FormalPara Key Points for Decision Makers

Non-initiation or delayed initiation of anti-coagulation in people with atrial fibrillation (AF) is associated with increased healthcare utilisation and costs, particularly in females and the frail.

There are significant opportunities for improvement in the prescribing and prompt initiation of anti-coagulation in an AF population.

As inpatient admissions are the primary cost driver in AF, future interventions should be aimed at mitigating healthcare resource utilisation by focussing on reducing inpatient admissions in people with AF.

1 Introduction

1.1 Background/Rationale

Atrial fibrillation (AF) is an abnormality of cardiac rhythm affecting approximately 33 million people globally [1]. Whilst its prevalence is currently around 2%, this is strongly correlated with increasing age; around 0.1% of people aged below 55 are affected, rising to around 10% of those aged over 80 [2, 3]. Owing to the ageing population, an estimated 1.8 million in the UK and 17.9 million in Europe are projected to be affected by 2060. AF elevates stroke risk by three to five times and is the risk factor most frequently implicated in cardioembolic strokes, which arise due to pooling of blood in the left atrium of the heart, with subsequent systemic embolisation. Such strokes are associated with greater morbidity and mortality, and longer hospital admissions than alternative subcategories of strokes, and although the incidence of strokes has declined overall, cardioembolic strokes have tripled in the preceding 2 decades in the UK and are projected to increase threefold by 2050 [4]. AF is also associated with a greater risk of heart failure, myocardial infarction, dementia, and chronic kidney disease [5]. Due to the significant costs associated with acute care, a principal determinant of the overall economic burden of AF costs is inpatient care [6, 7]. Furthermore, concomitant with the increasing prevalence of AF and increased AF-associated secondary care admissions, the cost of hospitalisation in people with AF is rising. Indeed, Patel et al. reported a 24% relative increase in mean AF hospitalisation costs over a period of 10 years [8]. AF and its sequelae thus exert an increasingly significant burden on healthcare resources, presenting a substantial challenge for health systems internationally.

Thromboprophylaxis with vitamin K antagonists (VKA) or direct oral anti-coagulants (DOACs) prevents approximately 70% of strokes in AF patients and is recommended for the majority of patients diagnosed with AF, representing a further key determinant of AF costs [1]. This includes not only expenditure for prescriptions, and for international normalised ratio (INR) monitoring in those prescribed VKAs, but also the costs for the management of adverse events associated with anti-coagulation, principally bleeding.

Thromboembolic risk and mortality risk are greatest in the month following diagnosis of AF, and these risks are mitigated in those prescribed anti-coagulation at diagnosis [9]. However, despite the considerable risk of stroke and systemic embolism (SSE) in individuals with AF, oral anti-coagulants (OACs) are often underutilised [10]. Whilst non-initiation, suboptimal adherence, and discontinuation of OACs are known to be associated with poorer clinical outcomes, the effect of late commencement of anti-coagulation on utilisation of healthcare resources and associated costs in people newly diagnosed with AF is not yet established [11]. The aim of this study was to assess comprehensive healthcare costs in people with AF, comparing those who were prescribed anti-coagulation in the immediate period following the index AF diagnosis date with those who had delayed initiation and those who never commenced anti-coagulation.

2 Methods

2.1 Study Design

This was a retrospective costing analysis of adults admitted to secondary care with an incident AF diagnosis in Scotland. A comprehensive approach, inclusive of inpatient, outpatient, prescription, and care home costs incurred by the respective AF cohorts, regardless of whether such costs were associated with the disease, was utilised. A medicalised approach on the other hand might not adequately represent all relevant costs as it relies on subjective assessment of which healthcare episodes are associated with the disease in question, in this case AF. Such judgment could be arbitrary [12, 13].

