This page separates reported trial information from statistical interpretation. The numerical results reported here are restricted to the trial data posted on ClinicalTrials.gov for DPPOS. Where ClinicalTrials.gov provides an endpoint but no formal statistical analysis for that endpoint, the page describes the appropriate analytical framework without creating an unreported result.
1. Trial at a Glance
DPPOS, the Diabetes Prevention Program Outcomes Study, was a completed randomized phase 3 prevention study sponsored by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). The registry reports a parallel design, no masking, 3 arms, and enrollment of 2779 participants.
| Feature | DPPOS |
|---|---|
| Brief title | Diabetes Prevention Program Outcomes Study |
| Phase | Phase 3 |
| Status | COMPLETED |
| Therapeutic area | Endocrinology |
| Conditions | Diabetes Mellitus; Cancer; CVD |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | NONE |
| Primary purpose | PREVENTION |
| Enrollment | 2779 |
| Lead sponsor | National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) |
| Sponsor type | NIH |
| Registered primary endpoints | 4 |
| Outcome measures posted | 17 |
| Statistical analyses posted | 2 |
| Reported method | Log-rank test |
| Reported effect measure | Hazard ratio |
| Hypothesis type for posted analyses | Superiority |
2. Clinical Question
The broad clinical question was how the randomized prevention strategies followed in DPP and DPPOS relate to long-term outcomes including diabetes, aggregate microvascular complications, cancer, and major adverse cardiovascular events.
Population
The DPPOS registry describes a long-term follow-up study of participants from the Diabetes Prevention Program. The statistical-analysis record states that DPPOS is the long-term follow-up of the 3234 participants randomized in the Diabetes Prevention Program and that all surviving DPP participants were invited to the follow-up study.
Interventions
The registry profile lists DPPOS Group Lifestyle, Metformin, DPPOS Boost Lifestyle, and Intensive Lifestyle Group Session among the interventions, with behavioral and drug intervention types represented.
Comparator structure
The statistical analyses posted on ClinicalTrials.gov compare Original Lifestyle with Original Placebo and Original Metformin with Original Placebo for mortality.
Primary question
The registered primary outcomes assess the occurrence of diabetes, aggregate microvascular complications, total cancer except non-melanoma skin cancer, and MACE consisting of myocardial infarction, stroke, or cardiovascular death.
3. Trial Design
Original Lifestyle
- Serious adverse events: 649/916 affected/at risk in the ClinicalTrials.gov record.
- Included in the reported mortality comparison with Original Placebo.
Original Metformin
- Serious adverse events: 657/927 affected/at risk in the ClinicalTrials.gov record.
- Included in the reported mortality comparison with Original Placebo.
Original Placebo
- Serious adverse events: 646/936 affected/at risk in the ClinicalTrials.gov record.
- Served as the comparator in both posted mortality analyses.
Registry-listed randomized groups
- Randomized Lifestyle During DPP: serious adverse events 235/1079.
- Randomized Metformin During DPP: serious adverse events 251/1073.
- Randomized Placebo During DPP: serious adverse events 238/1082.
4. Registered Primary Endpoints
The registry identifies 4 primary endpoints. All 4 are reported as having results posted, but the ClinicalTrials.gov record contains formal statistical analyses only for the secondary endpoint of mortality.
