The Impact of a Dermatology Information Source on Skin Problem Outcomes in Primary Care: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for The Impact of a Dermatology Information Source on Skin Problem Outcomes in Primary Care (NCT02922738), a randomized na study. The primary outcome is Number of Patients With Resolved Skin Problems.

2. Clinical Question

The study addresses whether a clinical information resource used during primary care might be associated with a patient outcome, rather than only with provider knowledge or decision-making. That distinction matters statistically: the assigned intervention was provider use of VisualDx, but the primary outcome was the patient’s report of resolution. The record describes a cluster-randomized controlled trial, with providers as clusters and patients grouped according to the provider they saw. This design makes the provider, rather than the individual patient, the unit of assignment. The registry does not report the number of provider clusters or the number of patients in each randomized group.

The primary question is about time to reported resolution during follow-up, not simply the proportion resolved at a single fixed assessment. Patients were interviewed by phone, and patients still described as unresolved at 90 days were censored. This framing allows the analysis to use the timing of reported resolution, subject to the assumptions and details of the Cox model. The record does not report whether the use of VisualDx changed provider decisions, what information patients received, or how any such changes might relate to the outcome.

3. Trial Design

The trial is described as randomized, single-masked, parallel, and cluster-randomized. Primary care providers were the clusters: providers were assigned either to refer to VisualDx when seeing a patient with a skin problem or to usual care without VisualDx referral. Patients inherited the randomized group status of the provider they saw. The registry records enrollment of 465, but does not report the number of provider clusters or the group-specific patient enrollment in the information available for this analysis.

Cluster assignment has important consequences for interpreting the result. Patients cared for by the same provider may have more similar experiences than patients cared for by different providers. Their outcomes therefore may not behave like independent observations. The sponsor’s reported analysis is a Cox proportional-hazards regression. The registry record does not report the cluster-adjustment details, such as whether the analysis accounted for clustering through a robust variance estimate, a frailty term, or another approach. It also does not identify who was masked, or describe how allocation was generated or concealed.

The intervention arm involved providers referring to VisualDx; the control arm involved usual care, which could include other information sources or none. The record describes patient interviews about problem status and follow-up visits at intervals after the index visit. For the primary endpoint, the stated assessment period was up to 3 months (90 days) after the index visit. Patients reported as unresolved at 90 days were censored. The registry does not provide the interview schedule or explain how missed interviews and other incomplete outcome information were handled.

The study status is recorded as completed. The recorded start was 2015-06 and primary completion was 2016-11. The record identifies the University of Vermont as sponsor and lists the condition categories as clinical decision support systems and skin diseases. The registry does not report details of the recruitment process, provider training, adherence to VisualDx referral, or cross-use of information sources between groups. Those details would help describe implementation, but their absence does not change the basic distinction between provider-level randomization and patient-level outcome assessment.

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

Analysis population for the primary outcome, as defined in the registry: Number of patient participants only (excludes providers)

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.

6. Primary Endpoint

The registry lists the primary outcome as Number of Patients With Resolved Skin Problems, measured over Period of assessment was up to 3 months (90 days) after index visit..

The primary endpoint is the number of patient participants with resolved skin problems. Resolution was assessed through a phone interview, based on the patient’s report of skin disease status. The registry states that the assessment period extended up to 3 months (90 days) after the index visit and that patients whose problems remained unresolved at 90 days were censored. The outcome is therefore connected to reported time to resolution, as reflected by the Cox proportional-hazards analysis, rather than only a reported status at a single visit.

Time-to-event methods can use information about when an event occurs while also accommodating follow-up that ends before an event is observed. In this trial, the event is reported resolution. Censoring an unresolved patient at 90 days means that the analysis does not count that patient as having resolved; it treats the observed follow-up as ending without an observed resolution event by that time. Censoring is not evidence that the problem later resolved or remained unresolved indefinitely.

