A Study to Test How Well Empagliflozin Works in Chinese Patients With Type 2 Diabetes Who Already Take Insulin: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study to Test How Well Empagliflozin Works in Chinese Patients With Type 2 Diabetes Who Already Take Insulin (NCT04233801), a randomized phase 3 study. The primary outcome is Change From Baseline in Glycosylated Haemoglobin A1c (HbA1c) at Week 24.

2. Clinical Question

The registry record describes a randomized, parallel-group comparison involving three groups. Enrollment was 219. The central statistical question is the difference in mean change in HbA1c between each empagliflozin dose and placebo, rather than whether a change within an individual group alone is statistically distinguishable from zero. The posted sponsor analysis reports adjusted comparisons from a repeated-measures model. Those estimates describe the average between-group contrast at Week 24 under that model; they do not, by themselves, establish how the result applies beyond the enrolled population or treatment period.

3. Trial Design

This was a Phase III, randomized, placebo-controlled, parallel-group study in Chinese adults with type 2 diabetes whose blood sugar remained insufficiently controlled while they were taking insulin. Participants could also be taking up to two oral anti-diabetic agents. The registry lists enrollment of 219 and a completed status. The official title describes the study as double-blind, while the registry’s masking classification is triple; which specific people were masked is not reported in the registry record.

Participants were assigned by chance to empagliflozin 10 mg, empagliflozin 25 mg, or matching placebo. Each was taken orally once daily for 24 weeks alongside background insulin therapy. A 2-week open-label placebo run-in preceded the first randomized dose. The trial description says participants were in the study for about 7 months, with about 8 study-site visits and one additional visit that could be at the site or by phone. Blood samples were collected to assess changes in blood sugar, and general health was also checked.

The primary outcome population was the modified Intention-to-Treat set, not necessarily every enrolled participant. It included randomized participants who received at least one dose, had a baseline HbA1c assessment, and had at least one on-treatment HbA1c value. The arm counts reported for the primary outcome were 73 for placebo, 72 for empagliflozin 10 mg, and 73 for empagliflozin 25 mg. The registry record does not report the original sample-size justification, the method used to generate the randomization sequence, or details identifying who was masked.

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: 4 patients did not meet eligibility criteria but randomized and treated in this trial. The rest of the subjects attended a specialist site which ensured that they (the subjects) strictly met all inclusion and none of the exclusion criteria.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Modified Intention-to-Treat (mITT) set: The mITT set consisted of all randomised patients who were treated with at least one dose of the study drug and had a baseline HbA1c assessment and at least one on-treatment glycosylated haemoglobin A1c (HbA1c) value. The assignment of patients to treatment groups were based on the planned randomised study drug at the time of randomisation.

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: Treated set (TS): The TS consisted of all patients who were randomised and treated with at least one dose of the study drug. The assignment of patients to treatment groups were based on the actual first study drug intake in the double-blind treatment period.

6. Primary Endpoint

The registry lists the primary outcome as Change From Baseline in Glycosylated Haemoglobin A1c (HbA1c) at Week 24, measured over At baseline (Week 0) and at Week 24.

The primary endpoint was change from baseline in glycosylated hemoglobin A1c, or HbA1c, at Week 24. HbA1c is the percentage of glycosylated hemoglobin and is used as a measure of blood glucose over a longer period than a single glucose reading. The endpoint therefore asks whether average HbA1c changed during the treatment period and how that change compared across randomized groups.

Change from baseline focuses the comparison on movement from each participant’s starting measurement, while the between-group contrast helps distinguish change associated with study-group assignment from change also seen under placebo and background therapy. The sponsor reported least-squares mean estimates using a repeated-measures model, rather than simply comparing raw group averages. The model included treatment, background therapy, visit, baseline HbA1c, baseline estimated glomerular filtration rate, and specified interactions. The registry record describes the endpoint timeframe as baseline (Week 0) and Week 24.

In general, a continuous endpoint such as HbA1c preserves information about the size of change across the measurement scale. It is distinct from a threshold endpoint, which classifies participants according to whether they cross a specified value. The registry also lists an HbA1c threshold outcome among the secondary outcomes, but that does not replace the continuous primary endpoint. Neither measure alone captures every aspect of diabetes control or the broader health status of participants.

