Adverse drug reactions, multimorbidity and polypharmacy: a prospective analysis of 1 month of medical admissions.

OBJECTIVE: To ascertain the burden and associated cost of adverse drug reactions (ADRs), polypharmacy and multimorbidity through a prospective analysis of all medical admissions to a large university teaching hospital over a 1-month period.
DESIGN: Prospective observational study.
SETTING: Liverpool University Hospital Foundation National Health Service (NHS) Trust, England. PARTICIPANTS: All medical admissions with greater than 24-hour stay over a 1-month period. MAIN OUTCOME MEASURES: Prevalence of admissions due to an ADR and associated mortality, prevalence and association of multimorbidity and polypharmacy with ADRs, and estimated local financial cost of admissions where an ADR was a contributing or main reason for admission with projected costs for NHS in England.
RESULTS: There were 218 identified patient admissions with an ADR giving a prevalence of 18.4%. The majority of these (90.4%) were ADRs that directly resulted in or contributed to admission. ADRs thus accounted for 16.5% of total admissions. Those with an ADR were on average taking more medicines (10.5 vs 7.8, p<0.01) and had more comorbidities than those without an ADR (6.1 vs 5.2, p<0.01). Drugs most commonly implicated were diuretics, steroid inhalers, anticoagulants and antiplatelets, proton pump inhibitors, chemotherapeutic agents and antihypertensives. 40.4% of ADRs were classified avoidable or possibly avoidable. The mortality rate due to an ADR was 0.34%. The average length of stay for those with an ADR was 6 days. Direct 1-month cost to the Trust from ADR admissions was £490 716. Extrapolated nationally, the projected annual cost to the NHS in England is 2.21 billion.
CONCLUSION: The local prevalence of admission and mortality from ADRs is higher than previously reported. Important factors that could be contributing to this include polypharmacy and multimorbidity. ADRs place a significant burden on patients and healthcare services with associated financial implications. Reducing inappropriate polypharmacy should be a major aim for preventing ADRs. © Author(s) (or their employer(s)) 2022. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

Over 1000 medical admissions were individually reviewed by specialists in clinical pharmacology and general internal medicine in this prospective analysis of adverse drug reactions (ADRs).

Standardised criteria, as listed in methods, were used to identify and classify ADRs. This improves the objectivity and reproducibility of the analysis.

Extrapolating the cost analysis nationally based on medical admissions locally may be unreliable due to differences including local population and services.

This study does not take into account how commonly each medicine that caused an ADR is prescribed in the local community.

Introduction

Improved living conditions and better access to and quality of medical care have led to increased life expectancy and the associated accumulation of long-term conditions (LTCs).1 According to a report by the Academy of Medical Sciences, multimorbidity is a growing issue globally, particularly in more economically developed countries where it is now considered the norm not the exception.2 Age is the single biggest risk factor for LTCs, such as cancer, cardiovascular disease and neurodegeneration, in developed countries. An ageing population is therefore at increased risk of polypharmacy.3 Care for people with multiple LTCs is often stretched across various single-organ specialists leading to siloed specialty prescribing and increasingly complex medication regimens.

Polypharmacy is the concurrent use of multiple medications by an individual. There is no consensus on the number of medications that defines polypharmacy because of the need to treat complex or multiple comorbidities with combinations of medicines. Thus, numerical definitions vary but perhaps the most common definition is taking five or more regular medications.4 The Wessex Academic Health Science Network has developed a set of prescribing comparators to better understand both the numerical and risk-related factors involved in the variation of prescribing of multiple medicines.5 In some individuals with complex or multiple conditions, polypharmacy may be appropriate, for example, when medicine use has been individually optimised and prescribed according to best evidence. In contrast, potentially inappropriate polypharmacy, where the risk of harms from individual medicines may outweigh their benefit in the context of the prescription as a whole, is associated with poor adherence and an increased risk of adverse drug reactions (ADRs) and interactions.6 ADRs are an important cause of morbidity and mortality, with significant health implications and associated economic burden. Landmark studies in the USA in 19987 and the UK in 20048 found ADRs to be related to 6.7% and 6.5% of hospital admissions, respectively. More recent systematic reviews have reported figures ranging from 3.6%9 to 15.6%.10 Potential reasons for variation in findings include the heterogeneity of methodologies and populations studied.

