Twitter and Diabetes: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Twitter and Diabetes (NCT02806700), a randomized na study. The primary outcome is Change in Patient Activation Measure (PAM) Score From Baseline.

2. Clinical Question

3. Trial Design

The registry describes a completed, randomized, parallel-group behavioral trial, with no masking. Enrollment was 628. Participants were assigned to the Twitter Diabetes Intervention or Twitter Diabetes Control group. The control group was contacted and asked to complete brief surveys; the intervention group was also asked to use Twitter for heart health. The registry does not report the method used to generate or conceal the random allocation sequence.

The primary outcome was change in Patient Activation Measure score from baseline to 6 months. The secondary outcome was change in systolic blood pressure over the same stated timeframe. The endpoint summaries are continuous outcomes reported as means and standard deviations. The primary outcome summary includes 202 participants in the control group and 129 in the intervention group. The secondary outcome summary includes 252 in the control group and 226 in the intervention group. The registry does not explain why the numbers represented in the outcome summaries differ across outcomes or from total enrollment.

The record lists a start date of 2016-06 and primary completion of 2018-05. It does not report a detailed protocol, eligibility criteria, recruitment flow, treatment adherence, or the timing and format of the Twitter activities. The original sample-size justification and design assumptions are not reported in the registry record. No analysis descriptions are posted, and the registry does not state which analysis population was intended. These omissions limit what can be concluded about the conduct and analysis beyond the design features and outcome summaries recorded.

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

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: Total number of participants who completed baseline survey and were randomized

6. Primary Endpoint

The registry lists the primary outcome as Change in Patient Activation Measure (PAM) Score From Baseline, measured over baseline and 6 months.

The primary endpoint is change in Patient Activation Measure score from baseline to 6 months. The registry describes PAM as a validated and reliable tool for assessing a person's ability to manage their health. Its total score ranges from 0 to 100, with higher scores indicating better outcomes. The measure also assigns four activation levels: Level 1 covers scores from 0 to 47.0; Level 2, 47.1 to 55.1; Level 3, 55.2 to 67.0; and Level 4, 67.1 to 100. The registry characterizes the levels as ranging from disengagement and limited confidence to confidence and proactive health management.

A change score focuses on how the measure differs from baseline at the stated follow-up, rather than only comparing scores at one time. It is relevant to this behavioral intervention because the trial's stated goal concerns patients' ability to manage their cardiovascular health condition. PAM captures activation and self-management, not every aspect of health or cardiovascular status. The registry does not report how individual item responses were handled, whether the level categories were used in the analysis, or the precise scoring rules applied to missing items.

7. Results Posted in the Registry

• Twitter Diabetes Intervention: mean 0.00 versus 3.40 on control; difference -3.40 (95% CI -7.70 to 0.90), standardized effect d = -0.18; z-test p = 0.121.

8. Secondary Efficacy Endpoints

The registry also posts results for secondary outcomes. For each outcome below, the endpoint type, the test applied to the posted summary statistics, the result and its interpretation are given in turn. Secondary endpoints are usually not adjusted for multiple testing, so they support rather than confirm the primary finding.

Change in Systolic Blood Pressure From Baseline

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard deviations.
• Result:
  • Twitter Diabetes Intervention: mean -18.50 versus -19.80 on control; difference 1.30 (95% CI -2.27 to 4.87), standardized effect d = 0.07; z-test p = 0.476.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Twitter Diabetes Intervention and the control. The interval gives the range of effects that remain compatible with the data.

The registry lists one secondary endpoint: change in systolic blood pressure from baseline to 6 months. Its posted group summaries and comparison do not indicate a clear between-group difference; the interval includes zero and the p-value is unadjusted. The secondary result is directionally and inferentially consistent with the broad pattern of an uncertain group comparison, but it measures a different aspect of health from PAM. It is not a second measurement of patient activation and does not independently reproduce the primary endpoint.

