De-Implementing Opioids for Dental Extractions: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for De-Implementing Opioids for Dental Extractions (NCT03584789), a randomized na study. The primary outcome is Opioid Prescribing at the Extraction Encounter.

2. Clinical Question

The trial addressed changes in opioid prescribing practice after dental extractions. Its overarching goal was to reduce reliance on opioid analgesics and increase reliance on non-opioid analgesics for postoperative pain management. The comparison therefore concerned strategies intended to inform clinician and patient decisions, rather than a direct comparison of pain medicines assigned to participants.

The registry reports two mixed-model odds ratios for the primary outcome. The first is 1.29, with a 97.5% confidence interval from .93 to 1.80 and p = .08. The second is 1.24, with a 97.5% confidence interval from .86 to 1.79 and p = .18. The registry does not state the reference-group coding or provide the event counts for the primary outcome by arm in the listed results. Those omissions limit how specifically the odds ratios can be translated into arm-by-arm prescribing patterns.

3. Trial Design

This was a randomized, parallel-group trial with three study arms and no masking. The registry describes the design as a prospective, provider-level, three-arm cluster randomized trial. In a cluster-randomized study, groups of people connected through a provider or setting are assigned together, rather than each individual being independently assigned. The provider-level assignment is important when interpreting the analysis because participants treated by the same provider may have more similar outcomes than participants treated by different providers.

The arms were standard practice, clinical decision support, and clinical decision support plus education. Standard practice was usual care with the dental provider. The clinical decision support strategy was intended to assist dentists in pain-management decisions without resorting to opioids when a non-opioid alternative would suffice. The education strategy addressed pain management before and after dental extraction, including information about the risks and benefits of different strategies. The full description of the decision-support content is not reported in the registry record.

The registry lists tooth extraction as the condition and 5722 as the enrollment. The study status is completed. The primary outcome was assessed at the extraction encounter. Other listed outcomes concerned exclusive non-opioid prescribing or recommendations, shared decision-making, and patient experiences of post-extraction pain.

Cluster assignment affects both design and analysis. Outcomes among participants treated by one provider may be correlated, so an analysis that treats every participant as fully independent could understate uncertainty. The registry reports a mixed-model analysis for the primary outcome. The number of providers or clusters, the cluster allocation details, and the extent of within-provider correlation 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: 73 participants declined use of their EHR data

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: 73 participants declined use of their EHR data

6. Primary Endpoint

The registry lists the primary outcome as Opioid Prescribing at the Extraction Encounter, measured over At the extraction encounter.

The primary endpoint was the number of participants prescribed an opioid at the extraction encounter. It is a binary outcome at the participant level: each participant either was or was not prescribed an opioid at that encounter. The registry names the outcome as opioid prescribing and specifies that the assessment occurred at the extraction encounter.

This endpoint directly addresses the trial’s focus on prescribing practice. It captures whether an opioid prescription was made at the encounter, rather than the amount of medication prescribed, whether it was filled, or whether a participant used it. Those other measures are not reported as the primary endpoint. The registry’s short outcome description does not indicate whether prescriptions issued outside that encounter were included.

For a binary outcome, a count describes how many participants experienced the event, while an odds ratio compares the odds of the event between groups. The registry reports the primary analyses as odds ratios from mixed models. It does not list the arm-specific primary-outcome counts or percentages in the reported analysis details. Without those counts and the reference-group coding, the odds ratios alone do not establish the absolute frequency of opioid prescribing in each arm.

Using prescribing at the encounter makes the outcome closely aligned with the clinical process targeted by the strategies. It does not, by itself, measure postoperative pain, pain relief, medication use, or the balance of benefits and harms for individual participants. Those outcomes require separate measures.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The primary outcome is binary: a participant was either prescribed an opioid at the extraction encounter or was not. The registry describes the endpoint as a participant count and reports odds ratios from mixed models. An odds ratio is a relative measure of odds, not a difference in probabilities. Odds and probabilities are related but are not interchangeable, so an odds ratio should not be read as a percentage-point change or as a direct statement about how many participants were prescribed an opioid.

What does the odds ratio compare?

An odds ratio compares the odds of an event in one group with the odds in another. A value of 1 corresponds to equal odds for the two groups under the specified comparison. Values above or below 1 indicate different odds according to which group is in the numerator. The registry reports values of 1.29 and 1.24, but the analysis description does not identify the reference group or explicitly state the ordering of the groups in each contrast. As a result, the numerical values should be reported as recorded without assigning a direction to a specific arm.

