Safety and Effectiveness of Omega 3-Fatty Acids, EPA Versus DHA, for the Treatment of Major Depression: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Safety and Effectiveness of Omega 3-Fatty Acids, EPA Versus DHA, for the Treatment of Major Depression (NCT00361374), a randomized phase 3 study. The primary outcome is Score on a Depression Severity Rating Scale Over Eight Weeks.

2. Clinical Question

3. Trial Design

This was a phase 3, randomized, quadruple-masked, parallel-group trial sponsored by Massachusetts General Hospital. The study status is completed. The registry record gives a start date of 2006-07 and a primary completion date of 2013-02. It describes the study as examining differences in the effects of EPA and DHA in the treatment of major depressive disorder, and lists placebo as a third arm.

The registry record reports that 196 of 389 screened patients were randomized. The arm-level outcome counts were 60 for EPA, 58 for DHA, and 59 for placebo. These are the numbers associated with the posted primary outcome results, and they total 177 evaluable subjects as reported in the outcome description. The same description says that 19 subjects dropped out before completing at least one post-baseline visit. The record does not detail the timing or reasons for those dropouts.

The supplement arms received EPA Omega-3, 1g/day, or DHA Omega-3, 1g/day. The placebo was a capsule containing 980mg soybean oil. The primary outcome was change in Hamilton D score over 8 weeks, with scores obtained every 2 weeks. The record names the analysis approach as mixed model repeated measures and describes covariates and a covariance structure. Allocation concealment, adherence monitoring, and any additional protocol procedures are not reported in the registry record. The direct contrast between EPA and DHA is not among the posted results; the reported comparisons are each supplement arm against placebo.

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: We randomized 196 of 389 screened patients. Nineteen subjects dropped out before completing at least one post-baseline visit, leaving 177 evaluable subjects

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: We randomized 196 of 389 screened patients. Nineteen subjects dropped out before completing at least one post-baseline visit, leaving 177 evaluable subjects for analysis.

6. Primary Endpoint

The registry lists the primary outcome as Score on a Depression Severity Rating Scale Over Eight Weeks, measured over 8 weeks.

The primary endpoint was change in score on the 17-item Hamilton D depression severity rating scale over 8 weeks. This scale sums responses across 17 items to assess depressive severity. Its possible total scores range from 0-52, with higher scores indicating greater depressive severity. The registry record describes scores of 7 or less as indicative of full remission, scores of 8-15 as mild depression, scores of 16-25 as moderate depression, and scores of 25 or greater as severe depression. The categories as written overlap at 25.

The endpoint is a continuous measure: the analysis focuses on average score change rather than only whether participants crossed a particular threshold. The record says scores were collected every 2 weeks over the 8-week treatment period, allowing the analysis to use measurements across visits rather than relying only on one endpoint measurement. The primary description specifies change from baseline to week 8 as the treatment effect of interest. A mean change conveys the average direction and magnitude of score change in each arm, while a comparison of mean changes describes the difference between an active arm and placebo.

Repeated measurement can make use of the pattern of observations over time, but interpretation still depends on how the model and missing observations are handled. The record names the model and some of its components; details about missing-data assumptions are not reported in the registry record.

7. Results Posted in the Registry

• Eicosapentaenoic Acid (EPA): mean -10.34 versus -9.49 on control; difference -0.85 (95% CI -2.55 to 0.85), standardized effect d = -0.18; z-test p = 0.328.
• Docosahexaenoic Acid (DHA): mean -9.26 versus -9.49 on control; difference 0.23 (95% CI -1.47 to 1.93), standardized effect d = 0.05; z-test p = 0.791.

8. Statistical Interpretation

9. Statistical Methods Explained

What does a mean difference represent?

For a continuous outcome, a mean difference expresses how far apart the average values are between two groups, on the original measurement scale. In this trial, the posted contrasts concern mean change in Hamilton D score for an active supplement arm compared with placebo. The EPA estimate is -0.85 units and the DHA estimate is 0.23 units. The negative EPA estimate indicates a lower change value for EPA than placebo under the reported subtraction order; the positive DHA estimate indicates a higher change value for DHA than placebo. The direction must be read alongside the definition of the outcome and the comparison. These contrasts do not directly compare EPA with DHA.

Why use a repeated-measures model?

