High Deductible Health Plans and Bipolar Disorder: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for High Deductible Health Plans and Bipolar Disorder (NCT02560701), a registered phase not stated study. The primary outcome is Inpatient Hospitalizations Among Bipolar Patients.

2. Clinical Question

3. Trial Design

The registry record describes an investigation of the impact of switching into high-deductible health plans on outcomes for patients with bipolar disorder. Its brief summary says the investigators use eleven years (2004-2014) of claims data from the largest US commercial health insurer, and also describes interviews with patients and caregivers recruited through a major advocacy group. The record lists Harvard Pilgrim Health Care as sponsor and gives the study status as COMPLETED.

The registry enrollment is 350823. That figure is the recorded enrollment; it should not automatically be read as the number of people contributing to every endpoint or to each comparison group. The record does not explain the relationship between enrollment, the claims-data population, the interview participants, and the endpoint-specific populations.

The primary endpoint is inpatient hospitalizations among bipolar patients, measured as a mean number per year at Year 3. Secondary outcomes concern access to outpatient services for bipolar disorder and medication adherence for psychotropic medications. The record lists six outcome measures, but the posted results include 0 analyses. The intervention entry names Patient Interview, while the study description centers on HDHP switching. How the interview component and the claims-based comparison fit together is not fully described.

Allocation is not stated. The registry explicitly cautions that this is not a randomized comparison and that the arms are not a like-for-like control comparison. Thus, the group labels alone do not show that participants had comparable characteristics or comparable chances of entering each group. Details of eligibility, exposure definition, follow-up procedures, covariate adjustment, and the statistical 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.

Before assignment: Study participants without either a) two years of continuous enrollment in a low-deductible health plan OR b) one year of enrollment in a low-deductible plan followed by a continuous year of enrollment in a high-deductible plan were excluded. This brought the sample size from 350,823 to 97,302.

Overall Study

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

6. Primary Endpoint

The registry lists the primary outcome as Inpatient Hospitalizations Among Bipolar Patients, measured over Year 3.

The primary endpoint is the mean number of annual inpatient hospitalizations among bipolar patients in Year 3. Its unit is hospitalizations per year, its summary measure is a mean, and the registry identifies standard deviation as the measure of dispersion. The description refers to the mean number of annual inpatient hospitalizations in the follow-up period.

Hospitalization frequency is a measure of inpatient service use. In a study of health-plan design, such a measure can describe whether care use differs between groups, but it does not by itself explain why a difference might occur. For example, the endpoint alone does not distinguish changes in access, underlying health, care-seeking, or other influences. Those are general interpretive considerations, not findings reported by this registry.

Using a mean gives a group-level average, while the standard deviation describes variation among observations around that average. For count outcomes such as hospitalizations, individual values are nonnegative and may be concentrated at low counts with a smaller number of larger counts. That general feature can make the mean sensitive to unusually high values. The registry record does not provide the actual means, standard deviations, sample sizes for the endpoint, or a confidence interval. Consequently, the magnitude and precision of the primary endpoint cannot be assessed from the posted results.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The registry identifies the primary endpoint as a continuous measure summarized by a mean and standard deviation. It does not report the analysis methods used to compare the groups. The explanations below are general statistical context, not a description of analyses confirmed in the registry.

What does the mean describe?

For a continuous outcome, the mean is the sum of observed values divided by the number of observations. For this endpoint, each value represents annual inpatient hospitalizations in the specified follow-up period. The mean describes average utilization in the analyzed group. It does not show how many people had no hospitalizations, how many had repeated hospitalizations, or how values are distributed across individuals. Different distributions can have the same mean.

The standard deviation describes how dispersed individual values are around the group mean. It is not the same as the uncertainty in the estimated mean. A standard deviation reflects variation among observations; uncertainty in a mean estimate also depends on the amount of usable data and on the analysis design. The registry supplies no primary endpoint standard deviations, group means, or confidence intervals, so these features cannot be evaluated for this comparison.

Why does the comparison design matter?

In a randomized comparison, allocation can help balance measured and unmeasured characteristics across groups on average. The registry explicitly says allocation is not stated and that the arms are not a like-for-like control comparison. For a nonrandomized exposure, differences in patient characteristics, health status, plan selection, or other factors may be related both to group membership and to hospitalization use. This is general confounding context; the registry does not identify which factors were measured or whether any adjustment was made.

An adjusted regression model can be used in some observational studies to account for measured covariates. But an adjusted result is only interpretable with information about the model, included covariates, assumptions, and the target contrast. None of those analysis details are reported in the registry record. It would therefore be inappropriate to describe a particular adjusted estimate or to assume that confounding was controlled.

How can count outcomes behave?

Hospitalization counts are nonnegative, discrete observations even though the registry classifies the endpoint summary as a mean of a continuous measure. General methods for counts may include models that account for the count distribution. Over-dispersion is a general condition in which variability is greater than a model would expect under a simpler count assumption; if ignored, standard errors and statistical tests can be poorly calibrated. Whether the endpoint showed over-dispersion, and whether any count-specific model was used, are not reported in the registry record. The classification as a mean does not reveal which model, if any, generated a comparison.

