Authors: Sarah D. Berry (1.Hebrew Senior Life, Hinda and Arthur Marcus Institute for Aging Research and Department of Medicine, Roslindale, Massachusetts; 2.Department of Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts), Kaleen Hayes (3.Department of Health Services, Policy, and Practice & Center for Gerontology, Brown University School of Public Health, Providence, Rhode Island.), Yoojin Lee (3.Department of Health Services, Policy, and Practice & Center for Gerontology, Brown University School of Public Health, Providence, Rhode Island.), Yuan Zhang (3.Department of Health Services, Policy, and Practice & Center for Gerontology, Brown University School of Public Health, Providence, Rhode Island.), Dae H. Kim (1.Hebrew Senior Life, Hinda and Arthur Marcus Institute for Aging Research and Department of Medicine, Roslindale, Massachusetts; 2.Department of Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts), Darae Ko (1.Hebrew Senior Life, Hinda and Arthur Marcus Institute for Aging Research and Department of Medicine, Roslindale, Massachusetts; 4.Section of Cardiovascular Medicine, Boston Medical Center and Boston University School of Medicine, Boston, Massachusetts), Douglas P. Kiel (1.Hebrew Senior Life, Hinda and Arthur Marcus Institute for Aging Research and Department of Medicine, Roslindale, Massachusetts; 2.Department of Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts), Lori Daielo (3.Department of Health Services, Policy, and Practice & Center for Gerontology, Brown University School of Public Health, Providence, Rhode Island.), Tingting Zhang (3.Department of Health Services, Policy, and Practice & Center for Gerontology, Brown University School of Public Health, Providence, Rhode Island.), Andrew R. Zullo (3.Department of Health Services, Policy, and Practice & Center for Gerontology, Brown University School of Public Health, Providence, Rhode Island.; 5.Department of Epidemiology, Brown University School of Public Health, Providence, Rhode Island.)
Categories: Article
Source: Journal of the American Geriatrics Society
Doi: 10.1111/jgs.18702
Authors: Sarah D. Berry, Kaleen Hayes, Yoojin Lee, Yuan Zhang, Dae H. Kim, Darae Ko, Douglas P. Kiel, Lori Daielo, Tingting Zhang, Andrew R. Zullo
Little evidence exists about the comparative effects of first-line antihypertensive medications (i.e., renin-angiotensin-aldosterone converting enzyme inhibitors (RAASi), amlodipine, or thiazide diuretics) in older adults with limited life expectancy. We compared the rates of injurious falls and short-term cardiovascular events between different first-line antihypertensive medication classes in adults receiving care in nursing homes (NH).
This was a retrospective cohort of Medicare fee-for-service beneficiaries receiving care in NHs. Patients newly dispensed first-line antihypertensive medications were identified using Part D claims (2015–2018) and linked with clinical assessments (i.e., Minimum Data Set). Fall-related injuries (FRI), hip fractures, and major adverse cardiac events (MACE) outcomes were identified using hospitalization claims. Patients were followed from the date of antihypertensive dispensing until the occurrence of outcomes, death, disenrollment, or 6-month follow-up. Inverse-probability-of-treatment-weighted (IPTW) cause-specific hazards regression models were used to compare outcomes between patients who were new users of RAASi, amlodipine, or thiazides.
Our cohort included 16,504 antihypertensive users (RAASi, n=9,574; amlodipine, n=5,049; thiazide, n=1,881). Mean age was 83.5 years (± 8.2), 70.6% were female, and 17.2% were non-white race. During a mean follow-up of 5.3 months, 326 patients (2.0%) experienced an injurious fall, 1,590 (9.6%) experienced MACE, and 2,123 patients (12.9%) died. The intention-to-treat IPTW hazard ratio (HR) for injurious falls for amlodipine (vs RAASi) use was 0.85 (95% confidence interval (CI) 0.66–1.08) and for thiazides (vs RAASi) was 1.22 (95% CI 0.88–1.66). The rates of MACE were similar between those taking anti-hypertensive medications. Thiazides were discontinued more often than other classes; however, inferences were largely unchanged in as-treated analyses. Subgroup analyses were generally consistent.
