Key Points
Question
What is the association of deprescribing antihypertensive medication with cognitive function in older residents in nursing homes?
Findings
This target trial emulation approach including 12 644 nursing home residents found that deprescribing antihypertensive medication was associated with less cognitive decline, particularly among those with dementia.
Meaning
These findings suggest the importance of patient-centered approaches to deprescribing antihypertensive medication, ensuring that regimens for older adults are optimized to preserve cognitive function and minimize potential harm.
Abstract
Importance
Antihypertensive medication deprescribing is common among nursing home residents, yet its association with cognitive decline remains uncertain.
Objective
To investigate the association of deprescribing antihypertensive medication with changes in cognitive function in nursing home residents.
Design, Setting, and Participants
This cohort study using a target trial emulation approach included VA long-term care residents aged 65 years or older with stays of at least 12 weeks from 2006 to 2019. Residents who were not prescribed antihypertensive medication, with blood pressure greater than 160/90 mm Hg, or with heart failure were excluded. Eligible residents with stable medication use for 4 weeks were classified into deprescribing or stable user groups and followed for 2 years or until death or discharge for intention-to-treat (ITT) analysis. Participants switching treatment groups were censored in the per-protocol analysis. Cognitive function measurements during follow-up were analyzed using an ordinal generalized linear mixed model, adjusting for confounders with inverse probability of treatment weighting. Per-protocol analysis included inverse probability of censoring weighting. Data analyses were performed from May 1, 2023, and July 1, 2024.
Exposures
Deprescribing was defined as a reduction in the total number of antihypertensive medications or a decrease in medication dosage by 30%, sustained for a minimum of 2 weeks.
Main Outcomes and Measures
Cognitive Function Scale (CFS) was classified as cognitively intact (CFS = 1), mildly impaired (CFS = 2), moderately impaired (CFS = 3), and severely impaired (CFS = 4).
Results
Of 45 183 long-term care residents, 12 644 residents (mean [SD] age 77.7 [8.3] years; 329 [2.6%] females and 12 315 [97.4%] males) and 12 053 residents (mean [SD] age 77.7 [8.3] years; 314 [2.6%] females and 11 739 [97.4%] males) met eligibility for ITT and per-protocol analyses, respectively. At the end of the follow-up, 12.0% of residents had a worsened CFS (higher score) and 7.7% had an improved CFS (lower score) with 10.8% of the deprescribing group and 12.1% of the stable user group showing a worsened CFS score. In the per-protocol analysis, the deprescribing group had a 12% reduction in the odds of progressing to a worse CFS category per 12-week period (odds ratio, 0.88; 95% CI, 0.78-0.99) compared to the stable user group. Among residents with dementia, deprescribing was associated with 16% reduced odds of cognitive decline (odds ratio, 0.84; 95% CI, 0.72-0.98). These patterns remained consistent in the ITT analysis.
Conclusions and Relevance
This cohort study indicates that deprescribing is associated with less cognitive decline in nursing home residents, particularly those with dementia. More data are needed to understand the benefits and harms of antihypertensive deprescribing to inform patient-centered medication management in nursing homes.
This cohort study investigates the association between antihypertensive deprescribing and changes in cognitive function in nursing home residents.
Introduction
Polypharmacy is common among older adults, with more than 40% taking 5 or more medications.1,2 Antihypertensive medications are important contributors to polypharmacy, with the prevalence notably high at 70%.3 While antihypertensive medications reduce cardiovascular risks, they also pose risks of adverse effects such as falls, orthostatic hypotension, and drug-drug interactions.4,5,6,7,8,9 The risk to benefit ratio of antihypertensive medication is unclear in adults with multimorbidity who are institutionalized given that they have been largely excluded from clinical trials. Considering that this population is at high risk for adverse effects, deprescribing—the strategic reduction or discontinuation of medications that may no longer be beneficial or could be associated with harm—may be clinically appropriate.1,10
The relationship between antihypertensive medication management and cognitive function in older adults is complex. While elevated blood pressure (BP) in midlife is a well-documented risk factor for cognitive decline, the optimal BP targets for older adults—especially those in nursing homes—remain unclear.11,12,13,14 Randomized clinical trials (RCTs) have shown mixed evidence on the effect of intensive BP control in reducing cognitive decline and the incidence of dementia.15,16 First, most trials enrolled relatively healthier older adults, specifically excluding nursing home residents and individuals with dementia who comprise a substantial segment of the aging population and who are at higher risk for cognitive decline. Second, several observational studies have observed an association between higher BP and less cognitive decline in older adults, especially among those with poor functional status.17,18,19,20 Therefore, a substantial knowledge gap persists regarding the long-term cognitive impact of antihypertensive treatment in frail older individuals, particularly those in nursing homes or those with dementia.
Our study aims to fill this gap by estimating the association of antihypertensive deprescribing with change in cognitive function of older nursing home residents, using a target trial approach using data from the US Department of Veteran’s Affairs (VA). We carried out a prespecified subgroup analysis based on dementia status given the paucity of data in this population and the potentially distinct responses to antihypertensive medications.21,22
Methods
This study received institutional review board approval from Stanford University and the VA Palo Alto Health Care System, with a waiver of informed consent because the research posed minimal risk and had procedures to protect confidentiality. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Target Trial Protocol Overview
We emulated a target trial to estimate the association of antihypertensive deprescribing on changes in cognitive function among veterans residing in a Community Living Center (CLC)—the VA term for nursing homes. The protocol is outlined in eTable 1 in Supplement 1. We identified residents as antihypertensive deprescribing users or stable users and recorded their cognitive function from time 0 to the final assessment of cognitive function or the end of the follow-up period. Subsequent analyses, emulating intention-to-treat (ITT) and per-protocol analyses, were executed to contrast the variations in the odds of worsening cognitive function between different treatment groups. Because protocol deviation is common in both RCTs and observational studies, we designated the per-protocol analysis as our primary approach to estimate the associations of deprescribing for patients who adhered to their assigned deprescribing or control groups.
