Evaluate the Efficacy and Safety of Saxagliptin Added to Insulin Monotherapy or to Insulin Combined With Metformin in Chinese Subjects With Type 2 Diabetes Who Have Inadequate Glycaemic Control: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Evaluate the Efficacy and Safety of Saxagliptin Added to Insulin Monotherapy or to Insulin Combined With Metformin in Chinese Subjects With Type 2 Diabetes Who Have Inadequate Glycaemic Control (NCT02104804), a randomized phase 3 study. The primary outcome is Change in HbA1c From Baseline to Week 24.

2. Clinical Question

The clinical question is framed for people receiving insulin alone or insulin with metformin whose glycaemic control was considered inadequate. It is not a comparison of saxagliptin with insulin, nor does it describe a general population with Type 2 diabetes. The record does not report the criteria used to define inadequate control, the eligibility thresholds, or how many participants were receiving each background regimen. The study title identifies the population as Chinese subjects in China, and its condition is Type 2 Diabetes Mellitus.

The main statistical focus is the mean change in HbA1c over the stated follow-up period. The associated estimate compares the groups after accounting for the analysis described by the registry, while the group-level means describe change within each group. These are related but not interchangeable questions: a within-group change does not by itself establish the size or precision of a treatment contrast. The registry also includes a secondary outcome about achieving HbA1c below 7%, along with other glucose and insulin-dose outcomes.

3. Trial Design

The registry record describes NCT02104804 as a randomized, parallel-group, Phase 3b trial sponsored by AstraZeneca. It is marked completed. The enrollment field reports 953. The study began on 2014-05-07, and primary completion is recorded as 2016-02-26. These registry entries describe the study record; they do not establish how many participants were included in each analysis.

The official title calls the trial multicenter and randomized, and describes it as double-blind. The registry masking field is coded as quadruple. The record does not explain which people or roles were masked, or how masking was maintained. Randomized allocation supports a between-group comparison, but details such as the randomization sequence, allocation concealment, and allocation ratio are not reported in the registry record. The parallel model indicates that the groups were compared alongside one another rather than through a crossover design.

The intervention was saxagliptin 5 mg, and the comparator was placebo for saxagliptin. Both were added to insulin alone or insulin combined with metformin, according to the arm descriptions. The group labels in the posted primary results are Saxagliptin Plus Insulin and placebo plus insulin; the safety descriptions explicitly state insulin with or without metformin. The registry does not report how participants were distributed across the background regimens or whether those subgroups had separate primary analyses.

The primary endpoint was change in HbA1c from baseline to Week 24, expressed as a least-squares mean with a 95% confidence interval. The registry lists five secondary outcomes, including postprandial glucose measures, the proportion reaching an HbA1c threshold, fasting plasma glucose, and daily insulin dose. The detailed protocol, statistical analysis plan, eligibility criteria, and sample-size justification are not reported in the registry record. In particular, the enrollment figure should not be treated as a description of the analysis population for every outcome.

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: 953 patients were enrolled, of which 641 patients took at least 1 dose of study drug and entered the lead-in period. A total of 312 patients did not enter the lead-in period (289 did not fulfill eligibility criteria, 21 due to subject decision, 1 each for severe non-compliance to protocol and other reasons) and 466 patients were randomized.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: 229 and 227 indicate the number of participants with non-missing baseline and Week 24 values in the Full Analysis Set of 232 and 230 respectively.

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: Note that 462 indicates the number of patients in the Full Analysis set (out of 466 randomized, 1 did not take study drug and 3 did not have 1 baseline or post baseline efficacy assessment).

6. Primary Endpoint

The registry lists the primary outcome as Change in HbA1c From Baseline to Week 24, measured over Baseline to 24 weeks.

