High‐risk prescribing in an Irish primary care population: trends and variation

Abstract

Aims

The aims of the present study were to examine the prevalence of high‐risk prescribing (HRP) in community‐dwelling adults in Ireland from 2011–2015 using consensus‐validated indicators, factors associated with HRP, and the variation in HRP between general practitioners (GPs) and in the dispensing of high‐risk prescriptions between pharmacies.

Methods

A repeated cross‐sectional national pharmacy claims database study was conducted. Prescribing indicators were based on those developed in formal consensus studies and applicable to pharmacy claims data. Multilevel logistic regression was used to examine factors associated with HRP and dispensing.

Results

There were significant reductions in the rates of most indicators over time (P < 0.001). A total of 66 022 of 300 906 patients at risk in 2011 [21.9%, 95% confidence interval (CI) 21.8, 22.1%], and 42 109 of 278 469 in 2015 (15.1%, 95% CI 15.0, 15.3%), received ≥1 high‐risk prescription (P < 0.001). In 2015, indicators with the highest rates of HRP were prescription of a nonsteroidal anti‐inflammatory drug (NSAID) without gastroprotection in those ≥75 years (37.2% of those on NSAIDs), coprescription of warfarin and an antiplatelet agent or high‐risk antibiotic (19.5% and 16.2% of those on warfarin, respectively) and prescription of digoxin ≥250 μg day^–1 in those ≥65 years (14.0% of those on digoxin). Any HRP increased significantly with age and number of chronic medications (P < 0.001). a) After controlling for patient variables, the variation in the rate of HRP between GPs was significant (P < 0.05); and b) after controlling for patient variables and the prescribing GP, the variation in the rate of dispensing of high‐risk prescriptions between pharmacies was significant (P < 0.05).

Conclusions

HRP in Ireland has declined over time, although some indicators persist. The variation between GPs and pharmacies suggests the potential for improvement in safe medicines use in community care, particularly in vulnerable older populations.

What is Already Known about this Subject

• High‐risk prescribing is known to increase the risk of drug‐related morbidity.

• Prescribing safety indicators have been developed to identify patients at risk of high‐risk prescribing.

What this Study Adds

• In Ireland, high‐risk prescribing declined over time from 2011–2015, although some indicators, including those associated with nonsteroidal anti‐inflammatory drug or warfarin prescriptions, persist.

• Increasing age and number of chronic medications are the main drivers of high‐risk prescribing.

• Variation in high‐risk prescribing between general practitioners, and in the dispensing of high‐risk prescriptions between pharmacies, is significant.

Introduction

The safety of medication use in primary care is an increasing priority for health policy and health services 1, 2. Prescribing errors in primary care have been shown to cause considerable harm, with systematic reviews estimating that between 2% and 4% of hospital admissions are due to preventable drug‐related morbidity 3, 4, 5, 6. Iatrogenic morbidity also has a significant impact on healthcare costs. In the US, the total cost of drug‐related hospitalizations was estimated to be US$38.9 billion in 2011 7, and in 2015 preventable drug‐related hospital admissions in England were estimated to cost commissioners around £530 million per year 8. However, hospital admissions account for only a small fraction of drug‐related harm and inconvenience to patients as many preventable adverse drug events (ADEs) are managed in primary care and do not require hospital admission 2. The drug classes most commonly implicated in preventable drug‐related hospital admissions are antiplatelet agents, diuretics, nonsteroidal anti‐inflammatory drugs (NSAIDs) and anticoagulants 4. Mortality has most frequently been associated with prescribing NSAIDs and antiplatelet agents 9. The majority of preventable harm is therefore due to therapeutic agents that are commonly used in primary care 2.

