FormalPara Key Points

This study highlights the prevalence of polypharmacy, increasing medication use over time, and possible overuse of some medicines with known risk of potential harm in cirrhosis.

General practitioners prescribe a greater proportion of ‘unsafe’ and ‘safety unknown’ medicines compared with consultants/specialists, and should be supported in the community with prescribing guidance and education.

These observational data provide valuable insight about prescribing patterns in the ‘real world’, and may be useful to inform future medication safety initiatives.

1 Introduction

Medication safety is an important but underrecognized opportunity to improve health outcomes for people with cirrhosis. Patients often see multiple healthcare providers and take many medicines to manage complications of decompensated cirrhosis including ascites, hepatic encephalopathy (HE) and variceal bleeding. Comorbidity burden in this group is also rising, particularly the prevalence of type 2 diabetes, renal disease and congestive heart failure [1], which often require additional pharmacotherapy.

Achieving optimal medication management in people with cirrhosis can be complex [2]. Progressive pathophysiological changes often result in alterations to drug pharmacokinetic (PK) disposition (e.g. absorption, distribution, metabolism or excretion) and/or pharmacodynamic (PD) response [3]. This can lead to adverse drug reactions, drug–drug interactions and drug–disease interactions with existing complications [4,5,6]. While the risk of medication-related harm is often dependent on the presence of clinical and patient-specific factors [2, 7], several commonly used medicines have well-recognised risks of precipitating adverse drug reactions or decompensation events in patients with cirrhosis. For example, non-steroidal anti-inflammatory drugs (NSAIDs) are often avoided in cirrhosis due to an increased risk of acute kidney injury and gastrointestinal bleeding, related to reduced prostaglandin synthesis [8, 9]. Impaired metabolism and clearance of opioids may lead to sedation, constipation and precipitation of HE [10,11,12]. Benzodiazepines and derivatives of the neurotransmitter gamma-aminobutyric acid, namely pregabalin and gabapentin, are also associated with an increased risk of HE [11, 13]. Among patients with cirrhosis and ascites, the use of proton pump inhibitors (PPIs) may be a risk factor for the development of HE and spontaneous bacterial peritonitis [14, 15], potentially due to changes in gut microbiota [16]. Use of angiotensin therapies (angiotensin II receptor blockers and angiotensin-converting enzyme inhibitors) by patients with ascites may also increase the risk of renal dysfunction due to progressive alteration in renal haemodynamics [17, 18]. Although there may be clinical indications for the use of these medicines, they should be prescribed judiciously with awareness of and monitoring for the risk of medication-related harm.

Non-hepatology specialists and general practitioners (GPs) are increasingly relied upon to manage day-to-day pharmacotherapy for multi-morbid patients with chronic liver disease. However, many GPs report uncertainty about managing patients with cirrhosis due to a lack of subspecialty knowledge [19]. To address gaps in non-specialist knowledge and develop specific guidance and education programmes aimed at preventing medication-related harm, greater understanding of current prescribing patterns and prevalence of medication use in the ‘real world’ is required.

The publicly funded universal healthcare system in Australia provides citizens and eligible residents with subsidised prescription medicines through the Pharmaceutical Benefits Scheme (PBS) [20]. Analysis of dispensing claims data can provide information about the prevalence of medication use in Australia. In this study, we aimed to characterise the medicines dispensed for Australians with cirrhosis and explore changes in use of medication groups over time. We also sought to assess the use of ‘potentially inappropriate’ medicines, especially in people with decompensated cirrhosis, for the purpose of tailoring future prescribing guidance and education programmes for non-hepatology specialists and GPs.

2 Patients and Methods

2.1 Study Population

A retrospective cohort study was conducted among CirCare participants. CirCare is a prospective multi-centre observational study of adults with cirrhosis who were recruited from five hospitals in the cities of Brisbane and Logan, Queensland, between July 2016 and December 2018. The study has been described previously [21, 22]. Briefly, consecutive adult patients with cirrhosis were recruited from the Royal Brisbane and Women’s, Prince Charles, Mater, Logan, or Princess Alexandra Hospitals when they attended hepatology/gastroenterology outpatient clinics, or were admitted for management of cirrhosis-related conditions. Patients were not invited to participate if the treating medical team advised of a cognitive or physical impairment that could interfere with participation (e.g. intellectual impairment, severe HE, admitted to the intensive care unit), or if an interpreter was not available and the patient was unable to communicate in English.

