Evaluation of the Effect of an Intervention on Potentially Inappropriate Medications (PIMS), Polypharmacy and Anticholinergic Burden Scores for People with Dementia; Results from the SMS Dementia Study†: A Quasi-Experimental Study

Background People with dementia (PWD) are at risk for medication related harm due to their impaired cognition and frequently being prescribed many medications. Few previous studies of PWD inpatients have been focused on medication safety interventions. This study aimed to evaluate an intervention designed to improve medication safety for people with dementia (PWD) and their carers during an unplanned admission to hospital. This article reports the effect of the intervention on potentially inappropriate medications (PIMs), polypharmacy and anticholinergic burden scores for PWD in the study. Methods A quasi-experimental pre-post design using an intervention site and a control site was conducted in 2017-2019, in a regional area in New South Wales, Australia. PIMs, polypharmacy and anticholinergic burden were measured at admission, discharge and three months after discharge. In addition, medication reconciliation at admission and scoring of pharmacists recommendations using severity and relevance scores were measured. Results There were 628 participants including 350 in the post-intervention phase. Polypharmacy for these admissions was high, and there was approximately 30% reduction in the number of medications at discharge. PIMs at admission were also high, and decreased signicantly at discharge however there was no treatment effect associated with the intervention. The mean anticholinergic burden score also decreased signicantly between admission and discharge, however, no treatment effect was seen. Conclusions High rates of polypharmacy and PIMs in this study indicate this study population was admitted with multiple comorbidities. Reduced PIMs at discharge were correlated with reduced anticholinergic burden. Medication reconciliation resulted in many recommendations that contributed to the reductions in medications. Although the study did not report a treatment effect, reductions in the number of medications and PIMs reduced medication related risk for PWD. for for Medication polypharmacy due to identifying potential drug interactions, side effects, dosage modications, and deprescribing. This study highlights need for focused medication management in this high risk population. The results indicate that the intervention feasible to implement and is helpful to informing future multisite studies that would have sucient power to demonstrate treatment effect.


Background
People with dementia (PWD) have a high risk for adverse health outcomes associated with medications due to being cognitively impaired [1]. This manifests as missing medications due to confusion or memory problems, or taking incorrect medications or dosages. The risk increases for PWD who don't have a carer, have multiple comorbidities and consequently have more than ve medications prescribed for them (polypharmacy) [2,3], and who are prescribed potentially inappropriate medications (PIMs) [4]. PIMs are de ned as "medications that pose potential risks that outweigh potential bene ts" [5]. PIMs pose a risk for PWD because they may have side effects that exacerbate confusion and balance problems (resulting in falls) and increase anticholinergic burden. When PWD are admitted to hospital, they may also suffer escalated behavioural and psychological symptoms of dementia (BPSD) and may be temporarily managed with PIMs such as psychotropics, sedatives or hypnotics.
Antipsychotics have been identi ed as being overused and having limited clinical bene t for BPSD [6,7] and should not be continued for more than three months [8]. Consequently, the risk for PWD remains high if these medications are not discontinued at discharge [9]. The prevalence of PIMs for PWD in the community has been reported in a systematic review, to range from 10-56% and is higher in nursing home settings [5]. However, the prevalence of PIMs in hospital has been reported in another systematic review to be 53-90% for inpatients with cognitive impairment [10]. These reviews provide evidence to con rm that PIMs are an important clinical issue that requires further attention. Two previous studies have compared PIMs between admission and discharge and reported that PIMs for PWD were signi cantly reduced at discharge (Mean 4.0 reduced to 3.3, difference 0.7, p < 0.0001) in a study of 277 admissions [1], and (Mean reduced from 0.8 to 0.4, difference 0.4, p = 0.01) in a study of 118 admissions [11]. When PWD are admitted to hospital, there is an opportunity to undertake medication reconciliation and to identify PIMs and other medications that may no longer be required. Where it is possible to reduce medications, the risk for PWD is also reduced. Some previous intervention studies have been conducted to reduce PIMs in community/primary care settings [12][13][14] and nursing homes, for older people [15]; however no studies have been conducted in hospital settings or on PWD for this purpose. Previous studies of older people in non-hospital settings used interventions such as education interventions, medication reviews and collaborative care approaches. Signi cant reductions in PIMs were reported from using medication reviews and an educational intervention. In addition, a prospective observational study of 991 pharmacist interventions for 557 patients during medication reconciliation in an emergency department, reported that medication errors were severe in 57% of cases, and that 65% of the interventions were relevant [16]. This suggests that pharmacist medication reconciliation can be used to reduce PIMs, and polypharmacy, as well as to reduce medication errors. An intervention study involving hospital pharmacists completing the medication management plan in the medical discharge summary also reported a signi cant reduction in the rate of medication errors at discharge [17]. Furthermore, a study that evaluated a collaborative care approach involving clinical pharmacist medication review, in which the clinical pharmacist was based in the community health centre with the general practitioner (GP); reported that 48% of recommendations were accepted by GPs in an elderly community population (n = 91) [13]. The results of these previous studies suggest that it is important to undertake research to evaluate the effectiveness of pharmacist interventions for PWD in the acute care setting to reduce PIMs (and associated anticholinergic burden), and polypharmacy and measure the frequency of medication prescribing errors for these vulnerable admitted patients.