High‐risk prescribing (HRP) can be defined as prescribing which may lead to adverse clinical outcomes or is not aligned with quality use of medicines 10. While only a small proportion of patients affected by high‐risk medication use will ultimately be harmed, averting preventable harm by avoiding HRP where possible, and regularly reviewing such prescribing, is imperative 2. Prescribing safety indicators have been developed to define prescribing patterns that may increase the risk of harm to the individual and should generally be avoided 11. Consensus‐validated sets of indicators of potentially inappropriate prescribing (PIP) focusing on older patients, such as the Beers criteria 12 and the Screening Tool of Older Persons potentially inappropriate Prescriptions (STOPP) criteria 13, 14, have been used in primary care but have limitations. For example, a proportion of listed items in the Beers criteria are rarely used in Europe and many of the drugs commonly associated with serious harm are not included 1, 15. The STOPP criteria are better associated with harm in older persons than the Beers criteria 14 but many of the indicators require information that is not consistently recorded in electronic healthcare databases, which limits routine or large‐scale application 1, 2, 15. These limitations led to the development of new instruments for use in primary care which can be applied to electronic databases 2, 15. In Scotland, a focused set of indicators targeting HRP of NSAIDs, warfarin, antipsychotic drugs, methotrexate and drugs which can aggravate heart failure, which could be implemented in routine primary care datasets, was developed through consensus between general practitioners (GPs) and pharmacists 1. The prescribing indicators in this set were considered to be able to identify the high‐risk use of drugs that have been shown to either commonly cause harm and/or cause severe harm in primary care. Their prevalence was measured in 315 general practices and HRP was shown to be common and variable between primary care practices 1.

Drug treatment in Ireland may be initiated in primary or secondary care but GPs prescribe the majority of drugs in the community, as is the case in most developed countries 16. Research has mainly focused on PIP in older persons, with studies consistently showing a high prevalence of PIP in primary care 17, 18, 19, 20. However, there is a lack of evidence at the population level on the prevalence of HRP in primary care in Ireland. In order to develop strategies to minimize HRP, a reliable evidence base on the magnitude of the problem and the identification of targets for improved prescribing is required 10.

The objectives of the present study were to: (i) examine the prevalence of HRP in those community‐dwelling adults at risk in Ireland from 2011 to 2015 using consensus‐validated indicators; (ii) examine patient factors associated with HRP in those adults at risk; and (iii) examine variation in HRP between individual GPs, and in the dispensing of high‐risk prescriptions between pharmacies.

Methods

Sample selection and study design

This was a repeated cross‐sectional national pharmacy claims database study. The Irish Health Service Executive (HSE) Primary Care Reimbursement Service (PCRS) pharmacy claims database was used to identify the study cohort: adults (aged 16 years and over) enrolled in the General Medical Services (GMS) scheme, who had been dispensed medicines or combinations of medicines considered as HRP in each year from 2011 to 2015. The HSE‐PCRS pharmacy claims database is used primarily to reimburse pharmacists for the provision of prescription medication in Ireland, through a number of national schemes, including the GMS scheme. The GMS scheme is a form of public health cover which provides approximately one‐third of the Irish population, subject to specific eligibility criteria, with access to free healthcare, routine dental work and prescription medication (although a small monthly copayment per prescription item has applied since October 2010). Eligibility for the GMS scheme is based on means testing, with a higher threshold for those over 70 years. The GMS scheme therefore over‐represents older persons, as well as women and more socially deprived individuals. Permission to use the HSE‐PCRS data for research purposes was obtained from the HSE‐PCRS. Ethics committee approval for the present study was not required. The World Health Organization Anatomical Therapeutic Chemical (ATC) classification was used to define the drug classes used 21.

Prescribing safety indicators were based on those that had been developed in formal consensus studies without modification 1, 15. Diagnostic information and outcome data are not available in the claims database, so analyses were restricted to HRP indicators that could be assessed using pharmacy claims data alone (Table 1). The indicators were applied over each year from 2011 to 2015, inclusive. For indicators involving coprescription with warfarin, the use of the coprescribed drug must have been at the same time as or between two prescriptions for warfarin within 12 weeks 1. For the indicator of NSAID prescription in patients aged ≥75 years without gastroprotection, this required there being no prescription of a gastroprotective drug in the 8 weeks prior to, or at the same time as, the NSAID prescription 1.

Table 1.

