This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
STRIDE was a randomized, double-masked, parallel-group phase 3 trial evaluating evidence-based tailored fall prevention against usual care in a prevention setting involving accidental falls and wounds and injuries.
| Feature | STRIDE |
|---|---|
| Phase | Phase 3 |
| Status | Completed |
| Study period | August 2015 through December 2019 |
| Lead sponsor | Brigham and Women's Hospital |
| Conditions | Accidental Falls; Wounds and Injuries |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Prevention |
| Enrollment | 5,451 |
| Primary endpoint type | Count / rate |
| Results posted | Yes |
| Statistical analyses posted | 7 |
2. Clinical Question
The central statistical question was whether an evidence-based tailored fall-prevention strategy changed the rate of first adjudicated serious fall-related injury compared with usual care during follow-up.
Population
Participants enrolled in the STRIDE phase 3 prevention trial for accidental falls and wounds and injuries.
Intervention
Evidence-based tailored fall prevention, described in the registry as the Fall Prevention Standard of Care intervention.
Comparator
Usual care, described as the control condition.
Primary question
Does the fall-prevention strategy alter the hazard of a first adjudicated serious fall-related injury relative to usual care?
3. Trial Design
Fall Prevention Standard of Care
- Evidence-based tailored fall prevention.
- Primary comparison against the control condition.
- Primary efficacy analysis used a Cox proportional-hazards model.
Usual Care
- Usual care was the randomized comparator.
- Primary and secondary analyses compared this group with the intervention group.
4. Endpoints
| Endpoint | Time frame | Type | Statistical approach |
|---|---|---|---|
| First Adjudicated Serious Fall-Related Injury | Enrollment through last completed follow-up or death interview (max 44 months) | Time-to-event; events per 100 person years of follow-up | Cox proportional-hazards model |
| First Self-reported Fall-related Injury | Enrollment through last completed follow-up or death interview (max 44 months) | Time-to-event; events per 100 person years of follow-up | Cox proportional-hazards model |
| Physical Function | 12-month and 24-month follow-up assessments; reported score is an average of follow-up scores | Continuous | Mixed-effects model |
| Disability | 12-month and 24-month follow-up assessments; reported score is an average of follow-up scores | Continuous | Mixed-effects model |
| Anxiety | 12-month and 24-month follow-up assessments; reported score is an average of follow-up scores | Continuous | Mixed-effects model |
| Depression | 12-month and 24-month follow-up assessments; reported score is an average of follow-up scores | Continuous | Mixed-effects model |
| Fear of Falling | 12-month and 24-month follow-up assessments; reported score is an average of follow-up scores | Continuous | Mixed-effects model |
5. Statistical Methodology
Primary time-to-event analysis
The primary endpoint was analyzed using a Cox proportional-hazards model. The reported effect measure was the hazard ratio comparing the Fall Prevention Standard of Care intervention with control.
The registry additionally notes that the primary analysis used a multistate model accounting for death as a semi-competing risk. This is important because death can prevent a participant from subsequently experiencing the injury endpoint, creating a particular form of event dependence in which death is terminal while the injury event is not necessarily terminal.
A hazard ratio below 1 indicates a lower estimated instantaneous event hazard in the intervention group relative to the control group, under the fitted model.
Longitudinal continuous-outcome analysis
Physical function, disability, anxiety, depression, and fear of falling were analyzed using mixed-effects models. The registry describes patients as nested within practices as random effects, while baseline score, randomization constraint variables, and predictors of missingness were included as fixed effects.
For the well-being outcomes, the reported effect was a difference in least-squares means across follow-up. The follow-up assessments occurred at 12 months and 24 months, and the reported score was an average of the follow-up scores.
| Component | Registry-described role |
|---|---|
| Random effect | Patients nested within practices |
| Fixed effect | Baseline score |
| Fixed effects | Randomization constraint variables |
| Fixed effects | Predictors of missingness |
| Follow-up assessments | 12 months and 24 months |
| Reported contrast | Difference in least-squares means across follow-up |
6. Primary Result: First Adjudicated Serious Fall-Related Injury
The registry reports a formal statistical analysis for the primary endpoint. The intervention was compared with control using a Cox proportional-hazards model, with the analysis notes specifying a multistate model accounting for death as a semi-competing risk.
Hazard ratio for first adjudicated serious fall-related injury
95% CI: 0.80–1.06 · P = 0.245
Two-sided 95% confidence interval; superiority hypothesis.
| Primary endpoint | Intervention vs control | 95% CI | P-value |
|---|---|---|---|
| First Adjudicated Serious Fall-Related Injury | HR 0.92 | 0.80–1.06 | 0.245 |
An HR of 0.92 corresponds to an estimated instantaneous hazard about 8% lower in the intervention group than in the control group, because 1 − 0.92 = 0.08. This is a relative model-based comparison, not an 8% absolute reduction in the number of participants experiencing an injury.