Eligibility Criteria
Eligible candidates included residents aged 65 years and older from 2006 to 2019 who had a minimum CLC stay of 12 weeks (to ensure that residents were not short-term rehabilitation patients). To identify a population in which there was clinical uncertainty regarding antihypertensive medication use, we excluded residents whose admission systolic/diastolic BP exceeded 160/90 mm Hg, those diagnosed with heart failure, or those not taking any antihypertensive medications at admission (Figure 1).
Figure 1. Cohort Flowchart for Target Trial Study Participants Exclusion and Censorship.
CFS indicates Cognitive Function Scale; CLC, community living center; and VA, US Department of Veterans Affairs.
Treatment (Deprescribing) Strategies and Assignment
We used VA Barcode Medication Administration data to extract daily antihypertensive regimens for residents. Antihypertensive medications included β-blockers, calcium channel-blockers, angiotensin-converting enzyme inhibitors, angiotensin receptor-blockers, loop diuretics, thiazide diuretics, α-blockers, vasodilators, and potassium-sparing diuretics. Medications were converted into weekly standard doses.23 Deprescribing was defined as either a reduction in the overall number of antihypertensive medications or a 30% decrease in medication dosage compared to the previous week and sustained for at least 2 weeks. Residents who did not undergo these medication changes were considered to be stable users of antihypertensive medications.
Outcomes and Follow-Up
The primary outcome was the Cognitive Function Scale (CFS), collected through the Minimum Dataset (MDS). This scale encompasses the Brief Interview for Mental Status assessment, which includes self- and staff-reported cognitive data.24 CFS is a 4-level ordinal variable, characterized as cognitively intact (CFS = 1), mildly impaired (CFS = 2), moderately impaired (CFS = 3), and severely impaired (CFS = 4). It was collected at least quarterly throughout the nursing home stay. For the ITT analysis, we collected residents’ CFS measurements from the event week to a maximum of 2 years (104 weeks). Residents were censored at either death or discharge within the 2-year follow-up period. For the per-protocol analysis, candidates were censored if medication intensification occurred in the deprescribing group (eg, eFigure 1 resident C in Supplement 1), or if medication deprescribing occurred in the stable user group within the 2 years (eg, eFigure 1 resident D in Supplement 1).
Covariates
For both the ITT and per-protocol analyses, we collected the following covariates for potential confounding and modeling: sociodemographic covariates collected from the electronic health record including age, sex, race and ethnicity, and US region of residence; admission comorbidity indicators based on codes from the International Statistical Classification of Diseases and Related Health Problems, Ninth and Tenth Revision (ICD) for acute kidney disease, atrial fibrillation, chronic obstructive pulmonary disease, cerebrovascular disease, depression, kidney failure, peripheral vascular disease, coronary heart disease, dementia, diabetes, any malignant neoplasm, metastatic cancer, ischemic or unspecified stroke, osteoarthritis; vital signs and health history covariates including systolic/diastolic BP, activities of daily living (ADL), number and median daily dose of antihypertensive medications, weight, smoking status, history of falls, and insulin use indicator; laboratory values including blood urea nitrogen, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycated hemoglobin, serum albumin, total cholesterol, carbon dioxide, estimated glomerular filtration rate, serum glucose, serum potassium, serum sodium, triglycerides, and serum creatinine; and Steinman Indices, composite measurements of multimorbidity for risk assessment, including ADL, instrumental ADL, hospitalization, death.25 Dementia was ascertained using ICD codes during 1 year before CLC admission (eTable 3 in Supplement 1). Covariates were sourced from the VA Corporate Data Warehouse.
Statistical Analysis
To estimate the association of deprescribing compared to the stable user, we used the ordinal logistic generalized linear mixed model (GLMM) to analyze the longitudinal CFS outcome. The proportional odds ratio (OR) assumption was examined via the Brant-Wald test, with P > .05 suggesting this assumption was upheld. We incorporated a random intercept to account for intrapatient heterogeneity and used the cumulative logit function to model the ordinal CFS with the covariates. For the estimation of the fixed-effects, the Gauss-Hermite quadrature method was applied to estimate the marginal likelihood function using 10 quadrature points.26,27 The unadjusted model included follow-up time (modeled per 12 weeks), treatment (deprescribing), and the interaction (time × treatment). We analyzed the entire population before stratifying by dementia status (dementia vs nondementia) to investigate possible effect modification.
To emulate the randomization process, we used the inverse probability of treatment weighting (IPTW) method, adjusting for potential confounders. Specifically, we calculated the propensity for deprescribing, converted this into weights, and used these weights in the outcome model. For the per-protocol analysis, we further modeled the probability of censoring due to protocol deviations and applied the inverse probability of censoring weighting (IPCW). We designated the per-protocol approach as our primary analysis which would estimate the effects of deprescribing if the patients remained in their deprescribing and stable user groups. We used the SuperLearner algorithm to model the probability of treatment and censoring, integrating a variety of algorithms such as the mean model, logistic regression, least absolute shrinkage and selection operator, interaction model, classification and regression tree, neural network, random forest, and gradient boosting. The balance of the covariates before and after applying IPTW and IPCW weighting is illustrated in eFigure 4 in Supplement 1. To account for residual confounding, we adjusted for the following covariates: age, sex, race, baseline ADL, systolic/diastolic BP, weight, median daily medication dose, falls within 30 days, depression, diabetes, peripheral vascular disease, presence of malignant and/or metastatic tumors, end-stage renal disease, and Steinman risk indices in the ordinal GLMMs. The weighting approach was implemented in both the unadjusted and the fully adjusted models.
We conducted sensitivity analyses using 2 outcomes, hearing ability and pain intensity level, as negative control studies. Both outcomes were ordinal variables with 4 naturally ordered levels, derived from the MDS. The results of these analyses are presented in eTable 5 in Supplement 1. eFigure 7 in Supplement 1 reports the weekly changes in systolic BP compared with baseline systolic BP across different treatment groups from time 0 to per-protocol censoring.
We present treatment effects as ORs with 95% CIs. The E-value, which assesses the robustness of an estimated effect against potential unmeasured confounding, was reported alongside ORs. A higher E-value indicates that the association is more robust to potential unmeasured confounding.28,29 Additionally, we displayed the predicted probabilities across various CFS levels between treatment groups. For the construction of the ordinal GLMMs, we used SAS, 9.4 (SAS Institute Inc) and for SuperLearner execution, we used R, version 4.1.1 (The R Foundation for Statistical Computing). Data analyses were performed from May 1, 2023, and July 1, 2024.
