Cost-Effectiveness of an Individualised Management Program after Stroke: A Trial-Based Economic Evaluation

Stroke is a leading cause of death and disability in Australia, affecting over 27,000 Australians and imposing an estimated 6.2 billion Australian dollars (AUD) in economic costs annually [1]. Therefore, it is important to implement evidence-based and cost-effective interventions for secondary stroke prevention, such as organisational disease management programs to potentially improve blood pressure [2], or nurse-led interventions to improve risk factors [3]. Furthermore, there is emerging evidence that pragmatic nationwide multidisciplinary post-stroke care programs with proper reimbursements, such as those provided in Australia, could reduce mortality and morbidity after stroke and improve quality of life (QoL) [4]. Consequently, an integrated care plan aimed at preventing recurrent stroke, improving functional and psychological status, and managing risk factors and comorbidities is recommended to optimise the long-term care of people with stroke [5].

In Australia, financial incentives are provided through the government-funded universal health insurance scheme (Medicare) to support primary care physicians (PCPs) to use individualised care plans for the ongoing management of people with chronic diseases (e.g., stroke). PCPs can claim reimbursement for chronic disease management (CDM) Medicare Benefits Schedule (MBS) items for developing and using these care plans [6]. Such claims have doubled over the last decade and are estimated to cost nearly AUD 1 billion annually [7]. In 2016, about one in two people with strokes were provided with these CDM plans [8]. Compared to survivors of stroke or transient ischaemic attack (TIA) without a claim for CDM items, those with a claim have 26% better survival between 19 and 30 months after stroke/TIA [9]. There is evidence from a model-based study that organised care for secondary stroke prevention is more cost-effective than usual care (UC) in Australia [10]. However, there is limited evidence on the cost-effectiveness of these disease management interventions in stroke based on patient-level, randomised controlled trial data. We assessed the cost-effectiveness of an individualised management program (IMP) for survivors of stroke/TIA, comprising nurse-led education in addition to provision with a CDM plan, to enhance the UC provided by their PCPs.

The economic evaluation was conducted alongside an RCT with a follow-up of 24 months from baseline, which was approximately 11 weeks after hospital discharge.

A multicentre, cluster RCT with a blinded assessment of outcomes was primarily designed to determine the effectiveness of an IMP in patients with stroke/TIA compared with UC [11‒13]. Patients hospitalised with stroke/TIA were recruited from four hospitals in Melbourne, Australia. Eligible participants were aged ≥18 years and lived within 50 km of the closest recruitment hospital. Patients were excluded if they were participating in another clinical trial, discharged to a nursing home, or were not expected to survive the duration of the study due to a rapidly deteriorating condition/disease. More details on participating hospitals, selection of participants, reasons for inclusion and exclusion of participants, and findings on primary and secondary health outcomes have been published elsewhere [11‒17]. Written informed consent was obtained from participants or their proxies. The trial was registered in the Australian and New Zealand Clinical Trials Registry (ACTRN12608000166370).

Participants were randomised by primary care practice to receive either UC or the intervention (in addition to UC) provided at baseline, 3 months, and 12 months. This IMP comprised: (1) a CDM plan developed by a nurse using a co-designed template aligned to the national stroke clinical guidelines that was also reviewed by a stroke specialist prior to mailing to the participant’s nominated PCP; and (2) stroke-specific tailored health education provided by a study nurse at the participant’s home. CDM plans included recommendations on the use of medications, the management of risk factors or comorbidities through adherence to recommended measurement thresholds and the management of mood. CDM plans were designed to meet the criteria that would ensure the PCP could be remunerated for using this plan, under the CDM items funded by the MBS. At 6 months and 18 months, a nurse reviewed CDM plans based on telephone assessments and mailed this revised plan to the PCP. Participants in the UC group received care as per the standard discharge care arrangement of the hospital and usual primary care practice. Study participants, outcome assessors, and stroke specialists were blinded to group allocation.

