Abstract
Background
The DCP (Diuretic Comparison Project), a pragmatic trial, evaluated whether chlorthalidone compared with hydrochlorothiazide would reduce the risk of nonfatal cardiovascular disease or noncancer‐related death. The intent‐to‐treat analysis found no difference in such comparison (hazard ratio, 1.04 [95% CI, 0.94–1.16]). The objective of the current study is to estimate the per‐protocol effect of chlorthalidone (12.5/25 mg daily) compared with hydrochlorothiazide (25/50 mg daily) in preventing major adverse cardiovascular events among older patients with hypertension.
Methods
The effect of adhering to treatment strategies was assessed by censoring at first instance of nonadherence, defined as a gap (>90‐day gap in drug coverage), switch (switching between study medications), and discontinuation (stop taking chlorthalidone/hydrochlorothiazide altogether for >90 days before the end of the study). The primary outcome was a composite of nonfatal myocardial infarction, stroke, heart failure resulting in hospitalization, urgent coronary revascularization for unstable angina, and noncancer‐related death. Using inverse probability weighting, we evaluated the per‐protocol effect of chlorthalidone and hydrochlorothiazide using DCP trial data.
Results
Nonadherence was found in 5476 (40%) participants; among 8047 (60%) adherents, 3905 (49%) were randomized to chlorthalidone and 4142 (51%) to hydrochlorothiazide. After censoring time when participants deviated from the assigned treatments, the estimated 5‐year risk ratio of the composite primary outcome of nonfatal cardiovascular disease and noncancer‐related death was 1.36 (95% CI, 0.96–2.12) in chlorthalidone compared with hydrochlorothiazide.
Conclusions
The per‐protocol analysis indicated a lower risk with hydrochlorothiazide compared with chlorthalidone in preventing nonfatal cardiovascular disease and noncancer‐related death; however, this difference was not statistically significant using dispensation data to identify adherence.
Registration
URL: https://www.clinicaltrials.gov; Unique identifier: NCT02185417.
Keywords: cardiovascular disease, diuretic, hypertension, nonadherence, per‐protocol
Subject Categories: Epidemiology, Myocardial Infarction, High Blood Pressure, Hypertension
Nonstandard Abbreviations and Acronyms
- DCP
Diuretic Comparison Project
- ITT
intention‐to‐treat
- MPR
medication possession ratio
- VA
Veterans Affairs
Clinical Perspective.
What Is New?
This per‐protocol analysis showed that there was no significant difference between the treatments of chlorthalidone and hydrochlorothiazide in the composite outcome, which included nonfatal myocardial infarction, stroke, heart failure resulting in hospitalization, urgent coronary revascularization for unstable angina, and noncancer‐related death when accounting for treatment adherence.
What Are the Clinical Implications?
Both chlorthalidone and hydrochlorothiazide may be used for the treatment of hypertension in older patients.
The prevalence of hypertension is rising globally, and identifying an effective treatment is important for public health 1 because preventive drugs are available at relatively low cost and are shown to be effective in reducing cardiovascular and hypertension‐induced health outcomes. 2 The first line of antihypertensive medications that prevent cardiovascular outcomes and lower blood pressure are thiazide diuretics. 3 Among these thiazide diuretics are hydrochlorothiazide and chlorthalidone.
The DCP (Diuretic Comparison Project) was a large, pragmatic comparative effectiveness research trial embedded within the Veterans Affairs (VA) health care system. 4 Corresponding study outcomes were captured from real‐world data encompassing nationwide VA electronic health records (EHRs) and non‐VA administrative claims supplied by the Centers for Medicare and Medicaid Services, as well as National Death Index data. There were 13 523 adults with hypertension aged ≥65 years randomized across 72 VA medical centers in the DCP trial over a 5‐year study period. The primary hypothesis was that chlorthalidone might be better than hydrochlorothiazide in preventing major nonfatal cardiovascular events among older patients with hypertension. However, the intention‐to‐treat (ITT) analysis showed no difference in the risk of the primary outcome between treatment groups, suggesting that chlorthalidone was not superior to hydrochlorothiazide in preventing major cardiovascular events and noncancer‐related death. 5 The ITT analysis provided a fundamental comparison of the 2 thiazide diuretics for antihypertensive treatment. A per‐protocol analysis was subsequently conducted as a secondary analysis according to the DCP study plan to evaluate treatment effects with consideration of overall adherence to the allocated interventions.
