Mobile Attention Retraining in Overweight Female Adolescents: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Mobile Attention Retraining in Overweight Female Adolescents (NCT02977403), a randomized na study. The primary outcome is Changes in Food-cue Visual Probe Task Attention Bias (AB) Reaction Time.

2. Clinical Question

The trial’s broader objective was to study relationships among eating patterns, body weight, and brain responses to images. Its summary describes interest in what happens in the brain when adolescents train their attention away from food using a smartphone program. The primary endpoint is a behavioral reaction-time measure, rather than a direct measure of brain activity, and the registry also lists frequency of loss-of-control eating episodes as a secondary outcome.

The comparison is between randomized groups, but the registry record does not report enough detail to establish how every outcome was analyzed or which analysis population was used. The posted results therefore describe specific recorded outcomes and estimates; they do not by themselves resolve the broader questions about eating patterns, weight, or brain responses.

3. Trial Design

The registry describes NCT02977403, titled “Mobile Attention Retraining in Overweight Female Adolescents,” as a completed, randomized, quadruple-masked, parallel-group trial. The official title is “Pilot Study of Mobile Attention Training in Overweight Female Adolescents.” The phase is recorded as NA. Enrollment was 82. The listed sponsor is the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD).

The recorded study start was 2017-02-10, and primary completion was 2023-06-08. The registry describes six visits over about 8 months. The first visit included screening procedures such as height, weight, blood pressure, waist measurements, medical history, physical examination, a urine sample, a DXA scan, and questions about health, social and psychological functioning, and eating habits. Parents or guardians of minor participants were also to answer questions about functioning and demographic information.

Before visits 2-6, participants were to avoid food and drink for about 12 hours. Those visits could include blood draws, MRI scans, meals with rating forms, and simple thinking tasks. The registry’s summary does not give a complete account of all visit procedures.

The active program trained attention using a consistent relation between the food-picture type and the location of the probe. In the sham program, the probe could replace either the neutral or the food picture at random. These descriptions distinguish the intended training contingency from the control experience; the record does not report adherence or engagement results for the programs.

The posted primary outcome is a change in food-cue visual-probe-task reaction-time attention bias, with a 2-weeks timeframe. The endpoint description specifies analyses of high-palatability food, low-palatability food, and non-food image pairings. The arm-level primary results include 22.0 participants in the retraining group and 21.0 in the sham group. These outcome counts are smaller than the total enrollment, and the registry notes computer and recording failures as sources of missing data. The record does not report a complete accounting of participant flow or the analysis population definitions.

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: Computer failures led to missing data for many participants. There were interruptions in data recording during magnetoencephalography sessions that were not predictable due to equipment failures.

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.

Population: 68 were randomized.

6. Primary Endpoint

The registry lists the primary outcome as Changes in Food-cue Visual Probe Task Attention Bias (AB) Reaction Time, measured over 2-weeks.

The primary endpoint is change in reaction time on a visual probe task, expressed in milliseconds. Participants saw image pairs involving high-palatability food, low-palatability food, or non-food images. A trial was classified as congruent when the probe appeared behind the more food-salient cue and incongruent when it appeared behind the less salient cue. The recorded score subtracts average reaction time on incongruent trials from average reaction time on congruent trials, using only trials with correct responses for the probe direction.

As defined in the registry, positive scores indicate quicker responses when the probe is behind the more palatable stimulus, which the record interprets as bias toward that stimulus. Negative scores indicate slower responses in that situation, interpreted as bias away. A score of 0 represents no bias toward or away from the more palatable stimulus. The measure is intended to quantify how attention is allocated in response to food-related cues, rather than to count eating episodes or measure body weight.

Reaction time is used because the task records how quickly participants respond to probes presented in relation to competing images. In general, a difference in response speed across congruent and incongruent trials can be used as an indirect indicator of attention allocation. It is not a direct reading of a participant’s thoughts or a clinical outcome. The registry reports a 2-weeks timeframe, but does not describe the interval meant by that label in greater detail.

The posted arm means are 1.85 for retraining and 4.27 for sham, each with a 95% confidence interval. The intervals are broad and include values on either side of 0. The registry reports a confidence interval for each arm; a confidence interval for an arm mean is not the same as a confidence interval for the between-group contrast.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The main statistics are arm-level means, confidence intervals, reported beta coefficients, and—in the secondary outcome—means, standard deviations, a recalculated mean difference, and a sponsor-reported generalized-estimation-equation result. Each answers a different question. Arm means describe the observed group averages. A contrast estimates how far apart groups are on a defined scale. A confidence interval describes uncertainty around an estimate under the analysis assumptions. A p-value measures how compatible the data are with a specified null hypothesis under the chosen method; it does not measure the probability that the null hypothesis is true.