These outcomes can contribute complementary information, yet the registry does not report whether they were expected to move together or how their results were intended to be interpreted jointly. A change in activation and a change in systolic blood pressure are not interchangeable, and a result on one cannot by itself establish what happened on the other. The posted secondary result therefore adds a separate piece of evidence, rather than changing the meaning of the primary estimate.

Unadjusted secondary results generally carry less confirmatory weight when several outcomes or comparisons are examined. Each additional test creates another opportunity for a result to appear noteworthy by chance. Statistical methods can address that issue by controlling error across a prespecified family of tests, but the registry does not report a multiplicity strategy or any adjusted secondary analysis. The record also does not describe an outcome hierarchy beyond identifying the primary and secondary endpoints. The secondary finding should consequently be treated as descriptive and interpreted cautiously, especially if used to support a claim that was not the primary question.

The registry lists two outcome measures in total and one secondary outcome. It does not report additional secondary analyses. That limited record should not be mistaken for proof that no other analyses were conducted; it means that they are not reported in the registry record. The available secondary information is useful for understanding the breadth of outcomes recorded, but it does not resolve the uncertainty around the primary result.

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

• Change in Systolic Blood Pressure From Baseline

9. Statistical Interpretation

10. Statistical Methods Explained

The primary and secondary outcomes are continuous measures, and the registry reports each arm's mean and standard deviation. A mean describes the average observed change in that group. A standard deviation describes the spread of individual outcome changes around the group mean; it is not the uncertainty of the mean itself. Large variation among participants can make a group average less precise, particularly when fewer outcome measurements are available.

What does the mean difference represent?

The reported comparison is the mean difference between the intervention group and control group. The direction is stated as intervention compared with control. A value below zero means the intervention group's average change is lower than the control group's average change; a value above zero means it is higher. Interpretation also depends on what a higher score means for the outcome. For PAM, higher scores indicate better outcomes, while the endpoint is a change score. The comparison is an unadjusted summary: the registry does not report a model adjusting for baseline values or other participant characteristics.

For systolic blood pressure, the mean difference is expressed in mm Hg. The sign of a change score should be interpreted in relation to the measure and the way change is defined. The record states the outcome as change from baseline but does not provide additional detail about the calculation or analysis of that change. No further transformation or covariate adjustment is described.

How should the confidence interval be read?

A confidence interval expresses uncertainty around an estimated difference. It gives a range of values compatible with the estimate under the statistical method used. A narrower interval indicates greater precision than a wider interval, all else being equal. If an interval includes zero for a difference, the data are compatible with no group difference; that does not prove the groups are identical. If the interval spans values on both sides of zero, it also permits effects in either direction. The posted intervals for both outcomes include zero.

A p-value measures how unusual the observed result, or a more extreme result, would be under a statistical model in which the groups have no difference. It is not the probability that the null hypothesis is true, nor does it measure the size or importance of a difference. The registry labels the posted p-values as unadjusted. The statistical test, its assumptions, and whether any alternative model was used are not reported in the registry record. A p-value should be considered alongside the effect estimate, confidence interval, study design, and outcome completeness, rather than as a stand-alone verdict.

Why do outcome counts and spreads matter?

The number of participants contributing to an outcome affects how much information is available for its estimate. The primary outcome has different reported group counts from the secondary outcome, and each outcome summary represents fewer participants than the enrollment figure. These registry entries do not explain why measurements are absent or not included. The registry does not report the reasons, missingness patterns, or handling procedures. Standard deviations also convey participant-to-participant variability. Variation affects precision, but a standard deviation is not itself a confidence interval and does not indicate how representative the participants are.

Randomization supports a comparison between assigned groups, but does not automatically resolve missing outcome data, non-adherence, or differences in how outcomes were measured. The trial is recorded as having no masking. The registry does not describe whether outcome measurement was standardized, whether people analyzing the data knew group assignments, or whether the reported group summaries follow an intention-to-treat approach. Those details are important to interpretation but cannot be inferred from the arm summaries.

What is known about the analysis plan?

