Communicating Multiple Disease Risks: A Translation of Risk Prediction Science: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Communicating Multiple Disease Risks: A Translation of Risk Prediction Science (NCT03255291), a randomized na study. The primary outcome is Change From Baseline to 90-day Follow-up in Self Reported Weekly Minutes of Exercise.

2. Clinical Question

3. Trial Design

NCT03255291, titled “Communicating Multiple Disease Risks: A Translation of Risk Prediction Science,” is recorded as a completed randomized study with a factorial model and single masking. The registry lists the phase as NA and enrollment as 554. The sponsor was Washington University School of Medicine. The recorded study start was 2017-06-27, and primary completion was 2019-01-03.

The study’s stated purpose was to translate epidemiological data about five diseases into visual displays of individualized risk. Those diseases were colon cancer, breast cancer (women), heart disease, diabetes, and stroke. The broader aim described in the record was to communicate how one behavior, physical activity, could relate to multiple disease risks. The listed condition was healthy volunteers.

The factorial assignment crossed two design features: risk display format and imagery behavior. Risk display formats were a risk ladder, table, and text; imagery behavior was labeled exercise or sleep. This produced six posted groups. The reference group for the recalculated comparisons is risk ladder with exercise imagery. The descriptions say that participants completed the Risk Assessment App with research-assistant help and took a baseline survey. Those descriptions state that the app presented personalized risk estimates for current activity and how estimates would change with regular exercise.

Other listed study activities included audio recordings, surveys, and text message reminders. The registry does not report the precise allocation procedure, masking details beyond “single,” or the identity of the masked party. It also does not describe the trial’s original sample-size justification or provide enough detail to infer how the factorial components were intended to be analyzed. Randomization and factorial labels identify design features, but do not alone establish that all outcome comparisons were prespecified or adjusted for multiple testing.

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: Additional participants who did not complete the 90 day follow up survey or completed the survey over 8 weeks late, were inattentive responders, had text messaging errors, or were considered an outlier (\>99th percentile) in change in minutes of exercise per week were excluded for analysis

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 to 90-day Follow-up in Self Reported Weekly Minutes of Exercise, measured over Baseline and up to 90 days.

The primary endpoint is change from baseline to follow-up in self-reported weekly minutes of exercise, with follow-up described as up to 90 days. It is a continuous measure: each participant’s reported exercise time can take a range of values rather than being only a yes-or-no outcome. The registry identifies the reported parameter as a least-squares mean and the dispersion as a standard error. The posted arm summaries list means and standard errors for the groups.

Exercise minutes provide a direct measure of reported behavior relevant to the study’s focus on communicating activity-related risk. Comparing change over time allows the analysis to address whether the groups differed in reported weekly activity change, rather than only comparing a single follow-up measure. Self-report can capture behavior that may not be available through routine measurement, but it is also subject to recall and reporting error. The record does not report an objective activity measure as the primary endpoint.

The endpoint’s interpretation depends on the analysis population. The registry says that additional participants were excluded if they lacked the 90-day follow-up survey, completed it over 8 weeks late, were inattentive responders, had text messaging errors, or were considered outliers, defined as above the 99th percentile in change in weekly exercise minutes. The numbers excluded for these reasons and the exact handling of other incomplete observations are not reported in the registry record.

7. Results Posted in the Registry

• Risk Display Format:Risk Ladder:Imagery Behavior:Sleep: mean 37.50 versus 92.20 on control; difference -54.70 (95% CI -90.89 to -18.51), standardized effect d = -0.51; z-test p = 0.003.
• Risk Display Format:Table:Imagery Behavior:Exercise: mean 63.10 versus 92.20 on control; difference -29.10 (95% CI -63.89 to 5.69), standardized effect d = -0.28; z-test p = 0.101.
• Risk Display Format:Table: Imagery Behavior:Sleep: mean 60.00 versus 92.20 on control; difference -32.20 (95% CI -67.68 to 3.28), standardized effect d = -0.30; z-test p = 0.075.
• Risk Display Format:Text:Imagery Behavior:Exercise: mean 50.10 versus 92.20 on control; difference -42.10 (95% CI -77.03 to -7.17), standardized effect d = -0.40; z-test p = 0.018.