The primary endpoint measures change in HbA1c from baseline to Week 24. HbA1c is a marker of glycaemic control that reflects glucose exposure over time, so change from baseline provides a way to describe whether the measured level moved during follow-up. The registry identifies the unit as percentage change and reports least-squares means with 95% confidence intervals. The precise measurement and laboratory procedures are not reported in the registry record.

The endpoint is useful because it gives a continuous outcome: it retains information about the amount of change rather than reducing the result to whether a participant crossed a cutoff. The group estimates describe average change under the stated analysis, while the between-group contrast addresses how those average changes differ. Both the size of that contrast and its uncertainty matter. A confidence interval conveys a range of values compatible with the data and analysis, subject to the model and its assumptions; it is not a guarantee that the true effect lies within that range.

The posted primary record states that 229 and 227 participants had non-missing baseline and Week 24 values in the Full Analysis Set, from groups of 232 and 230, respectively. This indicates that the reported endpoint values are based on participants with the specified measurements, but does not explain why measurements were missing or how missing values were handled in the broader analysis. Those details are not reported in the registry record. The endpoint should therefore be read as a Week 24 comparison based on the analysis described, not as a complete account of every participant’s trajectory.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

Statistical methods help distinguish a difference in observed group summaries from an estimate of a treatment contrast, and distinguish the size of an estimate from its uncertainty. The primary endpoint is continuous, while the posted secondary threshold outcome is binary. The methods and interpretations therefore differ across outcomes. The following explains the statistical ideas in general terms and relates them to what the registry record reports.

What does a least-squares mean represent?

A least-squares mean is a model-based estimate of a group mean. It is often used when an analysis adjusts for other variables or accounts for features of the study design. Unlike a simple arithmetic average, it can reflect the fitted statistical model and the adjustment terms included in that model. The registry reports least-squares mean changes for the primary endpoint and identifies ANCOVA as the method for the between-group contrast. The covariates included, model specification, and diagnostic checks are not reported in the registry record, so the exact adjustment cannot be reconstructed from the posted information.

The two group estimates are useful descriptions, but the direct comparison is generally the most relevant estimate for a randomized parallel-group trial. A change within one group can reflect treatment, background care, time, measurement variation, or other influences. A between-group contrast uses the randomized comparison to estimate how the groups differed under the trial conditions. Randomization supports this interpretation when implemented and analysed appropriately, but details of implementation and the full analysis plan are not available in the registry record.

How should a confidence interval be read?

A confidence interval expresses uncertainty about an estimated quantity under a statistical model and its assumptions. A narrower interval indicates greater precision than a wider interval for the same type of estimate. It does not show the probability that a particular fixed effect is inside the interval, and it does not cover uncertainty arising from every possible source, such as bias from missing data or departures from the model. The primary comparison is reported with a 95% confidence interval. Its endpoints show the range of values compatible with the estimate and analysis at that confidence level.

The group-level confidence intervals and the confidence interval for the treatment contrast answer related but distinct questions. The first describe each estimated mean change; the second describes the estimated difference between groups. Overlap or non-overlap of separate group intervals is not a substitute for the reported direct comparison. The ANCOVA contrast and its own interval should be used to assess the primary comparison.

What does the p-value add?

A p-value quantifies how incompatible the observed data are with a specified null hypothesis, under the analysis assumptions. A small p-value is not the probability that the null hypothesis is true, and it does not measure the size or importance of an effect. The primary result reports p < 0.001. That is evidence against the null comparison under the reported analysis, but precision and substantive importance still require examination of the estimated contrast, its confidence interval, and the clinical context. No threshold for clinical importance is reported in the registry record.

How do continuous and binary outcomes differ?

For the continuous primary endpoint, the result describes average change in HbA1c. The secondary threshold outcome instead counts the proportion reaching HbA1c below 7%. A binary endpoint is easy to interpret as a proportion meeting a criterion, but it discards the amount by which values are above or below the threshold. Its estimate can also be sensitive to the chosen cutoff. A risk difference describes an absolute contrast between proportions; a risk ratio or odds ratio describes a relative contrast. These are different scales, and none should be treated as interchangeable.