Older adults with limited life expectancy experience similar rates of injurious falls and short-term cardiovascular events after initiating any of the first-line antihypertensive medications.
Hypertension affects two-thirds of older adults,^1^ and it is the most common, modifiable risk factor for cardiovascular disease.^2,3^ Approximately 80% of adults with hypertension are prescribed one or more medications to reduce blood pressure.^4^ A recent systematic review found no difference in the cardiovascular benefits of commonly used antihypertensive medications.^5^ Therefore, the American College of Cardiology recommends selecting either a renin-angiotensin aldosterone system inhibitor (RAASi), amlodipine, or a thiazide as first-line antihypertensive medication.^4,6^
Older adults that receive care in a nursing home (NH) have a high burden of multi-morbidity and are particularly vulnerable to adverse drug events. Falls are a common adverse event experienced by patients in NHs with rates as high as 150 falls/100 bed-years.^7^ Evidence from observational studies suggest that antihypertensive medications may be associated with injurious falls (i.e., falls that result in a fracture or other trauma).^8^ Antihypertensive medications may contribute to injurious falls through their effects on bone and through effects on falls risk. For example, thiazides (compared to RAASi) have been demonstrated to reduce the risk of fracture, with effects mediated by increased calcium absorption and gains in bone mineral density (BMD).^9^ Conversely, thiazides have a propensity to cause hyponatremia, gout, and urinary urgency,^10^ which could lead to a fall. A post-hoc analysis of the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack (ALLHAT) trial, which randomized hypertensive patients to lisinopril, chlorthalidone, or amlodipine, found that patients using chlorthalidone had the lowest risk of fracture^11^, whereas patients using amlodipine had the greatest risk of falls initially.^12^ However, ALLHAT excluded persons with a history of symptomatic heart failure, recent cardiovascular events, and other co-morbidities that would limit life-expectancy over the six-year study.^13^
In older patients receiving NH services, the contribution of falls to fracture risk is particularly important such that understanding the role of anti-hypertensive medications in fall risk could have implications for prevention. Further, older adults requiring NH services have a limited life expectancy and the cardiovascular benefits of antihypertensive medications may differ in this population. Our objective was to compare the rates of injurious falls and short-term cardiovascular events in older adults who initiated a first-line antihypertensive medication (i.e., renin-angiotensin-aldosterone converting enzyme inhibitors (RAASi), amlodipine, or thiazide diuretics) while receiving care in a NH. We hypothesized that patients receiving thiazides would have a higher risk of FRIs and fractures, with no difference in cardiovascular benefits, as compared with patients receiving RAASi or amlodipine.
This retrospective cohort study included NH residents aged ≥ 65 years with a new-prescription for amlodipine, a thiazide, or a RAASi. Data included the MDS (version 3.0) files linked to the Medicare enrollment files, MedPAR, and Parts B, and D claims from 2011–2018. The MDS is a federally mandated resident assessment instrument containing over 400 items to be completed on all U.S. NH residents at the time of admission and quarterly thereafter.^14^ Data for the MDS is primarily ascertained by nurse assessment, and for characteristics such as pain, depressive symptoms, and cognitive function, through resident interview. The clinically rich MDS has the advantage of allowing for adjustment of differences in physical function and other characteristics between new-users of the different anti-hypertensive medications. Figure 1 illustrates the study design. This research was approved by the Institutional Review Board of Hebrew SeniorLife.
