A Study of Tirzepatide (LY3298176) Versus Insulin Lispro (U100) in Participants With Type 2 Diabetes Inadequately Controlled on Insulin Glargine (U100) With or Without Metformin: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study of Tirzepatide (LY3298176) Versus Insulin Lispro (U100) in Participants With Type 2 Diabetes Inadequately Controlled on Insulin Glargine (U100) With or Without Metformin (NCT04537923), a randomized phase 3 study. The primary outcome is Change From Baseline in Hemoglobin A1c (HbA1c) (Pooled Doses of Tirzepatide 5 mg, 10 mg and 15 mg).

2. Clinical Question

3. Trial Design

This was a randomized, Phase 3, open-label, parallel-group trial of tirzepatide and insulin lispro in participants with type 2 diabetes inadequately controlled on insulin glargine, with or without metformin. The registry lists 1428 participants enrolled and the study status as completed. The study began on 2020-10-19, with primary completion on 2022-10-11. The sponsor was Eli Lilly and Company.

Participants were assigned to one of four treatment groups: tirzepatide 5 mg, 10 mg, or 15 mg given subcutaneously once a week, or insulin lispro U100 given subcutaneously three times a day. The primary outcome combined the three tirzepatide dose groups, while some secondary outcomes report dose groups separately. All groups were receiving insulin glargine, and metformin use could vary.

Random allocation supports comparisons between assigned groups by reducing the influence of measured and unmeasured baseline differences, though it does not eliminate the effects of missing measurements or analysis choices. The open-label design means treatment assignment was known to participants and study personnel. That feature may matter for outcomes influenced by expectations or behavior; HbA1c is a laboratory measure, but the registry record does not describe whether laboratory assessment or other procedures were masked.

The primary endpoint was change from baseline in HbA1c at Week 52. The registry also lists secondary outcomes covering HbA1c targets, body weight, fasting serum glucose, self-monitored blood glucose, hypoglycemia-related target achievement, weight loss, and a health survey score. The registry record does not report the original sample-size justification, the intended enrollment rationale, or interim-analysis plans. The available record describes a mixed-model repeated-measures approach for the primary endpoint, but a complete statistical analysis plan is 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:

Analysis population for the primary outcome, as defined in the registry: All randomly assigned participants who took at least 1 dose of study drug and had a baseline and at least 1 post-baseline value for this analysis, excluding participants discontinuing study drug due to inadvertent enrollment and data after initiating rescue antihyperglycemic medication or prematurely stopping study drug. This analysis was planned to measure the outcome for pooled 5 mg, 10 mg and 15 mg tirzepatide.

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: All randomized participants.

6. Primary Endpoint

The registry lists the primary outcome as Change From Baseline in Hemoglobin A1c (HbA1c) (Pooled Doses of Tirzepatide 5 mg, 10 mg and 15 mg), measured over Baseline, Week 52.

The primary endpoint is change from baseline in hemoglobin A1c (HbA1c) at Week 52. HbA1c is the glycosylated fraction of hemoglobin A. It is measured primarily to identify average plasma glucose concentration over a prolonged period, so a change from baseline offers a way to compare longer-term glycemic measurements across treatment groups rather than relying on a single reading.

The reported measure is a least-squares mean, not a simple unadjusted average. The registry says the least-squares means were estimated using a mixed-model repeated-measures model with baseline HbA1c, pooled country, baseline metformin use, treatment, time, and treatment-by-time terms. Such adjustment can account for specified covariates and repeated outcome measurements under the model’s assumptions. The record identifies standard error as the dispersion measure for the group estimates.

The primary comparison pooled the 5 mg, 10 mg, and 15 mg tirzepatide doses and compared that group with insulin lispro. The posted estimates are -2.26 for pooled tirzepatide and -1.16 for insulin lispro. A negative change indicates a lower HbA1c measurement at follow-up relative to baseline in the reported scale. The reported comparison is -1.10 for pooled tirzepatide relative to insulin lispro, with a 95% confidence interval from -1.24 to -0.97.

As a general statistical point, a continuous endpoint retains information about the size of change rather than reducing each participant’s result to whether a target was met. That can make it informative for comparing average changes, while not describing the full distribution of individual responses.

