key: cord-0429064-duhtzeql
authors: Song, J.; Wang, X.; Wang, B.; Gao, Y.; Liu, J.; Zhang, H.; LI, X.; Li, J.; Wang, J.; Cai, J.; Herrin, J.; Armitage, J.; Krumholz, H.; Zheng, X.
title: Rationale and design of the Learning Implementation of Guideline-based decision support system for Hypertension Treatment (LIGHT) Trial and LIGHT-ACD Trial
date: 2021-03-12
journal: nan
DOI: 10.1101/2021.03.11.21253427
sha: 9564d8f6944736de85d30beed27ababac6e14f8e
doc_id: 429064
cord_uid: duhtzeql

Background: Computerized clinical decision support systems (CDSS) are low-cost, scalable tools with the potential to improve guideline-recommended antihypertensive treatment in primary care. Uncertainty remains about the optimal initial antihypertensive therapy in the settings of real practice. Methods: The Learning Implementation of Guideline-based decision support system for Hypertension Treatment (LIGHT) trial is a pragmatic, cluster-randomized controlled trial of CDSS versus usual care conducted in 100 primary care practices in China. The primary outcome is the proportion of hypertension visits with appropriate (guideline-recommended) antihypertensive treatment. Among patients recruited from primary care practices of the intervention group in the LIGHT trial, we further conducted a sub-study, the LIGHT-ACD trial, to compare the effects of initial antihypertensive therapy by randomizing individual patients to receive different antihypertensive regimens of initial monotherapy or dual therapy. The primary outcome of the sub-study is the absolute change in blood pressure from baseline to 9 months. Results: We hypothesize that the use of CDSS will result in a higher proportion of appropriate antihypertensive treatments being prescribed during visits for hypertension control in the LIGHT trial, and that particular choices of monotherapy or combinations of dual therapy lead to greater blood pressure change in the LIGHT-ACD trial. Conclusion: These nested trials will provide reliable evidence on the effectiveness of CDSS for improving adherence to guidelines for hypertension management in primary care, and data on the effectiveness of different initial antihypertensive regimens for blood pressure reduction.

INTRODUCTION 1 these outcomes' denominator (Supplement 1). 19 20

The CDSS was integrated into the EHR of the intervention sites and consisted of 22 three core components: (1) point-of-care decision support for antihypertensive 23 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. ; https://doi.org/10.1101 /2021 doi: medRxiv preprint P a g e 9 o f 3 3 therapy with regard to the class and dose of antihypertensive medication; (2) alerts for 1 referral, contraindications, and underdose or overdose of antihypertensive medication; 2 and (3) alerts for re-evaluation of the antihypertensive therapy if a physician does not 3 follow the CDSS recommendation. 4

The CDSS was developed by a multidisciplinary team including clinicians and 5 information technology (IT) experts. Clinicians and academics developed the CDSS 6 algorithm which provides patient-specific medication recommendations and alerts. 7

The algorithm was developed mainly based on hypertension management guidelines 8 for primary care in China. 18 Other guidelines from the USA and Europe were also 9 considered. 19-21 10 After the algorithm had been finalized, IT experts worked together to translate the 11 algorithm into computational logic. The CDSS logic was tested using simulated 12 patient data to trigger each possibility of the algorithm. After internal testing, the CDSS 13 was provided for clinicians for further validation. 14 We retrieved alerts and medication recommendations of CDSS for the patient 15 and compared them with the recommendation given by clinicians. The IT experts 16 were notified when any discrepancies were found so that the programming errors 17 could be identified. This process was repeated until no errors were observed in all test 18 cases. The user interface was tested by doctors in two excluded primary care 19 practices specifically to ensure the usability of CDSS. 20 21 Data collection, quality control and data management 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. ; https://doi.org/10.1101 /2021 doi: medRxiv preprint P a g e 1 0 o f 3 3

