Olfactory Contributions to Sleep-dependent Food Craving: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Olfactory Contributions to Sleep-dependent Food Craving (NCT04179838), a randomized na study. The primary outcome is Functional Magnetic Resonance Imaging (fMRI)-Based Decoding Accuracy of Food vs Non-food Odors in Piriform Cortex.

2. Clinical Question

3. Trial Design

This was a randomized, unmasked, crossover experiment. In a crossover design, participants receive each condition in sequence, so comparisons can use information from the same people under both conditions. The two possible sequences were sleep-deprived first and non-sleep-deprived second, or the reverse. The registry describes a 4-week washout between the two test periods. Randomizing sequence can help prevent order from being systematically tied to one sleep condition, but the registry record does not report whether order effects were assessed.

The sleep-deprived condition involved one night with 4 h of sleep; the non-sleep-deprived condition involved one night with 8 h of sleep. The primary brain measure was collected in the evening, approximately 19 h after the relevant night. Food intake was measured at a buffet, and the listed secondary outcomes also include energy density of consumed food and hormone measures.

Enrollment was 29. The primary outcome summaries report 25 observations in each condition. Because this is a crossover study, those condition-level counts should not be interpreted as separate groups of participants; the record describes participants undergoing both conditions. Reasons why the outcome count is lower than enrollment, the number of participants completing both periods, and the analysis population are not reported in the registry record.

The registry marks the study completed and identifies Northwestern University as sponsor. It does not report the trial’s original sample-size justification, a detailed statistical analysis plan, or whether interim analyses were planned or conducted. The registry’s posted primary test is a one-sided t-test. Further details about how that test handled the paired crossover observations are not reported in the registry record.

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.

First Intervention (1 Day)

Reasons for not completing this period:

Washout (4 Weeks)

Second Intervention (1 Day)

Reasons for not completing this period:

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 Functional Magnetic Resonance Imaging (fMRI)-Based Decoding Accuracy of Food vs Non-food Odors in Piriform Cortex, measured over A single time point, in the evening (~19 h) after both the night of the "Sleep-Deprived" and the "Non-Deprived" intervention..

The primary endpoint is fMRI-based decoding accuracy for distinguishing food odors from non-food odors using responses in the piriform cortex. The study team analyzed odor-evoked fMRI data with a searchlight-based support vector machine and leave-one-run-out cross-validation. In general, a classification accuracy describes how often a classifier assigns observations to the correct category; here, the categories were food and non-food odors. The endpoint is reported as a mean percentage correctly classified.

This measure is used to quantify how much information about odor category can be decoded from activity in a specified brain region. It is not a direct measure of food preference, craving, or the amount eaten. The registry describes two maps of decoding accuracy for each participant, one for each sleep condition, and says the measure of interest was extracted from the piriform cortex.

The primary outcome was assessed at a single evening time point after each sleep condition. That timing makes the comparison specific to the response measured at that time, rather than a measure of change throughout the day or over repeated nights. The endpoint’s standard errors describe uncertainty around the reported condition means; they do not by themselves describe the within-participant variability or the correlation between the two conditions.

7. Results Posted in the Registry

• Non-Sleep Deprived: mean 49.65 versus 54.38 on control; difference -4.73 (95% CI -6.80 to -2.66), standardized effect d = -1.26; z-test p = <0.001.

8. Secondary Efficacy Endpoints

The registry also posts results for secondary outcomes. For each outcome below, the endpoint type, the test applied to the posted summary statistics, the result and its interpretation are given in turn. Secondary endpoints are usually not adjusted for multiple testing, so they support rather than confirm the primary finding.

Energy-density of Consumed Food

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Non-Sleep Deprived: mean 3.68 versus 3.88 on control; difference -0.20 (95% CI -0.52 to 0.12), standardized effect d = -0.35; z-test p = 0.219.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Non-Sleep Deprived and the control. The interval gives the range of effects that remain compatible with the data.

Ghrelin

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Non-Sleep Deprived: mean 573.47 versus 584.33 on control; difference -10.86 (95% CI -123.31 to 101.59), standardized effect d = -0.05; z-test p = 0.850.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Non-Sleep Deprived and the control. The interval gives the range of effects that remain compatible with the data.

