Duloxetine for Social Anxiety Disorder: Prediction of Long Term Outcome: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Duloxetine for Social Anxiety Disorder: Prediction of Long Term Outcome (NCT00114127), a randomized phase 3 study. The primary outcome is Anxiety Symptoms as Assessed by Liebowitz Social Anxiety Scale.

2. Clinical Question

3. Trial Design

The registry describes this as a completed Phase 3 study with randomized allocation, double masking, and a parallel model. The reported enrollment is 28. The sponsor is Massachusetts General Hospital. The listed study start is 2004-06 and primary completion is 2008-11. The registry record does not report the original sample-size justification, the planned statistical analysis in sufficient detail, or how the study phases were sequenced beyond the descriptions of the arms.

The arm descriptions distinguish an open Phase 1 in which all participants entered Duloxetine 60mg/day for 6 Weeks from a randomized Phase 2 comparison. In Phase 2, participants were randomized to Duloxetine 60mg/Day + Placebo for 18 Weeks or Duloxetine 120mg/Day for 18 Weeks. The posted primary outcome includes 13 participants in the 60mg/day plus placebo group and 15 in the 120mg/day group. The registry does not explain whether those counts represent all randomized participants, participants with outcome data, or another analysis population.

The primary outcome timeframe is listed as 6 months, while the Phase 2 regimens are described as lasting 18 Weeks. The registry record does not report the precise timing of the outcome assessment in relation to the treatment period, or whether the posted endpoint represents follow-up after treatment. It also does not report the randomization procedure, concealment details, masking assessment, or handling of protocol deviations. These omissions limit reconstruction of the design and analysis from the registry alone.

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: Of the 39 participants who completed Phase I of this study, 28 moved on to Phase II. 1 subject was lost to follow up between baseline and week 1.

Overall Study

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 Anxiety Symptoms as Assessed by Liebowitz Social Anxiety Scale, measured over 6 months.

The primary endpoint is anxiety symptoms assessed with the Liebowitz Social Anxiety Scale (LSAS), reported as a mean score at 6 months. The LSAS is a 24-item scale with separate fear and avoidance scores covering social and performance situations. Its total scores include fear from 0-72, avoidance from 0-72, and an overall score from 0-144. Higher scores represent increased social anxiety.

The registry lists suggested interpretations of 55-65 for moderate social phobia, 65-80 for marked social phobia, 80-95 for severe social phobia, and greater than 95 for very severe social phobia. These ranges give context for score levels, but the posted group means do not indicate how individual participants were distributed across the scale or interpretation bands. The registry record does not say which LSAS total score was used for the primary analysis, beyond describing the outcome as anxiety symptoms assessed by the LSAS.

A continuous score preserves information about the degree of difference between observed scores rather than reducing each participant to a categorical response. Comparing group means therefore summarizes average scores, while not showing the range of experiences within either group. The endpoint is useful for describing symptom scores, but its mean alone does not establish clinical importance, explain changes for individual participants, or resolve uncertainty about the treatment comparison.

7. Results Posted in the Registry

• Duloxetine 120mg/Day for 18 Weeks (Phase 2): mean 53.70 versus 67.90 on control; difference -14.20 (95% CI -35.69 to 7.29), standardized effect d = -0.49; z-test p = 0.195.

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.

CGI-S

• 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:
  • Duloxetine 120mg/Day for 18 Weeks (Phase 2): mean 3.60 versus 4.42 on control; difference -0.82 (95% CI -2.28 to 0.64), standardized effect d = -0.41; z-test p = 0.271.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Duloxetine 120mg/Day for 18 Weeks (Phase 2) and the control. The interval gives the range of effects that remain compatible with the data.

The registry lists CGI-S as a secondary outcome. The overall pattern is broadly similar to the primary comparison in that the recalculated mean difference favors a lower posted mean in the 120mg/day group, while the interval includes zero. This does not independently settle the uncertainty in the primary result. Both outcomes come from the same trial context, and the registry record does not report whether the secondary analysis was prespecified in detail or how it relates to the primary estimand.