Prescribing safety indicator definitions within each year

Measure name	Numerator definition	Denominator definition
NSAID prescribed in patients ≥65 years currently using ACE/ARB and diuretic	No. of patients ≥65 years coprescribed NSAID with ACE/ARB and diuretic	No. of patients aged ≥65 years prescribed ACE/ARB and diuretic
NSAID prescribed in patients ≥75 years without gastroprotection	No. of patients ≥75 years prescribed NSAID (Mar‐Dec only) without prescription for gastroprotection in previous 8 weeks	No. of patients aged ≥75 years prescribed NSAID (Mar‐Dec only)
NSAID, antiplatelet agent a , high‐risk antibiotic b or oral azole antifungal prescribed to current warfarin user	No. of patients prescribed one of the specified drugs within 12 weeks of warfarin prescription	No. of patients prescribed warfarin
Verapamil/diltiazem prescribed to current beta‐blocker user	No. of patients coprescribed verapamil/diltiazem with beta‐blocker	No. of patients prescribed beta‐blocker
Digoxin prescribed at daily dose ≥250 μg in patients ≥65 years	No. of patients ≥65 years prescribed digoxin at daily dose ≥250 μg	No. of patients aged ≥65 years prescribed digoxin
Methotrexate 2.5 mg and 10 mg coprescription	No. of patients coprescribed 10 mg and 2.5 mg methotrexate	No. of patients prescribed methotrexate
Patients with at least one HRP (all indicators)	No. of patients with at least one HRP	No. of patients at risk of an HRP

ACE, angiotensin‐converting enzyme inhibitor; ARB, angiotensin receptor blocker; HRP, high‐risk prescription; No., number; NSAID, nonsteroidal anti‐inflammatory drug;

^aAspirin or clopidogrel

^bMacrolides, quinolones or metronidazole

An overall composite indicator, defined as whether or not the patient had received any high‐risk prescription (for all indicators) 1, was used to assess variation in HRP between individual GPs, and in the dispensing of high‐risk prescriptions between pharmacies, in 2015. Variation between GPs, after controlling for patient‐level variables, was examined using a funnel plot of the ratio of observed/expected numbers of patients with a high‐risk prescription by expected numbers for each GP 22. The expected number was calculated from the multilevel logistic regression model predicting HRP according to age, gender and number of chronic medications used in all patients. Variation between pharmacies, after controlling for patient‐level variables and the prescribing GP, was also examined using a funnel plot. GPs and pharmacies between 2 and 3 standard deviation (SD) limits were statistically significantly different from the average (P < 0.05), and GPs and pharmacies outside the 3 SD limits were highly statistically significantly different from the average (P < 0.005).

Statistical analyses

For each indicator, the proportion of patients receiving a high‐risk prescription was calculated with 95% confidence intervals (CI). Multilevel logistic regression was used to examine the association between receipt of at least one high‐risk prescription (for all indicators) and age, gender and number of chronic medications while adjusting for the clustering of patients within the GP, or the individual GP, or the prescribing GP. The following patient variables were included: age at first dispensed medicine [categorized into 16–39 (reference), 40–49, 50–59, 60–69, 70–79 and ≥80 years], gender [female (reference), male) and number of coprescribed chronic medications over the year [defined as the number of distinct second‐level ATC codes (e.g. A03), relating to only the following first‐level codes: A, B, C, G, H, L, M, N, R and S and excluding those in the denominator of the HRP indicators]. Chronic medication use was defined as at least three prescription items dispensed in the year for each drug class 17. The comorbidity variable was categorized as 0–2 (reference), 3–4, 5–6, 7–8, 9–10 and ≥11 distinct drug classes. A random intercept for the GP (level 2) variance was included. Adjusted odds ratios (ORs) and 95% CIs for the patient‐level variables and the error variance for the level 2 intercept were computed. Statistical significance at P < 0.05 was assumed. Statistical analyses were conducted using SAS® v 9.4 (SAS Institute Inc., Cary, NC, USA) and Stata v 14.0 (StataCorp, College Station, TX, USA).

Results

Figure 1 and Table S1 show the prevalence of high‐risk prescriptions for each indicator, and the composite indicator, for each year from 2011 to 2015. In total, 66 022 of 300 906 patients at risk in 2011 (21.9%, 95% CI 21.8%, 22.1%), and 42 109 of 278 469 patients at risk in 2015 (15.1%, 95% CI 15.0%, 15.3%), received at least one high‐risk prescription based on the indicators used in the present study. Overall, in 2015, the majority of patients with HRP (36 924 or 87.7% of those with any high‐risk prescription) triggered only one indicator, 4489 (10.7%) triggered two, and 696 (1.7%) triggered three or more.

