A Study to Compare a New Drug for Type 2 Diabetes to Placebo and to a Treatment Already Available for Type 2 Diabetes: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study to Compare a New Drug for Type 2 Diabetes to Placebo and to a Treatment Already Available for Type 2 Diabetes (NCT02119819), a randomized phase 2 study. The primary outcome is Change From Baseline in Hemoglobin A1c (HbA1c) at Week 12.

2. Clinical Question

The registry lists an enrollment of 420 participants and describes an expected study involvement of about 30 weeks. The posted Week 12 outcome is a measure of average change in HbA1c, not a direct count of clinical events. The randomized, double-masked design is intended to support comparisons between assigned groups while limiting the influence of participant and study-team knowledge of treatment assignment.

3. Trial Design

This was a completed phase 2 study with randomized allocation, double masking, and a parallel-group model. The registry lists 420 participants as the enrollment. The study began in 2014-04, and primary completion was in 2015-08. The registry does not report the original sample-size justification or the allocation ratio.

Participants were assigned to one of four LY2944876 dose groups, an exenatide extended-release group, or a placebo group. LY2944876 was administered subcutaneously once weekly for 24 weeks at doses of 10 mg, 15 mg, 30 mg, or 50 mg. Exenatide extended-release was administered subcutaneously at 2 mg once weekly for 24 weeks. The placebo group received placebo injections once weekly for 12 weeks and no injections during the second 12 weeks. Participants taking metformin at study entry remained on stable doses, as prescribed by their personal investigator.

The main purpose in the registry was to compare the safety and effectiveness of LY2944876 with exenatide extended-release and placebo in people with type 2 diabetes. The primary outcome was change from baseline in HbA1c at Week 12. The registry states that participants needed baseline and at least one post-baseline HbA1c value at Week 12 for the primary analysis population. Missing observations for that outcome were imputed using a Bayesian simple linear regression longitudinal model.

The record lists eight posted analyses and 15 outcome measures, and identifies 14 posted secondary outcomes. It does not report a full statistical analysis plan, the randomization procedure, the masking procedure, or the trial's own original assumptions for choosing its enrollment. These details should not be inferred from the group sizes in the posted outcome results.

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 randomized participants with baseline and at least one post-baseline HbA1c data at Week 12. Missing observations are imputed using the Bayesian simple linear regression longitudinal model.

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.

Only the first 6 of 7 groups are shown.

6. Primary Endpoint

The registry lists the primary outcome as Change From Baseline in Hemoglobin A1c (HbA1c) at Week 12, measured over Baseline, Week 12.

HbA1c is the glycosylated fraction of hemoglobin A. The registry describes it as a measure used to identify average plasma glucose concentration over prolonged periods of time. The primary endpoint measured change from baseline in HbA1c at Week 12, expressed as a percentage of HbA1c. A change score compares a participant's follow-up measurement with their baseline measurement, then the group result summarizes those changes.

This endpoint is useful in a diabetes trial because it reflects average glucose exposure over a prolonged period rather than a single measurement at one moment. That general property helps explain why HbA1c is used to assess changes over the course of a trial. It does not mean that this endpoint captures every aspect of diabetes or every outcome that matters to participants.

The registry reports mean changes with standard errors for each treatment group. Negative values indicate a decrease from baseline in the measure, according to the signed change values posted. The primary analysis population included randomized participants with a baseline value and at least one post-baseline HbA1c measurement at Week 12. Missing observations were imputed using the registry-described Bayesian simple linear regression longitudinal model. This population definition and imputation approach matter when interpreting the results: the posted group means are not simply the average of every randomized participant's observed Week 12 value.

The primary endpoint is a continuous measure. Its group mean describes the average change, while the standard error describes the estimated precision of that mean. Neither alone describes the range of individual participant responses.