Results
In all, 12 644 CLC residents (mean [SD] age, 77.7 [8.3] years; 12 315 [97.4%] males and 329 [2.6%] females; 49 [0.4%] American Indian, 155 [1.2%] Asian or Pacific Islander, 2207 [17.5%] Black, and 9247 [73.1%] White individuals, with 592 [4.7%] of Hispanic ethnicity) met the eligibility criteria and were included in the ITT analysis (Table 1). Of these, 1290 residents (10.2%) experienced deprescribing episodes and 11 354 (89.8%) remained stable users. Median (range) follow-up duration was 23 (9-65) weeks for the deprescribing group and 21 (5-77) weeks for the stable users. Within 2 years of follow-up, 383 participants (29.7%) in the deprescribing group and 2592 participants (22.8%) in the stable user group had died; 769 participants (59.6%) in the deprescribing group and 7342 participants (64.7%) in the stable user group were discharged from the CLC. Within these groups, 586 residents (45.4%) from the deprescribing group and 4551 stable users (40.1%) deviated from their assigned treatment group and were censored in the week of deviation. We excluded 591 residents (4.7%) without CFS measures between time 0 and the censored week; therefore, the per-protocol analysis included 12 053 residents (95.3%; Figure 1).
Table 1. Baseline Characteristics of Study Cohort, Stable Use of Antihypertensive Medication (Control) Arm Compared With Newly Deprescribed Arm.
| Variable | Analysis arm | |||||
|---|---|---|---|---|---|---|
| Intention to treat, No. (%) | Per protocol, No. (%) | |||||
| Stable use | Newly deprescribed | SMD | Stable use | Newly deprescribed | SMD | |
| Participants, No. | 11 354 | 1290 | NA | 10 935 | 1118 | NA |
| Demographic characteristic | ||||||
| Age, mean (SD), y | 77.6 (8.3) | 78.0 (8.3) | 0.04 | 77.6 (8.3) | 78.0 (8.4) | 0.05 |
| Female | 295 (2.6) | 34 (2.6) | 0.002 | 286 (2.6) | 28 (2.5) | −0.007 |
| Male | 11 059 (97.4) | 1256 (97.4) | 10 649 (97.4) | 1090 (97.5) | ||
| Race | ||||||
| American Indian | 44 (0.4) | 5 (0.4) | 0.15 | 42 (0.4) | 4 (0.4) | 0.17 |
| Asian/Pacific Islander | 145 (1.3) | 10 (0.8) | 138 (1.3) | 10 (0.9) | ||
| Black | 1998 (17.6) | 209 (16.2) | 1935 (17.7) | 171 (15.3) | ||
| White | 8270 (72.8) | 977 (75.7) | 7952 (72.7) | 857 (76.7) | ||
| Multiple races | 87 (0.8) | 6 (0.5) | 85 (0.8) | 5 (0.4) | ||
| Unknown | 810 (7.1) | 83 (6.4) | 783 (7.2) | 71 (6.4) | ||
| Hispanic ethnicity | 535 (4.7) | 57 (4.4) | −0.01 | 517 (4.7) | 49 (4.4) | −0.02 |
| US region of residence | ||||||
| Continental | 1457 (12.8) | 203 (15.7) | 0.15 | 1396 (12.8) | 171 (15.3) | 0.12 |
| Midwest | 2679 (23.6) | 308 (23.9) | 2577 (23.6) | 270 (24.2) | ||
| North Atlantic | 3600 (31.7) | 378 (29.3) | 3468 (31.7) | 326 (29.2) | ||
| Pacific | 1662 (14.6) | 160 (12.4) | 1618 (14.8) | 143 (12.8) | ||
| Southeast | 1956 (17.2) | 241 (18.7) | 1876 (17.2) | 208 (18.6) | ||
| Vital signs, mean (SD)a | ||||||
| Systolic BP, mm Hg | 129.6 (19.4) | 124.3 (19.9) | −0.27 | 129.7 (19.3) | 124.2 (19.9) | 0.03 |
| Diastolic BP, mm Hg | 70.2 (10.8) | 67.6 (10.8) | −0.24 | 70.3 (10.8) | 67.6 (10.7) | 0.04 |
| Weight, lb | 190.5 (47.4) | 188.5 (45.3) | −0.05 | 190.7 (47.5) | 188.3 (44.9) | −0.05 |
| Nursing home admission measures | ||||||
| ADL,b median (IQR) | 12 (6-19) | 13 (6-19) | 0.07 | 13 (6-19) | 13.5 (7-20) | 0.21 |
| Cognitive impairment, per CFS | ||||||
| Cognition intact | 3364 (29.6) | 364 (28.2) | 0.06 | 3290 (30.1) | 327 (29.2) | 0.10 |
| Mildly impaired | 4409 (38.8) | 511 (39.6) | 4219 (38.6) | 427 (38.2) | ||
| Moderately impaired | 2718 (23.9) | 305 (23.6) | 2606 (23.8) | 270 (24.2) | ||
| Severely impaired | 863 (7.6) | 110 (8.5) | 820 (7.5) | 94 (8.4) | ||
| Medication use | ||||||
| Insulin | 3844 (33.9) | 438 (34) | 0.002 | 3709 (33.9) | 370 (33.1) | −0.02 |
| Antihypertensive use, median No. | ||||||
| 1 | 5378 (47.4) | 532 (41.2) | 0.16 | 5199 (47.5) | 482 (43.1) | 0.13 |
| 2 | 3986 (35.1) | 460 (35.7) | 3830 (35) | 399 (35.7) | ||
| 3 | 1525 (13.4) | 231 (17.9) | 1461 (13.4) | 187 (16.7) | ||