2.2 Data Sources and Cohort

Access to fully anonymised data were granted by Public Health Scotland (PHS) for the purposes of a wider study that evaluated the comparative effectiveness of anticoagulation for thromboprophylaxis in individuals with AF using routinely collected healthcare data. Individuals aged 18 or older who were diagnosed with AF between January 1st 2012 and April 30th 2019 with a baseline CHA2DS2-VASc score (inclusive of one point for female sex), indicating thromboembolic risk, of 2 or greater were identified from Scottish Morbidity Records (SMR) 01. SMR01 includes data from inpatient and day case discharges for all specialities, except for psychiatry and obstetrics. The International Classification of Diseases, Tenth Revision (ICD-10) code I48, in any diagnostic position, was used to isolate the AF cohort (see Table S2 in the Electronic Supplementary Material [ESM]). SMR01 data were triangulated with the Scottish Stroke Care Audit (SSCA), a secondary care stroke dataset that includes data on AF diagnoses. Patient-level data linkage was conducted using prescribing data from the Prescribing Information System (PIS), mortality data from National Records of Scotland (NRS), outpatient care data from SMR00, and care home admissions data from the Scottish Care Home Census (SCHC) [14]. The analyses excluded people with alternative indications for anti-coagulant prescriptions, such as venous thromboembolism, valvular heart disease, or a mechanical heart valve, to ensure that AF was the exclusive indication for anti-coagulant prescriptions in our population (see Table S2 in the ESM).

The start date of the study was based on the availability of care home admission data. A 5-year lookback period from January 1st 2012 was implemented to allow for the exclusion of individuals in whom the index AF diagnosis occurred during the lookback period to prevent double-counting. Data were available until May 31st 2021, allowing for a follow-up period of 2 years to measure utilisation of health and social care resources and associated costs (Fig. 1).

Fig. 1
figure 1

Overview of AF cohort identification and sub-categorisation according to timing of first prescription of OAC and estimation of global costs. AF atrial fibrillation, ICD-10 International Classification of Diseases, Tenth Revision, OAC oral anti-coagulant, PIS Prescribing Information System, SMR Scottish Morbidity Records

Full size image

Data cleaning and pre-processing were completed in accordance with the accepted methodologies for routinely collected health datasets [15]. Since under 5% of records had missing data, and appeared to be missing completely at random, complete case analysis was undertaken without imputation.

Prescriptions in Scotland are available free of charge. The PIS dataset consists of prescribing, dispensing, and reimbursement data generated from prescriptions issued by doctors and other healthcare prescribers after being dispensed by a pharmacy [14]. Long-term repeat prescriptions including anti-coagulants are typically dispensed at intervals of 28 or 56 days. Three AF cohorts were assessed according to the timing of the first prescription of either warfarin or a DOAC (edoxaban, rivaroxaban, apixaban, dabigatran), if prescribed: never started, immediate OAC, and delayed OAC (Table 1). A threshold of 60 days was selected for the identification of late initiation since this appropriately accounted for the usual dispensing intervals for anti-coagulants and allows sufficient time for a repeat prescription of anti-coagulation to be dispensed for those commenced on anti-coagulation during their index admission and discharged with a supply. Only individuals that were anti-coagulant naïve prior to diagnosis were included in the analysis.

Table 1 Cohort definitions
Full size table

2.3 Unit Costs

Per diem costs for inpatient admissions, mapped according to specialty and health board, were derived from the R040 Specialty Group Costs in the Scottish Health Services Costs Book 2017/18; the total cost was calculated by multiplying per diem costs by the duration of the episode [16]. Costs for outpatient appointment attendance were derived from the Scottish Health Services Costs Book 2018/19, mapped according to specialty and health board using the average cost per attendance at consultant or nurse-led clinics [17]. Individual prescription costs were calculated from the PIS dataset by first determining the price per unit by dividing pack price by pack size then multiplying by the number of units dispensed [13].

Care home costs were calculated by mapping average weekly charges derived from the PHS care home census to council area, with the overall cost determined according to admission year, length of stay, and the provision of nursing care recorded in the SCHC dataset [18].

2.4 Econometric Model

A two-part model was utilised to account for the skewed distribution of healthcare costs, non-negative costs, and a considerable number of zero-cost observations reflecting no resource utilisation during a particular time period [13]. The first part used a probit model for the estimation of the probability of positive healthcare resource utilisation (see Equation S1 in the ESM), whilst the second part used a generalised linear model (GLM) with gamma distribution and log link to estimate costs, conditional on incurred costs being positive (see Equation S2 in the ESM). A gamma distribution and a log link function are widely used specifications for estimating healthcare expenditure. Modelling cost data, a gamma distribution offers means to account for heteroscedasticity and places less weight on very high costs. Identical explanatory variables were utilised in both parts of the model. Effect sizes are presented as cost-ratio regression coefficients with their respective 95% confidence intervals (CIs). The method of recycled predictions was used to determine and compare marginal effects for each of the included covariates, holding other covariates at their mean values.