This economic evaluation was conducted using both health system and societal perspectives [18]. Costs were estimated by applying unit costs (online suppl. Table I; for all online suppl. material, see https://doi.org/10.1159/000535638) in AUD for the 2021 reference year to each resource item utilised. We prospectively collected self-reported data on resource use and employment loss at 3 months, 12 months, and 24 months after baseline using a standardised cost assessment questionnaire. A discount rate of 5% was applied to all costs incurred in the second year, as recommended by the Australian Government [19].

The costs of intervention delivery included the time costs of stroke specialists and nurses who developed and revised CDM plans and the travel costs of nurses who provided stroke-specific education during home visits to participants. We also included costs for the development or review of a CDM plan from the MBS. The average per-person intervention cost amounted to AUD 683 (online suppl. Table II).

We estimated healthcare costs using self-reported data on the following healthcare resource items: rehospitalisation (number and type of occasions, and length of stay in days), inpatient rehabilitation (number of occasions and length of stay in days), outpatient rehabilitation (number of sessions), rehabilitation at home (number of sessions), ambulance (number of trips), primary care (frequency of visits), specialist care (number and type of service), allied health (number and type of service), respite care (number of days attended), medical tests (number and type of tests), and medications (number and class of prescriptions).

The following non-healthcare resource items were collected from all participants: community services (number and type of services), special equipment and aids (type of equipment/aids and quantity purchased), home modification (incurred costs), nursing home (number of days), informal care (type and hours of care received over the last week), and employment changes (employment status prior to and post-stroke/TIA). The cost of employment loss was estimated based on the average weekly total cash earnings by age group, sex, and employment status (online suppl. Table III).

Quality-adjusted life years (QALYs) were estimated from utility scores obtained from responses to the Assessment of Quality of Life 4-D (AQoL-4D) instrument at baseline and at 3 months, 12 months, and 24 months after baseline [14]. The AQoL-4D utility score ranged from −0.04 (worse than death) to 0 (equivalent to death) to 1 (full health) [20]. Participants who had died during the study were assigned a utility score of zero (n = 12 in UC and 10 in intervention). When QoL assessment was missing at baseline (n = 1 in UC), the 3-month value was used instead. For those with missing QoL assessments at 3 (n = 8 in UC and n = 13 in the intervention) or 12 months (n = 19 in UC and n = 19 in the intervention), changes in QoL were assumed to be linear between assessments. Participants who dropped out of the study were excluded. A discount rate of 5% was applied to the QALYs of the second year, as recommended by the Australian Government.

The economic evaluation was undertaken based on an intention-to-treat analysis. A median regression model adjusted for age and sex was used to determine differences in per-person QALYs and costs between the study groups. An incremental cost-effectiveness ratio (ICER) was calculated by dividing the difference in average total costs per person between two groups by the difference in average QALYs per person. Probabilistic sensitivity analyses were undertaken to characterise uncertainty, with costs and QALYs bootstrapped with 10,000 iterations. Bootstrapped incremental costs and QALYs were plotted on a cost-effectiveness plane. The intervention was considered to be cost-effecitve if the ICER was below the willingness-to-pay (WTP) threshold of AUD 50,000/QALY [21]. The probability of cost-effectiveness of the intervention was determined by quantifying the proportion of bootstrapped incremental costs and QALYs below the willingness-to-pay threshold. Analyses were performed using STATA version 16.0 (StataCorp LLC, College Station, TX, USA).

Of 563 participants with stroke/TIA recruited in the trial, 61 (10.8%) refused further participation or were lost to follow-up (n = 29 in UC and n = 32 in the intervention), and 502 (65% male, median age 69 years) were analysed, with 251 (50%) in the intervention group. Baseline characteristics were similar between the two trial groups [14]. There was no statistically significant difference in baseline characteristics between participants with and without missing data, except that the proportion of intracerebral haemorrhage was greater among participants with missing data [14]. Intervention fidelity ranged from 92% to 99% across time periods.