Comprehensive clinical and sociodemographic data for participants were obtained via recruitment interview, questionnaire tools, medical record review and linkage to the Queensland Hospital Admitted Patient Data Collection (QHAPDC), and the national PBS and Medicare Benefits Scheme (MBS). All patients received comprehensive review for decompensation complications at recruitment. Ethical approval for the CirCare study was obtained from the Human Research Ethics Committees of the Metro South Health (HREC/16/QPAH/628) and QIMR Berghofer Medical Research Institute (P2207).

2.2 Medication Data

For consenting CirCare participants, PBS data were extracted for all medicines dispensed between 1 January 2016 and 30 June 2020. The final dataset included prescriptions dispensed from cirrhosis diagnosis (or recruitment date if diagnosis date was unknown) until liver transplant or death (Fig. 1). PBS data included information about medicines eligible for subsidy that were dispensed by hospital and community pharmacies in Australia, including date of prescription and supply, drug name, strength, Anatomical Therapeutic Chemical (ATC) classification and patients’ reimbursement concession tier. PBS data did not include medications supplied by ‘private’ prescriptions or those supplied under the Repatriation Pharmaceutical Benefits Scheme; however, these comprise a minority of medicines expenditure in Australia [23, 24]. Medicines administered to inpatients at public hospitals, ‘over-the-counter’ medicines (including lactulose), vitamins (including thiamine and vitamin D) and complementary medicines that participants took without a prescription are not represented.

Fig. 1
figure 1

Study data timeline. Patients recruited to CirCare between July 2016 and December 2018 consented to extraction of all PBS and MBS data from 1 January 2016 to 30 June 2020. Patients contributed prescription data from cirrhosis diagnosis (or recruitment date if diagnosis date was unknown) until liver transplant or death. Example data timeline 1: patient with cirrhosis diagnosis (C1) prior to January 2016 who received a liver transplant (T1) in October 2019 – prescriptions from 1 January 2016 until date of liver transplant included in the final dataset. Example data timeline 2: patient with unknown cirrhosis diagnosis date was recruited (R2) in February 2017, and deceased (D2) after June 2020 – prescriptions from recruitment date until 30 June 2020 included in the final dataset. Example data timeline 3: patient with cirrhosis diagnosis (C3) in November 2016, and deceased (D3) in March 2018 – prescriptions from cirrhosis diagnosis until death included in the final dataset

Prescriber information was not available for our dataset. We inferred prescriber specialty (GP, consultant/specialist or ‘other’) via linkage with MBS data, which contains information about the date and type of subsidised medical services provided to eligible Australian citizens and residents. Linked MBS data was available from 1 January 2016 to 30 June 2020. Prescriptions written on the same day that patients received medical services from a provider with prescribing rights in Australia were assumed to be written by that provider. For prescriptions without aligning MBS data, we accessed QHAPDC (linked data available up to August 2019) to identify prescriptions written for hospital discharge, day admission, outpatients or in the emergency department, as these prescriptions are PBS eligible [23]. Prescriptions written during hospital encounters were assumed to be written under the authority of the treating consultant/specialist.

2.3 Measures

Severity of liver disease was classified by absence (compensated) versus presence (decompensated) of cirrhosis complications (ascites, hepatic encephalopathy, variceal bleeding or jaundice) at recruitment. Comorbidity burden was measured using the Charlson Comorbidity Index (CCI). Place of residence was classified using the Accessibility/Remoteness Index of Australia and the Index of Relative Socioeconomic Advantage and Disadvantage (where fourth and fifth quintiles represent ‘most disadvantaged’ areas) [25, 26].

Polypharmacy was defined as taking five or more unique medicines. Evidence-based recommendations [27] founded on PK/PD literature and clinical expertise were used to categorise medicines into safety categories; namely ‘safe’, ‘no additional risks known’, ‘additional risks known’, ‘safety unknown’ (e.g. insufficient PK/PD data to make a recommendation) and ‘unsafe’ using the Child–Pugh score. Medicines of clinical safety interest were also examined. Determining the indication for use and patient-specific risk factors for harm was outside the scope of this study.