The secondary outcome has a recalculated unadjusted risk difference and an unadjusted p-value, as well as the registry’s reported difference in proportions. The recalculation uses the posted summary counts and is not an adjusted model. It can be useful as a transparent check, but it does not replace a sponsor analysis or account for a prespecified model, covariates, or multiplicity. No discrepancy is flagged for this result. The assumptions used for any recalculation should still be distinguished from the trial’s own analysis.

Why does over-dispersion matter?

In general, over-dispersion occurs when observed variability is greater than a statistical model expects. If a model assumes less variability than the data contain, standard errors can be too small and confidence intervals too narrow, making evidence appear more precise than it is. Appropriate modelling may account for extra variability, depending on the outcome and design. The registry record does not report an over-dispersion assessment or a model for such a feature, and the primary endpoint is reported through an ANCOVA comparison. Over-dispersion should not be assumed to have occurred in this trial.

How should planning illustrations be used?

The planning illustrations use the observed binary effect as a reference and show participants per arm for 80% and 90% power at different fractions of that observed effect. They are not the trial’s sample-size justification and are not observed power. The stated assumptions are a two-sided significance level of 0.05, equal allocation, and a normal approximation; the effect to detect is the stated fraction of the observed effect. Such calculations are conditional illustrations: they depend on the chosen effect, design assumptions, and analysis method. They cannot establish that the original trial was adequately planned because the registry record does not report the original sample-size rationale or its assumptions.

Power is a prospective design concept: it concerns the chance that a planned analysis will detect a specified effect under stated assumptions. Calculating power after observing the result does not measure how persuasive or reliable that result is. The useful role of the planning illustrations is to show how required group sizes change when a smaller target effect is considered under the stated assumptions, not to make a claim about the observed trial’s power.

10. Analyses Reported by the Sponsor

The registry’s primary analysis is an ANCOVA comparison of least-squares means. It reports a difference of -0.58, with a 95% confidence interval from -0.72 to -0.45 and p < 0.001. That is the sponsor-reported analysis in the registry record and should be the main inferential result for the primary endpoint.

A recalculation from posted summary statistics is a simple unadjusted calculation and is not equivalent to the trial’s model-based analysis. It may not incorporate the same covariate adjustment, handling of missing data, or other prespecified features. The registry does not provide enough detail to reproduce the ANCOVA model or assess every analysis choice. For that reason, the recalculation is best treated as a descriptive cross-check, not as a replacement for the reported analysis.

No discrepancy is flagged between the posted primary result and a recalculation. The binary secondary outcome has both a registry-reported difference in proportions and a simple unadjusted recalculation; their p-values differ. That does not, by itself, establish an error: the methods may differ, and the record does not provide the full specification needed to reconcile them. The registry’s reported analysis remains the source for the trial’s own result, while the recalculation should be read as an independent, unadjusted summary of the posted counts.

11. Multiplicity, Missing Data and Analysis Populations

The registry record identifies a primary endpoint and lists secondary outcomes, but it does not report a full statistical analysis plan. In particular, the planned hierarchy among endpoints, the number of formal hypotheses, and any multiplicity adjustment are not reported in the registry record. Multiplicity matters because testing multiple outcomes or comparisons can increase the chance of at least one apparently positive result even when no corresponding effect exists. Without a stated testing strategy, secondary p-values should be interpreted cautiously and should not be treated as independent confirmation of the primary result.

The primary endpoint has a reported ANCOVA comparison with a 95% confidence interval and p < 0.001. The record does not explain whether this p-value was adjusted for any other testing, although the primary endpoint is distinguished from the secondary outcomes. The secondary HbA1c threshold analysis reports a sponsor p-value and an unadjusted recalculation. The methods and the lack of a reported multiplicity plan mean that these secondary findings have a different evidential status from the primary result.