The 95% CI of 0.80–1.06 describes the statistical uncertainty around the estimated hazard ratio. Because the interval includes 1.00, the reported estimate is compatible with both a lower and a higher hazard under the model.
The P-value of 0.245 describes the evidence against the null hypothesis under the specified statistical framework; it does not measure the size or clinical importance of the effect. A P-value is therefore not interchangeable with the hazard ratio or its confidence interval.
The interpretation also depends on the time-to-event framework, censoring, and the assumptions underlying the Cox model. The registry's multistate approach is particularly relevant because death can prevent subsequent observation of the injury endpoint.
7. Secondary Time-to-Event Result: First Self-reported Fall-related Injury
The first self-reported fall-related injury was analyzed as a secondary time-to-event endpoint using a Cox proportional-hazards model.
Hazard ratio for first self-reported fall-related injury
99% CI: 0.83–0.99 · P = 0.004
Two-sided 99% confidence interval; superiority hypothesis.
| Secondary endpoint | Effect estimate | Confidence interval | P-value |
|---|---|---|---|
| First Self-reported Fall-related Injury | HR 0.90 | 99% CI 0.83–0.99 | 0.004 |
An HR of 0.90 corresponds to a 10% lower estimated instantaneous hazard in the intervention group relative to control under the fitted Cox model. It does not mean that 10% fewer participants necessarily experienced an injury, because a hazard ratio is a relative time-to-event measure rather than an absolute risk difference.
The 99% CI of 0.83–0.99 is relatively narrow compared with the primary endpoint's interval and lies below 1.00. The 99% confidence level is notable because it provides a more stringent interval than the conventional 95% level.
The P-value of 0.004 provides evidence against the null hypothesis within the reported analysis. It does not quantify the magnitude of the treatment effect, which is described by the hazard ratio and its confidence interval.
This was a secondary endpoint rather than the primary endpoint. Consequently, the result should be interpreted in the context of the trial's overall endpoint structure rather than treated as interchangeable with the primary analysis.
8. Secondary Longitudinal Results
The registry reports five continuous secondary outcomes assessed at 12-month and 24-month follow-up. The analyses used mixed-effects models and reported differences in least-squares means or least-squares mean contrasts. All confidence intervals were two-sided 99% intervals.
| Outcome | Estimate | 99% CI | P-value |
|---|---|---|---|
| Physical Function | 0.72 | -1.10 to 2.54 | 0.309 |
| Disability | 0.59 | -1.80 to 0.93 | 0.528 |
| Anxiety | -0.90 | -2.00 to 0.20 | 0.037 |
| Depression | -1.19 | -2.36 to -0.02 | 0.009 |
| Fear of Falling | -0.07 | -1.38 to 1.25 | 0.897 |
Physical Function
Difference in least-squares means
99% CI: -1.10 to 2.54 · P = 0.309
The estimated difference in physical function was 0.72 units in the reported scale, with a 99% CI from -1.10 to 2.54. Because the interval crosses zero, the data are compatible with both a small negative and a positive mean difference under the model.
The P-value of 0.309 is a measure of statistical evidence against the null hypothesis, not a measure of the size of the difference. The magnitude of the estimated difference and the width of its confidence interval provide the more direct description of the effect and its precision.
Disability
Difference in least-squares means
99% CI: -1.80 to 0.93 · P = 0.528
The estimated difference was 0.59 units, with a 99% CI from -1.80 to 0.93. The interval includes zero, so the reported estimate is not precise enough to distinguish the direction of a small mean difference within the model's uncertainty.
The P-value of 0.528 should not be interpreted as the probability that the treatment has no effect. It is instead a measure of compatibility between the observed data and the specified null hypothesis under the analysis framework.
Anxiety
Difference in least-squares means
99% CI: -2.00 to 0.20 · P = 0.037
The reported mean difference was -0.90 units. The direction of a negative value indicates that the estimated follow-up score was lower in the intervention group under the reported contrast. The 99% CI of -2.00 to 0.20 includes zero, so the interval does not exclude no difference at the 99% confidence level.
The P-value of 0.037 should be read together with the 99% interval rather than in isolation. In particular, the P-value and confidence interval use different thresholds here: the registry reports a 0.037 P-value alongside a 99% CI that includes zero.
