Potentially inappropriate prescribing and its associations with health-related and system-related outcomes in hospitalised older adults: A systematic review and meta-analysis.

AIMS: To synthesise associations of potentially inappropriate prescribing (PIP) with health-related and system-related outcomes in inpatient hospital settings.
METHODS: Six electronic databases were searched: Medline Complete, EMBASE, CINAHL, PyscInfo, IPA and Cochrane library. Studies published between 1 January 1991 and 31 January 2021 investigating associations between PIP and health-related and system-related outcomes of older adults in hospital settings, were included. A random effects model was employed using the generic inverse variance method to pool risk estimates.
RESULTS: Overall, 63 studies were included. Pooled risk estimates did not show a significant association with all-cause mortality (adjusted odds ratio [AOR] 1.10, 95% confidence interval [CI] 0.90-1.36; adjusted hazard ratio 1.02, 83% CI 0.90-1.16), and hospital readmission (AOR 1.11, 95% CI 0.76-1.63; adjusted hazard ratio 1.02, 95% CI 0.89-1.18). PIP was associated with 91%, 60% and 26% increased odds of adverse drug event-related hospital admissions (AOR 1.91, 95% CI 1.21-3.01), functional decline (AOR 1.60, 95% CI 1.28-2.01), and adverse drug reactions and adverse drug events (AOR 1.26, 95% CI 1.11-1.43), respectively. PIP was associated with falls (2/2 studies). The impact of PIP on emergency department visits, length of stay, and health-related quality of life was inconclusive. Economic cost of PIP reported in 3 studies, comprised various cost estimation methods.
CONCLUSIONS: PIP was significantly associated with a range of health-related and system-related outcomes. It is important to optimise older adults' prescriptions to facilitate improved outcomes of care.

INTRODUCTION

The world's population is aging, with recent statistics showing that older people make up a considerable proportion of the world's population. In 2017, 1 in 8 people worldwide was aged 60 years or older and it is expected that this proportion will increase to 20% by 2050. This demographic transition has a number of implications to healthcare. Older adults are prone to multiple chronic conditions, necessitating the use of multiple medications, or polypharmacy. ,

Polypharmacy, commonly defined as the concurrent use of 5 or more regular medications, , is increasingly prevalent as the population ages. A recent population‐based study estimated the prevalence of polypharmacy among older Australians is high (36%), with the oldest old (aged 85 years or older), the most affected. The rate of polypharmacy is even higher in hospitalised patients (76%).

The use of polypharmacy may be clinically justifiable, but it is important to identify patients with inappropriate polypharmacy that may lead to adverse clinical outcomes. Older adults are particularly vulnerable to the negative impact of polypharmacy due to age‐related physiological changes that affect the pharmacokinetics and pharmacodynamics of medications, and their under‐representation in clinical trials, resulting in a lack of benefit/risk data. This vulnerability makes safe and effective prescribing a challenging and complex process in older adults, contributing to an increased risk of potentially inappropriate prescribing (PIP).

PIP involves prescribing medications that may not produce benefits relative to harm, or not prescribing medications that are recommended, which may pose significant harm to older adults. PIP encompasses potentially inappropriate medications (PIMs) and potential prescribing omissions (PPOs). PIMs are medications with a greater risk than benefit to a patient while PPOs are failures to prescribe medications of potential benefit. ,

Numerous tools are available in the literature to identify PIPs. These tools can be grouped into implicit (judgement‐based) and explicit (criterion‐based) tools, or a combination of both approaches. , Explicit tools can be easily applied with little or no clinical judgement, and the most studied explicit tool is the Beers list, which was first published in the USA in 1991 and last updated in 2019. Other explicit tools, the STOPP (Screening Tool of Older Person's Prescriptions) and START (Screening Tool to Alert doctors to Right Treatment) criteria, were developed in Europe in 2008 (later revised in 2014), , and have now become widely used tool in Europe and elsewhere. , , , The Beers and STOPP criteria address PIMs, whereas the START criteria enable detection of PPOs.

The link between polypharmacy and PIP is well established. , , , , As with polypharmacy, PIP is common in older adults and is associated with an increased use of healthcare resources and medication costs. , Previous systematic reviews have identified some links between PIPs and adverse drug events (ADEs) and hospitalisation, but are inconclusive on other outcomes such as mortality, emergency department (ED) visits and medication‐related hospital readmissions. , , , , These reviews have predominantly focused on studies using a limited number of tools, such as the Beers and STOPP criteria. It has not yet been established whether failure to prescribe medications of potential benefit, comprising PPOs, has clinical and resource implications. Also, this evidence has most often originated from either population‐based studies or analyses involving long‐term care residents, with limited data available from populations of older hospitalised patients. Moreover, the full range of outcomes associated with PIPs is not well established, especially in hospital settings. It is unclear whether prescribing of PIPs during inpatient care is associated with health‐related outcomes, such as ADEs and quality of life or with system‐related outcomes, such as mortality and hospital readmission. Thus, the aim of this systematic review was to synthesise the available literature on the associations of PIP in the inpatient hospital setting, identified through any validated and published tool, with health‐related and system‐related outcomes.

METHODS

This systematic review was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) statement, and the study protocol was registered on PROSPERO (CRD42020182598).

Data sources and search strategy

A comprehensive electronic search of the medication safety literature was undertaken using the following databases: Medline Complete (1916); EMBASE (1947); Cumulative Index to Nursing and Allied Health Literature (CINAHL) Complete (1937); PyscInfo (1806); Cochrane Central Register of Controlled Trials (1996); and International Pharmaceutical Abstracts (IPA; 1970) databases. The searches were limited to English language papers published between 1 January 1991 and 31 January 2021; the start date coincided with the first validated list of PIMs published in 1991. The search terms included synonyms related to inappropriate prescribing, older populations, hospital care, and health‐related and system‐related outcomes. These keywords were hand‐picked from the literature during preliminary literature searching. The key concepts were searched line by line and then combined using Boolean operators (OR, AND) to identify eligible studies. Keywords were customised to database‐specific Medical Subject Headings (MeSH) and indexing terms to capture relevant studies. In addition to language and year of publication, the database searches were also limited to studies with abstracts, and conducted on humans (Appendix 1). The university research librarian provided advice about setting up and conducting the search strategies for the various library databases.

In addition to electronic database searches, reference lists of relevant reviews and included articles were examined manually to identify any additional eligible studies. Search results were then imported into an EndNote library to manage article collections and remove any duplicate studies. The de‐duplicated search results were transferred to Covidence for independent blind screening of relevant papers.

Eligibility criteria and study selection

For inclusion in this review, older adults aged 65 years (60 years for low‐and middle‐income countries ) or older, who were admitted to hospital for inpatient services, irrespective of the types of admissions and ward specialities, were considered. Studies that involved multiple healthcare settings were required to clearly report separate data for each hospital setting. All observational cohort studies, cross‐sectional studies and case–control studies investigating the association between PIPs and health‐related outcomes were included. To be included, studies were required to employ validated criteria to identify PIPs, such as the Beers, STOPP and START criteria. Studies that employed modified versions of validated tools, and country‐specific tools were also considered. However, studies must have employed the tools in their entirety, and not been limited to specific medications or disease conditions. , ,

The primary outcomes of interest were health‐related, such as rates of adverse drug reactions (ADRs) and system‐related (e.g. all‐cause mortality, ED visits, hospital readmissions, length of stay). These outcomes could be measured across any period—before, during or after hospital discharge. However, studies that only measured PIP as an outcome (e.g. the impact of hospitalisation on the incidence of PIP) were not included. Secondary outcomes included health‐related quality of life, falls, functional decline, and cost‐related to PIPs. Similarly, these secondary outcomes could be measured any time, and data were extracted on these outcomes without any preset definitions, and hence, we adopted the definitions employed by each study.

