This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results on this page are restricted to the results posted in the ClinicalTrials.gov record; calculations added by Clinical Biostats are labelled as such.
1. Trial at a Glance
This page covers the registry results for Evaluation of Computer-Aided Lung Nodule Detection Software in Thoracic CT for Riverain Technologies LLC (NCT02440139), a randomized na study. The primary outcome is Comparing the Detection Accuracy, Sensitivity, and Specificity of Clinically Actionable Lung Nodules.
| Feature | Registry value |
|---|---|
| ClinicalTrials.gov identifier | NCT02440139 |
| Phase | NA |
| Condition | Lung Cancer |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Enrollment | 12 |
| Arms in the primary outcome | n/a |
| Lead sponsor | Virginia Polytechnic Institute and State University |
| Results posted | Yes |
2. Clinical Question
Population
Radiologists interpreted retrospective thoracic CT cases, described as including cases with cancer or clinically actionable benign nodules and cases without either finding. The brief summary planned approximately 300 cases: approximately 100 true-positive cases and 200 normal cases. It also described at least ten participating radiologists. These planned case and reader counts are not interchangeable with the registry enrollment figure, which is 12.
Intervention
ClearRead CT Insight, a computer-aided lung nodule detection system, was used to support interpretation of thoracic CT images. In the aided arm, radiologists marked locations of concern while using the software, then scored regions according to action and suspiciousness.
Comparator
The comparator was unaided interpretation of conventional thoracic CT images. Radiologists marked locations of concern and scored regions without ClearRead CT Insight. The record describes the study as randomized, double-masked, and parallel, but does not report how randomization or masking was implemented.
Primary question
The study sought to compare detection of clinically actionable lung nodules with and without software assistance. The primary objective was to establish superiority of aided over unaided reading for detecting actionable nodules. The record also describes reading time as a secondary concern. It does not report the observed detection estimates or the statistical comparison needed to determine the result.
3. Trial Design
The registry describes a retrospective, multi-reader, multi-case randomized reader study of ClearRead CT Insight for thoracic CT. Its listed allocation is randomized, masking is double, and model is parallel. These labels describe the registered design; details of the randomization sequence, allocation process, masking procedures, and the roles of readers or other personnel are not reported in the registry record.
The brief summary describes radiologists reading approximately 300 retrospective cases, including approximately 100 true-positive cases and 200 normal cases, and says at least ten radiologists would participate. The registry enrollment figure is 12. The record does not explain the relationship between that enrollment figure and the case or reader counts in the study description, so those figures should be read as distinct kinds of information rather than reconciled by inference.
In the unaided arm, readers marked locations of concern on conventional thoracic CT images and scored regions according to action and suspiciousness. In the aided arm, readers performed the assessment with ClearRead CT Insight. The description says reading time was measured by computer. It does not report the order of readings, whether readers assessed the same cases under both conditions, how cases were assigned, or whether washout procedures were used.
The registry lists the study as completed. Its recorded start is 2015-04, and its primary completion is 2017-09. The trial is associated with Lung Cancer, and the sponsor is Virginia Polytechnic Institute and State University. These record details provide context but do not establish how well the software performed.
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
| Stage | Arm 1 |
|---|---|
| STARTED | 12 |
| COMPLETED | 12 |
| NOT COMPLETED | 0 |
Analysis population for the primary outcome, as defined in the registry: Of the study chest CT cases, some contained cancers and/or clinically actionable benign nodules. They were considered targets in the study. The remaining study cases had no cancer nor clinically actionable nodules which served as true-negative for the 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: 12 US board-certified radiologists were invited to participate in the study. They were 8 males and 4 females.