2.4 Data Analysis

Characteristics of prescriptions and study participants are described as frequency (percentages, %), mean ± standard deviation (SD) or median [interquartile range (IQR)]. Differences between groups were analysed using Pearson’s chi-squared test (or Fisher’s exact test if expected cell counts were ≤ 5) for categorical variables, the independent samples t-test for continuous (normally distributed) variables and the Mann–Whitney U test for continuous (non-normally distributed) variables. Normality was assessed using normal curve histograms and the Shapiro–Wilk test. Data were analysed using IBM SPSS Version 28 and STATA SE 17.0.

Participants were censored at death, following receipt of a liver transplant or on 30 June 2020, whichever came first. Medications were defined using the ATC classification and grouped according to therapeutic class and/or clinical indication. Pattern of use was analysed in 6-monthly time intervals. Prevalence of medication use (period prevalence) was calculated by dividing the number of patients who were dispensed a particular medication (or group) at least once during a given time period by the total number of patients observed in that period (alive and included for at least 1 day). ‘Long-term’ medication use was defined as having a medicine dispensed at least twice in each of two consecutive time periods. This definition was based on the ‘PBS maximum quantity of dispensed units’ for chronic medicines, which typically covers 1–2 months of supply for a patient (e.g. most people with good adherence would fill three to six prescriptions during each 6 month time period). To capture non-persistent chronic medication use, we chose a conservative estimate of two prescriptions in each of two consecutive time periods, which represented at least four prescriptions dispensed over 12 months. This measure also captured intermittent use of opioids, NSAIDs and benzodiazepines, which have a typical ‘PBS maximum quantity of dispensed units’ to cover less than 1 month.

Generalised estimating equations (GEE) models were used to estimate the change in consumption of medicines among patients over time. The Quasi-likelihood under Independence model Criterion (QIC) was used to measure goodness of fit for the final multi-variable models. Linear combination of estimators was used to estimate the likelihood of medicine consumption at specific time points between compensated versus decompensated patients.

Change in the number of medicines dispensed was analysed using a GEE model with negative binomial distribution. Age, aetiology, CCI, cirrhosis stage and time period were included in the final model. A GEE model with binary logistic distribution was used to analyse the change in consumption of specific medications (taking versus not taking) over time. Cirrhosis stage, time period and cirrhosis stage × time period were included in the final model. A sensitivity analysis to evaluate the effect of age, aetiology and CCI in the multi-variable model found no difference in tested values in terms of magnitude or variation above or below the significance threshold (α = 0.050). However, the larger model encountered computational problems for medicines with lower consumption rates, and estimates could not be presented.

3 Results

3.1 Patient Characteristics

Of 746 patients invited to participate in the CirCare study, 581 were interviewed (77.9% response), of whom 529 (70.9%) gave consent for researchers to obtain their PBS records. Five PBS consent forms were incomplete, one patient was ineligible to receive PBS benefits and one patient was excluded as they were hospitalised (and unable to receive PBS benefits) from cirrhosis diagnosis to death. Complete PBS data for 522 patients (70.0%) are described hereafter. A flow diagram depicting recruitment and case inclusion is provided in the Supplementary Figure.

Patients’ mean age at recruitment was 59.7 ± 10.9 years, 69.7% were male, 72.0% were born in Australia, and 35.7% lived in ‘most disadvantaged’ areas. Two-thirds of participants had compensated cirrhosis at recruitment (Table 1). Alcohol-related liver disease was more prevalent among patients with decompensated cirrhosis, whereas non-alcoholic fatty liver disease (NAFLD)/non-alcoholic steatohepatitis (NASH) and metabolic comorbidities were more prevalent among those with compensated cirrhosis. Unsurprisingly, a greater proportion of decompensated patients were recruited during an inpatient admission.

Table 1 Clinical and demographic characteristics of the study population at recruitment

Over half the participants (n = 310; 59.4%) had a diagnosis of cirrhosis made on or before 1 January 2016, 92 (17.6%) were diagnosed during 2016, 79 (15.1%) during 2017 and 41 (7.9%) during 2018. During the follow-up period, 109 patients (20.9%) died and 24 patients (4.6%) received a liver transplant. Decompensated patients had a shorter median follow-up [40 months (IQR 28–53; range 1–53 months)] compared with compensated patients [53 months (IQR 40–53; range 2–56 months] (p < 0.001).