Participants in the UC group utilised slightly more healthcare resources (hospitalisations, inpatient rehabilitation, and ambulance transfers) than those in the intervention group. Resource use and loss of employment for 12 months are provided in online supplementary Table IV and for 24 months in Table 1. There was no statistically significant difference in per-person total costs between the groups within 12 months (β = -AUD 1,220, 95% CI: -AUD 4,976; AUD 2,534) and 24 months (β = -AUD 6,014, 95% CI: -AUD 14,001; AUD 1,972). Similarly, no statistically significant difference was found in per-person healthcare costs between the groups within 12 months (β = AUD 676, 95% CI: -AUD 358; AUD 1,711) and 24 months (β = AUD 376, 95% CI: -AUD 1,961; AUD 2,715).

Within 24 months, average per-person costs from a societal perspective were greater for the UC group (AUD 51,394, 95% CI: AUD 43,167; AUD 59,621) compared to the intervention (AUD 49,045, 95% CI: 39,127; AUD 58,962). From a health system perspective, it was the reverse (AUD 20,232, 95% CI: AUD 16,808; AUD 23,655 for the UC group and AUD 21,707, 95% CI: AUD 16,929; AUD 26,485 for the intervention). More details on costs within 24 months are provided in Table 2. Costs within 12 months are provided in the online supplementary Table V.

Mean QALYs for the intervention group were greater than those in the UC group by 0.028 within 24 months (Table 2). There was no statistically significant difference in per-person QALYs between the groups within 12 months (β = 0.006, 95% CI: −0.051; 0.063) and 24 months (β = 0.031, 95% CI: −0.070; 0.133) after adjustment.

Incremental costs of AUD 2,112 within 12 months resulted in an ICER of AUD 136,363/QALY from a health system perspective. From a societal perspective, incremental costs of AUD 1,524 translated to an ICER of AUD 87,027/QALY (Table 3). Within 24 months, the ICER was AUD 53,175/QALY from a health system perspective, but from a societal perspective, the intervention was dominant (i.e. less costly and more effective compared to UC).

There was a 32.7% probability of the intervention being cost-effective at a WTP of AUD 50,000/QALY from a health system perspective and 42.8% from a societal perspective within 12 months (online suppl. Fig. I). The probability increased to 46.7% from a health system perspective and to 60.5% from a societal perspective within 24 months (Fig. 1). There was considerable uncertainty, as shown in the cost-effectiveness planes from both perspectives at 24 months (Fig. 2-3) and 12 months (online suppl. Fig. II, III).

We provide evidence from a randomised trial of people living with stroke/TIA that the cost-effectiveness of an IMP, aligned to real-world primary care policy, is more likely to be detected beyond 12 months of follow-up. This evidence is from both societal and health system perspectives. According to studies, interventions with a probability of being cost-effective at or above 40% are recommended by decision-making committees [22]. In Australia, the WTP threshold for a QALY is approximately AUD 64,000. [23] Further decision analytic modelling using all available evidence should be undertaken to understand the long-term return-on-investment from the policy of CDM plans for managing stroke [24].

To the best of our knowledge, there have not previously been trial-based economic evaluations of CDM plans for stroke/TIA in Australia using patient-level data. Although it is difficult to compare economic evaluations of post-stroke care interventions across countries due to their complexity and differences in methodology, our findings within 12 months were in accordance with those from two previous trial-based studies with a 1-year follow-up. [25, 26]. The authors of these prior analyses concluded that stroke care incorporating structured and individualised care planning was not cost-effective over UC or an education-based intervention from both health system and societal perspectives [25, 26].