8. Statistical Interpretation

9. Statistical Methods Explained

The primary outcome is continuous, so the group summaries describe average change in weekly exercise minutes. A mean is useful for comparing groups on a common scale, but it can be influenced by unusually large or small changes. The registry says that observations above the 99th percentile in change were excluded as outliers for the primary analysis. That choice affects which data contribute to the estimate; the record does not give the number of excluded observations or a sensitivity analysis that retains them.

What does a least-squares mean represent?

In general, a least-squares mean is a model-based estimate of a group mean, potentially adjusted for other terms in the fitted model. It is not necessarily the same as a raw arithmetic average. The registry labels the primary parameter as a least-squares mean, but the reported group summaries do not identify the model terms used to obtain those means. The precise adjustment covariates, model specification, and treatment of factorial effects are not reported in the registry record. Accordingly, the recalculated comparisons based on posted group summaries are not equivalent to a full ANCOVA model.

What do the standard errors describe?

In general, a standard error describes the uncertainty of an estimated mean due to sampling variability; it is not a measure of how much individual participants vary. The posted arm summaries give standard errors, rather than standard deviations. The recalculations estimate pooled standard deviations from those standard errors and group sizes, under an approximation. The recalculated standard deviations are therefore approximate, not directly reported measures of participant-level spread. Any standardized effect sizes based on them inherit that approximation.

What is an unadjusted mean difference?

For each comparison, the recalculation subtracts the reference-group mean from the other group mean. The difference is expressed in minutes, preserving the endpoint’s original scale. The calculation is unadjusted: it does not incorporate the ANCOVA model described in the registry’s own analysis. Its confidence interval and p-value are consequently descriptive of this simple two-group calculation, rather than a reproduction of the sponsor’s full analysis. The registry’s own analysis takes precedence where results differ.

How should confidence intervals and p-values be read?

In general, a confidence interval gives a range of values compatible with the observed data under the assumptions of the analysis. Its width conveys precision: a narrower interval is more precise than a wider interval. For a mean difference, an interval that includes zero does not rule out no difference under that calculation. A p-value describes how unusual data at least as incompatible with a null hypothesis would be if that null hypothesis and the analysis assumptions held. It does not state the probability that the null hypothesis is true, and it does not measure clinical importance.

The recalculated p-values are unadjusted. The record lists multiple outcome measures and analyses, but does not report whether the sponsor adjusted for multiple comparisons or how the primary comparison was distinguished from other analyses. Thus, the displayed unadjusted p-values should not be treated as a familywise conclusion across all comparisons. The registry’s reported ANCOVA p-values are separate sponsor analyses; their associated estimates and intervals are not reported in the record.

What does factorial assignment add?

In general, a factorial design can assess more than one intervention feature within a study. Here the posted combinations cross display format with imagery behavior. A comparison between two groups can differ in one or both assigned features. For example, the recalculated reference comparisons change display format, imagery behavior, or both depending on the group. They are not a single isolated estimate of the effect of display format across imagery conditions, or of imagery across formats. The record does not provide a reported interaction estimate or a complete factorial-effects model, so the independent and combined contributions of the two features cannot be inferred from the listed group contrasts alone.

ANCOVA generally compares outcomes while accounting for other variables included in a model. The registry identifies ANCOVA for its reported analyses but does not state the covariates, coding, assumptions, or exact contrast definitions. It is therefore possible to describe the method named in the record, but not to reconstruct its adjusted effects from the posted summaries. The unadjusted calculations are useful as a transparent comparison of group summaries, not as a substitute for the sponsor’s model.

10. Confidence Intervals and Precision

The recalculated confidence intervals show substantial uncertainty around the estimated mean differences. The risk-ladder/sleep-imagery comparison has an estimate of -54.7 minutes and an interval from -90.89 to -18.51. This interval excludes zero under the unadjusted calculation, but still spans a range of possible differences. The intervals for the table comparisons extend from negative values to positive values: -63.89 to 5.689 for table/exercise imagery and -67.68 to 3.28 for table/sleep imagery. Those data are compatible with a range that includes no difference.

