA.

Finally, the design and the available summaries answer different statistical questions. A repeated-measures crossover analysis can account for within-participant dependence and the specified time structure. The simple recalculations describe differences between posted condition summaries without that information. Both can be informative in their proper roles, but they are not interchangeable. Details such as the analysis set, missing-data handling, model diagnostics, and sequence or period terms are not reported in the registry record.

10. Confidence Intervals and Precision

The unadjusted mean-difference estimates are 0.8 for Intervention 1 versus sham, 0.5 for Intervention 2 versus sham, and 0.2 for Intervention 3 versus sham, all in the outcome’s degrees celsius scale. Their intervals are -0.331 to 1.931, -0.741 to 1.741, and -1.041 to 1.441, respectively. Each interval includes zero, indicating that the summary-statistic calculations do not pin down the direction of the difference with precision.

In general, a confidence interval communicates uncertainty around an estimated contrast. A narrower interval indicates greater precision under the method used; a wider interval indicates that a broader range of values remains compatible with the data and assumptions. The intervals span values on both sides of zero, so they do not isolate a positive or negative contrast. The registry does not report a clinically meaningful threshold for interpreting the size of any difference, so statistical precision cannot be translated into clinical importance from the record alone.

These intervals are recalculated from the posted summary statistics and are unadjusted. They do not use the paired participant-level information that could matter in a crossover study, and they are not the intervals from the sponsor’s ANCOVA. The registry does not provide confidence intervals for the sponsor’s reported overall analysis. Its p-value of 0.592 therefore cannot be paired with an adjusted effect estimate or interval from the posted record.

The arm-level standard deviations are 1.5 for Intervention 1, 1.9 for Intervention 2, 1.9 for Intervention 3, and 2.1 for sham. Standard deviations describe variability in the observations, not precision of the estimated mean difference by themselves. Precision also depends on the amount of information and, for a crossover comparison, on the association between measurements from the same participants. That association is not reported.

These limits mean that the reported estimates should be read as uncertain summaries, not as exact differences. The registry’s missing adjusted intervals and model detail further limit how precisely the sponsor’s analysis can be interpreted.

11. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The planning figures are participants per arm, not total event counts. They are illustrations based on the observed effect and use a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. Each planning scenario uses the stated fraction of the observed effect as the effect to detect. The trial’s original sample-size assumptions and justification are not reported in the registry record. These figures are not observed power.

The 80% power column gives the estimated number of participants per arm for a design with an 80% chance of detecting the assumed effect under the stated planning assumptions, if that effect and the model assumptions apply. The 90% power column gives the corresponding per-arm planning figure for 90% power. These are prospective design illustrations, not descriptions of what the completed study achieved.

For Intervention 1, planning for 100% of the observed effect gives 82 participants per arm for 80% power and 110 participants per arm for 90% power. Planning for 75% of the observed effect gives 146 and 195 participants per arm, respectively. At 50% of the observed effect, the figures are 327 and 438 participants per arm. The corresponding planning figures increase substantially for Intervention 2 and Intervention 3, where the observed mean differences versus sham are smaller.

The general relationship is that smaller effects are harder to distinguish from random variation, so detecting them at the same power generally requires more participants. The displayed scenarios make that trade-off explicit by reducing the assumed effect to 75% and 50% of the observed effect. Because the observed effect is itself estimated with uncertainty, these values should be treated as planning illustrations rather than precise requirements for a future study. They also use equal allocation between two arms, whereas this trial is described as a randomized crossover with four conditions; the planning setup is not a reconstruction of the trial’s actual design assumptions.

12. How the Sample Size Is Derived

Power is the probability that a study of a given size rejects the null hypothesis when a specific true effect exists. It depends on four things: the effect to detect, the variability of the outcome, the significance level (here a two-sided α of 0.05, so z_1−α/2 = 1.960) and the desired power (z_1−β = 0.842 for 80% and 1.282 for 90%).

Formula

n is the number of participants in each arm, Δ is the difference in means to detect and σ is the common standard deviation.