This endpoint is relevant to the stated objective because it focuses on a patient outcome after a provider used, or did not use, the information source. It is also distinct from outcomes such as diagnostic accuracy, referral decisions, or provider confidence, none of which is identified as the primary endpoint. The record does not report a separate formal definition of resolution, corroboration by a clinician, or validation of the patient-reported status. These features should be considered when interpreting the endpoint as a measure of patient-reported problem resolution.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The reported method is Cox proportional-hazards regression. This is a time-to-event approach: it analyzes the time from a defined starting point to an event, while allowing some observations to end without the event having been observed. In this trial, the starting point is the index visit and the event is patient-reported resolution of a skin problem. The assessment period was up to 3 months (90 days), and patients unresolved at 90 days were censored according to the registry description.

What does a hazard ratio describe?

A hazard is the instantaneous event rate among people who have not yet experienced the event, conditional on remaining under observation up to that time. A hazard ratio compares that rate between groups. It is a relative measure, not a count of resolved patients and not a direct measure of the number of days between outcomes. The posted estimate is 0.92. The registry does not report the contrast orientation, so it is not possible to say from this record alone which group’s hazard is in the numerator and which is the reference.

The conventional interpretation of a Cox hazard ratio relies on proportional hazards: the relative hazards are assumed to remain constant over follow-up. This is a general model assumption, not a diagnostic result reported for this trial. The registry record does not report whether the assumption was assessed, whether any departures were found, or whether an alternative time-to-event model was considered. If relative hazards vary over time, a single hazard ratio may not fully describe the outcome pattern.

How should the interval and p-value be read?

The reported 95% confidence interval is 0.70 to 1.21. A confidence interval expresses uncertainty about the effect estimate under the model and its assumptions; it is not a guarantee that the true effect lies within the interval. Here the interval includes the null value for a relative hazard. It also includes estimates on either side of that value, so the data do not narrowly identify a single direction or magnitude. The reported p-value is 0.54. It is a measure of compatibility with the null model, not the probability that the null is true and not a measure of clinical importance.

When a confidence interval includes the null value, that does not establish that the groups are equivalent. Equivalence requires a prespecified range of differences considered acceptably small, along with an analysis aimed at testing that question. The registry does not report an equivalence or noninferiority objective, a margin, or a corresponding analysis. The superiority classification in the registry identifies the type of comparison, but does not turn a nonsignificant result into evidence of no effect.

Why does clustering matter?

Randomization occurred among providers, while the primary outcome was observed among patients. Patients treated by the same provider may share care processes and other characteristics, making their outcomes correlated. A model that treats every patient as fully independent can understate uncertainty if it ignores that dependence. Cluster-aware methods can address this in different ways, but the registry does not report which specific variance or model approach was used in the sponsor’s Cox analysis. It also does not report the number of clusters, cluster sizes, or an intracluster correlation estimate.

Because the design is cluster-randomized, an individual-level recalculation from posted summary statistics would not preserve the design information needed for valid uncertainty estimates. The registry specifically excludes a simple individual-level re-analysis as misleading and presents the sponsor’s analysis. No recalculated hazard ratio or recalculated p-value is reported. The sponsor’s own analysis therefore takes precedence over an unadjusted recalculation.

What does censoring contribute?

Censoring records that follow-up ended without an observed event by the time observation stopped. For this endpoint, unresolved status at 90 days was censored; that is different from classifying the participant as resolved or treating the outcome as missing in the same way as an absent interview. Valid time-to-event interpretation generally depends on censoring being handled appropriately in relation to the event process. The registry does not describe how censoring was implemented in the model, how loss to follow-up was handled, or whether the assumptions about censoring were examined.

The registry also does not report the model covariates, the treatment of ties in event times, diagnostic checks, or whether the hazard ratio was adjusted for baseline characteristics. These are not grounds to assume that a method was or was not used; they are details not reported in the registry record. The endpoint is patient-reported, and the record does not provide information about validation, the interview instrument, or adjudication of resolution.

How do planning calculations differ from trial results?

