Behavioral and Neural Response to Memantine in Adolescents With Autism Spectrum Disorder: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Behavioral and Neural Response to Memantine in Adolescents With Autism Spectrum Disorder (NCT01972074), a randomized phase 3 study. The primary outcome is Treatment Responder.

2. Clinical Question

3. Trial Design

This completed Phase 3 study, titled “Behavioral and Neural Response to Memantine in Adolescents With Autism Spectrum Disorder,” was sponsored by Massachusetts General Hospital. The registry lists a randomized, quadruple-masked, parallel-group design and enrollment of 84. The study began on 2015-02-17, and primary completion was recorded as 2018-05-07. The registry status is completed.

The treatment period was 12 weeks. The clinical comparison was memantine versus placebo, with the main posted outcome being a binary responder count at the end of the treatment period. The registry also describes pre- and post-treatment neuroimaging, including functional magnetic resonance imaging and hydrogen magnetic resonance spectroscopy. Healthy controls underwent neuroimaging twice, 12 weeks apart, and received no intervention during that window. They do not form a third arm in the posted responder comparison.

The primary responder analysis population was defined by exposure to study medication for at least 2 weeks. The control group was excluded from that analysis. The posted responder counts therefore use analysis denominators of 19 for memantine and 21 for placebo, while the adverse-event record uses different at-risk counts. These numbers describe different posted analysis sets and should not be treated as interchangeable.

The registry record does not report the original sample-size justification, the trial’s planned statistical model, or the details of allocation concealment. It also does not provide a complete account of which masking roles were included in quadruple masking. The design labels establish randomization, masking, and parallel treatment groups, but do not by themselves specify the methods used to implement or analyze those features.

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.

Before assignment: Per protocol, participants were enrolled when they provided informed consent. They then underwent screening procedures to confirm eligibility. If eligible and willing/able to participate, participants then began study procedures. Some participants dropped out prior to starting study procedures and were, therefore, excluded from this analysis.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Participants included in this analysis were exposed to study medication for at least 2 weeks. Participants in the Control Group were excluded from this analysis because they did not receive study medication and, therefore, cannot be categorized into Treatment Responders vs. Treatment Non-Responders.

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: Baseline Participants in the Memantine and Placebo groups completed the Week 0: Baseline study visit. Baseline Participants in the Control group completed the Week 100: Baseline Scan study visit. The Baseline assignment is different for the Control group because they received no intervention and so did not complete the Week 0: Baseline study visit.

6. Primary Endpoint

The registry lists the primary outcome as Treatment Responder, measured over 12 Weeks (from Baseline [Week 0] to Endpoint [Week 12]).

The primary endpoint is a binary participant-level outcome: each analyzed participant is categorized as a treatment responder or non-responder. A responder had to meet both parts of the registry definition at endpoint: a 25% reduction from baseline in the Social Responsiveness Scale, Second Edition: School-Age, Parent Report (SRS-2) total raw score, and an Autism Spectrum Disorder Clinical Global Impression-Improvement (ASD CGI-I) score of 2 or less.

The SRS-2 is a parent- or guardian-completed, 65-item measure used to assess ASD symptom severity. Its items use a four-point response scale, and higher scores indicate greater symptom severity. The ASD CGI-I is a clinician-rated measure of improvement, with higher scores indicating less improvement. Requiring both criteria makes the endpoint a joint threshold: meeting one criterion alone is not enough to count as a responder.

A responder count is useful for communicating how many participants crossed a prespecified clinical threshold, rather than only describing an average score change. However, converting continuous scale information into a yes-or-no outcome loses detail. Participants just below and just above the threshold are placed in different categories even if their measured outcomes are close. The posted result therefore addresses this combined responder definition, not every possible change in social symptoms or neuroimaging.

The timeframe is 12 weeks, from baseline at Week 0 to endpoint at Week 12. The registry record does not report additional primary endpoints in the posted outcome record.

7. Results Posted in the Registry

• Memantine: 47.4% (9/19) versus 19.0% (4/21) on control; risk difference 28.3 percentage points (95% CI 0.3 to 56.4), risk ratio 2.49 (0.91–6.77); unadjusted two-proportion z-test p = 0.056.