Addressing avoidable ADRs due to inappropriate prescribing is important to reduce the burden on patients and healthcare systems. There are few adult observational studies of admission due to ADRs in the UK that are not focused on specific populations or drug reactions. To our knowledge, there are only three in the last 20 years.8 11 12 In 2016, it was estimated that £1.3–£3 billion could be saved in the National Health Service (NHS) budget through reducing inappropriate and inefficient medicine usage.13

The aims of this study were to determine the current impact of ADRs on medical admissions and their association with multimorbidity and polypharmacy, and quantify the economic impact on the NHS. The population studied is broadly geographically comparable with that of the study by Pirmohamed et al.8

Methods

Study data were collected for 1 month in the city centre site of the Liverpool University Hospital Foundation NHS Trust, a large teaching hospital in Merseyside, England. The research question was developed due to the impact ADRs have on patients by causing admissions to hospital. There were no patient contributors or coauthors in this study. All patients referred via the medical assessment unit who were admitted for >24 hours were included. These were mostly via the emergency department but also included primary care referrals and admissions from outpatient clinics. Patients admitted via medicine but transferred to other centres for emergency treatment (such as primary coronary intervention) within 24 hours were included, as their expected inpatient stay would be >24 hours. Information including e-notes, community drug prescriptions and investigations were reviewed to determine if an ADR occurred. An ADR was defined using the Edwards and Aronson criteria.14 This does not include any type of drug overdose or relapse due to non-compliance. Cases were then defined as either the primary cause of admission, contributing factor or a co-incidental finding, and assessed against the following criteria:

Classification of the reaction as per Davies et al 15 into type A or type B reactions.

Causality as per the Liverpool Causality Assessment Tool (LCAT).16 This is a validated method of assessing the causality of ADRs that can be used by groups or individuals.

Severity as per the Adapted Hartwig Severity Scale,17 a widely used tool that categorises ADRs from severity level 1 (requires no change in treatment) to level 6 (directly or indirectly resulted in patient death).

Interactions as per the Drug Interaction Probability Scale (DIPS).18 DIPS assists practitioners in the assessment of drug interaction and evaluating causation in a specific patient.

Avoidability as per the Liverpool ADR Avoidability Assessment Tool (LAAT).19 This is a validated tool to support the assessment of the avoidability of ADRs based on available patient information.

Factors that suggested an ADR include the following: if it was consistent with the known adverse effect profile of the drug as per the British National Formulary,20 if there was a temporal relationship, and if alternate causes were excluded with history and investigation. Community drug prescription was verified and reviewed with patient electronic notes. These data were available for all admitted patients. If it was documented in the notes that a patient was not taking a medicine listed on their prescription, this was not included in our analysis.

Identification of ADRs and subsequent assessment of the above criteria were completed by authors RO and LW. Where consensus was not reached, it was assessed by a third reviewer (MP). Of the 258 patient episodes with possible identified ADRs, there was initial agreement on 236 (91%). For the remainder, consensus was obtained following joint review with MP.

Patient Level Information and Costing System data reporting healthcare resource group and hospital costs were obtained from the Liverpool University Hospitals NHS Foundation Trust finance office. Total costs were summed for episodes of admitted care resulting from ADRs (a), and for all other admissions (where ADR was a contributory factor, coincidental or unrelated (b)). In order to extrapolate costs, totals for non-elective short and long stay, and regular day or night admissions, for NHS England (c) were obtained for 2018–2019.21 Nationally projected costs were estimated as (a×c)/(a+b).

Patient and public involvement

The research question was developed due to the impact ADRs have on patients by causing admissions to hospital. There were no patient contributors or coauthors in this study.

Statistical analysis

Statistical analysis was performed using the SAS software (V.9.4, SAS Institute). The results are presented either as means and SDs or frequencies and percentages. Associations between patient characteristics and admission type (ADR/non-ADR) were investigated using univariable and multivariable logistic regression, with associations presented as ORs with 95% CIs. The multivariable model used backwards selection with a probability of exclusion of 0.1. A p value of <0.05 was regarded as statistically significant.

Results

There were 1187 admissions with 218 patients with an ADR (18.4%). As some of those had multiple ADRs, 235 were identified in total.

