Deprescribing Anticholinergic and Sedative Drugs to Reduce Polypharmacy in Frail Older Adults Living in the Community: A Randomized Controlled Trial

Abstract

Background

Polypharmacy is associated with poor outcomes in older adults. Targeted deprescribing of anticholinergic and sedative medications may improve health outcomes for frail older adults. Our pharmacist-led deprescribing intervention was a pragmatic 2-arm randomized controlled trial stratified by frailty. We compared usual care (control) with the intervention of pharmacists providing deprescribing recommendations to general practitioners.

Methods

Community-based older adults (≥65 years) from 2 New Zealand district health boards were recruited following a standardized interRAI needs assessment. The Drug Burden Index (DBI) was used to quantify the use of sedative and anticholinergic medications for each participant. The trial was stratified into low, medium, and high-frailty. We hypothesized that the intervention would increase the proportion of participants with a reduction in DBI ≥ 0.5 within 6 months.

Results

Of 363 participants, 21 (12.7%) in the control group and 21 (12.2%) in the intervention group had a reduction in DBI ≥ 0.5. The difference in the proportion of −0.4% (95% confidence interval [CI]: −7.9% to 7.0%) provided no evidence of efficacy for the intervention. Similarly, there was no evidence to suggest the effectiveness of this intervention for participants of any frailty level.

Conclusion

Our pharmacist-led medication review of frail older participants did not reduce the anticholinergic/sedative load within 6 months. Coronavirus disease 2019 (COVID-19) lockdown measures required modification of the intervention. Subgroup analyses pre- and post-lockdown showed no impact on outcomes. Reviewing this and other deprescribing trials through the lens of implementation science may aid an understanding of the contextual determinants preventing or enabling successful deprescribing implementation strategies.

Polypharmacy, the use of ≥5 medications, is common in older adults. Of adults, 60–79 years of age in the United States and Canada, 34.5% and 30.9% are prescribed ≥5 medications (1). In European countries, the prevalence in that age group is at least 26% (2) and may exceed 40% (3). In New Zealand, 31% of adults over the age of 65 are prescribed ≥5 medications (4).

Potentially inappropriate medications (PIMs), those with a greater risk of harm than benefit, are commonly prescribed for older adults, resulting in an increased prevalence of polypharmacy (3) and potential for medication-related harm such as falls, fractures, adverse drug reactions, hospitalizations, and mortality (5).

Medications with anticholinergic and sedative effects have the potential to cause more harm than good in older people, particularly when used long term. These medications can impair muscle strength and worsen cognition (6). Greater exposure to anticholinergic and sedative medications is associated with poorer physical and cognitive functioning and a higher risk of falls, fractures, impaired cognition, hospitalization, entry into residential care, death, and other adverse drug events (7–15). The Drug Burden Index (DBI), an internationally validated tool, calculates cumulative exposure to anticholinergic and sedative medications using pharmacokinetic and pharmacodynamic principles (16,17). Half a unit on the DBI is equivalent to 1 medication taken at its minimum efficacious dose. An increasing DBI score is linked to worsening cognitive and physical functioning in older people (9,11–13,16). Deprescribing, the process of reducing or discontinuing PIMs (18) may reduce negative health outcomes (11–13,19). Health care professionals should therefore strive to routinely deprescribe anticholinergic and sedative medicines to reduce potential medication-related harm in this vulnerable population.

There is a wealth of literature in the field of deprescribing trials. Use of the term “deprescribing trials” to search PubMed returned nearly 1 400 contributions since 2018, the year our study commenced. Three recent systematic reviews cover the field well. The first reviewed deprescribing interventions for community-dwelling older adults who had received comprehensive medication reviews (CMRs). It concluded that despite evidence that CMRs reduce the number of PIMs, there is little to no evidence that they reduce hospitalizations, quality of life, or falls. However, they may result in a slight reduction in all-cause mortality (20). The second review focused on deprescribing interventions among older people with frailty. It identified 3 pharmacist-led and 3 multidisciplinary-led interventions. All studies reported a reduction in the number of PIMs and some positive outcomes on clinical endpoints but no significant impact on quality of life (21). The third review, which reviewed reviews, focused more broadly on polypharmacy in older adults with multimorbidity, indicated that while interventions are associated with a reduction in the number of PIMs, there is minimal evidence to suggest they improve clinical and intermediate outcomes (22).