8. Statistical Interpretation

9. Statistical Methods Explained

The primary outcome is binary: a participant either met the combined responder criteria or did not. For a binary outcome, each group’s response proportion is the number classified as responders divided by the number included in that group’s analysis. The posted counts are 9 responders among 19 analyzed participants for memantine and 4 among 21 for placebo. The registry also reports the corresponding proportions as 0.474 and 0.191.

What does the risk difference describe?

The risk difference subtracts the placebo response proportion from the memantine response proportion. Its estimate is 0.283. This is an absolute-scale comparison: it expresses the difference in observed response proportions, rather than how many times as large one proportion is as the other. The estimate is specific to the responder definition and analysis population used in the registry. It should not be read as a difference in average symptom-score change.

A confidence interval for a risk difference gives a range of values compatible with the observed data under the interval’s statistical method and assumptions. The posted interval runs from 0.003 to 0.564. The width matters: a broad interval indicates that the estimate is imprecise and allows a range of possible differences. The interval is close to zero at one end, so its lower boundary provides little separation from no difference. It also extends to a considerably larger difference, reflecting uncertainty in the magnitude.

How should the p-value be understood?

The unadjusted p-value is 0.056. In general, a p-value measures how unusual data at least as discrepant as those observed would be under a specified null hypothesis and statistical model. It is not the probability that the null hypothesis is true, and it is not a measure of the probability that a result will be repeated. The registry record does not state the test or model underlying this value. Because the p-value is unadjusted, it should not be assumed to account for multiple outcomes or analyses.

The p-value and confidence interval are complementary summaries, not interchangeable evidence. The record does not specify whether they use the same inferential method. Their proximity to conventional decision boundaries makes it especially important to avoid turning either into a simple declaration of success or failure. The effect estimate, interval, endpoint definition, and analysis population all contribute to interpretation.

What do the relative measures add?

The posted risk ratio is 2.487, with a confidence interval from 0.914 to 6.767. It compares the response proportion in the memantine group with that in the placebo group on a multiplicative scale. The posted odds ratio is 3.825, with a confidence interval from 0.931 to 15.72. Odds are not the same quantity as probabilities, so the odds ratio should not be described as the ratio of response probabilities. Both relative-measure intervals are wide and include their respective null values, indicating considerable uncertainty in those estimates.

In general, relative measures can look more pronounced than absolute differences, especially when outcomes are not rare. They answer a different question from the risk difference and should be interpreted with the underlying group proportions. Here, the posted absolute proportions and risk difference make the observed scale of the result clear, while the relative estimates show that uncertainty also remains on multiplicative scales.

Why can over-dispersion matter?

Over-dispersion is a general concern when observed variation exceeds what a statistical model assumes. If a model understates variation, standard errors can be too small and intervals or tests can appear more decisive than warranted. For this trial’s posted primary outcome, the record reports participant counts but does not describe a model for over-dispersion, clustering, or other departures from a simple binary comparison. The issue cannot be evaluated from the registry summary alone, and no claim about its presence or absence is supported.

Randomization is intended, in general, to support a fair comparison by allocating participants to treatment groups in a way that limits systematic differences at baseline. Masking is intended to reduce the influence of treatment knowledge on care, outcome assessment, or reporting. The registry labels this trial randomized and quadruple-masked, but does not report enough implementation detail to determine how these protections were carried out or whether masking remained intact.

The treatment-responder analysis excludes healthy controls and requires at least 2 weeks of study-medication exposure. It is therefore not simply a comparison of every person enrolled or every person assigned. The registry record does not report an intention-to-treat analysis, a per-protocol analysis, a method for handling missing endpoint measurements, or an adjusted model for the primary comparison. Those omissions limit what can be inferred about the influence of exclusions and incomplete outcome data.