Leptin

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Non-Sleep Deprived: mean 9.68 versus 10.19 on control; difference -0.51 (95% CI -4.82 to 3.80), standardized effect d = -0.07; z-test p = 0.817.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Non-Sleep Deprived and the control. The interval gives the range of effects that remain compatible with the data.

The registry lists six secondary outcomes, but numerical summaries are posted for only three: energy density of consumed food, ghrelin, and leptin. For those outcomes, the condition means are somewhat higher in the sleep-deprived condition, while the recalculated intervals include zero. These are descriptive patterns, not evidence that the secondary outcomes follow the primary brain-decoding result or share its interpretation.

Numerical results for total calories of consumed food, insulin, and cortisol are not reported in the registry record. Their inclusion in the list of secondary outcomes does not establish the direction, size, or precision of any comparison. The available secondary results also do not provide a complete picture of food intake or the hormonal response.

Secondary outcomes can help describe whether a primary finding is accompanied by related changes, but each additional comparison creates further opportunity for chance findings. The registry does not report a multiplicity adjustment or a hierarchy specifying which secondary outcomes should be considered confirmatory. Accordingly, unadjusted secondary p-values carry less weight than a prespecified primary analysis when several outcomes are examined. They should be read as exploratory evidence and interpreted alongside their estimates and uncertainty, rather than as independent confirmation of the primary result.

The primary endpoint concerns fMRI classification accuracy, whereas the secondary endpoints concern food composition, total intake, or hormone concentrations. These outcomes measure different aspects of the experiment. A pattern in one domain does not, by itself, establish a corresponding effect in another.

Secondary outcomes registered for this trial, for reference (only those above were recalculated):

• Energy-density of Consumed Food
• Ghrelin
• Leptin
• Total Calories of Consumed Food
• Insulin
• Cortisol

9. Statistical Interpretation

10. Statistical Methods Explained

The study is randomized and uses a crossover model. In general, a crossover design exposes each participant to more than one condition and compares their measurements across conditions. This can reduce the influence of stable differences between participants, because each person can serve as their own comparator. The registry describes two treatment sequences and a washout period between conditions. The exact model used to account for within-person pairing, period, sequence, or carryover is not reported in the registry record.

What does the reported mean represent?

The primary outcome is continuous: decoding accuracy is measured as a percentage, and the registry reports a mean for each condition. The mean is the arithmetic average of the measured values in that condition. The primary summaries are 54.38 for the sleep-deprived condition and 49.65 for the non-sleep-deprived condition. These are condition means, not estimates of the outcome for every participant.

The accompanying standard errors are 0.81 and 0.68, respectively. In general, a standard error describes the sampling uncertainty of an estimated mean. It is not the same as the standard deviation, which describes variation among individual observations. In a crossover study, the standard error for a difference between conditions also depends on how paired observations vary together. That within-person covariance is not reported in the registry record, so it cannot be recovered from the two standard errors alone.

What does the comparison estimate?

The summary-statistic recalculation reports the mean difference as non-sleep-deprived minus sleep-deprived. Its estimate is -4.73 percentage points, with an interval from -6.803 to -2.657. A negative estimate in this ordering means the non-sleep-deprived condition had a lower mean than the sleep-deprived condition in the posted summaries. The interval describes uncertainty around that recalculated contrast under its method. The confidence level for the interval is not reported in the registry record.

The registry’s own primary analysis is reported as a one-sided t-test, with p = 0.001. A t-test evaluates how compatible the observed data are with a specified null hypothesis under its assumptions. A one-sided test evaluates evidence in a specified direction; a two-sided test allows departures in either direction. The registry does not report the detailed test specification, the direction specified in advance, or the treatment of the crossover pairing. Those details matter when interpreting the reported p-value.

How should a p-value be read?

In general, a p-value is the probability, assuming the null hypothesis and the test’s assumptions, of obtaining a test statistic at least as extreme as the observed one under the test’s definition of extremeness. It is not the probability that the null hypothesis is true, and it does not measure the size or practical importance of a difference. The sponsor’s p-value and the recalculated unadjusted p-value arise from different summaries and test descriptions; the registry’s reported analysis remains the trial’s own stated analysis.