Per‐protocol analysis assesses treatment effects if participants adhered to their prescribed treatment according to the trial protocol. By complementing the ITT effect of the diuretics with the effect estimates of the diuretics when patients adhere to their treatment as detailed in the protocol, also known as the per‐protocol effect, the clinicians and patients will be able to make more informed decisions about their hypertension treatment. 6 Both the ITT and per‐protocol results, obtained from a geographically and socioeconomically diverse sample, are crucial for guiding clinical decisions on antihypertensive treatment. The DCP trial randomized 13 523 patients to evaluate a less frequently used chlorthalidone and a more frequently prescribed hydrochlorothiazide. Our current analysis focused on outcome data and follow‐up time when the randomized treatments were maintained: to continue with the existing hydrochlorothiazide treatment regimen (25 or 50 mg daily) or switch to comparably dosed chlorthalidone (12.5 or 25 mg daily). This article aims to estimate the per‐protocol effects of chlorthalidone versus hydrochlorothiazide in preventing the primary composite outcome of nonfatal myocardial infarction (MI), stroke, heart failure (HF) resulting in hospitalization, urgent coronary revascularization for unstable angina, and noncancer‐related death, using real‐world clinical data collected in the DCP trial.
Methods
Data Availability Statement
Data collected for the trial will not be made available to others. VA clinical data cannot be effectively deidentified. However, researchers with an approved institutional review board and signed data agreement with investigator support may request the data.
Diuretic Comparison Project
The DCP was the first large‐scale pragmatic trial integrated with VA primary care, conducted between June 2016 and June 2022 (mean follow‐up, 2.4±1.4 years). Study design and recruitment were reported in detail elsewhere. 4 Briefly, 72 VA medical centers and 537 of their managed primary care clinics agreed to participate, with 4128 primary care providers (PCPs) providing informed consent via EHR.
Eligible patients were older (aged ≥65 years), had hypertension (baseline systolic blood pressure ≥120 mm Hg), and received hydrochlorothiazide at 25 or 50 mg daily. A total of 13 523 patients were enrolled and provided verbal consent. Those randomized to the hydrochlorothiazide group were asked to continue filling their existing prescription orders. For those randomized to chlorthalidone, physician order requests were sent to the associated PCPs to discontinue hydrochlorothiazide and start chlorthalidone prescriptions (12.5 or 25, respectively). PCPs could modify or discontinue the allocated treatments at any time during the study, analogous to usual clinical care.
All trial activities were performed in tandem with VA routine care practice. There were no standardized study clinical visits, and real‐world data were used for outcome ascertainment, safety surveillance, and patient follow‐up. Such a pragmatic trial design enabled patients from urban and rural areas to participate in the DCP study, including those from Alaska, Hawaii, and Puerto Rico. 7 This project was approved by the VA Central Institutional Review Board, with waivers of consent documentation and Health Insurance Portability and Accountability Act authorization.
Study Outcomes
Outcome definitions were consistent with the ITT analysis, which included a primary composite outcome of nonfatal MI, stroke, HF resulting in hospitalization, urgent coronary revascularization for unstable angina, and noncancer‐related death. 5 , 8 , 9 , 10 Matching data sets, containing all years of VA data, Medicare claims through 2021, and National Death Index records through 2019, were used. 5 Medicare data were used to capture inpatient encounters outside of the VA; National Death Index data were linked to confirm causes of death.
Treatment Monitoring
As mentioned, PCPs could alter the randomized dose, switch between study drugs, discontinue the study drug, or change to a different antihypertensive agent according to the patients’ needs (ie, per usual care). Corresponding drug orders were filled by the VA outpatient pharmacies, and resultant dispensation records were used for treatment monitoring. These data were obtained directly from administrative records found in the VA Corporate Data Warehouse, a national EHR data repository for the entire veteran population using the Veterans Health Administration.
Per‐Protocol Definition
The primary trial’s pragmatic study design allowed for PCPs to administer treatment as part of routine clinical care allowing for switching of medications, discontinuation of medications, and so on, as described in the above sections. However, this current analysis used a static treatment strategy (ie, participants were considered nonadherent regardless of the reason for stopping treatment) due to limitations of acquiring reasons for the PCP switching, discontinuing, gaps, and the like of the treatments to which participants were randomized for the trial. 6 As such, this analysis does not consider adherence to usual care to be per protocol. Instead, we thus refer to this analysis in terms of the patient being adherent to the treatment to which they are randomized, and the definition of adherence for this study is defined in further detail below.