| ≥4 | 465 (4.1) | 67 (5.2) | 445 (4.1) | 50 (4.5) | ||
| Antihypertensive use, mean (SD) | 1.8 (1.7) | 2.0 (1.8) | 0.27 | 1.8 (1.7) | 1.9 (1.7) | 0.10 |
| Comorbidities | ||||||
| Acute kidney injury | 2162 (19.0) | 264 (20.5) | 0.05 | 2082 (19.0) | 232 (20.8) | 0.04 |
| Atrial fibrillation | 2234 (19.7) | 261 (20.2) | 0.01 | 2164 (19.8) | 233 (20.8) | 0.03 |
| COPD | 1603 (14.1) | 142 (11.0) | −0.09 | 1565 (14.3) | 132 (11.8) | −0.07 |
| Cerebrovascular disease | 3900 (34.3) | 448 (34.7) | 0.008 | 3754 (34.3) | 384 (34.3) | 0 |
| Depression | 4648 (40.9) | 524 (40.6) | −0.006 | 4470 (40.9) | 451 (40.3) | −0.01 |
| Kidney failure | 2961 (26.1) | 352 (27.3) | 0.03 | 2836 (25.9) | 305 (27.3) | 0.03 |
| Peripheral vascular disease | 2807 (24.7) | 320 (24.8) | 0.002 | 2678 (24.5) | 268 (24.0) | −0.01 |
| Coronary heart disease | 4102 (36.1) | 485 (37.6) | 0.03 | 3946 (36.1) | 423 (37.8) | 0.04 |
| Dementia | 5350 (47.1) | 575 (44.6) | −0.05 | 5145 (47.1) | 496 (44.4) | −0.05 |
| Diabetes | 5384 (47.4) | 586 (45.4) | −0.04 | 5193 (47.5) | 505 (45.2) | −0.05 |
| Any malignant neoplasm | 3156 (27.8) | 400 (31) | 0.07 | 3026 (27.7) | 354 (31.7) | 0.09 |
| Metastatic cancer | 979 (8.6) | 156 (12.1) | 0.11 | 947 (8.7) | 140 (12.5) | 0.13 |
| Osteoarthritis | 2935 (25.8) | 349 (27.1) | 0.03 | 2813 (25.7) | 303 (27.1) | 0.03 |
| Ischemic/unspecified stroke | 2103 (18.5) | 236 (18.3) | 0.05 | 2033 (18.6) | 202 (18.1) | −0.01 |
| Fall and smoking history | ||||||
| Fall in 30 d prior | 5082 (44.8) | 599 (46.4) | 0.03 | 4869 (44.5) | 517 (46.2) | 0.04 |
| Smoking history | ||||||
| No | 6433 (56.7) | 755 (58.5) | 0.05 | 6199 (56.7) | 663 (59.3) | 0.05 |
| Yes | 2789 (24.6) | 300 (23.3) | 2694 (24.6) | 258 (23.1) | ||
| Unknown | 2132 (18.8) | 235 (18.2) | 2042 (18.7) | 197 (17.6) | ||
| Laboratory test results, mean (SD) | ||||||
| Blood urea nitrogen, mg/dL | 24.0 (11.9) | 27.0 (16.1) | 0.22 | 24.0 (11.9) | 26.7 (15.9) | 0.01 |
| HDL, mg/dL | 40.4 (13.4) | 39.9 (13.7) | −0.04 | 40.5 (13.4) | 39.6 (13.6) | 0.01 |
| LDL, mg/dL | 76.6 (30.3) | 76.6 (30.5) | 0 | 76.6 (30.2) | 76.6 (30.5) | −0.08 |
| Total cholesterol, mg/dL | 139.7 (37.1) | 139.1 (38.0) | −0.02 | 139.7 (37.1) | 138.8 (37.8) | −0.03 |
| Glycated hemoglobin, % | 6.3 (1.3) | 6.2 (1.3) | −0.03 | 6.3 (1.3) | 6.2 (1.3) | −0.24 |
| Albumin, g/dL | 3.3 (0.6) | 3.1 (0.6) | −0.19 | 3.3 (0.6) | 3.1 (0.6) | −0.002 |
| Carbon dioxide, mEq/L | 27.0 (3.5) | 26.7 (3.8) | −0.09 | 27.0 (3.5) | 26.8 (3.9) | 0.05 |
| Creatinine eGFR, mL/min/1.73 m2 | 75.4 (34.2) | 73.6 (35.0) | −0.05 | 75.6 (34.1) | 74.1 (34.8) | −0.002 |
| Glucose, mg/dL | 122.4 (50.6) | 123.8 (51.3) | 0.03 | 122.5 (50.6) | 123.2 (51.3) | −0.06 |
| Potassium, mEq/L | 4.2 (0.5) | 4.3 (0.5) | 0.05 | 4.3 (0.5) | 4.3 (0.5) | 0.02 |
| Serum creatinine, mg/dL | 1.4 (1.9) | 1.5 (2.4) | 0.06 | 1.4 (1.9) | 1.5 (2.5) | −0.02 |
| Sodium, mEq/L | 138.2 (5.4) | 137.7 (6.0) | −0.08 | 138.2 (5.5) | 137.7 (6.1) | −0.04 |
| Triglycerides, mg/dL | 118.5 (63.7) | 116.9 (64.3) | −0.03 | 118.6 (63.8) | 115.4 (62.4) | 0.01 |
| Creatinine increase of 50%c | 122 (1.1) | 29 (2.2) | 0.092 | 118 (1.1) | 25 (2.2) | −0.06 |
| Potassium counts <3.5c | 223 (2.0) | 42 (3.3) | 0.08 | 210 (1.9) | 32 (2.9) | 0.08 |
| Sodium count decline of 5c | 416 (3.7) | 58 (4.5) | 0.04 | 397 (3.6) | 46 (4.1) | −0.009 |
| Steinman Index,d mean (SD) | ||||||
| ADL | 2.3 (1.8) | 2.3 (1.8) | 0.02 | 2.3 (1.8) | 2.4 (1.9) | 0.04 |
| IADL | 2.7 (1.9) | 2.8 (1.9) | 0.03 | 2.7 (1.9) | 2.8 (1.9) | 0.03 |
| Hospitalization | 3.0 (2.4) | 3.1 (2.4) | 0.03 | 3.0 (2.4) | 3.1 (2.4) | 0.03 |
| Death | 2.4 (1.7) | 2.6 (1.8) | 0.08 | 2.5 (1.7) | 2.6 (1.8) | 0.08 |
Abbreviations: ADL, activities of daily living; CFS, Cognitive Function Scale; COPD, chronic obstructive pulmonary disease; EGFR, estimated glomerular filtration rate; HDL, high-density lipoprotein cholesterol; IADL, instrumental activities of daily living; LDL, low-density lipoprotein cholesterol; SMD, standardized mean difference.
