Clinical Evaluation of CARESTREAM Dual Energy and Digital Tomosynthesis: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Clinical Evaluation of CARESTREAM Dual Energy and Digital Tomosynthesis (NCT02531646), a randomized na study. The primary outcome is Radlex Scale for Diagnostic Quality Ratings - DE Predicate PA Chest.

2. Clinical Question

3. Trial Design

The registry identifies NCT02531646 as a completed clinical evaluation of CARESTREAM Dual Energy and Digital Tomosynthesis. It lists a randomized allocation, double masking, and a parallel model. Enrollment is recorded as 66. The study is listed as not applicable for phase, with a start in 2015-11 and primary completion in 2016-04. These record entries describe the registered design and timeline; they do not establish how every planned procedure was carried out.

The human-subject dual energy arm description says that thirty to forty (30-40) patients would receive a DR standard-of-care chest exam using the DRX Plus detector and a dual energy exam. It describes high energy and low energy exposures with the investigational device. The human-subject digital tomosynthesis arm description says that fifteen to twenty (15-20) patients would receive a DR standard-of-care chest exam and a digital tomosynthesis exam. That exam included a scout image (chest PA) and a tomosynthesis scan using investigational software. These descriptions use planned ranges; the registry record does not give actual enrollment or completion counts for these arms.

A separate arm describes eleven (11) phantoms of various anatomy imaged with linear tomography as predicate and digital tomosynthesis as investigational. Phantom imaging is distinct from imaging people, so it should not be read as a patient outcome group. The record does not explain the randomization procedure, allocation concealment, how masking was implemented, or whether the listed design descriptors applied in the same way to the phantom work. Those details are not reported in the registry record.

The registry lists Radiation as the intervention type. It does not report a detailed protocol, image acquisition settings, reader training procedures, or operational definitions for the predicate devices beyond the arm descriptions. The study is marked completed, but the completion status alone does not establish that every registered analysis was reported.

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: 224 image ratings from 7 radiologist readers (32 images rated x 7 readers) for Predicate \& Invest. DE - Human Subjects arm above.

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: 37 live human DE cases, 18 live human DT cases, 11 phantom images.

6. Primary Endpoint

The registry lists the primary outcome as Radlex Scale for Diagnostic Quality Ratings - DE Predicate PA Chest, measured over 3 months.

The primary endpoint is titled Radlex Scale for Diagnostic Quality Ratings - DE Predicate PA Chest. It assesses image quality using a diagnostic-quality scale with verbal categories: ratings from 1-1.9 indicate non-diagnostic quality; 2-2.9 indicate limited but acceptable quality; 3-3.9 indicate diagnostic quality; and 4 indicates exemplary quality. The descriptions distinguish images that provide little or no usable diagnostic information from images considered adequate or particularly suitable as examples.

Using a rating scale gives readers a structured way to classify image quality rather than relying only on an unstructured impression. The registry specifies a mean, which summarizes the average rating across the observations included in the outcome, and Standard Error, which describes the precision of that estimated mean under the analysis used. The endpoint description gives a timeframe of 3 months.

The population description for this endpoint is 224 image ratings from 7 radiologist readers, with 32 images rated x 7 readers, in the Predicate & Invest. DE - Human Subjects arm. Those are ratings, not a count of unique patients. The registry record does not report the observed mean, standard error, or any confidence interval for the endpoint. Nor does the endpoint title alone establish whether a formal comparison was made between the predicate and investigational images.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The registry describes a continuous endpoint summarized as a mean, with Standard Error as the dispersion measure. A mean gives the arithmetic average of the included ratings. The Standard Error describes the uncertainty associated with estimating that average under a particular statistical model. It is not the spread of individual ratings: ratings can vary substantially even when the mean is estimated relatively precisely. The registry record does not include the actual mean or Standard Error, so the endpoint's numerical estimate and precision cannot be discussed.

What does a rating-scale mean represent?