Adherence Definitions
DCP protocol initially proposed to apply a medication possession ratio (MPR) of <80% for nonadherence. This classification was revised subsequently on the basis of the clinical decisions made by patients’ PCPs and prescription fill patterns observed during the study. We accordingly redefined adherence to enhance the sensitivity of per‐protocol effect estimates. Both chlorthalidone and hydrochlorothiazide were typically filled with a 90‐day supply (96% of total fill records), with few instances of 30 days (3%) and 1% between 1 and 89 days. In combination with clinical inputs from cardiologists and previous studies also analyzing diuretics for antihypertensive treatment, a 90‐day cutoff was applied to define treatment nonadherence in this analysis. 11 , 12 Deviation of protocol occurred at the first instance of noncompliance (gap or switch) and the definition of discontinuation described below:
Discontinuation: This was detected by comparing time between last drug filled and the individual ending time point (earliest date of death; study withdrawal; or June 1, 2022). When dispensation records showed no fills of either study medication for >90 days before study end, follow‐up time was accordingly censored as last drug release date plus 30‐/90‐day supply plus 90‐day grace period. These individuals were considered drug discontinuation cases.
Gap: There was a >90‐day gap in drug coverage between fills of the assigned study medication. Time to follow‐up was determined as last drug release date plus 30‐/90‐day supply plus a 90‐day gap. Although the computation strategy was similar to discontinuation, this was captured by comparing the drug coverage between fills, rather than the last drug filled before study end.
Switch: Dispensation records showed a filling of the other study medication and did not switch back to the randomized treatment for >30 days. If records showed a brief period of switching with patients converted back to the randomized treatment within 30 days, then it was not considered a crossover. Time to follow‐up was censored when a steady switch was detected and was determined as the earliest date receiving the other study medication (contrary to the randomized order).
Study Covariates
The covariates included baseline prognostic factors (age, race, sex, diabetes status, and prior MI or stroke) that predict adherence to the study treatment. Age was used as a continuous variable, and sex was a categorical variable categorized as male or female. Race was categorized into White, Asian, Black or African American, Pacific Islander, American Indian or Native American/Hawaiian, multiple race, and unknown. History of diabetes was based on the past 2 years of outpatient records and determined as ≥2 occurrences of International Classification of Diseases (ICD) diagnosis or an ICD diagnosis plus a diabetes drug fill. History of MI/stroke was based on all VA records and both dichotomously defined as having or not having a clinical diagnosis before study randomization.
Time‐varying covariates included hypotension (mean arterial pressure <65 mm Hg), hypokalemia (serum potassium level <3.1 mmol/L), hyponatremia (serum sodium level <130 mmol/L), and hypomagnesemia (serum magnesium level below the lower reference range). Hypokalemia and hyponatremia were monitored via EHR data as specified in the DCP study protocol. Hypotension and hypomagnesemia were included in this analysis due to their clinical significance when using thiazides. 13 , 14 These measures were captured over the follow‐up duration, and new occurrences were determined as >7 days from last observed. During analysis, these were incorporated in the model by linking the laboratory or blood pressure date with the corresponding drug covering the period (within 30 days of the study medication dispensation date).
Statistical Analysis
We censored participants (stopped follow‐up) at the first instance of a protocol deviation, either having a gap, switch, or discontinuation to their randomized treatments. We used inverse probability weighting to adjust for prognostic and time‐varying variables that predict adherence to the protocol. To estimate the denominator of the weights, we fit a pooled logistic regression model for treatment with all the covariates, time‐varying factors, and time for follow‐up, and we fit separate models for the different treatment groups. For estimating the numerator of the weights, we fit a similar pooled logistic regression with time of follow‐up and the baseline covariates. We also fit separate pooled logistic regression models for censoring to estimate the denominator and numerator weights for censoring the end of follow‐up. We then calculated the stabilized weights and fit the model for the outcome, which included the baseline covariates. We then standardized over their empirical distribution and estimated the risks over 5 years. The analyses were performed with stratification according to the VA health care system as patients were randomized by a site‐stratification assignment schedule. We used 500 bootstrap samples to construct the 95% CIs. For analyses, we used SAS version 9.4 (SAS Institute, Cary, NC).