7. Results Posted in the Registry

• 10 mg LY2944876: mean -1.07 versus -0.29 on control; difference -0.78 (95% CI -1.10 to -0.46), standardized effect d = -0.88; z-test p = <0.001.
• 15 mg LY2944876: mean -1.09 versus -0.29 on control; difference -0.80 (95% CI -1.10 to -0.50), standardized effect d = -0.90; z-test p = <0.001.
• 30 mg LY2944876: mean -1.44 versus -0.29 on control; difference -1.15 (95% CI -1.45 to -0.85), standardized effect d = -1.30; z-test p = <0.001.
• 50 mg LY2944876: mean -1.33 versus -0.29 on control; difference -1.04 (95% CI -1.34 to -0.74), standardized effect d = -1.19; 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 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:
  • 10 mg LY2944876: mean -0.85 versus -0.38 on control; difference -0.47 (95% CI -0.86 to -0.08), standardized effect d = -0.46; z-test p = 0.018.
  • 15 mg LY2944876: mean -1.14 versus -0.38 on control; difference -0.76 (95% CI -1.13 to -0.39), standardized effect d = -0.74; z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the mean is statistically lower for 10 mg LY2944876 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 LY2944876 than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Percentage of Participants Requiring Rescue Therapy

• 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 LY2944876: 6.1% (4/66) versus 11.3% (8/71) on control; risk difference -5.2 percentage points (95% CI -14.5 to 4.1), risk ratio 0.54 (0.17–1.70); unadjusted two-proportion z-test p = 0.281.
  • 15 mg LY2944876: 2.8% (2/71) versus 11.3% (8/71) on control; risk difference -8.5 percentage points (95% CI -16.8 to -0.1), risk ratio 0.25 (0.05–1.14); unadjusted two-proportion z-test p = 0.049.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between 10 mg LY2944876 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 lower for 15 mg LY2944876 than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Pharmacokinetics (PK): Maximum Concentration (Cmax) of LY2944876

• 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 deviations.
• Result:
  • 15 mg LY2944876: mean 799.00 versus 607.00 on control; difference 192.00 (95% CI 82.66 to 301.34), standardized effect d = 0.58; z-test p = <0.001.
  • 30 mg LY2944876: mean 1,690 versus 607.00 on control; difference 1,083 (95% CI 901.82 to 1,264), standardized effect d = 1.92; z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the mean is statistically higher for 15 mg LY2944876 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 higher for 30 mg LY2944876 than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

The registry lists a broad set of secondary outcomes, including HbA1c at Week 24, body weight, fasting blood glucose, self-monitored blood glucose, lipids, a biomarker, rescue therapy, and anti-drug antibodies. Fourteen secondary outcomes are identified as posted, while only a subset has detailed numerical results in this record. The displayed secondary results therefore do not represent the full set of outcomes named in the registry.

Across the available secondary results, the overall pattern is that some measures provide additional context about glycemic change, while others address rescue therapy or drug concentration. The Week 24 HbA1c summaries offer a later measurement than the Week 12 primary endpoint. The rescue therapy outcome records the percentage requiring rescue, and the pharmacokinetic outcome summarizes maximum concentration across dose groups. These outcomes answer different questions and should not be read as interchangeable measures of the primary endpoint.

Secondary outcomes may support interpretation of the trial's broader pattern, but they do not independently establish the primary result. The sponsor's primary Bayesian analysis remains the central evidence for the specified Week 12 endpoint. Results from other time points and outcome types may be informative, but their interpretation depends on their own analysis methods, missing-data handling, and precision.

Unadjusted secondary analyses carry less weight when many outcomes are examined because the chance of observing apparently notable results by chance can increase with the number of comparisons. The registry record does not report a multiplicity adjustment for the secondary outcomes. Their estimates should therefore be regarded as exploratory context rather than a collection of independent confirmatory findings. The full secondary statistical plan and detailed results for every listed outcome are not reported in the registry record.

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

• Change From Baseline in HbA1c at Week 24
• Percent Change From Baseline in Body Weight
• Change From Baseline in Fasting Blood Glucose
• Change From Baseline in 7-point Self-Monitored Blood Glucose (SMBG) Values
• Change From Baseline in Lipids
• Change From Baseline in Fasting Fibroblast Growth Factor 21
• Percentage of Participants Requiring Rescue Therapy
• Percentage of Participants Developing Anti-Drug Antibodies to LY2944876

9. Statistical Interpretation

10. Statistical Methods Explained

What does a group mean describe?