Characteristics of ADRs

One hundred forty-five (66.5%) of the ADRs were the primary cause of admission, with 51 (23.4%) contributing to admission, and 22 (10.1%) co-incidental findings that alone would not have required hospital stay. Thus, ADRs directly caused or contributed to 16.5% of all admissions. Using the LCAT,16 45 (20.6%) were graded as definite, 79 (36.2%) as probable and 94 (43.1%) as possible ADRs. Forty (18.4%) were graded as avoidable and 46 (21.1%) as possibly avoidable using LAAT criteria.19 Sixty-four (29.4%) of ADRs were possibly or probably cause by a drug–drug interaction as per DIPS.18 One hundred eighty-eight (86.2%) were type A reactions as defined by Davies et al 15 and 30 (13.8%) type B. One hundred sixty-four (75.2%) of the ADRs were documented as recognised or acted on by the admitting medical team. The main drugs implicated in ADRs are listed in table 1. There were four ADRs (1.8%) that directly resulted in death and a further five that were implicated or a contributing factor to death (2.3%) (table 2). The mortality rate directly from ADRs, from all admissions, was therefore 0.42%. Median length of stay of patients with an ADR was 6 days.

Table 1

Drugs implicated in patient episodes with adverse drug reactions (ADRs)*

Drug class	No of ADRs (%)	Offending drug	ADR
Diuretics	31 (14.2)	Furosemide (13), spironolactone (8), bumetanide (6), bendroflumethiazide (2), co-amilofruse (1), indapamide (1)	Renal impairment (18), electrolyte derangement (12), postural hypotension (1)
Steroid inhaler	27 (12.4)	Steroid inhaler (27)	Pneumonia (26), oral thrush (1)
Anticoagulants	21 (9.6)	Warfarin (7), apixaban (5), edoxaban (4), rivaroxaban (4), enoxaparin (1)	Minor bleeding (10), anaemia (4), intracranial haemorrhage (4), gastrointestinal bleed (3)
Proton pump inhibitor	18 (8.3)	Lansoprazole (9), omeprazole (6), pantoprazole (3)	Hypomagnesaemia (11), hyponatraemia (6), Clostridium difficile (1)
Antiplatelet	16 (7.4)	Aspirin (13), clopidogrel (3)	Intracranial haemorrhage (5), gastrointestinal bleed (4), minor bleeding (4), anaemia
Chemotherapy	16 (7.3)	Chemotherapy (16)	Neutropenic sepsis (8), sepsis (4), constipation (1), deranged electrolytes (1), rash (1), thrombocytopenia (1)
ACE inhibitor/angiotensin receptor blocker	14 (6.4)	Losartan (4), ramipril (4), irbesartan (3), candesartan (1), lisinopril (1), perindopril (1)	Renal impairment (9), postural hypotension (3), hyperkalaemia (1), renal failure (1)
Antidepressants & antipsychotics	13 (6.0)	Mirtazapine (2), sertraline (2), sulpiride (2), carbemazapine (1), dosulepin (1), nortriptyline (1), olanzapine (1), risperidone (1)	Confusion (3), hyponatraemia (3), parkinsonism (3), constipation (1), gastrointestinal bleed (1), prolonged QTc (1)
Opiates	13 (6.0)	Codeine (5), morphine sulfate (3), oxycodone (2), tramadol (2), buprenorphine (1)	Constipation (6), confusion (4), respiratory depression (2), hallucinations (1)
Other	49 (22.4)	Other (49)	Other (49)

*In those with multiple ADRs, only the most severe ADR was included in this table, as defined by the Adapted Hartwig Severity Scale16 (see online supplemental material 1 for full list).

QTc, corrected QT interval.

Table 2

Deaths directly related to the adverse reaction (Adapted Hartwig class 7b)

Medication	Number	Cause of death
Chemotherapy	2	Neutropenic sepsis (2)
Aspirin	1	Intracranial haemorrhage
Edoxaban	1	Gastrointestinal bleed

Comparison of patients with and without ADRs

Table 3 shows descriptive statistics of patients with and without ADRs. In the patients with ADRs, as would be expected, liver and renal impairment were more prevalent compared with patients without ADRs (6.8% vs 2.8%, p=0.004).

