Statistical Analysis Plan for the Blood Pressure Postpartum (BP²) study

2.1 Preface

Hypertensive Disorders of Pregnancy (HDP) complicate 5–10% of pregnancies globally: 2–5% preeclampsia (PE), 3% gestational hypertension (GH), and 0.5–2% chronic hypertension (CH). It is well established that HDP’s adverse health effects do not end with the pregnancy: these conditions are associated with increased lifetime cardiovascular disease (CVD) risks. After HDP, women not only have a 3–4 times increased risk of chronic hypertension within 15 years, but a 2–3 times increased risk of CVD. This relationship is strongest for PE and CH but also exists for GH. Intervention data specific to the HDP population remain scarce. There has not yet been a randomised trial of specialised follow-up after HDP versus usual care, nor a sufficiently powered RCT after HDP to assess effect of lifestyle behaviour change strategies on presumed markers of later cardiovascular risk such as blood pressure and weight.

2.2 Scope of the analyses

The Blood Pressure Postpartum study (BP²) will provide evidence regarding the feasibility and effectiveness of postpartum lifestyle behaviour change interventions and structured clinical follow-up in improving cardiovascular health markers after HDP. This SAP applies to the 12-month follow up.

3.1 Study Objectives

In women who have had a hypertensive disorder of pregnancy, to compare the effect of three postpartum (PP) managementstrategies at 1 year PP, withfurther observational follow-up at 3 years PP (note that in this SAP we are referring to the 1 year/12 month follow from now on). We willcompare the effects of:

• 1) OptimisedUsualCare: Standard postpartum care afterHDP,with the additionof information packages sent to both womenand their primaryhealthcare practitionersprior to a 6-months postpartum visit with the primary healthcare practitioner

• 2) Brief Education Intervention: One-off cardiovascular risk assessment and diet/exercise education in a dedicated clinic 6 months postpartum

• 3) Extended Lifestyle Intervention: Cardiovascular risk assessment and diet/exercise education in a dedicated clinic 6 months postpartum, followed by an individualised dietary and exercise lifestyle program of six months duration (from 6-12 months PP)

on

• Maternal blood pressure (BP)

• Maternal positive lifestyle behaviour change

• Other maternal clinical cardiovascular risk indicators

• Non-invasivematernal measures of vascular structure and function

• Women’s health-related quality of life measures

• Infant growth and weight trajectories

• Healthcare costs, taking into account projected long-term savings

3.2 Endpoints

4.1 General Study Design and Plan

(ICH E3;9)

Study configuration and experimental design: three-arm, multicentre, randomized controlled trial with participants allocated 1:1:1 to control group, Brief Education Intervention and Extended Lifestyle Intervention.

PROBE design: Prospective Randomized Open Blinded Endpoint.

Type of Comparison: Superiority of intervention groups versus control

Type of control: optimized usual care

Level and method of blinding: Open blind, patients and clinicians will not be blinded, however clinicians taking outcome measurements and analyst team will be blinded to group allocation.

Method of treatment assignment: Randomization with stratification by study hospital, parity and body mass index (see Section 4.3).

4.2 Inclusion-Exclusion Criteria and General Study Population

(ICH E3;9.3. ICH E9;2.2.1)

General Study population: Australian teaching hospitals providing maternity care where staffing includes specialist obstetricians, physicians, paediatricians and anaesthetists/critical care specialists and >1500 births per year.

Inclusion criteria:

• Aged ≥18 years

• Given birth at one of the study hospitals (St George Public Hospital, Royal Prince Alfred Hospital, Liverpool Hospital, Campbelltown hospital, Royal Hospital for Women, Westmead Hospital) during the study period

• The index pregnancy was complicated by a hypertensive disorder of pregnancy (HDP): preeclampsia (PE), gestational hypertension (GH), chronic hypertension (CH), chronic hypertension plus superimposed preeclampsia (CH+PE)

Exclusion criteria:

• Stillbirth or neonatal death from HDP pregnancy

• Planning to move out of study area/unavailable for follow up during study period

• Active severe mental health condition or developmental disability precluding informed consent.

Participant Flow Diagram

• Download figure
• Open in new tab

4.3 Randomization and Blinding

(ICH E3; 9.4.3, 9.4.6. ICH E9; 2.3.1, 2.3.2)

Participants will be randomly assigned to Group 1, 2 or 3 with a 1:1:1 allocation as per a computer-generated randomisation schedule using REDCap software. Randomisation will be stratified by site, parity (1 versus >1) and body mass index (<=30 versus 30+), using blocks. Details relating to block randomisation will be kept in a separate document not available to anyone who is involved in trial recruitment. Block size will be revealed after analysis. Allocation concealment is ensured, as the randomisation code will not be released until after baseline questionnaires have been completed and the study researcher then accesses REDCap for randomisation to Group 1, 2 or 3.