10. Confidence Intervals and Precision

The risk-difference estimate is 0.283, and its confidence interval extends from 0.003 to 0.564. The interval conveys substantial uncertainty about the size of the observed group difference. Its lower endpoint is close to no difference, while its upper endpoint permits a considerably larger difference. The point estimate alone would conceal that spread.

The other posted measures reinforce the imprecision. The risk ratio is 2.487, with a confidence interval from 0.914 to 6.767. The odds ratio is 3.825, with a confidence interval from 0.931 to 15.72. Both intervals are broad and include the null value for their scale. The relative estimates therefore do not establish a precise magnitude of association.

Confidence intervals describe uncertainty under the method used to calculate them; they are not guarantees that the true value lies within a particular interval. The registry record does not identify the interval method or explain whether it matches the method used for the unadjusted p-value of 0.056. The posted risk-difference interval has a lower boundary above zero, while the p-value is above 0.05. Because the methods are not specified, this apparent difference in summaries should not be resolved by selecting one statistic and ignoring the other.

The responder analysis includes 19 participants in the memantine group and 21 in the placebo group. With relatively few participants represented in each posted group count, individual outcome classifications can have a visible effect on the estimated proportions. This is a general feature of binary outcomes in small analysis sets, not evidence of a particular error. The registry record does not provide additional information that would narrow the uncertainty.

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 events. They illustrate sample sizes under assumptions based on fractions of the observed effect: planning for 100% of the observed effect, planning for 75% of the observed effect, and planning for 50% of the observed effect. The figures are 43 and 57 participants per arm for 80% and 90% power when using 100% of the observed effect; 72 and 96 participants per arm when using 75%; and 150 and 201 participants per arm when using 50%.

In general, power is the chance that a study will detect an effect of a specified size under the assumed model and significance level. The 80% power column gives the illustrated participants per arm for a design with 80% power under the stated assumptions. The 90% power column gives the corresponding planning figure for 90% power. These are not summaries of the trial’s achieved power and should not be compared with the observed enrollment as though they were a result.

The planning illustration assumes a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation. The effect to detect is the stated fraction of the observed effect. Smaller effects require more participants for the same target power under these assumptions. The figures are illustrations computed from the observed effect, not the trial’s own design requirements; the original design assumptions are not reported in the registry record.

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, p₀ and p₁ are the control and treatment proportions, and p̄ is their average.

Worked calculation for this trial

1. Control proportion p₀ = 0.190; treatment proportion p₁ = 0.474; average p̄ = 0.332.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.6660 = 1.3054.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.6352 = 0.5346.
4. n = (1.3054 + 0.5346)² / (0.2832)² = 42.2, rounded up to 43 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
p0 <- 0.19047619047619                          # control proportion
p1 <- 0.473684210526316                          # treatment proportion
n_per_arm <- function(power, frac = 1) {
  q1 <- p0 + (p1 - p0) * frac               # smaller effect = fraction of the observed difference
  pbar <- (p0 + q1) / 2
  z_b <- qnorm(power)
  ceiling(round((z_a * sqrt(2 * pbar * (1 - pbar)) + z_b * sqrt(p0 * (1 - p0) + q1 * (1 - q1)))^2 / (q1 - p0)^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 (unrounded n per arm)
cat("power.prop.test:", round(power.prop.test(p1 = p0, p2 = p1, power = 0.80)$n, 1), "\n")

# Simulation: how often does the z-test reject at the 80%-power sample size?
set.seed(1972074)
n    <- n_per_arm(0.80)
reps <- 20000
x0 <- rbinom(reps, n, p0); x1 <- rbinom(reps, n, p1)
ph <- (x0 + x1) / (2 * n)
z  <- ((x1 - x0) / n) / sqrt(ph * (1 - ph) * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(z) > z_a, na.rm = TRUE), reps))

Output from running the code

100% of observed effect: n per arm = 43 (80% power), 57 (90% power)
 75% of observed effect: n per arm = 72 (80% power), 96 (90% power)
 50% of observed effect: n per arm = 150 (80% power), 201 (90% power)
power.prop.test: 42.2 
Simulated power at n = 43 per arm: 0.818 (20000 replicates)

The simulation draws many pairs of event counts at the planned sample size and counts how often the two-proportion z-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. Multiplicity, Missing Data and Analysis Populations

The registry record posts one primary outcome measure and reports no secondary outcome results. It also lists no posted analyses. This means the registry results do not provide a set of secondary estimates against which to assess consistency or explore other outcome dimensions. The study description includes neuroimaging aims, but the record does not post corresponding outcome results in the outcome record summarized by the primary responder counts.