Sociodemographic characteristics (age, gender, education, and health insurance), 2 physical measurements (blood pressure, heart rate, waist circumference, height, and 3 weight), cardiovascular risk factors and co-morbidities, current medications, 4 medication adherence, prescriptions, information on self-reported home monitoring 5 blood pressure, and side-effects related to antihypertensive medications of 6 participants are collected via EHR. Blood pressure is measured with the patient 7 seated, using the same validated automated sphygmomanometer (Omron 8 HBP-1300) 22 after at least a 5-minute rest at each visit. Two blood pressure readings 9 are taken 1-2 minutes apart and the average value is recorded. For primary care 10 practices in the intervention arm, if doctors do not follow the recommendations of 11 CDSS, the relevant reasons will be recorded. 12

All data are securely transmitted to the central server through automatic 14 electronic transfer and securely stored in an encrypted and password-protected 15 database. The database can be accessed only by approved staff members. At the 16 local sites, all staff members must use their own usernames and passwords to log into 17 the EHR, which will create an audit trail of all data entered or changed. Data 18 confidentiality policies on data collection, storage, and analysis have been strictly 19 imposed in order to ensure the confidentiality of personal information. 20

We developed a web-based platform to monitor real-time project progress and 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) quality, and to provide management support for primary health care practice staff. 1

On-site monitoring of recruitment, physical measurements, and accuracy of the data 2 documentation are regularly conducted by trained staff to ensure the quality of data 3 collection. All automated sphygmomanometers are calibrated annually. In addition, to 4 ensure the accuracy of the blood pressure value, research staff will randomly audit at 5 least one blood pressure values documented in the EHR against the recording in the 6 electronic blood pressure monitor from all sites on a daily basis. 7 8 Outcomes 9

The primary outcome is the proportion of hypertension visits with appropriate 10 treatment. Appropriate treatment is defined as the prescription compliant with all the 11 pre-specified evidence-based recommendations. These recommendations mainly 12 include titrating or switching treatment for patients with poor blood pressure control, 13 using a particular antihypertensive medication for patients with specific clinical 14 indications or without compelling contraindications or intolerance to their use. Detailed 15 recommendations specifications are shown in Supplement 2. 16

The secondary outcomes include the average change in systolic blood pressure, 17 blood pressure control rate at 9 months, and the proportion of hypertension visits with 18 acceptable treatment. Acceptable treatment is defined as either appropriate treatment 19 or non-appropriate treatment with reasons for failing to titrate treatment. Exploratory 20 outcomes include a composite of cardiac death, non-fatal stroke, and non-fatal 21 myocardial infarction. (Table 2)  22   . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 1

We initially assumed that at least 10 primary care practices are randomized to the 3 intervention arm and 10 to the control arm, and the baseline appropriate treatment 4 rate is 55% with maximum type II error of α=0.05. With 20 practices, assuming a 5 moderate intra-site correlation of 0.05, a within-patient correlation of 0.1, and a 6 statistical power of 90%, we needed 3 hypertension visits per patient for 50 patients at 7 each site in order to detect a 18% absolute difference in appropriate treatment rate 8 between the two arms. 9

Although we based our initial planning and site recruitment on this sample size 10 calculation, we currently have 100 sites that are or will be randomized. Under the 11 same assumptions as above but with 50 intervention and 50 control sites, we will be 12 able to detect a 4% absolute difference in appropriate treatment rate between arms. 13

The analyses and reporting of the results will follow the Consolidated Standards 14 of Reporting Trials guidelines for cluster randomized controlled trials. 23 All the 15 intervention evaluations will be performed on an intention-to-treat basis. Multiple 16 imputation by chained equations will be used to account for missing values, for both 17 explanatory and outcome variables. 18

The baseline characteristics of patients will be analyzed to assess cluster 19 differences between the intervention and control groups. We will summarize 20 continuous variables as median with interquartile ranges and categorical variables as 21 frequency with percentage. With all comparative outcomes, absolute differences with 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 95% CIs will be presented and adjusted by patient and site baseline characteristics. 1 Implementation stages will be treated as strata, with adjustment for calendar time to 2 account for secular trends. The analysis of both primary and secondary outcomes will 3 account for the clustering effect using mixed-effects models with primary care practice 4 as a random effect. The consistency of treatment effects on the primary outcome will 5 be explored in predefined subgroups, including age, gender, education, 6 implementation stage, and tertile of cluster-level endpoints. All statistical tests will be 7 performed using 2-sided tests at the 0.05 level of significance. 8 9