However, there is no clear evidence of the effectiveness of a medication review in reducing anticholinergic and sedative medicines as measured by the DBI in subgroups of frail older adults. Frailty in older adults describes the increasing vulnerability to adverse outcomes following minor insults and reflects depleting physiological reserves from the cumulative impact of such declines across the human body (23). Identifying older adults’ levels of frailty may help guide medication management and other aspects of clinical care. Specifically, frailty is associated with pharmacokinetic and pharmacodynamic changes and increased vulnerability to adverse drug effects and poor prognosis (24,25). Taking the degree of frailty into account when deprescribing can facilitate individualizing medication regimens and reduce the use of harmful medications older people take (26–29). However, we are unaware of any study where a frailty measure embedded in a routine, standardized assessment was used to identify older persons warranting a deprescribing intervention. In this study, we tested the utility of a frailty index (FI) based on standardized interRAI assessment data to target a medication decision-making intervention for older adults. The interRAI is a comprehensive suite of instruments used to standardize the evaluation of the complex care needs of older adults. In New Zealand, the Home Care (HC) assessment is routinely used for older persons seeking admission to publicly funded residential care; the shorter Contact Assessment (CA) is used to assess the provisioning of home-based support services (30-32). Both instruments contain questions across multiple domains that assess medical, social, and functional status (33).

Our primary aim was to test, by conducting a randomized controlled trial (RCT), whether patient-specific deprescribing recommendations developed by pharmacists following a medication review and provided to the patient’s family physician or general practitioner (GP) reduced the use of anticholinergic and sedative medications. In addition, we hypothesized that any resulting reduction in DBI score would be more pronounced for older adults with a greater level of frailty.

Method

A summary of the study protocol is provided here; we refer to the previously published study protocol for further details, such as recruitment, randomization, and intervention processes (34).

Design

We designed a 2-arm randomized controlled superiority trial to determine the effectiveness of a deprescribing intervention in a real-life clinical setting (35). After consenting to participate in the study, participants were randomly 1–1 allocated to either the intervention arm or usual care as the control arm. Intervention arm participants received a pharmacist-led assessment whereby recommendations to deprescribe targeted medications (Supplementary Table 1) were sent to their GP for clinical decision making. The DBI was used to quantify the use of targeted sedative and anticholinergic medications for each participant. We identified medicines with anticholinergic and sedative properties using medicines available in New Zealand and based on previously published literature (11,12,16). Medicines with sedative and anticholinergic properties were identified from the New Zealand Formulary and Medsafe datasheets maintained by the New Zealand Medicines and Medical Devices Safety Authority. Trial participants were stratified into low, medium, and high-frailty via the FI, which contained items common to interRAI-HC and CA assessments (36).

The study was approved by the New Zealand Health and Disability Ethics Committee (17/CEN/265) and registered prior to participant recruitment on the Australian New Zealand Clinical Trials Registry (ACTRN12618000729224).

Participants

Participants were recruited between September 25, 2018 and October 30, 2020 and included in the study if they:

1. were 65 years or more in age,

2. had undergone an interRAI-HC (31) or CA (32),

3. were community-dwelling in either the South Canterbury or the Canterbury district health board catchments of New Zealand, and

4. were taking at least 1 medication with anticholinergic or sedative effects regularly at the minimum registered daily adult dose (which would result in DBI ≥ 0.5).

Participants were excluded if they:

1. did not consent to the use of interRAI data for research, or

2. had a diagnosis of Alzheimer’s disease, dementia, schizophrenia, abnormal thought processes, delusions, or hallucinations as coded in the interRAI assessment, or

3. scored 3 or higher on the interRAI Cognitive Performance Scale (37), or

4. had a terminal illness with ≤6 months of life expectancy, or

5. were identified as not-frail (ie, no deficits on the FI), or

6. had a DBI of < 0.5, or

7. had a potentially life-threatening drug interaction requiring urgent medical attention during the study period.