General statistical explanation: when an outcome is measured repeatedly in the same participant, observations from that participant are related rather than independent. A repeated-measures model can represent this within-participant dependence while assessing patterns across treatment groups and study visits. The registry record says mixed model repeated measures analysis (MMRM) was used to examine treatment group effects on changes from baseline to week 8. It states that subjects were included as a random effect and treatment group and study week as fixed effects. In general, a random effect allows the model to account for participant-level variation, while fixed effects represent the specified group and time effects being assessed.

The record says an auto-regressive covariance structure was used because it provided the best fit to the data. General statistical explanation: a covariance structure specifies how measurements are related across visits. An auto-regressive structure represents a pattern in which measurements closer together in time tend to be more strongly related than measurements farther apart. The record does not provide model fit statistics or compare the selected structure with alternatives, so the stated reason cannot be independently assessed from the registry information.

What do the covariates do?

The registry record says site and baseline score were included as covariates in all models. General statistical explanation: covariate adjustment accounts for specified participant or study characteristics when estimating a treatment effect. Baseline score describes the starting value on the scale, while site represents the study location. Adjustment can improve the precision of an estimate or account for differences associated with those characteristics; the actual effect on the estimate depends on the observed data and model. The registry record does not post adjusted model estimates, so the extent to which these covariates changed the treatment comparisons is not reported.

How should the recalculated comparisons be read?

The posted arm summaries give means and standard errors. The comparison estimates are simple, unadjusted recalculations from those summaries, rather than reported adjusted MMRM results. Their p-values are likewise unadjusted. A p-value describes how compatible the observed comparison is with a specified no-difference model under the assumptions of the calculation; it is not the probability that a treatment effect is true, nor a measure of the size or clinical importance of a difference. The estimates and confidence intervals provide more direct information about possible magnitudes and directions.

General statistical explanation: standard error describes the estimated sampling variability of a mean estimate. It is not the standard deviation of individual participant scores. The reported standard errors for the arm means are 0.62 for EPA, 0.62 for DHA, and 0.61 for placebo. The mean differences and confidence intervals should be interpreted with the reported group sizes and the assumptions behind the recalculation in mind. Since the registry record describes an adjusted repeated-measures model for its primary analysis but does not post that model's estimates, the recalculated comparisons are a limited, unadjusted view of the results.

What is not specified?

The record does not report the original statistical analysis plan, the trial's sample-size justification, whether any interim analyses were planned or conducted, or multiplicity procedures. It also does not report details of model diagnostics, the precise analysis population definition, or the assumptions used to address missing outcome measurements. Those omissions limit the ability to reconstruct the full analysis from the registry record. General statistical explanation: a model's result can depend on its assumptions and on which measurements are included, so those details matter when evaluating a repeated-measures analysis. No unreported analysis choices can be inferred from the posted results.

10. Confidence Intervals and Precision

The confidence intervals describe a range of treatment differences compatible with the observed data under the calculation and its assumptions. They are not guarantees that the true difference lies within the interval, nor do they show the probability that one particular value is correct. Their width conveys precision: a wider interval leaves more uncertainty about the magnitude and direction of the comparison.

For EPA versus placebo, the estimated mean difference is -0.85 units. Its interval extends from -2.555 to a positive upper endpoint, so it includes zero and spans both negative and positive differences. The registry record's upper endpoint is not repeated numerically because it is reported with more than three decimal places. This interval is consistent with a range of possible effects that includes no difference; the estimate alone should not be read as a precise effect.

For DHA versus placebo, the estimated mean difference is 0.23 units and the interval runs from -1.475 to 1.935. It also includes zero and extends in both directions. The unadjusted p-values, 0.328 for EPA and 0.791 for DHA, are consistent with the fact that neither recalculated interval excludes no difference. They do not establish that the groups are equivalent or that the true effect is exactly zero.

The intervals and p-values are based on unadjusted comparisons recalculated from posted summary statistics. The registry record describes an adjusted MMRM analysis but does not report its estimates or confidence intervals. Precision for the adjusted primary model therefore cannot be evaluated from the posted results. Differences between estimates from the unadjusted calculations and a covariate-adjusted repeated-measures model are possible in general; the record does not quantify such a difference.

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 for these continuous-outcome comparisons, not total event counts. For example, planning around 100% of the observed EPA-placebo effect gives 490 participants per arm for 80% power and 655 participants per arm for 90% power. At 75% of the observed effect, the corresponding figures are 870 and 1,165 participants per arm; at 50%, they are 1,957 and 2,620. For the DHA-placebo comparison, the 100% figures are 6,565 participants per arm for 80% power and 8,789 participants per arm for 90% power. The larger estimates reflect the very small observed DHA-placebo difference used in that illustration.

The planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. Each row uses the stated fraction of the observed effect as the effect to detect. The 80% power column represents a planning target for the chance of detecting an effect of that specified size under the calculation assumptions if the effect exists. The 90% column represents a higher such target. They are not probabilities that the trial result is correct.

Smaller effects require more participants to detect under otherwise unchanged assumptions. The estimates are planning illustrations based on observed effects, not the trial's original sample-size rationale; that rationale is not reported in the registry record. This is not observed power.

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, Δ is the difference in means to detect and σ is the common standard deviation.

Worked calculation for this trial

1. Observed difference Δ = -0.850; pooled standard deviation σ = 4.745.
2. n = 2 × (1.96 + 0.842)² × 4.745² / -0.850² = 489.2, rounded up to 490 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 <- -0.85                      # observed difference in means
sigma <- 4.74485974683067                     # 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(361374)
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 = 490 (80% power), 655 (90% power)
 75% of observed effect: n per arm = 870 (80% power), 1,165 (90% power)
 50% of observed effect: n per arm = 1,957 (80% power), 2,620 (90% power)
power.t.test: 490.1 
Simulated power at n = 490 per arm: 0.799 (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.

13. Multiplicity, Missing Data and Analysis Populations

The registry record reports one primary outcome measure and no posted secondary analyses. It does not report whether there were additional planned outcomes or hypotheses beyond the posted endpoint, or whether a multiplicity adjustment was planned or used. A study may evaluate more than one treatment contrast or outcome; when multiple statistical tests are performed, the chance of obtaining at least one nominally unusual result can increase. General statistical explanation: multiplicity procedures are methods for controlling or describing that issue. No such procedure can be inferred from the unadjusted p-values reported for the two comparisons.

The primary outcome description states that 19 subjects dropped out before completing at least one post-baseline visit, leaving 177 evaluable subjects. The record does not report the reasons for dropout, the arm distribution of these subjects, the amount or pattern of missing measurements, or whether the dropouts differed from those who remained. It names MMRM but does not specify the assumptions used for missing observations or describe sensitivity analyses. General statistical explanation: repeated-measures models can use available observations under assumptions about the missing-data process; the validity of conclusions may depend on those assumptions. The registry information is not sufficient to judge how robust the results are to missing data.

The registry record reports arm-level outcome counts of 60 for EPA, 58 for DHA, and 59 for placebo. It does not give a full analysis-population definition or explain whether those counts reflect all randomized participants with any post-baseline measurement, a particular model population, or another inclusion rule. The record says 196 were randomized, but does not describe the treatment of every randomized participant in the primary analysis. The full analysis population and any per-protocol population are therefore not reported in the registry record.

14. 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 non-serious adverse event terms

The registry record lists serious adverse events affecting zero participants in the EPA group, zero in the DHA group, and zero in the placebo group. The numbers at risk were 60, 58, and 59, respectively. No serious adverse event terms are listed in the record. Deaths are not reported in the registry record.

For other adverse events, the registry lists 39 affected participants in the EPA group, 40 in the DHA group, and 43 in the placebo group. These are counts of participants affected, not a count of every event occurrence. The registry excerpt does not provide event descriptions, severity, timing, duration, or an account of how events were collected and classified. The figures therefore describe the reported participant counts, but do not allow detailed comparisons of specific types of events or assessment of their clinical context.

Zero reported serious events in these groups does not establish that no serious events could occur, and the counts for other adverse events should not be interpreted as evidence of comparative safety on their own. No medical conclusions follow from these counts.

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

The posted outcome data show average Hamilton D score changes of -10.34 for EPA, -9.26 for DHA, and -9.49 for placebo over the 8-week treatment period. The unadjusted EPA-placebo mean difference is -0.85 units; the unadjusted DHA-placebo mean difference is 0.23 units. Both confidence intervals include zero, and both comparisons have unadjusted p-values that do not distinguish the estimates from no difference under the calculations used. The intervals also allow for effects in either direction, so the estimates are not precise enough to identify a narrow range of differences.

These findings describe the posted summary data and unadjusted comparisons. They are not the adjusted MMRM results described in the primary outcome methods, because the registry record does not report those model estimates. The record also does not report a direct statistical comparison between EPA and DHA. It is therefore not possible to conclude from these posted contrasts whether EPA and DHA differ from each other.