There is evidence that disease management programs are cost-effective among people with medical conditions other than stroke/TIA. A CDM plan in primary care was cost-effective among Australians with coronary heart disease, but the estimates used in the study were projected and based on assumptions rather than patient-level data [27]. Researchers found that a nurse-led disease management program was cost-effective in a cluster RCT of 1,163 patients with coronary heart disease and chronic heart failure [28]. Self-management programs were also cost-effective in treating diabetes, as shown in a systematic review of 22 economic evaluations [29]. However, the reviewed studies were “predominantly of poor quality.” Similarly, authors of another systematic review of 22 economic evaluations assessing the cost-effectiveness of self-management interventions in chronic conditions (cardiovascular diseases = 3, diabetes = 4, others = 15) reported moderate methodological quality of the reviewed studies [30].

Our study was limited by the fact that we were unable to verify whether PCPs implemented CDM plans as recommended. However, because we had access to claims data, we discovered that the UC group was contaminated with some components of the intervention, as some participants in that group had claims for CDM items. Despite this, the proportion of our participants with a CDM claim was greater in the intervention group (73% at 12 months and 81% at 24 months) than in the UC group (33% at 12 months and 47% at 24 months) [31]. It is also important to note that our study, like many other economic evaluations, was not powered to detect differences in costs and QALYs between study groups, as designing a trial powered on cost-effectiveness would make that trial overpowered on clinical outcomes [32]. Therefore, our findings should be interpreted with caution. Additionally, long periods between data collections on resource use could pose a potential risk for recall bias. To address this, we are currently undertaking a study to validate the costs estimated from self-reported data on service utilisation in this clinical trial with government-held administrative data. Finally, missing data (11%) could have exaggerated or underestimated our costs and effects. These limitations and the use of an arbitrary WTP threshold, which may not reflect the true WTP threshold, should be taken into consideration when interpreting our findings.

Our study has several strengths. Firstly, we used a robust design for the economic evaluation alongside an RCT with a bottom-up approach to detailed cost estimation. Secondly, the cost-utility type of our economic evaluation allows the results to be compared across interventions and disease groups. Thirdly, we extended the follow-up period to 24 months, which allowed us to detect the effects of the intervention beyond the commonly funded 12-month period. Fourthly, we adopted a societal perspective, which provided a broad consideration of all incurred costs, making it particularly relevant for public health interventions. Finally, we used patient-level data collected by trained research personnel, which minimised assumptions.

In conclusion, we found that using an IMP for people with stroke/TIA was more effective and less costly than UC from a societal perspective over 24 months. However, from a health system perspective, the intervention was generally cost-effective for at least half of the iterations. In taking the field forward, we recommend that economic evaluations of post-stroke care programs have sufficient time horizons and consideration of costs beyond direct healthcare utilisation to demonstrate their value to society.

We appreciated the dedication and tireless efforts of the research nurses. The contribution of the participants, participating hospitals and the research team is acknowledged. We also thank Sharyn M. Fitzgerald (Monash University) and Judith Frayne (Alfred Hospital) who contributed to this study.

Ethics approval was obtained from relevant hospitals: Alfred Hospital Research and Ethics Committee (91/08), Eastern Health Research and Ethics (E101/0910), Southern Health Human Research Ethics Committee (10102B), Monash University Human Research Ethics Committee (EO2016/4/325). This research was undertaken with appropriate written informed consent of participants or guardians.

The authors have no conflicts of interest to declare.

The STANDFIRM trial was supported by the National Health and Medical Research Council project Grant (586605).

Z.O., M.T.O., A.G.T., D.A.C., and J.K. contributed to the study design and inception. M.T.O., A.G.T., D.A.C., T.P., M.R.N., D.U., V.K.S., C.F.B., R.P.G., and J.K. contributed to the design of the trial, acquisition, and collection of the trial data. Z.O., J.P., and J.K. collected information on unit costs. Z.O. undertook data cleaning and statistical analyses and drafted the manuscript. All authors provided a revision of the manuscript for critically important intellectual content. All authors read and approved the final version of the manuscript.

The data that support this study will be shared upon reasonable request to the corresponding author.