All GPs were advised that a person under their care had consented to participate in the study.

Process and Intervention

Participants were recruited following an interRAI-HC or interRAI-CA assessment. Assessment data were used to establish eligibility and frailty level. All eligible participants were stratified into 1 of the 3 frailty strata—low, medium, or high—based on the predefined cutoffs of a cumulative deficit-model FI. Although the cutoffs defined the tertiles in the target population, the cutoff values do not have clinical meaning. The FI was based on items common to the HC and CA instruments (36). This restricted the number of items but covered multiple domains and accounted for a maximum of 15 deficits. The FI was calculated by dividing the number of deficits by the number of scored items and validated using 324 249 interRAI assessment records of older persons in New Zealand. Despite the low number of deficit items, the FI demonstrated similar performance to a previously validated FI with 42 deficit items (38).

Each participant experienced, within their home, a pharmacist-conducted medication review (baseline review) by 1 of the 4 study pharmacists. All pharmacists received training specific to the intervention and followed the deprescribing guidelines defined in the study protocol. There was no preexisting relationship between study pharmacists and participants or their GP. No automated or systematized decision tools, such as software algorithms or decision trees, were used to guide deprescribing. During the review, the pharmacist recorded all medications and supplements the person was taking. However, the pharmacist did not have access to the participants’ clinical notes or medication records and all clinical decision making, including prescribing, remained with the GP. During coronavirus disease 2019 (COVID-19) lockdowns, recruitment into the study, and the intervention were suspended.

After each assessment, the participant’s DBI was calculated. If no grounds for exclusion were identified, the participant was randomized within their frailty strata through the use of 3 sets of concealed envelopes. Pharmacists knew participants’ DBI, frailty, and allocation but did not disclose this information to participants.

After the baseline review, the pharmacist discussed the following matters with the intervention participants: the medical condition for which the DBI medications had been prescribed; the timeframe for taking the medicines; the participant’s experiences with these medicines; and any concerns the participant had about any of the medicines. Concerns were elicited to help identify each participant’s need for any of the medications or any possible side effects from them.

The pharmacist sent a letter to the participant’s GP that outlined all medicines contributing to the DBI that the participant was taking, that person’s experiences with these medicines, and suggestions (where appropriate) as to which medications might be deprescribed. Neither frailty level nor DBI was disclosed to the GP. To ensure consistency of the intervention and subsequent recommendation, all letters to GPs were peer-reviewed by another clinical pharmacist.

The deprescribing intervention focused on reducing or stopping anticholinergic and sedative medications to reduce participants’ DBI scores, using a process adapted from Reeve et al. (39) and previously trialed in a New Zealand residential aged care setting (40). All medications, their dose and frequency of use, and the DBI calculations were recorded in the study database. The study’s senior pharmacist reviewed all DBI calculations.

In the control arm, all medications were recorded as per the intervention arm, after which the pharmacist’s role ended. The pharmacist neither discussed deprescribing options with the control participants nor sent deprescribing suggestions to the control participants’ GPs.

Participants underwent a second medication review (follow-up review) in their homes at least 6 months after the first. The follow-up review was conducted by a different pharmacist blind to participants’ allocation. Necessitated by COVID-19 lockdowns but also used in situations where participants had moved to aged residential care (ARC) or were unable to accommodate another visit, medication information for some participants during the second review was obtained through phone interviews, pharmacy dispensing records or the medication records of the aged care facility.

Outcomes

The primary outcome measure was the difference in DBI between medication review at baseline and follow-up. Secondary outcomes included emergency department (ED) visits, hospital admissions, admissions to ARC, and mortality. Secondary outcome measures were determined using routinely collected administrative data provided by the Ministry of Health. All New Zealand residents have a unique National Health Identification number that enables linkage of all interRAI, hospital, ARC, and mortality records. For secondary events, all patients were followed until the first event or the census date of June 5, 2021, the final date of the second review of all participants.