7. Results Posted in the Registry

• Empagliflozin 10 mg: mean -1.12 versus -0.13 on control; difference -0.99 (95% CI -1.27 to -0.71), standardized effect d = -1.16; z-test p = <0.001.
• Empagliflozin 25 mg: mean -1.12 versus -0.13 on control; difference -0.99 (95% CI -1.27 to -0.71), standardized effect d = -1.16; z-test p = <0.001.

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.

Percentage of Participants With HbA1c<7.0% at Week 24

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Empagliflozin 10 mg: 16.7% (12/72) versus 8.2% (6/73) on control; risk difference 8.4 percentage points (95% CI -2.2 to 19.1), risk ratio 2.03 (0.80–5.11); unadjusted two-proportion z-test p = 0.123.
  • Empagliflozin 25 mg: 30.1% (22/73) versus 8.2% (6/73) on control; risk difference 21.9 percentage points (95% CI 9.7 to 34.2), risk ratio 3.67 (1.58–8.51); unadjusted two-proportion z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Empagliflozin 10 mg and the control. The interval gives the range of effects that remain compatible with the data.
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically higher for Empagliflozin 25 mg than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Change in Body Weight From Baseline to Week 24

• 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:
  • Empagliflozin 10 mg: mean -1.99 versus 0.01 on control; difference -2.00 (95% CI -2.65 to -1.35), standardized effect d = -1.00; z-test p = <0.001.
  • Empagliflozin 25 mg: mean -1.31 versus 0.01 on control; difference -1.32 (95% CI -1.97 to -0.67), standardized effect d = -0.66; z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the mean is statistically lower for Empagliflozin 10 mg than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.
  • The 95% confidence interval excludes no difference, so the mean is statistically lower for Empagliflozin 25 mg than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Change From Baseline in Systolic Blood Pressure (SBP) at Week 24

• 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:
  • Empagliflozin 10 mg: mean -5.56 versus -1.95 on control; difference -3.61 (95% CI -7.59 to 0.37), standardized effect d = -0.30; z-test p = 0.075.
  • Empagliflozin 25 mg: mean -2.96 versus -1.95 on control; difference -1.01 (95% CI -4.95 to 2.93), standardized effect d = -0.08; z-test p = 0.615.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Empagliflozin 10 mg and the control. The interval gives the range of effects that remain compatible with the data.
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Empagliflozin 25 mg and the control. The interval gives the range of effects that remain compatible with the data.

The secondary outcomes extend the assessment beyond the primary continuous HbA1c endpoint. They include whether participants met an HbA1c threshold, changes in body weight and blood pressure, other glucose measures, and counts of hypoglycaemic events and adjudicated diabetic ketoacidosis events. The registry lists eight secondary outcomes with posted results. The available outcome summaries show that findings are not uniform across all types of measures: the threshold outcome and body-weight outcome have different comparative patterns from the blood-pressure outcomes. That variation is a reason to interpret each outcome on its own scale rather than treating secondary findings as one overall result.

The secondary outcomes are related to the primary result because they provide additional perspectives on glycemic status, other measured characteristics, and safety. They do not redefine the primary endpoint or independently establish the primary conclusion. The continuous HbA1c endpoint asks about average change, whereas the threshold outcome asks how many participants met a specific criterion. Body weight and blood pressure are separate measurements and should not be assumed to explain the HbA1c result. Results for additional listed outcomes include fasting plasma glucose, post-prandial glucose, hypoglycaemic events, and diabetic ketoacidosis events; their detailed numerical comparisons are not reported in the registry record.

In general, testing many secondary outcomes creates more opportunities for a nominally notable result to occur by chance. The registry record does not report a multiplicity adjustment or a hierarchy for interpreting secondary tests. Their p-values and confidence intervals should therefore be considered exploratory unless the analysis plan establishes otherwise. A lack of a reported adjustment does not show that a finding is false, but it means that isolated secondary evidence carries less weight than the primary analysis for which the main question was specified. The pattern across outcomes can inform context, while uncertainty about multiplicity and incomplete detail limits strong claims based on any single secondary result.