Worked calculation for this trial

1. Observed difference Δ = 0.800; pooled standard deviation σ = 1.825.
2. n = 2 × (1.96 + 0.842)² × 1.825² / 0.800² = 81.7, rounded up to 82 participants per arm for 80% power.

The smaller-effect rows in the table above repeat the same calculation with the effect multiplied by 0.75 and 0.5 (for ratios, applied to the logarithm of the ratio).

R code

alpha <- 0.05
z_a   <- qnorm(1 - alpha / 2)          # two-sided
delta <- 0.800000000000004                      # observed difference in means
sigma <- 1.82482875908947                     # pooled standard deviation
n_per_arm <- function(power, frac = 1) {
  ceiling(round(2 * (z_a + qnorm(power))^2 * sigma^2 / (delta * frac)^2, 6))
}
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: n per arm = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(n_per_arm(0.80, f), big.mark = ","), format(n_per_arm(0.90, f), big.mark = ",")))

# Check against R's built-in function (t-based, so very slightly larger)
cat("power.t.test:", round(power.t.test(delta = delta, sd = sigma, power = 0.80)$n, 1), "\n")

# Simulation: exact two-sample t-test using simulated means and variances
set.seed(3283072)
n    <- n_per_arm(0.80)
reps <- 20000
m_diff <- rnorm(reps, delta, sigma * sqrt(2 / n))             # difference in sample means
s2     <- sigma^2 * rchisq(reps, 2 * (n - 1)) / (2 * (n - 1))   # pooled sample variance
tstat  <- m_diff / sqrt(s2 * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(tstat) > qt(0.975, 2 * (n - 1))), reps))

Output from running the code

100% of observed effect: n per arm = 82 (80% power), 110 (90% power)
 75% of observed effect: n per arm = 146 (80% power), 195 (90% power)
 50% of observed effect: n per arm = 327 (80% power), 438 (90% power)
power.t.test: 82.6 
Simulated power at n = 82 per arm: 0.798 (20000 replicates)

The simulation draws many sets of group means and variances at the planned sample size and counts how often the two-sample t-test rejects the null. It should land close to the target power; at very small required sample sizes it can fall below it, because the normal approximation is rough there.

13. Analyses Reported by the Sponsor

The registry’s own analysis is labeled ANCOVA and reports p = 0.592 across the four intervention groups. The outcome description says that a two-way analysis of covariance with repeated measures was conducted, with time and intervention as factors and gender as a covariate. The record does not report adjusted differences, adjusted means, confidence intervals, model coefficients, or a detailed explanation of the tested term.

The separate recalculations compare each active-condition mean with the sham mean using posted summary statistics. They are unadjusted mean differences with intervals, not estimates from the sponsor’s repeated-measures ANCOVA. They do not account for participant-level pairing, time, or the gender covariate. Because the approaches and information used differ, their results need not match.

The sponsor’s p-value is the registry-reported inferential result, while the recalculated estimates are supplementary descriptions of the posted means and their uncertainty under a simpler comparison. Neither should be mistaken for the other. In particular, the unadjusted intervals cannot supply the missing adjusted effect estimates, and the overall p-value cannot reveal the size or direction of each adjusted comparison. The precise relationship between the overall test and individual active-versus-sham contrasts is not reported in the registry record.

14. Multiplicity, Missing Data and Analysis Populations

The registry lists three active-condition comparisons with sham in the recalculated summaries, as well as one reported ANCOVA result across the four groups. Multiple comparisons can increase the chance that at least one result appears notable by chance when many hypotheses are tested. In general, multiplicity procedures can adjust interpretation or testing to account for a family of comparisons. The registry record does not report whether the sponsor prespecified a family of contrasts, used a multiplicity adjustment, or treated any pairwise comparisons as primary. The p-value of 0.592 is reported for the overall analysis; no adjustment details are given.