Methods
This study aimed to evaluate an intervention designed to improve medication safety for PWD and their carers during an unplanned admission to hospital and the effect on the primary outcomes are reported separately. This article reports the effect of the intervention on PIMS, polypharmacy and anticholinergic burden scores for PWD in the study. In addition, we report the impact of the intervention on the frequency of medication reconciliation at admission and subsequent medication recommendations for PWD. A quasi-experimental pre-post design was used because participants could not be randomised at the study sites. The study was conducted at two regional hospitals in New South Wales, Australia, between October 2017 and September 2019. Usual care was delivered during phase one at both hospitals. In phase two, the intervention was delivered at the intervention hospital and the other hospital was used as a control site.
Participants were PWD or older people who had a positive screen for memory problems or confusion (excluding transient delirium), during an index admission via the emergency department (ED) during the study period. Proxy consent was provided by their carer or person responsible. Additional detail about participant eligibility is provided in a previous publication [1].
Data were collected using purpose-developed clinical audit instruments for admission (usually within 48 hours), and discharge (usually within 24 hours), and medications from medical records to measure PIMs, numbers of medications and to calculate the anticholinergic burden scores, using a modi ed anticholinergic burden score (mACB) (AUS) (See A1) at admission and discharge. Phone surveys of community pharmacists were conducted at three months after discharge to measure PIMs, numbers of medications, and to calculate the anticholinergic burden scores. In addition, severity and impact scores of prescribed medications were measured using the scoring system by Overhage and Lukes, as reported by Perez-Moreno et al [16] to evaluate the potential impact of the prescribing error and the effect on the patient's health. Errors were scored using ve categories of severity ranging from no error, to potentially lethal. Clinical relevance of the pharmacist recommendations (impact) were scored using six likely consequences in patient care, ranging from injurious to extremely signi cant.
General practitioner acceptance of pharmacist recommendations following home medication review at three months after discharge was also measured using a post discharge phone call by the study pharmacist.
The intervention comprised seven strategies delivered after admission and prior to discharge (see box).

Safe Medication Strategies (SMS) Intervention
After Admission -7 Contact GP about discharge medications including changes, and recommend to GP to arrange Home Medicines Review (HMR) or RMMR by Accredited Community Pharmacist Sample size calculations were based on the primary aim (treatment effect for readmissions/re-presentation to ED). The evaluations of medication use within the study reported in this paper are secondary outcomes; consequently, sample size calculations were not performed for these outcomes. Discharge and three month data collection was not performed if consent was withdrawn, or the participant died prior to that timepoint.
Descriptive statistics were summarised by phase and site using means (SD) and median (min, max) for continuous data, and counts and percentages for categorical data. The change in medication use and medication reconciliation was examined from admission to discharge, and from discharge to 3 months across sites and phases using mixed modelling (negative binomial, logistic, and linear mixed modelling as appropriate). Fixed effects included phase, site, time (categorical), all two-way interactions, and a 3-way interaction term (phase*site*time), and (given adequate response numbers) included PWD characteristics identi ed as being potentially unbalanced between the sites (age, gender, discharge destination). A random effect was included for participant to account for correlations within a person over time. The correlations between change in number of PIMs and mACB score (admission to discharge, and discharge to 3 months) were examined (averaged over site and phase) using Spearman correlation.
The treatment effect for mean count of medication recommendations (per participant) at admission was analysed using negative binomial regression. A zero-in ated negative binomial regression model was used to examine the treatment effect for "Signi cant (severity)" medication recommendations, medication recommendations not due to error, and "Relevant (impact)" medication recommendations; the odds of having at least one medication recommendation was modelled together with the average count of medication recommendations in participants who had at least one medication recommendation. Modelling included phase, site, and the interaction term (phase*site), and adjusted modelling included age, gender, and number of medications at admission.
