OHDSI Home | Forums | Wiki | Github

Study on comparison of combination treatment in hypertension

The new study below is planned to post to the OHDSI Research Network.

Comparison of combination treatment in hypertension

Objective: The objective of this study is to compare the effectiveness and adverse events between the combination treatments in hypertension

Rationale: The goal of antihypertensive therapy is to reduce cardiovascular end points including stroke, myocardial infarction, and heart failure by lowering blood pressure. Although it is evident that BP reduction per se is the primary determinant of CV risk reduction, the choice of initial drug therapy can exert some effect on long-term outcomes. Many large randomized trials have shown that two or more antihypertensive agents are required for reaching their treatment goals. Furthermore, recent data have suggested that the use of combination therapy in patients with hypertension may be beneficial for blood-pressure-lowering efficacy, obtaining blood pressure goals earlier, and reducing major adverse cardiovascular events. To date, However, the best combination treatment in hypertension have not been demonstrated. The evidence through OHDSI network can help clinicians to select the combination treatment for their patients.

Project Leads: Seng Chan You, Sungjae Jung, and Rae Woong Park from Ajou university

Please provide any comments or suggestions

Hi All,

Study on comparison of combination treatment in hypertension is available now.

The proposed OHDSI study protocol and analytic code has been posted on the Wiki and on GitHub.

Please provide any comments or suggestions. If you would like to join this study, please let me know.

Thanks,

Seng Chan You.

I have a question about CohortMethod package.

I use CohortMethod package for the analysis.

I’ve just found a strange thing in my study.

Figures below describes standardized difference of mean and balance before and after PS matching.

Target Cohort (T) : patients initiating ACEi/ARB + beta-blocker for hypertension
Comparator Cohort (C) : patients initiating ACEi/ARB + calcium channel blocker for hypertension
Outcome Cohort (O) : all-cause mortality

The differences in prevalence of COPD or diabetes mellitus between target and comparator cohort increase after matching.

Is there anyone who knows why this difference increases after matching in my analysis?

Settings for CohortMethod below:

covarSettings ← FeatureExtraction::createCovariateSettings(useCovariateDemographics = TRUE,
useCovariateDemographicsGender = TRUE,
useCovariateDemographicsRace = TRUE,
useCovariateDemographicsEthnicity = TRUE,
useCovariateDemographicsAge = TRUE,
useCovariateDemographicsYear = TRUE,
useCovariateDemographicsMonth = FALSE,
useCovariateConditionOccurrence = TRUE,
useCovariateConditionOccurrence365d = TRUE,
useCovariateConditionOccurrence30d = TRUE,
useCovariateConditionOccurrenceInpt180d = TRUE,
useCovariateConditionEra = TRUE,
useCovariateConditionEraEver = TRUE,
useCovariateConditionEraOverlap = FALSE,
useCovariateConditionGroup = TRUE,
useCovariateConditionGroupMeddra = TRUE,
useCovariateConditionGroupSnomed = FALSE,
useCovariateDrugExposure = TRUE,
useCovariateDrugExposure365d = TRUE,
useCovariateDrugExposure30d = TRUE,
useCovariateDrugEra = TRUE,
useCovariateDrugEra365d = FALSE,
useCovariateDrugEra30d = FALSE,
useCovariateDrugEraOverlap = FALSE,
useCovariateDrugEraEver = TRUE,
useCovariateDrugGroup = TRUE,
useCovariateProcedureOccurrence = TRUE,
useCovariateProcedureOccurrence365d = TRUE,
useCovariateProcedureOccurrence30d = TRUE,
useCovariateProcedureGroup = TRUE,
useCovariateObservation = FALSE,
useCovariateObservation365d = FALSE,
useCovariateObservation30d = FALSE,
useCovariateObservationCount365d = FALSE,
useCovariateMeasurement = FALSE,
useCovariateMeasurement365d = FALSE,
useCovariateMeasurement30d = FALSE,
useCovariateMeasurementCount365d = FALSE,
useCovariateMeasurementBelow = FALSE,
useCovariateMeasurementAbove = FALSE,
useCovariateConceptCounts = TRUE,
useCovariateRiskScores = TRUE,
useCovariateRiskScoresCharlson = TRUE,
useCovariateRiskScoresDCSI = TRUE,
useCovariateRiskScoresCHADS2 = FALSE,
useCovariateRiskScoresCHADS2VASc = FALSE,
useCovariateInteractionYear = FALSE,
useCovariateInteractionMonth = FALSE,
excludedCovariateConceptIds = excludedCovariateConceptIds,
includedCovariateConceptIds = includedCovariateConceptIds,
deleteCovariatesSmallCount = 50)

getDbCmDataArgs ← CohortMethod::createGetDbCohortMethodDataArgs(studyStartDate = “”,
studyEndDate = “”,
excludeDrugsFromCovariates = FALSE,
firstExposureOnly = FALSE,
removeDuplicateSubjects = TRUE,
washoutPeriod = 0,
covariateSettings = covarSettings)

createStudyPopArgs1 ← CohortMethod::createCreateStudyPopulationArgs(firstExposureOnly = FALSE,
washoutPeriod = 0,
removeDuplicateSubjects = TRUE,
removeSubjectsWithPriorOutcome = TRUE,
minDaysAtRisk = 0,
riskWindowStart = 0,
addExposureDaysToStart = FALSE,
riskWindowEnd = 0,
addExposureDaysToEnd = TRUE)