We first identified 2,795,409 Medicare fee-for-service beneficiaries aged 65 years or greater who spent one or more days in a NH in 2016. NH location was determined by the Residential History File, a tool that summarizes Medicare data to account for patient location daily.^15^ Next we identified patients with one or more dispensings for the antihypertensive drug classes of interest after their NH admission date and between July 2015 and June 2018. The date of the antihypertensive claim was designated as the start date of the episode. Because medications are not fully captured by Part D while a patient is under the SNF benefit, we excluded episodes if the medication start date was <180 days from SNFclaims. We then restricted the episodes to new users, defined as no claims for the antihypertensive in the 180 days before the start date. We further excluded episodes 1) without continuous enrollment in Parts A, B, and D, 2) without a valid (i.e., admission, quarterly, or annual) MDS assessment, and 3) with a hospice claim in the 180 days before the start date. Next we excluded episodes with any claim for comparator antihypertensive drugs in the 180 days before the start date and patients with a claim for a combination drug that included more than one of the antihypertensive comparator classes of interest. Episodes with use of a first-line antihypertensive medication >180 days before the start date or with concomitant use of non-first line antihypertensive medications (e.g., beta-blockers) remained. We required participants to have a diagnosis of hypertension by the MDS or CMS’s Chronic Conditions Data Warehouse (CCW) in the 3-years before the start date. We excluded episodes with a relative or absolute contraindication to any of the antihypertensive comparator classes (i.e., renal artery stenosis, hyperkalemia, end-stage renal failure, angioedema) as documented by MDS or MedPAR claims in the 180 days before the start date. Lastly, for persons with multiple episodes of new antihypertensive use during the study period, we sampled the first episode. Supplementary Figure S1 summarizes the selection of the study cohort.
The causal contrast of interest was defined as the average treatment effect of initiating amlodipine versus RAASi or initiating a thiazide versus RAASi, regardless of subsequent treatment discontinuation or switching. This is the observational study analogue of the intention-to-treat estimand in a randomized controlled trial. We chose RAASi as the reference group because prior literature suggests that these medications have little effect on falls or fracture.^16^ Medications were ascertained using Part D claims.^17^ Supplementary Table S2 includes a list of medications included.
Outcomes were ascertained using hospitalization claims in the primary or first secondary claim position using MedPAR. ICD-9 codes were used to ascertain outcomes between January 2015-September 2015, whereas ICD-10 codes were used beginning October 2015
We identified NH residents who sustained injurious falls and hip fractures using inpatient claims (Supplementary Table S3). The validated outcome of hip fracture is similar to that used by Wright et al. with a positive predicted value of 97%^18^ and the algorithm to ascertain all injurious falls has been published.^19^ The definition of major adverse cardiovascular events, defined as any hospitalization for myocardial infarction, cerebrovascular events, or congestive heart failure is listed in Supplementary Table S3. Mortality was ascertained using the date of death in the Medicare Beneficiary Summary File through 2018.
For the primary analysis, follow-up began on day zero (i.e., the day the anti-hypertensive medication was started) and continued until the first outcome event, death, Medicare disenrollment, hospice claim, 6-months follow-up, or until study end (December 31, 2018; Figure 1). Six-months was selected as the end of follow-up given our hypothesis that thiazides would impact fall risk immediately.
In a secondary analysis, to account for changes from the initial treatment strategy groups, we censored follow-up at the time the antihypertensive was stopped (>45 day gap), switched to a comparator class, or a comparator class was added. This analysis emulates an as-treated approach.
We used the MDS in combination with Medicare enrollment files, and MedPAR claims data to obtain resident characteristics including demographics, medical conditions, function, and medication use. We selected characteristics that were associated with either the exposure and outcomes OR the outcomes alone based on prior literature and clinical expertise. Race/ethnicity were determined by Medicare Master Beneficiary Summary File’s code modified using the Research Triangle Institute algorithm. Other characteristics were measured using the MDS assessment closest to and preceding the start date, or using the CCW or Medicare Parts A and D claims in the 180 days before the start date. Fall history, dementia severity (as defined by the validated Cognitive Function Scale^20^), and transfer independence were assessed by the MDS, which is completed by a nurse coordinator. We calculated a modified Gagne Comorbidity Score for each resident using ICD diagnosis codes and MDS diagnoses.^21,22^ Long-stay, defined as a resident who spent ≥ 100 days in the same facility with no more than a 10 day gap, was ascertained using the Residential History File. For facility characteristics (e.g., number of beds, ownership), we used Certification and Survey Provider Enhanced Reports (CASPER). Because distance to a hospital could affect the likelihood of receiving inpatient care for NH residents with the outcomes of interest, we considered distance from the facility to the nearest admitting hospital using the American Medical Association dataset and methods previously published.^23^
We adjusted for confounding using inverse probability of treatment weighting (IPTW). We estimated the propensity score (i.e., the probabilities of receiving each of the antihypertensive treatments, conditional on covariates) via a multinomial logistic regression model using 88 characteristics to predict the use of each of the three antihypertensive medication classes of interest. Supplements S4&S5 provide the full list of characteristics included in the propensity score model, with more than 20 co-morbidities, concomitant use of 26 different medication classes, and the total number of medication classes included in the propensity score models. Propensity scores were used to construct stabilized IPTWs. Standardized mean differences (SMD) were used to assess covariate balance between antihypertensive groups before and after IPTW. (Supplementary Tables S4&S5). No trimming of weights was conducted as no outlier weights were identified (mean IPTW 1.0 [SD 0.39]).