The record lists no reported analyses. It does not state whether the outcome comparisons were prespecified, whether the primary endpoint was analyzed using a particular model, how baseline scores entered the analysis, or what missing-data method was planned. It also does not report a multiplicity adjustment for the primary and secondary outcomes. The presence of a mean difference, confidence interval, and unadjusted p-value permits a basic reading of the posted comparison, but it does not fill in those methodological details.

The sample-size illustration is a separate planning exercise based on the observed effect, with a two-sided significance level of 0.05, equal allocation, and a normal approximation. It is not evidence about the trial's original design or the probability that this trial detected an effect. The original sample-size justification and assumptions are not reported in the registry record. General planning calculations help show how assumptions about the target effect and desired power influence enrollment needs; they do not replace a prespecified design or analysis plan.

11. Confidence Intervals and Precision

The primary mean difference is -3.4, and its confidence interval runs from -7.703 to 0.903. This interval includes zero. The estimate is therefore not highly precise about whether the intervention group's average change was lower or higher than the control group's: values on both sides of no difference remain compatible with the posted interval. Its width also signals appreciable uncertainty around the point estimate. The unadjusted p-value of 0.121 is consistent with the interval's inclusion of zero, but neither statistic determines whether the difference matters in practice.

The group standard deviations, 17.4 in control and 20.7 in the intervention group, indicate considerable variation in individual PAM changes relative to the group mean changes. That variability helps explain why a mean estimate can remain uncertain even when a numerical point estimate is available. The endpoint counts also differ between groups, with fewer observations reported for the intervention group. The record does not explain the outcome counts or report whether missingness may have affected the estimates.

For systolic blood pressure, the reported interval also includes zero, extending from -2.271 to 4.871 for the intervention-versus-control mean difference. This indicates uncertainty about direction as well as size. The secondary outcome's interval is a separate estimate and should not be combined mentally with the primary interval as if they measured the same quantity.

Confidence intervals describe statistical uncertainty under the analysis that produced them. The registry does not specify the confidence level or full analysis procedure, so interpretation should remain tied to the posted summaries. In particular, an interval that includes zero is not proof of no effect; it means the estimate does not distinguish a difference from no difference with the precision reflected in that interval. A narrower estimate, a clinically meaningful threshold, or a prespecified equivalence analysis would be needed to make stronger claims about the range of effects that can be ruled out. Such a threshold or equivalence analysis is not reported in the registry record.

12. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The planning figures are participants per arm for this continuous-outcome comparison; they are not total events. The illustration assumes two-sided significance level 0.05, equal allocation between the two arms, and a normal approximation. It uses the observed effect as a reference and shows planning for 100%, 75%, and 50% of that observed effect. The original design assumptions and sample-size justification are not reported in the registry record.

The 80% power column gives the estimated participants per arm under the stated assumptions for a design with 80% power to detect the effect used in that scenario. The 90% power column gives the corresponding participants per arm for 90% power. Power is a design property under assumed conditions; it is not a guarantee that a study will detect an effect, and it does not describe the probability that the observed result is true. These planning figures are not observed power.

Each row changes the assumed effect size relative to the observed estimate. The smaller the effect the design is meant to detect, the more participants are generally needed, because smaller group differences are harder to distinguish from random variation. The effect used for planning is therefore a substantive design choice, not merely a computational detail. The figures are illustrations based on the posted observed estimate and stated assumptions; they should not be read as a reconstruction of the trial's original enrollment target or as evidence that the trial was designed around those values.

The same planning logic applies to the secondary outcome, whose own observed effect is the reference for its illustration. The resulting participant requirements are specific to that outcome and its variability, rather than interchangeable with those for PAM. Interpreting either set requires attention to the chosen target effect, power, allocation, and model assumptions.

13. How the Sample Size Is Derived

Power is the probability that a study of a given size rejects the null hypothesis when a specific true effect exists. It depends on four things: the effect to detect, the variability of the outcome, the significance level (here a two-sided α of 0.05, so z_1−α/2 = 1.960) and the desired power (z_1−β = 0.842 for 80% and 1.282 for 90%).

Formula

n is the number of participants in each arm, Δ is the difference in means to detect and σ is the common standard deviation.