Statistical Analyses

Data were analyzed using an intention-to-treat analysis method. Summary data are presented as n (%) and means (standard deviation [SD]) or medians (interquartile range), depending on the skewness of the data. Proportions were compared with Pearson’s chi-squared test, and the difference was presented with a confidence interval (CI). Secondary analyses used a logistic regression model to account for confounding factors; we report odds ratios. Finally, Kaplan–Meier survival curves and competing risks Cox regression were used to analyze time-to-event data (41).

The study was powered for a clinically meaningful reduction in DBI of ≥0.5, with an effect size of a net 10% difference in the proportions of this reduction, with a power of 90% and α = 0.05. This calculation was based on previous pilot data that determined feasibility (40) and required 167 participants in each arm of the trial. Using the assumption of even numbers in each frailty stratum, we determined that 112 participants per stratum were required for an effect size of 20% with a power of 80% and α = 0.017 (0.05/3).

Prespecified subgroup analyses were done by (a) frailty strata and (b) recruitment district health boards (Canterbury and South Canterbury). Secondary analyses focused on time-to-event, time to any presentation at an ED, admission to hospital, entry into ARC, and all-cause death.

COVID-19 Local Lockdowns

On March 23, 2020, all of New Zealand went into one of the world’s most stringent lockdowns, which lasted until May 13, 2020 (42). As a result, we immediately suspended medication assessment visits and secured an ethics amendment for telephone-based follow-up medication reviews. These began on April 29, 2020 and continued until home visits could be restarted. Subsequent lockdowns were local to the Auckland and Northland districts and did not affect this study.

At the time, we hypothesized several ways how this lockdown might affect the study outcome: (a) fewer interRAI assessments being conducted, thus lower recruitment; (b) reduced GP capacity to process deprescribing recommendations, potentially lowering the effect of the intervention; (c) increased stress levels increasing prescribing of DBI-relevant sedative medications to address increased anxiety levels and sleeping problems; and (d) other unknown affects on patient behavior. We therefore prespecified a further subanalysis to our study when our work and the COVID-19 pandemic intersected. Our intention was to assess if there was an impact, not what nature such an impact might have been. For this subanalysis, we compared the prelockdown group with participants who completed the study impact-free before lockdown started with the post-lockdown groups for whom we hypothesized an impact of public health measures imposed in response to the pandemic.

Results

Of the 733 potential participants identified, 361 did not want to participate in the study or were excluded before the first medication review. During that review, another 9 with DBI < 0.5 were excluded. This left 363 participants who received a medication review; 184 (50.7%) were randomized to the intervention. Seven participants withdrew (3 control; 4 intervention) and 18 died (10 control; 8 intervention) before the second medication review, leaving 338 (171 intervention, 166 control) for whom the primary outcome could be calculated. Participants’ mean age was 79.9 (SD = 7.0) years, and the sample was predominantly female (66.4%) and New Zealand European (97.5%). Two participants, one aged 60 and one aged 63, were inadvertently included because their age was not available at the time of screening. For 116 of the 171 intervention participants, recommendations were articulated to change between 1 and up to 3 DBI medications. The 6 most commonly used medicines at the initial medication review were codeine, citalopram, zopiclone, doxazosin, gabapentin, and amitriptyline (Supplementary Table 1).

No exclusions occurred due to health or medication-related problems requiring urgent attention during the study. The study involved 226 GPs in 98 medical centers, with at least 1 participant under their care. Participant flow and descriptive statistics of the study cohort are shown in Figure 1 and Table 1, respectively.

Figure 1.

Consort diagram: participant flow for deprescribing randomized controlled trial (RCT).

Table 1.