3.2 Prescription Prevalence

A total of 89,615 prescriptions were dispensed to participants during the follow-up period, representing a median of 136 (IQR 62–237) prescriptions and a median of 16 (IQR 11–23) unique medicines per patient (total n = 9306 medicines). PPIs were the most prevalent medication class, accounting for 10,000 individual prescriptions (11.2%) and dispensed at least once for 383 (73.4%) patients. Opioids, angiotensin therapies and antidepressant prescriptions were also highly prevalent (Table 2). Period prevalence for medicines dispensed in each time period, including among people with compensated versus decompensated cirrhosis, are presented in the Supplementary Table.

Table 2 Commonly dispensed prescriptions and prevalence of medication use among people with cirrhosis

Approximately one-third of medicines (30.6%; n = 2844) were dispensed to patients once during the follow-up period, 13.9% (n = 1292) were dispensed twice, 17.2% (n = 1600) were dispensed three to five times and 38.4% (n = 3570) were dispensed six or more times. Among those medicines only dispensed once (n = 2844), the most common were antibiotics [22.5% (n = 640), predominantly amoxycillin ± clavulanate (6.6%; n = 188)] and opioids [10.3% (n = 294); predominantly oxycodone (4.7%; n = 133)]. Among those medicines dispensed six or more times (n = 3570), PPIs [9.6% (n = 343), predominantly pantoprazole (4.9%; n = 176)], diabetes medicines [8.8% (n = 313), predominantly metformin (4.6%; n = 166)] and diuretics [8.2% (n = 292), predominantly spironolactone (4.1%; n = 146)] were the most common.

GPs prescribed (initiated or repeated a medicine initiated by a specialist) 69.2% of dispensed prescriptions during the study period and specialists prescribed 10.4%. ‘Others’ (e.g. trainee clinicians, nurse practitioners, optometrists) prescribed 2.1%, and 1.3% were prescribed on a day that patients saw multiple healthcare providers (e.g. GP and specialist) and prescriber identity could not be confirmed. Prescriber information could not be linked for 17.0% of prescriptions. These may have been written prior to 1 January 2016 (MBS data not available), during an inpatient or outpatient encounter at a private hospital, during a public hospital encounter after August 2019 (QHAPDC data not available) or in the course of clinical care without a claimable MBS consultation on the date the prescription was written. Of the total 5170 unique medicines dispensed three or more times to patients during the study period, 18.5% were prescribed at least once by both a GP and consultant/specialist, 59.7% were only linked to a GP prescriber, 9.9% were only linked to a consultant/specialist prescriber, and no prescriber data could be linked for 11.9%.

3.3 Medicine Consumption Over Time

Polypharmacy was prevalent, with approximately two-thirds (59.4–68.6%) of observed participants in each time period dispensed five or more unique medicines (Supplementary Table). The median number of dispensed medicines increased from 6 (IQR 3–9) in January–June 2016 to 7 (IQR 3–9) in January–June 2020. Age, time period, NAFLD/NASH cirrhosis, comorbidity burden and cirrhosis stage were independently associated with a higher incidence rate of medication consumption (Table 3). The likelihood of taking common medicines also increased with each successive time period (Table 4), including PPIs (p < 0.001), opioids (p = 0.004), antidepressants (p = 0.003) and inhaled medicines (p = 0.008). NSAID use decreased over time (p = 0.040).

Table 3 Factors associated with the number of dispensed medicines over time
Table 4 Population likelihood of medication use over time

For some medicines, the pattern of use differed between patients with compensated and decompensated cirrhosis (Fig. 2). Decompensated patients experienced an increased (albeit non-significant) likelihood of spironolactone (OR 1.478, 95% CI 0.921–2.375; p = 0.106) and propranolol (OR 1.461, 95% CI 0.946–2.255; p = 0.087) dispensing over time. Among compensated patients, a significant increase in the likelihood of PPI (OR 1.418, 95% CI 1.166–1.724; p < 0.001) and statin (OR 1.361, 95% CI 1.095–1.691; p = 0.005) dispensing was observed, whereas decompensated patients were less likely to receive angiotensin therapies (OR 0.270, 95% CI 0.151–0.481; p < 0.001) and metformin (OR 0.364, 95% CI 0.188–0.703; p = 0.003) over time.