Figure 1.

Prevalence of high‐risk prescriptions (with 95% confidence intervals) for each indicator per year from 2011 to 2015. Antiplatelet agents comprised aspirin or clopidogrel; high‐risk antibiotics comprised macrolides, quinolones or metronidazole. ACE, angiotensin‐converting enzyme inhibitor; ARB, angiotensin receptor blocker; BB, beta‐blocker; HRP, high‐risk prescription; NSAID, nonsteroidal anti‐inflammatory drug

A comparison of the prevalence of high‐risk prescriptions for each indicator in 2011 vs. 2015 is provided in Table 2. The four indicators with the highest rates of HRP in both 2011 and 2015 were: (i) prescription of an NSAID to a patient aged ≥75 years without gastroprotection (47.9% in 2011; 37.2% in 2015); (ii) prescription of an antiplatelet agent to a current warfarin user (21.8% in 2011; 19.5% in 2015); (iii) prescription of a high‐risk antibiotic to a current warfarin user (17.8% in 2011; 16.2% in 2015); and (iv) prescription of digoxin at a daily dose of ≥250 μg to a patient aged ≥65 years (17.3% in 2011; 14.0% in 2015).

Table 2.

Comparison of the prevalence of high‐risk prescriptions in 2011 vs. 2015

Prescribing safety indicator	No. of patients receiving HRP/No. of patients at risk [% (95% CI)]		Z‐statistica	P‐valuea
	2011	2015
NSAID prescribed in patients ≥65 years currently using ACE/ARB and diuretic	13 610/86921 15.66 (15.42, 15.90)	6886/72613 9.48 (9.27, 9.70)	37.57	<0.001
NSAID prescribed in patients ≥75 years without gastroprotection	30 299/63313 47.86 (47.47, 48.25)	18 486/49750 37.16 (36.73, 37.58)	36.41	<0.001
NSAID prescribed to current warfarin user	7205/45808 15.73 (15.40, 15.90)	3490/35953 9.71 (9.40, 10.01)	26.08	<0.001
Antiplatelet agent b prescribed to current warfarin user	9987/45808 21.80 (21.42, 22.18)	7013/35953 19.51 (19.10, 19.92)	8.07	<0.001
High‐risk antibiotic c prescribed to current warfarin user	8129/45808 17.75 (17.40, 18.10)	5807/35953 16.15 (15.77, 16.53)	6.05	<0.001
Oral azole antifungal agent prescribed to current warfarin user	890/45808 1.94 (1.82, 2.07)	746/35953 2.07 (1.93, 2.22)	−1.33	0.183
Verapamil/diltiazem prescribed to current beta‐blocker user	3797/202095 1.88 (1.82, 1.94)	2681/194692 1.38 (1.33, 1.43)	12.51	<0.001
Digoxin prescribed at daily dose ≥250 μg in patients ≥65 years	3556/20544 17.31 (16.79, 17.83)	2017/14453 13.96 (13.39, 14.52)	8.58	<0.001
Methotrexate 2.5 mg and 10 mg coprescription	1701/10137 16.78 (16.05, 17.51)	945/11980 8.02 (7.54, 8.51)	19.96	<0.001
Patients with at least one HRP (all indicators)	66 022/300906 21.90 (21.79, 22.09)	42 109/278469 15.10 (14.99, 15.25)	67.19	<0.001
Total No. of patients dispensed medications	1 263 395	1 263 694
No. of patients with HRP/No. of patients dispensed medications	66 022/1263395 5.23 (5.19, 5.26)	42 109/1263694 3.33 (3.30, 3.36)	74.45	<0.001

ACE, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; CI, confidence interval; HRP, high‐risk prescription; No., number; NSAID, nonsteroidal anti‐inflammatory drug

^aFisher's exact test

^bAspirin or clopidogrel

^cMacrolides, quinolones or metronidazole

There were significant reductions in the rates of most of the HRP indicators over time (Table 2). The largest reductions from 2011 to 2015 were for prescription of an NSAID to a patient aged ≥75 years without gastroprotection (10.7% decrease, P < 0.001), coprescription of methotrexate 10 mg and 2.5 mg (8.8% decrease, P < 0.001), prescription of an NSAID to a patient aged ≥65 years and currently using an angiotensin‐converting enzyme (ACE) inhibitor or angiotensin receptor blocker (ARB) and a diuretic (6.2% decrease, P < 0.001), and prescription of an NSAID to a current warfarin user (6.0% decrease, P < 0.001).

