Testing an Implementation Science Tool to Increase Cervical Cancer Screening in Mombasa, Kenya: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Testing an Implementation Science Tool to Increase Cervical Cancer Screening in Mombasa, Kenya (NCT03514459), a randomized na study. The primary outcome is Cervical Cancer Screening.

2. Clinical Question

The trial addresses an implementation question: whether a clinic-level approach to examining screening and referral processes is associated with a difference in screening prevalence. The primary outcome is the number of eligible clients screened in relation to all eligible clients, aggregated across intervention and control clinics. It is not a measure of an individual participant’s change over time.

The registry reports a prevalence ratio of 1.84, with a 95% confidence interval from 1.54 to 2.20 and p < 0.001, for the reported comparison. These values describe the relative screening prevalence estimated by the sponsor’s analysis. The registry record does not report the underlying clinic-level counts or a detailed account of how the comparison was modeled beyond identifying Poisson regression.

Because clinics, rather than individual clients, were randomized, the result is a comparison between clinic groups. It should not be read as if each client were independently randomized. The finding describes the outcome recorded in the participating clinics during the study period; it does not, by itself, establish how the intervention would perform in other settings or under different implementation conditions.

3. Trial Design

This completed study used a randomized, parallel-group design in 20 family planning clinics in Mombasa County, Kenya. The clinic was the unit of randomization. The registry lists masking as none, so the study was open-label. The sponsor was the University of Washington. The recorded start date is 2020-01-29, and the primary completion date is 2021-07-31.

Clinics assigned to the intervention were introduced by study staff to the five steps of SAIA. The intervention description says that clinic-level cascade analysis was used to identify drop-offs in screening and referrals, with an Excel-based tool adapted from previous SAIA trials. The registry record does not provide a complete account of each step’s implementation or the extent to which clinics followed the approach.

Control clinics continued usual procedures. Their cervical cancer screening rates were evaluated every 3 months using family planning register data. The registry describes the outcome as aggregated data across 18 months. It also states that individual clients were not enrolled in the trial and were not followed over time. Thus, the people counted in the outcome are not a cohort of individually randomized or individually followed trial participants.

The study summary describes additional aims concerning the screening care cascade, correlates of failure to screen, and costs and budget impact. The registry record reports one outcome measure and one analysis. Results for the additional aims are not reported in the registry record. Details about allocation concealment, the randomization sequence, and the number of clinics assigned to each arm are not reported in the registry record.

This design is suited to evaluating a clinic-level implementation approach, but it also means the analysis must account for the fact that outcomes from clients within the same clinic may be related. The registry identifies the sponsor’s model as Poisson regression; further details about how the clustered design was handled 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.

Before assignment: Family planning clinics were the unit of randomization and whole clinics were randomized to intervention arm or control arm. Individual participants were not enrolled. Outcomes about women were aggregated across each clinic and not all individual level data including age were collected as we were using paper registry records with identifiable information redacted.

Overall Study

Analysis population for the primary outcome, as defined in the registry: These are aggregated participants in intervention and control clinics. Individual participants were not enrolled in the trial as this was a cluster randomized trial. Results are aggregated across all clinics and presented here but these are not RCT participants as they were not enrolled.

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: No individual participants were enrolled in the trial.

6. Primary Endpoint

The registry lists the primary outcome as Cervical Cancer Screening, measured over Aggregated data across 18 months, individual participants were not followed over time..

The primary endpoint is cervical cancer screening among family planning clients aged 21-65 years. It is defined as the proportion of all eligible clients who were screened: the number screened relative to the total number eligible. The outcome is reported as individuals screened, but the analysis uses aggregated clinic data rather than measurements from individually enrolled participants.

This endpoint is used to assess whether screening occurred among the intended group of eligible clinic clients. It focuses on the implementation outcome of reaching clients with screening, rather than on diagnosis, treatment, referral completion, or a health outcome after screening. The registry’s description of the wider project mentions follow-up referrals and treatment pathways, but those are not the primary outcome reported in this record.