The LIGHT-ACD trial 10

The LIGHT-ACD trial aims to include all participants in the intervention sites of 12 the LIGHT trial who are not taking antihypertensive medication or taking only one 13 medication which is not a beta-blocker and with a SBP ≥ 140 mm Hg. Key exclusion 14 criteria includes diabetes mellitus and intolerance to at least one class of 15 antihypertensive medications (Table 1) . 16

Participants in the LIGHT-ACD trial are categorized into 2 subpopulations. 17

Participants with a SBP of 140-159 mm Hg, and not taking any antihypertensive 18 medication are categorized as Population 1, the reminder as Population 2. 19 20

Populations 1 and 2 are randomized separately. Six three-step protocols are 22 integrated into the algorithm of CDSS. Population 1 are randomized to receive one of 23 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. ; https://doi.org/10.1101/2021.03.11.21253427 doi: medRxiv preprint

the initial monotherapies of A (angiotensin-converting enzyme inhibitor or angiotensin 1 receptor blocker), C (calcium channel blocker), or D (diuretics). Subsequently, the 2 participants initiated with A are randomized to add C or D following protocol 3 A-AC-ACD or A-AD-ADC, respectively, if necessary, to achieve blood pressure control. 4

Similar randomization procedures are applied in participants initiated with C or D. 5

Population 2 are randomized to receive one of the three initial dual therapies of AC, 6 AD, or CD, and then D, C, or A is added to achieve blood pressure control, 7 respectively ( Figure 2 ). Minimized randomization is used to ensure balance by age, 8 gender and education level among the three arms of the two populations. Neither 9 participants nor physicians are blinded to treatment allocation but the allocation is 10 concealed within the CDSS. 11

The assignment of treatment is presented as the medication recommendation 14 (class and dose) by the CDSS. The specific agent within each class is at the 15 physician's discretion based on the available medications at the primary care 16

practices. 17

For each case, the titration of antihypertensive medications is performed 18 automatically by CDSS according to the assigned treatment protocol. Participants 19 who are unable to follow their protocols because of a new onset of complications (e.g., 20 coronary heart disease) receive usual care. Those who are unable to follow their 21 regimen because of medication intolerance, are assigned to a new protocols 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. 

The primary outcome is the change in blood pressure from baseline to 9 months 4 of different regimens of initial therapy. Secondary outcomes include the proportion of 5 individuals with SBP <140 mm Hg and DBP <90 mm Hg at 9 months; the proportion of 6 individuals with SBP <160 mm Hg and DBP <100 mm Hg at 9 months; the proportion 7 of individuals who received monotherapy (only in Population 1), dual therapy, triple 8 therapy, and referral at 9 months; the proportion of individuals reported to have 9 antihypertensive drug related side-effects; and the proportion of individuals 10 transferred to usual care for any reasons. The exploratory outcome is the change in 11 blood pressure from baseline to 9 months of different protocols. (Table 2 ) 12 13

We assume approximately 25% of the LIGHT intervention patients are in 15

Population 1 and 75% in Population 2, with an 80% follow-up rate for the primary 16 outcome. For each population, we estimate the detectable difference in SBP between 17 treatment groups across a similar range of the intervention participants and statistical 18 power. We assume that the standard deviation in SBP is σ=10 mmHg, and that the 19 within-patient SBP correlation is R 2 =0.2 with a maximum type II error that is 20

Sidak-corrected for three comparisons, α=0.017. 24 With 100 LIGHT sites, we estimate 21 at least 2100 eligible LIGHT-ACD participants overall with complete follow-up. Under 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) these assumptions, we estimate that for Population 1 comparisons we will have 80% 1 power to detect a difference of 3.5 mm Hg in SBP, and for Population 2 comparisons, 2 we have 80% power to detect a difference of 2 mm Hg in SBP. 3

All the intervention evaluations will be performed on an intention-to-treat basis. 4

Multiple imputation by chained equations will be used to account for missing values, 5 for both explanatory and outcome variables. 6