The Radlex Scale for Diagnostic Quality Ratings uses ordered categories with qualitative interpretations. Averaging ratings treats the scale values as quantities that can be combined into a mean. That can be useful as a compact summary, but the mean does not show how ratings are distributed across categories. A given average can arise from different patterns of ratings, including different amounts of disagreement among readers. The registry record does not report category counts or the distribution of ratings, so those features cannot be assessed.

The scale descriptions set out thresholds for non-diagnostic, limited but acceptable, diagnostic, and exemplary image quality. The primary endpoint title identifies predicate PA chest images in the dual energy human-subject arm, while the study's general purpose concerns dual energy and digital tomosynthesis applications versus predicate devices. These are not interchangeable descriptions: the endpoint title is narrower than the overall study purpose. The available endpoint details do not define a numerical contrast or state whether ratings were paired across image types.

Why can repeated ratings affect uncertainty?

The endpoint population describes 224 image ratings from 7 radiologist readers, with 32 images rated x 7 readers. Ratings from one reader may resemble one another because readers can differ in how they use a scale. Multiple ratings of the same image may also resemble one another because they share image characteristics. These are general statistical considerations for reader-rating studies, not claims about the analysis actually used in this trial. When observations are related, methods that account for reader and image structure can produce uncertainty estimates that differ from methods assuming independent observations. The registry record does not report how the analysis handled those potential dependencies.

A Standard Error is often used to form a confidence interval, but a confidence interval is not reported in the registry record for this endpoint. In general, a confidence interval gives a range of values compatible with the data and the chosen statistical procedure, and its width communicates precision. A narrow interval indicates more precision than a wide interval, but neither alone establishes clinical importance. No interval can be reconstructed without the relevant estimates and analysis details.

What does a comparison require?

To interpret a comparison, readers need to know the contrast being estimated, the observations included, and the method used to account for the design. Randomized allocation is listed, but the record does not give the sequence-generation method, allocation concealment, or the connection between randomization and the specific image-rating endpoint. It also does not say whether the endpoint analysis was adjusted, whether ratings were paired, or how missing ratings were handled. Those choices matter because they affect both the estimate and its uncertainty.

The registry lists 9 outcome measures and 6 analyses. Multiple outcomes or comparisons raise the general question of multiplicity: when many statistical tests are conducted, the chance of at least one apparently notable result can increase. There are methods to address this, but the appropriate approach depends on the analysis plan and the role of each endpoint. The registry record does not describe a multiplicity strategy or identify the hierarchy of outcome measures. No adjustment method should therefore be inferred.

How should planning illustrations be read?

Planning illustrations based on the observed effect are hypothetical sample-size calculations, not observed trial results and not observed power. The stated assumptions are a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation; the effect to detect is a stated fraction of the observed effect. For time-to-event illustrations, the calculation concerns the total number of events; for other illustrations, it concerns participants per arm. These general planning assumptions do not establish the study's original design rationale. The registry record does not report a sample-size justification or indicate that these illustrations formed part of the sponsor's prospective plan.

Because the primary endpoint result and its analysis details are not reported in the registry record, readers cannot evaluate model fit, sensitivity analyses, precision, or the robustness of the primary comparison. Recalculated results, if presented elsewhere, would be simple unadjusted recalculations from posted summary statistics and would not replace the sponsor's own analysis. No such recalculated primary result is included in the recorded details.

10. Confidence Intervals for Each Arm

No two-arm comparison is shown for this trial, but each arm's mean can be given with a t-based 95% confidence interval computed from the posted mean, spread and number of participants.

R code and output

arm <- c("Predicate & Invest. DE - Human Subjects")
n   <- c(32)
m   <- c(3.454)
sp  <- c(0.0183)
sd  <- sp * sqrt(n)      # standard errors converted to SDs
for (i in seq_along(n)) {
  se <- sd[i] / sqrt(n[i]); h <- qt(0.975, n[i] - 1) * se
  cat(sprintf("CARM|%s|%d|%.3f|%.3f|%.3f|%.3f\n", arm[i], n[i], m[i], sd[i], m[i] - h, m[i] + h))
}

CARM  Predicate & Invest. DE - Human Subjects  32  3.454  0.104  3.417  3.491

11. Multiplicity, Missing Data and Analysis Populations

The registry lists 9 outcome measures and 6 analyses. Multiple measures and analyses can involve several related questions, but the record does not identify a prespecified testing hierarchy, a primary family of comparisons, or a correction for multiple testing. It is therefore not possible to determine from the registry record whether the reported analyses, if any, controlled the overall chance of false-positive findings across comparisons. A multiplicity approach should not be assumed.