7. Results Posted in the Registry

• Insulin Lispro: mean -1.16 versus -2.26 on control; difference 1.10 (95% CI 0.97 to 1.23), standardized effect d = 0.86; 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.

Change From Baseline in HbA1c

• 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:
  • 10 mg Tirzepatide: mean -2.27 versus -2.05 on control; difference -0.22 (95% CI -0.45 to 0.01), standardized effect d = -0.17; z-test p = 0.059.
  • 15 mg Tirzepatide: mean -2.46 versus -2.05 on control; difference -0.41 (95% CI -0.64 to -0.18), standardized effect d = -0.32; z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between 10 mg Tirzepatide 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 mean is statistically lower for 15 mg Tirzepatide than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Percentage of Participants With HbA1c Target Values <7.0%

• 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:
  • 10 mg Tirzepatide: 75.6% (180/238) versus 61.2% (148/242) on control; risk difference 14.5 percentage points (95% CI 6.3 to 22.7), risk ratio 1.24 (1.09–1.40); unadjusted two-proportion z-test p = <0.001.
  • 15 mg Tirzepatide: 79.7% (188/236) versus 61.2% (148/242) on control; risk difference 18.5 percentage points (95% CI 10.5 to 26.5), risk ratio 1.30 (1.16–1.47); unadjusted two-proportion z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically higher for 10 mg Tirzepatide 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 proportion with the event is statistically higher for 15 mg Tirzepatide than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Change From Baseline in Body Weight

• 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:
  • 10 mg Tirzepatide: mean -9.90 versus -6.90 on control; difference -3.00 (95% CI -4.03 to -1.97), standardized effect d = -0.53; z-test p = <0.001.
  • 15 mg Tirzepatide: mean -12.00 versus -6.90 on control; difference -5.10 (95% CI -6.14 to -4.06), standardized effect d = -0.88; z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the mean is statistically lower for 10 mg Tirzepatide 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 15 mg Tirzepatide than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

The registry lists secondary outcomes covering HbA1c change and target attainment, body weight, fasting serum glucose, self-monitored blood glucose, achievement of HbA1c and weight targets, hypoglycemia-related target attainment, and a physical component score from a health survey. The posted secondary results include outcomes with different scales and meanings, so they add context rather than serving as interchangeable measures of the primary endpoint.

Across the outcomes with results described in the record, the pattern is broadly aligned with the primary comparison: the tirzepatide groups have more negative mean changes in HbA1c than insulin lispro, and the posted target-attainment percentages are higher in the reported tirzepatide groups. Body-weight change estimates also differ by assigned group. These observations describe the posted estimates; they do not establish that every secondary outcome was tested with the same model or that all findings have the same evidential status.

The secondary results connect to the primary result by showing whether the comparison is also reflected in related glycemic and weight measures. HbA1c target attainment is a threshold-based outcome, while HbA1c change retains the continuous measurement. Weight change addresses a different clinical dimension. Concordance can be informative, but it does not remove uncertainty about missing data, analysis choices, or multiplicity.

The registry lists multiple outcomes and 28 analyses. When many outcomes and contrasts are examined, isolated small p-values can arise by chance unless multiplicity is addressed. The full adjustment strategy and prespecified hierarchy are not reported in the registry record. Accordingly, unadjusted secondary comparisons carry less weight than the primary analysis, especially where they were not clearly designated as confirmatory. They should be interpreted as supporting context and not as a collection of independent confirmations.

The record reports nine secondary outcomes, but it does not give a complete account of all intended analyses or their prespecified status. The specific displayed comparisons should not be generalized to unreported outcomes or treated as proof of a dose trend without a suitable formal analysis.

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

• Change From Baseline in HbA1c
• Percentage of Participants With HbA1c Target Values <7.0%
• Change From Baseline in Body Weight
• Change From Baseline in Fasting Serum Glucose
• Change From Baseline in Daily Average 7-Point Self-Monitored Blood Glucose (SMBG) Values
• Percentage of Participants Who Achieved HbA1c Target Value of <7.0% Without Hypoglycemia
• Percentage of Participants Who Achieved Weight Loss ≥5%
• Change From Baseline in 36-Item Short Form Health Survey (SF-36) Physical Component Summary (PCS) Score

9. Statistical Interpretation

10. Statistical Methods Explained

The primary analysis used a mixed-model repeated-measures approach to estimate least-squares mean HbA1c changes. A mixed model is a statistical framework for repeated observations that can account for the association between measurements from the same participant. It can use available repeated outcome measurements without requiring every participant to have a complete sequence, subject to the model and missing-data assumptions. The registry describes the model covariates, but the full model specification and all diagnostic details are not reported in the registry record.