Participant Demographics Prior to Stratification

	Control (N = 179)	Intervention (N = 184)	Total (N = 363)
Age
Mean (SD)	80.6 (7.0)	79.2 (6.8)	79.9 (6.9)
Range	60–97	65–96	60–97
Sex
Female (%)	123 (68.7)	118 (64.1)	241 (66.4)
Ethnicity
Māori (%)	2 (1.1)	5 (2.8)	7 (1.9)
Pacific peoples (%)	0 (0.0)	1 (0.6)	1 (0.3)
Asian (%)	1 (0.6)	0 (0.0)	1 (0.3)
NZ European (%)	175 (98.3)	175 (96.7)	350 (97.5)
N-Miss (%)	1	3	4
Frailty index
Mean (SD)	0.276 (0.128)	0.278 (0.124)	0.277 (0.126)
Range	0.033–0.717	0.020–0.667	0.020–0.717
Frailty strata
Low (%)	59 (33.0)	58 (31.5)	117 (32.2)
Medium (%)	80 (44.7)	87 (47.3)	167 (46.0)
High (%)	40 (22.3)	39 (21.2)	79 (21.8)

Note: SD = standard deviation.

Primary Analyses

Twenty-one (12.7%) participants in the control group had a reduction in DBI ≥ 0.5; 21 (12.2%) participants in the intervention group had a reduction in DBI ≥ 0.5. The difference in the proportion of participants (intervention–control) with a reduction in DBI ≥ 0.5 was −0.4% (95% CI: −7.9% to 7.0%, p ≈ 1). The negative sign indicates that a smaller proportion of participants in the intervention arm than in the control arm had a reduction in DBI ≥ 0.5. Logistic regression was performed to estimate the odds ratios for potentially confounding variables (Table 2). After accounting for the DBI at the first medication review, FI, age, sex, and time between medication reviews, the odds of reducing DBI by ≥0.5 in the intervention compared to the control was 0.96 (95% CI: 0.48 to 1.92). A reduction in DBI of ≥0.5 depended primarily on how large the DBI was at the baseline medicine review (the higher the DBI, the more likely a reduction). It also depended less on sex (males more likely) and time between medication reviews (longer times more likely). Figure 2 shows the distribution of baseline DBI for our study participants as a histogram.

Table 2.

Logistic Regression for Outcome: Drug Burden Index ≥0.5

	OR	OR 95% CI
Arm (reference = control)	0.96	0.78 to 1.93
Premedication assessment DBI (per 1.0)	3.26	1.98 to 5.64
Frailty index (per 0.1)	0.83	0.62 to 1.10
Age (per year)	1.04	0.99 to 1.10
Sex (reference = female)	1.94	0.95 to 3.97
Time between medication reviews (per week)	1.05	1.01 to 1.10

Notes: CI = confidence interval; OR = odds ratio.

Figure 2.

Number of participants by Drug Burden Index (DBI) at baseline medication assessment.

Subgroup Analysis

Prespecified subgroup analyses encompassed frailty and recruiting district. The mean DBI (SD) in each stratum were low-frailty 1.17 (0.57), medium-frailty 1.17 (0.67), and high-frailty 1.20 (0.64). Within each frailty stratum, the differences in the proportion of participants (intervention–control) with a reduction in DBI of ≥ 0.5 were low-frailty 9.7% (−4.2% to 23.5%), medium-frailty −5.5% (−17.4% to 6.4%), and high-frailty −5.2% (−22.2% to 11.8%; Figure 2).

In South Canterbury DHB, 4 of the 12 participants (33.3%) in the intervention arm had a reduction in DBI of ≥ 0.5 compared to 1 of 9 (11.1%) in the control arm (difference 22.2% [95% CI: −21.2% to 65.6%]). In the Canterbury DHB, the difference was −2.11% (95% CI: −9.8% to 5.6%).

A post hoc subgroup analysis was done to assess the potential impact of lockdown on the study data during the COVID-19 pandemic. At the time of lockdown, 84 participants had completed the study and were therefore not affected by the pandemic or any health measure put in place. Partially through the study, therefore potentially affected, were 99 participants who had their baseline assessment before lockdown. A further 155 participants were recruited and completed the study after the COVID-19 lockdown measure was lifted; 52 had already been identified before lockdown. To be counted into the impact-free prelockdown group, participants needed to have had their initial assessment and their 6-month follow-up completed. All other participants were pooled into the post-lockdown group. Of the prelockdown group, 16.7% (n = 7) in the intervention arm and 4.8% (n = 2) in the control arm had a reduction in DBI ≥ 0.5 (difference: 11.9% [95% CI: −3.5% to 27.3%]). After the beginning of the lockdown, the difference was −4.55% (95% CI: −13.6% to 4.5%). Figure 3 shows the subanalysis results in a Forest plot.