Worked calculation for this trial

1. Observed difference Δ = -3.400; pooled standard deviation σ = 18.753.
2. n = 2 × (1.96 + 0.842)² × 18.753² / -3.400² = 477.6, rounded up to 478 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 <- -3.4                      # observed difference in means
sigma <- 18.7530325815064                     # 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(2806700)
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 = 478 (80% power), 640 (90% power)
 75% of observed effect: n per arm = 849 (80% power), 1,137 (90% power)
 50% of observed effect: n per arm = 1,911 (80% power), 2,558 (90% power)
power.t.test: 478.5 
Simulated power at n = 478 per arm: 0.797 (20000 replicates)

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

14. Multiplicity, Missing Data and Analysis Populations

The registry identifies one primary outcome and one secondary outcome, but it does not report how statistical testing across outcomes was planned. There is no reported multiplicity adjustment, testing hierarchy, or family of hypotheses. When more than one outcome is tested, the chance of at least one apparently unusual result can rise if each is judged independently. General methods for managing this issue include prespecifying a hierarchy or adjusting inference across related tests. Without a reported approach, the unadjusted p-values should not be treated as if they account for all outcome testing.

Missing-data handling is also not reported. The numbers represented in the primary and secondary outcome summaries differ, and each group has different counts across the two outcomes. The registry does not state whether participants failed to complete surveys, lacked a measurement, were excluded for another reason, or had data handled through an imputation or other method. It is therefore not possible to infer why the counts differ or whether missingness was related to assignment or outcome values.

The analysis population is not reported. The record does not specify whether all randomized participants were included according to their assigned group, whether a per-protocol analysis was used, or whether another set of participants contributed to each outcome. It also does not report how participants who did not use Twitter as requested were handled. These are important distinctions: an analysis by randomized assignment estimates a different contrast from an analysis restricted to people who followed the intervention instructions.

The registry lists no reported analyses. Detailed model specifications, covariate adjustments, sensitivity analyses, handling of baseline measurements, and any interim analyses are not reported in the registry record. No interim analysis should be presumed. The absence of these details does not establish that a method was not used; it limits what can be verified from the registry entry. A fuller report would be needed to evaluate whether outcome definitions and analytic decisions followed a prespecified plan.

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

The posted primary results describe average changes in PAM for the two randomized groups and an estimated between-group difference. The point estimate favors a lower average change in the intervention group relative to control, but its confidence interval includes zero and the unadjusted p-value does not provide clear evidence of a group difference. The secondary systolic blood pressure comparison is also uncertain. These findings describe what the registry reports; they do not establish that the intervention caused a change in either outcome or that no effect is possible.

A result that does not clearly distinguish groups is not the same as evidence that the groups are equivalent. Equivalence requires a prespecified margin defining what differences would be considered sufficiently small, along with an analysis designed to assess that margin. The registry does not report such a margin or an equivalence analysis. The intervals still allow a range of possible differences, and the registry does not identify a threshold for a clinically important change in PAM or systolic blood pressure.

The trial also does not answer every question about social media and health. It concerns a particular behavioral intervention described as using Twitter for heart health, among people identified as having diabetes from Twitter, with outcomes measured at baseline and 6 months. The registry does not provide enough detail to characterize actual exposure, engagement, or the specific communications participants received. The results cannot be generalized automatically to other online platforms, other intervention formats, other populations, or other follow-up periods.

Randomization is relevant to interpreting the assigned-group comparison, but the outcome summaries alone do not explain adherence, missing data, or possible bias from unmasked participation. The registry's lack of a reported analysis plan also prevents assessment of whether the posted comparisons reflect the prespecified methods. These limitations do not erase the observed summaries; they define how far those summaries can support conclusions.

Accordingly, the results are best understood as imprecise posted estimates from a completed randomized trial, rather than as a definitive answer about the effects of Twitter use for heart health. They neither demonstrate a clear between-group difference on the reported outcomes nor establish that no difference exists.