To account for confounding by indication, arising from lack of randomisation in real-world studies, we have used propensity score (PS) methods to estimate the PSs feeding into our econometric model. Such methods estimate the probability of treatment assignment conditional on observed baseline characteristics. For our PS model, we have used inverse probability of treatment weights (IPTW) with a multinomial logit, typically used in scenarios with multiple interventions. Using a double robust approach, to further reduce any residual bias, we have used the same set of covariates described in our two-part model. The adequacy of model specification and covariate balance was assessed using mean standardised differences; differences in the means of covariates if below the threshold of 0.1 standard deviations were considered negligible.

In addition to the main explanatory variable, AF cohort, the model was also adjusted for demographic and clinical factors that were hypothesised to possibly have an impact on costs. These included age, sex, year, mortality, and an indicator of socio-economic status, the Scottish Index of Multiple Deprivation (SIMD), a scale of multiple deprivation for geographical locations, represented as quintiles, 1 denoting the most deprived and 5 indicating the least deprived areas, respectively. Geographical location was also included as a covariate, using the Scottish Government eightfold urban rural classification summarised into three strata: urban (1–2), small and large towns (3–6), and rural (7–8). This is measured based on population size and distance to the nearest settlement of more than 10,000 people. Further covariates were CHA2DS2-VASc score at baseline, an electronic frailty score calculated according to Gilbert et al., and comorbidity (determined using the Charlson Comorbidity Index [CCI]) [19]. Two interaction terms were included to account for interactions between age and mortality, and age and comorbidities. Indeed, there is a proportional relationship between increasing age, and mortality and comorbidities.

Separate models were run for each of the cost components, inpatient costs, outpatient costs, prescription costs, and care home costs, with a further model for overall costs. Cost components were analysed to identify the primary cost drivers, and the effect of the respective cost components on the total cost was evaluated.

STATA version 16 was utilised for statistical analyses. Descriptive statistics for categorical variables are reported as frequencies and percentages. Regression results are presented as coefficients with 95% CIs and standard errors.

2.5 Reasons for Inpatient Admission

The most frequent reasons for inpatient admissions within the SMR01 dataset, based on the ICD-10 codes recorded for the main condition, and other conditions in the first to fifth diagnostic positions for episodes of care, were determined for each cohort.

2.6 Projected Future Costs

Projected future costs for people with AF in Scotland, including inpatient costs, outpatient costs, prescription costs, and care home costs were estimated for 2030, 2040, and 2050. AF prevalence data were extracted from the Scottish Primary Care Information Resource (SPIRE) dataset for 2020 [20]. The prevalence of AF was assumed to increase by 2% annually, since this was comparable to the percentage increase in prevalence in the SPIRE dataset, with sensitivity analyses completed with larger annual increases in prevalence of 3% and 5%. Costs were assumed to increase by 3% annually from 2018, reflecting the timing of our costing source data, and were adjusted for an annual inflation rate of 2% [21].

3 Results

3.1 Characteristics of the AF Cohort

Of 54,385 people diagnosed with AF between January 2012 and April 2019, 27,580 individuals (50.7%) were prescribed an anti-coagulant (Table 2). Of all anti-coagulated individuals, 19,926 (72.2%) commenced anti-coagulation within 60 days of diagnosis of AF. In individuals with AF who were older, were frailer, had co-morbidities, or had received a prior anti-platelet prescription, anti-coagulation was more likely to be delayed or never initiated.

Table 2 Baseline characteristics by AF cohort
Full size table

3.2 Econometric Modelling Results

Propensity scores were visually inspected in terms of adequate overlapping distributions (see Figure S1 in the ESM). Following the first graphical assessment on the PS model’s specification, means standardised differences were used to assess the distribution of baseline covariates between groups. Results indicate that most of the baseline characteristics of patients in the delayed and immediate anti-coagulation groups have an adequate starting balance (see Figures S2 and S3 in the ESM); this is reflected in the differences in the means being below the threshold of 0.1 standard deviations. When weights were applied, as indicated by the adjusted standardised differences approaching to 0, a very good balance of patients’ covariates was obtained for both delayed and immediate groups.