Patient characteristics associated with HRP

Table 3 shows the prevalence of patients receiving at least one high‐risk prescription by a range of patient characteristics, and adjusted ORs derived from the multilevel logistic regression model, for 2015. Any HRP increased noticeably with age, rising progressively from 2.5% in patients aged <40 years to 20.2% in patients aged ≥80 years (adjusted OR 8.64, 95% CI 7.69, 9.71). Any HRP also increased progressively with the number of chronic drugs used, with 10.6% of patients prescribed 0–2 chronic drugs receiving any high‐risk prescription compared with 19.0% of those prescribed ≥11 chronic drugs (adjusted OR 1.35, 95% CI 1.29, 1.40). Females were slightly more likely to receive any HRP than males (15.2% vs. 15.1% in men, adjusted OR 1.06, 95% CI 1.04, 1.08).

Table 3.

Prevalence of patients receiving any high‐risk prescription in 2015, and multilevel odds ratios (both unadjusted and adjusted) for the clustering of patients within general practitioners, or within individual general practitioners

Patient‐level fixed effects (n)	Any HRP [% (95% CI)]	Multivariate odds ratio (95% CI)
		Unadjusted	Adjusted
Age (years):
<40 (12 440)	2.5 (2.3, 2.8)	1.00	1.00
40–49 (13 006)	4.1 (3.8, 4.5)	1.58 (1.37, 1.82)	1.56 (1.35, 1.80)
50–59 (23 342)	4.9 (4.6, 5.2)	2.84 (1.62, 2.09)	1.81 (1.59, 2.06)
60–69 (51 396)	9.7 (9.4, 10.0)	3.81 (3.38, 4.28)	3.73 (3.31, 4.20)
70–79 (100 574)	19.4 (19.2, 19.6)	8.48 (7.56, 9.51)	8.33 (7.42, 9.36)
≥80 (77 711)	20.2 (19.9, 20.5)	8.77 (7.82, 9.84)	8.64 (7.69, 9.71)
Gender:
Female (151 258)	15.2 (15.0, 15.4)	1.00	1.00
Male (127 211)	15.1 (14.9, 15.3)	1.07 (1.04–1.09)	1.06 (1.04, 1.08)
Number of chronic drugs:
0–2 (43 326)	10.6 (10.3, 10.9)	1.00	1.00
3–4 (52 826)	13.0 (12.7, 13.3)	1.05 (1.00, 1.09)	1.05 (1.01, 1.10)
5–6 (57 940)	14.9 (14.6, 15.2)	1.12 (1.07, 1.16)	1.12 (1.08, 1.17)
7–8 (47 741)	16.5 (16.2, 16.8)	1.19 (1.14, 1.23)	1.20 (1.15, 1.25)
9–10 (33 621)	17.5 (17.1, 17.9)	1.22 (1.17, 1.28)	1.23 (1.18, 1.29)
≥11 (43 015)	19.0 (18.6, 19.4)	1.33 (1.28, 1.39)	1.35 (1.29, 1.40)
Error variance (level 2 intercept)		0.091 (SE 0.005, P < 0.05)

CI, confidence interval; HRP, high‐risk prescription; n, number of patients; SE, standard error

Variation in HRP between GPs

There was considerable variation in the overall rate of HRP between individual GPs in 2015, ranging from 1.2% of patients receiving any high‐risk prescription to 36.0% (n = 2045). After controlling for patient‐level variables, significant variation between GPs remained (level 2 intercept 0.091, standard error 0.005, P < 0.05). Figure 2 shows the difference between the observed numbers of patients with a high‐risk prescription and those expected for each GP, after adjusting for patient case mix (age, gender and number of chronic drugs used). GPs varied from having 50% less than expected rates to 100% in excess. Just over 5% of GPs had significantly higher than average rates of HRP (P < 0.05), and just 1.2% were above the 3 SD limit (P < 0.005).

