Effect of Duloxetine on Opioid Use After Total Knee Arthroplasty: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Effect of Duloxetine on Opioid Use After Total Knee Arthroplasty (NCT03271151), a randomized phase 4 study. The primary outcome is Opioid Use.

2. Clinical Question

The clinical question concerns opioid use over the postoperative period up to the specified assessment, rather than pain intensity alone. The registry lists Pain Phenotype as a secondary outcome, but the primary outcome is cumulative opioid use, measured in oral morphine equivalents in milligrams. This common scale expresses opioid use as an equivalent amount of oral morphine, allowing use to be summarized despite possible differences among opioid medications.

The trial is described as randomized and parallel, with placebo as the comparator. Randomization is intended to make the treatment groups comparable on average, while a placebo comparison helps separate differences associated with assignment from differences that could arise under placebo. Those design features do not by themselves establish how well groups were balanced in this trial or how the analysis handled missing observations; those details are not reported in the registry record.

The posted result addresses the registered endpoint at postoperative day 14. It does not, by itself, answer whether the groups differed at other times, on pain outcomes, or on longer-term outcomes. The actual mix of opioid-use histories and other participant characteristics is not reported in the registry record, so the population represented by the analyzed results cannot be described more specifically.

3. Trial Design

This was a Phase 4, randomized, parallel-group trial sponsored by Hospital for Special Surgery, New York. The registry record lists an enrollment of 160 and a completed status. It gives a start date of 2017-09-28 and a primary completion date of 2020-12-03. These dates describe the record’s trial timeline; they do not provide details about when individual participants were recruited or assessed.

The official title calls the study a double-blinded randomized control trial, while the registry’s masking entry says quadruple. These descriptions are both present in the record. It does not specify which groups of people were masked, how masking was maintained, or whether masking was assessed. The record identifies duloxetine, named as Cymbalta, and placebo as the interventions.

The primary outcome is listed as opioid use, measured as cumulative oral morphine equivalents in milligrams at postoperative day 14. The posted results report a mean and standard deviation for each group. The duloxetine group has 78 participants in this outcome summary, and the placebo group has 77. These are the outcome denominators, not a full account of participant flow. Reasons for any difference between those counts and the enrollment figure are not reported in the registry record.

The brief summary describes post-discharge pain as a concern after knee arthroplasty and notes a goal of reducing opioid use. It also says the planned enrollment included non-opioid users and up to 15 chronic opioid users. The number of chronic opioid users actually enrolled or included in the outcome analysis is not reported in the registry record.

The record lists Pain Phenotype as a secondary outcome. Although it reports three outcome measures, no analyses are posted for that secondary outcome. Other details needed to fully assess conduct and analysis—including the treatment schedule, participant flow, analysis population definitions, and the trial’s original sample-size rationale—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

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 Opioid Use, measured over Post-operative day 14.

The primary endpoint is cumulative opioid use, expressed in oral morphine equivalents in milligrams, at postoperative day 14. Cumulative use summarizes opioid consumption over a period rather than capturing a single dose or a single moment of use. Expressing different opioids on a morphine-equivalent scale provides a shared unit for summarizing consumption.

In this trial, the registry reports the endpoint as a continuous measure and summarizes each group with a mean and standard deviation. The mean describes the group’s average cumulative use; the standard deviation describes how spread out individual values are around that average. The posted mean is 288.0 mg for the duloxetine group and 432.5 mg for the placebo group. The corresponding standard deviations are 226.0 mg and 374.0 mg.

Opioid consumption is the registered primary outcome, while pain phenotype is listed as a secondary outcome. The registry summary explains the interest in post-discharge pain and opioid use, but the endpoint is not itself a direct measure of pain intensity, function, or longer-term recovery. Those are distinct questions and should not be inferred from this measure alone.

General: a cumulative measure can reflect use across the period covered by its definition, but interpretation depends on how the period is defined and how observations are collected. The registry specifies postoperative day 14 as the timeframe and describes the measure as cumulative morphine equivalents. More detailed collection rules, including how missing or uncertain use was handled, are not reported in the registry record.