What does a least-squares mean represent?

A least-squares mean is a model-based group estimate adjusted for the covariates included in the analysis. Here, the described covariates were baseline HbA1c, pooled country, baseline metformin use, treatment, time, and a treatment-by-time term. Adjustment aims to compare groups at a common modeled setting for those included factors. It does not automatically correct for factors omitted from the model, measurement problems, or departures from model assumptions.

The primary outcome is continuous, and its group estimates are expressed as mean changes in HbA1c. The record gives standard errors for the group estimates. A standard error describes the estimated sampling variability of an estimate; it is not the spread of individual participant measurements. General statistical interpretation should distinguish uncertainty about the group mean from variation among individuals.

How should the treatment contrast be read?

The sponsor reported a mixed-model least-squares mean difference of -1.10, with a 95% confidence interval from -1.24 to -0.97 and p < 0.001. The comparison is oriented as pooled tirzepatide relative to insulin lispro. The recalculated comparison instead lists insulin lispro relative to pooled tirzepatide, giving 1.1 with an interval from 0.966 to 1.234 and p < 0.001. Opposite signs reflect reversed subtraction, not conflicting direction. The sponsor’s model-based contrast is the primary reported analysis; the recalculation is a simple unadjusted analysis from the posted summary statistics.

The sponsor’s method label for the primary result is mixed models analysis, with the least-squares mean difference as the estimand. The primary endpoint description says the model used Type III sums of squares. The registry labels the primary analysis as non-inferiority. However, a non-inferiority margin and its prespecification are not reported in the registry record, so the result cannot be fully evaluated as a non-inferiority analysis from this record alone. The reported test and interval should be understood as posted, without inferring missing protocol details.

Why can a recalculation differ in method?

A recalculation based on group estimates and their standard errors does not reproduce all features of a repeated-measures model. It uses a simpler comparison and depends on summary statistics, including an approximate pooled standard deviation for standardized comparisons. This can be useful as a transparent cross-check, but it may not use the same covariate adjustment, repeated-measures structure, missing-data handling, or population definition as the sponsor analysis. The registry record reports no discrepancy flag, but methodological differences remain important when interpreting the two estimates.

General statistical methods also distinguish unadjusted p-values from multiplicity-adjusted inference. When many outcomes or comparisons are examined, the chance of at least one apparently notable result can increase unless the analysis plan addresses multiplicity. The record lists multiple outcomes and analyses, but a complete multiplicity strategy is not reported in the registry record. Therefore, secondary p-values should be read as exploratory or supportive unless their role and adjustment are established in the full analysis plan.

Missing outcome data can affect model estimates. Mixed models may be valid under assumptions about the relationship between missingness and observed data, but the registry description alone does not establish whether those assumptions hold. The record says data after initiating rescue antihyperglycemic medication and premature discontinuation were excluded for the primary analysis, and it describes an analysis population requiring at least one post-baseline value. The full handling of missing values and sensitivity analyses is not reported in the registry record.

Randomization, adjustment, and modeling answer related but distinct questions. Randomization establishes the assigned comparison; the analysis model estimates an adjusted contrast using selected data and assumptions. Careful reading therefore considers the endpoint, treatment contrast direction, analysis population, model, interval, p-value, and missingness. No single statistic resolves all these considerations.

11. Confidence Intervals and Precision

The primary sponsor-reported difference is -1.10, with a 95% confidence interval from -1.24 to -0.97. This interval expresses uncertainty around the model-based mean contrast. Under the model and its assumptions, the interval is compatible with treatment differences within those limits at the stated confidence level. It is not a range containing 95% of individual participants’ changes, nor does it capture every source of uncertainty, including bias from missing data or unmeasured factors.