The full regression outputs for the two-part models for overall costs, inpatient costs, outpatient costs, prescription costs, care home costs, and proportions of zero cost observations are available in the ESM (see Tables S3–S8). Increasing frailty risk score and socioeconomic deprivation according to SIMD were associated with an increased probability of utilising healthcare resources and increased estimated costs (see Table S3 in the ESM).

Inpatient costs were the principal driver of costs in all cohorts, whilst prescription and outpatient costs comprised a relatively small proportion of the overall estimated costs (Fig. 2; Tables S9–S12 in the ESM).

Fig. 2
figure 2

Average annual costs for AF cohorts, by cost component, per patient diagnosed with AF. AF atrial fibrillation, OAC oral anti-coagulant

Full size image

The adjusted mean annual cost for the entire AF population was £7807 per person (unadjusted: £8491). The adjusted estimated mean annual cost was £7981 (unadjusted: £10,433) for those that never started, £6621 (unadjusted: £3976) for those that commenced anti-coagulation in the immediate period after AF diagnosis, and £9763 (unadjusted: £13,983) for those that initiated anti-coagulation late (Table 3). However, cost differences between those who initiated anti-coagulation late and those who never started were not statistically significant. Across the AF cohorts, females incurred statistically significantly greater costs than males, and frailty score was positively correlated with costs (Fig. 3a; Tables S3–S7 and S9–S12 in the ESM). There was a similar positive correlation between age and costs, aside from in the cohort that never commenced anti-coagulation, in which the 76–85 age category was associated with higher incurred costs than in those aged over 85 (Fig. 3b). Figure 4a, b show the distribution of inpatient costs as the main driver of overall costs by sex (Fig. 4a) and age group (Fig. 4b). Mortality was associated with considerably greater healthcare costs. Residence in towns was associated with greater costs than residence in urban or rural areas.

Table 3 Annual total costs per patient diagnosed with AF
Full size table
Fig. 3
figure 3

Average annual total costs for AF cohorts per patient diagnosed with AF, by sex (a) and by age category (b). AF atrial fibrillation, OAC oral anti-coagulant

Full size image
Fig. 4
figure 4

Average annual inpatient costs for AF cohorts per patient diagnosed with AF, by sex (a) and by age category (b). AF atrial fibrillation, OAC oral anti-coagulant

Full size image

3.3 Reasons for Inpatient Admission

Infections, primarily urinary and respiratory tract infections, were a principal cause of hospitalisations in our AF population (Table 4). Hip fracture and repeated falls were common reasons for admission to secondary care in those never commenced on anti-coagulation. Systemic embolism was a frequent cause of inpatient admission in the cohort with a delayed initiation of anti-coagulation. Chronic conditions such as type II diabetes, hypertension, and chronic kidney disease were frequent secondary causes of inpatient admission across the cohorts (see Tables S13–17 in the ESM).

Table 4 Frequency of main condition ICD-10 codes by cohort
Full size table

3.4 Projected Future Costs

Assuming a 2% annual increase in the prevalence of AF, the estimated costs associated with AF healthcare would be £1311.5 million in 2030, £2619.1 million in 2040, and £5230.3 million in 2050 (Table 5). However, if the annual percentage increase in AF prevalence is greater than 2%, the costs may be substantially higher; the forecasted cost in 2050 is £7008.7 million for a 3% annual increase, and £12,479.6 million for a 5% increase. These projections need to be interpreted cautiously and are subject to uncertainties related to future treatment options, strategies around screening and early detection of AF, and actual rate of inflation. However, we believe that projections provide an initial indication of the direction of healthcare cost development under different scenarios.