Cause specific hazards regression models estimating hazard ratios (HRs) and robust 95% confidence intervals (CIs) compared safety and effectiveness outcomes of amlodipine versus RAASi and thiazides versus RAASi accounting for the competing risk of death. Models were fit using the IPTW. In the secondary “as-treated” analyses, we calculated stabilized inverse probability of censoring weights (IPCW) using the same methods and variables as IPTW. In the IPCW models, we estimated the probability of switching (IPCW1) and the probability of stopping each of the antihypertensive medication classes (IPCW2), separately. We then created a combined weight for each person by multiplying the IPTWIPCW1IPCW2; this combination weight was used for secondary analyses (mean IPCW1=1.0 [SD 0.21]; mean IPCW2=1.0 [SD 0.18]). For the primary analysis, persons admitted to a NH before 2016 would have had to survive until 2016 to be included. To ensure that survivor bias was not influencing our findings, we conducted a sensitivity analysis where we restricted the study population to patients with a new anti-hypertensive medication dispensing in 2016 or later (n=12,757). All data were analyzed by SAS v9.4 (SAS Institute, Cary, NC, USA).
We conducted subgroup analyses to examine treatment effect heterogeneity for the following groups as we hypothesized they may result in differential safety effects of antihypertensive recent history of fall (yes versus none), dementia severity (intact or mild impairment versus moderate to severe impairment), and transfer independence (independent or mild assistance with transfers versus extensive or total assistance). For all subgroup analyses, we constructed new stabilized IPTWs where the propensity scores for the numerator were estimated conditional on effect measure modifiers (i.e., subgroup characteristics; example code provided in Supplementary Material S6), then used an interaction term within models to provide effects specific by levels of the effect modifier (e.g., by fall status).^24^ All analyses considered effect measure modification on the multiplicative scale to assess whether the combined effect of the subgroup characteristic with the antihypertensive medication exerted a greater association on each outcome than the effect of the characteristic or antihypertensive medication alone.
Our final cohort included 16,504 new users of antihypertensive medications (RAAS, n=9.574; amlodipine, n=5,049; and thiazide, n=1,881). The mean age was 83.5 years (± 8.2), 70.6% were female, and 17.2% were Black or non-white race. Most patients (88.6%) were long-stay residents. Select characteristics of the cohort according to first-line antihypertensive use are listed in Table 1. Before weighting, patients using amlodipine were more likely to be Black, to have dementia, and to have received care in an urban facility as compared with those using RAASi or thiazaides. Patients using RAASi were more likely to have coronary artery disease, congestive heart failure, and diabetes as compared with those using amlodipine or thiazides. Patients using thiazides were more likely to be female, have experienced a recent fall, and to have osteoporosis as compared with those using RAASi or amlodipine. Patients using amlodipine had more functional impairment than those using RAASi or thiazides (median ADL score 18 vs 16 and 17, respectively). After weighting, there was adequate balance across all characteristics (Supplementary Table S5): the SMD for all comparisons was <0.05.