The recalculated unadjusted p-value is < 0.001. “Unadjusted” means it is not corrected for testing multiple outcomes or comparisons. The registry does not report whether the sponsor’s primary p-value was adjusted. It also does not report a complete multiplicity plan, so the individual p-values should not be read as if they necessarily account for all analyses in the study.

Why does over-dispersion matter?

Over-dispersion is a general concern when observed variability is greater than a model assumes. It can make standard errors too small and uncertainty appear narrower than it is. For a continuous measure analyzed with a t-test, the relevant concern is whether the model’s assumptions about the distribution and variability are reasonable. The registry record does not report diagnostic checks or a specific over-dispersion assessment, so no conclusion about those checks can be drawn.

For crossover data, an additional issue is dependence: the two measurements from one participant are related. An analysis that treats the conditions as independent can fail to use that pairing correctly. The posted summary-statistic recalculation cannot incorporate the within-participant correlation because that correlation is not reported. The exact sponsor model and its handling of dependence are also not described beyond the reported one-sided t-test.

What can the posted summaries establish?

The means describe the observed condition averages, and the interval gives a measure of precision for the recalculated difference. Together with the registry’s test, they support a statistical description of the analyzed observations. They do not identify the mechanism behind the difference, establish whether a similar contrast would occur in a different sample, or answer every outcome question listed in the study. The registry record does not report an original sample-size justification, interim analysis plan, missing-data method, or detailed analysis population. These omissions limit assessment of how fully the posted analysis reflects the intended design.

11. Confidence Intervals and Precision

The recalculated primary difference is -4.73 percentage points for non-sleep-deprived relative to sleep-deprived, with an interval from -6.803 to -2.657. The interval does not cross zero. In general, a confidence interval gives a range of effect values compatible with the observed data and the assumptions of the calculation. Its width reflects precision: narrower intervals indicate less uncertainty under the method, while wider intervals indicate more.

The registry record does not state the confidence level of this interval. It should therefore be interpreted as the posted interval estimate without assigning it a coverage percentage that the record does not specify. Nor should the interval be treated as a range containing the true effect with certainty. It is conditional on the data and method used.

The outcome summaries report standard errors of 0.81 for sleep-deprived and 0.68 for non-sleep-deprived observations. These describe uncertainty around the respective means, not the paired difference directly. Because participants experienced both conditions, precision for the within-person contrast depends in part on the correlation between their two measurements. That correlation is not reported. The summary-statistic recalculation therefore cannot reflect all information that may be available to a paired analysis.

The registry’s own primary test reports p = 0.001 but does not include an effect estimate or interval. A p-value alone provides no direct description of the range of plausible effect sizes. The recalculated estimate and interval add that descriptive information, while retaining the limitations of a summary-based calculation. The two forms of reporting answer related but not identical questions.

12. 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 illustrations based on the observed effect, not the trial’s original design calculation. They assume a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation. The figures are participants per arm for 80% power or participants per arm for 90% power; they are not total events. This is not observed power.

The 80% power column describes the number of participants per arm that the planning calculation estimates would be needed to have an 80% chance of detecting the specified effect under those assumptions, if that effect and the assumptions were correct. The 90% column describes the corresponding planning target for a 90% chance. These are prospective design quantities, not probabilities that the result already observed is real.

The planning display varies the effect to be detected: it considers 100% of the observed effect, 75% of the observed effect, and 50% of the observed effect. Planning for a smaller effect requires more participants because a smaller signal is harder to distinguish from random variation. The registry’s original sample-size justification and design assumptions are not reported in the registry record. The illustration also describes an equal-allocation two-arm calculation; it should not be mistaken for a reconstruction of the crossover trial’s own planning process.

These figures use the observed effect as their starting point, so they are sensitive to the uncertainty and possible overestimation of that observed effect. They do not establish what sample size would be appropriate for another study without a separate design decision about the target effect, outcome variability, analysis, and study structure.