The proportion of medication recommendations classed as "Signi cant (severity)", as not due to error, and as "Relevant (impact)" at admission was compared across phase and site using logistic mixed modelling; modelling included phase, site, and the interaction term (phase*site), and adjusted modelling included age, gender, and number of medications at admission. A random effect was included for participant to account for correlations within a person over multiple medication recommendations.
Study data were collected and managed using REDCap (Research Electronic Data Capture) tools [18] hosted at the Hunter Medical Research Institute, Australia. Data were analysed using SAS v9.4 (SAS Institute Inc. Cary, NC); a priori, p < 0.05 (two-tailed) was used to indicate statistical signi cance.

Results
The nal sample comprised 278 participants for the pre-intervention and 350 participants for the postintervention phase. Patient characteristics by site and phase are presented in Table 1. There were some differences in gender, age and indigenous admissions between study sites and phases. For admissions from home, only 66% were discharged to home (343/523). Discharges to residential aged care facilities (RACF) increased by 69% compared with admissions from RACF (105 vs 178), and 5.1% died during the index admission. Descriptive data for medications are presented in Table 2 by phase, site and timepoint. Overall, participants were prescribed four less medications (approximately 30%) at discharge and this was sustained at three months after discharge. Overall 95-98% of participants were prescribed PIMs across timepoints. All sites showed a signi cant decrease in the mean number of PIMs from admission to discharge. After adjusting for age, gender and discharge destination, there was no signi cant treatment effect for PIMs at admission compared to discharge (p = 0.366), or at discharge compared to three months (p = 0.391). See supplementary tables C and D.
The mean mACB score decreased for all phase/site combinations from admission to discharge, however, no treatment effect was seen (p = 0.086). The mean mACB score increased from discharge to 3 months at the control site in both phases, and did not change signi cantly at the intervention site. See supplementary tables E and F.
Averaged over site and phase, signi cant moderate positive correlations were seen between PIMs change and mACB change from admission to discharge (rho = 0.48 p < 0.001), and from discharge to three months (rho = 0.55 p < 0.001).
Psychotropic and Sedative/Hypnotic PIMs categories are shown in Table 3 by site and phase, and timepoint. There were no differences in the proportion of participants on at least one psychotropic medication between sites, phases and timepoints (admission to discharge p = 0.275, discharge to three months p = 0.915). See supplementary tables G and H.
From admission to discharge, there was a signi cant decrease in the proportion of participants on at least one sedative/hypnotic medication at the control site in both phases and the intervention site in phase two and this is a clinically signi cant improvement in prescribing, however, it could not be shown that this was due to the intervention (admission to discharge p = 0.233, discharge to three months p = 0.807). See supplementary tables I and J.
Pharmacist medication reconciliation conducted at admission and prior to discharge is shown in Table 4 by site, phase and timepoint. The increase in the proportion of participants receiving pharmacist medication reconciliation at admission was clinically signi cant in the Intervention group between phase one and phase two, while the control group remained stable (numbers too low to support regression modelling); similar was seen at discharge.
Pharmacists' recommendations for medications that were identi ed during medication reconciliation as having a potential for harm or adverse reaction or prescribing error, were evaluated for their severity and relevance (impact of the service provided by the pharmacist). Severity and impact scores for pharmacist's medication recommendations (per participant) are shown in table 5. The proportion of participants with at least 1 medication recommendation at admission increased signi cantly in the Intervention group (OR 78.9, p < 0.001) and did not change at the control site, (OR 1.12, p = 0.676), and the overall effect was signi cant (OR 70.5, p < 0.001). The mean count of medication recommendations per participant increased signi cantly in the Intervention group between phase 1 and phase 2 (IRR 13.6, p < 0.001), while the control group remained stable (IRR 0.98, p = 0.923). See supplementary tables K and L.
Medication recommendations per participant scored as "Signi cant (severity)" was modelled as a 2-part model; the increase (phase 2 compared to phase 1) in the proportion of participants having at least one "Signi cant (severity)" medication recommendation was signi cantly more at the intervention site than the control site (OR 20.5 p < 0.001). While the mean count of "Signi cant (severity)" medication recommendations increased signi cantly in the intervention site from Phase 1 to Phase 2 (IRR 1.9 p = 0.022), and decreased signi cantly in the control site (IRR 0.6 p = 0.006), the mean count at the intervention site in Phase two was still lower than the mean count at the control site in Phase one. See supplementary tables M and N.