Depression
Difference in least-squares means
99% CI: -2.36 to -0.02 · P = 0.009
The estimated difference in depression was -1.19 units, with a 99% CI from -2.36 to -0.02. The interval lies below zero, indicating that the reported model-based contrast was negative throughout the 99% confidence interval.
The P-value of 0.009 provides statistical evidence against the corresponding null hypothesis under the reported model. The size of the estimated difference remains distinct from the P-value, and the clinical meaning of a one-unit change depends on the scale used for the outcome.
Fear of Falling
Difference in least-squares means
99% CI: -1.38 to 1.25 · P = 0.897
The estimated difference was -0.07 units, very close to zero on the reported scale. The 99% CI spans from -1.38 to 1.25, indicating substantial uncertainty around the point estimate.
The P-value of 0.897 does not establish that the two groups are identical. It indicates that the observed data provide little evidence against the null hypothesis under the specified model; the confidence interval is the more informative description of the range of compatible effects.
9. Analysis Populations and Longitudinal Modeling
The registry specifies that the physical function, disability, anxiety, depression, and fear-of-falling analyses were conducted in subsamples selected for well-being measures and with the relevant completed baseline measure.
| Outcome | Analysis population |
|---|---|
| Physical Function | Subsample selected for well-being measures with completed baseline LL-FDI |
| Disability | Subsample selected for well-being measures with completed baseline LL-FDI |
| Anxiety | Subsample selected for well-being measures and with completed baseline PROMIS Anxiety |
| Depression | Subsample selected for well-being measures and with completed baseline PROMIS depression measure |
| Fear of Falling | Subsample selected for well-being measures and with completed baseline FES |
This population definition matters statistically. A longitudinal estimate based on a specified well-being subsample is not automatically interchangeable with an estimate from the entire randomized cohort. The analysis population determines the participants contributing information to the model and therefore the population to which that particular estimate directly applies.
10. Statistical Methods Explained
Why was a Cox proportional-hazards model used for the injury endpoints?
The first adjudicated serious fall-related injury and first self-reported fall-related injury are time-to-event outcomes. A Cox model allows the analysis to use both the occurrence of an event and the amount of follow-up contributed by each participant, while accommodating right-censoring. The resulting hazard ratio summarizes the relative instantaneous event hazard between the randomized groups under the model.
What does a hazard ratio of 0.92 mean?
An HR of 0.92 means that the intervention group's estimated instantaneous event hazard was 0.92 times that of the control group under the fitted model. Equivalently, the point estimate corresponds to an 8% lower estimated hazard because 1 − 0.92 = 0.08. It does not mean an 8% reduction in absolute event probability or an 8% reduction for every individual participant.
Why is the confidence interval important?
The confidence interval communicates the statistical precision of the point estimate. For the primary endpoint, the HR was 0.92 with a 95% CI of 0.80–1.06. The interval therefore communicates considerably more information than the point estimate alone: the estimated effect is uncertain, and the interval includes the null value of 1.00.
Why was a mixed-effects model used for the well-being outcomes?
The outcomes were measured repeatedly at 12-month and 24-month follow-up, and patients were nested within practices. A mixed-effects model can represent this hierarchical and longitudinal structure by allowing patient/practice-related dependence to enter the model rather than treating every observation as statistically independent.
What does a mean difference of -1.19 mean for depression?
The reported least-squares mean difference was -1.19 units for the depression outcome. The negative sign identifies the direction of the modeled contrast. The estimate should be interpreted on the specific depression scale used in the trial; without a clinically established threshold in the registry information, the statistical magnitude should not be translated into a clinical severity category.
Why does the primary analysis account for death as a semi-competing risk?
Death can occur before a participant experiences the injury endpoint and can therefore prevent subsequent observation of that injury. In a semi-competing-risk setting, the terminal event changes the opportunity to observe the nonterminal event. The registry's multistate analysis explicitly accounts for this structure rather than treating death as an ordinary censoring event without qualification.
11. Understanding the Primary Hazard Ratio
The point estimate is below 1, but the confidence interval includes 1.00.
The point estimate of 0.92 represents an estimated 8% lower instantaneous hazard of first adjudicated serious fall-related injury for the intervention relative to control.
The HR does not directly provide the percentage of participants who experienced an injury, the absolute difference in injury risk, the number needed to treat, or the probability that an individual participant benefited.
The 95% CI of 0.80–1.06 expresses uncertainty around the estimated hazard ratio. The interval includes 1.00, so the reported estimate does not exclude a null hazard ratio within this confidence framework.
The P-value of 0.245 is evidence against the null hypothesis under the reported test. It is not an effect-size measure and should not be used as a substitute for the HR or confidence interval.