Measured prior to time 0.
Composed of 7 items: bed mobility, transfer, walking/locomotion, dressing, eating, toilet use, and personal hygiene that were rated on a 5-level scale from 0 (totally independent) to 4 (totally dependent), making the overall ADL score range from 0 to 28.
Calculated prior to time 0.
Steinman Index consists of 4 indices of multimorbidity developed from Medicare claims data and is designed to assess risks associated with ADL dependency, IADL dependency, hospitalization, and mortality.
SI conversion factor: to convert blood urea nitrogen to mmol/L, multiply by 0.357; HDL, LDL, and total cholesterol to mmol/L, multiply by 0.0259; glycated hemoglobin to proportion of total hemoglobin, multiply by 0.01; albumin to g/L, multiply by 10; carbon dioxide to mmol/L, multiply by 1; creatinine eGFR to mL/s/m2, multiply by 0.0167; glucose to mmol/L, multiply by 0.0555; potassium to mmol/L, multiply by 1; serum creatinine to µmol/L, multiply by 88.4; sodium to mmol/L, multiply by 1; triglycerides to mmol/L, multiply by 0.0113.
Residents in the control group had higher baseline BP, lower median ADL difficulty counts, and were taking fewer antihypertensive medications compared to the deprescribing group (Table 1). Residents in the deprescribing group had an elevated weekly average systolic BP compared to stable users (eFigure 7 in Supplement 1). No significant cognitive differences were observed at baseline. Stable users had a lower proportion of laboratory test result changes, with a creatinine increase of 50% or greater and potassium counts less than 3.5 mEq/L (to convert to mmol/L, multiply by 1) compared to the deprescribing group. Residents in the deprescribing group had a slightly higher Steinman index for mortality, indicating higher mortality risk. The standardized mean differences were all within 0.1, indicating balanced covariates after weighting (eFigure 4 in Supplement 1).30
Compared to their CFS at the beginning of the follow-up, 12.0% of the residents had a worsened CFS (higher score), while 7.7% had an improved CFS (lower score) by the end of the follow-up. Additionally, 10.8% of the residents in the deprescribing group and 12.1% in the stable user group had a worsened CFS score. Table 2 presents the primary findings. In the fully adjusted per-protocol analysis, which considered only those residents who followed their assigned group, residents in the deprescribing group experienced a 12% reduction in the odds of progressing to a worse CFS category per 12-week period (odds ratio [OR], 0.88; 95% CI, 0.78-0.99; P = .04) compared to stable users. This association remained consistent in ITT analysis, demonstrating a protective association, albeit with an attenuated magnitude (OR, 0.94; 95% CI, 0.90-0.98).
Table 2. Association of Deprescribing Antihypertensive Medication With Change in Cognitive Function Scale (CFS) Score, by Per-Protocol and Intention-to-Treat Analyses.
| Variable | Odds ratio (95% CI)a | |||
|---|---|---|---|---|
| Per-protocol analysis | Intention-to-treat analysis | |||
| Unadjustedb | Fully adjustedc | Unadjustedb | Fully adjustedc | |
| All nursing home residents d | ||||
| Deprescribing × time (12 wk) | 0.99 (0.99-1.01) | 0.88 (0.78-0.99) | 0.94 (0.90-0.98) | 0.94 (0.90-0.98) |
| E-value | 1.11 | 1.53 | 1.32 | 1.32 |
| Residents without dementia | ||||
| Deprescribing × time (12 wk) | 0.94 (0.76-1.18) | 0.94 (0.75-1.17) | 0.96 (0.90-1.04) | 0.95 (0.89-1.02) |
| E-value | 1.32 | 1.17 | 1.25 | 1.29 |
| Residents with dementia | ||||
| Deprescribing × time (12 wk) | 0.95 (0.82-1.09) | 0.84 (0.72-0.98) | 0.94 (0.90-0.99) | 0.94 (0.89-0.98) |
| E-value | 1.29 | 1.67 | 1.32 | 1.32 |
An odds ratio less than 1 suggests a protective association with deprescribing over time.
Unadjusted model: CFS approximate time + deprescribing + time × deprescribing.
Fully adjusted model: unadjusted + age, sex, race or ethnicity, baseline activities of daily living score, systolic/diastolic blood pressure, weight, median daily medication dose, occurrences of falls within 30 days, depression, diabetes, peripheral vascular disease, presence of malignant and metastatic tumors, end-stage renal disease, and 4 Steinman risk indices.
Overall, 12.1% of the residents transitioned to a worsened (higher) score on the CFS at the end of the follow-up period, 10.8% in the deprescribed group and 12.1% in the stable user group.
When stratified by dementia, we noticed similar trends of cognitive decline in both groups throughout the study. This decline was statistically significant given the association between dementia status, deprescribing, and time. Among residents with dementia, those in the deprescribing group had a 16% reduced odds of cognitive function worsening per 12-week period (OR, 0.84; 95% CI, 0.72-0.98). The E-values for the interaction terms in both analyses were modest (ITT E-value = 1.32; per-protocol E-value = 1.53). The main effects of deprescribing at baseline and time are detailed in eTable 2 in Supplement 1.
Figure 2 illustrates the marginal predicted probabilities for each cognitive function category over time based on the fully adjusted per-protocol analysis. At the start of the study (baseline cognitive state), residents in the stable user group were more likely to be estimated as cognitively intact (Figure 2A) or mildly impaired (Figure 2B), whereas they had a lower probability of being moderately impaired (Figure 2C) or severely impaired (Figure 2D) when compared to the deprescribing group. Over time, both the deprescribing and stable user groups experienced a decline in cognitive function with a higher probability of residents entering categories with higher levels of cognitive impairment (ie, more residents entering the categories of moderately or severely impaired). Residents in the deprescribing group demonstrated a slower rate of decline across all cognitive function categories compared to stable users. For example, the probability of being cognitively intact and mildly impaired for the deprescribing group decreased more gradually than among stable users, implying that deprescribing was associated with a slower transition to more severe states of cognitive impairment.