A mean is an average of the measurements included in an analysis. For this primary outcome, the posted values are average changes from baseline in HbA1c at Week 12. A mean compresses the participant-level data into one summary. It does not show the full distribution, how many participants improved or worsened, or whether some individuals had very different responses from the group average.

The signed change values retain direction: a negative change indicates a reduction in the measured value from baseline. Comparing group means can describe the difference between average changes, but the resulting estimate depends on which participants are included and how missing observations are handled. The registry defines the primary population and states that missing HbA1c observations were imputed with a Bayesian simple linear regression longitudinal model.

What does a standard error describe?

A standard error describes the uncertainty in an estimated mean, rather than the spread of individual observations. In general, a smaller standard error means that an estimate is more precise under the analysis assumptions. It is not interchangeable with a standard deviation: a standard deviation describes variation among observations, while a standard error describes uncertainty in an estimated summary. The primary outcome reports standard errors for the group means.

How should the sponsor's Bayesian comparisons be read?

The registry reports Bayesian posterior mean differences and 90% credible intervals for comparisons with placebo and exenatide extended-release. In general, a Bayesian analysis combines the observed data with model assumptions, including prior information, to produce a posterior distribution for the parameter of interest. A posterior mean difference is a summary of that distribution. A credible interval describes a range of values containing a stated share of posterior probability under the model and its assumptions.

The precise prior distributions, model specification, and interpretation of the reported p values are not described in the registry record. It is therefore important not to infer those details from the numbers alone. The reported Bayesian results are the sponsor's own analysis and are the primary results to use when the simple recalculations disagree with them.

What are the recalculated comparisons?

The recalculated mean differences are unadjusted comparisons derived from the posted group means and standard errors. They are simple summary-statistic recalculations, not a reproduction of the sponsor's Bayesian model. An unadjusted comparison does not account for a multiplicity procedure, and it does not incorporate analysis features that are not available in the posted summaries. The recalculated intervals should therefore not be substituted for the sponsor's reported analysis.

A confidence interval, in general, is an interval produced by a method that would cover the true parameter at a specified rate over repeated use of that method, if the method's assumptions hold. It communicates a range of values compatible with the data and method, not a guarantee that the true value lies in that specific interval. The recalculated intervals and the sponsor's credible intervals arise from different approaches and should not be treated as interchangeable.

Why do missing-data rules matter?

Missing outcome measurements can affect a group summary, particularly if the chance of having a measurement is related to participant characteristics or outcomes. Imputation fills in missing values using a statistical model; it does not turn unknown values into observed data. The registry specifies a Bayesian simple linear regression longitudinal model for missing primary HbA1c observations, but does not report the model details or the assumptions about why measurements were missing. The registry record also does not report sensitivity analyses for alternative missing-data assumptions.

Why is over-dispersion relevant in general?

For count or binary outcomes, a simple statistical model can sometimes assume less variability than the data actually contain. Over-dispersion means that observed variability is greater than the model assumes. When present, it can make standard errors too small and intervals too narrow if not handled appropriately. This general issue is relevant when assessing analyses of binary outcomes such as rescue therapy, but the posted rescue outcome does not include enough information to establish whether over-dispersion was assessed or how its analysis was conducted.

The registry does not report the full statistical analysis plan, the original sample-size calculation, interim analyses, or detailed model diagnostics. These omissions limit independent reconstruction of the analyses. The record does state the primary endpoint, the primary analysis population, the imputation approach, and the sponsor-reported Bayesian comparisons, allowing those elements to be described without assuming additional methods.

11. Confidence Intervals and Precision

For each LY2944876 dose compared with placebo, the sponsor reports a 90% credible interval for the posterior mean difference. The interval for 10 mg is -1.05 to -0.52; for 15 mg, -1.07 to -0.54; for 30 mg, -1.41 to -0.89; and for 50 mg, -1.30 to -0.78. Each of these reported intervals lies below zero. This describes the location and precision of the sponsor's model-based estimates, conditional on that analysis and its assumptions.