What does a confidence interval contribute?

In general, a confidence interval expresses uncertainty around an estimated quantity under a stated statistical procedure. A narrow interval indicates greater statistical precision than a wide interval, while both the estimate and interval are needed to understand the range of effects compatible with the analysis. A confidence interval is not a guarantee that the true value lies within its limits. No confidence interval for the primary endpoint is reported in the registry record, so the precision of the comparison cannot be described.

How should planning illustrations be read?

Sample-size calculations are planning tools that depend on assumptions about the effect to detect, variability, significance level, allocation, and analysis method. The registry description of the planning assumptions specifies a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation, with the target effect expressed as a fraction of the observed effect. Such illustrations do not establish that the study was designed using those assumptions, nor do they establish the probability that a result is true. They are not observed power. The registry does not report the trial's own sample-size justification, and the numerical primary effect needed to contextualize an observed-effect-based illustration is not reported in the registry record.

The record gives 0 analyses. It therefore does not identify a reported estimation procedure, model diagnostics, covariate set, or hypothesis test for the primary endpoint. General statistical principles can explain how such methods work, but they cannot fill in a missing analysis plan or result.

10. Confidence Intervals for Each Arm

No two-arm comparison is shown for this trial, but each arm's mean can be given with a t-based 95% confidence interval computed from the posted mean, spread and number of participants.

R code and output

arm <- c("High-Deductible Health Plan Group", "Control Group")
n   <- c(3517, 37776)
m   <- c(0.17, 0.17)
sp  <- c(0.6, 0.57)
sd  <- sp      # posted standard deviations
for (i in seq_along(n)) {
  se <- sd[i] / sqrt(n[i]); h <- qt(0.975, n[i] - 1) * se
  cat(sprintf("CARM|%s|%d|%.3f|%.3f|%.3f|%.3f\n", arm[i], n[i], m[i], sd[i], m[i] - h, m[i] + h))
}

CARM  High-Deductible Health Plan Group  3517  0.170  0.600  0.150  0.190
CARM  Control Group  37776  0.170  0.570  0.164  0.176

11. Multiplicity, Missing Data and Analysis Populations

The registry lists six outcome measures, including the primary endpoint and secondary outcomes related to outpatient service access and medication adherence. When a study evaluates multiple outcomes or subgroup comparisons, the chance of at least one apparently unusual result can increase if each is tested separately without accounting for multiplicity. General approaches include prespecifying a primary outcome, controlling a familywise error rate, or using other declared procedures. The registry record does not report a multiplicity strategy, the number of planned tests, or whether any outcome hierarchy was used. It also reports 0 analyses, so there are no listed outcome-specific estimates or tests to assess.

Missing-data handling is also not reported in the registry record. The record does not describe missing claims, incomplete follow-up, eligibility for the Year 3 measure, exclusions, or how participants with incomplete information were handled. It is therefore not possible to determine whether the endpoint analysis used complete cases, imputation, weighting, or another approach. These methods can lead to different interpretations when missingness is related to health or group membership; that is a general statistical concern, not a claim that a particular pattern occurred in this study.

The analysis population is not defined for the primary endpoint. The overall enrollment is 350823, but the registry does not state the endpoint-specific denominator, the number assigned to each analysis group, or the handling of people who changed plans or lacked follow-up data. Allocation is not stated, and the record warns that the comparison groups are not like-for-like. Without definitions of the eligible, exposed, comparison, and analyzed populations, readers cannot assess selection into the analysis or the exact population represented by any estimate.

These omissions limit reproducibility and make it difficult to connect the intended study question to a specific statistical contrast. They do not establish that missingness, multiplicity, or population selection caused bias; they mean the registry record does not report enough detail to evaluate those possibilities.

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 reports serious adverse event and other adverse event counts separately for the High-Deductible Health Plan Group and the Control Group. For the High-Deductible Health Plan Group, 3517 are listed as at risk for each category; the number affected is 0 for serious events and 0 for other events, and deaths are 0. For the Control Group, 37776 are listed as at risk for each category; the number affected is 0 for serious events and 0 for other events, and deaths are 0. No top serious-event terms are listed.

These entries are counts recorded in the registry, not a complete account of all possible safety information. The registry does not describe the method for collecting or coding events, the duration of safety observation, or whether the at-risk populations are comparable. The group sizes differ, and the record cautions that the groups are not a like-for-like control comparison. The count entries should not be converted into an inference about comparative safety without appropriate denominators, follow-up details, and an analysis plan.

Zero recorded events means the registry lists 0 affected participants in these categories. It does not establish that no events occurred outside the reported data or that the study could rule out uncommon events. The adverse-event counts are separate from the primary hospitalization endpoint and do not provide its missing results.

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

The registry record documents a study addressing whether switching into high-deductible health plans is associated with outcomes among patients with bipolar disorder. It identifies annual inpatient hospitalizations in Year 3 as the primary endpoint and names outpatient access and psychotropic medication adherence as secondary outcomes. These details establish what the investigators intended to examine, but they do not show the observed results.

