FormalPara Key Points

Implementing a multifaceted intervention that employs the Drug Burden Index as a risk assessment tool, when coupled with a stewardship program, may help hospital clinicians deprescribe inappropriate medications with sedative and/or anticholinergic effects.

Amongst sedative and anticholinergic medications, opioids were the most likely to be deprescribed as a result of the proposed approach.

Further studies are needed to establish more robust evidence regarding the generalisability of this care model across different populations and its effectiveness in deprescribing inappropriate medications in hospitals, with the goal of improving prescribing and clinical outcomes in older adults.

1 Background

Medications with sedative and anticholinergic properties are commonly used in older adults in the community [1], nursing homes [2] and in hospitals [3]. However, these medications have been associated with increased risk of medication-related harm including adverse drug reactions, falls, hospitalisation, and mortality [4,5,6]. Internationally, approximately 84% of older patients and 80% of caregivers agree that they would be willing to stop one or more of their or their care recipient’s medications if the prescriber said it was possible [7]. Therefore, deprescribing, which is a process of withdrawal of an inappropriate medication (one in which the risks outweigh the benefits in the individual including high risk and unnecessary medications), supported by a health care professional with the goal of improving outcomes, is important in older adults [8, 9].

The Drug Burden Index (DBI) is a clinical risk assessment tool that measures a patient’s total exposure to medications with anticholinergic and sedative properties[10]. International studies have shown that higher DBI scores are associated with poorer physical function, falls, frailty and death [11,12,13,14,15,16]. Although previous studies have shown that the DBI can identify patients at high risk of adverse drug events who may benefit from deprescribing, and can guide recommendations for deprescribing, the DBI score is challenging for clinicians to calculate at the point of care. [17] This has limited the use of the DBI in routine clinical practice.

Admission to hospital may be an opportunity to reduce the risk of medication-related harm through deprescribing. Previous research has shown that a computerised decision support system (CDSS) can be effective in optimising prescribing of medications and medication safety in older inpatients [18]. Therefore, a DBI score calculator was integrated into the electronic Medical Records (eMR) in an Australian acute hospital and formed the basis of a CDSS to facilitate deprescribing of anticholinergic and sedative medications (i.e. DBI-contributing medications) [19,20,21]. Given that deprescribing is a complex intervention, a multifaceted intervention bundle including deprescribing guides, consumer information leaflets and education materials on polypharmacy in older inpatients may help clinicians to facilitate implementation of a CDSS in clinical practice. We hypothesised that the DBI calculator integrated into the eMR with accompanying resources and a stewardship program could help identify and act on opportunities for deprescribing in older inpatients [22, 23]. The objectives of this study were to investigate the impact of the comprehensive intervention bundle using the DBI on (i) the proportion of older inpatients with at least one DBI-contributing medication stopped or dose reduced on discharge compared with hospital admission and (ii) deprescribing of different DBI-contributing medication classes during hospitalisation.

2 Methods

This study was reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) 2010 statement: extension to randomised pilot and feasibility trials [24].

2.1 Study Design, Setting, and Participants

This study was a non-randomised, controlled, before-and-after study conducted in a metropolitan tertiary referral public hospital in Sydney, Australia. The inclusion criteria were patients admitted to the aged care (acute geriatric medicine) service of the hospital for ≥ 48 h from December 3, 2020 to October 31, 2021 who were aged ≥ 75 years, and had a DBI > 0 at admission. Multiple admissions for the same patient were all included if they met the inclusion criteria. The number of patients with multiple admissions was evaluated. The nature of the before-and-after study design carries a risk of contamination bias. For example, if a patient admitted during the control period, but still in hospital during the intervention period, received care through the intervention bundle, the effect of the intervention could be underestimated (i.e. dilution) [18]. Therefore, patients whose hospitalisation crossed over the different evaluation periods (i.e. control, intervention and stewardship periods) were excluded. Sensitivity analysis that included these patients as part of the cohort was conducted to evaluate the potential influence of contamination bias.