13. 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 = -4.730; pooled standard deviation σ = 3.739.
2. n = 2 × (1.96 + 0.842)² × 3.739² / -4.730² = 9.8, rounded up to 10 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 <- -4.73                      # observed difference in means
sigma <- 3.73915097314885                     # 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(4179838)
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 = 10 (80% power), 14 (90% power)
 75% of observed effect: n per arm = 18 (80% power), 24 (90% power)
 50% of observed effect: n per arm = 40 (80% power), 53 (90% power)
power.t.test: 10.9 
Simulated power at n = 10 per arm: 0.760 (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.

14. Analyses Reported by the Sponsor

The registry’s own primary result is a one-sided t-test with p = 0.001. The sponsor’s posted record does not provide the corresponding primary effect estimate or confidence interval. This p-value should be treated as the reported result of the trial’s own analysis rather than as a direct estimate of the size of the contrast.

The summary-statistic recalculation gives a mean difference of -4.73 for non-sleep-deprived relative to sleep-deprived, with an interval from -6.803 to -2.657 and an unadjusted p-value of < 0.001. It is a simple recalculation from the posted condition summaries, not a reproduction of the sponsor’s full analysis. The registry describes a crossover study, but the within-participant correlation and detailed analysis model are not reported. Those features can affect the standard error and test result.

The sponsor’s one-sided test and the recalculation are therefore related evidence, but their p-values should not be treated as interchangeable. The sponsor’s analysis is the trial’s stated analysis; the recalculated estimate supplies a transparent summary-based contrast and interval. The registry does not report enough analysis detail to reconcile the methods further.

15. Multiplicity, Missing Data and Analysis Populations

The registry reports one primary endpoint and several secondary outcomes. It reports a one-sided t-test for the primary comparison and test results for some secondary outcomes, but a plan for controlling multiplicity is not reported in the registry record. When several hypotheses are tested, the chance of at least one apparently unusual result can increase. General approaches include adjusting the significance threshold, controlling a family-wise error rate, or treating secondary analyses as exploratory; the record does not say which approach, if any, was planned.

The registry does not state whether the primary outcome and its test direction were specified before examining the data, nor does it describe a hierarchy for secondary outcomes. These details affect how strongly isolated p-values should be interpreted. The reported primary test remains the sponsor’s stated result, while unadjusted secondary results should be viewed more cautiously in the context of multiple comparisons.

Missing-data handling and analysis populations are also not reported in the registry record. Enrollment was 29, while the primary outcome summaries include 25 observations per condition. The record does not explain why these counts differ, whether all participants completed both crossover periods, or how any incomplete measurements were handled. It does not identify an intention-to-treat, per-protocol, or other analysis population.

These omissions do not show that data were missing in a particular way or that the analysis was inappropriate. They mean that readers cannot assess from the registry record whether missingness may have affected the estimates, whether all randomized participants were included, or how robust the results are to alternative analysis choices. The detailed statistical analysis plan, missing-data procedures, and interim-analysis arrangements are not reported in the registry record.

16. 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 serious adverse events for 29 participants at risk in each condition, with zero participants affected in either condition. It also reports other adverse events for 29 participants at risk in each condition, with zero affected, and zero deaths in both conditions.

These figures describe what was recorded for the study’s two condition groups. In a crossover trial, the same participants may contribute to both condition records, so the at-risk counts should not be added as if they represented distinct people. The registry does not list serious adverse-event terms. Zero affected participants in the posted record means no such events were reported in these summaries; it does not prove that the conditions have no risk or establish safety in other settings or populations.

The record does not provide further detail about adverse-event definitions, collection procedures, duration of monitoring, or whether events were judged related to a condition. The safety figures are therefore limited to the counts posted in the registry.

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

The primary results describe a difference in average fMRI decoding accuracy between two sleep conditions in this crossover experiment. The sleep-deprived condition had a higher reported mean than the non-sleep-deprived condition. The summary-based interval and the sponsor’s one-sided test provide complementary statistical descriptions, but the interval’s confidence level and the sponsor’s detailed model are not reported in the registry record.

The endpoint is a measure of how accurately a classifier distinguished food from non-food odors based on fMRI responses in the piriform cortex. It is not a direct measurement of craving, appetite, food choice, or caloric intake. A difference in decoding accuracy should not be translated into a claim about those behaviors without corresponding evidence from those specific outcomes.