Multiplicity arises when several outcomes, time points, subgroups, or statistical tests are examined. In general, repeated testing can increase the chance of obtaining an apparently notable result by chance if the analysis does not account for the family of tests. The record does not report a multiplicity strategy, a hierarchy among outcomes, or whether interim analyses were planned or conducted. Those matters are not reported in the registry record and should not be inferred from the Phase 3 label or the posted p-value.

The analysis population is described as participants exposed to study medication for at least 2 weeks. Healthy controls were excluded because they did not receive study medication and could not be classified under the responder definition. The posted counts are 19 for memantine and 21 for placebo. The registry does not report how many assigned participants lacked endpoint data, why participants may have been excluded, or how missing response status was handled.

In general, missing-data methods can affect an estimate when outcome availability is related to treatment assignment or participant outcomes. The registry record does not identify an imputation method, a sensitivity analysis, or a primary analysis population such as intention-to-treat. It also does not report covariate adjustment or a model for the primary comparison. The analysis population and missing-data approach therefore cannot be reconstructed fully from the posted result.

The unadjusted p-value should not be interpreted as multiplicity-adjusted, because the registry record does not report such an adjustment. Equally, it would be inappropriate to assume that a particular correction was required or used. The available record supports description of the posted comparison, while leaving the broader analysis plan and handling of missing outcomes unreported.

14. 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 adverse-event record lists serious adverse events affecting 0 participants among 22 at risk in the memantine group and 0 participants among 21 at risk in the placebo group. Deaths are listed as 0 in both groups. The record lists no serious adverse-event terms. For the control group, the at-risk counts are 0, so those entries do not represent an observed safety comparison with treated participants.

The same adverse-event record lists other, non-serious adverse events affecting 14 participants in the memantine group and 9 in the placebo group. These are separate from the serious-event counts and should not be combined with them. The registry record does not provide event-level details for these other events in the posted summary.

These figures are counts within the registry’s stated at-risk groups, not estimates adjusted for exposure time or a complete safety analysis. The at-risk counts differ from the responder-analysis denominators, so the two sets should not be substituted for one another. Zero serious events in the posted counts does not establish that serious events are impossible or that a treatment is safe; it describes only what the record reports for these groups.

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

The posted result shows a higher observed responder proportion in the memantine group than in the placebo group under the registry’s combined endpoint definition. The estimated risk difference is 0.283, based on 9 responders among 19 analyzed participants in the memantine group and 4 among 21 in the placebo group. The confidence interval and the relative-measure intervals indicate that the magnitude is uncertain. The unadjusted p-value is 0.056, and the registry record does not report the test used to produce it.

This result concerns a thresholded, 12-week behavioral responder outcome. It does not describe every participant’s degree of change, establish effects on all ASD symptoms, or report the neuroimaging outcomes discussed in the study description. It also does not show how the result might vary across participant characteristics, because subgroup results are not posted.

The result should not be treated as proof that memantine caused the observed difference. Randomization and masking are important design features, but the registry summary does not provide the full analysis plan, missing-data methods, or enough detail to evaluate every potential source of bias. The difference between the responder-analysis group sizes and the adverse-event at-risk counts further illustrates that each result pertains to a defined analysis population.

Nor does the result establish that there is no difference between groups. The estimates are imprecise, and the confidence intervals allow a range of effects. A p-value above a conventional threshold is not evidence that the treatments are equivalent. The registry record does not report an equivalence or non-inferiority design, margins, or analyses.

In general, clinical interpretation depends on the outcome’s importance, the magnitude and precision of the estimate, study methods, and the broader evidence. The registry record supports a cautious description of the posted responder comparison; it does not by itself settle the clinical significance or generalizability of that result.

