Systematic review of societal costs associated with stroke, bleeding and monitoring in atrial fibrillation

The SLR was conducted according to the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines [27] and the Cochrane Handbook for Systematic Review of Interventions [28]. Details on data sources and study selection criteria are provided below.

The SLR identified 2862 unique records through electronic database searches. Of those, 184 were included in the review. One meeting abstract was identified from International Society of Pharmacoeconomic and Outcomes Research and one study from manual searches met the inclusion/exclusion criteria, resulting in 186 total publications (159 unique studies) with direct costs, resource use and/or indirect costs. Among the 186 publications, 15 reported indirect costs (14 unique studies and one related publication reporting on the same study population) [34–47]. Study attrition through the abstract and full-text screening levels are detailed in the PRISMA diagram in Figure 1.

Among the 14 indirect cost studies, eight were observational [34–36,39,41,43,44,47] and six were model-based, including economic evaluations or retrospective cost-of-illness studies based on published estimates reporting indirect cost inputs for their economic models [37,38,40,42,45,46]. Among the observational studies, two each were conducted in Denmark and Ireland and one each in the UK, Russia, Mexico and India. Indirect costs were evaluated for stroke in three studies, major bleeding in one and INR monitoring in four. Four studies specified that the population had NVAF, another four did not specify AF type and one included atrial flutter. Three studies were retrospective cohorts, three were prospective cohorts and two were cost-of-illness studies with real-world estimates. Five (62.5%) were funded by industry. Three studies were of high quality based on the GRACE quality checklist – studies which had a response of yes or not applicable to all or all but one checklist item, were rated as high quality. There were no trends in study quality by funding source. The remaining observational studies were of moderate quality since they lacked clear reporting of treatment, control of covariates and/or measurement of primary outcomes.

Among the model-based studies, two were conducted in Sweden, two were from the USA, and one study each was based in Denmark and Thailand. Three studies evaluated indirect costs of stroke and four addressed INR monitoring. No model-based analyses reported cost inputs for the indirect costs associated with major bleeding. Four studies specified that patients had NVAF, two did not specify type of AF, and one allowed inclusion of atrial flutter. Three studies were cost–effectiveness analyses (CEA), two were cost-of-illness studies with estimated population costs, and one was a budget impact model. Three studies (50%) were industry funded and half were of high quality based on the Drummond assessment, despite their funding source.

The indirect costs results were summarized by event type (stroke, major bleeding and INR monitoring) then by outcome type (indirect costs, indirect cost in relation to total cost) in the following sections.

Six studies, three of which were observational in design, reported data on the indirect costs of stroke among patients with AF. Model-based studies included one economic evaluation and two retrospective cost-of-illness analyses informed from published data. Two studies characterized the population to have NVAF, with the remaining four did not specify AF type. Four studies were conducted in Europe (Ireland, n = 1; Denmark, n = 1; Sweden, n = 2) and two studies were conducted in Asia (Thailand, n = 1; India, n = 1). The mean age of patients ranged from 61.4 to 78 years, where reported. The timeframe for which costs were reported varied (ranging from 1 to 2 years, or unspecified), with observational studies reporting costs as a function of time since onset of stroke. Model-based studies provided limited details on the timeframe. The study quality was rated as high in four studies and moderate in two. Data on monetary indirect costs associated with stroke are summarized in Table 2.

Four studies reported on productivity losses (or sick days) for patients, due to inability to work because of stroke [37–39,41]. Only three studies provided monetary values for productivity losses [37,39,41], with one reporting on the average number of sick days [38]. Nonmonetary costs (hours, days and proportions of patients) are summarized in Table 3. The number of cases and average duration of sick leave was reported by stroke type in one study (ischemic stroke [IS] vs unspecified type). The average duration of sick leave was similar by stroke type (403 vs 353 days, respectively [38]). Of the remaining studies providing monetary values for productivity loss, two observational studies used the human capital approach to estimate these [39,41]. One cost-of-illness analysis, conducted in Denmark, estimated lost earnings among patients aged 18–65 years, assuming that patients older than 65 yearswere unemployed [41]. This was the only study that presented costs at different timepoints after stroke onset [41]. Productivity loss attributed to stroke alone increased threefold from the year of the incident stroke up to the 1-year follow-up (from $288 to $843). A twofold increase was reported from 1 to 2 years following the stroke incident (from $843 to $1444) [41]. The remaining observational study, conducted in Ireland, collected data on employment status at the time of stroke onset and length of hospital stay. Productivity losses were estimated using the days spent in hospital and assigning the national median wage to patients who were employed and the minimum wage to stroke patients who were homemakers [39]. The average productivity loss at the end of a 2-year follow-up after stroke period was $930 [39]. A cost-of-illness analysis conducted in Sweden reported a productivity loss of €15,145 over lifetime post stroke (IS or hemorrhagic stroke); published estimates were used to inform the analysis [37].

Two studies, a retrospective cohort [44] and a CEA [42], considered the costs associated with informal care by unpaid caregivers. The observational study relied on estimates of informal care hours provided by patients, without providing details on assumptions used to convert informal care hours to a monetary value [44]. Caregivers spent an average of 3.2 h per day on informal care, representing an average annual loss of 13,370 [44]. The CEAs, conducted in Thailand, utilized informal care (not defined) costs per month by IS severity to inform their model. The cost inputs were obtained from published estimates and were higher for severe IS as compared with mild-to-moderate IS (฿12,260.59 vs ฿5023.93, respectively [42]).