Planning illustrations based on an observed effect are not observed trial outcomes and are not observed power. The stated assumptions for those illustrations are a two-sided significance level of 0.05, equal allocation between arms, and a normal approximation; the effect to detect is a stated fraction of the observed effect. For time-to-event calculations, the stated output is the total number of events using the Schoenfeld method; other calculations are expressed as participants per arm. These are planning conventions, not evidence that the trial was designed around those assumptions. The registry does not report the trial’s original sample-size justification or show that these planning assumptions governed its design.

10. Exact Estimates for Each Arm

No two-arm comparison is shown for this trial, but each arm's result can still be summarized with exact interval estimates computed from the posted counts. The Clopper-Pearson interval is exact and conservative; the Wilson interval is a common closer-to-nominal alternative; the Bayesian credible interval uses a uniform Beta(1, 1) prior. For small groups the three can differ noticeably.

R code and output

arm    <- c("Intervention - Used VisualDx", "Control - Usual Care")
n      <- c(158, 275)
events <- c(72, 135)
for (i in seq_along(n)) {
  ci <- binom.test(events[i], n[i])$conf.int          # Clopper-Pearson exact interval
  cr <- qbeta(c(0.025, 0.975), 1 + events[i], 1 + n[i] - events[i])   # Bayesian, uniform Beta(1,1) prior
  wi <- (function(x, n) { p <- x / n; z <- qnorm(0.975); d <- 1 + z^2 / n
    c((p + z^2 / (2 * n) - z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d, (p + z^2 / (2 * n) + z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d) })(events[i], n[i])
  cat(sprintf("ARM|%s|%d|%d|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f\n", arm[i], n[i], events[i], 100 * events[i] / n[i],
              100 * ci[1], 100 * ci[2], 100 * wi[1], 100 * wi[2], 100 * cr[1], 100 * cr[2]))
}

ARM  Intervention - Used VisualDx  158  72  45.6  37.6  53.7  38.0  53.3  38.0  53.4
ARM  Control - Usual Care  275  135  49.1  43.0  55.2  43.2  55.0  43.2  55.0

11. Analyses Reported by the Sponsor

The registry reports the sponsor’s own Cox proportional-hazards analysis: an estimate of 0.92, a 95% confidence interval from 0.70 to 1.21, and p = 0.54. It does not report an independent recalculation. The record explicitly notes that the study was cluster-randomized and that a simple individual-level re-analysis would be misleading, so only the sponsor’s analysis is shown. For that reason, the sponsor’s analysis takes precedence over any unadjusted recalculation from summary statistics.

This distinction is about the design and the uncertainty calculation, not an endorsement of an analysis method whose details are absent. With provider-level assignment and patient-level outcomes, statistical methods need to account for the grouping of patients within providers. The registry does not report how the sponsor’s Cox model accounted for clustering, the number of provider clusters, or the model’s adjustment variables. The reported hazard ratio and interval can be quoted as the registry’s result, while the details needed to independently reproduce the analysis are not reported in the registry record.

The estimate’s contrast orientation is also not identified in the registry information used. Therefore the numerical result should not be translated into a claim that one named arm had a higher or lower rate of resolution. The confidence interval includes the null value, and the p-value is 0.54; these describe statistical uncertainty under the sponsor’s model, not proof of equivalence or absence of an effect. Planning illustrations based on the observed effect are separate from this sponsor analysis and should not be read as observed power.

12. Multiplicity, Missing Data and Analysis Populations

The registry lists two outcome measures and two analyses, but the information available for this interpretation identifies the primary endpoint and its Cox result. The record does not report whether the analyses were prespecified in a statistical analysis plan, whether there were additional exploratory comparisons, or whether any method was used to account for multiple testing. Without those details, the degree to which the reported p-value was selected from a larger set of analyses cannot be assessed. The p-value should be interpreted as reported, without an assumption that multiplicity was or was not addressed.

The primary endpoint’s analysis population is described as patient participants only, excluding providers. The registry does not report the precise analysis population for the Cox model, such as how participants with incomplete interviews or other protocol deviations were handled. It also does not report whether an intention-to-treat principle was followed, whether patients were analyzed according to their provider’s assignment regardless of actual VisualDx use, or whether any per-protocol analysis was conducted. The description that patients inherited their provider’s group status establishes the assignment structure, but does not by itself specify all analysis-population rules.