Table 5 Forecasted costs inclusive of inpatient costs, outpatient costs, prescription costs, and care home costs, for the AF population in Scotland by annual percentage increase in prevalence of AF
Full size table

4 Discussion

Whilst the majority of our AF population who were prescribed anti-coagulation initiated treatment during the period immediately after diagnosis, a considerable number of eligible people were never prescribed anti-coagulation. This is consistent with previous studies; for example, a retrospective cohort study of people newly diagnosed with AF, utilising Quebec health insurance data, reported that 52% (n = 32,431) of their population commenced anti-coagulation, with 91% of those prescribed a VKA and 84% of those prescribed a DOAC initiating anti-coagulation within 90 days of incident AF diagnosis [22]. The remainder were not prescribed anti-coagulation during the study period.

Whilst late OAC initiation has previously been characterised in prescription adherence modelling studies of people newly diagnosed with AF, these studies considered more prolonged delays than that in our study. Indeed, a retrospective cohort study applying latent class mixed models to Medicaid claims data to evaluate patterns of OAC adherence identified a cohort in which anti-coagulation was initiated 6–10 months after diagnosis, which accounted for 13% of people prescribed warfarin and 8% prescribed a DOAC [23]. Furthermore, Hernandez et al. utilised group-based trajectory models to identify four trajectories of anti-coagulant adherence in the year after AF diagnosis in a Medicare cohort: 43.8% never commenced anti-coagulation, 7.6% had a delayed initiation between 3 and 8 months after diagnosis, and 49% commenced anti-coagulation early [24]. In both studies, delayed initiation was associated with female sex, as in our study, and an elevated HAS-BLED score.

In our population, delayed OAC initiation, followed by never started, was associated with higher healthcare resource utilisation and global costs compared to immediate OAC initiation. Delayed initiation of anti-coagulation may be associated with higher costs compared to never initiated, due to the initial prothrombotic effect of anti-coagulation, arising as a result of a transient hypercoagulable state, and adverse events, including major bleeding [25, 26]. Indeed, this difference in our study was principally due to higher inpatient costs in those with a late initiation of OACs. AF is commonly diagnosed secondarily during medical assessment for another event, such that clinical risk is greatest in the initial period following AF diagnosis [9, 27]. Healthcare utilisation may be greater in those with delayed OAC initiation compared to immediate anti-coagulation, since prompt anti-coagulation may mitigate thromboembolic risk in some individuals. Indeed, systemic embolism was a frequent cause of inpatient admission in the delayed initiation cohort. The costs associated with strokes are substantial, particularly in the first year post-stroke; Patel et al. estimated the mean costs from an NHS and Personal Social Services (PSS) perspective for ischaemic and haemorrhagic strokes in the first year to be £17,063 and £19,099 per person, respectively, and £7225 and £8230, respectively, thereafter [28]. Furthermore, strokes occurring in people with AF are associated with greater costs than those that occur in the absence of a diagnosis of AF; healthcare resource utilisation and costs for cardioembolic strokes, which primarily arise due to AF, are significantly greater than for non-cardioembolic strokes [29]. The costs in our study were greater than those reported previously for a general Scottish AF population by Ciminata et al., since our study used a cohort including those with a CHA2DS2-VASc score of ≥ 2 with higher thromboembolic risk, and more contemporary data with a preponderance of more expensive DOACs utilised for anti-coagulation [13].

Comparable to prior analyses of healthcare utilisation in people with AF, hospitalisation was a principal driver of costs in our population, particularly for those that were anti-coagulated. A recent costing analysis by Deshmukh et al. based on US insurance claims data reported significantly greater 12-month costs for a newly diagnosed AF population than a matched population without AF, primarily driven by the mean difference in all-cause inpatient admission costs and outpatient appointment costs [30]. As in our study, the economic burden of AF arising from utilisation of health services was particularly high in females, potentially arising from a greater frequency of presentation or referral to acute care services; females with AF are more likely to be symptomatic and managed conservatively, and thus may present with symptoms of advanced cardiovascular disease, associated with greater healthcare costs [31]. Hospitalisation as the principal cost driver in AF has also been recognised in studies forecasting the future economic burden of AF. Indeed, Burdett and Yip, utilising a prevalence-based approach, projected that the direct costs for AF in 2030 would be £2351 million, 1.11% of NHS expenditure, assuming a 3% annual increase in the AF prevalence, with secondary care admissions accounting for almost 60% of the overall AF costs [21]. Assuming an identical annual percentage increase in AF prevalence, our projection for 2030 was lower (£1445.9 million), but our estimate only considered the AF population in Scotland, did not include primary care costs, and the costs were based on those newly diagnosed with AF. The significant care home costs in those who never commenced anti-coagulation in our population likely reflect the considerably greater prevalence of dementia in this cohort compared to those that were anti-coagulated. In contrast, prescription costs and outpatient costs comprised a comparatively small proportion of the overall costs across all the cohorts in our AF population.