2.2 Interventions

The study consisted of three sequential periods (111 days each): control, intervention and stewardship periods. During the control period, patients received usual care that included medication reconciliation conducted by pharmacists without the intervention bundle. The automated DBI calculator that was incorporated into the local version of the state-wide eMR (Cerner PowerChart EMR) was not visible and was unavailable to clinicians.

During the intervention period, the following intervention bundle was provided: (1) the DBI clinician interface in the eMR, (2) deprescribing guides and consumer information leaflets, (3) DBI score integrated with the pharmacist patient list, and (4) an education module on polypharmacy in older inpatients.

  1. 1.

    The DBI clinician interface in the eMR

The DBI clinician interface integrated into the routinely used eMR displayed the patient’s total DBI score, a breakdown of the medications contributing to the score, and the contribution of each (Appendix 1) [20]. The interface was not visible to clinicians during the control period. The DBI score was calculated every time a medication was prescribed during hospitalisation. Medications prescribed as needed were excluded from the DBI calculation, preventing artificial inflation of the DBI score. The DBI score was visible on the opening page of the eMR and clinicians could click to access a separate DBI information page for each patient. The DBI clinician interface also provided a graphical display of the DBI score during the hospitalisation period, allowing clinicians to see when the DBI score increased and decreased. On the basis of commonly used published cut-off values for the DBI, the graph is colour-coded as per different risk categories, with DBI = 0 coloured green indicating low risk, 0 < DBI < 1 coloured yellow for moderate risk, and DBI ≥ 1 coloured red indicating high risk [25].

  1. 2.

    Deprescribing guides and consumer information leaflets

The DBI clinician interface has links to evidence-based deprescribing guides to support clinicians in safely reducing or stopping medications [26]. A link to printable consumer information leaflets was also included in the DBI clinician interface to assist in shared decision-making on deprescribing among patients, carers and clinicians [27]. The clinicians were encouraged to provide the patient/carer leaflets to their patients if they thought that they were indicated. The deprescribing guide and the consumer information leaflets are accessible through the link (https://www.nswtag.org.au/deprescribing-tools/).

  1. 3.

    Pharmacy patient list

The pharmacy patient list, which is routinely used by ward-based clinical pharmacists and contains key clinical information and results to guide pharmacy review, displayed a list of inpatients with their total DBI scores that could be sorted according to the DBI scores. This allowed pharmacists to prioritise patients for medication review on the basis of their DBI scores (Appendix 2) [20].

  1. 4.

    Education module on polypharmacy in older inpatients

Hospital staff involved in this study were encouraged by the research team to complete an education module on polypharmacy in older inpatients provided by the NSW Health Education and Training Institute to raise awareness of the importance of reviewing polypharmacy in older inpatients, roles of the interdisciplinary team and how to access resources for deprescribing [28].

In the stewardship period, a stewardship pharmacist (NM) with 9 years of geriatric clinical pharmacy experience across different Australian hospitals provided individualised patient-level advice to facilitate deprescribing of DBI-contributing medications and supported clinician use of the intervention bundle. For the purposes of this study, deprescribing included cessation and/or dose reduction of DBI-contributing medications during hospital admission. The stewardship process involved the following steps: (1) identifying patients with a DBI > 0 using the pharmacy patient list, (2) undertaking a medication review for eligible patients using patient’s medical conditions, medications prior to hospitalisation and medications in hospital, any drug allergies or intolerances, reason for visit, progress prior to stewardship review and any investigations, (3) creating a list of potential deprescribing opportunities for eligible patients, and (4) discussing deprescribing opportunities with the medical registrar, such as recommending deprescribing opportunities, asking for reasons for rejection of any recommendations, and promoting the use of deprescribing guides and consumer information leaflets to aid deprescribing. Detailed description of the stewardship program and the uptake of deprescribing recommendations were reported previously [23]. Detailed example of how the stewardship process was applied to a hypothetical patient is shown in Appendix 3.

