Home and Online Management and Evaluation of Blood Pressure (HOME BP) using a digital intervention in poorly controlled hypertension: randomised controlled trial

Abstract Objective The HOME BP (Home and Online Management and Evaluation of Blood Pressure) trial aimed to test a digital intervention for hypertension management in primary care by combining self-monitoring of blood pressure with guided self-management. Design Unmasked randomised controlled trial with automated ascertainment of primary endpoint. Setting 76 general practices in the United Kingdom. Participants 622 people with treated but poorly controlled hypertension (>140/90 mm Hg) and access to the internet. Interventions Participants were randomised by using a minimisation algorithm to self-monitoring of blood pressure with a digital intervention (305 participants) or usual care (routine hypertension care, with appointments and drug changes made at the discretion of the general practitioner; 317 participants). The digital intervention provided feedback of blood pressure results to patients and professionals with optional lifestyle advice and motivational support. Target blood pressure for hypertension, diabetes, and people aged 80 or older followed UK national guidelines. Main outcome measures The primary outcome was the difference in systolic blood pressure (mean of second and third readings) after one year, adjusted for baseline blood pressure, blood pressure target, age, and practice, with multiple imputation for missing values. Results After one year, data were available from 552 participants (88.6%) with imputation for the remaining 70 participants (11.4%). Mean blood pressure dropped from 151.7/86.4 to 138.4/80.2 mm Hg in the intervention group and from 151.6/85.3 to 141.8/79.8 mm Hg in the usual care group, giving a mean difference in systolic blood pressure of −3.4 mm Hg (95% confidence interval −6.1 to −0.8 mm Hg) and a mean difference in diastolic blood pressure of −0.5 mm Hg (−1.9 to 0.9 mm Hg). Results were comparable in the complete case analysis and adverse effects were similar between groups. Within trial costs showed an incremental cost effectiveness ratio of £11 ($15, €12; 95% confidence interval £6 to £29) per mm Hg reduction. Conclusions The HOME BP digital intervention for the management of hypertension by using self-monitored blood pressure led to better control of systolic blood pressure after one year than usual care, with low incremental costs. Implementation in primary care will require integration into clinical workflows and consideration of people who are digitally excluded. Trial registration ISRCTN13790648.


Web Appendix to accompany HOME BP main trial paper Appendix 1 Additional Tables and Figures
* Mean difference (95% confidence intervals) controlling for baseline blood pressure, age and BP target, with a random effect for practice $ Blood Pressure was measured using a BPTru 200 monitor set to record six values at 1 minute intervals. SBP systolic blood pressure in mmHg + Mean (standard deviation) * Mean difference (95% confidence interval) Controlling for baseline blood pressure, age and BP target, with a random effect for practice apart from age subgroup  *Participants could sign up to separate online interventions for each behaviour change, including standalone educational sessions regarding salt, healthy eating and alcohol with behaviour change techniques to increase self-efficacy. The physical activity intervention was an interactive online tool with tailored email prompts for increasing physical activity (Getting Active). The weight loss intervention was a complex 24-session tool (POWeR) based on self-management techniques such as goal-setting, which has been shown to be effective in a large scale RCT. 6

Figure A1
Scatterplot of joint distribution of incremental mean cost from NHS perspective (£s) and mean blood pressure reduction from baseline (mmHg) over 12 months

Figure A2
Cost effectiveness acceptability curve of the intervention and usual care groups based on blood pressure from baseline over 12 months.