Figure 2. Average Predicted Probabilities Over Time for the Cognitive Function Scale Categories, With 95% CI Bands.
Shading indicates 95% CI band for the predicted probabilities for each cognitive function category.
The analyses assessing the association between deprescribing, and 2 negative control outcomes are presented in eTable 5 in Supplement 1. Results showed that deprescribing was associated with worse hearing ability and no difference in pain intensity.
Discussion
Our study provides novel evidence regarding the association of antihypertensive deprescribing with cognitive outcomes among older nursing home residents. Based on a target trial emulation, residents who were deprescribed antihypertensive medications had slower cognitive decline when compared with residents who maintained a stable antihypertensive regimen. Notably, this association was stronger in persons with dementia. However, we interpret these results with caution because the estimates were close to null, the precision was limited, and the E-value suggests that moderate to strong unmeasured confounding could offset a potential protective effect. Our results align with emerging literature21,31,32 and suggest caution regarding intensive BP control in older adults living in nursing homes, particularly those with cognitive impairment.
Our results complement the existing evidence found in RCTs on deprescribing in community-dwelling older adults. The OPTIMISE trial33 randomized 569 patients (mean age, 84.8 years) to deprescribing and control arms with a primary outcome of systolic BP less than 150 mm Hg at the 12-week follow-up. Deprescribing was noninferior to usual care with an adjusted 1-sided relative risk of 0.98 (97.5% CI, 0.92 to ∞). The ECSTATIC trial34 assessed the predicted cardiovascular disease (CVD) risk between a deprescribing cardiovascular medication arm and a usual care arm, enrolling 1067 participants with a mean age of 55 years and including 319 in the per-protocol deprescribing arm. After a 2-year follow-up, the predicted CVD risk increased by 2.0% in the deprescribing arm and 1.9% in the usual care arm. Nevertheless, neither trial examined the impact of deprescribing on cognitive function, leaving a gap in our understanding of the broader impacts of deprescribing. The DANTE trial,35 involving 385 older adults residing in the community with mild cognitive impairment and no history of heart conditions, found no difference in cognitive or daily functional outcomes, but follow-up was limited to only 16 weeks. Most importantly, these 3 trials only included older adults living in the community, who are generally healthier than older adults living in nursing homes.
Although a robust association exists between midlife hypertension and cognitive decline and dementia in later life, the association may change with aging. Two longitudinal studies, the AGE-Reykjavik31 and the ARIC,36 showed that individuals with midlife hypertension who later exhibited lower BP measurements in late life had poorer cognitive performance. In contrast, those without a history of midlife hypertension generally maintain consistent cognitive function, regardless of their BP levels in later years. This pattern suggests that the long-term effects of hypertension on cognitive function can vary based on when hypertension first develops and its duration. Recognizing the differential effects of hypertension across the life course could help target the optimal timing for deprescribing. Deprescribing should involve a strategic and careful reduction or discontinuation of nonessential antihypertensive medications. This process must be person-centered, considering individual risk factors and medical history of BP control.9
Target trial emulation is a complementary approach to RCTs and proves useful when RCTs are challenging to implement or when assessing treatment effects in populations excluded from RCTs. Both situations apply in our study.37 However, similar to most observational studies, our study may be subject to residual confounding. Thus, to mitigate this bias, we implemented multiple methods including inverse probability weighting with SuperLearner, multivariable adjustment for residual confounding, and the evaluation of negative control outcomes, hearing loss and pain. The results from the analysis of change in hearing ability over time suggested a modest harmful effect of deprescribing; this is reassuring as it suggests that any bias of our design and analytic approach would result in a conservative bias. We used a robust and valid study design, aligning eligibility criteria, treatment assignment, and follow-up from time zero to mitigate biases, such as immortal time bias. Moreover, other studies have demonstrated that the results from well-designed target trial emulations largely align with those from RCTs, demonstrating the method’s effectiveness, although not its infallibility.38 Future studies could benefit from employing techniques such as instrumental variable analysis to better control for unmeasured confounding, especially given the robust design of our approach. Additionally, enhancing the representativeness of the sample by including more diverse populations beyond the VA, and using alternative statistical methods for further validation of the findings, could strengthen future research efforts.
Limitations
Our cohort study has additional strengths and limitations. We used a robust target trial emulation design involving a large cohort of older, co-morbid nursing home residents with a prolonged follow-up duration. We used detailed medication administered information to meticulously track medication adjustments in terms of quantity and dosage. Additionally, the use of MDS data captures a comprehensive and longitudinal view of cognitive outcomes throughout the study, offering ample repeated measurements conducive to sophisticated modeling approaches. Our study contains several limitations. First, the VA population is predominantly male and white, thus limiting the generalizability of our results to females and other racial and ethnic groups. Second, our cohort did not include patients with heart failure, limiting the generalizability of our results to this population. Third, the specificity of our dementia diagnosis is limited, as our current ascertainment lumps together various forms of dementia, such as Alzheimer disease and vascular dementia. Differentiating the impacts of deprescribing among these subgroups is challenging with the current ICD coding system, and the small sample sizes within each subgroup may not support valid analyses.