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

• Percentage of Participants With HbA1c<7.0% at Week 24
• Change in Body Weight From Baseline to Week 24
• Change From Baseline in Systolic Blood Pressure (SBP) at Week 24
• Change From Baseline in Diastolic Blood Pressure (DBP) at Week 24
• Change From Baseline in Fasting Plasma Glucose (FPG) at Week 24
• Change From Baseline in 2-hour Post-prandial Glucose (PPG) at Week 24
• Number of Participants With Confirmed Hypoglycaemic Events
• Number of Participants With Adjudicated Diabetic Ketoacidosis (DKA) Events

9. Statistical Interpretation

10. Statistical Methods Explained

The primary endpoint was analysed using a restricted maximum likelihood (REML) based mixed model repeated measures approach. The registry lists treatment, background therapy, and visit as fixed classification effects; baseline HbA1c and baseline estimated glomerular filtration rate as linear covariates; and treatment-by-visit and baseline-HbA1c-by-visit interactions. The reported outcome parameter was a least-squares mean, with standard error as the arm-level dispersion measure.

What does a repeated-measures model do?

In general, a repeated-measures model analyses observations collected at more than one visit while accounting for the fact that measurements from the same participant are related. It can estimate group means at visits and comparisons between groups while using the repeated outcome information in the model. The treatment-by-visit interaction allows the treatment-group contrast to vary by visit instead of requiring the same difference at every assessment.

The primary endpoint was measured at baseline and Week 24, and the listed model included visit as a classification effect. The registry record does not provide the full visit schedule used in the primary analysis or all model implementation details. It also does not report the missing-data assumptions or sensitivity analyses. A repeated-measures model can make use of available measurements, but the interpretation depends on its assumptions and on how missing observations arise. The model description alone does not establish that missing data were harmless or that all participants contributed the same amount of information.

What are least-squares means?

In general, least-squares means are model-based estimates of group means adjusted for the covariates and factors included in the model. They are not necessarily identical to unadjusted averages calculated directly from the observed values. Here the sponsor’s adjusted mean differences are based on the repeated-measures model, which incorporates the listed treatment, background therapy, visit, and baseline terms. The adjusted contrast is therefore the primary registry-reported comparison. The arm estimates and their standard errors are useful descriptive summaries, but they do not contain all the information used by the sponsor model.

Baseline HbA1c was included as a covariate and was also involved in an interaction with visit. In general, baseline adjustment can account for differences in starting measurements and can improve estimation when baseline is related to the outcome. Estimated glomerular filtration rate was also listed as a baseline covariate. The registry record does not report the exact coding of background therapy, the covariance structure for repeated measurements, or other implementation details needed to reproduce the model fully.

How do the recalculations differ?

The simple recalculations use posted arm-level summaries to estimate unadjusted comparisons. For continuous outcomes, the calculation uses group means and an approximate pooled standard deviation inferred from the reported standard errors. This is not the same as fitting the sponsor’s repeated-measures model, which uses participant-level repeated observations and covariate adjustment. The recalculated confidence intervals and p-values are therefore approximate checks on the summary data, not independent confirmation of the adjusted model.

For the primary outcome, the recalculated mean difference was -0.99 for each empagliflozin dose compared with placebo, with a recalculated 95% confidence interval of -1.267 to -0.713. These calculations use the posted arm summaries and are unadjusted. The sponsor reported adjusted estimates and intervals that are close but not identical in all displayed values. The registry record indicates no discrepancy flag; the sponsor’s model remains the source of the adjusted analysis. The recalculations should be read as a transparent, simplified analysis rather than as a substitute for the sponsor’s specified method.

What does a p-value contribute?

In general, a p-value measures how unusual data at least as inconsistent with a null hypothesis would be if that null model and its assumptions held. A small p-value is evidence against that null model, but it does not tell readers the probability that the treatment effect is real, quantify clinical importance, or establish that model assumptions are satisfied. It should be read alongside the estimate and confidence interval. For the primary comparison, the sponsor reports p < 0.001 for both dose groups versus placebo, together with adjusted mean differences and 95% confidence intervals.

What can the binary secondary endpoint show?

The HbA1c threshold outcome classifies participants according to whether they met the stated threshold at Week 24. In general, binary outcomes can be described by the proportion meeting the criterion, while a between-group risk difference describes the contrast in proportions. The registry also reports logistic regression analyses as odds ratios. Odds ratios and risk differences are different scales and should not be treated as interchangeable. The sponsor’s logistic regression estimates are adjusted analyses; simple calculations from the displayed counts are unadjusted. The registry record does not describe a multiplicity procedure for interpreting the set of secondary tests.