PURPOSE AND SCOPE OF THE PLAN
This document details the proposed analysis of the main paper(s) reporting results from the NIHR funded randomised controlled trial to evaluate self-management of raised BP through self-monitoring, medication adherence and lifestyle changes using the HOME BP online system with optional nurse support compared with standard care. The results reported in these papers should follow the strategy set out here. Subsequent analyses of a more exploratory nature will not be bound by this strategy, though they are expected to follow the broad principles set out here. The principles are not intended to curtail exploratory analysis (for example, to decide cut-points for categorisation of continuous variables), nor to prohibit accepted practices (for example, data transformation prior to analysis), but they are intended to establish the rules that will be followed, as closely as possible, when analysing and reporting the trial. The analysis strategy will be available on request when the principal papers are submitted for publication in a journal. Suggestions for subsequent analyses by journal editors or referees, will be considered carefully, and carried out as far as possible in line with the principles of this analysis strategy; if reported, the source of the suggestion will be acknowledged. Any deviations from the statistical analysis plan will be described and justified in the final report of the trial. The analysis should be carried out by an identified, appropriately qualified and experienced statistician, who should ensure the integrity of the data during their processing. Examples of such procedures include quality control and evaluation procedures. The health economics variables, incremental QALY and cost per patient, are included in this planned within-trial analysis. These economic results will inform a long term cost effectiveness model, which is detailed elsewhere.
Trial overview Blood pressure is a key risk factor for cardiovascular disease. The HOME BP trial aims to assess the feasibility, acceptability, effectiveness and cost-effectiveness of adding the HOME BP intervention (comprising HOME BP online digital intervention, self-monitoring, medication titration and lifestyle interventions with nurse support) into primary care for selfmanagement of hypertension, compared to usual care. 1.4 OUTCOME MEASURES Baseline measures are assessed by a Practice nurse or healthcare assistant at the GP Practice. Follow-up measures are assessed at 6 and 12 months by research nurses either in the patients' own practice or at their home. Data will be collected online or on paper questionnaires, if the online data was not completed for any reason. See Appendix II for a table of outcomes assessment schedule.

PRIMARY OUTCOME
The primary outcome is systolic blood pressure at 12 months.
Six measurements of BP are taken at the follow up visit. The mean of the 2 nd and 3 rd BP readings (conventional BP) will be used.
(iii) The analyses will be repeated taking a repeated measures approach and exploring the effect in both systolic and diastolic BP over the 12 month period (iv) The analyses (Systolic and diastolic BP at 6 and 12 months) will be repeated controlling for habituation. The mean of the 2-6 th measurements will be used.

Tertiary outcomes
Scales are included at Appendix III.

(i) Current medications
Data on the overall number of current medications will be reported. For individual drug classes, these will be converted into defined daily doses. The DDD is the assumed average maintenance dose per day for a drug used for its main indication in adults and is defined by the WHO (http://www.whocc.no/ddd/definition_and_general_considera/).

(ii)
Side effects and safety a) IPQ hypertension Adjusted symptoms subscale. List of 24 symptoms with dichotomous response options of yes or no to indicate whether each symptom has been experienced in the last 4 weeks, and one 'other' option with an open-text box to clarify. b) Adverse events such as admission to hospital, cardiovascular events and deaths will be recorded as part of the safety monitoring and for the economic analysis.

(iii)
Quality of life EQ5D QALY calculation by converting questionnaire scores using the standard eq5d rule set.

(iv) Resource use and costs
Cost (NHS and societal) will be estimated for each patient based on data on resources used multiplied by the relevant cost. Relevant resources include medications, those linked to adverse events (see above), and use of routine NHS services. Societal cost will comprise NHS cost plus any costs borne privately by patients. The cost of the intervention will be included for those in the intervention arm.

SAMPLE SIZE
The study requires a total sample of 244 patients per group. This allows 90% power to detect a difference in SBP of 5 mm Hg (SD 17 mm Hg) between intervention and usual care groups, based on the findings from the TASMINH-2 study. Allowing for a 15% participant drop out, we planned to recruit 287 participants per arm, resulting in a total sample size of 574 participants. Due to follow-up rates closer to 20%, the target sample size increased to 610 participants.

RANDOMISATION AND BLINDING IN THE ANALYSIS STAGE
Eligible participants were randomised in a 1:1 ratio to receive either usual care or the HOME BP intervention with optional nurse support using the HOME BP online system. Minimisation was used, taking into account participants' baseline SBP, BP target based on age (under 80/80 and over) and diabetes status, and practice. Patients were randomised to the optimal group 80% of the time. Any random numbers the minimisation routine needed were computer generated, therefore bypassing study team involvement. The statistician will analyse the data blind to group allocation.