Review articles, qualitative studies, conference abstracts without full‐text publications, case reports, editorials and commentaries were excluded. Studies that did not address outcomes of inappropriate medication use, including those exploring the prevalence of PIP per se, and risk factors for PIPs were also excluded.

Studies retrieved from all the databases and those located from the additional sources were screened independently by 2 reviewers (A.M., B.R./E.M.) for inclusion. Any discrepancies at the title and abstract level were resolved by a third reviewer (B.R./E.M.). Pilot testing on an initial sample of 15 studies demonstrated only moderate agreement between 2 independent reviewers (A.M., B.R.) in title and abstract screening (Cohen's κ = 0.47; % agreement = 73%). Further discussion resulted in additional detail in the eligibility criteria to improve agreement between reviewers. Studies deemed eligible after title and abstract screening passed into full text review. The full texts of potentially eligible studies were retrieved and assessed independently by 2 reviewers (A.M., B.R./E.M.) against the inclusion criteria, and ineligible papers were discarded. Any discrepancies at the full text level were again resolved by a third reviewer (B.R./E.M.).

Data extraction

A standardised, prepiloted document was employed for data extraction and quality assessment of the included studies. Items in the data extraction tool included general study characteristics (e.g. study authors, country of origin, study design, characteristics of the population), tools used to identify PIPs, medications associated with PIPs, and main results on health‐related outcomes due to PIP.

Quality assessment

As we proposed to include diverse study designs, we employed the Mixed Methods Appraisal Tool (MMAT v2018) for assessment of study quality. The MMAT was adopted for quality assessment of quantitative nonrandomised studies, which includes cohort, case–control and cross‐sectional analytic studies. In line with our study objectives, we set out a priori to consider only the control arm of interventional studies for quality assessment, using the same methodological criteria as the quantitative nonrandomised study designs.

Data analysis

Descriptive analysis was conducted on extracted data from all included studies. A meta‐analysis was conducted if 2 or more studies reported data suitable for quantitative synthesis. Health‐related or system‐related outcomes were pooled as an odds ratio (OR) or hazard ratio (HR) together with a 95% confidence interval (95% CI) using a random‐effects model with the generic inverse variance method. Meta‐analysis was performed for both crude and adjusted risk estimates. For studies that contributed 2 or more risk estimates for the same outcome, sensitivity analysis was conducted by selecting only the weakest association. We also conducted subgroup analyses based on various factors, such as the tool used to identify PIPs, study design and location, and quality score. All meta‐analyses were performed with Review Manager (RevMan) software (RevMan V.5.3, The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark). Pooled prevalence estimates were carried out using OpenMeta[Analyst] (http://www.cebm.brown.edu/openmeta/).

RESULTS

The database searches yielded 1821 results. After removal of duplicates, titles and abstracts of 1282 unique articles were independently screened, with 1116 excluded. The full texts of the remaining 166 studies were reviewed in detail using inclusion and exclusion criteria. Of these, 103 articles were excluded, mainly because studies reported a different outcome of interest (n = 58). The final screening identified 63 studies , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , suitable for inclusion in this review (Figure 1).

FIGURE 1

Flow diagram of the selection process

Characteristics of included studies

The included studies were conducted in 21 different countries (Table 1): 32 (52%) studies were performed in Europe, 13 (22%) in North America, 11 (14%) in Asia, 4 (7%) in Australia and 3 (5%) in Brazil, with publication dates between 2005 and 2020. Forty‐seven studies were cohort studies (25 were conducted prospectively), and 11 were cross‐sectional studies. The remaining studies were case–control or comparative retrospective (3 studies), mixed‐methods (involving a retrospective clinical audit) and a secondary analysis of a randomised controlled trial (each 1 study). Most studies (n = 44) were confined to single centres, mainly in geriatric or medical hospital wards. Sample sizes for included studies ranged from 52 to 45 809 individuals. The reported mean and median ages of participants in included studies ranged from 72.4–88.3 and 71–88 years, respectively. The average percentage of male participants among the included studies was 45%.

TABLE 1

Characteristics of included studies (n = 63)