The contribution of indirect costs to total cost was marginal in the two studies, with short-term follow-up data reported (1.1–1.4% [39,44]). In the study with 3-year data, indirect costs appeared marginal in relation to total costs in the year of the incident stroke event (1.4% [41]). However, the relative contribution of indirect costs increased to 11% of total costs in the first year of follow-up, and rose to 27.9% in the second year following the index event. In the cost-of-illness analyses that utilized lifetime costs, indirect costs made up almost a fifth of total costs (19.6% [37]).

The SLR identified only one study [43], conducted in Denmark, which reported indirect costs of bleeding among patients with AF. Data were used from a large Danish registry (n = 21,046) of patients with unspecified AF type or atrial flutter who were hospitalized from 1994 to 2012 to estimate indirect costs attributable to intracerebral hemorrhage (ICH), gastrointestinal (GI) bleeding, and other major bleeding events (not specified). Costs were estimated during the year of the bleeding event and over two additional years of follow-up. The average 3-year cost of productivity loss due to ICH events was estimated to be higher than that of GI and other major bleeding events. Total costs per patient were €2042 for patients with ICH compared with €317 and €1218 for GI and other major bleeding, respectively (Table 4). The productivity loss associated with ICH events amounted to a cost of €194 per patient in the incident year and increased sixfold to €1438 per patient during the 2 years after the event. By contrast, only a small increase was observed between the incident year and the 1-year follow-up among patients with GI bleeding. This cost decreased by 43% from the incident year to 3-year postevent. For other major bleeding events, an increase of about €200 per year per patient (50%) was observed. The study was rated of moderate quality due to the lack of reporting on treatment and analyses to test key assumptions that could impact results. Table 3 summarizes the characteristics of this study and the indirect costs per patient.

The Danish study reported both total direct and indirect costs. For ICH and other major bleeding events, the contribution of indirect costs to total costs was marginal in the incident year of the bleeding event as direct costs, due to hospitalization, were the initial cost driver. Indirect costs driven by productivity loss accounted for as much as 33% of total costs 2 years after an ICH event, and 48% of total costs for other major bleeding events (Table 4) [43].

The SLR identified eight studies (nine publications) [34–37,40,45–47,50], four of which were observational [34–36,47,50], reporting indirect costs associated with INR monitoring among patients with AF. The other four studies included two CEAs [40,45], a cost-of-illness study [37], and a budget impact model [46]. Two studies were conducted in the USA, and one each in the UK [51], Mexico [52], Ireland [46], Russia [36,50], Denmark [46] and Sweden [37]. AF type was specified to be nonvalvular in the majority of studies [34–36,40,45,46,50]. The mean age of patients ranged from 63.2 to 72.3 where reported. Costs were most commonly reported annually per patient or per test [34–37,40,47,50]. Three out of eight studies were rated of high quality and the remaining five were judged to be of moderate quality.

Indirect costs included productivity losses (worktime loss) and opportunity losses (lost leisure time) for patients and caregivers due to INR monitoring. Only three studies, two by the same first author [35,36,50] provided sufficient details on how indirect costs were defined and valued. The two observational studies by the same author, one based in Mexico and the other in Russia, reported costs in US dollars (USD) and were similar in both design and costs reported [36,52]. Both studies were conducted in the hospital setting, used a costing year of 2016, and reported data for patients with a mean age of approximately 65 years. Worktime loss was valued separately for patients who were employed, patients unemployed due to VKA use, and patients unemployed not due to VKA use. The median wage was used to determine productivity losses for employed patients and those without employment due to VKA; while the minimum wage was used to value the opportunity cost of lost leisure time for patients whose unemployment was unrelated to VKA. The cost of worktime loss for employed patients was similar in the two countries (Russia: $186.90 vs Mexico: $164.10) but the cost of opportunity loss for unemployed or retired patients was much higher in Russia compared with Mexico ($4458.30 vs $1337.20, respectively). The time spent at appointments and the hours taken off from work (among employed patients) was similar across both studies (Table 3). A retrospective study conducted in Ireland adopted a similar approach valuing lost time separately for employed and unemployed patients [47]. The average worktime lost/opportunity cost per patient per visit was €20.4, based on an average of 2.13 h spent at the clinic. The final observational study reported the mean cost of time off work for patients and caregivers of £1.2 and the mean travel costs of £33.9; however, there were limited details on whether the time off work was only composed of productivity losses and how those losses were valued [34].

Although all model-based evaluations reported indirect costs based on lost time due to INR monitoring, the cost components and how they were valued were less clear. The two CEAs used model inputs representing economic value of patient time for INR tests, at an average of about $1751 yearly or $153 per test [40,45]. The cost-of-illness analyses reported that the productivity loss per patient was €2.9 [37]; however, no further details were provided on the estimates. Finally, the budget impact analysis reported that the total societal cost associated with INR monitoring was $1212 [46].

Only three studies reported on the indirect costs associated with lost time of caregivers [34], with one providing limited details on valuation [34–36,50]. The remaining two observational studies used the national median wage [35,36,50] to assign value to caregiver time to attend INR visits. No distinction was reported for employed versus unemployed caregivers. The indirect cost for caregivers was different across the two studies. In the study conducted in Mexico, 70.8% of patients were accompanied by caregivers to their usual hospital appointments in contrast to 17.9% in the Russian study. The cost of worktime/opportunity loss for nonprofessional caregivers in Mexico was double that of Russia ($66.3 vs $34.7, respectively), but the worktime/productivity loss in hours per caregiver was lower in Mexico than in Russia (2.6 vs 5.7 h). All indirect costs of INR monitoring by study are summarized in Table 5.

Based on the available data, the contribution of indirect costs to total costs was calculated for each study and varied greatly, ranging from 1 to 73%. Indirect costs contributed to only 1% of total costs in a UK retrospective study, which appeared to only consider productivity losses [34]. Higher estimates were observed among studies considering the economic value of, or opportunity cost of, lost leisure time of patients [36,45].