Conclusions
Our study suggests that deprescribing antihypertensives may protect nursing home residents from future cognitive losses, especially for those living with dementia. This work highlights the need for patient-centered approaches to deprescribing, ensuring that medication regimens for older adults are optimized to preserve cognitive function and minimize potential harms.
eMethods. Additional Details and Explanations Regarding Statistical Methods and Results
eTable 1. Target Trial Protocol
eTable 2. Main Estimates of Effects of Deprescribing and Time Covariates
eTable 3. ICD9/10 Codes for Dementia Ascertainment
eTable 4. Missing Data Percentage
eTable 5. Association of Antihypertensive Deprescribing with Change in Hearing Ability and Pain Intensity: Per-Protocol and Intention-to-Treat Analyses with Odds Ratios (95% CI)
eFigure1. Target Trial Design Illustration
eFigure2. Overlap of Predicted Probabilities of Treatment in Deprescribing and Stable User Groups
eFigure 3. Variable Importance Plots for IPTW and IPCW SuperLearner
eFigure 4. Standardized Mean Difference Before and After Weighting
eFigure 5. Overlap of Predicted Probabilities of Treatment in Deprescribing and Stable User Arms in Imputed Datasets
eFigure 6. Overlay ROC Curves from Imputed Datasets
eFigure 7. Mean SBP Change Boxplots Throughout Followup by Treatment
Data Sharing Statement
Reference
- 1.Linsky A, Gellad WF, Linder JA, Friedberg MW. Advancing the science of deprescribing: a novel comprehensive conceptual framework. J Am Geriatr Soc. 2019;67(10):2018-2022. doi: 10.1111/jgs.16136 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Anderson TS, Steinman MA. Antihypertensive prescribing cascades as high-priority targets for deprescribing. JAMA Intern Med. 2020;180(5):651-652. doi: 10.1001/jamainternmed.2019.7082 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Muntner P, Carey RM, Gidding S, et al. Potential US population impact of the 2017 ACC/AHA high blood pressure guideline. Circulation. 2018;137(2):109-118. doi: 10.1161/CIRCULATIONAHA.117.032582 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Liu X, Steinman MA, Lee SJ, et al. Systolic blood pressure, antihypertensive treatment, and cardiovascular and mortality risk in VA nursing home residents. J Am Geriatr Soc. 2023;71(7):2131-2140. doi: 10.1111/jgs.18301 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Fried TR, O’Leary J, Towle V, Goldstein MK, Trentalange M, Martin DK. Health outcomes associated with polypharmacy in community-dwelling older adults: a systematic review. J Am Geriatr Soc. 2014;62(12):2261-2272. doi: 10.1111/jgs.13153 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Benetos A, Labat C, Rossignol P, et al. Treatment with multiple blood pressure medications, achieved blood pressure, and mortality in older nursing home residents: the PARTAGE study. JAMA Intern Med. 2015;175(6):989-995. doi: 10.1001/jamainternmed.2014.8012 [DOI] [PubMed] [Google Scholar]
- 7.Cai A, Calhoun DA. Antihypertensive medications and falls in the elderly. Am J Hypertens. 2018;31(3):281-283. doi: 10.1093/ajh/hpx203 [DOI] [PubMed] [Google Scholar]
- 8.Butt DA, Mamdani M, Austin PC, Tu K, Gomes T, Glazier RH. The risk of falls on initiation of antihypertensive drugs in the elderly. Osteoporos Int. 2013;24(10):2649-2657. doi: 10.1007/s00198-013-2369-7 [DOI] [PubMed] [Google Scholar]
- 9.Sheppard JP, Benetos A, McManus RJ. Antihypertensive deprescribing in older adults: A practical guide. Curr Hypertens Rep. 2022;24(11):571-580. doi: 10.1007/s11906-022-01215-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Reeve E, Gnjidic D, Long J, Hilmer S. A systematic review of the emerging definition of ‘deprescribing’ with network analysis: implications for future research and clinical practice. Br J Clin Pharmacol. 2015;80(6):1254-1268. doi: 10.1111/bcp.12732 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Onder G, Vetrano DL, Villani ER, et al. Deprescribing in nursing home residents on polypharmacy: incidence and associated factors. J Am Med Dir Assoc. 2019;20(9):1116-1120. doi: 10.1016/j.jamda.2019.01.130 [DOI] [PubMed] [Google Scholar]
- 12.Boockvar KS, Song W, Lee S, Intrator O. Hypertension Treatment in US Long-Term Nursing Home Residents With and Without Dementia. J Am Geriatr Soc. 2019;67(10):2058-2064. doi: 10.1111/jgs.16081 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Gottesman RF, Schneider AL, Albert M, et al. Midlife hypertension and 20-year cognitive change: the atherosclerosis risk in communities neurocognitive study. JAMA Neurol. 2014;71(10):1218-1227. doi: 10.1001/jamaneurol.2014.1646 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Jiang C, Li S, Wang Y, et al. Diastolic blood pressure and intensive blood pressure control on cognitive outcomes: insights from the SPRINT MIND Trial. Hypertension. 2023;80(3):580-589. doi: 10.1161/HYPERTENSIONAHA.122.20112 [DOI] [PubMed] [Google Scholar]
- 15.Hughes D, Judge C, Murphy R, et al. Association of blood pressure lowering with incident dementia or cognitive impairment: a systematic review and meta-analysis. JAMA. 2020;323(19):1934-1944. doi: 10.1001/jama.2020.4249 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Dallaire-Théroux C, Quesnel-Olivo MH, Brochu K, et al. Evaluation of intensive vs standard blood pressure reduction and association with cognitive decline and dementia: a systematic review and meta-analysis. JAMA Netw Open. 2021;4(11):e2134553-e2134553. doi: 10.1001/jamanetworkopen.2021.34553 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Miller LM, Peralta CA, Fitzpatrick AL, et al. The role of functional status on the relationship between blood pressure and cognitive decline: the Cardiovascular Health Study. J Hypertens. 2019;37(9):1790-1796. doi: 10.1097/HJH.0000000000002102 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Ogliari G, Sabayan B, Mari D, et al. Age- and Functional Status-Dependent Association Between Blood Pressure and Cognition: The Milan Geriatrics 75+ Cohort Study. J Am Geriatr Soc. 2015;63(9):1741-1748. doi: 10.1111/jgs.13616 [DOI] [PubMed] [Google Scholar]