7. Results Posted in the Registry

• Duloxetine ("Cymbalta"): mean 288.00 versus 432.50 on control; difference -144.50 (95% CI -241.94 to -47.06), standardized effect d = -0.47; z-test p = 0.004.

8. Statistical Interpretation

9. Statistical Methods Explained

The registry provides a continuous primary outcome, group means and standard deviations, and a comparison expressed as a mean difference. These summaries support a straightforward reading of the main result, but the record does not describe the full statistical analysis plan. The explanations below distinguish the posted quantities from general statistical concepts.

What does a mean difference describe?

A mean difference compares the average outcome in one group with the average in another. Here the reported contrast is duloxetine relative to placebo: the group means are 288.0 mg and 432.5 mg, respectively, and the posted difference is -144.5 mg. A negative value means the first group’s mean is lower under this contrast. This is a comparison of averages; it does not describe every participant’s experience or the whole shape of the outcome distributions.

General: a mean difference is most directly interpretable when the outcome is measured on the same scale in both groups and the meaning of the contrast is clear. The registry identifies oral morphine equivalents in milligrams and gives the group ordering, so the direction of the reported contrast can be read. The record does not describe any covariate adjustment, transformation, or alternative model for the primary result.

What does the standard deviation add?

The standard deviation describes variation among observed values within a group. The reported values are 226.0 mg for duloxetine and 374.0 mg for placebo. They provide context for the means: the outcome varies among participants, and the group averages are not individual-level predictions. A standard deviation is not the same as a standard error or a confidence interval. It describes spread in the data, while standard errors and intervals describe uncertainty about an estimated quantity.

General: continuous outcomes can be skewed or contain unusually large values. Cumulative medication-use measures may have a distribution that is not well described by a symmetric shape, but the registry record does not report distributional diagnostics for this trial. The standard deviations alone do not establish whether a normal model was appropriate.

How should the p-value be read?

The posted comparison gives an unadjusted p-value of 0.004. In general, a p-value describes how incompatible the observed data, or more extreme data, are with a specified null model, assuming that model and the test’s assumptions hold. It is not the probability that the null hypothesis is true, nor is it a measure of the size or clinical importance of the difference. The registry does not identify the exact test used for the posted comparison, so the test details and assumptions are not reported in the registry record.

The word unadjusted matters because it indicates that the reported p-value is not presented as corrected for multiple comparisons. The record lists a primary outcome and a secondary outcome, but it does not report a multiplicity strategy or whether other analyses were conducted outside the posted results. A p-value should therefore be interpreted alongside the estimated difference, its interval, and the available information about the analysis.

Why can over-dispersion matter?

General: over-dispersion occurs when the observed variability is greater than a statistical model expects. If a model assumes less variability than is present, standard errors can be too small and uncertainty can be understated. For count outcomes, over-dispersion is often considered in relation to a count model; this trial’s primary endpoint is reported as a continuous measure, not a count. The group standard deviations show variation, but the registry does not report a model assessment or an over-dispersion analysis. It would therefore be inappropriate to claim that over-dispersion was present or addressed.

What is known about the analysis?

The posted outcome summaries include 78 participants for duloxetine and 77 for placebo. The comparison reports a mean difference, interval, and unadjusted p-value. The record does not report whether the analysis followed an intention-to-treat principle, whether participants were analyzed according to assignment, how adherence was considered, or how missing outcome values were handled. The analysis population and any imputation or sensitivity analyses are not reported in the registry record.

The record reports zero analyses in its analysis listing, despite posting a primary outcome comparison. This means the registry record does not give a separately described analysis entry from which to recover detailed model specifications. The exact confidence level for the interval, the test procedure, and any covariate adjustment are not reported in the registry record. The posted statistics can still be described, but the underlying analysis choices cannot be reconstructed from these entries alone.