What does the primary confidence interval describe?

The registry reports least-squares means for the primary endpoint, with 95% confidence intervals. A least-squares mean is generally a model-based estimate of a group mean, which may account for features of a statistical model. The record does not describe the primary model or say whether covariates were included. Consequently, the arm-level least-squares means should not be treated as simple raw averages without qualification.

The 95% confidence interval attached to an estimate expresses uncertainty about that estimate. In general, if the same method were repeatedly applied under its assumptions, intervals constructed in this way would cover the true value at the stated rate. It is not a statement that there is a 95% probability that the particular interval contains the true value. For these primary arm estimates, the intervals cross 0, which is the endpoint’s defined no-bias value. That observation describes uncertainty about each arm mean; it does not replace a formal comparison between randomized groups.

How should the group contrast be read?

A between-group contrast asks whether the outcome differs between the retraining and sham groups. The primary registry record lists two beta coefficient estimates, each with a 95% confidence interval, but no method or p-value. Both intervals include 0, which is the null value for a difference-type estimate. Their widths indicate substantial uncertainty about the size and direction of the contrast. Because the record does not specify the model, contrast definition, or reason for two analyses, it is not possible to assign a more detailed interpretation to either coefficient.

In general, a model coefficient can represent a contrast after adjustment or transformation, depending on the model and coding. The coefficient’s meaning cannot be inferred solely from the label “beta coefficient.” A complete statistical interpretation would require the model form, covariates, coding choices, analysis population, and handling of missing observations. Those details are not reported in the registry record for these primary analyses.

What does the secondary recalculation add?

For the secondary eating-episode outcome, group means and standard deviations are posted for 26.0 participants per group. The registry also gives an unadjusted recalculation of the mean difference, with a 95% confidence interval and an unadjusted p-value. This is a simple comparison based on the posted summary statistics, not a replacement for the sponsor’s model-based analysis. The recalculated difference is -0.096, its interval runs from -1.265 to 1.073, and the unadjusted p-value is 0.872.

In general, a mean difference is easiest to interpret when the outcome is measured on a meaningful scale and the assumptions behind the comparison are reasonable. A confidence interval spanning 0 is consistent with a range of possibilities that includes no group difference. It also includes values in either direction, so it does not establish the direction of an underlying effect. An unadjusted p-value does not account for multiple outcomes or analyses unless a specific adjustment is applied.

The sponsor’s secondary analysis used generalized estimation equations. In general, generalized estimation equations estimate average relationships while accounting for within-participant or clustered correlation when the design requires it. Their appropriateness depends on the data structure, specified working correlation, outcome distribution, and variance estimation. The registry does not report those details, and the posted p-value is 0.256. The sponsor’s model-based result and the summary-statistic recalculation need not match because they are different analyses; the record does not provide enough information to reconcile them fully.

Why does over-dispersion matter?

General explanation: count outcomes can have variability greater than a simple statistical model expects. This is called over-dispersion. If a model assumes less variability than is present, standard errors and uncertainty intervals can be too small, and test results can appear more precise than warranted. The registry reports means and standard deviations for the eating-episode outcome, but does not report whether over-dispersion was assessed or how the generalized estimation equation handled the count distribution. This is a methodological question, not evidence that the analysis was or was not appropriate.

What are the planning figures?

Planning illustrations based on a stated fraction of the observed secondary effect give participants per arm for 80% power and participants per arm for 90% power. They are explicitly based on the observed effect rather than the trial’s own design. The stated assumptions are a two-sided significance level of 0.05, equal allocation between groups, and a normal approximation. The listed figures are planning illustrations, not observed power and not evidence about how many participants were actually needed for the original trial. The registry record does not report the trial’s own sample-size justification or design assumptions.

10. Analyses Reported by the Sponsor

The registry lists two primary beta coefficient analyses for the retraining and sham groups. Their estimates are -1.948 and -0.952, and their respective 95% confidence intervals are -20.790 to 16.894 and -35.280 to 33.377. Both intervals include 0. The record does not identify the methods used, explain why two analyses are listed, or provide p-values for them. That lack of detail limits how closely their interpretation can be connected to the arm-level least-squares means.

For the secondary loss-of-control eating outcome, the sponsor-reported analysis used generalized estimation equations and lists a p-value of 0.256. The registry does not post a confidence interval for that analysis. A separate, unadjusted recalculation from the posted group summary statistics gives a mean difference of -0.096, a 95% confidence interval from -1.265 to 1.073, and an unadjusted p-value of 0.872. These are not identical analyses: one is the sponsor’s model-based result, while the recalculation is an unadjusted comparison of the posted summaries. Differences between them cannot be attributed to a particular modeling choice because the model details are not reported.