Secondary outcomes can help describe different aspects of a study, but they are generally less decisive when many outcomes or analyses may be examined. Testing multiple outcomes increases the chance of observing a nominally notable result by chance if no adjustment is made. The registry record does not report a multiplicity procedure, a hierarchy of endpoints, or whether the secondary analysis was intended to support confirmatory inference. Its unadjusted result should therefore be treated as exploratory rather than as a standalone confirmation.

The record reports one secondary outcome and a sponsor test result for it, but does not provide a multiplicity-adjusted interpretation. Without a stated analysis plan, it is unclear whether all intended secondary outcomes and analyses are represented. The CGI-S result can be considered alongside the primary finding as descriptive evidence, with caution about its uncertainty and inferential status. It should not be used to strengthen the primary conclusion as though it were an independent replication.

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

• CGI-S

9. Statistical Interpretation

10. Statistical Methods Explained

These results involve continuous outcomes, for which each participant contributes a score and the group summary is a mean. A mean is the arithmetic center of the observed scores. It is informative about the group average, but does not show the distribution of scores, whether a few unusual observations influenced the average, or how participants' outcomes changed individually.

What does the mean difference express?

The mean difference expresses the difference between the two group means in the units of the outcome. For LSAS, the posted comparison is the 120mg/day group relative to the 60mg/day plus placebo group. A negative estimate means the mean in the first-named comparison group is lower than the reference group's mean. The estimate is an unadjusted recalculation from the posted summaries, not a separately reported adjusted treatment effect. It does not account for covariates or design features unless those are explicitly included in an analysis, and the registry record does not describe such adjustments.

What does a confidence interval communicate?

A confidence interval describes statistical uncertainty around an estimate under the assumptions of the analysis. Narrower intervals indicate greater precision; wider intervals indicate less precision. The 95% confidence interval for the recalculated LSAS mean difference extends from -35.69 to 7.286. Because it includes zero, the posted summaries are compatible with no mean difference as well as differences in either direction. A confidence interval is not a guarantee that the true value lies within its limits, nor does it directly describe the range of individual participant scores.

How should the p-values be read?

A p-value measures how unusual data at least as incompatible with a specified null hypothesis would be under the assumptions of a statistical test. It is not the probability that the null hypothesis is true, and it does not measure the size or clinical importance of an effect. The recalculated unadjusted p-value for LSAS is 0.195. The sponsor analysis is reported as a two-sided t-test with p <0.132. The registry does not report enough information to determine why these values differ or whether the tests used identical data and assumptions. A p-value should be read alongside the estimate, interval, design, and analysis details.

Why does the standard error matter?

The registry reports standard errors alongside the group means: 7.6 for the 60mg/day plus placebo group and 7.9 for the 120mg/day group. A standard error describes uncertainty in an estimated mean, rather than the spread of individual scores. It tends to be smaller when the mean is estimated more precisely. It should not be confused with a standard deviation, which describes variability among observations. The registry provides standard errors, and the recalculation uses them rather than treating them as standard deviations.

What does the test assume?

A two-sided t-test evaluates departures in either direction from the null comparison. Its interpretation depends on assumptions about the observations and analysis, including independence and the appropriateness of the model for the data. The registry lists a two-sided t-test for the sponsor's analysis, but does not report diagnostic checks, transformations, covariate adjustments, or the exact analysis population. The posted recalculation is an unadjusted comparison based on summary statistics, so it cannot reproduce unreported features of the sponsor's analysis.

Why is small-sample uncertainty important?

With limited observations, group averages and their differences can be sensitive to the outcomes of individual participants. Confidence intervals can consequently be wide, and estimates from small studies may not represent the magnitude that would be seen in a larger study. This is a general statistical concern, not proof that any particular observation influenced these results. The registry does not report participant-level data, distributions, or influence diagnostics, so the stability of the group means cannot be assessed directly.

What does over-dispersion mean in general?