Because the primary analysis uses a Cox model, interpretation depends on the model's time-to-event assumptions. The additional multistate treatment of death is also important because death changes the probability of subsequently observing a serious fall-related injury.
12. Confidence Intervals and P-values Across the Outcomes
The registry contains both 95% and 99% confidence intervals. The primary endpoint uses a two-sided 95% CI, while the reported secondary analyses use two-sided 99% CIs. These confidence levels should not be silently treated as equivalent.
| Outcome | Effect | CI level | Interval | P-value |
|---|---|---|---|---|
| First Adjudicated Serious Fall-Related Injury | HR 0.92 | 95% | 0.80–1.06 | 0.245 |
| First Self-reported Fall-related Injury | HR 0.90 | 99% | 0.83–0.99 | 0.004 |
| Physical Function | Mean difference 0.72 | 99% | -1.10–2.54 | 0.309 |
| Disability | Mean difference 0.59 | 99% | -1.80–0.93 | 0.528 |
| Anxiety | Mean difference -0.90 | 99% | -2.00–0.20 | 0.037 |
| Depression | Mean difference -1.19 | 99% | -2.36–-0.02 | 0.009 |
| Fear of Falling | Mean difference -0.07 | 99% | -1.38–1.25 | 0.897 |
For hazard ratios, the null value is 1. For mean differences, the null value is 0. That distinction is essential when reading the confidence intervals: an interval containing 1 for a hazard ratio or 0 for a mean difference includes the corresponding null effect.
13. Safety: Serious Adverse Events
The registry reports serious adverse events by randomized arm as the number affected divided by the number at risk.
| Arm | Participants with serious adverse events | At risk |
|---|---|---|
| Intervention - Fall Prevention Standard | 1,139 | 2,802 |
| Control | 1,108 | 2,649 |
The serious-adverse-event counts describe the number of affected participants and the corresponding denominators. They are not themselves a treatment-effect estimate, and the registry information does not provide a formal comparative statistical analysis for this safety measure.
14. Primary and Secondary Results in Context
| Outcome family | What the registry reports | Statistical interpretation |
|---|---|---|
| Primary serious fall-related injury | HR 0.92; 95% CI 0.80–1.06; P = 0.245 | The estimated hazard was below 1, but the confidence interval includes the null value. |
| Self-reported fall-related injury | HR 0.90; 99% CI 0.83–0.99; P = 0.004 | The reported point estimate corresponds to a 10% lower estimated hazard, with a 99% interval below 1. |
| Physical function | Mean difference 0.72; 99% CI -1.10 to 2.54; P = 0.309 | The interval includes zero and spans both directions of a possible mean difference. |
| Disability | Mean difference 0.59; 99% CI -1.80 to 0.93; P = 0.528 | The interval includes zero and indicates uncertainty in the direction of the difference. |
| Anxiety | Mean difference -0.90; 99% CI -2.00 to 0.20; P = 0.037 | The point estimate is negative, but the 99% interval includes zero. |
| Depression | Mean difference -1.19; 99% CI -2.36 to -0.02; P = 0.009 | The estimated difference is negative and the 99% interval lies below zero. |
| Fear of falling | Mean difference -0.07; 99% CI -1.38 to 1.25; P = 0.897 | The estimate is close to zero and the interval spans both negative and positive values. |
These outcomes are not interchangeable. The primary endpoint is a time-to-event injury endpoint, while the well-being outcomes are longitudinal continuous measures. Their effect estimates therefore live on different statistical scales and should be interpreted according to their respective models.
15. Limitations
- Primary endpoint uncertainty: the primary hazard-ratio estimate was 0.92, with a 95% CI of 0.80–1.06. The interval includes the null value of 1.00.
- Time-to-event assumptions: Cox-model interpretation depends on the model's assumptions about the relationship of hazards over time. A single hazard ratio is a summary measure rather than a complete description of the event process.
- Death as a competing event structure: death can prevent later observation of serious fall-related injury. The registry addresses this through a multistate model accounting for death as a semi-competing risk.
- Different confidence levels: the primary analysis reports a 95% confidence interval, whereas the posted secondary analyses use 99% confidence intervals.
- Secondary analysis populations: the continuous well-being outcomes were evaluated in specified subsamples with completed baseline measures rather than being described simply as analyses of all enrolled participants.
- Longitudinal dependence: repeated follow-up observations and patient clustering within practices require appropriate covariance and random-effects structures; the registry specifies a mixed-effects framework with patients nested within practices.
- Multiplicity: multiple secondary endpoints are reported. The individual P-values should therefore be interpreted in the context of the broader set of outcomes rather than as a collection of independent confirmatory tests.