Figure 2.

Funnel plot of the ratio of observed/expected number of patients (as a percentage) vs. expected number of patients with a high‐risk prescription for each general practitioner in 2015 (n = 2045). SD, standard deviation

Variation in dispensing of high‐risk prescriptions between pharmacies

There was considerable variation in the overall rate of dispensing of high‐risk prescriptions between pharmacies in 2015 (n = 1788). Figure 3 shows the difference between the observed numbers of patients with a high‐risk prescription and those expected for each pharmacy, after adjusting for patient case mix (age, gender and number of chronic drugs used) and the prescribing GP. Pharmacies varied from having 100% less than expected rates to 100% in excess. Approximately 3.6% of pharmacies had significantly higher than average rates of dispensing of high‐risk prescriptions (P < 0.05), and just 0.7% were above the 3 SD limit (P < 0.005).

Figure 3.

Funnel plot of the ratio of observed/expected number of patients (as a percentage) vs. expected number of patients with a high‐risk prescription for each pharmacy in 2015 (n = 1788). SD, standard deviation

Discussion

Overall, according to the indicator set used in the present study, the prevalence of HRP in Ireland appears to be decreasing over time, with significant reductions in the rate of high‐risk prescriptions for eight of the nine indicators examined between 2011 and 2015. Despite this, the prevalence of HRP remained high, with 15% of patients vulnerable to ADEs receiving one or more high‐risk prescriptions in 2015. Further, the rates of four of the nine indicators persisted above 13% of those at risk in 2015 (prescription of an NSAID in >75‐year‐olds without gastroprotection, coprescription of an antiplatelet agent or high‐risk antibiotic with warfarin, and prescription of high‐dose digoxin).

The persistently high rate of NSAID prescription without gastroprotection in older patients (≥75 years) is particularly of interest (48% in 2011 and 37% in 2015). Upper gastrointestinal (GI) haemorrhage is a major cause of morbidity, mortality and healthcare costs 23, and advanced age is a significant risk factor for serious upper GI events with NSAID use 24, 25, 26. Considering that proton pump inhibitors (PPIs) are well tolerated 23, 24 and very effective in the prevention of NSAID‐related upper GI complications 25, 26, 27, 28, 29, strategies to improve PPI prescribing in older patients requiring NSAID therapy are clearly needed. There were also relatively high rates of warfarin coprescription with an antiplatelet agent or high‐risk antibiotic. Combination therapy of warfarin and an antiplatelet agent may be indicated for patients with certain concomitant diseases, such as atrial fibrillation, thromboembolic disease or the presence of an artificial cardiac valve requiring warfarin with concomitant coronary artery disease, or peripheral vascular disease requiring treatment with an antiplatelet agent. Therefore, in some of these cases, dual therapy may have been justified in terms of benefit outweighing the risk of bleeding 30, 31. The prescription of a high‐risk antibiotic to a warfarin user may also be justified if a particular causative organism is susceptible only to the high‐risk antibiotic; however, there is more choice in terms of antibiotic therapy and in many of these cases there may have been options to use a low‐risk antibiotic to treat the infection. Exposure to high‐risk antibiotics has been shown to increase the risk of serious bleeding events by almost 50% when compared with low‐risk antibiotics in warfarin users 32. It is also possible that the international normalized ratio (INR) was closely monitored in those patients receiving warfarin, which may help to decrease the risk of serious bleeding events 32. The relatively high rate of high‐dose digoxin prescribing in older patients is also concerning. Older patients have a reduced capacity for elimination of digoxin, so if it is to be used, conservative doses are advised 33. Low digoxin doses (≤125 μg day^–1) have been shown to be effective in older patients, and the risk of digoxin toxicity is reduced 34.

