Comparative Effectiveness Trail to Reduce Child Maltreatment, Improve Client Outcomes and Examine Client Burden: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Comparative Effectiveness Trail to Reduce Child Maltreatment, Improve Client Outcomes and Examine Client Burden (NCT02549287), a randomized na study. The primary outcome is Parenting Young Children Scale-Supporting Positive Behavior Sub-scale.

2. Clinical Question

The registry describes a study of SafeCare, an evidence-based home visiting program, compared with Supportive Case Management, described as child welfare services as usual. Its primary outcome focused on parents’ supporting-positive-behavior skills, measured with a sub-scale of the Parenting Young Children Scale.

The broader study aims went beyond this quantitative comparison. They included identifying barriers and facilitators to implementation, and understanding parent and provider reactions to services. The record describes qualitative interviews, focus groups, session-by-session ratings, and organizational environment surveys as part of those aims. The primary result should therefore be read as one component of a mixed-methods study, not as a complete account of implementation or family experience.

3. Trial Design

The registry lists this as a randomized, parallel-group study with no masking and an enrollment of 289. It is marked completed. The study started in 2015-09, and primary completion was recorded in 2018-05. The condition listed is child maltreatment. The record identifies SafeCare and Supportive Case Management as the two interventions and describes Supportive Case Management as services as usual.

A design note in the registry identifies the study as cluster-randomized. That matters for statistical interpretation: outcomes from people in the same cluster may be correlated, so an analysis that treats every person as statistically independent could misrepresent uncertainty. The record’s analysis is a marginal model, and the registry note says a simple individual-level re-analysis would be misleading. The sponsor’s analysis is therefore the relevant reported inferential result; no independent individual-level re-analysis is presented.

The primary measure was assessed at baseline, after families were invited into the study during a provider visit, and at follow-up approximately 6-months later. The record also describes interviews with families at two time points, session-by-session family ratings of service reactions, provider ratings of family engagement, and provider organizational environment surveys at two time points.

The registry does not report the cluster unit, the method used to generate or conceal the allocation sequence, or details of how clustering was represented in the model. Masking is listed as none. The phase is listed as NA, and the study type is categorized as other. Further protocol details, including a sample-size justification and interim-analysis plan, are not reported in the registry record.

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: Data for four surveys were corrupted causing the Row population to differ from the Overall population. In addition, Some participants missed or refused to answer questions.

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 Parenting Young Children Scale-Supporting Positive Behavior Sub-scale, measured over Families were assessed at two time points: 1) after being invited into the study during a visit with their provider (Baseline) and, 2) approximately 6-months later (follow-up)..

The primary endpoint was the Supporting Positive Behavior sub-scale of the Parenting Young Children Scale. It assesses skills such as noticing and praising a child’s good behavior. The sub-scale contains 7 items; its score is the mean of those items on a 7-point scale, with higher scores representing a higher degree of positive parenting skills. The theoretical and actual ranges of means are both reported as 1-7.

Because the distribution was heavily skewed toward the positive end, the record says the measure was dichotomized using an approximate median split. Participants scoring 6 or below were classified as low, and those scoring higher than 6 were classified as high. The analysis therefore concerns classification into these two categories rather than a comparison of mean scale scores or changes in the full 7-point measure.

A dichotomized endpoint can make a comparison easier to express as odds of being in one category rather than the other. However, it also discards distinctions among scores within each category. For example, the analysis does not retain all the information about how far apart individual scores are. The registry does not report a rationale for choosing this cut point beyond the description of the approximate median split and the skew toward the positive end.

The endpoint was assessed at baseline and approximately 6-months later. The reported analysis does not specify in the registry record how the repeated assessments were represented in the model or which assessment contrast the estimate summarizes. That uncertainty should be kept in mind when interpreting the result.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The primary result is described as a marginal model, with generalized estimating equations (GEE) given as an example. In general, a marginal model estimates an average relationship for a population rather than a subject-specific relationship. GEE methods can account for correlation among observations, such as repeated observations or observations grouped within clusters, when the correlation structure and variance calculation are appropriately specified. The registry does not report the specific working correlation, variance estimator, or detailed model specification.

