Project PROTECT: Protecting Nursing Homes From Infections and Hospitalization: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Project PROTECT: Protecting Nursing Homes From Infections and Hospitalization (NCT03118232), a randomized phase 4 study. The primary outcome is Hospital Transfers Due to Infection.

2. Clinical Question

Project PROTECT is recorded as a randomized, parallel, phase 4 trial without masking. Its stated aim also included assessing antibiotic utilization and multi-drug resistant organism prevalence, but the primary result concerns hospital transfers due to infection. All hospital transfers is listed as a secondary outcome. These distinctions matter: a result for the primary endpoint does not, by itself, report effects on the other outcomes or establish how those outcomes were analyzed.

3. Trial Design

Project PROTECT compared nursing homes assigned to decolonization with nursing homes assigned to routine bathing. The registry describes the study as cluster-randomized, meaning assignment occurred at the nursing-home level rather than independently for each resident. Residents within one facility may share staff, routines, and care environments. Their outcomes can therefore be more alike than outcomes from residents in different facilities. Analyses that account for this clustering are important when estimating uncertainty.

The recorded design is randomized, parallel, and unmasked. Parallel means that the two assigned groups were followed as separate groups rather than crossing over between approaches. Unmasked means masking is recorded as none; the registry does not describe further details about who knew each facility’s assignment or how that might have affected outcome recording. The analysis population for the primary endpoint is described as residents, accounting for clustering in 28 nursing homes. The record does not state whether this is the total number of facilities in the trial.

The trial’s recorded enrollment is 13952. Its status is completed, with a start date of 2017-04-03 and a primary completion date of 2023-10-10. Those dates describe the registry record and do not, by themselves, explain how long each resident was followed. The primary endpoint timeframe is listed as 18 months.

The decolonization arm is described as using topical antiseptic products, while the comparison arm received routine bathing according to facility protocol. The registry’s summary refers to routine chlorhexidine bathing and periodic nasal antiseptic use. It does not report enough operational detail to establish how often each product was used, how adherence was assessed, or whether all facilities followed identical procedures. Randomization can support a comparison between assigned groups, but interpretation also depends on how implementation and outcomes were handled. Those details 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

Analysis population for the primary outcome, as defined in the registry: Analysis is of residents accounting for clustering in 28 nursing homes

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 Hospital Transfers Due to Infection, measured over 18 months.

The primary endpoint is titled “Hospital Transfers Due to Infection.” Its description defines the measure as the probability that a transfer to a hospital is due to an infection, and its unit is transfers. The assessment timeframe is 18 months. This is not simply a count of all hospital transfers: the outcome concerns transfers attributed to infection. All hospital transfers is listed separately as a secondary outcome.

This kind of endpoint is used to capture a clinically consequential event in the nursing-home setting. Transfers due to infection can reflect the occurrence of infection as well as decisions about when hospital care is needed. Consequently, the measure represents an outcome shaped by both residents’ health and the processes used to assess and transfer them. The registry does not describe the criteria for attributing a transfer to infection, the source of that classification, or any adjudication process. Those features affect how consistently the endpoint may have been measured across facilities.

The primary result is reported as a difference in risk ratio from a mixed models analysis. The registry record does not specify the order of the two analysis groups, the units or transformation used for the reported value, or a fuller mathematical definition of the contrast. It is therefore possible to report the estimate and its interval as recorded, but not to interpret the sign as favoring a particular arm or to translate the value into a percentage-point difference. The endpoint is relevant to the trial’s stated purpose, but interpretation should remain tied to this specific measure and its reporting limitations.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The registry identifies the primary analysis as a mixed models analysis and describes the reported parameter as a difference in risk ratio. Understanding why those labels matter begins with the trial’s assignment structure. Facilities, rather than individual residents, were randomized. Residents within one nursing home share aspects of their care environment, so their outcomes may be correlated. A model that treats every resident as fully independent can understate uncertainty when such correlation is present.

Why account for clustering?

In a cluster-randomized trial, the assigned intervention is delivered at the cluster level. Here, the clusters are nursing homes. The number of residents enrolled does not alone determine how much independent information the trial contains: residents sharing a home can contribute overlapping information about the assigned facility-level approach. A cluster-aware analysis accounts for this dependence when estimating effects and uncertainty. The primary population description explicitly says that the analysis accounts for clustering in 28 nursing homes. The registry does not report the number of residents in each home or the degree of within-home correlation.