The registry reports 97.5% confidence intervals. A confidence interval reflects both the estimated contrast and its uncertainty under the model and interval-construction method. A wider interval means the estimate is less precise than an otherwise comparable narrower interval. The first reported interval runs from .93 to 1.80; the second runs from .86 to 1.79. Both include 1, the equal-odds value. This does not prove there is no difference. It means that the interval estimates include equality as well as values on either side of it.

How should the p-values be read?

The registry gives p = .08 for one comparison and p = .18 for the other. In general, a p-value describes the compatibility of the observed data with a specified null hypothesis, conditional on the model and its assumptions. It is not a measure of effect size, a probability that a strategy is effective, or the probability that the observed estimate is due to chance. The record does not report a prespecified threshold for interpreting these tests, so the p-values should not be converted into a definitive success-or-failure label.

The confidence intervals and p-values are related summaries of uncertainty, but they are not substitutes for one another. Their interpretation depends on the model, comparison, and inferential procedure. The registry reports 97.5% intervals, but the reason for that confidence level and its relationship to any multiplicity adjustment are not reported in the registry record.

Why does clustering matter?

Participants treated by the same provider can have correlated outcomes because they share a clinical environment and provider practices. Cluster randomization recognizes that the assignment occurs at the provider level. A mixed model can account for the grouping structure, depending on its specification. If clustering were ignored, an analysis could treat related observations as more independent than they are and consequently misstate uncertainty. The registry specifically cautions that a simple individual-level re-analysis would be misleading and presents only the sponsor’s analysis.

The record does not report the number of providers, the cluster sizes, the estimated degree of within-provider correlation, or the precise mixed-model structure. Without those details, the analysis can be identified as a mixed-model analysis, but its implementation cannot be fully assessed from the registry record.

What is not established by the model label?

The phrase mixed models does not, on its own, specify which covariates were included, how repeated or grouped observations were represented, or how uncertainty was calculated. These details are not reported in the registry record. Nor does the label establish whether the analysis followed an intention-to-treat principle or another analysis population. The record identifies the main outcome and reports the model-based comparisons, but it does not give a full statistical analysis plan.

General interpretation should therefore distinguish the reported estimates from details that are unavailable. The odds ratios and their intervals describe the sponsor’s model-based contrasts as reported. They do not supply arm-specific absolute risks, a causal mechanism, or a complete account of the model assumptions. The cluster design makes the sponsor’s analysis more relevant than a simple participant-level recalculation would be, but the available record does not permit an independent reconstruction of that model.

10. Exact Estimates for Each Arm

No two-arm comparison is shown for this trial, but each arm's result can still be summarized with exact interval estimates computed from the posted counts. The Clopper-Pearson interval is exact and conservative; the Wilson interval is a common closer-to-nominal alternative; the Bayesian credible interval uses a uniform Beta(1, 1) prior. For small groups the three can differ noticeably.

R code and output

arm    <- c("Standard Practice", "Clinical Decision Support", "Clinical Decision Support + Education")
n      <- c(1720, 2269, 1660)
events <- c(443, 643, 418)
for (i in seq_along(n)) {
  ci <- binom.test(events[i], n[i])$conf.int          # Clopper-Pearson exact interval
  cr <- qbeta(c(0.025, 0.975), 1 + events[i], 1 + n[i] - events[i])   # Bayesian, uniform Beta(1,1) prior
  wi <- (function(x, n) { p <- x / n; z <- qnorm(0.975); d <- 1 + z^2 / n
    c((p + z^2 / (2 * n) - z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d, (p + z^2 / (2 * n) + z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d) })(events[i], n[i])
  cat(sprintf("ARM|%s|%d|%d|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f\n", arm[i], n[i], events[i], 100 * events[i] / n[i],
              100 * ci[1], 100 * ci[2], 100 * wi[1], 100 * wi[2], 100 * cr[1], 100 * cr[2]))
}

ARM  Standard Practice  1720  443  25.8  23.7  27.9  23.7  27.9  23.7  27.9
ARM  Clinical Decision Support  2269  643  28.3  26.5  30.2  26.5  30.2  26.5  30.2
ARM  Clinical Decision Support + Education  1660  418  25.2  23.1  27.3  23.2  27.3  23.2  27.3

11. Analyses Reported by the Sponsor

The registry reports the sponsor’s mixed-model analyses and specifically notes that a simple individual-level re-analysis would be misleading because the trial was cluster randomized at the provider level. No recalculated primary result is presented. The reported odds ratios, confidence intervals, and p-values should therefore be read as the sponsor’s own analysis, not as estimates independently reproduced from arm-level summary counts.