Figure 3.

Results of subgroup analyses by frailty, COVID-19 lockdown, and District Health Board (DHB) catchment. CDHB = Canterbury District Health Board; COVID-19 = coronavirus disease 2019; DBI = Drug Burden Index; SCDHB = South Canterbury District Health Board.

Secondary Analyses

During a median follow-up of 254 days, 218 (60.1%) participants visited the ED. The hazard ratio (HR) for an association of the intervention with a presentation to ED was 1.06 (95% CI: 0.82 to 1.39). During a median follow-up of 473 days, 60 (16.5%) participants were admitted to a hospital. The HR for an association of the intervention with hospital admission was 1.24 (95% CI: 0.74 to 2.06). During a median follow-up of 507 days, 49 (13.5%) participants entered ARC. The HR for an association of the intervention with entry to ARC was 1.14 (95% CI: 0.65 to 2.00). During a median follow-up of 516 days, 26 (7.2%) participants died. The HR for an association of the intervention with death was 0.98 (95% CI: 0.45 to 2.10).

Discussion

In this pragmatic RCT, the pharmacist-led medication review did not result in a clinically meaningful reduction in DBI scores for the intervention group. Additionally, the CI of the differences for each frailty stratum were so wide that we have to conclude that the intervention made no meaningful difference for participants in those strata.

Implementing deprescribing interventions is challenging (43). The low prevalence of reduction in DBI via this pharmacist-led intervention concurs with the findings of other community-based deprescribing trials conducted worldwide. For example, an Australian cluster RCT of the Goal-directed Medication Electronic Decision Support System (GMEDSS) that helped accredited pharmacists conduct home-medication reviews found an increase in recommendations to reduce anticholinergic and sedative drugs, thus DBI. However, no significant change in DBI was observed at the time of the 6-month review (44). In a Swiss cluster-randomized trial of older community-dwelling patients with polypharmacy, a patient-centered deprescribing procedure effectively reduced the number of medications taken immediately after the intervention but not after 6 and 12 months (45). An RCT based in a northern Netherlands community with a shorter 3-month intervention period found no difference in the proportion of participants with a DBI reduction of 0.5 between the control arm and the intervention arm (46).

Trials undertaken in hospitals and aged care facilities identified a high uptake of deprescribing recommendations (27,40,47). In these situations, as was evident in the study by Ailabouni et al. (40) that informed our intervention, once the decision to deprescribe has been agreed on, nurses deliver medication according to chartered lists from the prescribing physician.

Having pharmacists discuss medication use with the participants and provide recommendations for deprescribing to their GPs did not reduce participants’ use of anticholinergic or sedative drugs. Previous research has found that pharmacist-led deprescribing RCTs involving participant education succeeded in community and outpatient settings (47,48). In the D-PRESCRIBE trial (49), of the 489 community-dwelling older participants, 49% in the intervention group compared with 12% in the control group had medications deprescribed that were inappropriate according to Beers Criteria. A small RCT (150 participants) involving deprescribing non-benzodiazepine sedative-hypnotic drugs and participants randomized to usual care, educational information only, or educational information with pharmacist consultation demonstrated successful deprescribing in both intervention arms (47,48,50).

We found no clear indication in our study that frailty influenced deprescribing. This result may reflect conflicting influences on prescribers. On the one hand, prescribers may be reluctant to deprescribe for people with a higher degree of frailty because of concerns about disrupting homeostasis (aka rocking the boat). On the other hand, prescribers may be more willing to deprescribe in frailer older people because of evidence of their increased vulnerability to adverse drug reactions, limited time to benefit from preventative medications, and little available data on the benefit of particular drug treatments in this population (21,24,51).