Sensitivity Analysis
We used the parametric g‐formula to assess model misspecification and the robustness of our findings (Table S1). 15 This analysis was conducted using the publicly available SAS macro GFORMULA (http://www.hsph.harvard.edu/causal/software/). Sensitivity analyses were also conducted using adherence definitions originally defined in the analysis plan (Table S2). In brief, MPR was computed as the total number of days when the randomized drug was supplied during the study and divided by the patients’ follow‐up time. An overall adherence measure of MPR ≥80% was used as the cutoff for adherence and disregards the specific type of treatment deviations (gap, switch, or discontinuation as described in the current analysis).
Results
There were 13 523 patients randomized in the DCP trial. Table 1 shows that 8047 (60%) patients were considered adherent (3905 chlorthalidone versus 4142 hydrochlorothiazide) and 5476 (40%) nonadherent (2851 chlorthalidone versus 2625 hydrochlorothiazide). A total of 88 patients were censored at start, including 56 who never filled a chlorthalidone/hydrochlorothiazide prescription after randomization and 32 who never received a fill for the medication to which they were randomized. Overall, 2036 (15%) had a gap, 866 (6%) switched, and 2114 (16%) discontinued both study drugs (Figure 1). Among those who were nonadherent, a higher proportion had a comorbid condition compared with the adherent group (45% versus 44% with a history of diabetes; 9% versus 7% with HF; 12% versus 10% with prior MI or stroke). The average time to follow‐up was 1.8±1.1 years when taking medication adherence into consideration.
Table 1.
Baseline Characteristics by Adherence and Treatment
| N=13 523 | Adherence | Total (N=13 523) | |||
|---|---|---|---|---|---|
| Adherent—chlorthalidone (N=3905) | Adherent—hydrochlorothiazide (N=4142) | Nonadherent—chlorthalidone (N=2851) | Nonadherent—hydrochlorothiazide (N=2625) | ||
| Sex | |||||
| Female | 131 (3.35) | 132 (3.19) | 89 (3.12) | 79 (3.01) | 431 (3.19) |
| Male | 3774 (96.65) | 4010 (96.81) | 2762 (96.88) | 2546 (96.99) | 13 092 (96.81) |
| Race | |||||
| Asian | 4 (0.10) | 8 (0.19) | 11 (0.39) | 5 (0.19) | 28 (0.21) |
| Black or African American | 523 (13.39) | 543 (13.11) | 481 (16.87) | 480 (18.29) | 2027 (14.99) |
| Multiple race | 20 (0.51) | 28 (0.68) | 25 (0.88) | 15 (0.57) | 88 (0.65) |
| Pacific Islander, American Indian, or Native American/Hawaiian | 56 (1.43) | 67 (1.62) | 50 (1.75) | 35 (1.33) | 208 (1.54) |
| Unknown due to missing data | 214 (5.48) | 221 (5.34) | 143 (5.02) | 140 (5.33) | 718 (5.31) |
| White | 3088 (79.08) | 3275 (79.07) | 2141 (75.10) | 1950 (74.29) | 10 454 (77.31) |
| History of diabetes | 1702 (43.59) | 1857 (44.83) | 1265 (44.37) | 1205 (45.90) | 6029 (44.58) |
| History of HF | 287 (7.35) | 280 (6.76) | 238 (8.35) | 246 (9.37) | 1051 (7.77) |
| History of MI | 132 (3.38) | 129 (3.11) | 98 (3.44) | 129 (4.91) | 488 (3.61) |
| History of stroke | 296 (7.58) | 271 (6.54) | 238 (8.35) | 224 (8.53) | 1029 (7.61) |
| MI and stroke | 413 (10.58) | 383 (9.25) | 320 (11.22) | 339 (12.91) | 1455 (10.76) |
| Smoking | |||||
| Current | 868 (23.72) | 841 (21.80) | 652 (24.49) | 596 (24.33) | 2957 (23.41) |
| Former | 1701 (46.48) | 1786 (46.31) | 1244 (46.73) | 1109 (45.27) | 5840 (46.24) |
| Never | 968 (26.45) | 1109 (28.75) | 719 (27.01) | 690 (28.16) | 3486 (27.60) |
| Unknown due to missing data | 123 (3.36) | 121 (3.14) | 47 (1.77) | 55 (2.24) | 346 (2.74) |
| Age | 72.38±5.23 | 72.46±5.17 | 72.45±5.54 | 72.45±5.56 | 72.43±5.34 |
| Body mass index | 31.76±5.59 | 31.85±5.70 | 31.60±6.01 | 31.84±6.09 | 31.77±5.81 |
| Systolic blood pressure | 138.34±13.22 | 138.36±13.25 | 139.36±14.08 | 139.99±14.72 | 138.88±13.73 |
Categorical variables are reported as numbers and percentages, and continuous variables are reported as means±SDs. HF indicates heart failure; and MI, myocardial infarction.