That distinction matters because a calculation based only on participant counts could ignore the grouping of participants by provider. Such a calculation would not reproduce the intended handling of clustering and could give an inappropriate account of uncertainty. The mixed-model estimates are the relevant reported results for the design described in the registry.

The record does not provide enough model detail or primary-outcome counts by arm to verify the estimates independently. It also does not specify the reference-group coding for the reported contrasts. There is no stated discrepancy between a sponsor analysis and a recalculation: a recalculation is not shown. The sponsor’s reported analysis is therefore the only primary statistical analysis available in this record, with the limitations in reporting noted above.

12. Multiplicity, Missing Data and Analysis Populations

The registry lists a primary outcome and several secondary outcomes, and reports two analyses for the primary outcome. It does not report a multiplicity adjustment, a testing hierarchy, or a prespecified rule for interpreting multiple comparisons. It also does not explain why the confidence intervals are 97.5%. Without those details, the p-values should be treated as individual reported results rather than as a complete account of family-wise or other multiplicity-controlled inference. The record does not report whether the two primary comparisons were planned as co-primary contrasts or how they were intended to relate to the secondary outcomes.

For the primary outcome, the registry notes that 73 participants declined use of their EHR data. It does not report how these participants were distributed across arms, whether their outcomes were otherwise available, or how their absence from EHR data affected the analysis. The exact number included in each primary analysis and the handling of any other missing outcome data are not reported in the registry record. The enrollment count of 5722 should not be assumed to equal the analysis denominator.

The record also does not report whether the analysis used an intention-to-treat population, a per-protocol population, or another defined group. It does not describe imputation, missing-data assumptions, sensitivity analyses, or exclusions beyond the note about participants who declined EHR data use. These omissions matter because the interpretation of a model-based comparison depends partly on who contributed data and how missing outcomes were handled.

Interim analyses, stopping rules, and a detailed statistical analysis plan are not reported in the registry record. This does not establish that such procedures were absent; it means they cannot be described from this record. In the same way, the registry does not state whether the listed comparisons were selected before outcome data were reviewed. The available information supports a cautious reading of the reported analyses but not a complete evaluation of multiplicity or missing-data assumptions.

13. 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 in zero participants in each arm. For other adverse events, it reports zero in standard practice, one in clinical decision support, and zero in clinical decision support plus education. The corresponding numbers at risk are 1720, 2269, and 1660. No deaths are reported in any arm, and no leading serious adverse-event terms are listed.

These figures are counts, not comparative safety estimates. The registry does not provide an adverse-event definition, follow-up window, severity breakdown beyond the serious-versus-other distinction, or details of how events were collected. With no serious events listed, the record does not support a detailed comparison of serious-event risks across arms. A count of zero in the posted results should be read as no events reported in that category, not as proof that an event is impossible or that the strategies have equivalent safety.

The adverse-event figures also do not describe the primary prescribing outcome or patient experiences of post-extraction pain. They should be interpreted separately from the mixed-model efficacy-related outcome, and the available safety reporting is limited to the counts and group sizes recorded.

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

The primary result addresses whether opioid prescribing at the extraction encounter differed across trial arms under the sponsor’s mixed-model analysis. The registry reports two odds ratios, 1.29 and 1.24, with 97.5% confidence intervals from .93 to 1.80 and from .86 to 1.79, respectively. The corresponding p-values are .08 and .18. Both intervals include the equal-odds value of 1, and the registry does not specify the reference-group coding needed to assign a direction to each estimate.

These results do not establish that the study arms were identical. A confidence interval that includes 1 can be compatible with equality while also allowing differences in either direction. The intervals span a range of values, so the estimates are not highly precise about the comparative odds. The p-values do not measure the probability that a particular strategy did or did not affect prescribing, and they should not be treated as a verdict on clinical importance.

The endpoint records whether an opioid was prescribed at the encounter. It does not measure whether a prescription was filled or used, the amount prescribed, postoperative pain, pain relief, or participants’ subsequent medication choices. Although the trial also listed outcomes related to non-opioid prescribing, shared decision-making, and patient experiences of pain, results for those outcomes are not included in the primary-analysis details.