2.3 Outcomes

The primary outcome was the proportion of patients with at least one DBI-contributing medication stopped or dose reduced on discharge compared with admission. The secondary outcome was the change in deprescribing of different DBI-contributing medication classes during hospitalisation. To evaluate DBI-contributing medication changes during hospitalisation, medications prescribed at admission and discharge were classified using the Anatomical Therapeutic Chemical (ATC) classification of the World Health Organisation [29]. Specifically, after identifying the most appropriate fifth-level ATC code for each medication, its third-level ATC code (pharmacological or therapeutic subgroup) was identified. The classified DBI-contributing medications used at admission were then categorised into four prescribing patterns: stop, dose decreased, no change and dose increase. Newly started medications during hospitalisation were categorised as ‘start’.

2.4 Statistical Analysis

Patients’ baseline characteristics were summarised using descriptive statistics such as median and interquartile range (IQR). To examine the differences in patients’ characteristics between the evaluation periods, either the chi-squared test or the Fisher’s exact test was applied for categorical variables, and the Kruskal–Wallis test was performed for continuous data. For the analysis of the primary outcome, G-computation procedure was applied to estimate adjusted risk differences (i.e. average treatment effects) [30]. Specifically, the binary outcomes were firstly modelled as binomial distribution with logit link functions against a categorical variable describing the evaluation period (i.e. control, intervention and stewardship) and potential confounders (i.e. age, gender, length of stay, the number of DBI-contributing medications at admission). Second, using the estimated parameters from the logistic regression model, counterfactual outcomes were estimated for each patient under each evaluation period. Lastly, the average treatment effect for each evaluation period across all participants was calculated and used to estimate adjusted risk differences (aRD). The 95% confidence intervals (CI) for the point estimate were estimated using 1000 times bootstrap resampling [31]. The normality of the distribution in the bootstrap results was confirmed by checking a quantile–quantile plot and histogram of all parameter estimates. Using the estimated aRD, the number needed to treat (= 1/aRD, NNT) was also calculated.

As sensitivity analyses, we conducted the following three analyses to evaluate the impact of including and excluding patients whose hospitalisation crossed over the transition periods: (1) analysis that included patients in the transition periods as separate evaluation periods (i.e. transition period from the control to the intervention period, and transition period from the intervention period to the stewardship period), (2) analysis that included patients in the transition periods into the precedent period, and (3) analysis that included patients in the transition periods into the antecedent period.

For the analysis of the secondary outcome, the chi-squared test or the Fisher’s exact test was conducted to evaluate the difference of proportion of DBI-contributing medications deprescribed between the three evaluation periods. Bonferroni correction was applied for multiple comparisons by multiplying the raw p-values by three (i.e. the number of comparisons). As this was a pilot study in preparation for a multicentre randomised controlled trial, no sample size calculation was performed prior to the study. Data manipulation and statistical analyses were performed using Python version 3.9 (Python Software Foundation) and R version 4.2 (R Foundation). [30] For the g-computation procedure, the gComp function in the riskCommunicator package was used [32]. Two-sided p-values < 0.05 were considered statistically significant.

3 Results

A total of 1548 admissions for 1263 patients were initially considered for possible inclusion in the study (Fig. 1). Of 1548 admissions, 446 (28.8%) were excluded due to admission date and duration of hospitalisation. Of the 446 admissions, 82 admissions were excluded because their admissions crossed evaluation periods as shown in Fig. 1. The number of admissions in patients with DBI > 0 at admission whose hospitalisation crossed over the different evaluation periods was 22 between the control and the intervention periods and 18 between the intervention and the stewardship periods (Appendix 4). Out of the remaining 1102 admissions, 457 admissions (41.4%) were patients with DBI > 0 at admission: 45.2% (144/318) in the control period, 42.2% (176/417) in the intervention period and 37.3 (137/367) in the stewardship period. Of 409 patients with 457 hospitalisations, 368 (90.0%) patients were admitted only once. The patients’ baseline characteristics, including age, sex, interpreter required, length of stay and the number of medicines at admission, were all comparable among the three evaluation periods. The median age was 89.0 (IQR 86–92) years, with 63.5% (290/457) being female. Among the 457 admissions, the median duration of hospital stay was 6.3 days (IQR 4.0–10.8) and the median number of DBI-contributing medications at admission was 1 (IQR 1–2). Among the DBI-contributing therapeutic drug classes on admission, opioids were more likely to be prescribed in the stewardship period compared with the intervention period whilst dopaminergic agents were more likely to be prescribed to patients admitted during the intervention period compared with the control period (p < 0.05) (Table 1).