| Characteristic | Unit | Arm 1: Baseline Study |
|---|---|---|
| Participants analysed | 12 | |
| Age, Categorical: <=18 years | Participants | 0 |
| Age, Categorical: Between 18 and 65 years | Participants | 12 |
| Age, Categorical: >=65 years | Participants | 0 |
| Sex: Female, Male: Female | Participants | 4 |
| Sex: Female, Male: Male | Participants | 8 |
| Race and Ethnicity Not Collected | Participants | — |
| Region of Enrollment: United States | participants | 12 |
6. Primary Endpoint
The registry lists the primary outcome as Comparing the Detection Accuracy, Sensitivity, and Specificity of Clinically Actionable Lung Nodules, measured over 4 months.
| Element | Registry entry |
|---|---|
| Result type | Count Of Units |
| Measure of spread | not applicable |
| Unit | CT-case | nodule |
| Endpoint classification | not reduced to a single comparison |
The primary endpoint is described as comparing detection accuracy, sensitivity, and specificity for clinically actionable lung nodules between baseline reading and reading aided by ClearRead CT software. The measurement unit is CT-case | nodule, and the timeframe is 4 months. The target findings included cancers and clinically actionable benign nodules; cases without cancer or actionable nodules served as true-negative cases.
These measures address different aspects of classification. Accuracy concerns agreement with the case status overall. Sensitivity concerns identification of target findings, while specificity concerns correctly classifying cases without a target finding. Together, they help describe detection and false-positive classification rather than relying on a single summary. The brief summary also describes tabulating reader scores, constructing LROC curves, and assessing positive predictive value, negative predictive value, clinical actions, machine detection rate, and false positives per patient on normal cases.
Using these measures is relevant because a detection aid may affect both the identification of target nodules and the classification of cases without them. The registry record does not report the observed estimates, uncertainty intervals, or statistical tests for the primary endpoint. It therefore does not establish the direction or precision of any difference between reading conditions.
7. Why No Re-analysis Is Shown
8. Statistical Interpretation
The registry record states the primary objective: to test whether readers aided by ClearRead CT Insight detect actionable lung nodules better than unaided readers. It also describes a secondary objective concerning whether reading time is significantly increased with software assistance. However, the posted information does not include the observed detection estimates, a measure of uncertainty, or a reported test of the primary comparison. The stated objective should not be mistaken for evidence that the objective was achieved.
The outcome description says detection rates for accuracy, sensitivity, and specificity were computed and compared for baseline and aided studies. That description identifies the intended assessment, but it does not give the values or the result of the comparison. The record also mentions LROC curves, reader scores, positive and negative predictive values, clinical actions, machine detection rate, false positives per patient, and average reading time per case. Numerical results for these assessments are not reported in the registry record.
Without estimates and their uncertainty, it is not possible to judge the size or precision of any observed difference. A point estimate alone would describe the estimated contrast, while an interval would show the range of values compatible with the data under the analysis assumptions. Neither is available in the record. A significance statement or p-value is also not reported, so the stated superiority objective cannot be evaluated from the posted information.
The registry record identifies enrollment as 12, while the study description discusses approximately 300 retrospective cases and at least ten radiologists. It does not explain how these figures relate to each other or how the analysis accounted for multiple readers assessing cases. That distinction matters: a reader study has observations linked to both readers and cases, and those observations are not necessarily independent. No variance model or analysis procedure is reported to clarify how that dependence was handled.
The study is marked completed, but completion status does not itself convey an outcome. Likewise, the stated sample-size assumptions describe a planning-illustration framework, not a reported efficacy result. Sample-size illustrations are not observed power, and no numerical planning results are stated in the record summarized by these registry entries. In short, the record describes a question and planned measures, but it does not provide enough outcome information to estimate the benefit, harm, or precision of the software-assisted reading comparison.
9. Statistical Methods Explained
A multi-reader, multi-case study evaluates performance across readers and cases rather than treating a single reader or case as the entire study. This design can be useful for diagnostic technologies because reader skill and case difficulty both influence observed readings. General statistical consideration: results from multiple readings may be correlated when the same reader assesses several cases or cases are assessed by multiple readers. An analysis should account for the dependence relevant to its design. The registry record does not report the specific model or variance procedure used for this study.