The text comparisons have intervals from -77.03 to -7.174 for text/exercise imagery and from -78.52 to -7.28 for text/sleep imagery. Each excludes zero in the unadjusted recalculation. Exclusion of zero is not equivalent to precise estimation: these intervals still allow effects of meaningfully different sizes. Precision should be judged by the interval’s range and the quality of the analysis, not by the p-value alone.

These intervals come from simple recalculations using posted summaries. The group-level standard errors are used to approximate variability, and the recalculations do not reproduce the ANCOVA analyses. The registry does not report confidence intervals for its own ANCOVA results, so their precision cannot be assessed from the p-values alone. The reported p-values indicate evidence against the relevant null hypothesis under the analysis, but they do not supply the missing estimate or show how concentrated that estimate is.

More generally, confidence intervals are conditional on the statistical model and data included. Here, the primary analysis excluded specified participants and outliers, while counts by exclusion reason are not reported. The record also does not describe a multiplicity adjustment. Both the analyzed population and the number of comparisons matter when interpreting interval and p-value evidence. The recalculated intervals should therefore be read as approximate, unadjusted summaries, not as definitive adjusted uncertainty estimates for the factorial study.

11. 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 express participants per arm, not total events. They use the observed recalculated effect as a planning reference and show alternatives for detecting 100%, 75%, or 50% of that observed effect. For example, the row for planning for 75% of the observed effect means using a smaller target effect than the full observed difference; a smaller effect generally requires more participants to detect under the same assumptions.

The listed planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. Each illustration is based on the stated fraction of the observed effect. The original trial’s sample-size justification and design assumptions are not reported in the registry record. These calculations are planning illustrations, not observed power.

The 80% power column gives the estimated participants per arm for a future study to have an 80% chance of detecting the specified effect under the planning assumptions, if that effect and assumptions are correct. The 90% power column gives the corresponding number for a 90% chance. These are design targets, not probabilities that the observed results are correct, and not evidence that the current study had a particular probability of finding an effect.

The illustrations are comparison-specific because each observed difference has a different size. They should not be read as a single recommended enrollment target for the full factorial design. The calculations assume a two-arm comparison and do not incorporate the trial’s full factorial structure, the sponsor’s ANCOVA model, exclusions, or any multiplicity adjustment. Their purpose is to show how planning requirements change with the effect size used, rather than to reconstruct the trial’s original design.

12. 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 = -54.700; pooled standard deviation σ = 107.528.
2. n = 2 × (1.96 + 0.842)² × 107.528² / -54.700² = 60.7, rounded up to 61 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 <- -54.7                      # observed difference in means
sigma <- 107.527958877208                     # 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(3255291)
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 = 61 (80% power), 82 (90% power)
 75% of observed effect: n per arm = 108 (80% power), 145 (90% power)
 50% of observed effect: n per arm = 243 (80% power), 325 (90% power)
power.t.test: 61.6 
Simulated power at n = 61 per arm: 0.799 (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.

13. Analyses Reported by the Sponsor

The registry’s own analyses are reported as ANCOVA. They give p = 0.029 for the listed risk-ladder, table, and text groups with sleep imagery; p = 0.333 for the two text groups; and p = 0.021 for the listed text/sleep group. The record does not report the corresponding adjusted estimates or confidence intervals, so the direction and magnitude of those ANCOVA effects cannot be fully assessed from the posted results.

The recalculated contrasts are simpler, unadjusted comparisons of posted group means against the risk-ladder/exercise-imagery reference group. They are not the same contrasts as all of the sponsor’s analyses and do not reproduce ANCOVA. Their estimated differences, confidence intervals, and p-values may therefore differ. The reported analyses also do not include sufficient model detail to explain the discrepancy or to identify whether it arises from adjustment, contrast definitions, or another aspect of the analysis.