Supplementary Table S7 provides the unadjusted associations and association after IPTW between the medication classes and outcomes. During follow-up, 326 patients (2.0%) experienced an injurious fall and 2,123 patients (12.9%) died. The overall rate of injurious falls was 44.3/1000 person years. After weighting, the HR for injurious falls for amlodipine (vs RAASi) was 0.85 (95% CI 0.66 to 1.08) and the HR for thiazides (vs RAASi) was 1.22 (95% CI 0.88, 1.66). Figure 2 presents the cumulative incidence of injurious falls according to medication type. Results were similar in the sensitivity analysis among the 77.3% sample with a new antihypertensive dispensing in 2016–2018 (Supplementary Table S8).
Overall, 193 patients experienced a hip fracture (26.1 fractures/1000 person years). After weighting, the HR for hip fracture for amlodipine (vs RAAS) was 0.78 (95% CI 0.55 to 1.08) and for thiazides (vs RAAS) was 1.34 (95% CI 0.88, 1.96).
Supplementary Table S9 provides the unadjusted associations and associations after IPTW between the medication classes and outcomes. During follow-up, 303 patients (1.8%) experienced a myocardial infarction or cerebral vascular accident, 613 (3.7%) heart failure, and 771 (4.7%) any MACE. The overall rate of major adverse cardiovascular events was 105.9/1000 person years. After weighting, the HR for cardiovascular events for amlodipine (vs RAASi) was 0.94 (95% CI 0.81 to 1.10) and for thiazides (vs RAASi) was 0.92 (95% CI 0.73, 1.16). Figure 3 presents the cumulative incidence of major adverse cardiovascular events according to medication type. Results were similar among the 77.3% sample with a new antihypertensive dispensing in 2015–2018 (Supplementary Table S10).
Discontinuation and switching occurred more commonly in individuals prescribed thiazides versus RAASi or amlodipine. For example, by 6-months follow-up, 17.4% of patients using RAASi and 15.7% using amlodipine discontinued the medication, as compared with 24.1% of those using thiazides. After accounting for IPCW, adjusted associations of safety and effectiveness outcomes were similar to the ITT results. For example, the HR for injurious falls for amlodipine (vs RAAS) was 0.85 (95% CI 0.63, 1.14) and for thiazides was 1.35 (95% CI, 0.89, 1.95). For major adverse cardiovascular events, the HR for amlodipine (vs RAAS) was 1.01 (95% CI 0.84, 1.22) and for thiazides was 0.78 (0.56, 1.08).
Figure 4 shows the results of the subgroup analyses. Estimates were similar in patients with and without a history of fall and regardless of dementia severity. Patients prescribed thiazides (versus RAASi) who also were dependent with transfers (n=9,735, 59.0%) had an increased risk of injurious falls (HR: 1.68, 95% CI 1.06, 2.57), but these medications had no association with injurious falls in patients who were independent with transfers (HR: 0.87, 95% CI 0.53, 1.34).
In a large sample of older Medicare beneficiaries receiving care in a NH, we found no statistically significant difference in the rates of injurious falls and cardiovascular events between patients initiating any of three types of first-line antihypertensive medications over 6 months follow-up. Discontinuation occurred more commonly among individuals using thiazides, but adjusting for potential informative censoring did not change our results. Our results support current guidelines that suggest any of these medication classes are appropriate first-line choices for older adults initiating a medication for hypertension.
Most studies that have conducted a head-to-head comparison of first-line antihypertensives have focused on long-term effectiveness outcomes, including cardiovascular events. For example, in the ALLHAT trial, there were fewer cardiovascular events among the thiazide users as compared with angiotensin-converting enzyme inhibitor or amlodipine users^25^ It is possible that these results are due to more optimal control of blood pressure with thiazides or because the thiazide dispensed in this trial was chlorthalidone. Chlorthalidone is seldom used clinically due to concerns of hypokalemia,^26^ and its use was rare in our study (<5% thiazide users). In a meta-analysis of 58 trials including ALLHAT that compared various antihypertensive regimens, there was no difference in the risk of cardiovascular events among RAASi, amlodipine and thiazide users.^27^ While we did not necessarily expect to see an adjusted difference in major cardiovascular events over 6-months follow-up, our results support these findings.