What does “odds ratio, log” mean?

The modeled outcome had two categories: high and low supporting-positive-behavior skills. Odds compare the chance of an outcome category with the chance of the other category. An odds ratio contrasts those odds between groups; reporting it on the log scale means the estimate and confidence limits are expressed after taking a logarithm. On that scale, zero is the null value. The reported estimate is 0.55, and its 95% confidence interval runs from -0.11 to 1.20. Since this interval crosses zero, it includes the null on the scale used in the registry.

The record does not map the analysis group codes to the named intervention arms. Consequently, the estimate’s direction cannot be described as favoring a named arm based only on the registry entry. The reported p-value is 0.12. It reflects the test associated with the sponsor’s model; it is not a measure of the probability that the null hypothesis is true, nor does it quantify the importance of the finding.

Why does the cluster design matter?

In general, cluster randomization assigns groups rather than treating each participant as an independently randomized unit. People within a cluster may share experiences or conditions, making their outcomes more alike than outcomes from unrelated clusters. When this correlation is ignored, standard errors and confidence intervals can be misleading. The registry specifically warns that a simple individual-level re-analysis would be misleading and says only the sponsor’s own analysis is shown. The marginal-model description is consistent with an approach that can be used for correlated data, but the registry does not provide enough detail to evaluate precisely how clustering was handled.

The study also assessed families at baseline and follow-up. Repeated measures from the same family can be correlated. In general, a marginal model can handle correlated measurements, but the registry does not say whether the reported model included both assessments, focused on follow-up, or estimated a particular time contrast. The result should not be described as a quantified change from baseline unless that analysis is specified.

What does dichotomization do?

The scale’s mean score was reduced to a high-or-low classification, with scores of 6 or below categorized as low and scores higher than 6 as high. The record gives the reason as a distribution heavily skewed toward the positive end and describes the split as approximate. This allows an odds-based comparison, but it removes score detail within each category. People with different scores can be grouped together, and people near the cut point can be placed in different categories. The reported estimate therefore concerns the binary classification, not the full distribution of scale scores.

General statistical interpretation also distinguishes a test result from the size and precision of an effect. A p-value summarizes compatibility between the data and a null model under specified assumptions; it does not convey all uncertainty or practical importance. The confidence interval is useful because it displays a range of effect values compatible with the fitted model and data. Here, the interval includes the null value, so the analysis does not clearly distinguish the comparison from no difference on the modeled scale.

How should planning calculations be read?

The registry describes a planning illustration based on the observed effect, with a two-sided significance level of 0.05, equal allocation between arms, and a normal approximation. It says the target effect is a stated fraction of the observed effect, and that non-time-to-event illustrations are expressed as participants per arm. These are planning assumptions, not observations from the trial and not “observed power.” The registry does not report a design justification or enough detail to assess whether the original enrollment target was based on those assumptions. A planning calculation does not alter the interpretation of the trial’s estimated effect or its confidence interval.

Finally, outcome and analysis counts matter when reading a result among multiple study questions. The record lists 13 outcome measures and 13 analyses, but does not describe any multiplicity adjustment, prespecification details, or a complete statistical analysis plan. Those missing details limit what can be concluded about the role of this test among all analyses.

10. Analyses Reported by the Sponsor

The registry identifies the reported estimate as the sponsor’s marginal-model analysis and cautions that a simple individual-level re-analysis would be misleading because the study was cluster-randomized. For that reason, the sponsor’s analysis is the appropriate result to use for statistical interpretation in this record.

No independent recalculated estimate is reported in the registry record. There is therefore no recalculation to compare numerically with the sponsor’s result. The posted sponsor result is 0.55 on the log odds-ratio scale, with a 95% confidence interval from -0.11 to 1.20 and p = 0.12. The interval includes zero, the null value on that scale.