Authors, year	Country	Study design	Study setting, specialty	Sample size	% male	Age (y), mean (SD)	Study period, follow–up duration (mo)	PIP tool	Data source of PIP	Outcomes assessed	PIP prevalence, %
Akkawi et al. 2019 33	Malaysia	C	Single centre, medical & surgical wards	502	51.4	72.4 (5.9)	Apr–Oct 2016 & Apr–Oct 2017, NR	STOPP/START v2	Medical record	HRQoL	STOPP v2: 28.5; START v2: 45.6; STOPP/START v2: 59.2
Bachmann et al. 2018 34	Switzerland	PC	Single centre, geriatric inpatient rehabilitation	210	53.8	75.5	Feb–Nov 2014, 0.75	STOPP 2008	Referral letter	HRQoL, mobility	43.3
Basnet et al. 2018 35	USA	RC	Multicentre, medical & surgical units	24 204	45	78 (9)	Sep 2011–Dec 2013, 1	Beers 2012	Electronic medical record	HR	58.9
Bo et al. 2018 36	Italy	PC	Multicentre, internal medicine & geriatric wards	1000	45.5	81.9 (7.7)	Dec 2015–Jun 2016, 6	Beers 2015	Medical record, patient interview	M, HR	63
Brunetti et al. 2019 37	Italy	PC	Multicentre, geriatric & internal medicine wards	611	51.6	81.6 (7.0)	Mar–June 2017, 6	STOPP/START v2	NS	M, HR	STOPP v2:54.8; START v2: 47.3; STOPP/START v2: 71.7
Cabré et al. 2018 38	Spain	C	Single centre, acute geriatric unit	3292	39.9	84.7 (6.6)	Jan 2001–Dec 2010, NR	Beers 1991, STOPP v1	Medical record	ARA	STOPP v1: 20; Beers 1991: 9
Cheong et al. 2019 39	UK	CC	Single centre, NR	200	34.5	83.8 (5.68)	Jan–Dec 2015, NR	Beers 2015	Electronic discharge summary record	HR	33
Corsonello et al. 2009 40	Italy	PC	Multicentre, acute care medical wards	506	45.7	80.1 (6.0)	Apr– Jun 2007, 12	Beers 2003	Medical and nurse records	FD, ADE	Admission: 20.6; during hospital stay: 9.7
Counter et al. 2018 41	UK	RC	Single centre, general medical unit	259	49	77	Nov 2013–Jun 2014, 41.5	STOPP/START v2	Inpatient clinical notes, electronic records of outpatient clinic review, GP referral & discharge letters	HR, M	STOPP v2: 59.1; START v2: 69.1; STOPP/START v2: 83.8
Dalleur et al. 2012 42	Belgium	C	Single centre, NR	302	37.4	Median (IQR): 84 (81–88)	Dec 2007–Nov 2008, 12	STOPP/START v1	Electronic medical record	ARA	STOPP v1: 47.7; START v1: 62.9
De Vincentis et al. 2020 43	Italy	PC	Multicentre, medical wards	2631	48.6	Median: 79.6	2010–2016, 3	Beers 2019, STOPP 2015	REPOSI registry	M, HR, FD	Beers 2019: 31.1; STOPP v2: 25.6
Eshetie et al. 2020 44	Australia	PC	Multicentre, general medicine wards	181	45.3	Median: 87.5	5 Jun–7 Jul 2017, 1.25	STOPP 2015, Beers 2019	Medical record	ARA	People with dementia: Beers [A: 79.1, D: 84.6]; STOPP [A: 78, D: 79.1]; people without dementia: Beers [A: 81.1, D: 85.6]: STOPP [A: 87.8, D: 85.6]
Fabbietti et al. 2018 45	Italy	PC	Multicentre, acute care wards of geriatric medicine	647	51	80.1 (6.9)	Jan–Dec 2013, 12	STOPP 2015, Beers 2015	MEDELNET‐AC project	HR	STOPP v2: 30; Beers 2015: 27.7
Fabbietti et al. 2018 46	Italy	PC	Multicentre, geriatric and internal medicine acute care wards	733	45.2	80.06 (7.01)	Jun 2010–May 2011, 3	STOPP 2015, Beers 2015	CRIME project	FD	STOPP v2: 40.2; Beers 2015: 35.9
Fahrni et al. 2019 47	Malaysia	PC	Multicentre, general medical or surgical services	301	54.8	Median (IQR): 72 (67–77)	Jun–Dec 2014, 7	STOPP START v1	Medical record	ADE, ARA	STOPP v1: 34.9; START v1: 37.9; STOPP/START v1: 58.5
Floroff et al. 2014 48	USA	RC	Singe centre, neuroscience IcU	112	45.5	65–74 y: 36.6%; 75–84 y: 36.6%; ≥85 y:26.8%	Mar–Jul 2011, 5	Study specific tool	Electronic medical record	LoS, M, time to recovery	81.3
Forget et al. 2020 49	Canada	RC	Single centre, preoperative clinic	252	46	Median (IQR): 72 (69–76)	Jan 2017–Jan 2018, 3	MedSafer	Community pharmacy	ED visit, LoS	78
Fromm et al. 2013 50	Germany	RC	Multicentre, geriatric units	45 809	30.8	Median (IQR): 82 (78–86)	Jan 2009–Dec 2010, 24	PRISCUS list	Geriatrics in Bavaria databank	FD	25.9
Gallagher et al. 2008 51	Ireland	PC	Single centre, medical and surgical services	715	46	Median (IQR): 77 (72–82)	2007, 4	STOPP v1, Beers 2003	Medical record, GP referral letter, patient, pharmacist	ARA	STOPP v1: 35; Beers 2003: 25
Gallagher et al. 2008 52	Ireland	PC	Single centre, medical & surgical services	597	46	77 (7)	NR, 3	Beers 2003	Medical record, GP referral letter, GP, pharmacy	ARA	32
Galli et al. 2016 53	Brazil	C	Single centre, medical or cardiovascular ICU	599	54.9	Median (IQR): 71 (65–77)	Jan–Dec 2013, NR	Beers 2012	Medical record	ADR	98.2
Gillespie et al. 2013 54	Sweden	RCT	Single centre, internal medicine wards	368	41.3	86.7 (4.1)	Oct 2005–Jun 2006, 12	MAI, STOPP START v1	Electronic case notes	HR, ARA	NR
Glans et al. 2020 55	Sweden	CR	Single centre, NR	720	49.5	Case: 80 (8); control: 78 (8)	2017, 1	SNBHW criteria	Electronic medical record	HR	NR
Gosch et al. 2014 56	Austria	RC	Single centre, geriatrics and internal medicine	457	17.5	80.61 (7.07)	2000–2004, 38	STOPP START v1	Discharge summary	M	STOPP v1: 53.4; START v1: 79.9; STOPP/START v1: 90.4
Gutiérrez‐Valencia et al. 2017 57	Spain	RC	Single centre, acute geriatric unit	200	35	88.3 (5.7)	Jan–Feb 2015, 6	Beers 2015, STOPP/START v2	Medical record, discharge summary	HR, M, ED visit	STOPP v2 [A: 68.5; D: 71.5]; START v2 [A: 58; D: 58]; Beers 2015: [A: 71; D: 71.5]
Hagstrom et al. 2015 58	USA	PC	Single centre, NR	560	53	NR	May 2012–Apr 2013, NR	Beers 2012	NR	LoS, HR, cost	67.8
Hamilton et al. 2011 59	Ireland	PC	Single centre, medical and surgical services	600	40.2	Median (IQR): 77 (72–83)	NR, 4	Beers 2003, STOPP v1	Medical record, patient/care giver interviews	ADE	STOPP v1: 56.2; Beers 2003: 28.8
Hattori et al. 2020 60	Japan	RC	Signe centre, geriatric hospital	116	42.2	85.3 ± 10.2	2016–2018, 24	START v2	Electronic medical record	M	53.3
Iaboni et al. 2017 61	USA	PC	Multicentre, NR	477	24.5	User: 78.5 (8.4); nonuser: 78.4 (9.1)	2008–2012, 12	Beers 2012	Medication record	Time to functional recovery	51
Jensen et al. 2014 62	Denmark	PC	Single centre, acute medical unit	71	55.00	Median: 78.7	Oct–Dec 2011, 1	Red–yellow–Green list (Danish criteria)	Personal electronic medication record, patient interview/care giver interview	HRQoL, FD	84.5