Mixed models are a broad family of statistical models that can represent both overall group differences and variation associated with clusters or other grouping structures. Depending on the specification, a model can include facility-level random effects, fixed effects, or other terms. The registry provides the method name but not the model formula, link function, covariates, estimation procedure, or diagnostics. It would therefore be inappropriate to claim which particular mixed model was fitted or what assumptions its implementation made.

What does a risk-ratio contrast represent?

A risk ratio compares risks between groups. A difference in risk ratios would compare risk-ratio quantities, but the registry’s label alone does not identify the exact formula, scale, or group ordering used for this result. The result is 16.6 with a 95% confidence interval from 11.0 to 21.8. Because units and contrast order are not specified, the estimate should not be translated into a percentage-point change, a relative reduction, or a count of transfers. The endpoint description concerns the probability that a hospital transfer is due to infection, but that description does not resolve the model’s reporting scale.

How should the confidence interval be read?

A confidence interval expresses uncertainty about an estimated parameter under the statistical model and its assumptions. A 95% confidence interval is generated by a procedure that, over repeated studies conducted under the same conditions, would capture the target parameter at the stated rate. For an individual study, it is not a guarantee that the interval contains the true value. The reported interval’s width reflects statistical precision on the stated scale, but the clinical meaning of that width depends on knowing the scale and contrast definition. Those are not fully explained in the registry record.

The p-value is reported as p < 0.001. In general, a p-value measures how unusual data at least as inconsistent with a specified null hypothesis would be if that hypothesis and the analysis assumptions were correct. It does not give the probability that the null hypothesis is true, establish causation on its own, or quantify clinical importance. The null hypothesis and test details are not reported in the registry record, so the p-value should be interpreted alongside the sponsor’s stated model and estimate rather than treated as a standalone conclusion.

Why can over-dispersion matter?

For count or event data, a basic model may assume a particular relationship between the mean and variance. If observed variability is greater than that assumption permits, the data are over-dispersed relative to the model. Ignoring excess variability can make standard errors too small and intervals too narrow. Cluster dependence can also create extra variability compared with an individual-level independence assumption. These are general statistical considerations; the registry does not report a dispersion estimate or state whether over-dispersion was assessed for this analysis.

The endpoint is described as probability-based, while the result parameter is described as a difference in risk ratio and the outcome unit is transfers. The record does not provide enough methodological detail to determine the exact data structure or link between the underlying observations and reported parameter. No further characterization of the outcome model should be inferred from those labels alone.

What cannot be reconstructed from the record?

The registry does not report the full analysis plan, model equation, adjustment variables, group order, handling of missing outcome data, or sensitivity analyses. It also does not report whether interim analyses occurred, how the primary analysis population was defined beyond residents accounting for clustering in 28 nursing homes, or the sample-size justification. Without these details, an independent analyst cannot reproduce the sponsor’s model from the reported summary alone. General descriptions of mixed models explain the method family, but they cannot fill in trial-specific choices absent from the record.

Because assignment was clustered, a simple re-analysis that treats residents as independently randomized would be misleading. The sponsor’s own analysis is the relevant reported analysis for the primary result. The statistical labels, estimate, confidence interval, and p-value can be described accurately, but a more detailed assessment requires model and data information not reported in the registry record.

10. Analyses Reported by the Sponsor

The registry reports the sponsor’s result as a mixed models analysis accounting for the cluster-randomized design. The estimate is labeled “Difference in Risk Ratio” and is 16.6, with a 95% confidence interval from 11.0 to 21.8 and p < 0.001. This is the analysis to use when describing the primary result in the registry.

No simple recalculated estimate is reported alongside it. For this trial, a resident-level re-analysis that ignores the fact that nursing homes were randomized would not be an appropriate check: residents in the same home can have correlated outcomes, and the trial assignment occurred at the facility level. The registry specifically identifies the primary analysis as accounting for clustering in 28 nursing homes. A valid comparison or re-analysis would need to preserve that structure and have sufficient information to reconstruct the model.

The result’s scale and group ordering are not specified in enough detail to translate the estimate into an arm-specific direction or an intuitive percentage. The arm mapping for the reported analysis groups is also not stated. Accordingly, the sponsor’s numerical result can be reported as recorded, but it cannot be converted into a statement that decolonization increased or decreased the endpoint relative to routine bathing from this record alone. The sponsor’s analysis takes precedence over any naïve individual-level calculation because that calculation would omit the trial’s clustering.