Fewer studies have compared injurious falls among patients newly prescribed first-line antihypertensive medications. A post-hoc analysis of the ALLHAT trial suggested thiazide users had the lowest risk of fracture.^11^ Amlodipine users had the greatest risk of falls in the first year only,^12^ with no differences in orthostatic hypotension or syncope between groups. In a prospective cohort of community dwellers, angiotensin–converting enzyme inhibitors and calcium channel blockers were associated with a decreased risk of falls, compared with no antihypertensive treatment.^28^ This study did not directly compare risk of falls in persons taking medications, although rates appeared to be similar. In a retrospective cohort study of Medicare beneficiaries, there appeared to be no difference in injurious falls rates among RAASi, calcium channel blockers, or diuretic users.^8^ We also found no statistically significant difference in fall rates among older adults withmultimorbidity initiating a first-line antihypertensive medication. Our non-significant results do not preclude the possibility of a true increased or decreased risk of falls.
Although rates of injurious falls were considerably lower than rates of cardiovascular events in our study, older adults with multimorbidity may prioritize safety over prevention of cardiovascular outcomes. Tools exist to help align treatment with what matters most to patients;^29^ however, the majority of these tools were developed in the community setting or if developed in the NH setting, were used to facilitate advanced directive conversations. It is important that tools to guide treatment include patient preferences and that they are tailored and tested in the NH setting.
We found no difference in heterogeneity of treatment effects according to dementia severity or history of falls. Among individuals with transfer dependence, thiazides (versus RAAS) were associated with a greater risk of injurious falls, but we did not observe this association in persons who were independent in transfers. This observation is counter to our hypothesis that thiazides would increase fall and injury risk through urinary symptoms in persons who were ambulatory. It is possible that individuals with dependence in transfers are particularly sensitive to adverse effects of thiazides, perhaps by trying to get out of bed to use the toilet and falling to the ground because of mobility limitations; however, it is likely this observation represents type-I error given multiple hypothesis testing.
There are potential limitations to our study. First, given the observational study design, confounding may exist. Covariates were measured using the MDS or claims data and we did not have information on laboratory values (e.g., creatinine clearance) or clinical presentation that may have influences prescribing. We handled measured confounding using IPTW, but it is possible that unmeasured confounding remains. Second, we had no data on blood pressure control. A randomized repeat-crossover trial of adults with hypertension in Sweden demonstrated substantial heterogeneity in treatment response to first-line antihypertensive medications, with on average greater systolic blood pressure reductions with RAASi as compared with amlodipine or thiazides.^30^ The optimal blood pressure target for older adults with multimorbidity remains controversial,^31^ and additional study is warranted to identify preferred targets. Third, discontinuation occurred more often with thiazide users and it is possible that had thiazide users continued the medication that the results would differ. We conducted an as-treated analysis, and found similar results; however, we acknowledge that we did not have detailed information on reasons for discontinuation (e.g., hypotension, hypokalemia or hyponatremia) forthe analyses. Fourth, the definitions of outcomes were restricted to inpatient claims, and less severe, non-hospitalized events were not captured. Inclusion of outpatient claims would likely identify false positives or historic, rather than incident, events.^32^ Finally, this study was conducted among NH residents with a high burden of multimorbidity but also having a greater likelihood of adherence with the medications when administered by NH staff, and results may not be generalizable to healthier populations. The evidence base is already much stronger for healthier populations, and the need to generalize this study’s evidence to these individuals is low.
In summary, when selecting a first-line antihypertensive medication for older adults with multimorbidity, providers must balance effectiveness and safety concerns. In a population of adults requiring NH services, we found no statistically significant difference in the rates of injurious falls and cardiovascular events among patients initiating RAASi, amlodipine, or thiazides that would guide treatment decisions. Treatment decisions should instead be informed by co-indications, such as heart failure, or monitoring requirements, like phlebotomy. Thiazide diuretics were discontinued more frequently. Given the known variability in blood pressure response to these medications and the uncertain blood pressure target in this population, additional inquiry is needed to guide the first-line choice of treatment for hypertension in adults with more limited life expectancy.