16. Why This Trial Matters Statistically

This trial is statistically interesting because it evaluates a behavioral intervention delivered through a communication channel that the registry describes as potentially reaching broad demographic segments and conveying health-related messages. The design turns that broad idea into a comparison between an intervention group asked to use Twitter for heart health and a survey-contacted control group. That makes clear why outcome definition and allocation matter: online activity is the intervention, while patient activation and systolic blood pressure are distinct measures of its possible relevance.

The primary endpoint focuses on a person's reported ability to manage health, rather than on a clinical event. Its continuous change score allows the comparison to retain information about the amount of change instead of reducing participants to categories alone. At the same time, the reported variability and confidence interval show why an average difference can remain uncertain. Statistical interpretation requires looking beyond whether an estimate has a sign to how precisely it is estimated and what values remain compatible with the data.

The record also illustrates the importance of transparent reporting. Enrollment is stated, but outcome counts differ by endpoint, and no explanation of missing measurements or analysis populations is reported. The registry posts group summaries and unadjusted comparisons, but no analyses are described. This makes it difficult to assess the analysis choices, multiplicity handling, and the relationship between planned and reported methods. Those are practical issues in interpreting pragmatic or digitally delivered behavioral interventions, not merely technical details.

Finally, the planning illustration makes the role of the target effect visible: detecting a smaller effect generally requires more participants under the same assumptions. Because those calculations are based on the observed effect rather than the original design rationale, they cannot answer whether the trial was adequately planned. The distinction between design planning and observed results is central to reading the statistical evidence responsibly.

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.

Several limitations are specific to the population and intervention recorded. Participants were identified as having diabetes from Twitter, which makes the recruitment context part of the target population. The registry does not report how diabetes status was confirmed, how people were selected or contacted, or whether those who did not use Twitter could participate. The findings therefore may not represent people with diabetes who were not reachable through this channel. Additional demographic and clinical characteristics are not reported in the registry record.

The intervention description is broad. It gives examples of tweeting, following, and receiving tweets for heart health, but not the amount or content of activity, the schedule, or whether engagement was monitored. The control description specifies survey contact, but does not clarify whether participants had access to other health information or Twitter activities. Without these details, it is difficult to characterize the contrast in exposure or assess how consistently the intervention was delivered.

The trial was unmasked, and the primary endpoint is a self-management measure reported through surveys. Participant awareness of assignment could affect responses, while the registry does not state whether outcome assessors or analysts were masked. This is a potential interpretive concern, not evidence that responses were biased. Follow-up and survey completion details are not reported, and the different outcome counts cannot be explained from the registry entry alone.

The record also lacks reported analyses and a sample-size justification for the original design. It does not provide enough detail to assess covariate adjustment, missing-data handling, analysis populations, adherence, or multiplicity. The adverse-event records list zero counts for participants at risk as well as zero event counts in each group. Because the numbers at risk are also zero, those entries should not be read as evidence that no adverse events occurred among the enrolled participants.

18. Related Tutorials

19. Related Statistical Calculators

20. Sources

• ClinicalTrials.gov: Twitter and Diabetes — NCT02806700.
• Registry record last updated 2024-10-21; status last verified 2024-10.
• Results first posted 2024-10-21.
• 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

NCT02806700, titled Twitter and Diabetes, is recorded as a completed randomized, parallel-group behavioral trial with no masking. Enrollment was 628, and the registry lists diabetes and cardiovascular disease as conditions. The intervention group was asked to use Twitter for heart health; the control group was contacted and asked to complete brief surveys.

The primary outcome was change in Patient Activation Measure score from baseline to 6 months. The posted mean difference was -3.4, with a confidence interval from -7.703 to 0.903 and an unadjusted p-value of 0.121. The secondary outcome was change in systolic blood pressure over the same timeframe; its posted comparison also had a confidence interval including zero and an unadjusted p-value of 0.476. The registry reports no analyses and does not report the original sample-size justification, missing-data procedures, or analysis populations. The endpoint summaries do not show a clear between-group difference, while leaving uncertainty about the possible size and direction of effects.