The registry describes the primary outcome analysis as a repeated-measures ANCOVA, but the detailed statistical analysis plan, analysis population definition, and model specification are not reported in the registry record. It is therefore unclear whether all enrolled participants were intended to be included, how participants with incomplete outcome data were handled, or whether any exclusions occurred. The record reports 22 enrolled and 20 observations per condition, but does not explain the difference or identify the participants contributing to each summary.

Missing data matter in a crossover study because a participant may have information for some conditions or time points but not others. In general, the consequences depend on why data are missing, how much information is absent, and the analysis method’s assumptions. The registry does not report missingness by condition or time point, reasons for missing measurements, imputation, or a sensitivity analysis. No conclusion about the extent or impact of missing data can be drawn from the available record.

The registry also does not report analysis by randomized sequence, period effects, or carryover effects. These details can matter when interpreting crossover data because condition and order may be related in ways that affect the outcome. This identifies information absent from the record; it does not establish that a specific bias occurred. The available report supports a description of the listed design and posted results, but not a full audit of the sponsor’s analysis choices.

15. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 0% reporting threshold.

The registry reports that 22 participants were at risk for serious adverse events and that 0 were affected. It also reports 22 participants at risk for other adverse events and 0 affected. The death count is 0. No serious adverse-event terms are listed among the top reported terms.

These figures describe events recorded in the registry for the study population; they do not establish that no unrecorded or unreported events occurred. The record does not provide event definitions, collection procedures, observation time for safety, or condition-specific adverse-event results. In a crossover study, these details would help clarify how events were attributed across exposure conditions, but they are not reported in the registry record.

The figures should therefore be read as the posted counts: no serious adverse events, other adverse events, or deaths are recorded among the stated at-risk population. The counts do not compare the safety of the active conditions with sham, since condition-specific rates and estimates are not given. Nor can these data alone characterize the frequency of uncommon events or establish a safety profile beyond the participants and observation recorded in the trial.

16. What the Result Does, and Does Not, Mean

The registry records average thermal-threshold values after three acute intermittent hypoxia conditions and sham in a randomized, double-masked crossover study of healthy adults. The posted means range from 45.9 for sham to 46.7 for Intervention 1. The unadjusted differences versus sham are 0.8, 0.5, and 0.2 for Interventions 1, 2, and 3. Their intervals all include zero, and the sponsor’s reported overall ANCOVA result is p = 0.592.

These results describe the measurements and analyses posted in the registry. They do not establish that any condition produces a sensory change that is important to participants. The record does not define the clinical meaning of higher or lower thermal thresholds, provide a threshold for a meaningful difference, or report adjusted effect estimates and intervals from the sponsor’s repeated-measures model. The overall p-value also does not quantify the size of a contrast.

The design is specific to acute exposures and a post-intervention outcome averaged over measurements every 10 minutes for 60 minutes. The record does not support conclusions about longer-term effects, other sensory outcomes, other populations, or clinical use. The title identifies healthy adults, so findings should not be generalized automatically to people with a health condition. The registry describes the study’s stated purpose as examining sensory effects, and the posted primary outcome addresses thermal thresholds rather than every possible sensory function.

Statistical uncertainty matters: each unadjusted interval covers values below and above zero, and the intervals are based on summary statistics rather than the paired repeated-measures analysis. The sponsor’s result and the recalculations are not directly interchangeable. Together they show the reported summaries and the limits of the available inference; they do not provide a definitive estimate of an adjusted active-versus-sham effect.

The appropriate reading is therefore bounded. The registry records no statistically separated unadjusted contrast from sham in the posted recalculations and reports an overall ANCOVA p-value of 0.592. It does not report enough information to determine the adjusted contrast sizes, their precision, the handling of missing observations, or the clinical importance of any difference. No treatment recommendation follows from these data.

17. Why This Trial Matters Statistically

This trial raises a useful statistical issue: a crossover design and a simple comparison of arm-level summaries answer different questions. The design allows participants to experience multiple conditions in different sequences, which can make within-participant information relevant. A repeated-measures analysis can account for dependence among observations from the same participant. In contrast, a recalculation from condition means and standard deviations cannot recover the pairing or covariance information.