Figure 1. Percentage of gaps, switches, and discontinuation at the end of follow‐up by treatment group.
With the per‐protocol analysis when we censored at the first instance of nonadherence, the number of primary outcome events (unweighted) decreased to 467 for hydrochlorothiazide and 448 for chlorthalidone when compared with the ITT analysis. 5 The estimated 5‐year risk of the primary outcome if everyone had initiated and continuously taken hydrochlorothiazide is 0.15, and the estimated 5‐year risk of the primary outcome if everyone had initiated and continuously taken chlorthalidone is 0.20. After adjusting for pre‐ and postrandomization variables, the estimated 5‐year risk ratio of the primary composite outcome (nonfatal MI, stroke, HF resulting in hospitalization, urgent coronary revascularization for unstable angina, and noncancer‐related death) if everyone had initiated and continuously taken chlorthalidone over the follow‐up is 1.36 (95% CI, 0.96–2.12) times than if everyone had taken hydrochlorothiazide over the follow‐up. Also, there were no differences seen on the cumulative incidence risk curves of the primary end point (Figure 2).
Figure 2. Cumulative incidence risk curves (unweighted) for the composite outcome of nonfatal cardiovascular disease event or noncancer‐related death by treatment.
As part of the sensitivity analysis, we assessed the data using adherence as defined by the MPR. There were slightly more participants in the adherent group when applying the ≥80% MPR specification. The HR (hazard ratio) was 1.19 (95% CI, 1.01–1.39), which is similar to the main per‐protocol HR estimate even though it was in a subgroup analysis with a different definition of adherence (Table S2). The parametric g‐formula was used for assessing model misspecification and robustness of the data, and it showed similar results to the inverse probability weighting analysis (Table S1).
Discussion
Using trial data from the DCP, we estimated the per‐protocol effect of chlorthalidone compared with hydrochlorothiazide on the composite outcome, which included nonfatal MI, stroke, HF resulting in hospitalization, urgent coronary revascularization for unstable angina, and noncancer‐related death among older patients with hypertension. As with the primary ITT analysis, 5 the per‐protocol analysis also found that there was no significant difference in treatments with respect to the primary outcome after accounting for medication adherence. However, we were able to see that those on chlorthalidone were more likely to switch to hydrochlorothiazide than those randomized to continue with hydrochlorothiazide. Interestingly, patients on hydrochlorothiazide had more instances of gaps over 90 days than those on chlorthalidone by the end of follow‐up.
The expectation based on previous clinical trials was that chlorthalidone would be more effective in lowering events, given the difference in pharmacodynamics between the 2 diuretics (chlorthalidone is much longer acting and more potent at similar doses as compared with hydrochlorothiazide). Nevertheless, DCP’s primary ITT and current per‐protocol results do not support such theory. According to the per‐protocol analysis, the HRs and adjusted risk ratios indicated a similar directional effect, but the significance was inconsistent. Although the per‐protocol analyses do suggest a potential clinically important harm with chlorthalidone versus hydrochlorothiazide and the sensitivity analysis (based on MPR) is statistically significant, it is possible that the doses predominantly used in DCP could influence the results. In the previous randomized outcome trials showing benefits for each drug, the lowest target dose for chlorthalidone was 25 mg and for hydrochlorothiazide was 50 mg: Only 5% of DCP participants were on these doses at randomization due to the pragmatic nature of the trial. Participants could not have their doses increased by the trial investigators because there were no local investigators or coordinators to monitor the effects of dose adjustments. 16 , 17
The pragmatic trial design of the DCP allowed for assessing the effect of the 2 different diuretics for treating hypertension by embedding the trial within the EHR system. This allowed for the elimination of trial‐related visits, dedicated study staff at the local site, and interventions that were outside the scope of routine clinical visits. The pragmatic design along with the ITT analysis provided effect estimates on real‐world clinical outcomes based on randomization of the treatment without taking adherence into consideration. The per‐protocol effect allowed another step forward in informing clinicians and patients of how the clinically available medications affect outcomes if they are taken as prescribed. 6 The pragmatic design allowed us to assess adherence as represented by real‐world use, unlike traditional trials. The medication adherence levels in our study, nevertheless, were what is seen in clinical practice and similar to data seen with the VA health care system for treatment with diuretics. 18 , 19 Furthermore, the adherence mimics what we see in clinical practice, and our results from the per‐protocol analysis are similar to those from the main analysis indicating that patients will get similar benefit from the 2 drugs.