Only one previous study of HRP was appropriate for comparing the prevalence of specific indicators with those in the current study, although this previous study of 315 general practices in Scotland was based on 2007 data 1. There was a higher prevalence of six of the seven comparable indicators measured in 2011 in the current study than in the earlier Scottish study. Coprescription of an NSAID with warfarin was nearly fivefold higher in the current study compared with the Scottish study (current study 15.7% vs. Scottish study 3.4%). Coprescription of an antiplatelet agent with warfarin was approximately double (21.8% vs. 9.6%), as too were coprescription of a high‐risk antibiotic with warfarin (17.8% vs. 7.9%), coprescription of an azole antifungal with warfarin (1.9% vs. 0.7%) and coprescription of an NSAID, ACE/ARB and diuretic in patients 65 years and over (15.7% vs. 8.8%). Coprescription of methotrexate 10 mg and 2.5 mg was also higher in 2011 in the current study compared with the Scottish study (16.8% vs. 11.8%). By contrast, the prevalence of prescription of an NSAID without gastroprotection in patients over 75 years was slightly lower in the current study (47.9% vs. 50.5%). These differences may be partly due to the different type of datasets and populations analysed in each study. The Scottish study used a primary care database, whereas the current study was based on a pharmacy claims database, which over‐represents older persons and more socially deprived individuals and may have led to the higher rates observed.

Consistent with previous studies, increasing age and number of chronic drugs used were associated with increasing rates of HRP in multilevel logistic regression analysis, although the association of age was considerably stronger in the current study than in previous studies in the UK [e.g. adjusted OR 8.5 for age 70–79 years compared with <40 years (current study) vs. adjusted OR 1.18 (comparing age 70–79 years with <40 years) and adjusted OR 2.84 (comparing age 71–80 years with 18–50 years) in previous UK studies] 1, 11. These previous studies found polypharmacy to be more strongly associated with HRP than age in multivariate analysis [e.g. adjusted OR 1.33 for ≥11 chronic drugs compared with 0–2 (current study) vs. adjusted OR 7.90 (comparing 0 chronic drugs with ≥11) and adjusted OR 9.40 (comparing 0–1 chronic drugs with >10) in previous UK studies] 1, 11. Importantly, older patients, and patients with a high comorbidity burden, all appear to be particularly vulnerable to HRP.

The high variation in the prevalence of any HRP in those at risk between individual GPs suggests that there are opportunities for improving prescribing safety in primary care 1. Information on the practice in which the GP operates was not available in the current study but may, in part, explain the variation between GPs. Previous studies have suggested that differences between practices may be an important source of variation, although practice‐level variables had much less influence than patient‐level variables in those studies 1, 11. Further, in a study examining the extent of variation between individual GPs and between the practices in which they work, much greater variation in HRP was observed between GPs than between practices 35. Thus, interventions aimed at improving safety in primary care prescribing should not only target practices, but also encourage practices to investigate variation between individual GPs in the practice 35. An important role of the pharmacist is to identify and act on drug‐related problems to ensure patient safety. The high prevalence of high‐risk prescriptions being dispensed, and the variation in the dispensing of high‐risk prescriptions between pharmacies, suggest that interventions to improve medication safety in the community should also target community pharmacists and the pharmacies in which they work.

Studies of interventions to improve prescribing safety in primary care are limited and have largely focused on pharmacist‐led interventions, with variable effectiveness 36, 37. A complex pharmacist‐led information technology intervention (PINCER), including educational outreach, feedback and dedicated support, significantly reduced a range of medication errors in UK general practices with computerized records 38. More recently, a complex intervention combining professional education with informatics to facilitate review, and financial incentives to review Data‐driven Quality Improvement in Primary care (DQIP), significantly reduced the rate of targeted HRP in primary care in the UK, and was associated with significant reductions in the rate of hospital admissions for GI ulcer or bleeding and heart failure but not for acute kidney injury 39. However, implementing complex interventions on a large scale is relatively difficult and costly 16. Most recently, a randomized controlled trial conducted in Scotland Effective Feedback to Improve Primary Care Prescribing Safety (EFIPPS) found that a simple intervention involving feedback of prescribing safety data was effective at reducing HRP relating to the use of antipsychotic agents, NSAIDs and antiplatelet agents 16. As electronic medical records or pharmacy claims data are increasingly used in primary care, this type of intervention may be feasible to implement at scale, and the potential for improving prescribing safety with this intervention is promising 16.