In addition, there was an increase in proportion of participants with at least one medication recommendation that was "not due to error", at the intervention site (OR 104, p < 0.001) and overall (OR 65.  The proportion of medication recommendations that are "Signi cant (severity)" decreased signi cantly at the intervention site between phase 1 and phase 2 (OR 0.37 p = 0.006), although this decrease was not signi cantly different to the decrease seen at the control site (treatment effect OR 0.63 p = 0.328). See supplementary table R.
There was a signi cant increase in the proportion of medication recommendations that did not involve a prescribing error at both sites in phase two, compared to phase one (Intervention OR 3 p = 0.003, Control OR 2.4 p = 0.008), although the overall treatment effect was not signi cantly different (OR 1. A signi cant moderate positive correlation was seen between severity and impact of medication recommendations at admission (n = 1446, Rho = 0.58 p < 0.001).
The GP acceptance of community pharmacist's recommendations at three months after discharge was 68% (n = 104/156 recommendations).

Discussion
Polypharmacy was high overall (> 90%) and this re ects high comorbidity in this study population, and increasing comorbidity has been reported to be signi cantly associated with higher polypharmacy [19]. This result is higher than the rate reported in a study of older people including participants with cognitive impairment (n = 373) with a rate of 69% for polypharmacy [19] and a study that included 10,528 participants with dementia in primary care of 57% [3].
The number of medications at discharge reduced by 30% across study sites and phases and this was sustained at three months, suggesting that medication reconciliation and deprescribing is practiced to some extent in the delivery of usual care and may explain why the study was not able to report a treatment effect. Nonetheless reduced prescribing reduces medication related risk for PWD [1,19,20].
PIMs prescribing was very high (> 90%) overall in this study. Previous studies have reported lower rates of PIMs prescribing (of at least one PIM) for PWD in the community. A multi-country study (n = 2004) reported a rate of 60% [4] and a nationwide study (n = 2190) reported a rate of 67% [21], however a recent systematic review reported prevalence of PIMS ranged from 53-90% for inpatients with cognitive impairment [10]. The mean number of PIMs decreased signi cantly (by 25%) from admission to discharge and this also reduced medication related risk for PWD [1,4], however no treatment effect was identi ed.
The mean mACB score decreased signi cantly from admission to discharge and this suggests reduced risk for PWD, however no treatment effect was identi ed. The signi cant moderate positive correlations between PIMs reduction and mACB reduction also indicate reduced medication related risk for PWD in this study.
Prescribed psychotropic medications did not vary signi cantly from admission, however 44% of patients were prescribed these medications. A report from Alzheimer's Australia states that up to 20% of PWD who receive antipsychotic medications derive bene t from them [22], so there is potential for inappropriate prescribing for PWD in this study.
The proportion of participants on at least one sedative/hypnotic was signi cantly reduced at discharge at both sites (by 60%) -this is a clinically signi cant improvement in prescribing because it reduces risk for PWD, however no treatment effect was identi ed. This reduction may have been in uenced by a recent initiative by the Australian Commission on Safety and Quality in Health Care that published National Safety and Quality Health Service Standards and targeted inappropriate prescribing for BPSD (https://www.safetyandquality.gov.au/publications-and-resources/resource-library/reducing-inappropriate-useantipsychotics-people-behavioural-and-psychological-symptoms-dementia-bpsd-infographic).
Medication reconciliation at admission and discharge increased signi cantly in the intervention site in phase two in this study. Medication review has been reported to signi cantly improve the appropriateness of prescribing in aged care facilities [14,15] and primary health care [13].
There was a signi cant increase in the mean number of medication recommendations identi ed by pharmacists during medication reconciliation at the intervention site in phase two. There was a high proportion of participants in the intervention group who received at least one medication recommendation (93%) and this was higher than the proportions for usual care/phase 1 participant groups (16-35%). This result is higher than the rate (19%) reported in an observational study (n = 2984) of admissions to emergency department (ED) [16], and may indicate that PWD require more modi cations of their medications.
There was a signi cant increase in the proportion of participants having at least one "Signi cant (severity)" medication recommendation for the intervention group (72%) compared with the usual care groups (21%). Increased severity scores for the intervention group may have patient safety implications or may be the result of increased medication reconciliation at admission. The observational ED study reported a rate of 57% of signi cant severity recommendations [16].