The timeframe is aggregated data across 18 months, and individual participants were not followed over time. Accordingly, the estimate compares screening prevalence across the clinic groups over the stated period. It does not show whether particular clients were screened at multiple visits, how long individual clients waited, or whether a person’s screening status changed. Those details are not reported in the registry record.

The registry does not give the underlying numerator and denominator counts, clinic-specific results, or a breakdown by client characteristics. Without those details, readers cannot assess variation in screening across clinics or subgroups from the reported outcome alone. The stated endpoint nevertheless provides a direct measure of the central implementation question: how much of the eligible family planning clinic population was screened during the study period.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The registry reports Poisson regression with a prevalence ratio as the effect measure. The reported estimate is 1.84, with a 95% confidence interval of 1.54 to 2.20 and p < 0.001. The analysis is marked as a superiority comparison. The registry does not report model covariates, variance-estimation details, or a fuller statistical analysis plan.

What does a prevalence ratio express?

In general, a prevalence ratio compares the prevalence of an outcome between groups. A value of 1 represents equal prevalence. A value above 1 means the prevalence in the numerator group is higher than in the reference group, while a value below 1 means it is lower. The interpretation depends on which group is placed in the numerator and which is used as the reference. The registry’s reported comparison is for the SAIA and control arms, with an estimate of 1.84.

A prevalence ratio is a relative measure. It does not directly state the percentage-point difference between groups. To interpret absolute screening levels, readers would need the group-specific proportions or counts; these are not reported in the registry record. The ratio also does not describe the number of additional clients screened without knowing the underlying eligible population and group-specific screening prevalence.

How should the confidence interval be read?

In general, a confidence interval gives a range of values compatible with the observed data and the statistical model at a stated confidence level. The reported 95% confidence interval extends from 1.54 to 2.20. It expresses uncertainty about the estimated relative prevalence, rather than uncertainty about each client’s outcome. The interval does not include 1, the value corresponding to equal prevalence under the ratio scale. The interval’s width reflects estimation uncertainty, but does not by itself account for unreported design or model assumptions.

What does the p-value add?

In general, a p-value measures how incompatible the observed data are with a specified null hypothesis, assuming that hypothesis and the model are correct. Here, the reported value is p < 0.001. It is not the probability that the intervention has no effect, and it does not quantify the size or importance of the difference. It also does not replace the confidence interval, which communicates the estimate’s precision on the prevalence-ratio scale.

Why does clinic randomization matter?

In general, cluster randomization assigns groups such as clinics rather than assigning each person separately. Outcomes among people who attend the same clinic can be correlated because they share staff, procedures, records, and local conditions. Treating every person as an independent observation can make uncertainty appear smaller than it is. Appropriate analysis needs to reflect the randomization unit and the dependence within clusters.

This trial randomized 20 clinics, while its primary outcome is based on aggregated clients. The registry identifies Poisson regression as the method but does not report whether the model used cluster-robust standard errors, a random effect, a correction for a small number of clusters, or another approach to account for clustering. These are examples of possible general methods, not claims about the method used in this trial. The registry’s statistical detail is insufficient to identify the exact implementation.

Why is an individual-level re-analysis not appropriate?

The outcome population consists of aggregated clients, not individually enrolled trial participants. The intervention was assigned to clinics, not clients. A re-analysis that treated the aggregated clients as independently randomized observations would disregard how assignment occurred and could misstate precision. The registry explicitly notes that a simple individual-level re-analysis would be misleading, and only the sponsor’s own analysis is shown.

General explanations of regression do not establish which variables were included or how the model was specified in this trial. The registry record does not report an adjusted covariate set, missing-data procedures, sensitivity analyses, or diagnostics. Consequently, the reported ratio and interval can be described, but the details needed to reproduce or fully evaluate the fitted model are not available in the registry record.

10. Analyses Reported by the Sponsor

The sponsor’s analysis is the analysis reported for the primary comparison: Poisson regression estimating a prevalence ratio of 1.84, with a 95% confidence interval of 1.54 to 2.20 and p < 0.001. The registry does not present a separate simple recalculation from summary statistics. It specifically notes that a simple individual-level re-analysis would be misleading for this cluster-randomized design, so only the sponsor’s own analysis is shown.