Table 3

Characteristics of patients with and without adverse drug reactions (ADRs)

	ADR group	Non-ADR group	Total
Number of admissions	218	969	1187
Age, mean (SD)	73.2 (14.5)	66.7 (19.2)	67.9 (18.6)
Male (%)	106 (48.6)	455 (47.0)	561 (47.3)
Number of medicines, mean (SD)	10.5 (4.6)	7.8 (5.1)	8.3 (5.1)
Polypharmacy (%)	199 (91.3)	706 (72.9)	905 (76.2)
Number of comorbidities, mean (SD)	6.1 (3.0)	5.2 (3.3)	5.4 (3.2)
Multimorbid (%)	99.1	90.3	91.9
Liver impairment* (%)	6.9	2.8	3.5
Renal impairment† (%)	11.0	6.8	7.6

*Liver impairment defined as chronic liver disease.

†Renal impairment defined as chronic kidney disease stage IV or V.

Logistic regression results are presented in table 4. Patients with ADRs were older than those without (mean age 73.2 (14.5) vs 66.7 (19.2), OR 1.02 (1.01 to 1.03)) and were taking more medicines (mean 10.5 (4.6) vs 7.8 (5.1), OR 1.11 (1.07 to 1.14)), with polypharmacy present in 91%, compared with 73% in the non-ADR group. They had more comorbidities (mean 6.1 (3.0) vs 5.2 (3.3), OR 1.08 (1.04 to 1.13)), although this variable was not included in the multivariable model (due to its correlation with number of medicines). Those with ADRs were more likely to have liver impairment (6.9% vs 2.8%, OR 2.58 (1.35 to 4.93)) and renal impairment (11.0% vs 6.8%, OR 1.69 (1.04 to 2.78)).

Table 4

Logistic regression analysis of patients with and without adverse drug reactions

	Univariable OR(95% CI)	P value(Wald χ2)	Multivariable OR(95% CI)	P value(Wald χ2)
Age	1.02 (1.01 to 1.03)	<0.001	1.02 (1.01 to 1.03)	<0.001
Sex (male)	1.07 (0.80 to 1.44)	0.659
Number of medicines	1.11 (1.07 to 1.14)	<0.001	1.10 (1.07 to 1.13)	<0.001
Number of comorbidities	1.08 (1.04 to 1.13)	<0.001
Liver impairment (CLD)	2.58 (1.35 to 4.93)	0.004	3.23 (1.63 to 6.40)	<0.001
Renal impairment (CKD stage IV or V)	1.69 (1.04 to 2.78)	0.036

CKD, chronic kidney disease; CLD, chronic liver disease.

Healthcare Resource Group (HRG) costs were available for 214 (98.2%) patients in the ADR group and 950 (98.0%) patients in the non-ADR group. Mean costs per episode of care were £2293 (95% CI 1918 to 2668) and £2131 (95% CI 1899 to 2364), respectively. The total costs of admissions resulting from ADR were £309 207, representing 12.3% of the costs of the whole cohort over the 1-month sampling frame. Admissions where ADR was a contributing factor cost £138 762 (5.5%); and where ADR was coincidental, the cost was £42 747 (1.7%). The total costs of non-elective short and long stays, and regular day or night admissions across all NHS Trusts and NHS Foundation Trusts in England were £17.98 billion in 2018–2019, of which we estimate £2.21 billion were due to admissions resulting from ADRs.

Discussion

This study found ADRs in 18.4% of hospital admissions. In 16.5% of admissions, it was the primary cause or a contributing cause suggesting ADRs have a significant burden on hospital admissions. This is over twice as high as the 6.5% found in the study by Pirmohamed et al, 8 which consisted of broadly the same geographical area. Most commonly implicated medicines included diuretics, steroid-based inhalers, anticoagulants, proton pump inhibitors (PPIs), antiplatelets, chemotherapy agents, antihypertensives, opiates and antidepressants/antipsychotics (table 1). It is important to consider this study does not reflect how often each of these medicines is prescribed in the community. Some of the medicines implicated with the highest number of ADRs and deaths may be a reflection of how commonly they are prescribed. Furthermore, the use of these medicines also provides clinical benefits that reduce morbidity, mortality and need for hospital admissions, and this is not taken into account by our data. Direct mortality from ADRs was 0.42%, which is also an increase from Pirmohamed et al’s study (0.15%)8 and twice as high as a recent meta-analysis.22