We will use frequencies with percentages to describe categorical variables and 7 means with SDs to describe continuous variables unless skewed, which we present 8 as medians and interquartile ranges. The differences between the three groups will be 9 assessed either by univariate analyses of variance (ANOVA) or by χ 2 tests. For 10 pairwise testing of primary outcomes, multiple Student t tests or Mann-Whitney U 11 tests will be used; P values will be adjusted for multiple comparisons by using the 12 Sidak method. As secondary analyses, the primary end points will be adjusted for 13 baseline blood pressure values by analysis of covariance (ANCOVA). For secondary 14 outcomes, log-binomial regression will be used to compare groups and calculate 15 relative risk of outcomes at 9 months. 16

Additionally, we will perform pre-specified subgroup analyses of outcomes by age, 17 sex, education, smoking status, and tertile of baseline blood pressure. 18 19 DISCUSSION 20

The LIGHT trial, to the best of our knowledge, is the largest pragmatic 21 randomized trial showing the feasibility and effectiveness of a new model of delivering 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) high-quality care for hypertension in primary care settings. Moreover, by adopting a 1 streamlined study design, we embed a patient-randomized controlled trial 2 (LIGHT-ACD study) into a cluster-randomized trial through an algorithm-enabled 3 CDSS tool, representing a contemporary paradigm of clinical research to improve the 4 efficiency of the trial and accelerate the generation of evidence through electronic 5 health systems, standardized treatment regimens, and decision support systems. 6

Our studies have several strengths. First, we developed a usable CDSS, which 7 can seamlessly integrate into clinical routine workflow and provide tailored 8 antihypertensive recommendations at the point of care. These features are highly 9 correlated with effective CDSS for improving process of care and patient 10 outcomes. 25, 26 The use of a CDSS in primary care may reduce the heterogeneity of 11 care due to the lack of qualified doctors for hypertension management in China. As 12 recommendations and alerts of CDSS are generated automatically by the built-in 13 algorithm, which was developed based on current guidelines, this approach can thus 14 assist primary care doctors, even those with less training, in making informed and 15 evidence-based medical decisions. 16 Second, we have built a streamlined framework for a clinical trial that enables us 17 to compare the effectiveness of several guideline-based initial antihypertensive 18 therapies. Earlier randomized clinical trials such as the ALLHAT 14 and 19 ACCOMPLISH 27 trials, had provided a direct comparison among several 20 monotherapies or dual therapies. In contrast with these standalone trials, the conduct 21 of the LIGHT-ACD trial is embedded into the existing framework of the LIGHT trial. We 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) have incorporated a series of stepped treatment protocols into the CDSS, whereby 1 the randomized allocation of recommended medication can be performed 2 automatically by following the algorithm-consistent order at each encounter where 3 decision support is delivered. 28 While assessing the effectiveness of CDSS, we can 4 also compare the effectiveness of common initial antihypertensive monotherapies or 5 dual therapies in an unobtrusive manner. 16,27,29,30 6 Third, the pragmatic design of both trials, are built on the infrastructure of the 7 electronic health records already routinely used in primary care practices. Although 8 traditional explanatory trials remain a key tool for demonstrating the efficacy of 9 intervention/treatment in highly controlled settings, the pragmatic design can deliver 10 real-world effectiveness with greater external validity. 31, 32 In contrast to trials with 11 study-specific visits, the enrollment and follow-up of patients, and the collection of 12 outcome data in our trial are incorporated into routine clinical practice. Moreover, the 13 exclusion criteria are kept to a minimum to enroll a diverse spectrum of the population. 14 These considerations improve the efficiency of trials and enhance generalizability of 15 the study results. 32 16

Fourth, the two studies are further distinguished by their efforts to build a learning 17 decision support tool. Apart from basic functions of CDSS such as medication 18 recommendation and alerts, the tool itself can generate new knowledge in terms of 19 the effectiveness of treatment strategies embedded in the CDSS from the ongoing 20 delivery of care. These study results, in turn, can be used to adaptively improve CDSS 21 by shifting the randomization ratio of stepped antihypertensive protocols toward the 22 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. ; https://doi.org/10.1101/2021.03.11.21253427 doi: medRxiv preprint P a g e 1 9 o f 3 3 more effective group. 33 The updated CDSS can also be iteratively implemented, 1 tested, and improved. 2