Because the registry record flags a discrepancy, the sponsor’s own analysis takes precedence over the recalculated results. The recalculations remain useful as transparent checks of the posted summaries, but they should not be treated as corrected sponsor results or as replacements for the registry’s stated analysis. Without the adjusted estimates, confidence intervals, model details, and full analysis population information, an independent reconstruction of the sponsor’s conclusion is not possible from the registry record.

14. Multiplicity, Missing Data and Analysis Populations

The record lists 12 outcome measures, 8 posted secondary outcomes, and 14 analyses. It reports several group comparisons and multiple outcome domains, including self-efficacy, attitudes toward exercise, imagery vividness, and planning. When many outcomes or contrasts are examined, the chance of at least one apparently notable result can increase. In general, a multiplicity strategy can address this issue by adjusting error rates or by clearly separating primary, secondary, and exploratory questions.

Whether the sponsor used a multiplicity adjustment, which comparisons were prespecified, and how the primary analysis was distinguished from other analyses are not reported in the registry record. The recalculated p-values are unadjusted. They should not be interpreted as though they were corrected for the full set of analyses. The record’s ANCOVA p-values are reported without effect estimates or confidence intervals, and the statistical plan needed to understand their role is not reported in the registry record.

The primary outcome description states that participants were excluded for missing or late follow-up, inattentive responses, text messaging errors, or outlying change in exercise minutes. The record does not provide counts by reason, describe imputation, explain whether missingness was assessed, or report sensitivity analyses addressing these exclusions. It also does not state whether all randomized participants were included in the primary analysis. The stated analysis population is therefore narrower than enrollment, but the extent and possible consequences of that restriction cannot be quantified from the record.

In general, missing-data methods matter because results can depend on whether unavailable outcomes are related to participants’ observed or unobserved outcomes. Excluding participants with unavailable follow-up can affect comparability if the reasons for missingness differ across groups. The registry does not report enough detail to determine whether this occurred. It also does not provide group-specific analyzed counts alongside the outcome summaries in a way that resolves the exclusions described for the primary endpoint.

The record reports zero serious adverse events, other adverse events, and deaths in the listed groups. This is a description of the registry entries, not a substitute for information about completeness of event ascertainment or follow-up. No further adverse-event detail is reported in the registry record.

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

The posted results describe reported changes in weekly exercise minutes among participants included in the primary analysis. Group means varied across the six combinations of risk-display format and imagery behavior. The recalculated contrasts quantify how those group means compare with the risk-ladder/exercise-imagery reference group, with confidence intervals indicating uncertainty around each contrast. They do not describe the change in disease risk or incidence.

The study’s rationale concerns communicating multiple disease risks in a way that may help people understand how physical activity relates to several health outcomes. However, the primary endpoint is self-reported exercise minutes, not comprehension of the risk display, disease prevention, or a clinical outcome. Although the registry lists secondary measures related to self-efficacy, attitudes, imagery, and planning, the posted information in this record does not provide their results. The intended public-health rationale should not be confused with evidence that any particular display changes long-term behavior or health.

A randomized factorial study can support comparisons between assigned groups, but interpretation still depends on the analysis used, the population analyzed, adherence and measurement details, and the handling of multiple comparisons. The registry’s own analyses use ANCOVA, while the straightforward recalculations are unadjusted and do not match every sponsor contrast. With the discrepancy flagged and the sponsor’s adjusted estimates absent, the recalculations should be treated as limited summaries rather than definitive treatment-effect estimates.

The outcome is self-reported and the registry describes exclusions that include missing or late follow-up and outlying changes. The numbers excluded and the sensitivity of the results to those decisions are not reported. The record also does not report a multiplicity strategy. These limitations make it difficult to assess how robust the findings are across plausible analysis choices.

Accordingly, the results show differences in posted group means and varying degrees of statistical uncertainty in unadjusted comparisons. They do not establish that the intervention improved exercise behavior, that one format is superior in a general population, or that the observed differences translate into better disease outcomes. Those broader claims would require evidence beyond the posted primary outcome summaries.