For continuous secondary outcomes, the sponsor reports repeated-measures estimates for body weight and blood pressure. Their analysis descriptions resemble the primary endpoint’s model approach, but the primary model details should not automatically be assumed to apply identically to every secondary analysis unless stated in the registry record. The statistical analysis details for outcomes beyond those explicitly described are not reported in the registry record.

11. Confidence Intervals and Precision

The sponsor reported a 95% confidence interval of -1.28 to -0.71 for the adjusted HbA1c comparison of empagliflozin 10 mg with placebo. For empagliflozin 25 mg compared with placebo, the 95% confidence interval was -1.26 to -0.70. These intervals give a range of values compatible with the observed data and the sponsor’s model at the stated confidence level. Both intervals exclude zero, and both are on the negative side of the comparison scale.

In general, a confidence interval communicates both the estimated direction and the uncertainty around an estimate. It is not a range containing a fixed proportion of individual participant responses, nor does a 95% interval mean there is a 95% probability that the particular fixed effect lies inside the interval. Its usual repeated-sampling interpretation depends on the analysis procedure and assumptions. The interval’s width is useful for judging precision: narrower intervals indicate less statistical uncertainty than wider intervals on the same scale.

The recalculated, unadjusted 95% confidence interval for each primary comparison was -1.267 to -0.713. It is close to, but not the same as, the sponsor’s adjusted intervals. The recalculation uses approximate standard deviations inferred from standard errors and does not reproduce the repeated-measures model. The sponsor’s adjusted intervals are therefore the appropriate intervals to use when describing the registry-reported primary analysis.

Precision is also distinct from clinical importance. The registry record does not state a threshold for a clinically important HbA1c difference, and statistical intervals cannot supply that threshold. Nor does an interval for an average contrast describe how much individual responses varied. The reported standard errors and confidence intervals concern estimation of group-level means and contrasts, not prediction for a particular participant.

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 planning figures are participants per arm for the continuous primary outcome, not total event counts. They are planning illustrations based on the observed effect, not the trial’s original design justification. The stated assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, a normal approximation, and an effect to detect set at the stated fraction of the observed effect. The trial’s own original sample-size assumptions and justification are not reported in the registry record.

The 80% power column gives the number of participants per arm estimated to yield an 80% chance of detecting the specified effect under those planning assumptions, if that effect and the calculation’s assumptions were correct. The 90% power column gives the corresponding per-arm planning number for a 90% chance. These are design calculations, not outcomes observed in the trial. This is not observed power.

The rows consider planning for 100%, 75%, and 50% of the observed effect. A smaller target effect is harder to distinguish from sampling variation, so it requires more participants under the same assumptions. The planning curve expresses the same general relationship across effect sizes: as the effect to detect becomes smaller, the estimated sample requirement rises. These illustrations are sensitive to the observed effect used as their starting point and should not be treated as a recommendation for a future trial without a separately justified target effect, outcome variability, and design.

The assumptions simplify the trial setting. Equal allocation is used for the pairwise planning comparison, although the trial itself had three randomized groups. The normal approximation and observed-effect basis may not represent all features of the sponsor’s adjusted repeated-measures analysis. The planning values are therefore useful for reading how assumptions relate to required sample size, not for inferring the trial’s achieved power or the adequacy of its original enrollment.

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 = -0.990; pooled standard deviation σ = 0.851.
2. n = 2 × (1.96 + 0.842)² × 0.851² / -0.990² = 11.6, rounded up to 12 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 <- -0.99                      # observed difference in means
sigma <- 0.851489850224279                     # 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(4233801)
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 = 12 (80% power), 16 (90% power)
 75% of observed effect: n per arm = 21 (80% power), 28 (90% power)
 50% of observed effect: n per arm = 47 (80% power), 63 (90% power)
power.t.test: 12.6 
Simulated power at n = 12 per arm: 0.777 (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 sponsor’s primary analysis used a mixed model repeated measures approach and reported adjusted mean differences. For empagliflozin 10 mg versus placebo, the adjusted estimate was -0.99, with a 95% confidence interval of -1.28 to -0.71 and p < 0.001. For empagliflozin 25 mg versus placebo, the adjusted estimate was -0.98, with a 95% confidence interval of -1.26 to -0.70 and p < 0.001.