Kanaan et al. 2013 63	USA	PC	Multicentre, NR	731	48.4	78.8 (7.1)	Aug–Dec 2010, 1.5	Beers 2012	Medical record	ADE	NR
Kersten et al. 2015 64	Norway	RC	Single centre, medical and geriatric wards	232	40.9	86 (5.7)	2012, 8	Study specific tool	Medical record, GP referral letter	LoS, FD	Admission: 39.2; discharge: 37.9
Komagamine et al. 2019 65	Japan	PC	Single centre, internal medicine ward	739	47.4	Median (IQR): 82 (74–88)	May 2017–Nov 2018, 1	Beers 2015	Electronic medical record	HR	Admission: 47.2; discharge: 32.2
Kose et al. 2020 66	Japan	RC	Single centre, rehabilitation ward	569	33.6	Median (IQR) 79 (73–85)	July 2010–October 2018, NR	Beers 2019	Medical record	FD	NR
Laroche et al. 2006 67	France	PC	Single centre, acute medical geriatric unit	2018	30.6	85.2 (6.6)	Jan 1994–Apr 1996; May 1997–Jan 1999; 49	Beers 1997	Prescription, patient/care giver interview, GPs	ADR	66
Lau et al. 2017 68	China	RC	Singe centre, medical wards	165	39.4	83.35 (5.49)	1–31 May 2016, 1	STOPP v2	Medical record	HR	27.3
Lester et al. 2019 69	Canada	C	Single centre, level 1 trauma centre	319	64.9	76	Jan 2013–Dec 2014, 1	Beers 2015	Medical record	M, LoS	63.9
Manias et al. 2015 70	Australia	RC	Single centre, medical ward	200	42.5	81.4 (7.16)	May 2012–April 2013, NR	STOPP START v1	Medical record	ADE	STOPP: 51; START: 74
Manias et al. 2019 71	Australia	MM	Single centre, ED and general medical units	249	38.6	Median (IQR): 88 (86–91)	Jan–Dec 2016, NR	STOPP START v2	Medical record	ADE	PIMs (ED: 51; T1: 37.1; T2: 40.4; D: 36.9); PPOs (ED: 44.6; T1:43.8; T2:41.8; D: 36.9)
Mansur et al. 2009 72	Israel	PC & RC	Single centre, acute geriatric ward	212	38.2	81.1 (7.25)	Jul 2004–Jun 2005, 3	Beers 2003	Medical record	HR, M	A: 43.5; D: 44.4
Nagai et al. 2020 73	Japan	RC	Multicentre, surgical ward	253	13.4	75.6 (8.6)	Oct 2014–Dec 2018, 12	STOPP‐J	Electronic medical record	F, FD	42.3
Ni Chroinin et al. 2016 74	Australia	RC	Single centre, medical & surgical wards	534	51.7	78 (9)	Jan 2013, 1	STOPP v1	Medical record	ARA	[A: 54.8, D: 60.8]
Olsson et al. 2011 75	Sweden	PC	Singe centre, NR	140	37.9	83.4 (5.0)	Sep 2006–May 2007, 12	MAI	Medical record, prescription, medication lists	HRQoL	Mean MAI score: 61.3
O'Connor et al. 2012 76	Ireland	PC	Single centre, medical & surgical services	513	44	Median (IQR): 77 (72–82)	Jul–Oct 2010,4	STOPP v1	NR	ADR	51
Onder et al. 2005 77	Italy	RC	Multicentre, NR	5152	47.8	78.8 (8.4)	1997–1998, 24	Beers 2003	GIFA database	M, LoS, ADR	28.6
Ozalas et al. 2017 78	USA	RC	Single centre, acute care for elders unit	340	41.8	84 (11)	Jan–May 2011, NR	Beers 2003 & 2012	Medical record	M, LoS, ADE	Beers 2003: 42.1; Beers 2012: 67.4
Page et al. 2006 79	USA	RC	Single centre, internal medicine services	389	31.1	79	Mar 2000–Aug 2001, 18	Beers 2003	Medical record	M, ADE, LoS	27.5
Pardo‐Cabello et al. 2017 80	Spain	C	Single centre, internal medicine unit	275	43.6	Median (IQR): 82 (76–86)	Feb–Apr 2016, NR	STOPP 2	Medical record, discharge summary	Cost	41.5
Parekh et al. 2018 81	UK	PC	Multicentre, medical wards	1280	42	Median (IQR): 82 (75–87)	2013–2015, 12	Beers 2015	PRIME study	HR, M, ADR, ARA	21.6
Pasina et al. 2014 82	Italy	C	Multicentre, internal medicine & geriatric wards	844	48.8	78.8 (7.4)	2008–2010, 3	Beers 2003, 2012	REPOSI registry	HR, M, ADE	Beers 2003: 20.1; Beers 2012: 23.5
a Passarelli et al. 2005 83	Brazil	PC	Single centre, internal medicine service	186	38.7	73.6 (9.1)	Sep 2002–May 2004, NR	Beers 2003	Medical record, patient interview	ADR	67.4
Rahman et al. 2019 84	USA	RC	Single centre, medical ICU	346	56.4	65–74 y: 51.7%; ≥ 75: 48.3%	Jan–Dec 2014, 12	Beers 2012, 2015, STOPP v1	Medical record	HR, M, LoS	STOPP v1 [A: 44.5, D: 42.9]; Beers 2012 [A: 58.1, D: 63.6], Beers 2015 [A: 68.5, D: 77.4]
Sevilla‐Sanchez et al. 2017 85	Spain	C	Single centre, acute care geriatric unit	235	34.5	86.80 (5.37)	Nov 2014–Aug 2015, 10	MAI, STOPP v2	Patient‐centred prescription	M, LoS	STOPP v2: 88.5; MAI: 97.4
Sevilla‐Sánchez et al. 2018 86	Spain	C	Single centre, acute geriatric unit	235	34.5	86.80 (5.37)	Nov 2014–Aug 2015, 10	STOPP frail	Patient‐centred prescription	M, ADE, ARA, LoS	67.2
Slaney et al. 2015 87	Canada	RC	Single centre, alternate level of care	52	58	82.69 (8.03)	Sep 2012, NR	Beers 2012	Electronic medical record	ADE	92
b Tachi et al. 2019 88	Japan	RC	Single centre, NR	1236	BCJV: 60.3; GM2015:59.2	BCJV: 77.9 (6.8); GM2015:77.7 (7.2)	Oct–Nov 2014, NR	BCJV, GL2015	Electronic medical record	ADR, cost	BCJV: 24; GL2015: 72.4
Tosato et al. 2014 89	Italy	PC	Multicentre, geriatric & internal medicine	871	46.8	80.2 (7)	Jun 2010–May 2011, NR	Beers 2012, STOPP v1	CRIME project	ADR, FD	STOPP v1: 50.4; Beers 2012: 58.4; Combination: 75
van der Stelt et al. 2016 90	Netherlands	CC	Multicentre, NR	338	47.3	Cases: 79.4; control: 78.5	Sep 2005–Jun 2006, 2	Beers 2012; STOPP/START V1	HARM study	ARA	STOPP v1: 34.1; START v1: 57.7; STOPP/START v1: 68.9; Beers 2012: 44.4
Varallo et al. 2011 91	Brazil	C	Single centre, internal medicine ward	308	NR	NR	Aug–Dec 2008, NR	Beers 2003	Medical record, patient/care giver interview	ARA	19.1
Walker et al. 2019 92	USA	RC	Single centre, level 1 trauma centre	2181	48	78.5	Jan 2014–Aug 2017, NR	Modified Beers criteria	Electronic medical record	F	71.2
Wang et al. 2019 93	China	PC	Single centre, comprehensive department	508	61.40	84.2 (5.9)	Jun 2015–Dec 2017, 36	Beers 2015, Chinese criteria 2017	NR	HR, M	Beers 2015: 69.3; Chinese criteria: 66.7
Weir et al. 2020 94	Canada	PC	Multicentre, internal medicine, cardiac & thoracic surgery wards	2402	57.5	Median (IQR): 76 [70–82]	Oct 2014–Nov 2016, 1	Study specific tool	Pharmacy claims database, medical record	ADE, composite outcome	66
Zhang et al. 2017 95	China	C	Single centre, geriatrics department	456	73.20	81.8 (7.8)	May–Dec 2015, NR	Beers 2015, Beers 2012	Medical record	ADR	Beers 2012: 44.7; Beers 2015: 53.5