The record supports a description of the trial design, the primary endpoint, the sponsor’s model-based estimates, and the reported adverse-event counts. It does not give enough information to reconstruct absolute prescribing rates by arm or to assess all model and missing-data assumptions. The most defensible conclusion is therefore limited: the reported estimates are uncertain comparisons of odds from a provider-level cluster-randomized trial, and the registry entries do not justify stronger claims about the direction, size, or practical importance of differences.

15. Why This Trial Matters Statistically

This trial is statistically notable because the unit of randomization was the provider, while the primary outcome was measured for participants. That structure creates a distinction between the unit assigned to an intervention and the unit on which outcomes are counted. Participants treated by the same provider may share aspects of care, so their outcomes may not be statistically independent. Analyses need to reflect the grouping induced by the design.

The registry’s warning against a simple individual-level re-analysis illustrates a general principle: a large participant count does not by itself resolve concerns about clustering. The number of independent assignment units and the within-cluster similarity can affect the precision of estimates. Those design details are not reported in the registry record, but the use of mixed models and the sponsor’s analysis are explicitly identified.

The trial also highlights the difference between relative and absolute results. The registry reports odds ratios, but not arm-specific primary-outcome counts in the listed analyses. Without the reference-group coding and baseline event frequencies, the relative estimates cannot be translated into absolute prescribing patterns from this record. Both forms of information are valuable: relative measures compare groups, while absolute counts help readers understand event frequency.

Finally, the reported confidence intervals and p-values demonstrate why statistical interpretation should consider uncertainty, model structure, and reporting completeness together. The estimates are not precise enough to settle the direction of the contrasts as described, and important details about multiplicity, missing data, and model specification are not reported. This makes the trial a useful example of how design-aware analysis and transparent reporting shape what can responsibly be concluded from registry results.

16. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The registry record does not provide the primary-outcome counts or event frequencies by arm, the reference-group coding for the odds ratios, or the exact analysis denominators. These omissions prevent a direct reading of absolute prescribing patterns and limit independent verification of the reported contrasts. The 73 participants who declined EHR data use are noted, but their distribution by arm and the handling of their outcomes are not reported in the registry record.

Important design and model details are also absent. The number of providers or clusters, cluster sizes, the degree of within-provider correlation, covariate adjustments, and the specific mixed-model structure are not reported in the registry record. Because assignment occurred at the provider level, these details are particularly relevant to assessing the precision and assumptions of the estimates.

The record does not report a statistical analysis plan, multiplicity procedures, interim analyses, stopping rules, or the reasoning behind the 97.5% confidence intervals. It also does not identify the analysis population or describe how missing data beyond the EHR-use note were handled. The absence of these details from the record should not be taken to mean that the procedures did not exist.

For the intervention descriptions, the clinical decision-support text is incomplete in the registry record. Further details about implementation fidelity or how consistently the strategies were used are not reported. The adverse-event record gives counts but not detailed definitions, collection procedures, or follow-up information. These limitations constrain conclusions about both the primary comparison and the broader experience of participants.

17. Related Tutorials

18. Related Statistical Calculators

19. Sources

• ClinicalTrials.gov: De-Implementing Opioids for Dental Extractions — NCT03584789.
• Registry record last updated 2022-08-30; status last verified 2022-05.
• Results first posted 2022-07-19.
• 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.

20. Record Summary

NCT03584789, titled De-Implementing Opioids for Dental Extractions, is a completed, randomized, parallel-group, provider-level cluster trial with no masking. The registry lists enrollment of 5722 and three arms: standard practice, clinical decision support, and clinical decision support plus education. Its primary outcome was the number of participants prescribed an opioid at the extraction encounter.

The sponsor reports two mixed-model odds ratios: 1.29, with a 97.5% confidence interval from .93 to 1.80 and p = .08; and 1.24, with a 97.5% confidence interval from .86 to 1.79 and p = .18. The registry cautions that a simple individual-level re-analysis would be misleading because of the cluster-randomized design. The reference-group coding and arm-specific primary-outcome counts are not reported in the registry record.

Serious adverse events and deaths are reported as zero in all arms; one other adverse event is reported in the clinical decision support arm. Further details on missing data, multiplicity, analysis populations, and model specification are not reported in the registry record.