16. Why This Trial Matters Statistically

This trial raises a statistical question that is common in behavioral and communication research: how should outcomes be analyzed when an intervention combines multiple design features? The six groups reflect the crossing of display format and imagery behavior. A comparison between selected groups may change one feature, both features, or the combination, so a set of pairwise differences is not automatically an estimate of each feature’s independent contribution.

The distinction between the registry’s ANCOVA analyses and unadjusted recalculations also matters. A model-based adjusted result and a simple contrast of group means answer related but not necessarily identical questions. Without the model specification and adjusted estimates, readers cannot determine why the results differ or judge the adjusted effect’s size and precision. Reporting estimates and confidence intervals alongside p-values would make that assessment more informative.

The record also illustrates why analysis populations and exclusions should be described in detail. The primary endpoint excludes participants for several follow-up, response-quality, messaging, and outlier criteria. Knowing how many participants were affected, how exclusions were distributed across groups, and how alternative handling changes the estimates would help readers evaluate robustness. Those details are not reported in the registry record.

Finally, the study’s motivating question concerns communication of multiple disease risks, while its primary outcome is self-reported exercise change over follow-up up to 90 days. Statistical interpretation should keep the measured endpoint distinct from the broader public-health rationale. This distinction helps prevent estimates about reported behavior from being overstated as evidence about disease outcomes or durable behavior change.

17. 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 describes a specialized communication intervention focused on personalized risk estimates and physical activity. Its population is identified as healthy volunteers, and the record does not report demographic characteristics that would clarify how broadly the findings might apply. Enrollment is recorded as 554, but the primary outcome analysis excluded participants for several reasons. Counts by reason and group are not reported, limiting assessment of selection effects.

The primary endpoint is self-reported weekly exercise minutes. Self-report may be affected by recall or reporting behavior, and the record does not report an objective activity measure to corroborate it. Follow-up was described as up to 90 days, but the timing distribution and handling of surveys completed over 8 weeks late are not detailed beyond their exclusion from the primary analysis. The degree to which the reported outcome reflects sustained behavior is therefore not established by this endpoint.

The intervention descriptions say that the app presented changes associated with regular exercise, while the assigned imagery-behavior labels include both exercise and sleep. The registry record does not fully clarify the content or implementation of each imagery condition, adherence to assigned activities, or exposure to each display. The precise masking procedure and allocation details are also not reported in the registry record.

The sponsor’s ANCOVA estimates and confidence intervals are absent, and the model specification is not reported. The unadjusted recalculations use approximate variability derived from posted standard errors and do not reproduce the sponsor’s model. The record also does not report a multiplicity plan, missing-data sensitivity analyses, or the original sample-size justification. These gaps restrict independent evaluation of precision, robustness, and the relationship between planned and analyzed comparisons.

18. Related Tutorials

19. Related Statistical Calculators

20. Sources

• ClinicalTrials.gov: Communicating Multiple Disease Risks: A Translation of Risk Prediction Science — NCT03255291.
• Registry record last updated 2020-01-18; status last verified 2020-01.
• Results first posted 2020-01-18.
• 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.

21. Record Summary

NCT03255291 is a completed randomized factorial study sponsored by Washington University School of Medicine. It enrolled 554 healthy volunteers and tested combinations of risk-display format and imagery behavior in a study about communicating individualized risks for multiple diseases. The primary endpoint was change from baseline to follow-up up to 90 days in self-reported weekly minutes of exercise.

The posted group means ranged from 37.5 to 92.2 minutes. Simple unadjusted recalculations compare each non-reference group with the risk-ladder/exercise-imagery group. The registry also reports ANCOVA p-values, but not the corresponding adjusted estimates or confidence intervals. Because the record flags a discrepancy, the sponsor’s own analysis takes precedence over the recalculations.

Interpretation is limited by the outcome’s self-reported nature, stated exclusions from the primary analysis, and missing detail on multiplicity, missing-data handling, and model specification. The registry results address reported exercise change, not disease incidence or clinical outcomes.