4.1 Strengths and Limitations

Our study used contemporary, high-quality, linked, national, individual patient data mapped to detailed costing data to provide a robust estimation of the healthcare resource utilisation, and associated costs, in people with AF in Scotland. Furthermore, few other studies have also considered the costs of care home admission in people with AF, despite forecasts projecting the costs of long-term nursing home to rise to £928 million by 2040, accounting for approximately 13% of indirect costs associated with AF [13, 21].

A limitation of the study is that we did not have access to patient-level data for primary care, and primary care costs, which meant that our cohort likely did not capture all those diagnosed with AF in Scotland during the observation period [32]. However, given the increased prevalence of AF with advancing age, we would anticipate that the majority would have at least one secondary care encounter, in which they would be captured in our dataset, which linked SMR01 inpatient data and SSCA data. Furthermore, primary care costs have previously been shown to comprise only a small proportion of the overall costs of AF, so their inclusion would likely have had minimal impact on our findings.

Since our population consisted of a secondary care AF population at elevated thromboembolic risk, who are likely to have greater clinical risk than individuals managed exclusively in primary care, our projected costs for Scotland may be over-estimated.

Furthermore, our prescribing data did not include those dispensed in secondary care, and thus it is possible that some people were misclassified as delayed initiators, despite being commenced on anti-coagulation in hospital. However, we would anticipate that most people initiated on anti-coagulation in hospital would have required at least one subsequent community prescription within the 60 days following diagnosis.

Our analyses did not consider adherence to anti-coagulation after initiation; discontinuation of anti-coagulation is associated with poorer clinical outcomes, and likely greater healthcare resource utilisation and costs [33].

4.2 Future Research

Finally, this study did not include sub-group analyses according to DOAC or VKA prescribing, although prior studies suggest that prescription of DOACs would reduce healthcare costs. Indeed, modelling by Blann et al. projected that the prescribing of edoxaban rather than warfarin in 75% of the eligible AF population would decrease thromboembolisms and deaths by 22,916 and 206,449, respectively [34]. The enhanced safety profile of DOACs in comparison to warfarin would mitigate costs associated with adverse events, particularly bleeding. Furthermore, based on the RE-LY, ROCKET-AF, and ARISTOTLE randomised controlled trials, undertaken from a US payer perspective, Deitelzweig et al. estimated that usage of rivaroxaban, dabigatran, and apixaban in preference to warfarin would decrease medical costs (US dollars) in a patient-year by $89, $179, and $485, respectively, due to a reduction in clinical events, including ischaemic stroke, haemorrhagic stroke, and pulmonary embolism; drug costs and monitoring costs were not included in these analyses [35]. The effect of discontinuation of anti-coagulation on healthcare resource utilisation and costs in people with AF should also be explored in future studies.

Future research on healthcare resource utilisation in AF should consider the effects of the forthcoming expiration of DOAC patents, and availability of generics, which would be anticipated to mitigate costs further [34].

Given that inpatient admissions are the primary cost driver in AF, it is integral that the future development of interventions aimed at mitigating healthcare resource utilisation focuses on reducing inpatient admissions in people with AF. Indeed, the potential for this has been suggested in prior work by Pastori et al., who reported in their Italian study that the clinical optimisation of people with AF per the Atrial fibrillation Better Care (ABC) pathway, an integrated, holistic management strategy that prioritises the proactive mitigation of cardiovascular risk, was associated with a reduction in healthcare costs of $3137 per person-year (US dollars) [36].

4.3 Conclusion

AF and its sequelae exert a substantial economic burden, with non-initiation and delayed initiation of anti-coagulation associated with increased healthcare utilisation, particularly in individuals who are female or frail. Healthcare costs in AF are primarily driven by inpatient costs, and to a lesser extent, care home costs; costs for prescriptions and outpatient appointments account for a comparatively small proportion of costs.