What does randomization contribute?
Randomization is intended to reduce systematic differences between the conditions being compared by using an allocation process rather than choosing condition assignment based on expected outcomes. In a reader study, the practical meaning depends on what was randomized: readers, cases, reading sessions, or another unit. The record labels allocation randomized and the model parallel, but does not report the randomized unit, sequence generation, concealment, or allocation details. Those omissions prevent a more specific account of what the randomization protects against in this study.
What does masking mean in this record?
The registry labels the study double-masked. Masking can reduce the influence of knowledge about condition assignment on assessments or study conduct. In a software-aided reading study, implementation details are especially relevant: the record does not say who was masked, how the aided condition was concealed or presented, or whether readers could identify when software was in use. The double-masked label should therefore be reported as registered, without assuming an unreported procedure.
How should the diagnostic measures be understood?
Sensitivity describes how often target cases are identified as positive, while specificity describes how often cases without the target are identified as negative. Accuracy summarizes agreement across classifications, but can be difficult to interpret on its own when the mix of target and non-target cases affects the summary. Positive predictive value describes the proportion of positive classifications that are true target cases; negative predictive value describes the proportion of negative classifications that are true non-target cases. These predictive values depend on the case mix as well as classification performance, so estimates from a selected reader-study set need not represent performance in another case population.
The brief summary also describes LROC curves. General explanation: localization receiver operating characteristic methods consider whether a reader not only identifies a case as positive but localizes the finding, with performance examined across scoring thresholds. The registry says LROC curves were to be constructed but does not provide curve estimates or a reported comparison. Clinical action scores and suspiciousness scores provide additional descriptions of reader assessment; their scoring rules and analysis are not reported in the registry record.
Why assess false positives and reading time?
A detection system may influence both detection of target nodules and calls on cases without a target. The planned measurement of false positives per patient on normal cases is relevant to that balance. The record also lists average reading time per case as a secondary outcome. General explanation: a time comparison can describe the operational burden of a reading aid, but interpretation requires the observed times, their variability, and a defined comparison. None of those numerical results is reported in the registry record.
The endpoint timeframe is 4 months. This identifies the recorded assessment window, but the record does not explain how it relates to reading sessions, case collection, or follow-up. The listed unit is CT-case | nodule, while the brief summary describes reader-level scores and case status. How these levels were combined in the primary statistical analysis is not reported.
What can a confidence interval or p-value tell a reader?
General explanation: a confidence interval expresses uncertainty around an estimate under the assumptions of the statistical method. A narrower interval indicates greater precision in that framework; an interval spanning values compatible with materially different interpretations may leave the result uncertain. A p-value assesses how unusual the observed data would be under a specified null model and assumptions. It does not measure the size or clinical importance of an effect. The registry record does not report an estimate, confidence interval, or p-value for the primary comparison, so those tools cannot be applied to interpret the posted outcome.
The stated planning framework uses a two-sided significance level of 0.05, equal allocation between arms, and a normal approximation. It describes illustrations based on fractions of an observed effect. Such illustrations concern hypothetical planning choices, not observed power; they do not establish the precision of this completed study. The record does not report the trial's own sample-size justification, and no numerical planning result is stated.
10. Multiplicity, Missing Data and Analysis Populations
The registry describes several assessment dimensions: accuracy, sensitivity, specificity, reader scores, LROC curves, positive and negative predictive values, clinical actions, machine detection rate, false positives per patient, and average reading time per case. It lists 3 outcome measures and 2 analyses. The record does not identify which comparisons were designated as confirmatory beyond describing the primary objective, nor does it report a statistical plan for handling multiple outcomes, thresholds, readers, or analyses.