| Primary endpoint | Time frame | Registry definition | Formal statistical analysis reported? |
|---|---|---|---|
| Number of Participants With Diabetes. | Outcomes were assessed from 1996-2020 for median follow-up of 21 years and a maximum follow-up of 23 years including a median of 3 years in DPP in all participants enrolled in DPPOS. | Primary outcome defined according to American Diabetes Association criteria (fasting plasma glucose level ≥ 126 mg/dL [7.0 mmol/L] or 2-hour plasma glucose ≥ 200 mg/dL [11.1 mmol/L], after a 75 gram oral glucose tolerance test (OGTT), and confirmed with a repeat test). | No formal analysis in the posted results. |
| Prevalence of Aggregate Microvascular Complication | Outcomes were assessed from 2012-2013 (approximately 2 years). | Aggregate microvascular disease is defined as the average prevalence of 3 components: (1) retinopathy measured by photography (ETDRS of 20 or greater); (2) neuropathy detected by Semmes Weinstein 10 gram monofilament, and (3) nephropathy based on estimated glomerular filtration rate (eGFR by chronic kidney disease (CKD-Epi) equation ) (<45 ml/min, confirmed) and albumin-to-creatinine ratio in spot urine (> 30mg/gm, confirmed). | No formal analysis in the posted results. |
| Number of Participants With Total Cancer Except Non-melanoma Skin Cancer | Outcomes were assessed from 1996-2020 (approximately 24 years). | All primary cancers except non-melanoma skin cancer that occurred after randomization in DPP | No formal analysis in the posted results. |
| Number of Participants With Major Adverse Cardiovascular Events (MACE): Myocardial Infarction (MI), Stroke, or Cardiovascular Death (CVD) | Outcomes were assessed from 1996-2019 over a total median follow-up of 21 years since DPP randomization | MACE includes MI, stroke or CVD death that occurred after randomization and adjudicated by an outcomes committee who are blinded to treatment assignment. | No formal analysis in the posted results. |
5. Planned Analysis of the Primary Endpoints
ClinicalTrials.gov reports that results are posted for each of the 4 primary endpoints, but the ClinicalTrials.gov record does not provide a formal treatment comparison for any of them.
Number of Participants With Diabetes
This is a binary clinical outcome: each participant either meets the registry definition of diabetes during the relevant assessment period or does not. For a randomized comparison of cumulative diabetes incidence over time, a time-to-event framework such as Kaplan-Meier estimation and a log-rank test can be used when time to diabetes onset and censoring information are available. A Cox model can then provide a hazard ratio. If only fixed-time proportions are available, an appropriate binary-outcome comparison could instead be used.
Prevalence of Aggregate Microvascular Complication
The registry defines aggregate microvascular disease from the average prevalence of 3 components: retinopathy, neuropathy, and nephropathy. Because the stated endpoint is prevalence over an approximately 2-year assessment period, the appropriate analysis depends on the exact participant-level structure and analysis plan. A binary comparison of prevalence can be considered when the endpoint is represented as affected versus unaffected at the specified assessment, while the composite construction requires careful attention to how the component measures are combined.
Total Cancer Except Non-melanoma Skin Cancer
The registered endpoint counts primary cancers except non-melanoma skin cancer occurring after DPP randomization. If cancer occurrence is analyzed as time to first qualifying cancer, Kaplan-Meier methods, a log-rank comparison, and a Cox model are natural time-to-event approaches. Competing events and censoring would need to be handled according to the prespecified statistical analysis plan.
Major Adverse Cardiovascular Events
MACE includes myocardial infarction, stroke, or cardiovascular death after randomization, with events adjudicated by an outcomes committee blinded to treatment assignment. Because the endpoint is a composite of clinically meaningful time-to-event outcomes, a time-to-first-MACE analysis would commonly use Kaplan-Meier estimation, log-rank testing, and a Cox model when the statistical analysis plan specifies that framework.
6. Results: Reported Mortality Analyses
The statistical analyses posted on ClinicalTrials.gov concern mortality, a secondary time-to-event endpoint. Both comparisons use the log-rank method and report hazard ratios with two-sided 95% confidence intervals under a superiority hypothesis.
Original Lifestyle vs Original Placebo
Hazard ratio for mortality
95% CI: 0.81–1.28 · P = 0.87
Analysis: Log Rank · Effect measure: Hazard Ratio (HR)
| Feature | Reported value |
|---|---|
| Outcome measure | Mortality |
| Endpoint type | Time-to-event |
| Groups compared | 1 Original Lifestyle vs 3 Original Placebo |
| Method | Log Rank |
| Effect measure | Hazard Ratio (HR) |
| Estimate | 1.02 |
| 95% confidence interval | 0.81–1.28 |
| P-value | 0.87 |
| Hypothesis type | Superiority |
The reported hazard ratio of 1.02 is the estimated relative hazard of mortality for Original Lifestyle compared with Original Placebo under the reported survival analysis. An HR of 1 would represent equal estimated hazards; an HR above 1 means the estimated hazard is higher in the first-named group under the model.