- Clinical meaning of continuous outcomes: the registry reports changes in scale units but does not provide a clinical threshold for interpreting the magnitude of each reported mean difference.
16. Why This Trial Matters Statistically
STRIDE is a useful statistical teaching case because it combines two different analytical problems within the same randomized trial: time-to-event outcomes for fall-related injury and repeated continuous outcomes for physical and psychological well-being.
| Concept | How it appears in STRIDE |
|---|---|
| Randomization | Randomized allocation to two parallel study arms. |
| Blinding | The registry describes the trial as double masked. |
| Time-to-event endpoint | First adjudicated serious fall-related injury and first self-reported fall-related injury. |
| Cox proportional-hazards model | Used for the two reported time-to-event injury analyses. |
| Hazard ratio | Primary and secondary injury effects are reported as HRs. |
| Confidence intervals | Primary analysis uses a 95% CI; secondary analyses use 99% CIs. |
| Mixed-effects model | Used for physical function, disability, anxiety, depression, and fear of falling. |
| Repeated measures | Well-being outcomes were measured at 12-month and 24-month follow-up assessments. |
| Hierarchical data | Patients were nested within practices in the longitudinal models. |
| Missingness predictors | Predictors of missingness were included as fixed effects in the mixed-effects analyses. |
| Semi-competing risk | The primary analysis used a multistate model accounting for death as a semi-competing risk. |
17. Statistical Interpretation: Relative Effects vs Absolute Outcomes
Relative effect
The primary analysis reports an HR of 0.92. This expresses the relative relationship between the estimated event hazards under the Cox model.
Absolute event burden
The registry also describes the primary endpoint in events per 100 person years of follow-up. An absolute rate answers a different question from a hazard ratio.
Longitudinal effect
The well-being analyses report mean differences across 12-month and 24-month follow-up assessments rather than hazard ratios.
Uncertainty
Confidence intervals should accompany the point estimates because they describe the precision of the statistical estimates under the corresponding models.
One of the most important statistical lessons is that the primary endpoint's HR of 0.92 cannot be directly compared numerically with a mean difference such as -1.19 for depression. These are different estimands measured on different scales and derived from different statistical models.
18. Longitudinal Trial History
STRIDE begins
The registry records the study start as August 2015.
Follow-up period completed
The registry records primary completion as December 2019.
Results posted
The registry reports posted outcome measures and seven posted statistical analyses, including one formal analysis of the primary endpoint.
19. What the Statistical Results Do — and Do Not — Establish
The primary estimate of HR 0.92 indicates a lower point estimate for the intervention group's instantaneous hazard of first adjudicated serious fall-related injury, but the 95% CI of 0.80–1.06 includes the null value.
The first self-reported fall-related injury analysis reports HR 0.90 with a 99% CI of 0.83–0.99. This corresponds to a 10% lower estimated instantaneous hazard at the point estimate, while the confidence interval describes uncertainty around that estimate.
The longitudinal outcomes produce estimates on their respective measurement scales. The depression estimate of -1.19 and the anxiety estimate of -0.90, for example, should not be interpreted using the same framework as a hazard ratio.
A P-value addresses statistical evidence under a specified hypothesis test. It does not determine whether an effect is clinically important. Clinical interpretation requires knowledge of the outcome scale, meaningful thresholds, absolute event rates, and the context in which the intervention is used.
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Calculators
22. Sources
- ClinicalTrials.gov: NCT02475850.
- PubMed: PMID 38740298.
- PubMed: PMID 37533073.
- PubMed: PMID 35932279.
- PubMed: PMID 33678038.
- PubMed: PMID 33037632.
Continue through Clinical Biostats
Connect the statistical methods used in STRIDE with deeper tutorials and practical statistical tools.
23. Record Summary
STRIDE provides a useful example of how a randomized prevention trial can require different statistical frameworks for different outcomes. The primary endpoint, first adjudicated serious fall-related injury, was analyzed with a Cox proportional-hazards model within a multistate framework accounting for death as a semi-competing risk. Its reported HR was 0.92 with a 95% CI of 0.80–1.06 and P = 0.245. A secondary time-to-event endpoint, first self-reported fall-related injury, produced an HR of 0.90 with a 99% CI of 0.83–0.99 and P = 0.004.
The trial also illustrates the distinction between survival analysis and longitudinal continuous-outcome modeling. Physical function, disability, anxiety, depression, and fear of falling were analyzed using mixed-effects models that accounted for patients nested within practices and included baseline scores, randomization constraint variables, and predictors of missingness. Their reported effects therefore require interpretation on the relevant outcome scales rather than through the hazard-ratio framework.