11. Multiplicity, Missing Data and Analysis Populations

The registry lists six outcome measures and two analyses. It identifies hospital transfers due to infection as the primary outcome and all hospital transfers as a secondary outcome. The broader trial summary also mentions antibiotic utilization and multi-drug resistant organism prevalence. The record does not give a complete account of how all outcomes and analyses were related or prioritized beyond those labels.

When a trial examines multiple outcomes, time points, subgroups, or model specifications, the chance of at least one apparently notable result can increase. Multiplicity procedures can address this by defining a primary comparison or adjusting inference across a family of tests. The registry record does not report whether any multiplicity adjustment was planned or used, which outcomes formed a testing family, or whether the reported p-value was adjusted. It should therefore be read as the p-value reported for the stated analysis, not assumed to reflect a particular correction strategy.

Missing-data handling is not reported in the registry record. The entry does not state how unavailable observations were treated, whether follow-up differed between arms, or whether sensitivity analyses examined missingness. These details matter because incomplete outcome ascertainment can affect estimates, especially when reasons for missingness are associated with facility assignment or resident outcomes.

The primary population description says the analysis is of residents accounting for clustering in 28 nursing homes. It does not provide a fuller definition of the analysis population, such as all randomized residents, residents with available outcome information, or another group. The registry record also does not report the treatment of facilities or residents who may not have contributed data to the analysis.

Interim analyses and stopping rules are not reported in the registry record. Nor is the statistical analysis plan described in enough detail to determine whether the reported analysis followed prespecified choices. These omissions do not invalidate the reported result, but they restrict assessment of selective reporting, multiplicity, and robustness. The result should be interpreted as the reported primary analysis, with these areas of uncertainty kept distinct from the numerical estimate itself.

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.

Most frequent non-serious adverse event terms

The registry reports serious adverse events and deaths as zero in both arms. For the decolonization group, the serious-event population at risk is 7388 and serious events affected 0; deaths are 0. For the routine-bathing group, the serious-event population at risk is 6564 and serious events affected 0; deaths are 0. The registry’s serious adverse event terms list is empty.

These figures describe the recorded serious-event counts in the stated at-risk groups. They do not mean that every possible adverse effect was assessed or that no non-serious events occurred. The registry also lists other adverse events: 35 affected participants in the decolonization group and 0 in the routine-bathing group, with other-event populations at risk of 7388 and 6564, respectively. The record does not provide event terms for those other events in the information, so their nature and clinical significance cannot be characterized.

Zero recorded serious events in both groups is a descriptive finding, not proof that serious events are impossible or that the approaches have identical safety profiles. The registry does not report the event-collection procedures, observation duration for safety outcomes, or further detail about how events were classified. These counts should be read as registry-reported safety information, separate from the primary effectiveness endpoint.

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

The registry result establishes what the sponsor reported for the primary endpoint: a mixed models estimate labeled “Difference in Risk Ratio” of 16.6, with a 95% confidence interval from 11.0 to 21.8 and p < 0.001. The analysis description says it accounts for clustering in 28 nursing homes. This supports a precise description of the reported statistical finding, while leaving important interpretation questions unresolved.

The result does not, from the registry record alone, show which arm had the higher endpoint value. The analysis groups are not mapped to the arm names in the reported contrast, and the contrast ordering is not stated. The units and exact mathematical definition of the reported parameter are also not given. Thus the estimate should not be reframed as a percentage, percentage-point difference, relative reduction, or number of hospital transfers. The confidence interval and p-value do not remedy those reporting gaps.

The result is specific to hospital transfers due to infection over the primary endpoint timeframe of 18 months. It cannot automatically be generalized to all hospital transfers, antibiotic utilization, or multi-drug resistant organism prevalence. Those outcomes are distinct, and the registry identifies all hospital transfers as secondary. Results for the other outcomes are not included in this primary analysis description.

Random assignment at the nursing-home level is relevant to interpreting a comparison of assigned groups, but the result is still dependent on the trial’s outcome definitions, implementation, and statistical model. The registry does not report enough detail to assess every assumption or to reproduce the analysis independently. The data also do not establish that all nursing homes or residents would experience the same outcome pattern.