CHARACTERISTICS OF PARTICIPANTS
Baseline characteristics of the patients (i.e. sociodemographic data, duration of hypertension, past medical history, height and weight, blood pressure, current antihypertensive medication, adherence to medication) will be reported by randomised group. Continuous data will be summarised in terms of the mean, standard deviation, and number of observations or, where skewed, median and lower & upper quartiles. Binary/categorical data will be summarised in terms of frequency counts and percentages.
Participant flow, from screening through randomisation, follow up and analysis will be presented in a CONSORT flow chart and include reasons for withdrawal.

DEFINITION OF POPULATION FOR ANALYSIS
All data will be included in the analysis as far as possible to allow full ITT analysis. Patients will be analysed in the groups they were allocated, irrespective of whether they received that intervention or not.

PRIMARY ANALYSIS
Analyses will be performed using Stata version 13 or above. All tests will be two-tailed with point estimates, 95% confidence intervals and exact p-values for the treatment effect presented.
Analyses using regression models will adjust for the factors used in minimisation (baseline SBP, and BP target based on age and diabetes status). Baseline SBP and age will be entered into the model as continuous variables. Practice was also a minimisation factor but this will be included as a random effect (see 2.1 below). No formal adjustment for multiple significance testing will be applied. The primary approach for analysis will be with imputation of missing data to allow full ITT analysis. Patients will be analysed in the groups they were allocated, irrespective of whether they received that intervention or not.

PRIMARY OUTCOME
Descriptive summaries of mean blood pressure at baseline, 6 months and 12 months will be presented for each group.
The primary analysis will use general linear mixed modelling to compare intervention and usual care SBP at follow-up adjusting for minimisation factors and empirical confounders as defined above. The unadjusted and adjusted mean difference between groups will be presented The assumptions of the normality of the residuals from the fixed part of the model and the normality of the random effects at the cluster level will be checked. Appropriate transformations will be considered if there is some suggestion that the assumptions for the linear model may not be met.

HANDLING MISSING AND UNREALISTIC DATA
The primary analysis for all outcomes will be with all missing values imputed using an individual chained equations multiple imputation model. This model will impute the 2 nd and 3 rd blood pressure measurements and then take an average of these to form a dataset with complete primary outcome measurement rather than imputing the average measurement. The model will include all blood pressure measurements taken as well as the minimisation variables and sociodemographic characteristics.
A sensitivity analysis will present the results for a complete cases analysis. It is possible that some blood pressure values may be entered erroneously. A macro will be used to identify outliers and these values will be replaced with the average of their two immediate neighbours from the same visit. (See Appendix V)

SECONDARY OUTCOMES
Analysis of systolic BP at 6 months (mean of 2 nd /3 rd measurements) will be derived from the primary outcome model. Additional blood pressure outcomes as listed below will be analysed as per the primary outcome analysis in section 4.1. Analysis of diastolic BP will adjust for baseline DBP rather than baseline SBP. 1. Systolic BP at 6 months (mean of 2 nd /3 rd measurements) 2. Diastolic BP at 6 and 12 months (mean of the 2 nd /3 rd measurements) 3. The analyses (Systolic and diastolic BP at 6 and 12 months) will be repeated controlling for habituation. The mean of the 2-6 th measurements will be used.
A secondary analysis will explore the effect over the 12 month period with readings at baseline, 6 months and 12 months using a repeated measures modelling approach. We will use a multilevel mixed model (MMLM) framework with observations (level 1) nested within participants (level 2) nested within practices (level 3). Results will be presented adjusting for minimisation variables and significant confounders as per the primary analysis. The model will use all the observed data and makes the assumption that missing blood pressure measurements are missing at random given the observed data.
As there may not be a constant treatment effect over time, a treatment/time interaction will be modelled and included if significant, with time treated as a random effect. The model will include a random effect for practice (random intercept) and patient (random intercept and slope on time) to allow for between patient and practice differences at baseline and between patient differences in the rate of change over time (if significant), and fixed effects for baseline covariates. An unstructured covariance matrix will be used.
The assumptions of the normality of the residuals from the fixed part of the model and the normality of the random effects at the cluster level will be checked. Appropriate transformations will be considered if there is some suggestion that the assumptions for the multilevel linear model may not be met.