The interval does not include zero, and the reported p-value is < 0.001. This is evidence against a zero difference under the sponsor’s analysis framework, but the p-value does not state the probability that the null hypothesis is true and does not measure practical importance. The magnitude and context of the endpoint remain necessary for interpretation.

The recalculated contrast is expressed in the opposite direction: 1.1, with an interval from 0.966 to 1.234. It is based on a simple unadjusted recalculation from posted summaries. Its interval is not identical to the sponsor’s interval because it does not reproduce the sponsor’s full mixed-model analysis. The sponsor-reported estimate and interval should be used to describe the primary analysis.

For the group estimates, the registry reports standard errors of 0.048 for pooled tirzepatide and 0.049 for insulin lispro. Standard errors indicate estimated precision of the group means, whereas the confidence interval for a contrast describes uncertainty about the difference. Precision does not establish absence of bias, and a narrow interval cannot compensate for incomplete reporting of the analysis plan or population.

As a general principle, intervals are more informative than a significance label alone because they display both direction and plausible magnitude under the model. They remain conditional on how participants, outcomes, covariates, and missing measurements were handled. The registry record does not report enough information to assess every modeling assumption or sensitivity analysis.

12. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The sample-size figures are planning illustrations based on observed effects, not the trial’s original design calculations. For continuous outcomes, the entries are participants per arm; time-to-event figures, if present, would instead be total events. The listed assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. Each planning scenario uses the stated fraction of the observed effect as the effect to detect.

The 80% power column describes the number of participants per arm that would give an 80% chance of detecting the specified effect under those assumptions if that effect were real. The 90% power column gives the corresponding planning count for a 90% chance. These are probability targets used when planning a hypothetical study, not guarantees about any actual study or result.

Rows that use a smaller fraction of the observed effect represent more modest differences to detect. Smaller effects are harder to distinguish from random variation, so the planning figures generally require more participants per arm. The resulting values depend on the assumed effect size, variability, test, allocation, and approximation. They should not be read as a reconstruction of the trial’s protocol or its actual sample-size rationale; that rationale is not reported in the registry record.

This is not observed power. Observed power is not a useful substitute for interpreting the estimate and its confidence interval after results are known. These figures are only planning illustrations calculated from the observed effect and stated assumptions; the actual trial enrollment and analysis were not determined by these illustrative counts.

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 = 1.100; pooled standard deviation σ = 1.283.
2. n = 2 × (1.96 + 0.842)² × 1.283² / 1.100² = 21.4, rounded up to 22 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 <- 1.1                      # observed difference in means
sigma <- 1.28303890757081                     # 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(4537923)
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 = 22 (80% power), 29 (90% power)
 75% of observed effect: n per arm = 38 (80% power), 51 (90% power)
 50% of observed effect: n per arm = 86 (80% power), 115 (90% power)
power.t.test: 22.4 
Simulated power at n = 22 per arm: 0.788 (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 result is a mixed-model least-squares mean difference of -1.10 for pooled tirzepatide relative to insulin lispro, with a 95% confidence interval from -1.24 to -0.97 and p < 0.001. It is the reported analysis for the primary endpoint and reflects the registry’s described model, including baseline and treatment-related covariates and repeated measurements.

The simple recalculation is oriented in the reverse direction, insulin lispro relative to pooled tirzepatide. It gives 1.1, with a 95% interval from 0.966 to 1.234 and p < 0.001. The opposite signs arise from the order of subtraction. The recalculation uses posted summary statistics and does not reproduce the sponsor’s full model, so small differences in estimates and intervals are expected from the distinct methods.

The registry record does not flag a discrepancy between the recalculation and sponsor analysis. Even so, the primary interpretation should emphasize the sponsor’s model-based result rather than treating a simplified reanalysis as equivalent. The record labels the primary analysis non-inferiority, but the non-inferiority margin is not reported in the registry record. That missing detail limits independent assessment of the non-inferiority framework.

15. Multiplicity, Missing Data and Analysis Populations

The registry reports 10 outcome measures and 28 analyses, including a primary endpoint and multiple secondary outcomes and comparisons. Examining many statistical tests creates a multiplicity concern: even if no true differences existed, the chance of obtaining one or more apparently notable results would rise as the number of tests increases. This is a general statistical issue, not a claim that any particular result is false.