The estimate does not mean that 1.02 times as many participants died, nor does it represent a 2% difference in cumulative mortality. A hazard ratio is a relative time-to-event measure rather than a direct difference in the proportion of participants who experience the event.
The two-sided 95% confidence interval of 0.81–1.28 describes statistical uncertainty around the estimated hazard ratio. Because the interval spans 1, the data are compatible with a range of relative mortality hazards in either direction under the model.
The P = 0.87 value addresses the statistical evidence against the null hypothesis used for the comparison. It does not measure the magnitude of the effect, the probability that the null hypothesis is true, or the clinical importance of the observed estimate.
As with other Cox-model hazard ratios, interpretation depends on the time-to-event analysis and its assumptions, including the proportional-hazards framework when a Cox model is used. The ClinicalTrials.gov record reports the log-rank method and hazard ratio but does not provide enough information here to independently evaluate proportional hazards, censoring patterns, or the underlying survival curves.
Original Metformin vs Original Placebo
Hazard ratio for mortality
95% CI: 0.79–1.25 · P = 0.95
Analysis: Log Rank · Effect measure: Hazard Ratio (HR)
| Feature | Reported value |
|---|---|
| Outcome measure | Mortality |
| Endpoint type | Time-to-event |
| Groups compared | 2 Original Metformin vs 3 Original Placebo |
| Method | Log Rank |
| Effect measure | Hazard Ratio (HR) |
| Estimate | 0.99 |
| 95% confidence interval | 0.79–1.25 |
| P-value | 0.95 |
| Hypothesis type | Superiority |
The reported hazard ratio of 0.99 indicates an estimated mortality hazard very close to 1 for Original Metformin relative to Original Placebo in the registry-reported analysis. Under the reported hazard-ratio framework, values below 1 favor the first-named group and values above 1 favor the comparator in terms of the estimated instantaneous event rate.
The estimate does not establish identical mortality experience for every participant, and it is not an absolute risk difference. It summarizes the relative time-to-event comparison represented by the reported analysis.
The two-sided 95% confidence interval of 0.79–1.25 includes 1 and therefore reflects substantial uncertainty around the point estimate. The interval is more informative about precision than the point estimate alone.
The P = 0.95 value describes the statistical evidence under the specified superiority testing framework. A p-value is not an effect-size measure and should not be interpreted as the probability that the treatment effect is zero.
The ClinicalTrials.gov record does not provide the underlying number and timing of deaths, Kaplan-Meier estimates, censoring distribution, or formal model diagnostics. Those omissions limit how far the reported hazard ratio can be interpreted beyond the registry result itself.
7. Statistical Methodology
Time-to-event analysis
Mortality is represented in the statistical analyses posted on ClinicalTrials.gov as a time-to-event endpoint. Unlike a simple binary mortality proportion, a time-to-event analysis retains information about when an event occurs and can accommodate participants whose event time is not observed during follow-up through censoring.
The survival function S(t) represents the probability of remaining event-free beyond time t. In a mortality analysis, the event is death and the time scale begins according to the prespecified study definition.
Kaplan-Meier estimation
A Kaplan-Meier estimator would describe the estimated mortality-free survival experience over time. It is particularly useful because participants can contribute follow-up until death, last known follow-up, or another prespecified censoring point. The registry-reported DPPOS statistical-analysis records do not post Kaplan-Meier estimates, survival probabilities, or a survival curve, so no such values are reproduced here.
Here di is the number of events at event time ti, and ni is the number at risk immediately before that time.