The outcome also has a time dimension. Thermal thresholds were measured every 10 minutes for 60 minutes after intervention, while the registry reports an average. The analysis description names time and intervention and identifies repeated measures. This makes the specification of time, the dependence among repeated observations, and the interpretation of an average important to understanding what the estimated effect represents. The record does not provide enough detail to evaluate those choices fully.

The comparison of sponsor-reported and unadjusted results illustrates why an overall p-value is not a substitute for effect estimates and intervals. The reported ANCOVA p-value of 0.592 communicates a test result, but the record does not give adjusted contrast sizes or precision. The unadjusted differences and intervals offer a scale-based description, yet they do not reproduce the crossover model. Statistical reporting is most informative when estimates, uncertainty, and model details accompany significance testing.

Finally, the record separates enrollment from outcome counts without explaining the relationship: 22 participants were enrolled, while each condition summary has 20 observations. This makes clear analysis-population and missing-data reporting important. Without those details, readers cannot assess who contributed to each estimate or how incomplete measurements may affect the result. The trial matters statistically as an example of how design, endpoint timing, analysis structure, and reporting completeness all shape interpretation.

18. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The principal trial-specific limitation in interpreting the posted summaries is the lack of detail about the crossover data. Four sequences are listed, but sequence-specific outcomes, period effects, carryover assessments, and washout information are not reported in the registry record. These omissions make it difficult to assess how order and repeated exposure were handled in the sponsor’s analysis.

The registry reports 22 participants enrolled and 20 observations for each condition, but does not explain the counts. It does not identify the analysis population, describe participant flow, report missing outcome measurements, or state how incomplete records were handled. The reported standard deviations and means therefore cannot show whether the same participants contributed to all four condition summaries.

The primary analysis description is brief. It names repeated-measures ANCOVA, time, intervention, and gender as a covariate, but does not provide the model equation, covariance structure, term-specific results, adjusted contrasts, or confidence intervals. The overall p-value is reported without a corresponding effect estimate. The unadjusted recalculations use summary statistics and cannot address these gaps or exploit the crossover pairing.

The endpoint is limited to average thermal thresholds during the stated post-intervention window. The registry does not report a clinical interpretation threshold or explain how a difference in degrees celsius relates to participants’ experience. This restricts interpretation of the estimates’ practical importance. The study title identifies healthy adults, and the registry does not establish applicability to other populations or outcomes.

Finally, details about masking, intervention delivery, safety observation, and multiplicity are limited. The record identifies double masking and device interventions but does not explain implementation. It does not report adjustment for multiple comparisons or condition-specific adverse-event results. These are limitations of what can be established from the registry record, not evidence that any particular design or analysis problem occurred.

19. Related Tutorials

20. Related Statistical Calculators

21. Sources

• ClinicalTrials.gov: Effects of Acute Intermittent Hypoxia on Sensory Function in Healthy Adults — NCT03283072.
• Registry record last updated 2023-08-21; status last verified 2023-08.
• Results first posted 2023-08-21.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

22. Record Summary

NCT03283072 is a completed randomized, double-masked crossover study titled “Effects of Acute Intermittent Hypoxia on Sensory Function in Healthy Adults,” sponsored by the University of Florida. The registry reports 22 participants enrolled and lists three acute intermittent hypoxia device conditions and one sham condition.

The primary outcome is mean thermal threshold in degrees celsius, averaged from measurements every 10 minutes for 60 minutes after intervention. The registry reports 20 observations per condition, an overall ANCOVA p-value of 0.592, and no adjusted estimates or confidence intervals for that analysis. Unadjusted recalculations versus sham have intervals that include zero.

The registry records 0 serious adverse events, 0 other adverse events, and 0 deaths among 22 participants at risk. Details on missing data, the analysis population, crossover period handling, and the sponsor’s full statistical model are not reported in the registry record.