This distinction matters because randomization occurred at the clinic level, whereas the outcome counts eligible clients in aggregated clinic data. A calculation that treated all clients as individually randomized could fail to reflect within-clinic dependence and would not reproduce the trial’s allocation structure. The sponsor’s estimate is therefore the relevant reported statistical result, although the registry record does not provide enough model detail to independently assess how clustering or other modeling choices were handled.

The result can be interpreted on its reported scale as a relative prevalence comparison. The record does not report group-specific screening proportions or counts, so an absolute difference cannot be derived from the reported estimate alone. The estimate and its uncertainty should be read as the sponsor’s model-based result, with the limitations of the available methodological description kept in view.

11. Multiplicity, Missing Data and Analysis Populations

The registry reports one outcome measure and one analysis. It does not report whether the broader study aims or other analyses were included in a prespecified multiplicity plan, or whether any adjustment for multiple testing was planned or used. The absence of a reported plan does not establish that no plan existed; these details are not reported in the registry record.

In general, multiplicity arises when several outcomes, time points, subgroups, or statistical comparisons are tested. If many tests are considered, the chance of at least one apparently unusual result can increase. Whether adjustment is appropriate depends on the study questions and analysis plan. The record’s single listed primary analysis does not describe all analyses that may have been performed for the study’s broader aims, so the full multiplicity context cannot be assessed.

Missing-data methods are not reported in the registry record. The outcome is based on aggregated family planning register data, but the record does not state how incomplete, unavailable, or inconsistent register entries were handled. It also does not report whether data completeness differed between intervention and control clinics. This limits assessment of the potential influence of missing records on the prevalence estimate.

The analysis population is described as aggregated clients in intervention and control clinics; individual participants were not enrolled and were not followed over time. The record does not provide a detailed analysis-population definition beyond eligible family planning clients aged 21-65 years, nor does it report exclusions, clinic-level inclusion criteria, or handling of clinics with incomplete data. It is therefore not possible to determine from the registry record whether every eligible client or clinic contributed to the reported analysis.

The registry identifies the analysis as Poisson regression but does not report adjustment variables, model diagnostics, sensitivity analyses, or the handling of the cluster structure in detail. These omissions are relevant to interpretation of precision and robustness. They do not change the value reported, but they restrict independent evaluation of how the estimate was produced.

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 registry’s adverse-event entries report zero serious adverse events in both the control and SAIA groups, and zero deaths in both groups. Each group is listed with 10 at risk. The group labels state that no individual participants were enrolled. These entries should therefore not be read as a conventional participant-level safety comparison without clarification of what the at-risk counts represent.

The primary outcome concerns aggregated screening data, and the registry explains that individual clients were not enrolled as trial participants. The adverse-event figures do not identify individual event descriptions, follow-up completeness, or how the at-risk counts relate to the clinic-level design. The registry record does not report further detail to resolve those questions.

Zero reported serious events means that the record lists none in these entries; it does not establish that there was no possibility of harm or that all events were captured. The figures also do not provide a basis for comparing safety between groups when the underlying unit and ascertainment details are not described. They should be interpreted as registry-reported counts with important limits on their clinical meaning.

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

The reported result is evidence of a difference in screening prevalence between the clinic groups in the trial’s aggregated 18-month data. The estimated prevalence ratio is 1.84, with a 95% confidence interval from 1.54 to 2.20 and p < 0.001. The registry identifies the result as a superiority analysis. These statistics characterize the sponsor’s comparison; they do not by themselves explain the mechanisms behind the observed difference.

The outcome is a clinic-level implementation measure based on the proportion of eligible family planning clients screened. It is not a measure of individual clients’ health outcomes, the accuracy of screening, referral completion, treatment, or longer-term consequences. The registry does not report those outcomes in the primary analysis. Nor does it report group-specific proportions, so the relative estimate cannot be translated from the record into an absolute difference in screening.

The finding applies to the participating clinics and the data collected during the stated period. Clinic randomization strengthens the basis for comparing the assigned clinic groups, but the registry does not provide enough statistical detail to determine exactly how the analysis accounted for clustering. The number of clinics is 20; clients within a clinic may share conditions and therefore should not be treated as independently randomized.