Log-rank test
The registry explicitly reports Log Rank as the method for both mortality comparisons. The log-rank test compares the observed and expected numbers of events between groups across follow-up and is designed for randomized time-to-event comparisons.
Conceptually, the test aggregates evidence over event times rather than comparing only the final percentage of participants who died. This is why the analysis can use information about the timing of mortality events as well as whether they occurred.
Hazard ratio
The registry reports a hazard ratio as the effect measure. A hazard ratio compares the instantaneous event rates represented by the survival analysis. An HR of 0.99, for example, is not equivalent to saying that 0.99 times the final percentage of participants died.
The hazard ratio is a relative time-to-event measure. Its interpretation depends on the analysis model and does not directly provide an absolute mortality difference.
Confidence intervals
The reported mortality analyses provide two-sided 95% confidence intervals. These intervals quantify statistical uncertainty around the hazard-ratio estimates under the relevant estimation framework. A confidence interval that crosses 1 is compatible with no difference in hazard as well as with effects in either direction within the interval.
Superiority testing
Both registry-reported mortality analyses identify Superiority as the hypothesis type. This means the statistical question was framed around whether the compared strategies differed in the specified direction of a superiority framework rather than around demonstrating that two strategies were sufficiently similar.
8. Statistical Methods Explained
Why was a log-rank test used for mortality?
Mortality is a time-to-event outcome. A log-rank test is designed to compare survival distributions between groups while incorporating the timing of observed events. It is therefore more informative for a longitudinal mortality comparison than simply comparing two end-of-study proportions.
What does a hazard ratio of 0.99 mean?
For Original Metformin versus Original Placebo, the reported HR is 0.99. This is an estimated relative mortality hazard very close to 1 under the reported analysis. It does not mean that mortality differed by exactly 1%, because a hazard ratio is not an absolute risk difference or a percentage of participants experiencing the event.
Why does the confidence interval matter?
The point estimate alone does not communicate precision. For the Original Metformin comparison, the 95% CI is 0.79–1.25. For Original Lifestyle versus Original Placebo, it is 0.81–1.28. Both intervals span 1, showing that the point estimates should not be interpreted without their associated uncertainty.
Why is a p-value not an effect size?
The p-value measures how compatible the observed data are with the null hypothesis under the specified testing procedure. It does not describe how large or clinically meaningful an effect is. The magnitude and uncertainty of the estimated effect are conveyed more directly by the hazard ratio and confidence interval.
Why can the primary endpoints not simply be assigned the same analysis?
The four registered primary endpoints are not identical statistical objects. Diabetes and cancer can be represented as event outcomes over time, whereas aggregate microvascular complication is defined as a prevalence measure based on 3 components. MACE is a composite cardiovascular endpoint. The appropriate analysis depends on the exact endpoint construction, observation schedule, censoring rules, and prespecified analysis plan. The ClinicalTrials.gov record does not provide those complete analysis specifications.
What is the difference between the DPP randomization population and DPPOS enrollment?
The registry-reported mortality analysis states that DPPOS is the long-term follow-up of the 3234 participants randomized in the Diabetes Prevention Program and that all surviving DPP participants were invited to participate. The DPPOS ClinicalTrials.gov record separately reports enrollment of 2779. The registry caveat explicitly states that not all participants randomized in the original DPP enrolled in the DPPOS follow-up study.
9. Analysis Population and Follow-up Considerations
The registry-reported mortality analyses identify DPPOS as the long-term follow-up of the 3234 participants randomized in the original Diabetes Prevention Program. The analysis-population description further states that all surviving DPP participants were invited to the follow-up study.
| Feature | Registry information reported |
|---|---|
| Original DPP randomized population referenced by mortality analysis | 3234 participants |
| DPPOS enrollment | 2779 |
| Mortality follow-up period | 1996 to 2019 |
| Death ascertainment | National Death Index search conducted in 2019 using early release data as of Dec 2018. |
| Mortality comparison 1 | Original Lifestyle vs Original Placebo |
| Mortality comparison 2 | Original Metformin vs Original Placebo |
10. Safety: Serious Adverse Events
The ClinicalTrials.gov record provides serious adverse events as affected participants over participants at risk for six labeled groups. These figures are presented exactly as reported in the registry and are not converted into percentages or subjected to an unreported comparison.