General: randomization is designed to support fair comparisons by assigning participants to groups by chance. It does not guarantee identical groups, ensure that all randomized participants contribute an endpoint, or replace transparent reporting of analysis choices. Assessment of these issues requires details about participant flow and analysis populations, which are not reported in the registry record.

How should the result be kept in context?

The primary result concerns a single endpoint at postoperative day 14. The record also lists Pain Phenotype as a secondary outcome, but no analysis for that outcome is posted. Statistical evidence for the primary comparison does not automatically establish findings for secondary outcomes, other timepoints, or other populations. The registry’s displayed result is useful for describing this endpoint, while unanswered questions about the analysis and broader outcomes remain.

10. Confidence Intervals and Precision

The posted comparison gives a mean difference of -144.5 mg, with an interval from -241.9 to -47.06. The interval expresses uncertainty around the estimated difference under the analysis used. The registry record does not report the interval’s confidence level, so it should not be labelled with a particular confidence percentage. The method used to construct it is also not reported in the registry record.

The interval spans values that all have the same negative direction under the posted contrast of duloxetine relative to placebo. It nevertheless covers a range of possible differences, rather than identifying one exact value. The estimate is therefore more informative when read together with the interval than when read alone. The interval does not show the range of individual participants’ outcomes; the group standard deviations address variability in observed values, while the interval concerns uncertainty in the group comparison.

General: precision refers to how narrowly an estimate is determined by the data and method. A narrower interval generally indicates greater precision, while a wider interval leaves more uncertainty about the size of the estimate. Precision is distinct from whether a result is important in practice. The registry does not report a threshold for a clinically important difference, so the clinical importance of the estimated difference cannot be determined from the statistical interval alone.

The endpoint summaries include 78 participants in the duloxetine group and 77 in the placebo group. The registry does not explain why those outcome counts differ from the enrollment figure of 160, or whether missingness affected the estimate. Such information would help assess how representative the analyzed observations are of all randomized participants. The interval should consequently be read as precision for the posted comparison, not as a complete account of uncertainty arising from trial conduct or missing data.

The unadjusted p-value is 0.004, but a p-value does not replace the interval or describe the estimate’s precision by itself. The confidence level, analysis specification, and multiplicity handling are not reported in the registry record. These omissions limit how fully the interval and p-value can be evaluated.

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 sample-size table is a planning illustration based on the observed effect, not a report of the trial’s original design calculation. Each entry gives participants per arm, not a total across both arms. The assumptions listed for the illustration are a two-sided significance level of 0.05, equal allocation between the two arms, a normal approximation, and an effect to detect set as a stated fraction of the observed effect. The trial’s own original sample-size assumptions and justification are not reported in the registry record.

The table considers planning for 100% of the observed effect, 75% of the observed effect, and 50% of the observed effect. For 100% of the observed effect, the entries are 72 participants per arm for 80% power and 96 participants per arm for 90% power. For 75% of the observed effect, they are 128 participants per arm for 80% power and 171 participants per arm for 90% power. For 50% of the observed effect, they are 287 participants per arm for 80% power and 384 participants per arm for 90% power.

The 80% power column describes the number of participants per arm that the planning calculation estimates would be needed for an 80% chance of detecting the specified effect under its assumptions. The 90% power column describes the corresponding planning target for a 90% chance. These are design calculations, not a measure of how likely the posted result is to be correct. Smaller effects need more participants to detect under the stated assumptions, which is why the entries increase as the assumed fraction of the observed effect becomes smaller.

The curve should be read as a visual presentation of the planning relationship, not as additional observed trial evidence. The displayed calculations are planning illustrations, 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 = -144.500; pooled standard deviation σ = 308.522.
2. n = 2 × (1.96 + 0.842)² × 308.522² / -144.500² = 71.6, rounded up to 72 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 <- -144.5                      # observed difference in means
sigma <- 308.521966675906                     # 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(3271151)
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 = 72 (80% power), 96 (90% power)
 75% of observed effect: n per arm = 128 (80% power), 171 (90% power)
 50% of observed effect: n per arm = 287 (80% power), 384 (90% power)
power.t.test: 72.5 
Simulated power at n = 72 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 does not report a plan for handling multiplicity. It identifies Opioid Use as the primary outcome and Pain Phenotype as a secondary outcome, but does not state whether the primary comparison was protected from other comparisons by a prespecified hierarchy, whether a correction was planned, or whether additional outcomes or analyses were considered. The posted p-value is explicitly unadjusted. Without a stated multiplicity strategy, it cannot be interpreted as adjusted evidence across a family of outcomes.