All participants who give consent for participation and who complete the baseline questionnaires within the scheduled 3 week timeframe will be randomised. Randomisation will be performed in REDCap by staff responsible at each study site for recruitment, which will usually be the study nurse/midwife at that site. This staff member will then organise information package and instructions regarding GP follow-up (Group 1) or for information package and postpartum clinic visit (Groups 2 and 3). This staff member will not be involved in outcome assessment in postpartum clinic at 6 month, 12 month or subsequent visits.

Due to the nature of the intervention, neither postpartum clinic staff nor participants can be blinded to allocation. Regarding analysis, a PROBE design (prospective randomised open blinded endpoint) is adopted, with analysis occurring blinded to group allocation (Ford & Norrie, 2016). After all participant data including outcome data is entered by study staff, it will be exported with group allocation concealed for blinded analysis by the trial statistician and trial health economist as applicable.

4.4 Study Assessments

(ICH E3; 9.5.1. ICH E9; 2.2.2)

6.1 Timing of Analyses

The final analysis will be performed when:

• All randomised subjects have either completed the 12 month visit or dropped out of the study prior to 12 months.

• The data for analysis meets the cleaning and approval requirements of the main analyst:

  • Attributing baseline levels to any missing 12 month data

  • Checking missing data values for validity and consistency

  • Checking data dictionary matches with dataset

  • Removal of Group 3 GHS compliance questions (and any other questions that are only answered by one of the randomized groups)

  • Removal of group names, randomisation of groups with numbers or letters so analyst is blinded

  • Check for invalid entries eg birthweight of 1g and either fix or set to missing

• After the finalization and approval of this SAP document by the analyst, the principal investigator and the study coordinator.

6.2 Analysis Populations

(ICH E3; 9.7.1, 11.4.2.5. ICH E9; 5.2)

6.3 Covariates and Subgroups

(ICH E3; 9.7.1, 11.4.2.1. ICH E9; 5.7)

Covariates: adjust analyses for hospital.

Subgroup analysis for the stratified randomisation and pre-specified groups (below).

Primary and secondary outcomes to be assessed (secondary results will go in supplement). Significance of the treatment*subgroup interaction term in the regression model at a level of p ≤ 0.01 will be required to conclude there exists heterogeneity of treatment effect.

• Parity (1 v >1)

• BMI (<30 v ≥ 30 at baseline)

• Type of hypertensive disorder of pregnancy (chronic hypertension, gestational hypertension, preeclampsia, preeclampsia superimposed on chronic hypertension)

6.4 Missing Data

(ICH E3; 9.7.1, 11.4.2.2. ICH E9;5.3. EMA Guideline on Missing Data in Confirmatory Clinical Trials)

End of recruitment: discussion with study management team and project statistician to discuss the proportion of missing data and strategies to minimise missing data prior to final data cleaning and analysis.

Missing data will be assessed to determine whether the 12-month outcome data is missing at random. Strategies to deal with missingness will depend on the nature of missingness and could include:

1. Baseline data: impute a value of 0 for missing covariates, and then include a missing-indicator variable for the covariate as an additional covariate in the regression model [White and Thompson (2005)].

2. Outcome data: where GP, clinic-based measures are missing e.g. waist circumference, weight, then self-report measures can be used.

3. Lost to follow up: as specified with Intention to Treat, there are three possible approaches to handling the missing 12-month. Using baseline measures for primary outcomes in those who drop out is the best way to mimic real world practice where people do not continue with an intervention, however it has the potential to bias the results towards the null especially if there is a large loss to follow up, therefore, we will consider multiple imputation of missing data based on the nature of missingness.

Reporting: proportion of missing data at baseline and loss to follow up will be reported as well as methods to deal with both issues.

6.5 Multiple Testing

(ICH E3; 9.7.1, 11.4.2.5. ICH E9; 2.2.5)

The primary analysis compares 12 months (outcome, t=1) and 6 months (baseline, t=0), therefore only two points in time. Subsequent analyses at later times may include 12 month data and longitudinal analysis accounting for multiple time points, but that is not the focus of the current analysis.

However, in the initial analysis given that there are a number of primary and secondary outcomes this increases the risk of Type 1 error due to multiple hypothesis testing. This will be dealt with one of two methods: 1) by controlling for the false discovery rate (FDR) as proposed by Benjamini & Hochberg 1995 (J R Statistic Soc: 57; 289) or controlling for the familywise error rate (FWER) proposed by Hothorn, Bretz & Westfall 2008 (Biometric J:50; 346) through the R package multcomp. The Hothorn, Bretz & Westfall approach allows simultaneous inferences to be made over both parametric models and all relevant hypotheses, while accounting for correlation between the test statistics and allowing for stepwise testing procedures.