Multiplicity arises when several outcomes or statistical comparisons are examined. General explanation: evaluating many comparisons can increase the chance that at least one appears notable by chance, so a prespecified testing strategy can help readers interpret evidence. No multiplicity adjustment or hierarchy of testing is reported in the registry record. It is therefore not possible to tell whether any nominal finding, if one were reported elsewhere, would have been adjusted or treated as exploratory.
The record does not describe missing readings, excluded cases, incomplete reader assessments, or rules for handling unavailable observations. It also does not report an analysis population for each endpoint or explain whether all randomized units were included in the analyses. These details matter because exclusions or missing outcomes can affect both the estimate and its interpretation. The registry enrollment figure and the study description of retrospective cases and participating readers do not resolve the analysis population. Methods for missing-data handling and the populations actually analyzed are not reported in the registry record.
11. Safety Results
The registry reports serious adverse events as affected participants divided by participants at risk.
| Arm | Participants with serious adverse events | Participants at risk | Percentage |
|---|---|---|---|
| Arm 1 | 0 | 12 | 0.0% |
| Arm 2 | 0 | 12 | 0.0% |
Deaths
| Arm | Deaths (participants affected of at risk) |
|---|---|
| Arm 1 | 0 of 12 |
| Arm 2 | 0 of 12 |
Other (non-serious) adverse events
Participants with at least one non-serious event at or above the registry's 0% reporting threshold.
| Arm | Participants affected of at risk |
|---|---|
| Arm 1 | 0 of 12 |
| Arm 2 | 0 of 12 |
The registry lists serious adverse events for Arm 1 and Arm 2. Each arm has 12 at risk and 0 affected by serious adverse events; deaths are listed as 0 in each arm. Other adverse events are also listed as 0 affected among 12 at risk in each arm. These are the recorded figures, not estimates of comparative risk.
A zero count means that no affected events are listed in the registry for the recorded group and category. It does not establish that an event is impossible or that the risk is zero. With no event counts beyond zero, the record does not support a numerical estimate of how the arms compare in safety. The registry does not provide confidence intervals or further detail about event collection, definitions, reporting duration, or the population monitored.
The study concerns a software aid used in reading retrospective CT cases, and the safety figures should be read within that registered context. They do not describe the clinical consequences of individual detection decisions or establish safety in settings not represented by the record. The absence of reported serious events is distinct from evidence about diagnostic performance, which is also not numerically reported.
12. What the Result Does, and Does Not, Mean
The record documents a randomized, double-masked, parallel reader study evaluating ClearRead CT Insight against unaided thoracic CT interpretation. It identifies clinically actionable nodule detection as the primary question and reading time as a secondary concern. It also describes a retrospective multi-reader, multi-case approach and names diagnostic measures intended to characterize performance.
What the record does not establish is whether software assistance improved detection, how large any difference was, or how precisely it was estimated. No numerical results for accuracy, sensitivity, specificity, reader scores, LROC performance, predictive values, clinical actions, machine detection, false positives, or reading time are reported. The primary objective and planned analyses are not outcomes. Completion status likewise does not indicate success or failure.
The registry enrollment figure is 12, while the study description refers to approximately 300 cases and at least ten radiologists. Because the record does not explain how those quantities relate or identify the analysis population, they should not be used to infer the number of independent readings or the precision of an effect estimate. The statistical model and treatment of reader-case dependence are also not reported.
The listed adverse-event counts show 0 affected by serious adverse events and 0 deaths in each arm, with 12 at risk in each arm. Those registry entries describe reported counts and do not quantify comparative safety or rule out unrecorded events.
Accordingly, the defensible conclusion is about the question and design, not the efficacy of the software. The record presents a planned comparison and completion information, but it does not supply the outcome estimates needed to judge the direction, magnitude, or uncertainty of the detection results. This absence should not be interpreted as proof of no effect; it means that the result cannot be determined from the registry record.