The sponsor-reported comparisons with exenatide extended-release have 90% credible intervals of 0.08 to 0.61 for 10 mg, 0.07 to 0.59 for 15 mg, -0.28 to 0.24 for 30 mg, and -0.17 to 0.35 for 50 mg. The intervals for the 30 mg and 50 mg comparisons include zero; the intervals for the 10 mg and 15 mg comparisons do not. These interval statements describe the sponsor's reported Bayesian analyses and should not be confused with the recalculated confidence intervals.

In general, an interval conveys more than whether it includes zero: its width indicates the range of values compatible with the data and method, while its position indicates the direction and scale of the estimate. A narrow interval is more precise than a wide interval, all else equal. Interval width does not account for every source of uncertainty, such as unreported modeling choices or multiplicity.

The registry's own interval estimates take precedence because the simple recalculations differ from the reported analyses. The recalculated intervals are unadjusted and based on posted summary statistics; they do not reproduce the Bayesian analysis. The registry record does not report a multiplicity adjustment or enough model detail to assess how uncertainty was handled across all comparisons.

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 illustrations describe hypothetical future planning based on the observed effect in a particular posted comparison. They are not the original trial's enrollment justification, which is not reported in the registry record. For the continuous outcomes shown, the figures are participants per arm, not a combined total. The registry's planning description specifies a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. The effect to detect is set to the stated fraction of the observed effect.

The 80% power column gives the planned number of participants per arm for a design that would have an 80% probability of detecting the selected effect under the stated planning assumptions, if that effect and the assumptions were correct. The 90% power column gives the corresponding number per arm for a 90% probability. These are planning illustrations, not descriptions of the trial's actual power or of the probability that a result is true. This is not observed power.

Rows planning for 100% of the observed effect use the full observed effect as the target; rows planning for 75% or 50% use smaller target effects. Smaller effects are harder to distinguish from random variation, so they require more participants under otherwise similar assumptions. The figures can rise substantially when the planned effect is reduced, particularly for outcomes with modest observed differences or greater variability.

For time-to-event outcomes, the registry's general planning description says the figures represent total events using the Schoenfeld approach. That distinction does not change the interpretation for the continuous outcomes described in this trial, where the figures are participants per arm. The original design assumptions, including the effect size used for the trial's enrollment target, are not reported in the registry record.

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 = -0.780; pooled standard deviation σ = 0.885.
2. n = 2 × (1.96 + 0.842)² × 0.885² / -0.780² = 20.2, rounded up to 21 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 <- -0.78                      # observed difference in means
sigma <- 0.885432918402887                     # 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(2119819)
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 = 21 (80% power), 28 (90% power)
 75% of observed effect: n per arm = 36 (80% power), 49 (90% power)
 50% of observed effect: n per arm = 81 (80% power), 109 (90% power)
power.t.test: 21.2 
Simulated power at n = 21 per arm: 0.794 (20000 replicates)

The simulation draws many sets of group means and variances at the planned sample size and counts how often the two-sample t-test rejects the null. It should land close to the target power; at very small required sample sizes it can fall below it, because the normal approximation is rough there.

14. Analyses Reported by the Sponsor

The registry's reported primary comparisons use a Bayesian method and present posterior mean differences with 90% credible intervals. Separately, the recalculated comparisons use posted means and standard errors to produce simple unadjusted estimates and p values. These methods do not necessarily yield the same results because the recalculations do not reproduce the sponsor's Bayesian model or its assumptions.

The registry record marks a discrepancy between the sponsor's results and the recalculations. In that situation, the sponsor's own analysis takes precedence for interpretation of the trial's primary results. The recalculations can serve as a transparent check based on the summary statistics, but they are not a replacement for the analysis reported by the sponsor.

The 90% credible intervals and reported p values should be described as the sponsor presents them. The registry record does not explain the p-value calculation in enough detail to assign it a more specific interpretation, nor does it provide the full Bayesian model, prior distributions, or complete statistical analysis plan. Accordingly, the reported posterior estimates and intervals are the most faithful account of the sponsor's analysis, while the simplified recalculations are best treated as a separate, unadjusted comparison.