The registry does not flag a discrepancy between the sponsor results and the recalculation. The sponsor’s analysis remains the trial’s reported model-based analysis; the recalculation is a supplementary, unadjusted check rather than a substitute. Neither result alone resolves the limitations created by missing detail about the analysis model, multiplicity, or analysis population.

11. Multiplicity, Missing Data and Analysis Populations

The registry record lists 54 outcome measures and 107 analyses, but it does not report a multiplicity strategy for the outcomes and analyses. A multiplicity plan states how investigators will interpret or adjust for the increased chance of chance findings when several hypotheses are tested. General statistical context: without such a plan, an isolated p-value among many analyses is less persuasive as confirmatory evidence. This concern is relevant to the secondary result, whose recalculated p-value is explicitly unadjusted. The record does not establish whether any adjustment was used elsewhere.

The record describes computer failures that led to missing data for many participants. It also notes unpredictable interruptions in data recording during magnetoencephalography sessions due to equipment failures. These explanations identify technical sources of missingness, but the registry does not quantify the missing primary endpoint data by reason, report whether missingness differed between groups, or describe an imputation or other missing-data method. It does not state whether the equipment interruptions affected all listed outcomes in the same way.

The arm-level primary results include 22.0 participants in retraining and 21.0 in sham; the secondary outcome has 26.0 participants in each group. The registry does not report a complete participant flow or define the populations used for each analysis. It is therefore not possible to determine from the record whether the results use an intention-to-treat population, a per-protocol population, or another analysis set. The relationship between the enrollment of 82 and the available outcome counts is not fully described.

Other key elements of the statistical plan are also not reported in the registry record. These include the primary model specification, covariate adjustment, interim analyses, the trial’s sample-size justification, and detailed rules for handling missing observations. The posted planning figures are illustrations based on the observed effect, not the trial’s own design assumptions. These omissions do not establish that the original protocol lacked such plans; they mean only that these details are not reported in the registry record.

Missingness can affect precision and, depending on why observations are missing, can also affect the validity of an estimate. General explanation: an analysis that uses only available observations may be unbiased under some missing-data conditions but biased under others. Without an account of the missingness mechanism and the analysis approach, readers cannot assess that issue fully. The broad primary confidence intervals and reduced endpoint counts reinforce the importance of treating the estimates cautiously.

12. 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.

Most frequent non-serious adverse event terms

The posted serious adverse event figures list 32 participants at risk in the retraining group and 36 in the sham group. Serious adverse events affected 0 participants in each group, and deaths were 0 in both groups. These are the recorded serious-event counts for the groups and should be read as such; the registry does not list any serious adverse-event terms.

The record also lists other, non-serious adverse events affecting 4 participants in retraining and 2 in sham, among the participants at risk in those groups. These counts are separate from the serious-event figures. The registry does not provide further information in the posted summary about the nature, timing, severity, or relationship of these other events to the programs. It also does not explain whether every participant in the primary outcome population was included in the safety population.

In general, an observed count of zero serious events does not demonstrate that a risk is absent. The available counts describe what was recorded in this trial’s safety reporting, not the full range of possible events or what might occur in other settings. The registry’s at-risk counts and event counts should not be used to infer comparisons beyond the information recorded.

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

The results describe a randomized comparison of smartphone attention-bias retraining with a sham program in overweight female adolescents. For the primary reaction-time measure, the posted group estimates and primary beta coefficient intervals are imprecise and include 0. The secondary loss-of-control eating estimate is also imprecise: the unadjusted recalculated interval includes 0, while the sponsor’s model-based result is reported without a confidence interval. Together, these statistics describe uncertainty in the posted results rather than a definitive conclusion about the programs.

The findings do not establish that attention retraining changes attention bias, eating episodes, body weight, or brain responses. The registry’s primary endpoint is a reaction-time score, and the one posted secondary outcome is a count of eating episodes. The trial’s broader objective mentions eating patterns, body weight, and brain responses to images, but the results summarized in the record do not establish effects across those domains. The registry also does not provide enough detail about missing data, analysis populations, or primary models to fully assess the analyses.

General statistical explanation: a confidence interval that includes the null value does not prove that there is no effect. It indicates that the data and method are compatible with a range of estimates that includes no difference. Conversely, a point estimate alone does not prove an effect, particularly when its uncertainty is broad. Interpretation should consider effect size, uncertainty, design, measurement quality, and the analysis plan together.

These results are specific to the participants, program descriptions, outcomes, and follow-up timeframe recorded in this trial. The registry describes right-handed females ages 12-17 who were overweight, and does not establish how the findings apply to other groups or contexts. The results should therefore be understood as estimates from a pilot study with documented recording problems and incomplete reporting of important analysis details, not as a universal answer to the broader research question.