A, admission; ADE, adverse drug event; ADR, adverse drug reaction; ARA, adverse drug reaction/event related hospital admission; BCJV, Beers Criteria–Japanese Version; CR, Comparative retrospective; CRIME, CRIteria to Assess Appropriate Medication Use among Elderly Complex Patients; D, discharge; ED, emergency department; HR, hospital readmission; M, mortality; FD, functional decline; F, falls; C, cross sectional; CC, case–control; RC, retrospective cohort; RCT, randomised controlled trials; PC, prospective cohort; GL2015, Guidelines for Medical Treatment and its Safety in the Elderly 2015; HARM, hospital admissions related to medication; HRQoL, health‐related quality of life; LoS, length of stay; NR, not reported; PIP, potentially inappropriate prescribing; IQR, interquartile range; SD, standard deviation; STOPP, Screening Tool of Older Person's Prescriptions; STOPP‐J, Screening Tool for Older Persons' Appropriate Prescriptions for Japanese; START, Screening Tool to Alert to Right Treatment; SNBHW, Swedish National Board of Health and Welfare; NORGEP, Norwegian General practice; MM, mixed methods; ICU, intensive care unit; MAI, medication appropriateness index.

The study design was not stated but assigned by the authors of this review, considering the methodological procedure described in the study.

Sex proportion and mean age was calculated only for patients exposed to inappropriate medication use.

Over half of the studies (n = 36) assessed PIP exposure using any versions of the Beers criteria, followed by STOPP (26 studies) and START criteria (12 studies). Other tools employed to assess the appropriateness of medication use included Medication Appropriateness Index (3 studies), PRISCUS list and STOPP Frail (each 1 study) and other study or country specific tools (8 studies). Study or country specific tools were derived mainly from a mix of tools, such as the Beers and STOPP criteria. Medical record review, either paper or electronic, was the main source of data for PIP identification. Some studies followed‐up patients for assessment of outcomes, ranging from 3 weeks to 49 months (Table 1). Based on the MMAT, 42 studies fulfilled at least 4 of the 5 items (Appendix 2).

PIP prevalence and common medications involved in PIPs

Based on different sets of PIP criteria, more than 1 prevalence estimate was reported in 25 studies, and discrete prevalence estimates for care transitions (e.g. admission, discharge) per study were reported in 8 studies. Overall, the pooled PIM prevalence was estimated at 47% (95% CI 37–56), 46% (95% CI 39–53), and 56% (95% CI 40–72) according to the different versions of Beers, STOPP and study or country‐specific criteria, respectively. The overall estimated PPO prevalence, from the pooled analysis of the START criteria, was 55% (95% CI 46–64) (Appendix 3). The most frequently reported PIMs or medication classes were benzodiazepines, antipsychotics, antihistamines/anticholinergics and antithrombotics, whereas the most frequently reported PPOs were: antiplatelet therapy with documented history of coronary, cerebral or peripheral vascular disease; and vitamin D and calcium supplement in patients with known osteoporosis or previous fragility fracture. Commonly reported PIMs contributing to adverse outcomes related to medications from benzodiazepine, opioid and antipsychotic classes (Appendix 4).

Association of PIP with outcomes

A total of 39 included studies reported results based on adjusted estimates. The key covariates that were adjusted for included age, sex, disease comorbidities, and number of medications (Appendix 5).

PIPs and mortality

Nineteen studies measured the association between PIP and mortality. , , , , , , , , , , , , , , , , , , Four studies reported in‐hospital mortality, , , , the remainder assessed mortality outcome after hospital discharge. Bo et al., apart from reporting the association between the full PIP exposure (inclusive of all types of medications) and mortality, also reported the association of specific PIPs with mortality 6 month after hospital discharge. Full PIP exposure did not have a significant association with mortality; however, the prescription of specific PIPs, such as antipsychotics (adjusted OR [AOR] 1.65, 95% CI 1.12–2.44) and digoxin dosage ≥ 0.125 mg/d (AOR 1.77, 95% CI 1.06–2.98) were associated with higher odds of mortality.

Only 4 , , , of 19 studies found an increased risk of mortality from either full or specific PIP exposure. Three meta‐analyses for the association of PIPs with mortality were conducted to combine results from different risk estimates. Results from a pooled analysis of ORs did not show a significant difference between PIP users and nonusers (AOR 1.10, 95% CI 0.90–1.36, P = .35; Figure 2A), and the same for pooled crude ratios (OR 1.15, 95% CI 1.00–1.31, P = .05; Table 2). Similarly, the effect estimates of 2 studies , evaluating the association of the numbers of PIPs (measured as continuous variable) and mortality, did not produce a significant result (AOR 1.49, 95% CI 0.98–2.26, P = .06; Figure 2b), as was for studies reporting risk estimates using hazard ratio (adjusted HR [AHR] 1.02, 95% CI 0.90–1.16, P = .75; Figure 2c).

FIGURE 2

(A) Forest plot of adjusted odds ratio for an association between PIP users (compared with nonusers) and all‐cause mortality. (B) Forest plot of adjusted odds ratio for an association between the numbers of PIPs (measured as continuous variable) and all‐cause mortality. (C) Forest plot of adjusted hazard ratios for an association between PIP and all‐cause mortality. Studies with ≥2 outcome data using various tools are shown with the type of tool. AORs, adjusted odds ratios; AHRs, adjusted hazard ratios; PIP, potentially inappropriate prescribing

TABLE 2

Pooled odds ratio and sub‐group analysis, stratified by covariate adjustment, potentially inappropriate prescribing (PIP) tool, country, study design and quality score (n = 21)