15. Multiplicity, Missing Data and Analysis Populations

The primary outcome includes comparisons of several LY2944876 doses with placebo and with exenatide extended-release. Testing multiple comparisons creates a multiplicity issue: when many hypotheses are evaluated, the chance of at least one apparently unusual result can increase if each test is considered separately. In general, a prespecified multiplicity strategy can control or otherwise address that issue. The registry record does not report whether or how the primary comparisons were adjusted for multiplicity.

The record also lists numerous secondary outcomes, and the available secondary results include different measurement types and time points. The registry does not report a multiplicity adjustment for those analyses. Consequently, secondary results should be interpreted cautiously, especially when considered alongside many outcomes. The record does not establish which secondary comparisons, if any, were prespecified as confirmatory.

For the primary endpoint, the analysis population consisted of randomized participants with baseline and at least one post-baseline HbA1c measurement at Week 12. The registry states that missing observations were imputed using a Bayesian simple linear regression longitudinal model. It does not report the model specification, assumptions about the missingness process, details of prior distributions, or sensitivity analyses under alternative assumptions. Imputation can retain participants who lack an observed outcome, but the resulting estimates rely on the model and its assumptions.

The registry record does not report a complete statistical analysis plan, interim analyses, or the trial's original sample-size justification. It also does not give enough information to reconstruct all analysis populations or determine how every secondary endpoint handled missing data. These omissions constrain independent assessment of the planned hierarchy of analyses and the robustness of results to alternative methods.

16. Safety Results

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

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 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%), cataract (1.4%), endometrial adenocarcinoma (1.4%), cataract operation (1.4%), branchial cyst (1.4%).

The adverse-event record lists the number at risk and the number affected in each group. For serious adverse events, the affected counts are 0 among 66 at risk for 10 mg LY2944876, 2 among 71 for 15 mg, 2 among 73 for 30 mg, 3 among 70 for 50 mg, 3 among 69 for exenatide extended-release, and 2 among 71 for placebo. These are counts, not comparisons adjusted for differences between groups or follow-up.

The registry lists several serious-event terms at 1.4% each: unstable angina, cataract, endometrial adenocarcinoma, cataract operation, branchial cyst, and lower abdominal pain. The record does not provide group attribution for those listed terms in the information. Death counts are not reported in the registry record.

Other adverse events affected 34 among 66 at risk in the 10 mg group, 47 among 71 in the 15 mg group, 39 among 73 in the 30 mg group, 46 among 70 in the 50 mg group, 38 among 69 in the exenatide group, and 16 among 71 in the placebo group. These descriptive counts do not establish that a difference between groups is caused by treatment. The registry record does not provide the event-specific detail needed for a fuller safety comparison.

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

The posted results describe changes in a laboratory measure and other recorded outcomes among participants assigned to the study groups. They show group averages and model-based comparisons, not the outcome for every participant. At Week 12, mean HbA1c change was negative in each group, and the sponsor's reported comparisons with placebo have 90% credible intervals below zero. Comparisons with the active comparator had a different pattern, with some intervals including zero and others not.

These results do not, by themselves, establish how a participant should be treated or whether a particular treatment is appropriate. They do not capture every potential benefit, harm, or outcome relevant to an individual. This trial's primary endpoint was HbA1c change at Week 12, and the registry reports a specific analysis population and imputation method. Interpretation is therefore limited to that endpoint and analysis framework.

Statistical estimates also do not eliminate uncertainty. The sponsor's Bayesian results depend on a model whose full details are not reported in the registry record. The unadjusted recalculations are a separate approach and should not override the sponsor's analyses because the record indicates a discrepancy. The record also does not report a multiplicity strategy, full analysis plan, or all details needed to assess missing-data assumptions.