The result does not establish that the same relative difference would occur in other clinics or settings. The record does not provide information on implementation fidelity, clinic-by-clinic variation, or external validation. It also does not report results for the study’s other aims concerning the care cascade and costs. Those questions cannot be answered from the primary estimate.

The appropriate reading is therefore specific: the sponsor’s analysis reports a relative difference in aggregated screening prevalence for the randomized clinic groups, with statistical uncertainty described by the confidence interval. The estimate should not be converted into a claim about individual benefit or used to infer outcomes that the primary endpoint did not measure.

14. Why This Trial Matters Statistically

This trial illustrates an important distinction in biostatistics: the unit assigned to an intervention may differ from the people whose outcomes are counted. Clinics were randomized, while the primary outcome was compiled from family planning clients in those clinics. That structure requires attention to clustering and makes a simple individual-level re-analysis unsuitable.

It also shows why the effect measure must be matched to the question. The reported prevalence ratio compares screening prevalence across clinic groups. It communicates a relative contrast, but not the absolute number or proportion of clients screened in either group. Without the group-specific values, readers cannot assess the absolute scale of the result from the ratio alone.

The registry’s reporting also demonstrates the value of distinguishing a model result from its supporting detail. The estimate, confidence interval, p-value, and regression method are reported, but the model specification and handling of clinic-level dependence are not. Readers can interpret the stated statistical result while recognizing that they cannot fully reproduce the analysis from the registry record.

Finally, the outcome reflects an implementation process rather than an individually measured clinical endpoint. Aggregated register data can address whether screening was recorded among eligible clients, but cannot show individual trajectories or later outcomes. Keeping the design, endpoint, effect measure, and reporting limits aligned is essential to interpreting the finding without extending it beyond what was measured.

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 has important limits for interpreting the result. It reports a clinic-randomized design and a Poisson regression analysis, but does not describe the model specification in enough detail to determine how within-clinic correlation was handled. That matters because assignment occurred at the clinic level and outcomes were aggregated from clients attending those clinics.

Group-specific screened counts and eligible-client denominators are not reported. The relative prevalence estimate therefore cannot be paired with absolute screening proportions using the registry record alone. The record also does not provide clinic-specific outcomes, so variation across the 20 clinics cannot be assessed from the reported result.

The outcome is based on aggregated register data across 18 months, and individual clients were not enrolled or followed. The record does not describe missing or incomplete register entries, data validation, or whether data availability differed across arms. It also does not report implementation fidelity or the extent to which clinics used the intervention components.

Details about allocation concealment, the randomization sequence, covariate adjustment, sensitivity analyses, and the analysis population are not reported in the registry record. The record lists one primary outcome and one analysis but does not report a multiplicity plan. It describes broader aims concerning the care cascade and costs, but results for those aims are not reported in the registry record.

These limitations do not negate the reported estimate, but they constrain independent assessment of its precision, robustness, and broader applicability. The result is most directly informative about the aggregated screening comparison in the participating clinics, under the sponsor’s reported analysis.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Testing an Implementation Science Tool to Increase Cervical Cancer Screening in Mombasa, Kenya — NCT03514459.
• Registry record last updated 2024-05-22; status last verified 2024-05.
• Results first posted 2024-05-22.
• 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

NCT03514459 is a completed randomized, parallel-group trial titled “Testing an Implementation Science Tool to Increase Cervical Cancer Screening in Mombasa, Kenya.” The registry lists the sponsor as the University of Washington, masking as none, and phase as NA. The study started on 2020-01-29, with primary completion on 2021-07-31.

The trial randomized 20 family planning clinics. Its primary outcome was the proportion of eligible clients aged 21-65 years screened for cervical cancer, measured from aggregated data across 18 months. Individual clients were not enrolled or followed over time.

The reported analysis used Poisson regression and estimated a prevalence ratio of 1.84, with a 95% confidence interval of 1.54 to 2.20 and p < 0.001. The registry reports one outcome measure and one analysis. It does not report detailed model specifications, group-specific screening counts, or results for the project’s other stated aims.