Contextual Considerations

Two aspects of our study need to be considered because of their possible impact on our findings.

First, the global COVID-19 pandemic disrupted our recruitment of participants and participant interviews. With considerable attention directed toward managing the pandemic within the health care system, GPs faced capacity issues. We, therefore, decided to conduct a retrospective subgroup analysis for participants who completed the study before the March 2020 lockdown. In the prelockdown subgroup, more participants in the intervention arm than in the control arm had a reduction in DBI of ≥0.5. However, the numbers of participants involved were very small, and we observed the opposite pattern post-lockdown in favor of the control. Because we can offer no reasonable explanation for this finding beyond random chance, we cannot suggest that findings favoring the intervention might have occurred in the absence of the Covid-related lockdown.

Second, after the study’s proposal and funding phases but before participant recruitment commenced, a primary health care organization in Canterbury initiated targeted educational activities on polypharmacy. This education program has a demonstrated effect on subsequent prescribing behavior (52). However, because the mean time to extinction of effect is 14.5 months, there is little overlap with the study baseline and likely no effect on any seen between-group differences. It is also theoretically possible that the education program primed GPs to respond to the letter regarding control group patients, thus reducing the between-group differences; as noted, there was a 12% reduction in the control arm. This reduction could have affected the potential for increasing deprescribing compared with usual care. However, we do not see this as likely because the same education initiative did not occur in South Canterbury, where the decrease in the control arm was similar.

Limitations

First, pharmacists did not have access to participants’ medical records and therefore relied on information conveyed by each participant. Likewise, participants’ medication records were unavailable before the pharmacists’ first home visits. These limitations may have attenuated the effect of a pharmacist-led medication review. To assess the accuracy of medication lists, pharmacists compared pharmacy labels on bottles and containers with what participants said they were taking.

Second, the intervention did not emphasize GP–pharmacist collaboration. Although pharmacists advised GPs of potential medications for deprescribing, the intervention had no mechanism to influence the GPs’ or the participants’ uptake.

Third, the study’s design did not include individual follow-up with GPs or participants to determine reasons for acting or not acting on the suggestions made, nor did it include other means of collecting implementation-related data to determine where and why the intervention worked or failed.

Fourth, because the primary outcome of our study was extremely stringent, more so than primary outcomes in similar studies such as GMEDSS (53), it may have missed the small changes in DBI that occur during the weaning process of many anticholinergic and sedative drugs. We assessed no other prescribing or deprescribing outcomes that might have arisen from the pharmacists’ reviews, including medication changes that may have fewer adverse effects but do not alter the measured DBI. Furthermore, we measured the net change in DBI affected by deprescribing and new prescribing, not deprescribing alone.

Fifth, the study was powered for a clinically significant but large absolute difference in change in deprescribing. However, we did not meet our prespecified target number of patients for the high-frailty stratum, suggesting the analysis for this stratum is underpowered.

Finally, while our recruitment successfully achieved sample size, we did not engage a sufficient number of Māori to represent New Zealand’s indigenous people at the corresponding general population level.

Strengths

Our use of the interRAI assessments, collaboration with local health delivery organizations, and participant recruitment method and process allowed us to exceed our recruitment target of 336 participants despite unfavorable circumstances due to the pandemic. As a result, the study was well-powered, with the potential to show the effectiveness of the intervention.

The population studied was narrowly focused on community-dwelling older adults taking at least 1 DBI medicine and showing frailty. This RCT is, to the best of our knowledge, the first to look at a pharmacist-led deprescribing intervention in the community and to embed frailty as an aspect influencing deprescribing. To achieve this, we designed and included a cumulative deficit frailty measure applicable to multiple interRAI assessment instruments. Similar studies aim their deprescribing intervention at a broader, less specific target group. They, therefore, dilute the potential for effectiveness for people with specific conditions or circumstances.

Future Directions

Our study highlighted several areas for future investigations directly linked to our intervention method.