The primary endpoint description specifies a mean and Standard Error, but does not report the analysis population, exclusions, missing-rating rules, or the handling of incomplete assessments. The record also does not state whether all planned readers and images contributed to each analysis, or whether a sensitivity analysis assessed the effect of missing observations. These details are important because the number and pattern of missing ratings can affect both the estimate and its precision.

The registry's enrollment entry is 66, while the arm descriptions give planned ranges for human-subject imaging and a separate phantom imaging group. The record does not provide an account reconciling planned ranges with actual arm-level enrollment, nor does it give an analysis-set flow. The endpoint description reports ratings and readers, but does not explain how those observations were selected for analysis or how their dependence was modeled. The randomization method and analysis populations are not reported in the registry record.

Consequently, interpretation should distinguish the design information that is registered from analysis details that remain unreported. A completed status and listed outcome measures do not establish that there were no missing observations or that all outcomes were analyzed as planned.

12. Safety Results

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

Other (non-serious) adverse events

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

The adverse-event entries report zero serious affected participants and zero other affected participants for each listed group. The corresponding at-risk counts are 32 for Predicate & Invest. DE - Human Subjects, 17 for Predicate & Invest. DT - Human Subjects, and 11 for Predicate & Invest. DT - Phantom Images. These entries describe the registry's recorded counts; they should not be read as proof that no event of any kind occurred outside the recorded categories or reporting period.

The registry gives no death count for the groups. A missing death value is not the same as a reported zero. The top serious terms list is empty, and the registry does not provide further event descriptions or a detailed safety follow-up account in the registry record. The phantom group consists of phantom images, not people, so its at-risk entry should not be interpreted as a patient safety denominator.

These safety counts are descriptive. They do not provide a comparative estimate, confidence interval, or statistical test of differences between groups. With no further detail about event definitions, timing, ascertainment, or the reporting window, the figures do not support broader conclusions about safety. They only state the recorded affected counts and the listed groups at risk.

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

The registry establishes that a clinical imaging evaluation was registered as randomized, double masked, and parallel, and that its stated purpose was to evaluate dual energy and digital tomosynthesis against commercially available predicate devices. It also identifies a primary diagnostic-quality rating endpoint and describes the rating scale. These features make clear what kind of question the study was intended to address: image quality assessment, rather than a patient health outcome.

The registry entries do not include the primary mean, Standard Error, contrast between image types, confidence interval, or p-value. As a result, they do not establish the direction or size of any difference in ratings, nor how precisely such a difference might have been estimated. The endpoint's category definitions explain how ratings are interpreted, but definitions alone are not findings.

The record describes a set of image ratings by radiologist readers. The number of ratings is not equivalent to a count of unique patients. The record also includes human-subject and phantom imaging descriptions, which represent different kinds of evidence and should not be combined as if they were the same population. The registry does not report sufficient analysis detail to determine how repeated ratings, missing data, or multiple outcomes were handled.

The zero affected counts in the adverse-event entries are specific registry entries, not a complete account of every possible safety outcome. The record does not include death counts or detailed follow-up information for these entries. Likewise, completed status indicates the trial's recorded status, not that every planned result or analysis is available.

Any planning illustration based on an observed effect remains a hypothetical sample-size calculation. It is not evidence that the study had adequate power or that an effect was detected. Overall, the registry record provides useful design and endpoint context but, without the numerical primary result and analysis methods, does not permit a conclusion about the observed diagnostic-quality comparison.

14. Why This Trial Matters Statistically

This trial is statistically interesting because its endpoint concerns ratings of medical images, an outcome that can involve repeated observations from readers and images. The endpoint description reports 224 image ratings from 7 radiologist readers, with 32 images rated x 7 readers. Such data have a structure that may be more complex than a set of unrelated observations. Reader-specific rating tendencies and image-specific characteristics are general sources of dependence that an analysis may need to consider.

