Analysis of Electrocorticographic Signals: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Analysis of Electrocorticographic Signals (NCT03785028), a na na study. The primary outcome is Qualitative Measure: Prioritization Cue-related Changes in the Neural Representation of Stimuli Reported as Binary for Prioritized and Unprioritized Item Decodability.

2. Clinical Question

3. Trial Design

This is a completed, single-group basic science study, registered as NCT03785028 under the title “Analysis of Electrocorticographic Signals.” The record lists the condition as epilepsy, the sponsor as University of Wisconsin, Madison, and enrollment of 5. The study began on 2019-07-19, reached primary completion on 2021-02-21, and has a status of COMPLETED. Its phase is listed as NA.

The one experimental arm involved working-memory and attention tasks. The study description says signals were collected from epilepsy patients having surgery as part of their treatment while they performed two cognitive tasks. The primary endpoint description focuses on decoding remembered faces, scenes, or words from electrocorticography signals and comparing items according to their priority for upcoming task demands. There is no separate control arm and no randomized allocation. Thus, the prioritized-versus-unprioritized contrast is an analysis of task-related neural representations, not a treatment effect estimated from parallel trial groups.

The primary outcome timeframe is listed as twenty minutes. The endpoint is presented as a qualitative measure reported for the group of all participants. The primary outcome description notes that data from 4 participants were reported as a group: three completed 5 blocks of 20 trials each, and one completed 4 blocks. One participant struggled with the task, and that participant’s data were not interpretable. These details describe the outcome population; the registry does not explain how the enrolled total relates to the reported group beyond those statements.

The stated research objectives concern working memory and behavior, and the study description says results from each of the proposed experiments would be published in peer-reviewed journals. Whether those publication endpoints were achieved is not reported in the registry record. A protocol-level sample-size justification, detailed statistical analysis plan, and explanation of the study’s allocation and masking entries 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.

Overall Study

Analysis population for the primary outcome, as defined in the registry: Data collected from 4 participants, reported here as a group. Three of them completed 5 blocks of 20 trials each, resulting in a total of 100 trials, and one of them completed 4 blocks, 80 trials in total. One participant struggled to perform the task so data were not interpretable.

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 Qualitative Measure: Prioritization Cue-related Changes in the Neural Representation of Stimuli Reported as Binary for Prioritized and Unprioritized Item Decodability, measured over Twenty minutes.

The primary endpoint concerns whether electrocorticography signals carry information about the memory items participants are holding in mind. The study description says a machine-learning method, multivariate pattern classification, is used to decode the face, word, or scene being remembered from signals measured by electrodes. Support Vector Machines are described as the classifiers, with performance assessed using a 10-fold cross-validation procedure.

The central contrast is between a prioritized item and an unprioritized item. Priority is defined by the demands of the upcoming task: a cue indicates that one of the items presented on a trial will be tested first. The endpoint is intended to assess whether the neural representation of an item differs according to that priority, using decodability as an operational measure. In general, decodability asks whether a classifier can distinguish the relevant categories from the signal patterns; it is not itself a direct measure of subjective experience or memory accuracy.

The registry identifies the outcome as qualitative and says it is reported categorically for the group of all participants. It does not give the category counts, classifier performance values, uncertainty estimates, or a numerical comparison between priority conditions. Those results are not reported in the registry record. Consequently, the endpoint’s purpose and analysis approach can be described, but the observed findings cannot be characterized from this record.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The endpoint uses multivariate pattern classification to ask whether measured electrocorticography signals contain patterns that distinguish remembered items. In general, a classifier learns a relationship between input features and labels. Here, the described inputs are brain signals, and the labels correspond to memory items such as faces, scenes, or words. The scientific interest is whether decoding differs according to whether an item is prioritized or unprioritized for the upcoming task demands.

What classification performance represents

In general, classification performance describes how well a model assigns observations to their intended categories. It depends on the signal, the labels, the model, and the way performance is evaluated. A higher ability to distinguish categories would indicate that the measured signal contains information useful to the classifier. It would not, by itself, prove that a person consciously represents an item in a particular way, nor would it directly measure clinical benefit. The registry calls the endpoint qualitative and categorical, but does not report the categories or the resulting performance values.

Why cross-validation is used

Cross-validation is a general approach for assessing how a model performs on data not used to fit that particular model. The endpoint description specifies a 10-fold cross-validation procedure. In general, the data are divided into folds; a model is fit using some folds and evaluated on another, with this process repeated so that each fold is used for evaluation. This can offer a more informative assessment than reporting performance on the same observations used to train the classifier.

Cross-validation does not automatically remove every source of bias. Its usefulness depends on how observations are assigned to folds and whether related observations are kept appropriately separated. For example, if observations from the same participant or closely related trials appear in both training and evaluation data, the performance estimate may not represent generalization to new participants. How folds were constructed, whether they were stratified, and whether participant-level separation was used are not reported in the registry record. Those details matter for interpreting what population or data the cross-validated performance might generalize to.