The proportion of patients having at least one medication recommendation not due to error increased signi cantly at the intervention group (85%) compared with the usual care/phase 1 groups (5-20%). This suggests that pharmacists may have recommended modi cations in medications rather than agging potential drug interactions or adverse effects.
There was a signi cant increase in the proportion of participants having at least one relevant pharmacist medication recommendation in the intervention group (90%), indicating a clinically signi cant impact of the service provided by the pharmacist. The usual care/phase 1 rates ranged from 11-33% and these were lower than the rate reported in the previous ED study (65%) [16]. The signi cant moderate positive correlation between severity and impact scores in this study was similar to the correlation reported in the previous ED study (Rho = 0.73 p < 0.001) [16].
The signi cant severity scores (by number of medication recommendations) were signi cantly lower at the intervention site in phase two compared with the control site in phase one. However, there was variation between the sites in phase one, and between the phases at the control site, and this nding should be interpreted and extrapolated cautiously.
The proportion of medication recommendations that were not due to error (by number of medication recommendations) increased signi cantly at the intervention site between phase one and two, however no treatment effect was identi ed.
There was a signi cant increase in the relevance (impact) of medication recommendations (by number of medication recommendations) at the intervention site between phase one and two, however no treatment effect was identi ed.
GP acceptance of at least one HMR recommendation at three months after discharge was 68%, and this was higher than an observational study in primary care (n = 91) in which GP acceptance of 304/625 pharmacist recommendations (48%) was reported [13], and a study (n = 1021) reporting GP acceptance of ED pharmacist recommendations (49%) [23].
Previous studies have reported the effect of interventions on medication safety for older people in the community or primary care settings [12][13][14][15], however this study evaluated the effectiveness of a pharmacist intervention for PWD inpatients and the effect on polypharmacy and PIMS. Medication safety for PWD is particularly important because of the risks associated with medications for PWD. The Australian Health Ministers Advisory Council has established nine National Health Priority Areas, including dementia, and in 2019 they announced that quality use of medicines and medicine safety will be the 10th National Health Priority Area in Australia (https://vivacommunications.com.au/blog/medicines-safety-now-a-national-health-priority/ ). The addition of this priority area should emphasize the importance of this issue for future research and practice improvement initiatives.
Clinical Implications PWD or cognitive impairment are not always identi ed at admission. Consequently, clinicians may not recognise that this vulnerable group of patients needs particular attention regarding their medications. PWD often have polypharmacy and many associated medication safety concerns. Polypharmacy may be a consequence of their complex comorbidities. Pharmacist-led medication reconciliation is a valuable means of ensuring medication safety for PWD and can result in them having improved outcomes due to reductions in polypharmacy, PIMs and deprescribing.

Strengths and Limitations:
Few previous studies have been conducted exclusively on PWD in acute care and focused on their medication safety. PWD or cognitive impairment is not always identi ed at admission. Having an intervention that focuses on PWD in an inpatient setting requires robust systems for identi cation of PWD that are sensitive to their needs and values. This study undertook screening to identify PWD at admission, and evaluated the effect of a pharmacist-led intervention on polypharmacy and PIMs for PWD in acute care. The study design only used two sites and participants were not randomised, and this limits the internal validity and generalisability of the results. In addition, values of all regression ndings should be interpreted cautiously due to low numbers and only two study sites.

Conclusions
This study has identi ed that admission to hospital presents an opportunity to undertake medication reconciliation and minimise risk for medication related poor outcomes for PWD. Medication reconciliation can contribute to reducing polypharmacy and PIMs and result in recommendations for improved medication safety, due to identifying potential drug interactions, side effects, dosage modi cations, and deprescribing. This study highlights the need for focused medication management in this high risk population. The results indicate that the intervention is feasible to implement and is helpful to informing future multisite studies that would have su cient power to demonstrate treatment effect. Declarations CO was an investigator and contributed to funding acquisition, participated in the design of the study, software management and data analysis, and contributed to manuscript development and revisions DP was an investigator and contributed to funding acquisition, participated in the design of the study, and contributed to manuscript development and revisions.
AS was an investigator and contributed to funding acquisition, participated in the design of the study, and contributed to manuscript review and editing.
RL contributing to manuscript review and editing.
RB and WM contributed to data collection, delivery of the intervention, and reviewing and editing of the manuscript.
JA was an investigator and contributed to funding acquisition, participated in the design of the study, and contributed to manuscript review and editing.
AK was the lead investigator and responsible for funding acquisition, participated in the design and coordination of the study and design of study instruments, contributed to data collection and analysis, and drafted the manuscript.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.