| Registry group | Affected / at risk |
|---|---|
| 1 Original Lifestyle Enrolled in DPPOS | 649/916 |
| 2 Original Metformin Enrolled in DPPOS | 657/927 |
| 3 Original Placebo Enrolled in DPPOS | 646/936 |
| 1B Randomized Lifestyle During DPP | 235/1079 |
| 2B Randomized Metformin During DPP | 251/1073 |
| 3B Randomized Placebo During DPP | 238/1082 |
These are serious-adverse-event counts, not estimates of treatment effect. The ClinicalTrials.gov record does not provide a statistical comparison, confidence interval, or p-value for these safety figures. They also distinguish between groups enrolled in DPPOS and the labeled randomized groups during the original DPP.
11. Registry Limitations and Follow-up Caveats
- Incomplete follow-up enrollment: not all participants randomized in the original Diabetes Prevention Program enrolled in the DPPOS follow-up study.
- Different ascertainment procedures: outcome ascertainment and adverse-event collection procedures differed between the original DPP and DPPOS follow-up phases.
- Subset assessments: some assessments were conducted only in subsets of participants who completed specific evaluations during follow-up.
- Primary-endpoint analysis detail: the ClinicalTrials.gov record reports that results are posted for the four primary endpoints, but they do not provide formal statistical-analysis records for those endpoints.
- Mortality analysis detail: the statistical analyses posted on ClinicalTrials.gov provide the log-rank method and hazard ratios, but not the underlying event counts, Kaplan-Meier estimates, or survival curves.
- Population interpretation: the DPPOS enrollment figure and the original DPP randomized population referenced in the mortality analysis are different quantities and should not be treated as interchangeable.
- Assessment subsets: because some assessments occurred only in subsets, prevalence-type results may not represent every enrolled participant in the same way.
12. Multiplicity, Stratification, and Other Design Topics
The ClinicalTrials.gov record supports only a limited description of these design features. The mortality analyses identify a superiority hypothesis and two comparisons against Original Placebo. The data do not specify an alpha allocation, multiplicity-adjustment procedure, interim-analysis plan, Bayesian method, non-inferiority margin, or crossover strategy.
| Design topic | What the ClinicalTrials.gov record supports |
|---|---|
| Multiplicity | Two mortality comparisons are posted; no multiplicity-adjustment procedure is reported. |
| Hypothesis | Superiority for both posted mortality analyses. |
| Statistical method | Log-rank test. |
| Effect measure | Hazard ratio. |
| Interim analysis | Not specified in the ClinicalTrials.gov record. |
| Missing-data / imputation method | Not specified in the ClinicalTrials.gov record. |
| Stratification factors | Not specified in the ClinicalTrials.gov record. |
| Bayesian methods | Not specified in the ClinicalTrials.gov record. |
| Non-inferiority margin | Not applicable to the registry-reported superiority analyses; no margin is reported. |
| Crossover | Not specified in the ClinicalTrials.gov record. |
In particular, absence of a registry-reported detail should not be interpreted as evidence that the underlying protocol contained no such procedure. It means only that the ClinicalTrials.gov record does not support a more specific statement.
13. Interpreting the Two Mortality Estimates Together
| Comparison | HR | 95% CI | P-value |
|---|---|---|---|
| Original Lifestyle vs Original Placebo | 1.02 | 0.81–1.28 | 0.87 |
| Original Metformin vs Original Placebo | 0.99 | 0.79–1.25 | 0.95 |
The two point estimates are both close to 1, while both confidence intervals include 1. The analyses posted on ClinicalTrials.gov therefore provide estimated relative mortality hazards close to the null value, with uncertainty extending on both sides of 1.