Approximately 40% of ADRs were classified as avoidable or possibly avoidable. This is consistent with previous studies which found significant proportions of ADRs that lead to hospital admissions are potentially avoidable.23 Furthermore, as expected, the majority (86.2%) of ADRs were ‘type A’ reactions, meaning that they were the result of the expected pharmacological action of the medicine and therefore potentially more predictable and avoidable. Given this, future efforts should be targeted at reducing these preventable admissions. Key strategies that can mitigate ADRs include stratifying patients by susceptibility prior to medication initiation using key information such as comorbidities, concomitant medications and renal and hepatic function. This is particularly required in elderly patients who are at risk of accumulating multiple age-related health deficiencies that require drug therapy.24 Where available and appropriate, pharmacogenomic testing can also be used to identify those at high risk of an ADR to guide medication choice or optimal dose. Following initiation, management plans such as appropriate blood test monitoring and scheduled clinical review for ongoing indication can also reduce the risk of an ADR.25 A total of 29.4% of ADRs were possibly or probably caused by drug–drug interactions as per DIPS.18 Deprescribing, defined as the process of dose reduction or stopping of medicines by a healthcare professional, has been proposed as an important tool to reduce the burden of ADRs. The optimal use of medicines should include the entire prescribing spectrum including starting, dose adjustment and stopping at the point at which harm outweighs benefit.

Comparison with Pirmohamed et al’s study

The study by Pirmohamed et al 8 was a large prospective study of admission due to ADRs in two Liverpool hospitals, the large university teaching hospital used in this study and a smaller district general hospital. This found an ADR prevalence in hospital admissions of 6.5%, suggesting there has been a significant increase since 2004. Numerous clinical reasons could have influenced this including changes in population demographics, increased morbidity and prescribing patterns. Some of the increase may be because pharmacovigilance has improved over the last 20 years, and the adverse reaction profile of drugs is more comprehensive. For example, following a large case–control study by Suissa et al, 26 an increased risk of pneumonia and dose-dependent 30-day mortality from steroid-based inhalers in patients with Chronic Obstructive Pulmonary Disease (COPD) was added as a side effect to the British National Formulary. Over 10% of the ADRs in this study are attributable to this. Some ADRs were also secondary to newer therapies including chemotherapies and monoclonal antibodies that have been developed since 2004.

In 2004, two of the main causative medicines were Non-Steroidal Anti-Inflammatory Drugs (NSAIDs) (11.8%) and antiplatelets (aspirin and clopidogrel) (23.8%). However, in this more recent study, these medicines were implicated in only 0.85% and 7.4% of the ADRs, respectively. Despite the increase in total ADRs from 2004, this would suggest a large proportional reduction. This could be due to greater awareness of these ADRs in older people leading to enhanced pharmacovigilance in prescribers along with changes in prescribing practice including co-administration of PPIs. However, this change has promoted PPIs as a cause of ADRs from very few cases to being responsible for 12.1% of ADRs in this study. The majority of the reactions were only mild transient electrolyte disturbances, with only a single severe associated ADR of Clostridium difficile. In the case of antiplatelets, two factors are likely to have contributed to this change: (a) there has been an active programme of reduction in their use for primary prevention of cardiovascular disease; and (b) changes in atrial fibrillation guidelines have led to a greater use of anticoagulants rather than antiplatelets.

In recent years, concern about an opiate crisis due to excessive community prescription has occurred in the USA.27 In this study, opiate medications accounted for 5.1% of ADRs which is similar to the 6.0% found in 2004. This suggests that proportionally, there has not been a significant increase in prescription opiate-related admissions locally. Of the related 13 events, the majority were non-lethal, with only 2 cases exhibiting respiratory depression but with no permanent harm following reversal.

Changes in prescribing patterns and subsequent ADRs may reflect increasing multimorbidity and polypharmacy. However, as such data were not previously collected, this cannot be directly compared. Furthermore, methodological differences may have contributed as this study did not include any data from a district general hospital or surgical admissions. Additionally, screening and data collection was completed by medical doctors and clinical pharmacologists, whereas previously it was completed by a number of healthcare professionals. Thus, differences in clinical and diagnostic experience could also be responsible for some increased identification of ADRs. Finally, since 2004, the Liverpool University Hospital Foundation NHS Trust has adopted electronic health records which may have assisted in the identification of ADRs in this study.