- 19.Mossello E, Pieraccioli M, Nesti N, et al. Effects of low blood pressure in cognitively impaired elderly patients treated with antihypertensive drugs. JAMA Intern Med. 2015;175(4):578-585. doi: 10.1001/jamainternmed.2014.8164 [DOI] [PubMed] [Google Scholar]
- 20.Sabayan B, Oleksik AM, Maier AB, et al. High blood pressure and resilience to physical and cognitive decline in the oldest old: the Leiden 85-plus Study. J Am Geriatr Soc. 2012;60(11):2014-2019. doi: 10.1111/j.1532-5415.2012.04203.x [DOI] [PubMed] [Google Scholar]
- 21.Gottesman RF. Should hypertension be treated in late life to preserve cognitive function? Con side of the argument. Hypertension. 2018;71(5):787-792. doi: 10.1161/HYPERTENSIONAHA.117.09336 [DOI] [PubMed] [Google Scholar]
- 22.Iadecola C, Yaffe K, Biller J, et al. ; American Heart Association Council on Hypertension; Council on Clinical Cardiology; Council on Cardiovascular Disease in the Young; Council on Cardiovascular and Stroke Nursing; Council on Quality of Care and Outcomes Research; and Stroke Council . Impact of hypertension on cognitive function: a scientific statement from the American Heart Association. Hypertension. 2016;68(6):e67-e94. doi: 10.1161/HYP.0000000000000053 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Law MR, Wald NJ, Morris JK, Jordan RE. Value of low dose combination treatment with blood pressure lowering drugs: analysis of 354 randomised trials. BMJ. 2003;326(7404):1427. doi: 10.1136/bmj.326.7404.1427 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Thomas KS, Dosa D, Wysocki A, Mor V. The minimum data set 3.0 cognitive function scale. Med Care. 2017;55(9):e68-e72. doi: 10.1097/MLR.0000000000000334 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Steinman MA, Jing B, Shah SJ, et al. Development and validation of novel multimorbidity indices for older adults. J Am Geriatr Soc. 2023;71(1):121-135. doi: 10.1111/jgs.18052 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Fitzmaurice GM, Laird NM, Ware JH. Applied longitudinal analysis. John Wiley & Sons; 2012. [Google Scholar]
- 27.Capanu M, Gönen M, Begg CB. An assessment of estimation methods for generalized linear mixed models with binary outcomes. Stat Med. 2013;32(26):4550-4566. doi: 10.1002/sim.5866 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Mathur MB, Ding P, Riddell CA, VanderWeele TJ. Website and R package for computing E-values. Epidemiology. 2018;29(5):e45-e47. Link. doi: 10.1097/EDE.0000000000000864 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.VanderWeele TJ, Ding P. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med. 2017;167(4):268-274. doi: 10.7326/M16-2607 [DOI] [PubMed] [Google Scholar]
- 30.Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009;28(25):3083-3107. doi: 10.1002/sim.3697 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Muller M, Sigurdsson S, Kjartansson O, et al. ; Age, Gene/Environment Susceptibility-Reykjavik Study Investigators . Joint effect of mid- and late-life blood pressure on the brain: the AGES-Reykjavik study. Neurology. 2014;82(24):2187-2195. doi: 10.1212/WNL.0000000000000517 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Walker KA, Sharrett AR, Wu A, et al. Association of midlife to late-life blood pressure patterns with incident dementia. JAMA. 2019;322(6):535-545. doi: 10.1001/jama.2019.10575 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Sheppard JP, Burt J, Lown M, et al. ; OPTIMISE Investigators . Effect of antihypertensive medication reduction vs usual care on short-term blood pressure control in patients with hypertension aged 80 years and older: the OPTIMISE randomized clinical trial. JAMA. 2020;323(20):2039-2051. doi: 10.1001/jama.2020.4871 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Luymes CH, Poortvliet RKE, van Geloven N, et al. Deprescribing preventive cardiovascular medication in patients with predicted low cardiovascular disease risk in general practice−the ECSTATIC study: a cluster randomised non-inferiority trial. BMC Med. 2018;16(1):5. doi: 10.1186/s12916-017-0988-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Moonen JE, Foster-Dingley JC, de Ruijter W, et al. Effect of discontinuation of antihypertensive treatment in elderly people on cognitive functioning—the DANTE Study Leiden: a randomized clinical trial. JAMA Intern Med. 2015;175(10):1622-1630. doi: 10.1001/jamainternmed.2015.4103 [DOI] [PubMed] [Google Scholar]
- 36.Gottesman RF, Albert MS, Alonso A, et al. Associations between midlife vascular risk factors and 25-year incident dementia in the Atherosclerosis Risk in Communities (ARIC) cohort. JAMA Neurol. 2017;74(10):1246-1254. doi: 10.1001/jamaneurol.2017.1658 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Unroe KT, Saliba D, Hickman SE, Zimmerman S, Levy C, Gurwitz J. Evaluation of medical therapies in the nursing home population: gaps, challenges, and next steps. J Am Geriatr Soc. 2024. doi: 10.1111/jgs.18829 [DOI] [PubMed] [Google Scholar]
- 38.Wang SV, Schneeweiss S, Franklin JM, et al. ; RCT-DUPLICATE Initiative . Emulation of randomized clinical trials with nonrandomized database analyses: results of 32 clinical trials. JAMA. 2023;329(16):1376-1385. doi: 10.1001/jama.2023.4221 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eMethods. Additional Details and Explanations Regarding Statistical Methods and Results
eTable 1. Target Trial Protocol
eTable 2. Main Estimates of Effects of Deprescribing and Time Covariates
eTable 3. ICD9/10 Codes for Dementia Ascertainment
eTable 4. Missing Data Percentage
eTable 5. Association of Antihypertensive Deprescribing with Change in Hearing Ability and Pain Intensity: Per-Protocol and Intention-to-Treat Analyses with Odds Ratios (95% CI)
eFigure1. Target Trial Design Illustration
eFigure2. Overlap of Predicted Probabilities of Treatment in Deprescribing and Stable User Groups
eFigure 3. Variable Importance Plots for IPTW and IPCW SuperLearner
eFigure 4. Standardized Mean Difference Before and After Weighting
eFigure 5. Overlap of Predicted Probabilities of Treatment in Deprescribing and Stable User Arms in Imputed Datasets
eFigure 6. Overlay ROC Curves from Imputed Datasets
eFigure 7. Mean SBP Change Boxplots Throughout Followup by Treatment
Data Sharing Statement