The record does not report group means for inpatient hospitalizations, an estimated comparison, or uncertainty around an estimate. As a result, it cannot show whether hospitalization frequency differed, how large any difference was, or how precisely it was estimated. The absence of a reported numerical result is not evidence of no difference. It is a limit on what can be concluded from this registry record.

The design description also constrains interpretation. Allocation is not stated, and the registry cautions that the groups are not like-for-like controls. Even if a numerical difference were available, a nonrandomized group contrast would not by itself establish that plan design caused the difference. General considerations include differences between people entering each plan, differences in underlying health, and other influences on care use. The registry does not report whether these issues were addressed analytically.

The overall enrollment of 350823 should not be mistaken for an endpoint-specific analysis size. The record does not explain how many participants contributed to the Year 3 endpoint or how missing observations were handled. Likewise, the adverse-event entries are not substitutes for primary outcome results, and the planning assumptions do not constitute evidence of an effect or observed power.

The most supportable conclusion is descriptive: the registered study set out to examine health-plan switching and several outcomes for patients with bipolar disorder, but the posted results do not report the primary endpoint analysis. The record therefore cannot support a quantitative claim about the hospitalization comparison or a causal conclusion about HDHPs.

14. Why This Trial Matters Statistically

This study raises a statistical question that is important in health services research: how to interpret utilization differences when exposure to a health plan is not randomly allocated. Claims data can describe service use across large populations, but a large enrollment alone does not remove confounding or make comparison groups equivalent. The registry's own warning that the arms are not like-for-like makes the distinction between association and causal effect especially important.

The endpoint also illustrates how a summary measure can conceal important distributional features. A mean annual hospitalization count provides an average, while the standard deviation describes individual variation. Neither alone reveals the distribution of counts or whether a small part of the population contributes substantially to the mean. Those are general interpretive issues for count-valued outcomes. The registry does not report the endpoint values needed to determine whether they are relevant in this comparison.

The study description connects claims-based outcomes with patient and caregiver interviews, which signals interest in both recorded utilization and lived experience. The registry lists Patient Interview as the intervention, while its summary describes HDHP switching as the policy exposure. Clarifying how these components relate is important for understanding what was measured and which population each finding would represent.

Finally, the registry record demonstrates why transparent reporting of results and methods matters. A named endpoint and completed status do not reveal the estimate, its precision, the analysis population, or the handling of missing data. With 0 analyses reported, readers cannot evaluate the statistical evidence for the main question. The value of the record is in identifying the research question and documenting important design cautions; it does not supply the result needed for a quantitative conclusion.

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 primary limitation for interpretation is that the registry record does not report a numerical result for the primary endpoint. It gives the endpoint definition and dispersion measure but no group means, contrast, confidence interval, or test result. The record lists 0 analyses, leaving no reported statistical estimate to assess.

The design is not described in enough detail to reconstruct the comparison. Allocation is not stated, and the record explicitly says the arms are not a like-for-like control comparison. The HDHP exposure definition, the timing and process for identifying a switch, the criteria for the control group, and methods for addressing confounding are not reported in the registry record. Therefore, the intended policy question cannot be translated into a clearly defined causal contrast from the available information.

The registry also does not explain how the overall enrollment relates to endpoint-specific eligibility, follow-up, or analysis. The population field for the primary outcome is empty. Details of the Year 3 follow-up period, missing claims or outcome data, and analysis-population rules are not reported in the registry record. These gaps prevent assessment of selection and completeness for the primary endpoint.

Further details that would help interpret the study—including the analysis model, covariates, multiplicity handling, subgroup results, and the sample-size justification—are not reported in the registry record. The summary describes interviews, but their methods and their relationship to the claims analysis are not specified in the posted results. These limitations do not show that the study was conducted improperly; they describe what cannot be evaluated from the registry record.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: High Deductible Health Plans and Bipolar Disorder — NCT02560701.
• Registry record last updated 2020-08-06; status last verified 2019-05.
• Results first posted 2019-07-24.
• 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 record concerns high-deductible health plans and outcomes among patients with bipolar disorder. It identifies Harvard Pilgrim Health Care as sponsor, lists the status as COMPLETED, and records enrollment of 350823. The study summary describes claims data from 2004-2014 and interviews with patients and caregivers.

The primary endpoint is mean annual inpatient hospitalizations in Year 3, with standard deviation as the dispersion measure. Secondary outcomes address outpatient service access and psychotropic medication adherence. The intervention entry is Patient Interview, while the study question concerns switching into HDHPs. Allocation is not stated, and the registry cautions that the arms are not a like-for-like control comparison.

The registry lists six outcome measures and 0 analyses. It does not report a numerical primary endpoint result, its uncertainty, the endpoint-specific analysis population, or the methods used to address missing data and confounding. The adverse-event entries list 0 affected participants and 0 deaths in both groups. These record entries do not provide a quantitative result for the primary comparison.