Multimorbidity and polypharmacy

Multimorbidity and polypharmacy were both associated with admission due to ADRs on univariate analysis. Those with an ADR were taking on average 35% more medicines than those without (10.5 vs 7.8, p<0.001), which is an established risk factor for ADRs.28 Despite this, polypharmacy must not be conflated with inappropriate prescribing as some patients, particularly those who are multimorbid, require multiple medicines to optimise their LTCs with associated positive outcomes. This study did not assess the appropriateness of all community prescriptions, but only of those that directly caused an ADR via the avoidability assessment tool.

The mean number of comorbidities for the entire admitted population was 5.4. Although we do not have direct data from 2004 for comparison, it is reported that the incidence of multimorbidity has been increasing.2 The ADR group on average had 17% more comorbidities than the non-ADR group (6.1 vs 5.2, p<0.001), which is a known risk factor.28 However, the number of comorbidities was not part of the logistic regression because of the correlation between the number of medicines and number of comorbidities. Although the total number of comorbidities is relevant, it does not give insight into disease severity, for which the number of medications being taken may be a better proxy. For example, hypertension or type 2 diabetes managed with lifestyle factors would produce less medication burden than more advanced disease. Furthermore, some conditions and their management are known to predispose to prescribing cascades and therefore polypharmacy.28 29

Cost analysis

ADRs are a significant cost burden on the NHS, and in this study accounted for £1 in every £8 spent on the care of non-elected hospital admissions. Putting this into perspective, the total cost of admissions related to ADRs over the 1-month study period (£490 716) is comparable with the annual cost of chemotherapy procurement by the hospital in which the study was conducted. When extrapolated nationally, our estimate of £2.21 billion for admissions resulting from ADRs exceeds the costs of all outpatient procedures for NHS England. Previous cost analyses of medication-related harm in England provide annual estimates of £1.9 billion based on an extrapolation from Pirmohamed et al,30 £529 million for potentially avoidable ADR-related admissions31 and £396 million for discharged elderly people.32

Strengths and weaknesses

Key strengths include the following: data were collected prospectively and notes were reviewed by specialists in clinical pharmacology and general internal medicine. This optimised the reliability of collected data and identification of ADRs. The availability of patient-level cost data, reflecting the actual spend on hospital care, represents another strength over many previous cost analyses.

Using standardised criteria to identify and classify ADRs improves objectivity and reproducibility as evidenced by levels of concordance >90% between reviewers. However, some elements of the criteria can be subjective and rely on reviewer clinical experience. Furthermore, many of the medical conditions and side effects attributed to an ADR may have occurred regardless of prescription, for example, regarding steroid inhalers and the increased risk of pneumonia in patients with COPD. A limitation of our study is that we have not concurrently assessed the benefits of taking medicines in individual patients.

It must be emphasised that causality assessment is a time-consuming process requiring clinical insight and therefore it is challenging to do this in time-limited real-world clinical practice. In the future, efforts to enhance the usability of electronic healthcare records, using time-saving approaches, such as artificial intelligence and machine learning, could make medicines optimisation more efficient.

Liverpool is ranked as the most deprived major city in England, an established factor in predicting increased morbidity.32 With the disparity between the most and least deprived areas in England having increased since the 1990s,33 34 changes in local population may have influenced differences found from 2004 as well as limit the utility of extrapolation of data nationally. In addition, generalisability of these data to more ethnically diverse populations is limited as Liverpool is 91% white.

Conclusion

This study found ADRs contributed or directly caused 16.5% of all admissions with an associated mortality rate of 0.34%. Factors associated with an ADR on logistic regression included age, number of medications and liver impairment. The data suggest ADRs place a significant and increasing burden on patients and healthcare services with associated financial implications. Using patient-level cost data, the projected annual cost of ADR admissions to the NHS in England is £2.21 billion. With 39.4% of these ADRs identified as avoidable or potentially avoidable, future efforts should be directed to reduce this burden. Reducing inappropriate polypharmacy should be a major aim for preventing ADRs.