The trial was able to gather many key variables as well, including the important time‐varying factors that require clinical biometric and laboratory assessments. Hypokalemia and hyponatremia were captured as part of the expected adverse events during trial follow‐up and thus important to adjust for in our analysis. 5 In a meta‐analysis by Khenhrani et al assessing effectiveness and safety of chlorthalidone and hydrochlorothiazide in patients with hypertension, a higher rate of hypokalemia was found in the chlorthalidone group when compared with hydrochlorothiazide. 20 The DCP trial showed similar observations with patients on chlorthalidone (6.0%) having more hypokalemia than those on hydrochlorothiazide (4.4%). 5 , 13 Additionally, hypomagnesemia and hypotension variables were captured due to their clinical significance. Studies have shown an increased risk of hypomagnesemia with the use of thiazides, which indirectly causes hypotension due to volume depletion and electrolyte imbalances. 13 , 14 , 21 Through successful pragmatic trial design and EHR implementations, the known diuretic responses including hypotension, hypokalemia, hyponatremia, and hypomagnesemia were also assessed under real‐world situations.
Our study had some limitations. Although we conducted a per‐protocol analysis of a randomized trial, it is an observational analysis, and we thus needed to adjust for all important prognostic factors that predict adherence. We were able to adjust for most of these important factors but did not have information about reasons for gaps in the fills, switching of medications, and discontinuation of medications. Physicians, along with their patients, had the option to change medications, but the reasons for doing so were not captured in our data. However, we tried to capture proxies for some potential reasons using laboratory values such as potassium, sodium, and magnesium. Future studies can potentially use the growing methodologies of large language models and artificial intelligence to further fulfill the data gap of patient/physician reasonings for medication switches, gaps in fills, and discontinuation and thus improve any model misspecification. Another limitation was not having the exact dates of factors that led to nonadherence, such as when switches occurred and the like, which can affect follow‐up time and censoring. However, we were able to use pharmacy fill dates to inform our censoring. We also were not able to adjust for some proxy variables for overall health such as body mass index, smoking, and so on due to missingness and timeliness of the data collected with the pragmatic design of the trial in relation to the randomization date. Additionally, health care engagement proxy variables such as clinical visits and all‐cause hospitalization were not adjusted for in the models due to multicollinearity. Another limitation was not being able to assess the effectiveness of the high doses for adherence or outcomes as a majority of the patients were receiving only low doses of 12.5 mg for chlorthalidone and 25 mg for hydrochlorothiazide.
On the other hand, our study had many strengths. First, we had a large sample size that allowed us to assess the per‐protocol analysis with stringent criteria for adherence, and the large sample size allowed for more precise estimates. Second, with the pragmatic trial design, we were able to gather information on a majority of the important prognostic factors that predict adherence, adding to the robustness of our analysis. The per‐protocol analysis results also support the main trial outcomes, which further suggests that in usual care, clinicians can pick either diuretic medication. Additionally, we were able to use the parametric g‐formula, which is another analytical method with modeling assumptions different from inverse probability weighting, to assess the model misspecification with our results, and we found results that were similar to our inverse probability weighting. 15
Conclusions
Overall, taking adherence into consideration using a per‐protocol assessment, we did not find a difference between chlorthalidone and hydrochlorothiazide in the treatment of hypertension in preventing major cardiovascular outcome events or noncancer‐related deaths. Future pragmatic trials should ensure the implementation of adherence measures necessary for estimating the per‐protocol effect in the analysis plan apart from the naïve per‐protocol analysis included in most trials that use MPR and other methods for measuring adherence.