The scale also illustrates a common distinction in clinical research: a concise numerical summary can be useful, yet it may hide the distribution and interpretation of the underlying ratings. The Radlex categories give descriptive meaning to ranges of the rating scale, while the mean summarizes values across observations. Without the distribution, uncertainty estimate, and analysis details, readers cannot tell whether the average reflects consistent ratings or substantial variation among readers or images.

The trial lists 9 outcome measures and 6 analyses, making transparency about endpoint selection and multiplicity relevant. Knowing which comparison was primary, how other analyses were treated, and whether the reader-image structure was modeled would help readers interpret the evidential strength of any result. The registry record does not report those details or the numerical primary result.

The study also includes phantom imaging alongside human-subject imaging. Keeping those evidence sources distinct is important: phantom image comparisons can inform technical evaluation, while human-subject ratings address images obtained in patients. Their roles and analysis populations cannot be inferred beyond the arm descriptions. The record's limitations illustrate why design labels and endpoint definitions are not substitutes for complete results and transparent statistical methods.

15. 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 record does not report the numerical result for the primary endpoint, including its mean, Standard Error, confidence interval, or statistical test. This prevents assessment of the observed rating level, the precision of the estimate, and any contrast involving predicate and investigational images. The endpoint title identifies dual energy predicate PA chest images, but does not by itself specify a complete comparative estimand.

The primary endpoint description reports 224 image ratings from 7 radiologist readers, with 32 images rated x 7 readers. It does not state whether ratings were analyzed as independent observations, whether reader or image clustering was accounted for, or whether assessments were paired. Those omissions limit assessment of the uncertainty calculation. The record also does not give the rating distribution, reader agreement, or missing-rating information.

The registry lists randomized allocation, double masking, and a parallel model, but does not describe the randomization procedure, allocation concealment, implementation of masking, or how these design features related to image review. Planned ranges appear in the human-subject arm descriptions, but actual arm-level enrollment and analysis-set counts are not reported. The enrollment entry of 66 does not supply that missing flow information.

The record lists 9 outcome measures and 6 analyses without describing multiplicity control, the statistical analysis plan, model choice, or sensitivity analyses. Sample-size justification and interim analyses are not reported in the registry record. Details of image acquisition, reader qualifications beyond the stated radiologist readers, and the predicate-device comparison procedures are also limited or absent from the recorded information. These gaps constrain both reproducibility and interpretation.

The safety entries do not provide death counts or detailed event follow-up. The phantom imaging group is not a patient group and should not be used to interpret patient safety. The registry's completed status does not resolve these reporting limitations.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Clinical Evaluation of CARESTREAM Dual Energy and Digital Tomosynthesis — NCT02531646.
• Registry record last updated 2017-03-09; status last verified 2015-08.
• Results first posted 2016-11-02.
• 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.

19. Record Summary

NCT02531646 is a completed study titled Clinical Evaluation of CARESTREAM Dual Energy and Digital Tomosynthesis. The registry identifies it as randomized, double masked, and parallel, with enrollment recorded as 66 and phase listed as not applicable. The recorded study period starts in 2015-11, with primary completion in 2016-04.

The study evaluates investigational dual energy and digital tomosynthesis imaging against commercially available predicate devices. Its primary endpoint is the mean Radlex Scale for Diagnostic Quality Rating for DE Predicate PA Chest, with Standard Error as dispersion and a timeframe of 3 months. The endpoint description identifies 224 image ratings from 7 radiologist readers, with 32 images rated x 7 readers. The registry record does not report the numerical primary result or enough analysis detail to interpret a comparative effect.

Three arm descriptions cover human-subject dual energy imaging, human-subject digital tomosynthesis imaging, and phantom imaging. Recorded affected counts for serious and other adverse events are zero in each group; death counts are not reported. Further details on analysis populations, missing data, multiplicity, and statistical methods are not reported in the registry record.