The findings should be read as results from a completed phase 2 comparison involving 420 enrolled participants, with the limitations of the reported analyses kept in view. They do not establish how results would generalize beyond the participants and study conditions represented in the registry, nor do they address outcomes not measured or reported. The secondary and safety summaries provide additional descriptive context but cannot fill in unreported analyses or determine causality on their own.

18. Why This Trial Matters Statistically

This trial is statistically notable because it compares multiple doses of one investigational drug with both placebo and an active comparator while measuring a continuous primary endpoint. That structure makes it important to distinguish the within-group mean changes from the between-group comparisons. A group mean can appear substantial while the question of interest remains whether its change differs from another group's change, with uncertainty considered.

The record also illustrates why the sponsor's analysis and a simple recalculation should not be conflated. The sponsor reports Bayesian posterior comparisons, while recalculations based on group summaries produce unadjusted estimates. When these approaches disagree, the sponsor's analysis takes precedence as the trial's reported inferential result. Transparent separation of these analyses helps prevent a summary-statistic check from being mistaken for the trial's original model.

Several comparisons and numerous secondary outcomes also make multiplicity relevant. The registry record does not report how multiplicity was handled. That absence affects the weight that can be placed on unadjusted secondary findings and on separate comparisons considered in isolation. Missing-data handling matters as well: the primary outcome used a specified imputation model, but the record does not provide enough detail to assess its assumptions fully.

Finally, precision is not conveyed by a mean alone. The reported credible intervals give readers a sense of uncertainty around the sponsor's comparisons, while the group standard errors describe uncertainty around group means. Together, these details support a more careful reading than a simple ranking of group averages. The trial is a useful example of interpreting endpoint definitions, analysis populations, model-based estimates, and evidence limitations as connected parts of a statistical result.

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, allocation ratio, full randomization procedure, or complete statistical analysis plan. As a result, the planned assumptions underlying the 420-participant enrollment cannot be assessed from the record. Planning illustrations based on observed effects do not reconstruct the trial's original design rationale.

The primary analysis includes randomized participants with baseline and at least one post-baseline HbA1c value at Week 12, rather than necessarily every randomized participant. Missing observations were imputed using a Bayesian simple linear regression longitudinal model, but details of the model and missingness assumptions are not reported. Sensitivity analyses for alternative assumptions are also not reported in the registry record.

The sponsor's Bayesian results cannot be fully reproduced from the record because the model specification and prior distributions are not reported. The record indicates a discrepancy with the simpler unadjusted recalculations, making it especially important not to treat those recalculations as the sponsor's analysis. The multiplicity approach for the primary comparisons and secondary outcomes is not reported.

Although 15 outcome measures and 14 posted secondary outcomes are listed, the record contains detailed numerical results for only a subset of secondary endpoints. The available secondary results cannot establish the overall behavior of every outcome named in the registry. Safety summaries are also limited: the record gives group counts for serious and other adverse events, but not a complete event-by-event comparison in the information reported. Death counts are not reported in the registry record.

These limits affect what can be independently checked and how broadly the results can be interpreted. They do not change the figures that are posted, but they leave important questions about analysis choices, missing data, multiplicity, and the completeness of outcome reporting unresolved.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: A Study to Compare a New Drug for Type 2 Diabetes to Placebo and to a Treatment Already Available for Type 2 Diabetes — NCT02119819.
• Registry record last updated 2021-05-27; status last verified 2021-05.
• Results first posted 2021-04-14.
• 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, double-masked, parallel-group phase 2 trial of LY2944876 in participants with type 2 diabetes. Enrollment was 420. Participants were assigned to one of four LY2944876 doses, exenatide extended-release, or placebo; stable metformin use continued for participants taking it at entry.

The primary endpoint was change from baseline in HbA1c at Week 12. The registry reports Bayesian comparisons with placebo and exenatide extended-release, as well as group means and standard errors. The primary analysis population and a Bayesian longitudinal imputation method for missing observations are specified, but the full model and analysis plan are not reported.

Secondary and safety results add descriptive context, with limitations in multiplicity reporting and outcome detail. The registry's own Bayesian analysis takes precedence over the unadjusted recalculations because the record indicates a discrepancy.