Sources of Funding
This work was supported by the Veterans Affairs Cooperative Studies Program through a grant to the Diuretic Comparison Project.
Disclosures
Dr Cushman reported receiving grants from ReCor and George Medicines institutional grant for hypertension study outside of the submitted work. The remaining authors have no disclosures to report.
Supporting information
Tables S1 and S2
Acknowledgments
The authors thank Alejandro Szmulewicz and Hanna Gerlovin for their insights into causal inference for per‐protocol analyses. The authors also acknowledge the leadership and dedication of their late colleague, Dr Frank Lederle.
This manuscript was sent to Yen‐Hung Lin, MD, PhD, Associate Editor, for review by expert referees, editorial decision, and final disposition.
Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/JAHA.125.046142
For Sources of Funding and Disclosures, see page 8.
References
- 1. Mills KT, Stefanescu A, He J. The global epidemiology of hypertension. Nat Rev Nephrol. 2020;16:223–237. doi: 10.1038/s41581-019-0244-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Lu WL, Yuan JH, Liu ZY, Su ZH, Shen YC, Li SJ, Zhang H. Worldwide trends in mortality for hypertensive heart disease from 1990 to 2019 with projection to 2034: data from the Global Burden of Disease 2019 Study. Eur J Prev Cardiol. 2024;31:23–37. doi: 10.1093/eurjpc/zwad262 [DOI] [PubMed] [Google Scholar]
- 3. Whelton PK, Carey RM, Aronow WS, Casey DE Jr, Collins KJ, Dennison Himmelfarb C, DePalma SM, Gidding S, Jamerson KA, Jones DW, et al. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline for the prevention, detection, evaluation, and management of high blood pressure in adults: a report of the American College of Cardiology/American Heart Association task force on clinical practice guidelines. Hypertension. 2018;71:e13–e115. doi: 10.1161/HYP.0000000000000066 [DOI] [PubMed] [Google Scholar]
- 4. Ishani A, Leatherman SM, Woods P, Hau C, Klint A, Lew RA, Taylor AA, Glassman PA, Brophy MT, Fiore LD, et al. Design of a pragmatic clinical trial embedded in the electronic health record: the VA’s diuretic comparison project. Contemp Clin Trials. 2022;116:106754. doi: 10.1016/j.cct.2022.106754 [DOI] [PubMed] [Google Scholar]
- 5. Ishani A, Cushman WC, Leatherman SM, Lew RA, Woods P, Glassman PA, Taylor AA, Hau C, Klint A, Huang GD, et al. Chlorthalidone vs. hydrochlorothiazide for hypertension‐cardiovascular events. N Engl J Med. 2022;387:2401–2410. doi: 10.1056/NEJMoa2212270 [DOI] [PubMed] [Google Scholar]
- 6. Hernan MA, Robins JM. Per‐protocol analyses of pragmatic trials. N Engl J Med. 2017;377:1391–1398. doi: 10.1056/NEJMsm1605385 [DOI] [PubMed] [Google Scholar]
- 7. Hau C, Efird JT, Leatherman SM, Soloviev OV, Glassman PA, Woods PA, Ishani A, Cushman WC, Ferguson RE. A centralized EHR‐based model for the recruitment of rural and lower socioeconomic participants in pragmatic trials: a secondary analysis of the Diuretic Comparison Project. JAMA Netw Open. 2023;6:e2332049. doi: 10.1001/jamanetworkopen.2023.32049 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Hau C, Woods PA, Guski AS, Raju SI, Zhu L, Alba PR, Cushman WC, Glassman PA, Ishani A, Taylor AA, et al. Strategies for secondary use of real‐world clinical and administrative data for outcome ascertainment in pragmatic clinical trials. J Biomed Inform. 2024;150:104587. doi: 10.1016/j.jbi.2024.104587 [DOI] [PubMed] [Google Scholar]
- 9. Huang H, Turner M, Raju S, Reich J, Leatherman S, Armstrong K, Woods P, Ferguson RE, Fiore LD, Lederle FA. Identification of acute decompensated heart failure hospitalizations using administrative data. Am J Cardiol. 2017;119:1791–1796. doi: 10.1016/j.amjcard.2017.03.007 [DOI] [PubMed] [Google Scholar]