TERTIARY OUTCOMES
For all tertiary outcomes, linear regression will be used for continuous outcomes if the assumptions are met. Otherwise non-parametric analyses will be used. Logistic regression will be used for dichotomous outcomes and a suitable count model, as determined by goodness of fit measures, for count data. All analyses will control for stratification variables and potential confounders. Medication defined by DDD will be analysed using mixed effect models following a similar strategy to that outlined in the analysis of the primary outcome. Where available, the baseline value will be included as a covariate along with minimisation variables. Current medication will be analysed by drug class and overall and summarised as per the TASMIN-SR paper (table 3 below). Number of antihypertensive medications will also be analysed.

Adverse events
The number and percent of people experiencing each symptom from the 12 month questionnaire will be tabulated as per Table 4 of the TASMIN-SR study (McManus et. al JAMA 2014) in order from the most commonly reported to least commonly reported. All symptoms will be reported in the statistical report, not just the top 10. Hypoglycaemic episodes for diabetic patients in last 6 months will also be reported (recorded at 6m and 12m FU) Comparisons between groups will be conducted at 6 months and at 12 months using a generalised linear mixed effects model for binary data, adjusted for baseline values (symptom experienced at baseline Y/N), with a random effect for practice.

Estimating QALYs from EQ5D5L data
To derive a quality of life value at each time point based on the EQ5D5l requires a weighted average of these five values. The weights, based on a survey of public values are provided on the Health Economics journal website (https://onlinelibrary.wiley.com/doi/full/10.1002/hec.3564. Supporting Information files hec3564-sup-0001-supplementary.zip application/x-zip-compressed ) The value for each patient at each time point will be a value between 0 and 1. Any missing EQ5D5L values will be imputed using the same chained equations approach as the primary outcome. QALYs will be analysed using an area under the curve approach. For each patient the relevant formula is QALY=(Q2-Q1)/2*T, where Q1 is initial quality of life, Q2 is the end of period quality of life and T is time. If related deaths occur and are to be included, Q2 would take the value 0. The relevant time might be the point of death if known. Cost Cost will be estimated by multiplying the units of service used by the price. The health economist will provide a relevant look up table for NHS costs.
Costs per unit change in the primary outcome will be calculated and reported for each trial arm.

SUBGROUP ANALYSES
Although we are not powered to look at subgroups, we will explore the impact of key subgroups that could plausibly modify intervention effectiveness: • (i) Age (two categories split at median) • (ii) BP target (standard hypertension/older hypertension/diabetes) • (iii) Sex (Male/Female) • (iv) Baseline BP (two categories split at median) • (v) IMD (two categories split at median) • (vi) Comorbidity (Three categories -none, one, multiple) • (vii) Number of medications • (viii) Previous experience of self-monitoring at baseline (None, Did previously but stopped, Current) For this exploratory analysis, estimates of the interaction between subgroup and intervention will be provided with 95% confidence intervals and the estimates of the intervention effect when the subgroup is selected. Clear evidence of benefit in a subgroup will require the interaction term for that subgroup to be significant at the 5% level. Although a formal adjustment for multiplicity will not be carried out, the results will be interpreted cautiously, as at least 1 significant result might be expected by chance 5 SENSITIVITY ANALYSES

SENSITIVITY ANALYSIS FOR MISSING DATA
A complete cases analysis will be carried out to explore if the findings from this sensitivity analysis are similar to the main analysis.

SENSITIVITY ANALYSIS FOR PRACTICE WHO DIVERGED FROM PROTOCOL
One practice diverged from protocol as the nurse prescriber decided to see all usual care patients at regular intervals to monitor their blood pressure. A sensitivity analysis will be undertaken excluding all participants recruited from this practice to explore whether there is any impact on inferences from the primary analysis.

SAFETY ANALYSIS
Serious adverse events for the full population will be summarised descriptively according to randomised group. No statistical comparisons will be undertaken on these data

CHANGES TO PROTOCOL OR PREVIOUS VERSIONS OF SAP
All changes from the protocol or from previous versions of the stats plan will be detailed in the report.

Economic measures
Patient quality of life (EQ-5D) X X X Patient side effects (IPQ hypertension: Adjusted symptoms subscale) X X Costs of equipment and drugs X(NR) Health professional time X(NR) Patient time X(NR) Qualitative process analysis Patient experience and views of the DI X HCP experience and views of the DI X