Why uncertainty matters

In general, a point estimate of classification performance is incomplete without information about its uncertainty. Results can vary across participants, task trials, and the particular data used for model fitting and evaluation. A confidence interval or another uncertainty summary can show how precise an estimate is under stated assumptions. No estimate or uncertainty measure for classification performance is reported in the registry record. As a result, the record does not support an assessment of precision or consistency.

The outcome is described as a group measure, with the group identified as all participants, while the population note says data from 4 participants were reported as a group. Group-level aggregation can conceal differences between individuals. In general, a pooled result may reflect a pattern that is not present equally in each participant. Participant-specific estimates or a description of between-participant variation would help distinguish a consistent group pattern from one driven by particular observations; such information is not reported in the registry record.

What the registered contrast does and does not test

The prioritized-versus-unprioritized contrast is based on the task’s cue-related demands. It concerns the relative decodability of items within the study’s cognitive task. It is not a randomized comparison of treatment groups, because the record describes a single experimental arm and no control arm. In general, task contrasts can address focused mechanistic questions, but their interpretation depends on the design, measurement, and analysis details. They should not be translated into a claim about treatment effectiveness.

The primary outcome description gives a timeframe of twenty minutes and describes the classifier analysis, but it does not report a full statistical plan. The precise performance metric, handling of repeated trials, model tuning procedure, decision threshold, and inferential approach are not reported in the registry record. Without those details, the classification workflow cannot be fully reconstructed from the registry entry.

Over-dispersion and clustered observations

In general, over-dispersion occurs when observed variability is greater than a statistical model expects. It can make standard errors too small if the model treats correlated observations as independent. Repeated task trials from the same participant are often related, so an analysis may need to account for clustering or dependence. The registry describes blocks and trials for participants, but does not report how within-participant dependence was handled. This is a general consideration, not evidence that the analysis had over-dispersion.

The registry lists 0 analyses and does not report numerical outcome findings. It is therefore not possible to evaluate model fit, assess the adequacy of uncertainty estimates, or compare an unadjusted calculation with a sponsor analysis. Any sample-size illustration based on an observed effect is a planning exercise, not observed power; the record does not report a numeric result from which the endpoint’s actual precision could be judged.

10. Multiplicity, Missing Data and Analysis Populations

The registry lists 3 outcome measures, while the study description refers to two experiments. The primary outcome concerns cue-related changes in neural representation of prioritized and unprioritized items. How the other outcome measures relate to the primary endpoint, and whether they were treated as confirmatory or exploratory, are not reported in the registry record.

When a study evaluates multiple outcomes, comparisons, or analysis choices, the chance of at least one apparently unusual result can increase. In general, multiplicity procedures or a prespecified hierarchy can help clarify which comparisons are primary and how evidence should be interpreted. The registry record does not report a multiplicity adjustment, outcome hierarchy, or detailed statistical analysis plan. No inference about whether such procedures were used can be made from the listed outcome count alone.

The outcome description says one participant struggled to perform the task and that the data were not interpretable. It also says data from 4 participants were reported as a group, while enrollment is listed as 5. The reason that other data do not appear in the described group, the criteria for interpretability, and any missing-data handling rules are not reported in the registry record. The available description does not establish whether the exclusion was prespecified or how it might affect the result.

The analysis population is described for the primary measure, but the record does not report a broader population definition, participant-level inclusion rules, or a sensitivity analysis. There are 0 analyses listed, and no numerical results or analysis-specific denominators are reported. Accordingly, the extent to which missing or uninterpretable data could influence the endpoint cannot be quantified from the registry record.

11. 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 adverse-event figures list 5 participants at risk for serious adverse events, 0 affected, 5 at risk for other adverse events, and 0 affected. The record also lists 0 deaths. These are the registry’s reported counts for the experimental arm. No serious adverse-event terms are listed.

These counts describe what is recorded, not proof that adverse events could not occur. In general, a zero count means no event of that type was recorded among the stated at-risk group during the relevant reporting period; it does not establish that the underlying risk is zero. The reporting period and event-collection details are not reported in the registry record. The record also does not give further information about event definitions, severity beyond the serious-versus-other distinction, or attribution.

This was a single-group study involving behavioral tasks and electrocorticography signals in people having surgery as part of epilepsy treatment. The adverse-event counts should not be interpreted as a comparative estimate, because there is no control arm. The registry does not provide information that would support comparing safety between interventions or attributing events to a particular study procedure.

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

The record establishes that a completed single-group study examined electrocorticography signals during working-memory and attention tasks in people with epilepsy undergoing surgery. It describes an analysis using Support Vector Machines and a 10-fold cross-validation procedure to assess decodability of remembered items according to their priority. The endpoint is framed as a group-level qualitative measure of prioritized and unprioritized item decodability.