Strengths and limitations

The present study was a large national study of prescribing safety in primary care over consecutive years in Ireland, using multiple indicators based on explicit prescribing safety advice previously described 1, 15. The study design enabled examination of the variation in HRP between patients and GPs, and dispensing of high‐risk prescriptions between pharmacies, as well as trends in HRP over time, in a highly representative population.

We acknowledge the following limitations. Firstly, HRP as defined by the indicators, may not always be inappropriate. The prescription may be high risk but still justified for clinical reasons, after balancing risk and benefit in conditions of uncertainty 1, 11. Nevertheless, given that all the prescribing indicators examined in the study are stated as being contraindicated or to be avoided in routine practice in UK national guidance, the observed high prevalence of, and large variation in, HRP should be consistent with a significantly appropriate proportion of prescribing examined 1.

Another notable limitation involves the constraints of using only pharmacy claims data, which does not include pre‐existing disease or monitoring information. As such, the indicator set employed in the current study is not comprehensive and only represents a selection of measures of HRP. For example, indicators relating to warfarin monitoring, diabetic patients, heart failure patients, dementia patients or patients with peptic ulcer disease could not be assessed. It is therefore likely that the present study substantially underestimated the true prevalence of HRP by GPs. Additionally, the pharmacy claims database does not include drugs purchased over the counter, such as NSAIDs and gastroprotective drugs. However, the GMS scheme provides these items for free (or for a small copayment) on prescription and therefore the risk of bias is expected to be low.

Finally, although the study identified significant variation in HRP between individual GPs, it did not examine variation between practices. Therefore, it could not distinguish between practices with individual prescribers who were more likely than average prescribers to engage in HRP and practices with generalized HRP, and both may be important 1. However, prescribing is more likely to be an individual rather than practice responsibility 11.

Implications for policy and practice

The results of the present study highlighted several areas where improvement in primary care prescribing safety is needed. Levels of NSAID prescription without gastroprotection in older patients, coprescription with anticoagulants and high‐dose digoxin prescribing remain high, so more intensive interventions to alert prescribers and community pharmacists to the risks associated with these drugs are needed. Due consideration should also be given to the risks of prescribing multiple medications and the importance of regular medication reviews 11, particularly for older patients.

A composite indicator, such as that used in the current study, could be used to identify GPs with high rates of HRP, and pharmacists with high rates of dispensing high‐risk prescriptions, for the purposes of clinical governance and performance management 1. For instance, in the UK, the Quality Outcomes Framework incentives scheme provides UK GPs with financial incentives based on performance indicators 11.

Conclusions

The prevalence of HRP in Ireland appears to have declined over time, although there are some indicators, such as those associated with NSAID prescription, anticoagulant coprescription and high‐dose digoxin, which still persist. The high variation observed between GPs, and between pharmacies, suggests that there is potential for improvement in primary care prescribing and dispensing safety. Future research should focus on interventions to reduce HRP and the dispensing of high‐risk prescriptions in primary care, particularly in vulnerable older populations.

Competing Interests

All authors declare no competing interests.

This research was supported by the Health Research Board Ireland (RL‐2015‐1579). The authors acknowledge the Health Service Executive Primary Care Reimbursement Service (HSE‐PCRS) for providing access to the administrative pharmacy claims data used in this study.

Contributors

K.B. conceived and designed the study. Data were acquired by K.B. (Health Service Executive Primary Care Reimbursement Service). K.B. carried out the statistical analyses. K.B. and C.B. interpreted the data. The manuscript was drafted by C.B. and all authors were involved in the critical revision of this and approval of the final manuscript for submission.

Supporting information

Table S1 Prevalence of patients receiving high‐risk prescriptions for each indicator in the years 2011–2015

Byrne, C. J. , Cahir, C. , Curran, C. , and Bennett, K. (2017) High‐risk prescribing in an Irish primary care population: trends and variation. Br J Clin Pharmacol, 83: 2821–2830. doi: 10.1111/bcp.13373.