Over-dispersion is a general issue for outcomes whose observed variability is greater than a statistical model assumes. When present and ignored, uncertainty can be understated. The primary and secondary outcomes in this record are continuous mean scores, and the registry does not report an over-dispersion assessment. It would therefore be inappropriate to claim that over-dispersion occurred or to apply a correction based on the registry information alone.

Overall, the posted statistical summaries support a descriptive comparison and an unadjusted recalculation, but not a full reconstruction of the analysis. Important details about the analysis population, missing observations, model checks, and prespecified methods are not reported in the registry record.

11. Confidence Intervals and Precision

The recalculated LSAS mean difference is -14.2, with a 95% confidence interval from -35.69 to 7.286. The interval is wide relative to the point estimate and crosses zero. This indicates limited precision: the summaries do not tightly establish either the size or direction of the difference. The endpoints are not separate observed effects; they mark the uncertainty range under the recalculation's assumptions.

The group means are reported with standard errors of 7.6 and 7.9. These quantify uncertainty in the estimated group means, not the spread of participants' LSAS scores. The recalculated interval reflects uncertainty in the contrast between the means. It should not be read as an interval containing most individual scores, and it does not describe how score changes varied among participants.

The sponsor's reported analysis gives p <0.132 but no estimate or confidence interval. As a result, its precision cannot be assessed from the sponsor result as reported. The registry record does not explain the difference between the sponsor's p-value and the recalculated p-value of 0.195, or provide sufficient detail to establish whether the two analyses used the same data and assumptions.

Confidence intervals describe statistical precision, not clinical importance. The registry provides interpretation ranges for LSAS scores, but the interval for a difference between group means is not itself an individual score or a direct measure of how many participants experienced meaningful change. Participant-level distributions and a prespecified threshold for meaningful change are not reported in the registry record.

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 sample-size figures are planning 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 the two arms, and a normal approximation. Each scenario uses the stated fraction of the observed effect as the effect to detect. The figures are participants per arm, not total event counts. They are not observed power.

For the LSAS comparison, planning for 100% of the observed effect gives 67 participants per arm for 80% power and 89 participants per arm for 90% power. Planning for 75% of the observed effect gives 118 participants per arm for 80% power and 158 participants per arm for 90% power. Planning for 50% of the observed effect gives 265 participants per arm for 80% power and 355 participants per arm for 90% power.

The 80% power column represents the illustrative number of participants per arm under which the assumed effect would have an 80% chance of meeting the specified statistical threshold, if the planning assumptions held. The 90% column applies the same idea at a higher target chance. These values are not guarantees about a future study: actual precision and results depend on the true effect, variability, design, and analysis.

Smaller target effects require more participants per arm in these planning scenarios. That pattern is expected because a smaller difference is harder to distinguish from random variation. The registry record does not report the trial's own original sample-size assumptions or justification, so the illustrations should not be interpreted as evidence about how the completed trial was designed.

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 = -14.200; pooled standard deviation σ = 29.166.
2. n = 2 × (1.96 + 0.842)² × 29.166² / -14.200² = 66.2, rounded up to 67 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 <- -14.2                      # observed difference in means
sigma <- 29.1657465056317                     # 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(114127)
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 = 67 (80% power), 89 (90% power)
 75% of observed effect: n per arm = 118 (80% power), 158 (90% power)
 50% of observed effect: n per arm = 265 (80% power), 355 (90% power)
power.t.test: 67.2 
Simulated power at n = 67 per arm: 0.797 (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 reports the sponsor's comparison as a two-sided t-test with p <0.132 for the two Phase 2 groups. It does not give the sponsor's estimated mean difference, confidence interval, test statistic, or further analysis details. The sponsor result therefore indicates a reported test outcome but provides limited information about the magnitude and precision of the comparison.

A separate unadjusted recalculation from the posted group summaries gives a mean difference of -14.2 and a 95% confidence interval from -35.69 to 7.286, with p = 0.195. This recalculation uses the posted means and standard errors and cannot reproduce any unreported adjustments, analysis-population choices, or other features of the sponsor's analysis. The registry record does not explain why the reported p-values differ.