- 10. Raju S, Hau C, Woods P, Flynn M, Sadatis C, McPherson J, Tella A, Ishani A, Ferguson RE, Leatherman SM. Ascertainment of stroke from administrative data to support a pragmatic embedded clinical trial. Contemp Clin Trials. 2023;130:107214. doi: 10.1016/j.cct.2023.107214 [DOI] [PubMed] [Google Scholar]
- 11. Burke TA, Sturkenboom MC, Lu S‐e, Wentworth CE, Lin Y, Rhoads GG. Discontinuation of antihypertensive drugs among newly diagnosed hypertensive patients in UK general practice. J Hypertens. 2006;24:1193–1200. doi: 10.1097/01.hjh.0000226211.95936.f5 [DOI] [PubMed] [Google Scholar]
- 12. Mancia G, Zambon A, Soranna D, Merlino L, Corrao G. Factors involved in the discontinuation of antihypertensive drug therapy: an analysis from real life data. J Hypertens. 2014;32:1708–1716. doi: 10.1097/HJH.0000000000000222 [DOI] [PubMed] [Google Scholar]
- 13. Kieboom BCT, Zietse R, Ikram MA, Hoorn EJ, Stricker BH. Thiazide but not loop diuretics is associated with hypomagnesaemia in the general population. Pharmacoepidemiol Drug Saf. 2018;27:1166–1173. doi: 10.1002/pds.4636 [DOI] [PubMed] [Google Scholar]
- 14. Efstratiadis G, Sarigianni M, Gougourelas I. Hypomagnesemia and cardiovascular system. Hippokratia. 2006;10:147–152. [PMC free article] [PubMed] [Google Scholar]
- 15. Hernan M. Causal inference: what if. Chapman & Hall/CRC; 2020. [Google Scholar]
- 16. ALLHAT Officers and Coordinators for the ALLHAT Collaborative Research Group , The Antihypertensive and Lipid‐Lowering Treatment to Prevent Heart Attack Trial . Major outcomes in high‐risk hypertensive patients randomized to angiotensin‐converting enzyme inhibitor or calcium channel blocker vs diuretic: the Antihypertensive and Lipid‐Lowering Treatment to Prevent Heart Attack Trial (ALLHAT). JAMA. 2002;288:2981–2997. doi: 10.1001/jama.288.23.2981 [DOI] [PubMed] [Google Scholar]
- 17. Amery A, Birkenhäger W, Brixko P, Bulpitt C, Clement D, Deruyttere M, De Schaepdryver A, Dollery C, Fagard R, Forette F, et al. Mortality and morbidity results from the European working party on high blood pressure in the elderly trial. Lancet. 1985;1:1349–1354. doi: 10.1016/s0140-6736(85)91783-0 [DOI] [PubMed] [Google Scholar]
- 18. Siegel D, Lopez J, Meier J. Antihypertensive medication adherence in the Department of Veterans Affairs. Am J Med. 2007;120:26–32. doi: 10.1016/j.amjmed.2006.06.028 [DOI] [PubMed] [Google Scholar]
- 19. Hynes CA, Hau C, Woods P, Leatherman S, Anand ST, Glassman P, Taylor A, Cushman WC, Ishani A, Ferguson R. Medication adherence in the real world: lessons from the diuretic comparison project. J Clin Med. 2025;14:5695. doi: 10.3390/jcm14165695 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Khenhrani RR, Nnodebe I, Rawat A, Adwani R, Ghaffar A, Devi S, Afzal MS, Usama M. Comparison of the effectiveness and safety of chlorthalidone and hydrochlorothiazide in patients with hypertension: a meta‐analysis. Cureus. 2023;15:e38184. doi: 10.7759/cureus.38184 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Palmer BF, Naderi ASA. Metabolic complications associated with use of thiazide diuretics. J Am Soc Hypertens. 2007;1:381–392. doi: 10.1016/j.jash.2007.07.004 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Tables S1 and S2
Data Availability Statement
Data collected for the trial will not be made available to others. VA clinical data cannot be effectively deidentified. However, researchers with an approved institutional review board and signed data agreement with investigator support may request the data.