Missing-data handling is also important. The primary endpoint description reports 229 and 227 participants with non-missing baseline and Week 24 values in the Full Analysis Set, from groups of 232 and 230. The record does not state why values were missing, whether missingness differed by reason or group, or how missing outcomes were handled in the model. It does not report sensitivity analyses that test alternative assumptions about missing values. The posted estimate should therefore be read in light of the available endpoint measurements and the analysis described, without assuming a particular imputation method.

The phrase Full Analysis Set is used in the endpoint description, but the registry does not give a full definition of that population or explain its relationship to all randomized participants. The enrollment field reports 953, but that figure alone does not specify how many people were randomized, treated, or included in each outcome analysis. Details of exclusions and analysis populations are not reported in the registry record. The record also does not report interim analyses, stopping rules, or the sample-size justification. These omissions limit assessment of the prespecified nature of the analyses and of how closely the posted results follow the original plan.

12. Safety Results

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

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 0% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: diabetes mellitus inadequate control (0.9%), cerebral infarction (0.9%), deafness neurosensory (0.4%), vitreous haemorrhage (0.4%), intestinal obstruction (0.4%).

The safety record gives serious adverse events for two groups: placebo once daily plus insulin with or without metformin, and saxagliptin 5 mg once daily plus insulin with or without metformin. The registry reports 14 serious adverse events among 231 at risk in the placebo group and 12 among 234 at risk in the saxagliptin group. These figures describe the posted serious-event counts and the associated at-risk group sizes; they are not, on their own, a complete safety comparison.

The record also lists other adverse events: 96 among 231 at risk in the placebo group and 89 among 234 at risk in the saxagliptin group. The serious-event terms listed include diabetes mellitus inadequate control and cerebral infarction at 0.9% each, and deafness neurosensory, vitreous haemorrhage, intestinal obstruction, and acute pancreatitis at 0.4% each. These terms are descriptions of reported events, not evidence about cause. The record does not provide detailed event narratives, exposure-adjusted rates, or a basis for attributing an event to study treatment.

Deaths are not reported in the registry record. The figures should be interpreted descriptively: the record does not give a statistical safety comparison, an account of event severity beyond the serious-event classification, or enough detail to establish whether the groups had comparable observation time. Differences in raw counts cannot establish that one regimen is safer. Safety conclusions require the full study context, including definitions, exposure, follow-up, and the handling of events.

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

The primary result is an estimate of the difference in mean HbA1c change between the randomized groups over the stated baseline-to-Week 24 period. The registry reports a negative between-group least-squares mean difference, with a 95% confidence interval that does not include zero and p < 0.001. This indicates a statistically distinguishable group contrast under the reported model. It does not, by itself, establish how important that difference is to patients or whether it meets a threshold for clinical importance; no such threshold is reported in the registry record.

The result applies to the study population and treatment conditions described: Chinese subjects in China with Type 2 diabetes and inadequate control on insulin alone or insulin with metformin, with saxagliptin or placebo added. It should not automatically be generalized to people with different characteristics, other background treatments, or different follow-up. The registry does not report subgroup-specific primary estimates, so the overall comparison should not be interpreted as proof of the same effect within each background-treatment subgroup.

The result also does not establish longer-term outcomes. The primary endpoint is measured at Week 24, and the registry information does not provide a longer-term primary comparison. Nor does the HbA1c contrast alone answer questions about complications, quality of life, or the overall balance of benefits and harms. Those outcomes are not reported as part of the primary result.

Statistical significance should not be confused with certainty. The confidence interval describes precision under the analysis assumptions, while missing-data uncertainty, possible departures from model assumptions, and unreported design details remain relevant. The registry record does not report enough information to assess all of these issues. The best reading is therefore specific: the posted analysis estimates a primary HbA1c contrast in this trial, with the stated interval and p-value, while broader clinical and safety conclusions require information beyond that estimate.