13. Why This Trial Matters Statistically
Diagnostic reading aids are evaluated through more than a single tally of detected nodules. A reader may identify a target, localize it, assign a level of suspicion, and make a clinical-action judgment. The registry’s description of sensitivity, specificity, accuracy, LROC curves, scores, and false positives reflects these different aspects of performance. Examining them together can help distinguish detection from the tendency to call findings positive.
The multi-reader, multi-case design is statistically important because performance can vary with both the reader and the case. General explanation: evidence that accounts for these sources of variation is more informative about how results might extend beyond a particular reader-case pairing. The registry does not report how its analysis addressed these dependencies, so the planned design cannot be translated into a precision assessment.
The comparison also includes an operational measure: average reading time per case. Diagnostic performance and time are different questions. A detection comparison alone would not describe time burden, and a time comparison would not establish detection performance. The registry describes both concerns but does not report numerical results for either.
Finally, this record illustrates why a completed status and detailed endpoint descriptions are not substitutes for posted estimates. Without outcome values and uncertainty, readers cannot assess whether the observed difference was small or large, precise or uncertain, or consistent across readers and cases. The study question is relevant to statistical education precisely because the design and analysis details needed for interpretation matter, and because the registry record leaves key result and method details unreported.
14. 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 principal limitation for interpreting this trial is that numerical primary and secondary outcome results are not reported in the registry record. There are no posted estimates for the detection measures, no uncertainty intervals, and no reported significance tests. Consequently, the direction, magnitude, and precision of the comparison cannot be assessed.
The record identifies a randomized, double-masked, parallel design but does not describe the randomized unit, sequence generation, allocation procedures, or who was masked. It also does not explain how readings were sequenced, whether readers saw cases under both conditions, or how potential effects of reading order were handled. These omissions limit assessment of how the design supports a causal interpretation.
The study description refers to approximately 300 retrospective cases and at least ten radiologists, while the enrollment field lists 12. The relationship among these quantities is not reported. The record does not identify the number of analyzed cases or readers for each endpoint, nor does it describe exclusions, missing readings, or the analysis population. It also does not report how statistical analyses accounted for correlation among readings from the same reader or case.
The case set is described in terms of cancer, clinically actionable benign nodules, and normal cases. The record does not provide details sufficient to assess case selection, representativeness, or whether the case mix reflects other reading settings. Predictive values and overall accuracy can be sensitive to case composition, so generalization beyond the study set cannot be judged from the available description.
Statistical analysis details, including the model, uncertainty estimation, and handling of multiple outcomes, are not reported. The registry’s outcome descriptions and completion status do not fill those gaps. These limitations concern what can be concluded from the registry record, not proof that unreported analyses or outcomes do not exist elsewhere.
15. Related Tutorials
16. Related Statistical Calculators
17. Sources
- ClinicalTrials.gov: Evaluation of Computer-Aided Lung Nodule Detection Software in Thoracic CT for Riverain Technologies LLC — NCT02440139.
- Registry record last updated 2021-11-04; status last verified 2021-11.
- Results first posted 2021-10-04.
- 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.
Continue through the Clinical Biostats statistical pathway
Use this trial's endpoint as a starting point for deeper study of effect measures, confidence intervals, and sample-size planning.
18. Record Summary
NCT02440139 is a completed randomized reader study of ClearRead CT Insight for thoracic CT interpretation. The registry lists double masking and a parallel model. Its primary question concerns detection of clinically actionable lung nodules with aided versus unaided reading; average reading time per case is listed as a secondary outcome.
The study description characterizes the design as retrospective and multi-reader, multi-case. It discusses approximately 300 CT cases, including approximately 100 true-positive cases and 200 normal cases, and at least ten radiologists. The enrollment field lists 12. The registry does not explain how these figures relate.
The recorded study start is 2015-04, with primary completion in 2017-09. The sponsor is Virginia Polytechnic Institute and State University. The record describes diagnostic and operational assessments but does not report their numerical results, statistical uncertainty, or detailed analysis methods. It lists 0 serious adverse events and 0 deaths in each arm, with 12 at risk in each arm.