The record reports p-values for the primary analysis and several secondary comparisons. It does not report a complete multiplicity strategy, such as an adjustment method, a testing hierarchy, or which secondary comparisons were confirmatory. Without that information, secondary p-values should not be interpreted as if they were all protected by a single prespecified confirmatory plan. The primary result has a distinct status as the named primary endpoint, but full evaluation of its testing framework also requires the protocol and analysis plan, which are not reported in the registry record.

The primary analysis population required random assignment, at least one dose of study drug, baseline measurement, and at least one post-baseline measurement. The description excludes data after rescue antihyperglycemic medication and refers to premature discontinuation and inadvertent enrollment; the description is incomplete in the registry record. It does not provide a full flow of participants through each analysis or explain all exclusions for all outcomes.

Missing-data handling matters because participants without follow-up measurements may differ from those with observed outcomes. Mixed-model repeated-measures methods can use available measurements under assumptions about missingness and the relationship between missingness and observed data. The registry identifies the model but does not report the assumptions in detail, the amount or pattern of missingness, or sensitivity analyses that test alternative assumptions. Those details are not reported in the registry record.

The posted analysis should therefore be read as a model-based result in a specified analysis population, not as a complete account of every participant’s outcome. The registry also does not report interim-analysis plans or a full statistical analysis plan, so it is not possible to determine from this record alone how all analysis decisions were prespecified.

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: hypoglycaemia (4.0%), covid-19 (1.0%), covid-19 pneumonia (1.0%), pneumonia (0.8%), ischaemic stroke (0.8%).

The registry reports serious adverse events and other adverse events by treatment group. For serious adverse events, the affected counts were 20 of 243 in the 5 mg tirzepatide group, 16 of 238 in the 10 mg group, 17 of 236 in the 15 mg group, and 81 of 708 in the insulin lispro group. The reported at-risk counts are specific to these safety summaries and should not be assumed to match the primary endpoint analysis population.

The registry also reports other adverse-event counts of 94 of 243, 116 of 238, 121 of 236, and 99 of 708, respectively. Death counts were 3, 3, 1, and 11 in those groups. These are descriptive counts, not adjusted comparisons, and the record does not establish that differences between groups are statistically reliable or attributable to treatment.

The listed top serious-event terms include Hypoglycaemia at 4.0%, Covid-19 at 1.0%, Covid-19 pneumonia at 1.0%, Pneumonia at 0.8%, Ischaemic stroke at 0.8%, and Acute myocardial infarction at 0.8%. These terms and percentages should be interpreted within the registry’s safety reporting context; definitions, exposure-adjusted rates, and a full comparative safety analysis are not reported in the registry record.

Serious adverse event counts do not summarize all tolerability or risk. They do not on their own establish causality, severity patterns beyond the registry classification, or differences in duration of exposure. The figures are descriptive and should be read as such.

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

The primary result describes the difference in model-estimated HbA1c change between the pooled tirzepatide groups and insulin lispro at Week 52 among the analysis population specified in the registry. The sponsor-reported contrast is -1.10, with a 95% confidence interval from -1.24 to -0.97 and p < 0.001. This is a precise model-based estimate under the stated analysis, and its direction and uncertainty are conveyed by the estimate and interval.

It does not mean that every participant experienced the group-average change. A least-squares mean describes a modeled group average, not the range of individual outcomes. Nor does a p-value state the probability that a treatment effect is true, or determine whether a difference is important to an individual. The endpoint measures HbA1c, not every dimension of diabetes management, health status, or safety.

The comparison was randomized and open-label, and all groups were on insulin glargine with or without metformin. Interpretation is therefore specific to the study’s population, background treatment, assigned regimens, follow-up, and analysis. The trial does not directly answer questions about other populations or treatment combinations not described in the registry record.

The sponsor’s primary analysis is model-adjusted, while the simple recalculation uses posted summary statistics and a different approach. The recalculation is a cross-check, not a replacement for the sponsor’s analysis. Missing-data assumptions, multiplicity handling, and the non-inferiority margin are not fully reported in the registry record, which limits independent assessment of the full statistical framework.

Safety counts add important descriptive information but do not establish comparative causality. The record’s serious and other adverse-event totals cannot, by themselves, describe all potential risks or how exposure time and event definitions affected reporting. Overall, the posted results provide evidence about a specific Week 52 HbA1c comparison, with limits that should remain visible alongside the estimate.