Fig. 2
figure 2

Proportion of patients who were dispensed: a medicines for decompensation complications, and b other medicines, where change in use differed between people with compensated and decompensated cirrhosis. Solid lines represent people with compensated (C) cirrhosis and broken lines represent people with decompensated (D) cirrhosis. p-Values (p ≤ 0.200 shown) were derived from GEE population estimates and represent within-group change in the likelihood of being dispensed a medicine in January–June 2020 compared with January-June 2016. *p-Value for rifaximin unable to be calculated due to estimation divergence (small n patients)

3.4 Prevalence of ‘Potentially Inappropriate’ Medication Use

A safety classification could be assigned to 59,961 (66.9%) dispensed medicines and one-third had no safety recommendation (not assessed by Weersink et al. [27]; Fig. 3). Among those medicines with a safety classification, 13.0% were ‘unsafe’ and 2.3% were ‘safety unknown’ in all stages of cirrhosis. A further 19.6% were ‘unsafe’ and 7.2% were ‘safety unknown’ in Child–Pugh B and/or C cirrhosis.

Fig. 3
figure 3

Safety classifications for dispensed medicines according to evidence-based (PK/PD) recommendations [27]. n = 29,654 dispensed medicines did not have a safety recommendation (white) and included n = 6610 antidepressants, n = 2292 benzodiazepines, n = 1946 gabapentinoids, n = 1261 anticoagulants, n = 3433 inhaled medicines, n = 1596 ophthalmological medicines and n = 1306 topical preparations

When stratified for individual patients’ severity of cirrhosis, the majority of dispensed medicines were ‘safe’ or ‘no additional risks known’ (78.2%; Table 5), while 14.2% (n = 8526) were classified ‘unsafe’ for the patient they were dispensed to [predominantly PPIs (n = 5134; 60.2%) and statins (n = 2220; 26.0%)] and 3.9% (n = 2315) were ‘safety unknown’ [predominantly antibiotics (n = 597; 25.8%) and fenofibrate (n = 617; 26.7%)]. GPs prescribed 71.5% of ‘unsafe’ and 73.1% of ‘safety unknown’ medicines. A greater proportion of total medicines prescribed by GPs were classified as ‘unsafe’ or ‘safety unknown’ compared with consultants/specialists (18.6% versus 12.1%; p < 0.001; Fig. 4). If all ‘prescriber unknown’ medicines were allocated to consultant/specialists, this significant difference was still observed.

Table 5 Commonly dispensed prescriptions in each patient-specific safety classification according to liver disease severity
Fig. 4
figure 4

Proportion of medicines prescribed by GPs, consultant/specialists and ‘others’ according to patient-specific safety classification (stratified for individual patients’ severity of cirrhosis). ‘Others’ included trainee clinicians (‘medical practitioners’ without a specialty qualification training under the supervision of a GP or consultant/specialist), nurse practitioners and optometrists

Figure 5 illustrates the proportion of patients who were dispensed medicines of clinical safety interest during the study period. Of concern, 90 (26.1%) compensated and 26 (14.7%) decompensated patients were dispensed NSAIDs at least once during the study period. PPI use was prevalent and statistically more frequent among decompensated patients (p < 0.001), whereas angiotensin therapies were more common among people with compensated cirrhosis (p < 0.001). Whilst overall opioid use was not significantly different between groups (p = 0.087), 62/129 users with decompensated cirrhosis took opioids long-term compared with 82/226 users with compensated cirrhosis (p = 0.030). Long-term use of angiotensin therapies (~80%), benzodiazepines (~33%), gabapentinoids (~56%) and PPIs (~73%) was equally common among those with and without decompensation.

Fig. 5
figure 5

Proportion of patients with compensated and decompensated cirrhosis who were dispensed medicines of clinical safety interest at least once during the follow-up period. The textured proportion of the bars represent long-term users, defined as having a medicine dispensed at least twice in each of two consecutive 6-monthly time periods. p-Values represent statistical difference between total users (Pearson’s Chi-square test). NSAIDs non-steroidal anti-inflammatory drugs, PPIs proton pump inhibitors

4 Discussion

Pharmacotherapy has an important role in the management of cirrhosis complications and comorbidities. However, little is known about the number and type of medicines dispensed to people with cirrhosis, as this predominantly occurs in the community. In this study of 522 Australians with cirrhosis (33.9% with decompensated disease at recruitment), high rates of prescription medication use were observed over a 4.5 year time period. The median number of dispensed medicines increased from 6 (IQR 3–9) in January–June 2016 to 7 (IQR 3–9) in January–June 2020, with a large proportion of patients dispensed antibiotics (88.9% at least once), PPIs (73.4%), opioids (68.0%), diuretics (62.8%) and beta-blockers (51.9%).