Fig. 1
figure 1

Flow diagram of study cohort

Full size image
Table 1 Patients’ baseline characteristics (n = admissions)
Full size table

Figure 2 shows the five most commonly used DBI therapeutic drug classes at admission during the overall study period: antidepressants (23.6%), opioids (25.4%), antiepileptics (24.5%), dopaminergic agents (17.7%) and antipsychotics (13.1%). Combining data from all study periods, most of these DBI-contributing medications used at admission continued to be used without changing their dosage on discharge. However, the prevalence of deprescribing (i.e. dose reduction or cessation) DBI-contributing medications was different depending on the drug classes, ranging from 18.8% for dopaminergic agents (21/112) to 55.0% for antipsychotics (33/60). In addition, although 33.6% (39/116) of opioids taken at admission were deprescribed, a similar number of opioids (34) were newly prescribed by the time of discharge.

Fig. 2
figure 2

The five most commonly used Drug Burden Index contributing drug classes at admission and their prescribing patterns. Data is extracted from all 457 admissions studied without separating into study periods and describes overall patterns of drug use on and during admission

Full size image

The proportion of patients who had at least one DBI-contributing medication stopped or dose reduced on discharge increased from 29.9% (43/144) in the control period to 37.5% (66/176) in the intervention period and 43.1% (59/137) in the stewardship period. Using the control period as the reference, the impact of the intervention bundle with stewardship program was statistically significant (aRD 12.1%, 95% CI 1.0–24.0%) while that of the intervention bundle only was not (aRD 6.5%, 95% CI −3.2–17.5%). For every nine patients admitted during the period when the intervention bundle plus the stewardship program were available to clinicians, one additional patient had one or more DBI-contributing medications stopped or dose reduced on discharge compared with usual care before the intervention was introduced (NNT, 9). Length of stay was positively associated with the odds of DBI reduction (adjusted odds ratio 1.06, 95% CI 1.03–1.09), while age, sex and the number of DBI-contributing medications at admission were not. None of the sensitivity analyses significantly altered the results (Appendices 4–6).

Figure 3 shows the proportion of prescriptions of DBI-contributing therapeutic drug classes present on admission that were deprescribed in hospital. In the stewardship period, 45.7% (21/46) of opioids were stopped or dose reduced, while this only occurred for 17.9% of opioids in the control period (p = 0.04). There was also a trend towards greater deprescribing of antipsychotics and antiepileptics during the intervention and stewardship periods, but these were not statistically significant in this study. The proportion of antipsychotics deprescribed was more than 50% in all three evaluation periods (Table 2).

Fig. 3
figure 3

Proportion of Drug-Burden-Index-contributing drug classes deprescribed during admission. p-Value: *< 0.05. Bonferroni correction was applied for multiple comparisons

Full size image
Table 2 Proportion of patients with at least one anticholinergic and/or sedative medication stopped or dose reduced on discharge
Full size table

4 Discussion

This study found that integrating the comprehensive intervention bundle with a stewardship pharmacist was accompanied by increased deprescribing of sedative and anticholinergic medications from 31% to 43% (aRD 12.1%) of patients, with an estimated NNT of 9. The proportion of opioids that were stopped or dose reduced during admission significantly increased statistically from 17.9% in the control period to 45.7% in the stewardship period. The effect of the intervention bundle with a pharmacist steward was statistically significant whilst that of the intervention bundle alone was not adequate. This indicates the importance of the role of stewardship pharmacists in facilitating deprescribing of inappropriate medications with sedative and anticholinergic effects. This result is in line with previous studies reporting that providing actionable recommendations to clinicians and asking clinicians the reason for rejecting the recommendations were two success factors of CDSS [33].