Missing-data methods are not reported in the registry record. The registry states that patients unresolved at 90 days were censored, but does not describe how missed interviews, withdrawal, or loss to follow-up were handled. Censoring an observed unresolved patient at the assessment limit is part of the stated time-to-event outcome definition; it should not be conflated with an absent outcome report. The record does not report sensitivity analyses for missingness or censoring.

Other unreported analysis details include the handling of provider clustering, covariate adjustment, proportional-hazards checks, and the rules for including observations with incomplete timing information. These omissions limit independent appraisal of the model and its assumptions. They do not justify inventing a method or concluding that no method was used. They mean that the posted result can be described, but the full analysis pathway and robustness to missing information or multiplicity cannot be established from the registry record.

13. Safety Results

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

Deaths

Other (non-serious) adverse events

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

The registry reports zero serious adverse events and zero other adverse events in the intervention group, with 175 at risk for each category. In the control group it reports zero serious adverse events and zero other adverse events, with 290 at risk for each category. The registry also reports zero deaths in each group. No serious adverse event terms are listed.

These figures describe the adverse-event reporting recorded for the trial; they do not establish that no adverse events of any kind occurred outside the recorded categories or observation procedures. The registry does not report the method or duration of safety surveillance, event definitions, severity grading procedures, attribution rules, or completeness of follow-up for adverse events. The groups’ numbers at risk are not the same as proof that all participants had identical safety observation.

The study intervention was use of an information source by providers, and the adverse-event figures should be read within that study context. They do not establish a general safety profile for clinical care or for use of the information source in other settings. The outcome analysis and the safety reporting answer different questions: the Cox analysis concerns time to patient-reported resolution, while adverse-event counts describe recorded events. The registry reports no events in the listed categories, but the limits of the safety information remain relevant.

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

The posted result is a statistical estimate from a cluster-randomized trial asking whether provider use of VisualDx was associated with time to patient-reported resolution of skin problems. The Cox estimate is 0.92, with a 95% confidence interval from 0.70 to 1.21 and p = 0.54. The interval includes the null value, and the reported test does not provide strong evidence against the null under the model used. Since the contrast orientation is not reported, the estimate should not be described as favoring either the VisualDx arm or usual care.

The result does not establish that the groups were equivalent. A confidence interval that includes the null can still be compatible with meaningful associations in either direction, depending on the clinical context and the values within the interval. The registry does not report a threshold for clinical importance or a formal equivalence objective. It is therefore not possible to conclude from this result alone that any possible difference would be unimportant.

The result also does not establish that the information source caused a change in resolution. Randomization supports a causal comparison in principle, but interpretation depends on the design, analysis, adherence, missing information, and clustering being handled appropriately. The record does not report the details needed to assess all of these issues. Patients’ outcomes were linked to their provider’s assignment, and the provider-level clustering needs to be reflected in the analysis. The registry states that a simple individual-level recalculation would be misleading and reports only the sponsor’s analysis.

The endpoint is patient-reported resolution after the index visit, not independently verified disease status, diagnostic accuracy, quality of care, or referral appropriateness. The registry does not report the interview instrument or a formal adjudication process. Patients unresolved at 90 days were censored, and the record does not describe how other incomplete follow-up was handled. These details affect how broadly the result can be interpreted.

Accordingly, the finding is best described as an uncertain estimate with a confidence interval that spans the null value. It does not show a clear statistical distinction in the reported analysis, but lack of a clear distinction is not evidence that the groups had identical outcomes. It also does not establish that VisualDx is beneficial, ineffective, or harmful. The conclusion should remain limited to the trial’s recorded comparison, its patient-reported endpoint, and the uncertainty and reporting limitations.

15. Why This Trial Matters Statistically

This trial is statistically notable because the intervention was assigned to providers while the primary outcome was measured among patients. That arrangement is common in evaluations of clinical tools: a provider’s behavior or access to information is changed, and the consequences are assessed in patient care. It also creates a hierarchy in the data. Patient outcomes may be more alike within a provider’s practice than across different providers, so the analysis must respect the unit of randomization rather than treating every patient as unrelated.