The record does not flag a discrepancy between the posted summaries and the reported analysis. Even so, the two results should be kept distinct: one is the sponsor's reported test, and the other is an unadjusted recalculation from summary statistics. The sponsor's analysis cannot be fully evaluated from the information reported, while the recalculation has the limitations of relying only on those summaries.

15. Multiplicity, Missing Data and Analysis Populations

The registry record does not report whether the primary outcome was the sole confirmatory endpoint, whether secondary outcomes were tested under a prespecified hierarchy, or whether statistical significance thresholds were adjusted for multiple comparisons. Multiplicity matters when several outcomes, timepoints, or analytical choices are examined: the more opportunities there are to find a nominally unusual result, the less straightforward it is to interpret an unadjusted p-value as confirmatory evidence. No adjustment should be assumed in the absence of a reported method.

The registry lists a primary LSAS outcome and a secondary CGI-S outcome, with analysis results for both. It does not report whether other analyses were conducted but not posted, how the outcomes were prioritized, or whether any subgroup analyses were planned. The number of posted analyses does not establish the full scope of analyses considered. Consequently, the p-values are best read as reported or recalculated results without assuming control of the overall false-positive risk.

Missing-data handling is also not reported in the registry record. The posted group counts for the primary outcome are 13 and 15, but the record does not state whether these are randomized counts, observed-outcome counts, or counts in a particular analysis population. It does not describe imputation, likelihood-based methods, complete-case analysis, or sensitivity analyses. The effect of missing outcomes on the estimates therefore cannot be evaluated.

The registry record does not report the analysis population definition, such as whether the primary analysis followed randomized assignment regardless of treatment adherence. It also does not describe protocol deviations or exclusions. Without these details, the analysis cannot be reconstructed and the potential for bias related to missing data or population selection remains uncertain. These are reporting limitations, not evidence that a particular method was or was not used.

16. Safety Results

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

Other (non-serious) adverse events

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

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: elevated liver function tests (7.7%), study medication overdose (7.7%).

The registry reports serious adverse events for both Phase 2 groups. In the Duloxetine 60mg/Day + Placebo for 18 Weeks group, 2 serious adverse events are reported among 13 at risk. In the Duloxetine 120mg/Day for 18 Weeks group, 0 are reported among 15 at risk. The record lists Elevated Liver Function Tests and Study medication overdose as serious-event terms, each with 7.7%. The registry does not provide further event narratives or establish whether these terms correspond to all serious events reported.

Other adverse events are listed for 0 among 13 at risk in the 60mg/day plus placebo group and 2 among 15 at risk in the 120mg/day group. The record does not report deaths for these groups. These counts describe the registry entries; they do not establish a comparative safety profile. Sparse event counts are difficult to interpret, and the record does not provide detail about severity, timing, relatedness, duration, or the method for collecting events.

Serious adverse-event counts should be distinguished from all adverse events and from a formal statistical comparison. The registry record does not report a comparative safety analysis or enough information to determine whether the groups were similarly monitored. These figures are descriptive and should not be interpreted as evidence that one regimen is safer.

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

The registry shows that, among participants represented in the posted summaries, the mean LSAS score was lower in the 120mg/day group than in the 60mg/day plus placebo group at the stated 6 months. The unadjusted recalculated difference is uncertain: its 95% confidence interval includes zero and covers values in both directions. The sponsor's reported two-sided test is p <0.132, while the recalculated p-value is 0.195. The record does not explain the difference between those results.

These data do not by themselves establish that the regimens differ, that any observed difference is clinically important, or that one regimen should be preferred. Statistical uncertainty is substantial, and the record omits details needed to assess the analysis population, missing outcomes, and prespecified methods. The posted means are group summaries and do not reveal individual responses or score distributions.

The outcome is a symptom scale with suggested interpretation ranges, but the record does not report a threshold for meaningful change or the number of participants who improved by a meaningful amount. A group mean cannot determine whether a particular participant experienced benefit or harm. The reported safety counts are also limited descriptive data, not a complete basis for comparing safety.