Fixing seed for reproducability:

createPsArgs ← CohortMethod::createCreatePsArgs(prior = Cyclops::createPrior(“laplace”,
exclude = c(0),
useCrossValidation = TRUE),
control = Cyclops::createControl(noiseLevel = “quiet”,
cvType = “auto”,
tolerance = 2e-07,
cvRepetitions = 10,
startingVariance = 0.01,
threads = 1))

matchOnPsArgs1 ← CohortMethod::createMatchOnPsArgs(caliper = 0.25,
caliperScale = “standardized”,
maxRatio = 1.0)

omExcludedConcepts ← c()
omIncludedConcepts ← c()
fitOutcomeModelArgs1 ← CohortMethod::createFitOutcomeModelArgs(modelType = “cox”,
stratified = FALSE,
useCovariates = FALSE,
excludeCovariateIds = omExcludedConcepts,
includeCovariateIds = omIncludedConcepts)


Hi Seng: The study diagnostics suggest that the propensity score
adjustment strategy did not achieve adequate balance in many covariates.
This can happen for several different reasons, including: 1) the size of
one or both of the cohorts was small, so that the propensity score model
did not have sufficient data to fit a good predictive model. 2) the two
cohorts were very different and there are one or more highly predictive
covariates that could discriminate the group (of which, some may be
covariates that should be excluded from model fitting, 3) the adjustment
strategy (matching/stratification) did not create well-balanced strata,
Given what you have shared here, I suspect it might be either #1 or #2
(not likely #3 since you are doing 1-1 matching).

Thank you, @Patrick_Ryan
I think you’re right (#1). Because of small number of subjects in cohort, this problem occurred.
I learned a good lesson from this. Thanks!!

Hi Chan,

Some crude tests I did a while back indicate that you need at least 1,000 subjects in each group to get an accurate propensity model.

Hope that helps.

The abstract using Korean data for this study (‘Comparison between first line dual-combination therapies in hypertension: nationwide population based cohort study’) was accepted for Advances In Science Presentation in European Society of Cardiology Annual Congress 2017. This abstract will be presented in combination with ‘state of the art’ expert lecture about ‘Strategies to improve the treatment of hypertension’.

If anyone join the ESC 2017 in Barcelona, please let me know.

Finally, I and @zai have completed the code for study on comparison of combination treatment in hypertension.
See details on the wiki ,full protocol , and code.

The Korean result of this study was accepted to present in the ‘Highlight’ Session in European Society of Cardiology Congress 2017.

Please provide any comments or suggestions.
If you would like to join as a contributing investigator, contributing to the analysis and write-up of this work, please let us know.

Thanks,

Chan and Sungjae

Thank you for ‘Best community contribution award’ in OHDSI symposium 2017. @Patrick_Ryan

As I presented, We’re recruiting the data partners to join the study for ‘hypertension combination treatment’.

We found that there is no significant difference between combination treatment in Korean population. However, the main outcome of hypertension is myocardial infarction in Western patients whereas the main outcome is stroke Asian patients. So I think there is heterogeneity of the result from combination treatment across the world. Until now, no one knows best regimen for initiating hypertension treatment. I’m convinced the evidence generated from this study can help vast majority of population in the world.

The protocol is available in wiki: http://www.ohdsi.org/web/wiki/doku.php?id=research:comparison_of_combination_treatment_in_hypertension
The code is available in GitHub: https://github.com/OHDSI/StudyProtocolSandbox/tree/master/HypertensionCombination

Please send me an email to ‘applegna@gmail.com’ if you have any question or can join this study.
Thank you again!

@SCYou We will try to run this on our Truven dataset here at GT.

Thanks!

Jon

Is this a known and established fact, or should we test that first, before any treatment decision?

@Christian_Reich, I always admire your keen point!

‘Asian countries such as Japan, South Korea, China, Hong-Kong, Taiwan and Thailand have the characteristic features of higher stroke mortality and morbidity than of coronary heart disease, while Western countries have vice versa. It has been discussed for a long time whether or not Japanese people and some other Asian populations have different risk factors for CHD than Western populations.’ [Ueshima et al., Circulation, 2008]

We don’t know whether this heterogeneity is caused by hypertension. But still, the main fatal outcome after hypertension is stroke in Asia whereas it is myocardial infarction in Western countries. Available evidences from RCTs are conflicting about combination treatment in hypertension(one from US[Jamerson et al., 2008 NEJM], the other one from Japan[Matsuzaki et al., J Hypertension 2011]). I think one of reasons accounting for these competing results is the variance in incidence of CHD and stroke between Asian and Western countries.

Although each class of anti-hypertensive agent has similar protective effects for overall risk reduction, each class has its own strength over others (eg., CCB has minor additional effect in prevention stroke [Law et al., BMJ, 2009]). For these reasons, I think the result might be different across the world in this study.

@jon_duke, Thank you for your help! please tell me, if there is any problem in my code or any question!

Chan

t