What the record does not establish is the result of that analysis. It does not report classification performance values, participant-specific findings, uncertainty intervals, or statistical test results. Therefore, no direction, magnitude, or precision of a prioritized-versus-unprioritized difference can be stated. The lack of numerical results should not be interpreted as evidence of no difference, and it does not allow a conclusion of equivalence.

The study is not a randomized treatment comparison. Its single arm and within-task contrast address a basic science question about neural signal patterns during a cognitive task. The endpoint does not, by itself, measure a clinical benefit, and the record does not report a clinical outcome comparison. The study’s stated objectives concern how the brain keeps information in mind and uses it to guide behavior.

Interpretation is also limited by the small outcome group described in the registry and by the note that one participant’s data were not interpretable. The entry does not provide participant-level results, the complete analysis procedure, or an account of missing-data handling. These omissions make it impossible to judge how representative a group finding would be or how robust the classification result was.

The record’s status is COMPLETED, but completion status is not a substitute for reported findings. The stated publication endpoints concern publication of results from each proposed experiment; their status is not reported in the registry record. The supported conclusion is therefore about the study’s question and methods, not about whether the analysis detected a neural representation difference.

13. Why This Trial Matters Statistically

This study illustrates a statistical distinction between measuring a signal and making a claim about what that signal means. A classifier can be used to assess whether patterns in electrocorticography data distinguish remembered items. The interpretation depends not only on the model but also on the validation procedure, the unit of analysis, and how repeated observations are handled. The registry identifies a cross-validation approach, but does not give enough analysis detail to evaluate all of those choices.

The design also shows why the scientific comparison should be described precisely. The prioritized-versus-unprioritized contrast concerns task-related neural decodability within a single group. It is not a randomized treatment contrast and cannot be read as a treatment effect. Clear separation of mechanistic questions from clinical comparisons helps prevent statistical language about “effects” from being mistaken for evidence about patient outcomes.

The outcome population makes participant-level variation important. The record describes group reporting for data from 4 participants and notes that one participant’s task data were not interpretable. In general, a pooled estimate can obscure whether a pattern is shared or concentrated in a subset of participants. Reporting individual variation, alongside an appropriate group summary and its uncertainty, can make a small mechanistic study easier to interpret. Such findings are not reported in the registry record.

Finally, a completed study can still have an incomplete statistical record. The entry explains the endpoint and the classifier approach but lists 0 analyses and no numerical outcome results. This distinction matters: methods can describe what researchers intended to measure, while results are needed to assess magnitude, uncertainty, and consistency. For this study, the registry supports discussion of the design and measurement question, but not a quantitative judgment about the observed neural signal.

14. 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 offers limited information for independently evaluating the analysis. It describes Support Vector Machines and a 10-fold cross-validation procedure, but does not report the exact classification metric, model-tuning process, fold construction, or inferential method. In particular, it does not say whether observations were separated by participant when the classifier was evaluated. These details affect what the performance estimate would represent and how well it might generalize.

The outcome is described as a group measure, and the record says data from 4 participants were reported as a group. It does not provide participant-level performance or a measure of variability across participants. A group summary without such information can make it difficult to distinguish a broadly shared pattern from participant-specific results. The entry also notes that one participant’s data were not interpretable, but does not specify the criteria used to make that determination or the missing-data approach.

The listed enrollment is 5, while the outcome description refers to 4 participants with reported data and one participant whose data were not interpretable. The registry does not explain this accounting further. The number of trials and blocks is described for participants in the outcome population, but the record does not say how those observations were weighted or modeled.

There is no control arm, and the registry lists no randomized allocation. The study therefore cannot support a treatment comparison. The record lists 3 outcome measures but does not describe multiplicity handling or a complete analysis plan. Finally, numerical primary-outcome results and uncertainty estimates are not reported in the registry record. These limitations prevent assessment of the observed effect’s size, precision, and consistency.

15. Related Tutorials

16. Related Statistical Calculators

17. Sources

• ClinicalTrials.gov: Analysis of Electrocorticographic Signals — NCT03785028.
• Registry record last updated 2024-05-01; status last verified 2024-04.
• Results first posted 2024-05-01.
• 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.

18. Record Summary

NCT03785028 is a completed, single-group basic science study titled “Analysis of Electrocorticographic Signals,” sponsored by University of Wisconsin, Madison. The record lists epilepsy as the condition, enrollment of 5, a start date of 2019-07-19, and primary completion on 2021-02-21. The study involved working-memory and attention tasks during electrocorticography signal collection.

The primary endpoint concerns machine-learning classification of remembered faces, scenes, or words, with decodability considered for prioritized and unprioritized items. The record describes Support Vector Machines and 10-fold cross-validation. It says data from 4 participants were reported as a group and notes that one participant’s data were not interpretable.

No numerical endpoint findings or analysis results are reported in the registry record. The entry lists 0 analyses. It reports 0 serious adverse events, 0 other adverse events, and 0 deaths among the stated at-risk counts. The design has no separate control arm, so the neural contrast is not a randomized treatment comparison.