The separate recalculation is a simple, unadjusted comparison based on posted arm summaries. It estimates -0.99 for each dose comparison and gives a 95% confidence interval of -1.267 to -0.713. The calculation uses approximate standard deviations derived from the reported standard errors, so it does not reproduce the sponsor’s covariate-adjusted repeated-measures analysis. Small differences between the sponsor estimates and recalculated values are consistent with the analyses using different information and methods.

The registry record does not flag a discrepancy between the posted results and recalculation. The sponsor’s adjusted analysis is the principal result to report because it is the registry-reported model analysis for the primary endpoint. The recalculation serves as a transparent, approximate check on the arm summaries. It should not be used to replace the sponsor’s adjusted estimate, interval, or p-value.

15. Multiplicity, Missing Data and Analysis Populations

The registry record describes the primary outcome model and lists multiple secondary outcomes, but a multiplicity strategy is not reported in the registry record. It is not reported whether the sponsor controlled the family-wise error rate, used a testing hierarchy, or applied another adjustment across the secondary comparisons. Without that information, the secondary p-values should not be interpreted as though each were the only hypothesis tested. In general, when many hypotheses are examined, the chance of at least one apparently unusual result can increase unless the analysis accounts for the collection of tests.

The record also does not report whether interim analyses were planned or conducted, or whether any formal stopping rules applied. The original sample-size justification and the protocol’s statistical analysis plan are not reported in the registry record. These omissions mean the registry summary does not establish how the study’s enrollment target was chosen or how the full set of analyses was prespecified. They should not be filled in by inference from the final sample size or posted results.

The primary population is defined as the modified Intention-to-Treat set: randomized participants who received at least one dose and had baseline and at least one on-treatment HbA1c assessment. This is narrower than a definition that includes every randomized participant regardless of treatment or outcome availability. The registry record does not report the full disposition of participants, reasons for missing outcome assessments, or whether sensitivity analyses explored alternative missing-data assumptions.

In general, repeated-measures models can use available observations without requiring a single completed value for every participant, but validity depends on assumptions about the outcome and missingness process. The listed REML-based model does not, by itself, reveal whether those assumptions were plausible or how robust estimates were to departures from them. Details on imputation, missing-data sensitivity analyses, and analysis populations for outcomes beyond the stated primary population are not reported in the registry record. The posted model specification is informative but not a complete substitute for the protocol or full statistical analysis plan.

16. 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 5% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: angina unstable (1.4%), arrhythmia (1.4%), arteriosclerosis coronary artery (1.4%), deafness neurosensory (1.4%), hyperparathyroidism (1.4%).

The registry reports participants affected by serious adverse events and deaths for the randomized groups. Serious adverse events were reported for 6 participants in the placebo group, 10 in the empagliflozin 10 mg group, and 7 in the empagliflozin 25 mg group. The at-risk count listed for each of those groups was 73. No deaths were reported in these groups. For the run-in period, the record lists 2 participants affected by serious adverse events among 219 at risk, with no deaths.

These are counts, not an adjusted comparison or a measure of causation. They do not show whether an event was caused by study treatment, how long each participant was followed for safety, or the full clinical context of each event. The registry lists several top serious adverse event terms at 1.4% each, including unstable angina, arrhythmia, coronary artery arteriosclerosis, neurosensory deafness, hyperparathyroidism, and thyroid mass. The available counts should be read as descriptive safety reporting and not as proof that an intervention caused or prevented an event.

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

The primary posted analysis estimates differences in average HbA1c change at Week 24 between each empagliflozin group and placebo in the study’s modified Intention-to-Treat population. The sponsor’s adjusted estimates are -0.99 for empagliflozin 10 mg and -0.98 for empagliflozin 25 mg versus placebo. Both reported 95% confidence intervals exclude zero, and both p-values are p < 0.001. These results describe the statistical evidence and precision for the model-based comparisons in this trial.

They do not establish that every participant experienced the group-average change, nor do they predict an individual response. They also do not establish the clinical importance of the estimated differences because a threshold for clinical importance is not reported in the registry record. Statistical significance is not a synonym for clinical benefit, and the registry summary alone does not assess all outcomes that may matter to patients or clinicians.

The findings apply to the enrolled setting: Chinese adults with type 2 diabetes and insufficient glycemic control while taking insulin, with or without up to two other oral anti-diabetic agents, over the recorded 24-week treatment period. The results do not establish what would happen in populations with different characteristics, under different background therapies, or over longer follow-up. The registry record does not report enough information to assess how representative the participants were of all people with type 2 diabetes who use insulin.