A key finding was the high prevalence of patients dispensed medicines with an increased risk of medication-related harm. PPIs were widely used in our cohort, consistent with reports from The Netherlands (period prevalence 68.2%) [28] and USA (46.0%) [29]. We found that compensated patients had an increased likelihood of PPI dispensing over time (p < 0.001). Short-term PPI use is recommended after variceal bleeding and endoscopic band ligation [30], following upper gastrointestinal ulcer bleeding [31] and for management of gastroesophageal reflux disease in appropriate patients [32]. However, longer-term use of PPIs (which occurred in 53.6% of study participants) is discouraged due to an increased risk of HE [14, 15], serious infections [33] and greater hospital readmission rate [16].

Pharmacologic management of pain in cirrhosis can be complex due to altered and unpredictable drug metabolism [34]. We observed a significant increase in the likelihood of opioid dispensing over time and ‘long-term’ use in 27.6% of patients. Whilst opioid use for chronic pain management is not advised and requires careful monitoring (due to an association with HE [11, 12], increased length of hospital admissions [35], hospital readmissions [36] and poor health-related quality of life [37]), our findings reflect ‘real-world’ clinician choices in the absence of effective alternatives. Benzodiazepines similarly have several clinical indications in chronic liver disease which must be balanced against the increased risk of HE [11]. Short-term use (< 10 days), but not long-term use (> 28 days), has been associated with first-time HE among decompensated patients with ascites [13]. Whilst our definition of ‘long-term’ use differed, most benzodiazepine users (~67%) in our study were dispensed these medicines once/intermittently. Prevalence of benzodiazepine use in our cohort (dispensed at least once for 41.0% of patients) was higher than in other studies (14.2–31.4%) [28, 29], and may reflect sample differences and prescribing practices between countries.

In terms of non-opioid analgesics, gabapentinoids (99.9% pregabalin) were dispensed for almost one-quarter of participants during the study period, and the likelihood of taking these medicines increased over time (non-significantly, p = 0.051) despite an increased risk of HE [11]. Alternative therapies for neuropathic pain such as tricyclic antidepressants (dispensed at least once for 17.6% of patients) may have similar adverse cognitive effects. Determining the extent of simple analgesic use (e.g. paracetamol/acetaminophen, NSAIDs) in our cohort was more difficult, as these medications are also available ‘over-the-counter’. NSAIDs were dispensed at least once to 22.2% of patients and there was a marked decrease in their likelihood of being dispensed over time (p = 0.040). In contrast, low-dose paracetamol (≤ 2 g/day) is regarded as safe and is the preferred initial analgesic [34, 38]. Paracetamol was dispensed at least once for 41.2% of participants and a non-significant increase in dispensing was observed over time.

Diuretics, beta-blockers and rifaximin are essential in the management of decompensated cirrhosis, and the increased prescription of these medicines likely reflects progression of liver disease severity over time. While overall prescriptions for comorbidities such as type 2 diabetes, hypertension and dyslipidaemia remained relatively stable, their likelihood of being dispensed over time differed between patients with compensated and decompensated cirrhosis. For example, use of angiotensin therapies decreased among patients with decompensated cirrhosis, consistent with pathophysiological changes due to portal hypertension and clinical guidance about the risk of renal impairment in patients with ascites [39, 40]. Increased period prevalence for newer type 2 diabetes medicines (dipeptidyl peptidase-4 inhibitors, glucagon-like peptide-1 analogues and sodium-glucose co-transporter 2 inhibitors) was observed in line with wider PBS availability over time, but the significance of this could not be computed due to model convergence errors with medicines of lower consumption rates.