The record lists one secondary outcome and reports three outcome measures, but no analyses are posted for the secondary outcome. The number of posted analyses is zero. This does not explain whether unposted analyses were conducted, whether the secondary outcome was collected, or why its results are absent. Those details are not reported in the registry record.

Missing-data methods are also not reported in the registry record. The primary outcome summaries include 78 participants assigned to duloxetine and 77 assigned to placebo, while total enrollment is listed as 160. The record does not explain the difference between the enrollment figure and these outcome counts. It does not say how many participants lacked an outcome, why any values were unavailable, whether missingness differed by group, or whether imputation or sensitivity analyses were used.

The analysis population is not defined. The registry does not state whether the primary comparison included all randomized participants, only participants with observed endpoint data, or another population. It also does not describe how treatment discontinuation, non-adherence, or protocol deviations were handled. These omissions matter because analysis population and missing-data choices can affect interpretation of a randomized comparison.

General: multiplicity can increase the chance of at least one apparently notable result when many hypotheses are tested. Missing-data methods can also affect estimates if participants with unavailable outcomes differ from those with observed outcomes. These are general concerns, not claims about what occurred in this trial. The registry record does not provide enough information to determine whether either issue altered the posted result.

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 serious adverse event summary lists 80 participants at risk in each group. It reports zero serious adverse events in the duloxetine group and zero in the placebo group, zero deaths in each group, and other adverse event counts of 2 for duloxetine and 3 for placebo. These counts are presented as recorded; the registry does not provide event descriptions.

The safety denominators differ from the primary outcome denominators, which are 78 for duloxetine and 77 for placebo. The registry record does not explain this difference or provide further information about who contributed to each safety summary. Do not treat the safety counts as if they use the primary endpoint analysis population.

Counts alone do not describe event severity beyond the categories shown, timing, duration, or whether events were judged related to the intervention. The record does not report the definitions used for the serious and other adverse-event categories in this summary. Zero events in a group means none were recorded in that category in the posted summary; it does not establish that no such event could occur. These figures should be read as a limited account of the events listed, not as a complete safety assessment.

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

The posted primary results show different group means for cumulative opioid use at postoperative day 14: 288.0 mg for duloxetine and 432.5 mg for placebo. The reported mean difference, with duloxetine compared with placebo, is -144.5 mg. Its interval runs from -241.9 to -47.06, and the unadjusted p-value is 0.004. These statistics describe the posted comparison and its uncertainty under the analysis used.

They do not show that every participant in the duloxetine group used less opioid than every participant in the placebo group. The reported standard deviations show variation within both groups, and group averages do not describe individual responses. Nor does the result establish a difference in pain intensity or other outcomes: the primary endpoint is cumulative opioid use, and the registry does not post an analysis for the listed secondary outcome Pain Phenotype.

The result also does not establish the size or persistence of any difference outside the registered timeframe. The primary endpoint is specified at postoperative day 14. Findings at other timepoints, later opioid use, function, or longer-term outcomes are not reported in this primary comparison. The registry record does not give enough detail to assess the analysis population, missing-data handling, or multiplicity strategy, all of which are relevant to a comprehensive interpretation.

Randomization and placebo comparison are features of the design, but interpreting a randomized estimate also requires information about trial conduct and analysis. The registry’s masking entries differ in wording: the official title says double-blinded, while the masking entry says quadruple. The record does not specify the masked roles or provide details to resolve that difference.