Secondary outcomes and serious adverse event counts add context but do not remove these limits. The secondary findings cover distinct measures and the registry does not report a multiplicity procedure. Safety counts are descriptive and do not establish causation. Overall, the posted data support a precise statistical description of the primary model comparisons in this trial, while broader clinical meaning, durability, and applicability remain questions not answered by the registry record alone.

18. Why This Trial Matters Statistically

This trial illustrates why a clinical result is more than a comparison of two displayed means. The primary analysis used repeated measurements and adjustment for background therapy and baseline covariates. Its adjusted estimates address the Week 24 contrast under the specified model, while a simple calculation from arm summaries provides a different, unadjusted view. Reading both carefully helps distinguish the sponsor’s planned model-based result from an approximate summary-statistic recalculation.

It also shows the value of confidence intervals alongside p-values. The adjusted HbA1c estimates are accompanied by 95% intervals, allowing readers to consider both direction and precision rather than relying only on a significance threshold. The recalculated intervals are close to the sponsor’s reported intervals, but they arise from a simpler method and do not reproduce the repeated-measures analysis. Understanding that distinction prevents a convenient recalculation from being mistaken for the primary analysis.

The trial’s secondary outcomes illustrate another general issue: a clinical study may examine several measures that answer related but non-identical questions. HbA1c change, reaching an HbA1c threshold, body weight, blood pressure, and safety events should not be merged into one statistical claim. Because multiplicity procedures are not reported in the registry record, secondary findings deserve more cautious interpretation than the primary comparison.

Finally, planning illustrations based on observed effects are not the same as a trial’s original design rationale or an estimate of achieved power. The record does not report the original sample-size justification. Separating planning assumptions, sponsor analyses, and descriptive safety counts makes the evidence easier to interpret without extending it beyond what the registry reports.

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 enrolled population is specifically described as Chinese adults with type 2 diabetes, insufficient glycemic control, and insulin use, with or without up to two additional oral anti-diabetic agents. The registry record does not report participant characteristics in enough detail to judge how broadly the results apply to people outside that setting. The primary outcome population further requires treatment exposure and available baseline and on-treatment HbA1c measurements, so it is not defined as every randomized participant.

The treatment period was 24 weeks. The record does not establish longer-term outcomes beyond that period. Although the study is listed as completed, detailed participant disposition, adherence, reasons for discontinuation, and missing outcome patterns are not reported in the registry record. Those details would help readers assess how much the analysis population may differ from the randomized population and how missingness could affect the model-based estimate.

The sponsor’s primary analysis is described as a REML-based repeated-measures model, but the registry does not report every implementation detail, including the repeated-measures covariance structure and missing-data sensitivity analyses. The simple recalculation is also limited: it uses arm-level summaries, approximate standard deviations inferred from standard errors, and no covariate adjustment. It cannot validate all assumptions of the sponsor’s model.

The record does not report the original sample-size justification, multiplicity strategy, interim analyses, or full statistical analysis plan. The secondary outcomes are numerous, and the absence of a reported multiplicity method limits interpretation of isolated secondary findings. Finally, reported serious adverse event counts are descriptive; the registry summaries do not establish attribution, event timing, or a causal comparison.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: A Study to Test How Well Empagliflozin Works in Chinese Patients With Type 2 Diabetes Who Already Take Insulin — NCT04233801.
• Registry record last updated 2023-12-15; status last verified 2023-03.
• Results first posted 2023-12-15.
• 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

NCT04233801 is a completed Phase III randomized, placebo-controlled, parallel-group study of empagliflozin in Chinese adults with type 2 diabetes who were taking insulin and had insufficient glycemic control. Enrollment was 219. Participants received empagliflozin 10 mg, empagliflozin 25 mg, or matching placebo once daily for 24 weeks, following a 2-week open-label placebo run-in.

The primary endpoint was change in HbA1c from baseline to Week 24. The sponsor used a REML-based mixed model repeated measures analysis and reported adjusted mean differences versus placebo of -0.99 for the 10 mg group and -0.98 for the 25 mg group; both reported p-values were p < 0.001. The registry also lists secondary outcomes and serious adverse event counts. The original sample-size justification, multiplicity strategy, and detailed missing-data sensitivity analyses are not reported in the registry record.