The results should therefore be understood as a registry-based account of a small randomized comparison, not as a definitive statement about efficacy or safety. Further interpretation would require fuller reporting of the protocol, participant flow, outcome availability, analysis methods, and the sponsor's estimated effect with its uncertainty.

18. Why This Trial Matters Statistically

This trial is statistically instructive because it pairs a randomized comparison with a continuous symptom measure and reports group means, standard errors, and a sponsor test result. Those ingredients allow an unadjusted comparison to be recalculated from summary statistics, but the recalculation remains distinct from the sponsor's analysis. The contrast highlights why readers need both an estimate and an account of how it was obtained.

The confidence interval is especially important. The point estimate alone might appear to suggest a sizable difference in mean LSAS scores, but the interval extends across zero and is compatible with differences in either direction. This illustrates the distinction between the observed estimate and the precision with which an effect is known. A p-value alone would not communicate that range of plausible values.

The registry also presents an example of how reporting omissions constrain interpretation. The sponsor's analysis is identified as a two-sided t-test, but its estimate and confidence interval are not reported. Details on missing data, analysis populations, multiplicity, and the original sample-size rationale are also absent. These omissions make it difficult to reproduce the analysis or judge how closely it followed a prespecified plan.

Finally, the planning illustrations show how sample-size requirements change when the assumed target effect is made smaller. They are planning scenarios based on the observed effect, not evidence that the completed trial had a particular power. The distinction between design planning, observed estimates, and post hoc calculations is central to sound interpretation of clinical-trial statistics.

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 leaves important features of this trial unclear. Although it identifies the study as Phase 3, the intervention descriptions refer to a Phase 1 open trial and a Phase 2 randomized comparison. The record does not fully explain the sequence of phases, participant flow between them, or how the posted outcome population was defined. The primary outcome timeframe is listed as 6 months, while the randomized regimens are described as lasting 18 Weeks; the relation between treatment duration and outcome assessment is not reported.

Baseline participant characteristics, eligibility criteria, allocation procedures, concealment, and masking details are not reported in the outcome information. The record does not say whether the posted group counts represent all randomized participants or only participants with available outcome data. Missing-data methods, protocol deviations, and analysis-population definitions are also not reported, limiting assessment of potential bias.

The primary endpoint description does not specify which LSAS total score was analyzed. The record provides group means and standard errors, but not participant-level distributions or change-from-baseline summaries. It does not report the sponsor's effect estimate or confidence interval, nor explain the difference between the sponsor's p-value and the unadjusted recalculation. Multiplicity methods and the original sample-size justification are not reported.

The safety information is sparse and does not provide event timing, relatedness, severity details, or a comparative analysis. These gaps do not prove that a feature was absent from the actual study; they mean it is not reported in the registry record. The available information supports cautious description, not a full appraisal of trial conduct or results.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: Duloxetine for Social Anxiety Disorder: Prediction of Long Term Outcome — NCT00114127.
• Registry record last updated 2014-06-10; status last verified 2014-06.
• Results first posted 2012-06-07.
• 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

This completed randomized, double-masked, parallel Phase 3 record concerns duloxetine for social anxiety disorder and lists enrollment of 28. The posted Phase 2 comparison is Duloxetine 60mg/Day + Placebo for 18 Weeks versus Duloxetine 120mg/Day for 18 Weeks. The primary outcome is the mean LSAS score at 6 months.

The posted LSAS means are 67.9 and 53.7, respectively. The unadjusted recalculated mean difference is -14.2, with a 95% confidence interval from -35.69 to 7.286 and p = 0.195. The sponsor reports a two-sided t-test with p <0.132. The registry does not explain the difference between these results.

Important details about the analysis population, missing data, multiplicity, and original sample-size rationale are not reported in the registry record. The results are best read as uncertain descriptive evidence from the posted comparison, rather than a definitive estimate of efficacy or safety.