14. Why This Trial Matters Statistically

This trial illustrates why the contrast between groups matters more than reading each group’s change in isolation. Both group summaries describe movement from baseline, but the randomized comparison estimates the difference between those movements. The direct ANCOVA result, its confidence interval, and its p-value provide complementary information: the estimate describes the contrast, the interval communicates precision, and the p-value addresses compatibility with a null hypothesis under the analysis assumptions.

It also shows the distinction between a continuous endpoint and a threshold response. Mean change in HbA1c preserves the magnitude of change across the measurement range, while the HbA1c-below-7% outcome classifies participants around a cutoff. The results can be related without being redundant. A reader should not expect the two measures to answer exactly the same question, and should consider the separate methods used for each.

The record is a useful example of model-based reporting. Least-squares means and an ANCOVA contrast communicate an adjusted analysis, but readers need details of the model and analysis population to fully assess it. Those details are not reported in the registry record. Recalculation from summary counts can offer a transparent check for a binary endpoint, yet it is unadjusted and does not reproduce every feature of a sponsor analysis.

Finally, the case demonstrates why secondary results, missing-data handling, multiplicity, and safety denominators matter. A statistically significant primary comparison does not establish consistency across all outcomes, and a list of adverse-event counts does not by itself determine comparative safety. The registry’s omissions leave important questions about prespecification and analysis choices unanswered. Clear separation of the primary finding, supporting outcomes, and limitations helps prevent a narrow statistical result from being interpreted more broadly than the record warrants.

15. 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 detailed statistical analysis plan, including the covariates in the primary ANCOVA, model checks, or a complete description of analysis populations. This makes it difficult to reproduce the primary estimate or assess all assumptions. Although the endpoint description identifies the Full Analysis Set and gives the number with non-missing baseline and Week 24 values, the set’s full definition and the reasons for missing measurements are not reported.

The trial’s enrollment field reports 953, but that figure does not explain the numbers randomized, treated, or included in each analysis. Group counts differ across endpoint and safety descriptions, and the registry does not explain those differences. The record also does not report how participants were divided between insulin alone and insulin with metformin, or provide subgroup-specific estimates. Consequently, the overall result cannot establish whether effects were consistent across those background regimens.

The registry lists secondary outcomes, but detailed results are not posted for all of them in the information available. It does not report whether multiplicity was controlled, whether interim analyses were conducted, or how the original sample size was justified. The planning illustrations are based on the observed binary effect under stated assumptions and cannot fill that gap. Safety information is likewise limited: deaths are not reported, and event counts alone do not describe exposure time, event adjudication, or causal attribution.

Finally, the primary follow-up is baseline to Week 24. The result should not be extended to longer-term outcomes or populations not described by the record. These limitations do not erase the reported primary estimate, but they define what can be concluded from the registry information and where additional protocol, analysis, and full-report details would be needed.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Evaluate the Efficacy and Safety of Saxagliptin Added to Insulin Monotherapy or to Insulin Combined With Metformin in Chinese Subjects With Type 2 Diabetes Who Have Inadequate Glycaemic Control — NCT02104804.
• Registry record last updated 2017-10-02; status last verified 2017-04.
• Results first posted 2017-10-02.
• 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.

19. Record Summary

NCT02104804 is a completed randomized, parallel-group Phase 3b trial sponsored by AstraZeneca. It evaluated saxagliptin 5 mg versus placebo added to insulin alone or insulin with metformin in Chinese subjects in China with Type 2 diabetes and inadequate glycaemic control. Enrollment is recorded as 953.

The primary endpoint was change in HbA1c from baseline to Week 24. The registry reports a least-squares mean difference of -0.58, with a 95% confidence interval from -0.72 to -0.45 and p < 0.001. Five secondary outcomes are listed, including an HbA1c threshold outcome. Serious and other adverse-event counts are also posted. Details of the full statistical plan, multiplicity handling, sample-size justification, and missing-data methods are not reported in the registry record.