That description is deliberately narrower than saying that the interventions "had no effect." A nonsignificant p-value does not prove equivalence, and a confidence interval spanning 1 does not establish that the true effect is exactly zero. It indicates that the registry-reported analysis does not provide strong statistical evidence against the null value under the stated testing framework.
Point estimate: 1.02 for Original Lifestyle versus Original Placebo and 0.99 for Original Metformin versus Original Placebo.
Precision: the 95% confidence intervals extend below and above 1 in both comparisons.
Statistical evidence: the reported p-values are 0.87 and 0.95, respectively.
What remains unknown from the ClinicalTrials.gov record: the underlying event counts, Kaplan-Meier survival estimates, censoring patterns, and complete model specification are not included in the ClinicalTrials.gov record.
14. Why This Trial Matters Statistically
DPPOS is a useful teaching case because it combines randomized prevention, long-term follow-up, several clinically different endpoint types, and survival analysis. It also illustrates why the structure of the endpoint determines the appropriate statistical framework.
| Concept | How it appears in DPPOS |
|---|---|
| Randomization | The trial allocation is reported as RANDOMIZED. |
| Parallel design | The registry identifies PARALLEL as the design model. |
| Prevention trial | Primary purpose is PREVENTION. |
| Binary endpoints | Diabetes and total cancer are registered as binary endpoint types in the ClinicalTrials.gov record. |
| Prevalence endpoint | Aggregate microvascular complication is represented as an "Other / unclear" endpoint with a prevalence definition. |
| Composite endpoint | MACE combines myocardial infarction, stroke, or cardiovascular death. |
| Time-to-event analysis | Mortality is explicitly identified as a time-to-event endpoint. |
| Log-rank test | Used for both registry-reported mortality comparisons. |
| Hazard ratio | Reported as the mortality effect measure. |
| Confidence interval | Both mortality HRs include two-sided 95% confidence intervals. |
| Long-term follow-up | Mortality outcomes were assessed throughout follow-up from 1996 to 2019. |
| Population attrition | Not all participants randomized in the original DPP enrolled in DPPOS. |
| Assessment subsets | Some follow-up assessments were performed only in participants completing specific evaluations. |
15. Related Tutorials
Learn more about the methods used in this trial:
16. Related Calculators
17. Sources
- ClinicalTrials.gov: DPPOS, NCT00038727.
- Linked publication: PubMed 19878986.
- Linked publication: PubMed 26377054.
- Linked publication: PubMed 42295772.
- Linked publication: PubMed 41413398.
- Linked publication: PubMed 40675492.
Continue through the Clinical Biostats tutorials and calculators
Explore the statistical methods that underlie randomized prevention studies, survival analysis, confidence intervals, and time-to-event comparisons.
18. Record Summary
DPPOS is a randomized, parallel phase 3 prevention study with 2779 participants enrolled in the follow-up study and 4 registered primary endpoints addressing diabetes, aggregate microvascular complications, total cancer except non-melanoma skin cancer, and MACE. The registry-reported formal statistical analyses concern the secondary time-to-event endpoint of mortality, analyzed with the log-rank test and reported using hazard ratios.
For Original Lifestyle versus Original Placebo, the reported mortality HR was 1.02 with a two-sided 95% CI of 0.81–1.28 and P = 0.87. For Original Metformin versus Original Placebo, the reported HR was 0.99 with a two-sided 95% CI of 0.79–1.25 and P = 0.95. These estimates should be interpreted as relative time-to-event measures, not as absolute mortality differences.
The broader statistical lesson is that long-term clinical-trial analysis requires attention to the exact endpoint definition, analysis population, follow-up structure, censoring, and statistical method. DPPOS illustrates this particularly clearly because its registered endpoints span binary outcomes, prevalence, composite cardiovascular events, and time-to-event mortality, while the ClinicalTrials.gov record provides formal hazard-ratio analyses only for the latter.