16. Why This Trial Matters Statistically

This trial illustrates why a binary responder result needs to be read together with its definition and denominators. Response required both a specified reduction in parent-reported SRS-2 score and a qualifying clinician-rated ASD CGI-I score. That joint threshold is understandable as a clinical summary, but it compresses different patterns of score change into a single responder category and depends on both measures being available.

The comparison also shows the importance of absolute and relative effect measures. The risk difference describes the observed separation in response proportions, while the risk ratio and odds ratio express the comparison on relative scales. Their intervals are broad, emphasizing that a prominent point estimate does not necessarily imply a precise estimate. The unadjusted p-value of 0.056 is another summary, not a substitute for examining the effect estimates and uncertainty.

Analysis populations matter as well. The responder counts include participants exposed to study medication for at least 2 weeks and exclude healthy controls, whereas the adverse-event counts use different at-risk group sizes. Keeping those denominators distinct prevents an incorrect comparison between outcomes or groups that were not analyzed on the same basis.

Finally, the study description includes behavioral and neural aims, but the posted primary outcome result is a behavioral responder count. The absence of posted secondary results and details on missing-data handling, multiplicity, and the statistical plan limits what can be concluded beyond that specific comparison. These are general lessons for interpreting registry results: the endpoint definition, analysis population, effect scale, uncertainty, and completeness of reporting all shape what a numerical result can support.

17. 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 the trial’s original sample-size justification, planned statistical model, allocation-concealment procedures, or detailed implementation of quadruple masking. It does not specify the test behind the unadjusted p-value or the method used for the confidence intervals. The apparent difference between the risk-difference interval and the p-value therefore cannot be explained from the posted record.

The responder analysis is restricted to participants exposed to study medication for at least 2 weeks. The record does not describe exclusions in detail, report the number or reasons for missing endpoint measurements, or identify how missing response status was handled. It also does not state whether the primary comparison was adjusted for baseline characteristics or analyzed under an intention-to-treat framework.

The outcome is a dichotomized combination of parent-reported and clinician-rated measures. The posted result does not include continuous score changes, distributions, or separate results for the two component criteria. The registry description identifies neuroimaging objectives, but the outcome record does not post those results or secondary outcome findings.

For safety, the record provides counts of participants with serious and other adverse events, but does not supply event-level detail for the other events or exposure-adjusted rates. Serious adverse-event counts are zero in both treatment groups, but the posted counts alone cannot characterize all possible safety outcomes. The adverse-event at-risk group sizes also differ from the responder-analysis denominators.

These reporting limits do not show that a method was absent or that an unreported analysis was not conducted. They mean that those aspects cannot be evaluated from the registry record. The findings should therefore be understood as the posted results for the stated outcome and analysis population, with uncertainty and reporting gaps kept in view.

18. Related Tutorials

19. Related Statistical Calculators

20. Sources

• ClinicalTrials.gov: Behavioral and Neural Response to Memantine in Adolescents With Autism Spectrum Disorder — NCT01972074.
• Registry record last updated 2024-07-01; status last verified 2024-06.
• Results first posted 2019-09-11.
• 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.

21. Record Summary

NCT01972074 is a completed Phase 3 randomized, quadruple-masked, parallel-group trial sponsored by Massachusetts General Hospital. It studied memantine versus placebo in adolescents with autism spectrum disorder over 12 weeks and included pre- and post-treatment neuroimaging. The registry lists enrollment of 84.

The primary posted outcome is treatment response, defined by a 25% reduction in SRS-2 total raw score and an ASD CGI-I score of 2 or less. Responders were 9 of 19 in the memantine group and 4 of 21 in the placebo group. The estimated risk difference is 0.283, with a confidence interval from 0.003 to 0.564; the unadjusted p-value is 0.056.

The record does not post secondary outcome results or analyses and does not report the original sample-size rationale, missing-data methods, or full statistical plan. Serious adverse events and deaths are listed as 0 in both treatment groups in the posted safety record.