The DBI measures the drug burden across several medication classes, notably psychotropics (54), opioids, and a wide range of anticholinergics. Although these classes of drugs are known to be harmful, they are difficult to deprescribe. There is limited trial evidence in the literature, other than for psychosocial interventions (55), on the most effective processes for discontinuing these medications. Research developing and testing clear clinical pathways/guidance, patient-facing material, and evidence from deprescribing trials to inform these may be useful.

Overall, because the DBI measures the person’s total drug burden, a simple measure of reduction in DBI does not capture all changes in anticholinergic or sedative medications. For example, if a benzodiazepine is weaned and an antidepressant is started, DBI may stay the same, decrease, or increase depending on the relative doses.

In older adults, medical conditions and medications to treat them develop into more complex comorbidities and polypharmacy issues. Deprescribing may only be possible under certain narrow sets of circumstances and prognoses. Identifying and researching the conditions under which deprescribing is feasible and beneficial may improve the implementation of deprescribing protocols.

Furthermore, deprescribing within a health care system may benefit from focusing the available resources on people with a high need and a high potential for successful deprescribing. The use of routinely gathered administrative data, such as dispensing records or standardized interRAI assessment data, may allow algorithms to be developed that identify individuals who could benefit from and have a high potential for deprescribing.

As with the findings of our study, multiple previous deprescribing interventions have been unsuccessful. Translating possible interventions into clinical practice thus remains challenging. The reasons for such failings and difficulties are rarely fully understood and unlikely to result from a single shortcoming in any intervention. We consider that the complexity of deprescribing, with its enabling and inhibiting aspects, needs to be better understood if we are to develop efficacious and practicable interventions. Accordingly, a systematic review of the deprescribing process through the lenses of the different “players” and their respective contexts could aid the development of deprescribing interventions. Applying the lens of implementation science to this and previous studies may also answer broader systemic and contextual questions regarding deprescribing.

Conclusion

Our pharmacist-led medication assessment of frail older people living in the community was ineffective at reducing the anticholinergic/sedative load measured by DBI within 6 months. We found no evidence to suggest that this intervention resulted in substantial deprescribing within any frailty stratum. Consequently, we have no evidence whereby we can recommend a frailty measure that triggers a medication review for older people with DBI indices ≥0.5. Secondary outcome measures also showed no difference between the intervention and control groups. Management of the COVID-19 pandemic may have negatively affected the outcome of the study. A review of the study through the lens of implementation science may identify possible improvements to the study’s design and intervention, while a broader systemic review of deprescribing interventions may provide more holistic information for designing and implementing deprescribing processes.

Clinical Trials Registration Number: ACTRN12618000729224

Funding

This study was funded by the Health Research Council of New Zealand (grant 17/363).

Conflict of Interest

None declared.

Author Contribution

H.J. (PI) codesigned the study, obtained funding, and oversaw its conduct; P.S.N., the senior pharmacist in the team, peer-reviewed all medication reviews and Drug Burden Index (DBI) calculations and advised on applying the DBI in the study; H.U.B. codesigned the study and managed the trial, including ethics approvals, project reporting, database design, and data management; S.N.H. gave permission to use the DBI, advised on its use and protocol, and provided critical input to analysis and interpretation; N.J.A. contributed to gaining ethics approval, codesigned the study, trained pharmacists, and helped interpret the study findings; S.B.G. cocreated the project proposal and funding, led the qualitative assessment of the frailty measure; S.K.W. led the pharmacist team, training, and peer review of medication reviews and GP letters, and reviewed all electronic medication data of the study; R.A.-N. designed and validated the frailty measure; J.W.P. codesigned the study, conducted the statistical analyses, oversaw the conduct of the study when H.J. was on extended leave; D.M. supported the study as senior clinical reviewer and advisor. All authors contributed to the writing of this manuscript.

Data-Sharing Statement

Study materials are provided as Supplementary Materials. New Zealand’s ethics laws do not permit free sharing of data. However, aggregate and deidentified data may be provided to collaborating research groups under an appropriate data-sharing agreement.