The trial also lists food intake and hormone outcomes, but the registry contains numerical summaries for only some secondary measures. The unreported results cannot be inferred from the primary result or from the names of the outcomes. The reported data therefore answer a focused question about an imaging measure, rather than establishing a broad account of how sleep affects eating or physiology.

The experiment compares one night of partial sleep deprivation with one night of normal sleep, with a washout between conditions. It does not, from the posted information, establish what would occur after repeated nights, in other circumstances, or in populations not described in the record. Participant characteristics and detailed eligibility criteria are not reported in the registry record.

Statistical evidence should also be separated from practical importance. A p-value does not quantify the size or importance of an effect, and the recalculated interval is subject to the limits of the available summary data. The results are best understood as the observed contrast in this study and its reported precision, with uncertainty about analysis details and incomplete secondary outcome reporting.

18. Why This Trial Matters Statistically

This study illustrates a statistical challenge common to crossover experiments: each participant contributes measurements under multiple conditions, and the relationship between those repeated measurements affects the precision of the contrast. Reporting condition means and their standard errors is useful, but without the within-participant correlation or detailed model, readers cannot fully reconstruct the paired analysis from summary data.

It also shows why an effect estimate and interval add information beyond a test result. The sponsor’s reported p-value indicates the outcome of a one-sided test, while the recalculated mean difference and interval describe the contrast’s direction, magnitude, and uncertainty under a summary-based method. Neither should be mistaken for a complete account of the analysis.

Several outcomes are listed, spanning imaging, food intake, and hormones. Without a reported multiplicity plan, the evidence from secondary comparisons needs cautious interpretation. In addition, incomplete numerical reporting makes it difficult to assess whether the measured brain outcome aligned with the behavioral and physiological outcomes that motivated the broader question.

Finally, the difference between enrollment and the condition-level outcome counts underscores the value of reporting analysis populations and missing-data handling. Those details help readers judge how the randomized design translated into the analyzed results. Their absence limits statistical appraisal even when the principal numerical result is available.

19. 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 registry record does not report participant characteristics or detailed eligibility criteria, limiting assessment of who the findings represent. It also does not explain why enrollment was 29 while the primary summaries include 25 observations per condition, whether participants completed both periods, or how missing measurements were handled.

Although the study is described as randomized and crossover, the registry does not report the statistical model’s handling of paired observations, period effects, sequence effects, or possible carryover. The 4-week washout is described, but no assessment of carryover is reported. The condition means and standard errors alone do not provide the within-participant correlation needed to reconstruct a paired contrast.

The registry’s primary test is one-sided, but the direction specified in advance and the detailed analysis plan are not reported. The confidence level for the recalculated interval is not stated in the registry record. The record also does not report an original sample-size justification, interim analyses, or a multiplicity strategy.

Numerical results are available for only three of the six listed secondary outcomes. Detailed results for total calories, insulin, and cortisol are not reported in the registry record. Adverse-event summaries give counts but do not describe event definitions, monitoring procedures, or relatedness assessments. These reporting gaps constrain interpretation of both the analysis and the scope of the findings.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: Olfactory Contributions to Sleep-dependent Food Craving — NCT04179838.
• Registry record last updated 2020-03-25; status last verified 2020-03.
• Results first posted 2020-03-25.
• 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.

23. Record Summary

NCT04179838, “Olfactory Contributions to Sleep-dependent Food Craving,” is a completed randomized, unmasked crossover study sponsored by Northwestern University. It enrolled 29 participants and compared one night of partial sleep deprivation, described as 4 h of sleep, with one night of normal sleep, described as 8 h of sleep, using a 4-week washout between conditions.

The primary outcome was fMRI-based decoding accuracy for food versus non-food odors in the piriform cortex. The registry reports a one-sided t-test with p = 0.001. Condition summaries include 25 observations per condition; the registry does not explain the difference from enrollment. Selected secondary outcome results are posted, while numerical results for other listed outcomes are not reported in the registry record. Serious adverse events, other adverse events, and deaths are reported as zero in both condition records.