Stratification a	Mortality			Hospital readmission			ADRs/ADEs
	n	OR (95% CI)	SD (I 2), P	n	OR (95% CI)	SD (I 2), P	n	OR (95% CI)	SD (I 2), P
Unadjusted estimates (all studies)	19	1.15 (1.00, 1.31)		15	1.22 (1.03, 1.44)		7	1.80 (1.48, 2.21)
PIP tool
Beers 1997/2003	4	1.06 (0.78, 1.43)	0%, .63	1	0.77 (0.48, 1.24)	59.2%, .04	2	1.77 (1.38, 2.27)	0%, .61
Beers 2012/2015/2019	6	1.16 (0.88, 1.52)		7	1.08 (0.91, 1.30)		3	1.60 (1.05, 2.44)
STOPP	4	1.04 (0.80, 1.36)		4	1.75 (1.01, 3.01)		1	2.78 (1.33, 5.81)
START	3	1.25 (0.82, 1.92)		1	1.67 (1.18, 2.36)		0	―
Study/country specific	2	1.62(0.95, 2.74)		2	1.18 (0.87, 1.60)		1	2.20 (0.84, 5.76)
Country
America	4	1.90 (1.19, 3.03)	63.9%, .07	0	―	22.5%, .26	2	1.67 (1.14, 2.45)	0%, .8
Europe	12	1.08 (0.93, 1.25)		11	1.14 (0.99, 1.31)		4	1.88 (1.48, 2.40)
Asia	3	1.28 (0.88, 1.85)		4	1.67 (0.88, 3.19)		1	1.40 (0.45, 4.36)
Study design
Prospective cohort	8	1.18 (1.02, 1.36)	68.5%, .04	12	1.20 (1.05, 1.37)	92%, <.00001	3	2.31(1.46, 3.65)	0%, .48
Retrospective cohort	8	1.32 (0.98, 1.78)		1	6.48 (3.00, 14.02)		3	1.72 (1.37, 2.16)
Cross sectional	3	0.76 (0.53, 1.08)		2	0.80 (0.58, 1.10)		1	1.40 (0.45, 4.36)
Quality score									0%, .61
5	9	1.01 (0.81, 1.25)	6.2%, .11	6	1.00 (0.84, 1.19)	82.3%, .02	5	1.91(1.42, 2.56)
<5	10	1.25 (1.07, 1.47)		9	1.44 (1.13, 1.85)		2	1.72 (1.30, 2.26)
Adjusted estimates (all studies)	12	1.10 (0.90, 1.36)		8	1.11 (0.76, 1.63)		15	1.26 (1.11, 1.43)
PIP tool
Beers 1997/2003	4	1.03 (0.76, 1.40)	74.3%, .0004	1	0.72 (0.43, 1.21)	25.5%, .26	6	1.24 (0.98, 1.57)	22.9%, .27
Beers 2012/2015	4	0.91 (0.83, 1.01)		4	0.88 (0.65, 1.18)		4	1.16 (0.90, 1.49)
STOPP	1	1.09 (0.63, 1.89)		2	3.16 (0.79, 12.57)		2	1.65 (0.87, 3.12)
START	2	1.87 (1.16, 3.01)		0	―		0	―
STOPP/START	1	2.51 (1.20, 5.25)		0	―		0	―
Study/country specific	0	―		1	0.99 (0.57, 1.72)		3	1.38 (1.13, 1.70)
Country
America	3	1.49 (0.86, 2.58)	17.5%, .30	0	―	0%, .54	3	1.42 (1.06, 1.91)	4.5%, .19
Europe	8	1.04 (0.83, 1.31)		6	0.98(0.78, 1.23)		9	1.14 (0.97, 1.35)
Asia	1	1.84 (0.70, 4.84)		2	2.00 (0.20, 19.58)		0	―
Others	0	―		0	―		3 b	1.44 (1.16, 1.78)
Study design
Prospective cohort	1	0.75 (0.48, 1.17)	7.7%, .03	5	1.01 (0.76, 1.35)	91.3%, <.0001	7	1.28 (1.07, 1.54)	42.3%, .18
Retrospective cohort	9	1.29 (0.99, 1.68)		1	6.56 (2.89, 14.88)		5	1.37 (1.12, 1.68)
Cross sectional	2	0.72 (0.45, 1.17)		2	0.75 (0.53, 1.06)		3	0.93 (0.65, 1.34)
Quality score
5	6	0.90 (0.66, 1.21)	61.6%, .11	4	0.84(0.62, 1.14)	59.4%, .12	11	1.34 (1.11, 1.63)	0%, .37
<5	6	1.17 (0.94, 1.73)		4	1.65(0.75, 3.63)		4	1.20 (1.02, 1.40)

Data for other outcomes not reported, not enough subgroups;

Canada, Brazil; ADRs/ADEs, adverse drug reactions/adverse drug events; n, total number of screenings (>1 screening may be contributed by a single study); SD, sub‐group difference; CI, confidence interval; OR, odds ratio.

PIPs and hospital readmissions

Eighteen studies provided data on all‐cause hospital readmissions. , , , , , , , , , , , , , , , , , Bo et al. reported both the associations between full (inclusive of all medications) and specific PIPs exposure with hospital readmissions. Irrespective of the screening criteria and PIP measurement (as dichotomous, continuous and categorical), only 5 of these studies , , , , demonstrated a positive association between PIPs and hospital readmissions. The number of PIPs (continuous) as predictors of hospital readmission were reported by 5 studies, , , , , with only 1 study showing a significant positive association. We did not perform meta‐analysis using PIPs as a continuous variable, because summary risk estimates were provided in different formats or studies did not provide sufficient detailed information. Also, PIPs (measured dichotomously) were reported in 13 studies, but only 7 studies , , , , , , gave data suitable for adjusted meta‐analysis. The pooled estimate for full PIP exposure and all‐cause hospitalisations did not reach statistical significance (AOR 1.11, 95% CI 0.76–1.63, P = .59; AHR 1.02, 95% CI 0.89–1.18, P = .74; Figure 3) although meta‐analysis of the crude odds ratios showed a positive association (OR 1.22, 95% CI 1.03–1.44, P = .02; Table 2). The meta‐analysis of AOR was associated with a significant heterogeneity (I 2 = 76%) that was minimised on removal of Lau et al. (I 2 = 29%, P = .5).

FIGURE 3

(A) Forest plot of adjusted odds ratios for an association between PIP (measured dichotomously) and all‐cause hospital readmission. (B) Forest plot of adjusted hazard ratio for an association between PIP and all‐cause hospital readmission. Studies with ≥2 outcome data using various tools are shown with the type of tool. AORs, adjusted odds ratios; AHRs, adjusted hazard ratios; PIP, potentially inappropriate prescribing

PIPs and ADE‐related hospital admissions

Overall, 12 studies evaluated the impact of PIPs on medication‐related hospital admissions: 7 studies , , , , , , reported the prevalence of hospital admissions due to PIPs (as judged by an expert panel) and 5 studies , , , , assessed the association between PIPs and ADE‐related hospital admissions. A pooled analysis of hospital admissions due to PIP estimated that PIP use was causal or contributory to admission in 11% of patient admissions (95% CI 8–15%). A meta‐analysis also showed that PIP use was associated with a 91% increased odds of ADE‐related hospital admissions (AOR 1.91, 95% CI 1.21–3.01, P = .005; Figure 4a). However, on sensitivity analysis, the association between PIPs and ADE‐related hospital admissions was not statistically significant when only the weakest association from a study contributing 4 AOR estimates using various PIP tools, was included in the pooled analysis (AOR 1.65 95% CI 0.75–3.62; P = .21).

FIGURE 4

(A) Forest plot of adjusted OR for the association between PIPs (measured dichotomously) and ADE‐related hospital admissions. (B) Forest plot of adjusted OR for the association between PIMs and ADRs/ADEs. (C) Forest plot of adjusted odds ratio for the association between PIMs (measured as a continuous variable) and ADRs/ADEs. Studies with ≥2 outcome data using various tools are shown with the type of tool. AORs, adjusted odds ratios; AHRs, adjusted hazard ratios; PIP, potentially inappropriate prescribing

PIPs and ED visits

Three studies reported the association between PIPs and ED visits, either as a separate outcome , or as part of a composite outcome. Using an electronic prescribing tool, Forget et al. did not show a significant association between the numbers of PIMs and ED visits in the 90 days post hospital discharge, irrespective of frailty status. Likewise, Gutiérrez‐Valencia et al. reported that the presence of Beers, STOPP or START criteria did not show an association with ED visits at 6 months. By contrast, Wier et al. (using a study specific tool) reported that each additional new PIM prescribed at discharge, was associated with an increased risk of composite outcome (ED visit, rehospitalisation, or death) in the 30 days following hospital discharge (AHR 1.13, 95% CI 1.03–1.26). Also, receiving at least 1 new PIM prescription (new PIM users) was marginally associated with the composite outcome (AHR 1.22, 95% CI 1.00–1.49). Alternatively, chronic use of PIMs (e.g. PIMs continued from the community), measured as either discrete or continuous variable, did not show any independent significant association with the composite outcome.