14. Why This Trial Matters Statistically

This trial is statistically informative because it makes a behavioral training comparison in a population and setting where attention to food cues is of research interest. Its randomized, parallel-group design and sham comparator allow the groups to be compared under different probe-location contingencies. The quadruple-masking designation is recorded, although the registry does not describe which roles were masked or how masking was maintained.

The endpoint also illustrates the distinction between a measure’s operational definition and the broader concept it is intended to represent. The visual-probe task turns reaction-time differences on congruent and incongruent trials into a score interpreted as attention bias toward or away from more palatable stimuli. This offers a measurable outcome, but it remains an indirect behavioral indicator, not a direct measure of attention or brain activity. The score’s meaning depends on correct responses, the image pairings, and the trial-level recording process.

The posted results foreground a recurring statistical issue: an estimate without adequate precision cannot settle a research question. The primary means and coefficients have confidence intervals that include 0, and the secondary recalculation is also uncertain. The registry’s note that computer failures caused missing data for many participants, together with unpredictable recording interruptions, makes reporting of missingness and analysis choices especially important.

The record is also a useful example of why a sponsor-reported model analysis and a simple unadjusted recalculation should be distinguished rather than blended. The secondary generalized-estimation-equation result and the summary-statistic recalculation are different analyses. Reporting both can help readers understand what is directly available, while the missing model details limit reconciliation. Similarly, planning illustrations based on an observed effect are not evidence of achieved power and do not replace an original design justification.

More complete reporting of the primary model, analysis populations, multiplicity approach, sample-size rationale, and handling of missing data would allow a clearer assessment of the trial’s estimates. The posted statistics still offer a transparent account of the measured outcomes and their uncertainty, while showing why interpretation depends on both numerical results and the quality of the statistical record.

15. 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.

Several limitations specific to this trial affect interpretation. The registry reports computer failures that led to missing data for many participants and unpredictable recording interruptions during magnetoencephalography sessions. It does not quantify missingness for each outcome, report missingness by group, or explain how the primary analysis addressed missing observations. Because the primary endpoint depends on recorded reaction times, the consequences of incomplete recording cannot be evaluated fully from the posted information.

The primary outcome counts are 22.0 and 21.0, compared with enrollment of 82. The secondary outcome counts are 26.0 per group. The record does not provide a complete participant-flow account explaining how these counts arose, or define the analysis population for each endpoint. The registry also does not report whether participants adhered to the smartphone programs or how much training they completed.

The primary results include least-squares means and two beta coefficient analyses, but the model methods, coefficient interpretation, and reason for reporting two analyses are not described. No primary p-values are posted. The 2-weeks timeframe is recorded, but its precise relationship to visits or intervention exposure is not explained. The registry does not provide a primary sample-size justification, interim-analysis information, a multiplicity plan, or detailed missing-data procedures.

The secondary outcome is the only secondary measure with posted summary results in this record, despite the registry listing 54 outcome measures and 107 analyses. The relation between that single result and the larger set of measures is not described. The sponsor-reported secondary model has no confidence interval posted, while the recalculation is unadjusted. These reporting limits leave uncertainty about both precision and the role of multiple testing.

Finally, the eligibility summary describes a specific group: right-handed females ages 12-17 who were overweight. The record does not establish applicability to other populations. These limitations are constraints on interpretation of the registry record; they do not, on their own, show that the original study procedures or analyses were inappropriate.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Mobile Attention Retraining in Overweight Female Adolescents — NCT02977403.
• Registry record last updated 2025-05-31; status last verified 2025-05.
• Results first posted 2024-08-20.
• 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.

19. Record Summary

NCT02977403 is a completed randomized, quadruple-masked, parallel-group pilot study of smartphone attention-bias retraining versus a sham program in overweight female adolescents. Enrollment was 82. The primary outcome was change in food-cue visual-probe-task reaction-time attention bias at the 2-weeks timeframe. The registry also posts frequency of loss-of-control eating episodes as a secondary outcome.

The primary arm estimates and reported coefficient intervals include 0 and are imprecise. The secondary outcome has an unadjusted recalculated mean difference of -0.096, with a 95% confidence interval from -1.265 to 1.073; the sponsor-reported generalized-estimation-equation analysis lists a p-value of 0.256. Computer failures and recording interruptions are noted as sources of missing data, while details about analysis populations, primary model methods, multiplicity, and missing-data handling are not reported in the registry record.

The posted results describe uncertain estimates from this trial and do not establish conclusions about broader effects on eating, weight, or brain responses.