In Australia, a large number of medicines taken by patients with cirrhosis are non-liver-related therapies [41]. We found GPs prescribed 69.2% of total dispensed medicines, 71.5% of ‘unsafe’ and 73.1% of ‘safety unknown’ medicines in the current study, similar to findings from The Netherlands [28]. Whilst medicines may have been initiated (e.g. drug and dose selection) by the prescriber, some GP prescriptions may have been repeated/written upon instruction from a specialist, and some specialist prescriptions may have been for a medication not specific to their specialty. Regardless of the initiating prescriber, regular medication review and monitoring are essential steps in the medication management cycle that should be undertaken with every prescription. The large proportion of GP prescriptions in our study highlights the reliance on primary care to manage day-to-day pharmacotherapy, including of potentially ‘unsafe’ drugs, for people with cirrhosis. This is an important observation because although specialist clinical practice guidelines provide information about prescribing and monitoring in cirrhosis, this information may not be readily available to clinicians in the community. There may also be limited information about safety and dosing for newer therapies in severe hepatic impairment. Medication reconciliation, improved availability of practical prescribing guidance, and targeted deprescribing interventions represent opportunities to reduce adverse medication-related outcomes for people with cirrhosis.

This study has several strengths and limitations. The sample was drawn from a large, heterogeneous, ‘real-world’ cohort of patients with cirrhosis. A high participation rate (77.9%) minimised the risk of selection bias, and assessment of cirrhosis severity, aetiology and comorbidities by clinicians minimised the risk of misclassification [21]. We used data from the PBS, which is the principal federal mechanism for providing Australians with reliable, timely and affordable access to medicines in the community, and captures the majority of outpatient prescription medicine use in Australia. However, data about ‘private’ prescriptions, ‘over-the-counter’ medicines, vitamins and supplements were not available. Several medicines may therefore be underrepresented in this study, including carvedilol (only PBS subsidised for heart failure; use for portal hypertension is ‘private’), paracetamol/acetaminophen, NSAIDs (e.g. ibuprofen, diclofenac), vitamin D, thiamine, lactulose, low-dose PPIs and salbutamol. Furthermore, medicines administered to inpatients are State/Territory funded and not captured by national PBS data. Therefore, sicker/decompensated patients with frequent or prolonged hospital admissions may have had their medication use underestimated because they spent less time in the community. The indication and dose of dispensed medicines was also not available.

Whilst prescriber information was not available in our dataset, we were able to infer prescriber identity (GP, consultant/specialist or ‘other’) for 83% of dispensed medicines using linked MBS and QHAPDC data. This method may have misclassified a number of prescriptions and 17% of dispensed medicines could not be linked. Therefore, limited conclusions may be drawn regarding differences in safety classifications between prescriber groups. We were unable to explore the association between use of specific medicines and decompensation events, as specific dates for decompensation events were not captured. We were also unable to reliably determine state transition from compensated to decompensated cirrhosis during the study period. The published transition rate of ~5–7% per year [42] suggests that ~20–30% of compensated patients may have changed classification during the median 53 month follow-up period. Similarly, we were not able to capture instances of re-compensation (e.g. with abstinence). While this limitation may affect the safety classification of medicines prescribed after transition and impact analyses comparing usage trends both within and between cirrhosis stage, study findings are in line with clinical expectations.

The final observed time period in this study (January–June 2020) coincided with the global COVID-19 pandemic; however, we are confident that our results reflect genuine medication usage trends over time. Australia benefited from a strong early response to COVID-19, which included rapid implementation of MBS-subsidised telehealth consultations to enable healthcare access whilst staying at home. Many pharmacies also offered home delivery services for multi-morbid ‘at-risk’ patients to ensure continuity of medication supply. In this study, GEE analyses with an unstructured correlation matrix minimised assumptions about within-subject and between-subject variables over time. Furthermore, we observed an increase in total medication use over time prior to the COVID-19 pandemic (see Supplementary Table). A very small (non-significant) decline in the median number of dispensed medicines was observed in January–June 2020 compared with the previous two time periods, which may be attributed to the social and economic impacts of COVID-19 (e.g. people not accessing healthcare) or attrition of ‘sicker’ (e.g. multi-morbid on polypharmacy) patients over time.

5 Conclusions

This study has enabled novel insights into the prevalence and patterns of prescriptions dispensed for people with compensated and decompensated cirrhosis over a 4.5 year period. These observational data suggest that polypharmacy is common, and that PPIs, opioids and benzodiazepines may be overused in Australian patients with cirrhosis. Further exploration of indication and potential alternative therapies is required. Pharmacovigilance and future medication safety efforts should target high-risk prescribing practices, educating GPs and promote medication rationalisation in the community. These data provide insights at an international level about potential high-risk prescribing practices that may not be recognised and contribute to adverse patient outcomes.