The prevalence of older patients with DBI > 0 at admission in this study (41%) was consistent with previous studies conducted in Australia (33–53% [34,35,36,37]), Finland (37% [25], 38% [38]), the UK, (48% [39]), the USA (34% [40]), and New Zealand (32% [41], 43% [42]). The impact of the intervention bundle with stewardship program was different for different medication classes. Among the DBI-contributing medication classes, opioids were more likely to be deprescribed in the stewardship period compared with the control period. However, opioids were also reported to be the most common drug class where a stewardship pharmacist’s recommendations were rejected by clinicians, usually because of a current indication that was not clear to the steward from review of the medical record [23]. In Australia, more than 1.9 million adults initiate opioid therapies every year [43], with the majority of prescriptions being issued for maintenance treatment for non-cancer chronic pain [44]. Given opioid deprescribing is perceived as a complex and challenging practice at the medication, patient, prescriber and health system levels [45], close interdisciplinary collaboration among medical teams, pharmacists and patients is required to reduce suboptimal opioid use, which may have been achieved through the stewardship model.

Although the use of CDSS is expected to optimise prescribing of medications and medication safety in older adults [18], the comparability of research findings between different CDSS is relatively low because of the differences in the design, features, user interface, data entry source and degree of integration into workflow in daily practice, which are rarely reported in detail [46, 47]. A systematic review on the effectiveness of CDSS interventions to reduce potentially inappropriate medications (PIMs) for older adults reported that CDSS facilitated such deprescribing, although these differences were not always statistically significant [48]. Despite the potential efficacy of CDSS in reducing PIMs, there is little evidence of the impact of this on patients’ clinical outcomes (e.g. adverse drug events, ADE) [49]. One of the possible explanations is variance of ADE risks among medications classified as PIMs in the various studies. Relatively low-risk PIMs (e.g. stool softeners) have been one of the drug classes most deprescribed in CDSS interventions, but are less likely to cause ADEs [50]. Given that our CDSS focused on sedative and anticholinergic medications on the basis of the validated clinical risk assessment tool (DBI), deprescribing of these high-risk medications may lead to improved patient clinical outcomes. Further studies are required to investigate this.

This study had several limitations. Firstly, we only included patients aged 75 years or older and so the results may not be generalisable to younger patients. Secondly, this study applied a before-and-after study design as a pilot study with limited datasets, the results of which can be confounded by unmeasured factors (e.g., reason for admission). This pilot study design aimed to test the feasibility and effect size of the intervention to inform subsequent implementation and clinical trials. To mitigate these potential biases, several patients’ baseline characteristics were adjusted, and average treatment effect was estimated. Thirdly, this study did not assess long-term sustainability of this model of care. While a CDSS with an accompanying stewardship program is more likely to be successful than simply providing an intervention bundle, it is more resource intensive, and cost-effectiveness was not measured. Lastly, as this study was conducted within a single service, this study lacked generalisability in different populations from a variety of services across this and other hospitals. On the basis of this study’s findings, a stepped wedge randomised controlled trial of the intervention involving six hospitals has commenced in New South Wales, Australia, to evaluate the impact on deprescribing and clinical outcomes.

5 Conclusions and Implications

This study found that integrating the comprehensive intervention bundle with an accompanying stewardship program is a promising strategy to facilitate the deprescribing of sedative and anticholinergic medications for older adults in acute hospitals. Opioids were more likely to be deprescribed in the stewardship period compared with the control period. Further studies are needed to build more robust evidence for the effectiveness of this model of care for the purpose of deprescribing inappropriate medications in hospital to improve prescribing and clinical outcomes in older adults.