For the analysis of primary outcomes the type 1 error rate for either multiplicity testing method will be set to α= 0.05 and for the secondary outcomes α = 0.01.

7.1 Subject Disposition

As shown in the Participant Flowchart above (page 9).

7.2 Derived variables

As shown in the 4.4.3 variable list.

7.3 Protocol Deviations (COVID-19 pandemic)

Due to the COVID-19 pandemic essential clinical and outcome measures (blood pressure, weight, and blood/urine tests) 6 and 12 months postpartum were performed, depending on LHD/site policy (a) by the research midwife, with visits kept to <15 minutes, and separate from antenatal/outpatient clinic spaces (i.e. no waiting time and woman able to maintain social distancing) (b) where possible via the woman’s GP, to coincide with necessary visits of her infant (for 6 and 12 month check and vaccinations).

Women were reminded that they and their baby must be well/asymptomatic of COVID-19 to be attending such visits. Therefore, there may be some missing data due to COVID-19 infections. These women will be treated as outlined in section 6.4 after consideration of the amount of missingness.

In addition, due to pandemic restrictions one of the secondary outcomes-maternal vascular structure and function was not measured. As this is a secondary outcome and the amount of missing data will be considerable this outcome will not be included in the primary analysis.

7.4 Demographic and Baseline Variables

As shown in Section 4.4.3 ‘Variables’ under ‘Descriptives’. Presented as outlined in Section 7 ‘Summary of study data’ and Table 1 Section 14.

7.5 Concurrent Illnesses and Medical Conditions

As shown in Section 4.4.3 ‘Variables’ under ‘Descriptives’. Presented as outlined in Section 7 ‘Summary of study data’ and Table 1 Section 14.

7.6 Compliance

For groups 1 and 2 all participants will be considered to have complied, as baseline (6 months visit) data cannot be obtained without participants having been compliant to this allocation. For group 3, participants will be considered (a) non-compliant if they did not sign up to the extended intervention (Get Healthy Service) and have at least one assessment with GHS (b) partially compliant if there is a record of GHS sign-up and initial assessment/engagement (c) fully compliant if they are noted as having graduated from the GHS program.

8.1 Primary outcome Analysis

There are six primary hypotheses resulting from an all-pairwise comparison of the three study groups (two intervention groups and control) in the two primary outcomes of

1. maternal systolic blood pressure decrease

2. binary threshold on maternal weight (≥ 4kg reduction) and/or waist circumference (≥ 2cm reduction)

change from 6-12 months postpartum.

The three pairwise hypotheses relating to primary outcome 1 (BP) will be incorporated into an ANOVA model, and the three pairwise hypotheses relating to outcome 2 (composite measure) incorporated into a logistic regression model. The model fit statistics (F-test and χ²-test) from each of the models will be used to determine (with a threshold of α = 0.05) whether the relevant individual pairwise hypotheses should be evaluated. If the threshold is met, one of the multiple testing methods described in Section 6.5 will be applied. All (three or six) tests that pass to the second stage of testing will be considered to be one family of tests and tested as two-sided and based on a null hypothesis of no treatment effect. A multiplicity adjusted p-value of less than 0.05 will be considered significant in any of the tests for the primary outcomes.

Analyses will be adjusted for hospital site.

Subgroup analysis will occur as stipulated in Section 6.3.

8.2 Secondary outcomes Analyses

Analysis of the secondary outcomes will proceed in a similar manner to the primary outcomes, with tests for differences across groups performed as a first step using ANOVA testing for continuous outcomes and chi-squared test (from a logistic regression model) for categorical outcomes. For ordinal outcomes we will first do an ordinal regression to compare 12m and 6m categories within treatment groups, we will then compare groups. However, there will be a reduction in the required significance level to α = 0.01. Where there is evidence of differences amongst groups in the secondary outcomes, the same multiple testing method will be used to determine between which groups the differences exist, while controlling the familywise type I error rate at α = 0.01.

Analyses will be adjusted for hospital site.

Subgroup analysis will occur as stipulated in Section 6.3.

8.3 Secondary Analyses of primary and secondary endpoints

Use of the Get Healthy Service (GHS) will be analysed in a secondary analysis of both primary and secondary endpoints comparing women with GHS usage ≥ 4 phone calls versus no evidence of signing up to GHS. Analyses will be adjusted for randomisation group to assess whether outcomes are associated with GHS usage independently of the effect of the intervention.

9.1 Adverse Events

We don’t expect any adverse events for these interventions.

9.2 Clinical Laboratory Evaluations

Laboratories are accredited by NATA and use the same reference ranges. Evaluation is not needed.