A confidence interval that includes zero does not demonstrate equivalence or prove that there is no effect. Equivalence requires a defined margin and an analysis designed to assess whether differences fall within that margin; such a margin and analysis are not reported in the registry record. Likewise, a numerical change within each group does not isolate the effect of the assigned supplement, since the relevant treatment comparison is between groups.

The primary measure is a score of depressive severity, and the registry record provides scoring ranges and categories. The posted mean changes do not show how many individuals reached remission or moved between severity categories. Such participant-level threshold outcomes are not reported in the registry record. The data also do not establish effects beyond the 8-week outcome period described for the primary endpoint.

16. Why This Trial Matters Statistically

This trial is statistically informative because it separates three distinct questions that can be confused: whether average scores changed within each group, whether each supplement differed from placebo, and whether EPA differed from DHA. The registry gives arm-level mean changes and reports the two supplement-versus-placebo summaries, but does not report a direct EPA-versus-DHA contrast. The results therefore answer only part of the comparative question raised by the study.

It also illustrates why an estimate must be read together with its uncertainty. EPA has a negative mean difference, while DHA has a small positive mean difference, but both intervals include zero and extend to values in the opposite direction. The point estimates alone do not establish either the direction or the magnitude of a reproducible difference. The very large planning estimates for the DHA-placebo contrast reflect the small observed difference used in that planning illustration; they are not evidence about the trial's original design or a guarantee of future results.

The registry describes a repeated-measures model with participant-level random effects, treatment group and week effects, and adjustment for site and baseline score. That description helps readers understand the intended primary analysis. However, because the model's estimates are not posted, the unadjusted recalculations cannot fully represent the reported analysis plan in action. Missing-data details and multiplicity procedures are also not reported. These gaps matter for judging how the treatment comparisons were estimated and how confidently they can be interpreted.

Finally, the registry distinguishes serious from other adverse events, but the posted counts do not include enough detail to evaluate event types or clinical context. Careful statistical reading preserves what the record does show while avoiding stronger claims than its estimates, intervals, and analysis reporting 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 primary results available in the registry record are limited to one continuous outcome and do not include posted secondary analyses. The record describes an adjusted MMRM analysis but does not report its treatment-effect estimates, standard errors, confidence intervals, or model-based p-values. The available supplement-versus-placebo estimates are unadjusted recalculations from arm summaries, so they should not be mistaken for the full primary model results.

The record reports that 19 subjects dropped out before completing at least one post-baseline visit, but does not give dropout reasons, timing, group distribution, or missing-measurement patterns. It does not report the analysis population definition or sensitivity analyses for missing data. Consequently, the impact of attrition and the assumptions used to include participants with incomplete repeated measurements cannot be assessed from the posted information.

Although the study question concerns differences between EPA and DHA, the posted comparisons are each against placebo. A direct EPA-versus-DHA contrast is not reported in the registry record. The record also does not provide participant-level remission counts or transitions between the severity categories described for the scale. The mean score changes alone do not reveal how individual participants responded.

Further trial details that could help interpret the findings—including participant characteristics, adherence, allocation concealment, the original sample-size justification, interim analyses, and multiplicity methods—are not reported in the registry record. Safety information is also limited: counts of affected participants are posted, but detailed event terms, severity, timing, and deaths are not available. These limitations do not change the reported estimates, but they constrain the questions those estimates can answer.

18. Related Tutorials

19. Related Statistical Calculators

20. Sources

• ClinicalTrials.gov: Safety and Effectiveness of Omega 3-Fatty Acids, EPA Versus DHA, for the Treatment of Major Depression — NCT00361374.
• Registry record last updated 2014-07-18; status last verified 2014-07.
• Results first posted 2014-07-18.
• 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

The registry record describes a completed, phase 3, randomized, quadruple-masked, parallel-group study of EPA, DHA, and placebo in major depressive disorder. It reports that 196 of 389 screened patients were randomized. The primary endpoint was change in 17-item Hamilton D score over 8 weeks, with measurements every 2 weeks.

Arm-level mean changes were -10.34 for EPA, -9.26 for DHA, and -9.49 for placebo. Unadjusted recalculated mean differences versus placebo were -0.85 for EPA and 0.23 for DHA; both confidence intervals included zero. The record describes an adjusted repeated-measures analysis but does not post its estimates. It does not report a direct EPA-versus-DHA comparison or secondary analyses. Serious adverse events affected zero participants in each arm; deaths are not reported in the registry record.