PIPs and length of stay

Ten studies described the relationship between PIP and length of stay (hospital or intensive care unit). , , , , , , , , , Across the studies, there was no clear association between PIP and length of stay. However, there was some indication that prescription of Beers medications (especially 2 or more) was associated with an increased length of hospital stay. , , , Conversely, 1 study reported that the use of PIM as determined by the STOPP was significantly associated with an increased intensive care unit and hospital stay but no association with the Beers criteria.

PIPs and ADRs/ADEs

Twenty‐three studies assessed the impact of PIPs on the occurrence of ADRs/ADEs, either through analysing the association between PIMs and ADRs/ADEs , , , , , , , , , , , , , or simply reporting only the share of PIMs in the occurrence of ADRs/ADEs. , , , , , , , , Links between PPOs and ADRs/ADEs were not reported by any study. Two meta‐analyses were conducted to determine the association between PIMs and ADRs/ADEs. The first meta‐analysis pooled adjusted odds ratios of the association between PIMs (measured dichotomously) and ADRs/ADEs, indicating that PIM users (compared with nonusers) were associated with a 26% increase in the odds of ADRs/ADEs (AOR 1.26, 95% CI 1.11–1.43, P = .0003; Figure 4a). Likewise, the direction of effect was the same using pooled crude OR (Table 2). The second meta‐analysis combined results to estimate the association between PIMs (measured as a continuous variable) and ADRs/ADEs, implying that for every additional PIM, there was a 73% increased odds of ADEs/ADEs (AOR 1.73, 95% CI 1.26–2.37, P = .0008; Figure 4b). However, this meta‐analysis was associated with significant statistical heterogeneity (I 2 = 91%).

PIPs and functional outcomes

Twelve studies reported the association between PIMs and functional status, expressed in terms of mobility, , , hand‐grip strength, , time to functional recovery , and functional independency. , , , , , , , , No study reported these outcomes for PPOs. None of the studies , , reported a significant association between PIMs and mobility (measured using the timed up‐and‐go test). Two studies , reported that PIM users were significantly associated with longer time to achieve recovery than non‐PIM users. The use of PIMs was also associated with lower handgrip strength, which was measured using dynamometer, in 1 study but not in the other study. However, exposure with multiple specific PIMs; that is, a concomitant use of 3 and more psychotropic or opioid medications was associated with reduced hand‐grip strength.

Functional independence was measured using various instruments: the Barthel Index , , , ; the ADL (activity of daily living) score , , ; the FIM (functional independence measure) score ; and the new mobility score. A meta‐analysis of an association between PIMs and functional decline, defined as the loss of independence in at least 1 ADL, was conducted. The pooled estimate showed that the use of PIMs increased the odds of functional decline by 60% (AOR 1.60, 95% CI 1.28–2.01, P < .0001: Figure 5). However, this association was not significant on limiting the analysis to include the weakest estimate from studies contributing 2 or more estimates (AOR 1.24 95% CI 0.86, 1.79, P = .25).

FIGURE 5

Forest plot of adjusted odds ratio for the association between PIPs (dichotomous) and functional decline. Studies with ≥2 outcome data using various tools are shown with the type of tool. AORs, adjusted odds ratios; PIP, potentially inappropriate prescribing

PIPs and falls

Two studies , reported falls as an outcome. The prescription of Beers medications was significantly associated the incidence of falls. Similarly, the number of PIMs prescribed (according to STOPP for the Japanese version) was associated with increased occurrence of subsequent falls 1 year after hospital discharge.

PIPs and health‐related quality of life

The association between PIPs and health‐related quality of life (HRQoL) was reported in 4 studies, , , , all using the EuroQol‐5 dimensions (EQ‐5D). Two studies , additionally employed the EuroQol‐Visual Analogue Scale (EQ‐VAS) to measure self‐rated HRQoL. Using STOPP/START criteria, 1 study did not find a difference between patients who had PIM/PPO and those who did not in associations with EQ‐5D index and EQ‐VAS. Another study, using the medication appropriateness index (MAI) but the same HRQoL measures, reported lower medication quality was associated with a lower HRQoL. Associations were not clear in the remaining studies; for example, inappropriate medication use (screened via STOPP and a country‐specific tool ) was significantly associated with reduced HRQoL but only when PIMs were measured dichotomously and only red PIMs (defined as medications that should be avoided irrespective of diagnosis, according to the Danish Criteria ) were included, respectively.

Cost implications of PIPs

Three studies reported the economic costs of PIMs. , , No studies reported cost implications of PPOs. Hagstrom et al. reported those individuals with 3 or more PIMs compared with those with 1 PIM had statistically significant higher hospital costs in the USA. Pardo‐Cabello et al. evaluated the mean cost of PIMs using STOPP v2 and determined that the cost associated with PIM use was €18.75 ± 4.24 per patient per month (€225.14 ± 50.91 per patient per year), with opioids accounting for the highest percentage of the expenditure. Similarly, Tachi et al. calculated the extra cost for treatment of adverse reactions per inpatient who was prescribed drugs listed in the Beers Criteria–Japanese Version and Guidelines for Medical Treatment and its Safety in the Elderly 2015 and was estimated to range from 497 to 13 371 yen per patient (≈7–180 AUD), which corresponds to a national cost of 2.18–381.42 (≈0.03–5 AUD) billion yen per year. Overall, whether the estimation was on total hospital costs, or the extra costs due to PIMs and treatment of PIM‐related ADRs, the use of PIM was associated with higher economic cost.

DISCUSSION

The systematic review showed a pooled PIM estimate of between 46 and 56%, depending on the tool used, and a pooled PPO estimate of 55% based on the START criteria. Substantial exposure of PIPs during hospital care had significant associations with a range of health‐related and system‐related outcomes, including medication‐related hospitalisation, ADRs/ADEs, functional decline, falls and health care costs. However, based on adjusted estimates, PIP did not show a significant association with all‐cause mortality and hospital readmissions. Additionally, inconsistent findings were noted for other outcomes, such as ED visits, length of stay and HRQoL. Most importantly, PIP outcomes were most often related to PIMs; none of the included studies explored links between PPOs and ADRs, ADEs, functional decline, falls and cost.

Comparison with existing literature

Previous systematic reviews have examined associations between PIMs and various outcomes, mainly in heterogeneous healthcare settings, which included community setting, nursing home and hospital, , , , or only in primary care. , The findings of our review are consistent with previous reviews on all‐cause mortality, but not hospital readmissions. For example, a systematic review and meta‐analysis by Xing et al. included 33 studies from various healthcare settings reporting that PIMs (identified by Beers and STOPP criteria) were significantly associated with an increased risk of ADRs/ADEs and hospital readmission but not mortality. Likewise, other reviews have also reported PIPs did not affect mortality , and yet the impact on hospital readmission was significant, whether in a primary care or across healthcare settings. ,

It should be noted that the methodology used in our review has identified 4 important differences (apart from settings) compared with previous reviews. , , , , , First, we separately analysed all‐cause hospital readmissions from ADE‐related admissions. Interestingly, when doing so, there was a significant association between PIPs and ADE‐related hospital admissions. The current review found that approximately 1 in 10 hospital admissions were related to PIMs, as a primary or contributory cause. Second, we did not combine risk estimates from various measures of PIP, which typically may lead to erroneous conclusions. Here, we explored the association between PIPs and health‐related and system‐related outcomes, considering PIPs as a dichotomous variable (PIP users vs. nonusers), and the number of PIPs as both a continuous and as a categorical (0, 1, 2 and ≥3 PIP) variable. However, this way of classification was not without challenges, especially when conducting meta‐analysis using PIPs as a continuous variable. Very few studies gave data in a suitable format for meta‐analysis. Third, meta‐analysis was conducted using the full PIP exposure, especially for all‐cause mortality, in which 1 study provided data for both full and specific PIPs. While the full PIP exposure did not show a significant association with mortality, the prescription of specific medications, such as antipsychotics and digoxin dosage ≥0.125 mg/d were associated with higher odds of mortality. There is evidence showing that prescriptions of these medications are associated with all‐cause mortality. , Fourth, we pooled data using a random effects model (as opposed to fixed effects) considering the variation in the tools employed to measure PIPs.