The registry explicitly recognizes this issue by excluding a simple individual-level recalculation. That choice is important: a calculation based only on patient-level summaries can appear straightforward while failing to represent the randomized clusters and the associated uncertainty. The sponsor’s Cox analysis is therefore the result to interpret, even though the registry record does not report the cluster-adjustment details needed for independent reproduction.

The endpoint also illustrates the distinction between a binary status and a time-to-event outcome. The recorded outcome concerns patients whose skin problems were reported as resolved, while the analysis uses a Cox model and follow-up up to 3 months (90 days). Censoring unresolved patients at 90 days makes the timing and handling of follow-up part of the statistical question. A time-to-event estimate is not interchangeable with a proportion resolved at a single time point.

Finally, the reported confidence interval conveys uncertainty that a p-value alone cannot. The estimate is 0.92 and the 95% confidence interval ranges from 0.70 to 1.21; the interval includes the null value, and p = 0.54. Together, these results show why statistical interpretation should avoid a binary reading of “significant” versus “not significant.” The record supports a description of the observed estimate and its uncertainty, while leaving open questions about clinical importance, clustering details, missing data, and model checks.

16. 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 how this trial is described in the registry. The primary outcome was based on patient-reported skin disease status collected by phone interview. The record does not report the interview instrument, the operational definition of resolution, validation of the report, or clinical confirmation. Patient-reported status is directly relevant to the stated outcome, but its interpretation depends on how the question was asked and how responses were classified.

The cluster-randomized design requires analysis that accounts for provider-level assignment and potential correlation among patients cared for by the same provider. The registry notes that an unadjusted individual-level re-analysis would be misleading, but does not report the number of provider clusters, cluster sizes, or the sponsor’s specific approach to clustering. These omissions prevent an independent assessment of whether the model’s uncertainty estimates fully reflect the design.

The report also leaves important time-to-event details unspecified. Patients unresolved at 90 days were censored, but the handling of missed interviews, withdrawal, and other incomplete follow-up is not reported. The proportional-hazards assumption, model diagnostics, covariate adjustment, and contrast orientation are also not reported. Without the contrast orientation, the numerical hazard ratio cannot be attributed to a particular arm’s relative event rate.

The registry does not report adherence to VisualDx referral, use of other information sources in practice, possible overlap between groups, or provider-level implementation details. These factors could affect how the assigned intervention translated into actual care, but no direction or magnitude can be inferred from their absence. The registry also does not report the original sample-size justification, analysis plan, multiplicity strategy, or a threshold for clinical importance. Consequently, the result can be interpreted as the sponsor’s reported estimate, but not as a complete account of the trial’s design rationale, robustness, or practical significance.

17. Related Tutorials

18. Related Statistical Calculators

19. Sources

• ClinicalTrials.gov: The Impact of a Dermatology Information Source on Skin Problem Outcomes in Primary Care — NCT02922738.
• Registry record last updated 2019-07-30; status last verified 2019-07.
• Results first posted 2019-07-30.
• 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.

20. Record Summary

NCT02922738 is a completed, randomized, single-masked, parallel cluster trial sponsored by the University of Vermont. Providers were assigned to refer to VisualDx for patients with skin problems or to provide usual care; patients took the group status of the provider they saw. The registry records enrollment of 465 and describes patient interviews about outcomes after the index visit.

The primary endpoint was the number of patient participants with resolved skin problems, assessed by phone interview up to 3 months (90 days) after the index visit. Patients unresolved at 90 days were censored. The sponsor’s Cox proportional-hazards result is 0.92, with a 95% confidence interval from 0.70 to 1.21 and p = 0.54. The contrast orientation and details of cluster adjustment are not reported in the registry record.

The registry reports zero serious adverse events, zero other adverse events, and zero deaths in each group. These are recorded safety counts and should be interpreted alongside the limited surveillance details available in the record.