The record does not provide enough detail to reconstruct the model or verify its handling of clusters and repeated assessments. It also does not map the analysis group codes to the named arms. The reported value should therefore be read as the sponsor’s model-based comparison, without assigning its direction to SafeCare or Supportive Case Management and without presenting an individual-level recalculation as a substitute.

11. Multiplicity, Missing Data and Analysis Populations

The record lists 13 outcome measures and 13 analyses. It does not report whether the primary outcome was designated in a protocol or analysis plan before results were examined, how the analyses were prioritized, or whether any multiplicity adjustment was used. The p-value of 0.12 is therefore the reported value for this particular analysis; the record does not establish how it relates to a family-wise or other error-control procedure across the full set of analyses.

Missingness is also relevant. The record says data for four surveys were corrupted, causing the row population to differ from the overall population. It also says some participants missed or refused to answer questions. The registry does not report the missing-data method, the extent of missingness for this primary endpoint, or whether the analysis used an intention-to-treat population, a complete-case population, or another analysis population. These details are not reported in the registry record.

The primary scale was assessed at baseline and approximately 6-months later. The record does not state how incomplete repeated assessments were handled or whether missingness differed between arms. Without those details, it is not possible to determine whether the reported comparison may be sensitive to assumptions about the missing responses.

The cluster-randomized design raises a separate analytic consideration: the number and size of clusters and the treatment of within-cluster correlation can affect precision. The registry warns against simple individual-level re-analysis, but does not report the cluster count, the cluster-level allocation details, or the specific variance method. This limits independent assessment of the reported uncertainty. These omissions do not change the reported estimate; they define what cannot be evaluated from the registry record alone.

12. 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 adverse-event record lists zero serious adverse events and zero other adverse events for both SafeCare and Supportive Case Management. It also records zero deaths in both arms. The at-risk counts are reported separately for serious and other adverse events: 193 for SafeCare and 96 for Supportive Case Management in each category.

These entries describe what was recorded in the registry. They do not establish that no adverse experiences occurred outside the events captured, reported, or categorized for this record. The registry does not provide serious adverse-event terms, and no top serious-event terms are listed. It also does not describe the event-collection procedures or provide a narrative about adverse events.

The study concerns services and parenting outcomes, and the adverse-event counts should be interpreted within that context, without treating zero recorded events as proof of absence of risk. The registry’s figures are counts, not a comparison adjusted for the cluster-randomized design or for differences in time under observation. No safety inference beyond the recorded counts is warranted from these entries alone.

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

The posted result quantifies the sponsor’s marginal-model comparison for a binary classification of supporting-positive-behavior skills. The reported log odds-ratio estimate is 0.55, with a 95% confidence interval from -0.11 to 1.20 and p = 0.12. Because the interval crosses the null value of zero on the log scale, the data do not clearly distinguish the comparison from no difference under the reported analysis.

This does not demonstrate equivalence, and it does not establish that the interventions have the same effect. A result that does not clearly distinguish a comparison from the null may reflect uncertainty, and the confidence interval includes estimates on both sides of zero. The record does not report an equivalence or non-inferiority design or margin, so those interpretations are not supported.

The result also cannot be translated into a claim that one named arm had higher odds than the other, because the registry does not connect the analysis group codes to the named arms. Nor is it a direct estimate of improvement in mean scale scores: the analyzed outcome categorized participants as high or low using the stated cut point. The record does not specify the exact contrast across baseline and follow-up represented by the estimate.

Beyond the primary outcome, the study sought to understand implementation barriers and facilitators, parent and provider reactions, burden, engagement, satisfaction, perceived impact, and organizational environment. The primary model result does not answer all of those questions. Qualitative and other quantitative measures are described, but their results are not included in the registry entries available for this interpretation.