18. Why This Trial Matters Statistically

This trial presents several statistical issues that are useful to distinguish when reading clinical results. First, the primary endpoint is continuous HbA1c change, and the primary estimate is a covariate-adjusted least-squares mean contrast from a repeated-measures model. That differs from a raw difference between simple group averages, even when the direction of the estimates is consistent.

Second, the main analysis pools three tirzepatide doses, while secondary results include individual dose groups. Pooling answers a question about the combined dose groups as defined for the primary endpoint; separate-dose estimates address different comparisons and should not be substituted for the primary contrast. A pattern across dose groups is descriptive unless a formal dose-response analysis is reported.

Third, the reported primary analysis is labeled non-inferiority, but the registry record does not report the non-inferiority margin. A non-inferiority conclusion depends on a prespecified margin and a suitable analysis framework, so the label alone is insufficient for independent evaluation. The reported estimate, confidence interval, and p-value remain useful descriptions of the posted model-based comparison.

Fourth, the open-label design and analysis-population exclusions matter for interpretation even with randomization. Treatment awareness can affect behavior and some outcome processes, and exclusions or missing measurements can change the analyzed population. The registry describes aspects of the primary model and population but not a complete missing-data strategy, sensitivity analysis, or multiplicity plan.

Finally, the planning illustrations distinguish a plausible observed effect from a smaller effect that a future study may wish to detect. They are not evidence of the trial’s actual power or design rationale. Reading the estimate together with its uncertainty, analysis assumptions, and reporting gaps is more informative than relying on a single significance label.

19. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The registry record does not report the original sample-size justification, so the planning illustrations based on observed effects cannot be used to infer the trial’s design target or planned power. It also does not report a non-inferiority margin, although the primary analysis is labeled non-inferiority. This leaves a central element of that framework unavailable for independent assessment.

The primary analysis population excludes some observations after rescue antihyperglycemic medication and premature discontinuation, and the registry description is truncated. The full participant flow, reasons for missing measurements, and handling of missing data are not reported in the registry record. The mixed-model method depends on assumptions, and the record does not provide sufficient diagnostics or sensitivity analyses to evaluate them.

The study was open-label. The registry does not state whether outcome assessors or laboratory personnel were masked, so that detail is not reported in the registry record. HbA1c is a laboratory measure, but the trial’s open-label nature may still be relevant to other outcomes and participant behavior.

There are numerous outcome measures and analyses, while the full multiplicity strategy and prespecified status of secondary comparisons are not reported. Secondary findings should therefore be treated as less conclusive than the primary result unless additional documentation clarifies their role. The record also does not provide a complete description of all intended analyses.

Safety information is presented as counts and selected serious-event terms. Exposure-adjusted comparisons and a complete account of event definitions and follow-up are not reported in the registry record. These limits mean that descriptive safety counts should not be treated as a full comparative safety assessment.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: A Study of Tirzepatide (LY3298176) Versus Insulin Lispro (U100) in Participants With Type 2 Diabetes Inadequately Controlled on Insulin Glargine (U100) With or Without Metformin — NCT04537923.
• Registry record last updated 2023-11-02; status last verified 2023-10.
• Results first posted 2023-11-02.
• 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

The registry describes a completed randomized, Phase 3, open-label, parallel-group trial enrolling 1428 participants with type 2 diabetes inadequately controlled on insulin glargine, with or without metformin. Participants received once-weekly tirzepatide at one of three doses or insulin lispro U100 three times daily, in addition to background insulin glargine.

The primary endpoint was change from baseline in HbA1c at Week 52. The pooled tirzepatide least-squares mean change was -2.26 and the insulin lispro estimate was -1.16. The sponsor-reported mixed-model difference was -1.10, with a 95% confidence interval from -1.24 to -0.97 and p < 0.001.

The record includes secondary outcomes and descriptive safety counts. It does not report the non-inferiority margin, original sample-size rationale, complete multiplicity strategy, or full missing-data sensitivity analyses. These omissions limit independent evaluation of parts of the statistical plan and reinforce the need to interpret the estimates within their reported analysis framework.