Statistical significance is not equivalent to clinical importance. The p-value describes compatibility with a null hypothesis under specified assumptions; it does not state whether the observed contrast is large enough to matter in practice. Since the scale and group direction are not clear, the practical magnitude cannot be judged from the estimate alone. The findings should therefore be conveyed as the recorded sponsor analysis, with uncertainty about direction, scale, and broader outcomes made explicit.

14. Why This Trial Matters Statistically

This trial illustrates why the unit of randomization should guide statistical analysis. Project PROTECT randomized nursing homes, not individual residents. Residents within a facility can share staffing, routines, and environmental exposures, creating dependence in their outcomes. Treating them as independent observations would fail to represent the design and could misstate precision. The registry’s description of a mixed models analysis accounting for clustering is therefore a central part of how the primary result should be read.

It also shows that a result’s label is not always enough to make it clinically interpretable. The reported estimate, confidence interval, and p-value are available, but the registry does not state the ordering of the contrast groups or clearly define the reported scale. An estimate can be statistically precise on its own scale while remaining difficult to translate for readers if the scale and direction are not explicit. Transparent reporting of these details is important alongside reporting uncertainty.

The endpoint itself combines an event with a care decision: a hospital transfer attributed to infection. Such measures can reflect both health events and clinical or operational practices. In a nursing-home trial, understanding the criteria used to classify infection-related transfers would help readers assess consistency across facilities. Those criteria are not reported in the registry record.

Finally, this example separates statistical evidence from broader claims. The primary analysis concerns one outcome, while the trial’s summary describes additional objectives. A p-value for the primary result cannot stand in for evidence about every outcome, nor can it resolve missing details about multiplicity, missing data, or the model. Careful reading preserves the value of the reported analysis without extending it beyond what the registry states.

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 reporting gaps limit interpretation. The registry identifies the primary endpoint and timeframe, but does not describe how a transfer was classified as due to infection, who made that determination, or whether classification was reviewed consistently across nursing homes. Because the endpoint involves a transfer decision and attribution to infection, those details would help assess how comparable outcome measurement was across facilities.

The reported contrast is labeled as a difference in risk ratio, but its scale, formula, and group ordering are not fully specified. The analysis group identifiers are not mapped to decolonization and routine bathing in the reported result. This prevents a reliable arm-specific directional interpretation and makes the estimate difficult to translate into an intuitive measure. The confidence interval and p-value are reported, but they do not supply the missing definition.

The record names mixed models analysis and says the primary analysis accounted for clustering in 28 nursing homes, but does not report the model specification, assumptions, covariates, or diagnostic checks. The number of residents contributing to each facility-level estimate is not stated. The record also does not report adherence, implementation variation, missing-data methods, or sensitivity analyses.

The registry does not report the sample-size justification, an interim-analysis plan, a detailed multiplicity strategy, or a full statistical analysis plan. It lists six outcome measures and two analyses, but does not describe the complete relationship among them. These omissions restrict assessment of how analysis choices were prespecified and how uncertainty across outcomes was handled.

Finally, the primary result does not provide findings for the trial’s other stated aims, including antibiotic utilization and multi-drug resistant organism prevalence. The safety information is limited to event counts and at-risk populations; details about event terms and collection procedures are not reported for all categories. These are limitations of the information available in the registry record, rather than evidence that the analyses or trial procedures were necessarily inadequate.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Project PROTECT: Protecting Nursing Homes From Infections and Hospitalization — NCT03118232.
• Registry record last updated 2023-12-27; status last verified 2023-12.
• Results first posted 2023-12-27.
• 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

Project PROTECT is a completed phase 4, randomized, parallel trial with no masking recorded. Nursing homes were assigned to decolonization using topical antiseptic products or routine bathing per facility protocol. The registry records enrollment of 13952, a start date of 2017-04-03, and a primary completion date of 2023-10-10.

The primary outcome was hospital transfers due to infection over 18 months. The primary analysis is described as a mixed models analysis accounting for clustering in 28 nursing homes. The reported difference in risk ratio is 16.6, with a 95% confidence interval from 11.0 to 21.8 and p < 0.001. The record does not state the contrast ordering or scale in enough detail to assign the direction to an arm or translate the estimate into a percentage.

Serious adverse events and deaths were recorded as zero in both arms. Other adverse events affected 35 participants in the decolonization group and 0 in the routine-bathing group. Missing-data methods, multiplicity procedures, a detailed statistical plan, and the complete definition of the reported contrast are not reported in the registry record.