The findings are also bounded by missing or corrupted survey data and by unreported details of the analysis population and model specification. The cluster-randomized design makes the sponsor’s analysis important to preserve; a naive individual-level calculation would not be an appropriate substitute. The record supports a cautious description of the reported estimate and its uncertainty, not a broad conclusion about the overall effectiveness of either service.

14. Why This Trial Matters Statistically

This study raises several statistical issues that are common in applied clinical and social-service research. First, cluster randomization changes the unit of assignment and the structure of the data. Outcomes from people within a cluster may be correlated, so analysis must reflect that dependence. The registry explicitly cautions that a simple individual-level re-analysis would be misleading, underscoring why design details matter when interpreting a reported estimate.

Second, the primary measure was converted from a scale to a binary category because the distribution was heavily skewed toward the positive end. This makes an odds-based comparison possible, but it changes the question: the analysis concerns crossing a cut point, not the full range of parenting-skills scores. The choice and consequences of dichotomization deserve attention in any interpretation of the result.

Third, the result is accompanied by a confidence interval that crosses the null on the reported log scale. Reading the interval alongside the p-value helps distinguish a lack of clear evidence from evidence of equivalence. The record does not report an equivalence framework, and the estimate should not be used to claim that the arms are the same.

Finally, the study’s broader mixed-methods aims matter. Implementation barriers, service reactions, engagement, and perceived burden are not interchangeable with the primary endpoint. A single model estimate cannot summarize all these dimensions. The registry lists multiple outcomes and analyses, while leaving multiplicity handling, missing-data methods, and several model details unreported. Taken together, these features make transparent reporting of the estimand, analysis population, cluster adjustment, and outcome definition especially important.

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.

Several trial-specific limitations are apparent from the registry record. The primary endpoint was dichotomized at the stated threshold after the measure was described as heavily skewed toward the positive end. The resulting analysis does not preserve differences among scores within the high and low categories. The record does not report a sensitivity analysis using the full scale or alternative cut points.

The study is identified as cluster-randomized, but the cluster unit, number of clusters, allocation details, and model specification are not reported in the registry record. The record cautions against a simple individual-level re-analysis, yet does not give enough detail to assess the sponsor’s treatment of clustering. It also does not make clear which time contrast the reported estimate represents, despite assessments at baseline and approximately 6-months later.

Data for four surveys were corrupted, and some participants missed or refused questions. The record does not report the extent or pattern of missingness for the primary outcome, the analysis population, or the method used to handle incomplete data. The mapping from analysis group codes to the named arms is also not reported, preventing an interpretation of the estimate’s direction by intervention name.

The registry lists 13 outcome measures and 13 analyses but does not report a multiplicity strategy or the full statistical analysis plan. It does not report the sample-size justification or interim-analysis plan either. These gaps limit assessment of the planned status of the endpoint and the precision procedures. They do not negate the reported result, but they constrain the conclusions that can be drawn from the registry entry.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Comparative Effectiveness Trail to Reduce Child Maltreatment, Improve Client Outcomes and Examine Client Burden — NCT02549287.
• Registry record last updated 2020-01-13; status last verified 2019-12.
• Results first posted 2020-01-13.
• 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

The registry describes a completed, randomized, parallel-group study with no masking and enrollment of 289. It compared SafeCare, an evidence-based home visiting program, with Supportive Case Management, described as child welfare services as usual. The study is identified as cluster-randomized, and the registry cautions against a simple individual-level re-analysis.

The primary endpoint was the high-versus-low classification on the Parenting Young Children Scale-Supporting Positive Behavior sub-scale. The sponsor’s marginal-model result was a log odds ratio of 0.55, with a 95% confidence interval from -0.11 to 1.20 and p = 0.12. The interval includes the null value on the reported scale, and the analysis group codes are not mapped to named arms.

The broader study included implementation and family-response aims. The registry notes corrupted survey data and missed or refused questions. Details about missing-data handling, multiplicity, the analysis population, and several aspects of the statistical model are not reported in the registry record.