Our study has some potential limitations. First, our study outcomes are focused 3 on surrogate outcomes instead of clinical outcomes such as death or vascular events. 4

To examine the effectiveness on clinical events, a much larger and longer trial would 5 be required. However, it is expected that improvements in blood pressure control over 6 time would favorably affect clinical outcomes. Second, given the nature of CDSS, 7 which delivers its recommendation directly to physicians, blinding was not feasible in 8 both studies. We minimized the potential bias by using objective measures as primary 9

and secondary outcomes. Third, due to the limited timeframe of the study, an 10 extended follow-up period was not included following the 12-month intervention to 11 measure persistence of effects after the intervention ceases. 12

In conclusion, these two trials will provide reliable evidence regarding the 13 effectiveness of CDSS on improving adherence to guidelines for hypertension 14 management in primary care, and data on the effectiveness of different initial 15 antihypertensive regimens for blood pressure reduction in the real-world setting. 16 17 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. ; https://doi.org/10.1101/2021.03.11.21253427 doi: medRxiv preprint P a g e 2 6 o f 3 3 algorithm of the CDSS. All authors contributed to critical revisions and approved the final version of the article.

Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

This project was supported by the CAMS Innovation Fund for Medical Science Health Information for work to advance intelligent disease prevention and health promotion; collaborates with the National Center for Cardiovascular Diseases in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. ; https://doi.org/10.1101 /2021 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. ; https://doi.org/10.1101/2021.03.11.21253427 doi: medRxiv preprint P a g e 2 9 o f 3 3

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

Population-1: Participants with systolic blood pressure (SBP) of 140-159 mm Hg, and were not taking any antihypertensive medication. Population-2: Participants with SBP ≥ 160 mm Hg and were not taking any antihypertensive medication or taking one antihypertensive medication which was not beta-blocker, or those with SBP 140-159 mm Hg and were taking one antihypertensive medication which was not beta-blocker. A: angiotensin-converting enzyme inhibitor or angiotensin receptor blocker; C: calcium channel blocker; D: diuretic.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Currently in an acute episode of disease Currently pregnant or breastfeeding, or planning a pregnant or breastfeeding during the study Cognitive or communication disorders *Participants who were not eligible for the LIGHT study at the first screening visit were re-assessed for eligibility at the subsequent visits until the end of recruitment. # Participants who were not eligible for the LIGHT-ACD study at the first screening visit would not be re-assessed for eligibility at the subsequent visits. † Including angina, myocardial infarction, coronary artery bypass grafting, percutaneous coronary intervention, >50% stenosis of coronary artery, .

CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021. Absolute change in blood pressure at 9 months of different regimens of initial therapy Secondary Absolute change of systolic blood pressure at 9 months Proportion of individuals with SBP <140 mm Hg and DBP <90 mm Hg at 9 months Blood pressure control rate at 9 months Proportion of individuals with SBP <160 mm Hg and DBP <100 mm Hg at 9 months Acceptable treatment rate among all post-randomization hypertension visits Proportion of individuals who received monotherapy*, dual therapy, triple therapy, and referral at 9 months Proportion of individuals with antihypertensive drug side-effects Proportion of individuals transferred to usual care for any reason

A composite of cardiac death, non-fatal stroke, and non-fatal myocardial infarction

Absolute change in blood pressure at 9 months of different protocols* SBP, systolic blood pressure; DBP, diastolic blood pressure. Primary and secondary outcomes of LIGHT-ACD were assessed among initiating therapies; exploratory outcomes of LIGHT-ACD were assessed among protocols. *Only assessed in Population 1, who are not currently taking any antihypertensive medication with systolic blood pressure 140-159 mm Hg, and initiated with monotherapies . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 12, 2021.

; https://doi.org/10.1101/2021.03.11.21253427 doi: medRxiv preprint

Quality of primary health care in China: 13 challenges and recommendations. The Lancet

Launch of the health-care reform plan in China