General: statistical evidence and clinical importance are separate questions. A p-value summarizes evidence relative to a null model; it does not establish that an estimated difference matters to patients. No threshold for a clinically important difference is reported in the registry record. The findings should therefore be treated as a result for this endpoint and timepoint, rather than as a complete assessment of treatment effects or a basis for individual care decisions.

16. Why This Trial Matters Statistically

This trial illustrates why a clinical study result should be read through both its endpoint and its statistical summary. The endpoint is cumulative oral morphine equivalent use at postoperative day 14, not a general measure of pain or recovery. Keeping that distinction visible prevents a difference in one outcome from being treated as evidence about outcomes that were not analyzed or reported.

The group means and standard deviations provide complementary information. The means show the average values in the two groups, while the standard deviations show that individual outcomes vary within each group. The mean difference and interval then summarize the between-group comparison and its uncertainty. Together, these quantities give a fuller description than a p-value alone.

The posted unadjusted p-value is 0.004, but the registry does not report a multiplicity strategy, analysis population, or missing-data method. The result therefore also demonstrates the importance of transparent reporting: a numerical finding can be stated precisely while important details about how it was generated remain unavailable. The difference between the outcome and safety denominators is another reason to consult participant-flow and analysis information, which is not reported in the registry record.

The sample-size illustration emphasizes that design targets depend on the effect size one intends to detect. Its figures are planning calculations based on the observed effect and specified assumptions, not a reconstruction of the trial’s original planning. This distinction matters when interpreting a completed trial: planning numbers are not evidence that the study had a particular probability of finding its observed result.

More broadly, this record offers a concise example of how to separate the observed comparison, uncertainty, and unanswered questions. The statistical result is informative for the listed endpoint, but its interpretation remains bounded by the reported timepoint, missing analysis details, and absence of posted results for the secondary outcome.

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 leaves several trial-specific details unclear. It does not report the actual number of chronic opioid users enrolled, even though the summary says the trial planned to include up to 15 chronic opioid users. This limits description of the population represented by the posted outcome. Other participant characteristics and eligibility details are also not reported in the registry record.

The primary outcome summaries include 78 duloxetine participants and 77 placebo participants, while enrollment is listed as 160. The record does not explain the outcome counts, participant flow, or reasons that observations may not have contributed to the endpoint summary. The safety summary uses 80 participants at risk in each group, and the record does not explain how these denominators relate to the primary outcome counts.

The registry does not report the treatment schedule in enough detail to describe it. The displayed duloxetine arm description is incomplete, and the placebo description does not fill in the missing intervention details. The masking information also differs between the official title, which says double-blinded, and the masking entry, which says quadruple; the masked roles are not specified.

The posted primary comparison lacks a stated confidence level, full test specification, analysis population, missing-data approach, and multiplicity plan. No analyses are posted for the secondary outcome Pain Phenotype. The record also does not report a threshold for clinical importance or results for other timepoints. These limits do not negate the posted numbers, but they constrain what can be concluded from them and what can be checked independently.

18. Related Tutorials

19. Related Statistical Calculators

20. Sources

• ClinicalTrials.gov: Effect of Duloxetine on Opioid Use After Total Knee Arthroplasty — NCT03271151.
• Registry record last updated 2024-12-27; status last verified 2024-09.
• Results first posted 2022-04-26.
• 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 4 randomized, parallel-group trial sponsored by Hospital for Special Surgery, New York, comparing duloxetine, listed as Cymbalta, with placebo after total knee arthroplasty. The record lists enrollment of 160 and a primary completion date of 2020-12-03.

The primary outcome is cumulative oral morphine equivalent use in milligrams at postoperative day 14. The posted means are 288.0 mg for duloxetine and 432.5 mg for placebo. The reported mean difference is -144.5 mg, with an interval from -241.9 to -47.06 and an unadjusted p-value of 0.004.

The registry also lists Pain Phenotype as a secondary outcome but posts no analysis for it. Important details—including missing-data methods, analysis populations, multiplicity handling, the original sample-size rationale, and the confidence level for the interval—are not reported in the registry record.