Our results found that evidence for the associations between PIP and other system‐related outcomes, such as ED visits and length of hospital stay, were inconclusive. This was consistent with findings from a previous review across healthcare settings. However, there is some evidence that PIP in primary care has an association with ED visits. Despite our inconclusive findings about PIP and ED or hospital usage, this review does provide evidence about the association between prescription of multiple PIMs and increased length of hospital stay. In particular, the prescription of PIMs at hospital discharge was significantly associated with composite outcomes (comprising ED visit, hospital readmission and mortality). The higher risk of hospital discharge PIMs (compared to community PIMs) may be due to the possibility of medication discontinuation before patients' hospitalisation if they had already experienced an adverse event.

In the present review, the PIMs that most often contributed to adverse health‐related outcomes were medications from benzodiazepine, opioid and antipsychotic classes. These groups of medications have been associated with increased risk of falls. , , Although only 2 studies, , in the current review, assessed the association between full PIM exposure and falls as a primary outcome, detecting a significant positive relationship; many of the included studies , , , , demonstrated PIMs that increased fall‐risk were largely responsible for medication‐related hospital admissions. This is particularly important given that >2/3 of medication‐related hospital admissions are likely to be preventable.

Implications for practice and research

The present review suggests that interventions targeting PIM use may prevent medication‐related harm and improve health outcomes among hospitalised older adults. Our findings showed significant associations between PIMs and medication‐related hospitalisation, ADRs/ADEs and functional decline. Hospitalisation offers an opportunity for medication review and rationalisation although a high rate of PIM, including new PIMs, is also likely at hospital discharge. , The strength of associations with health outcomes was consistently highest for new PIMs. It is, therefore, recommended to have a comprehensive assessment of medication use, especially during care transitions such as hospital discharge, in order to prevent new PIMs from occurring during the patient's journey, and not cascaded into the community. In contrast, the evidence about associations between PPOs and health outcomes (e.g. ADRs/ADEs, functional outcomes, falls) are both limited and unclear, hence indicating a need for further studies. Although limited studies evaluated PPOs, the predictive validity of the START criteria for mortality outcome appears promising and needs further investigation.

Deprescribing interventions are generally feasible to reduce PIMs in a hospital setting, but the evidence is limited about the impact on clinical outcomes. In addition to deprescription, strategies to reduce omission of important medications, such as vitamin D and calcium supplementation in patients prone to falls, can reduce risk of fractures and falls. In our current review, the most frequently reported PPOs were vitamin D and calcium supplement in patients with known osteoporosis or previous fragility fracture. It is possible that many PIP‐related adverse outcomes are preventable by amalgamating screening tools with practice measures, such as medication reconciliation and medication review.

Strengths and limitations

This systematic review provides a comprehensive exploration of the association between PIPs and a range of health‐related outcomes among older adults in hospital settings. Multiple electronic databases and rigorous screening were used to locate studies evaluating all types of PIP (consisting of PIMs and PPOs), without restricting to specific screening tool for identification of PIPs.

We performed meta‐analysis using both adjusted and unadjusted data providing opportunity to examine consistency of the evidence and detect confounding heterogeneity. It is evident that adjusted estimates control confounding, but if used alone may lead to an overestimation of the association. ,

Our review has several limitations that merit consideration. First, there were some studies that did not apply the full screening criteria, mainly those studies employing the Beers criteria. Many of the included studies , , , , , , that employed the different versions of the Beers criteria, only adopted the criteria for PIM use independent of diagnosis. Similarly, there were also studies that did not apply the full version of STOPP. , , , , , , , , These may have caused the heterogeneities and variations in estimates, but we did not perform subgroup analysis based on the completeness of tool because of fewer studies per outcome. Second, included studies varied in terms of adjustment for confounding variables. While many included studies adjusted for multiple confounders, there are still studies that did not sufficiently control for relevant confounders, such as number of medications. , , , , , , , , , The number of medications is the most consistent determinant of PIM use across settings. Also, it is debatable whether the health outcomes are due to the PIPs or the disease/condition itself. Several studies , , , , , , , , failed to adjust for comorbidities. The heterogeneity in adjustment may be 1 of the factors why pooled estimates from the adjusted vs. unadjusted model vary in the magnitude/direction of effect, specifically for the outcome related to hospital readmissions. Third, combining 2 or more risk estimates from a single study for a same outcome may carry a risk of bias. For instance, sensitivity analyses confirmed that the associations between PIPs and ADE‐related hospital admissions, as well as with functional decline were not statistically significant when limiting the analyses to estimates with the weakest association. Fourth, some studies were not designed to investigate the impact of PIPs on health‐related outcomes. For example, PIMs were counted as covariates in the assessment of ADRs , , or hospital readmission, , rather than as a primary exposure of interest.

CONCLUSION

Our systematic review and meta‐analysis revealed a substantial proportion of patients had PIP during hospitalisation and exposure to PIP had a significant association with a range of important health and system‐related outcomes in the inpatient hospital setting. These outcomes included medication‐related hospitalisation, ADRs/ADEs, functional decline, falls and health care cost. However, PIPs (whether dichotomously or continuously measured) did not show an association with all‐cause mortality or hospital readmissions based on adjusted estimates. The impact of PIPs on other outcomes, such as ED visits, length of stay and HRQoL, was inconclusive. PIP‐related adverse outcomes are amendable by incorporating common screening tools within interventions designed to optimise older adults' prescriptions at hospital transitions.

COMPETING INTERESTS

There are no competing interests to declare.

CONTRIBUTORS

A.B.M., B.R. and E.M. were involved in conceptualisation, design and framing the research question. A.B.M conducted literature searches, and study selection with B.R. and E.M. helping an independent screening. A.B.M. conducted quality appraisal. A.B.M., B.R. and E.M. were involved in the preparation of the manuscript, including data analysis and interpretation of results. B.C. critically revised the initial manuscript draft for important intellectual content. All authors made suggestions and approved the final version of the manuscript.

Data S1 Appendix 1. The search strategyAppendix 2. Quality assessment using the mixed‐methods quality appraisal toolAppendix 3. Pooled PIP prevalence estimates, stratified based on the PIP toolAppendix 4. Summary of commonly involved PIP medicationsAppendix 5. Covariates adjusted, extracted from each included study

Click here for additional data file.