A systematic review of health state utility values for older people with acute myeloid leukaemia

A systematic review of the literature was conducted to identify articles reporting HSUVs of older people with AML in sufficient detail to be used in economic evaluations. The review was conducted and reported according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist and flow diagram [25]. Published guidelines on conducting systematic reviews of HSUVs informed the review to enhance quality, rigor, and reproducibility [20, 26, 27]. The protocol was registered prospectively with PROSPERO (CRD42022301344).

A broad and extensive search was conducted of electronic databases covering medical, nursing, and economic literature, including PubMed, EMBASE, CINAHL, PsycINFO, Cochrane (Database of Systematic Reviews and Registry of Controlled Trials databases), and EconLit for studies published from inception until February 2023. This broad time period was used to ensure no potentially relevant publications were missed. Forward searching of citations was conducted in Google Scholar as well as hand searching of reference lists of included studies. ProQuest Dissertation and Theses Database was searched for grey literature.

Eligibility criteria for the systematic review are presented in Table 1. Studies were included if they reported on HSUVs of people with AML aged 60 years or older. This age was chosen so no potentially relevant studies were missed, as the literature defines ‘older’ as 60–65 years and over. Studies that reported data all age groups of people with AML, or multiple cancers were included if HSUVs for older people with AML were reported separately. If data were not reported separately, at least 80% of the sample had to be older people with AML to be included. This age restriction was applied to capture HSUVs that accurately reflected the experience of older people with AML, who have unique issues compared to younger people.

Studies reporting HRQoL in sufficient detail to be converted to utility values with published algorithms were also eligible (HRQoL terms displayed in Supplementary file 1 – indirect valuation methods). Currently available published algorithms for HRQoL measures were sourced from a recent systematic review focused on HSUVs of people with a cancer diagnosis [28]. Studies that reported the EQ-5D Visual Analogue Scale alone (and not the individual domain responses or overall utility scores) were excluded. Published conference abstracts were included if they reported sufficient information on HSUVs and other eligibility criteria could be sourced from original publications (i.e., age of sample). Three relevant systematic reviews were used for reference cross-checking but not included [8, 21, 29].

A combination of key terms and synonyms, and Medical Subject Headings (MeSH) were developed, guided by the current related literature [8, 20, 21], and tailored to each electronic database (see Supplementary file 1 for search terms). Database search terms and results are displayed in Supplementary files 1 and 2.

Title, abstract and full-text screening of studies was conducted by two reviewers independently (EB, NG) against selection criteria in Covidence online systematic review software [30]. Ambiguities were discussed with other members of the research team (NM, HC).

Data were extracted independently by two reviewers (EB, NG) using a data extraction form based on the relevant literature and guidelines [20, 21, 27] and included publication details (year published, country), study aim, design, years of data collection, intervention and control (if applicable), sample (number characteristics of participants), age, description of HSUVs (measure of valuation and tariff), and the health state and utility values reported (via mean or median, and variance if reported). If feasible, HRQoL measures were extracted and converted to HSUVs using published algorithms. A range of non-preference based HRQoL scores can be converted to utility values including the Quality of Wellbeing Scale, Short Form Health Survey, Assessment of Quality of Life, and the Health Utilities Index [31]. The EORTC-QLQ-C30 is the most widespread HRQoL tool used in the oncology and AML literature that can be mapped to utility values [21, 29].

Quality assessment of included studies was assessed by two independent reviewers (EB, NG) using guidelines on searching and selection of HSUVs from the literature (developed from the NICE [National Institute of Clinical Excellence] Decision Support Unit Technical Support Document 9) [32]. These guidelines provide an overview of the key criteria to assess quality of data including the following: (1) sample size; (2) respondent selection and recruitment; (3) inclusion/exclusion criteria; (4) response rates to instrument used; (5) loss to follow-up; (6) missing data; (7) any other problems with the study; and 9) appropriateness of measure [32]. This method was chosen as it has been commonly used in other systematic reviews of HSUVs [33, 34], is easy and practical to use, and covers the key quality domains in the International Society Pharmaceutical Outcomes Research guidance on reviewing the literature on HSUVs [35]. Further quality assessment was performed for completeness, using the relevant CASP (Critical Appraisal Skills Programme) checklist [36]. These checklists were selected as they provide a clear, systematic, consistent approach to assess quality of various study designs [36].

Utility values were mapped against health states (see Table 2) according to the clinical characteristics of the study sample. These health states were developed based on clinical knowledge and published literature, and their validity was confirmed via expert review (two experienced haematologists [JB, TL], and two senior haematology nurses [AT, NG]. Due to the limited number of studies included in the review and differences between the studies in population and interventions evaluated, pooling data was considered inappropriate [20]. Additional data were extracted from studies with HRQoL measures reported in sufficient detail to be mapped to a relevant HSUV. Mapping algorithms were selected for the relevant HRQoL instruments once selection of studies was complete [37].

Database (n = 658) and hand searching (n = 3) identified 532 studies, following removal of duplicates. After screening, seven studies were included in the review (see Fig. 1. PRISMA flow diagram, and Supplementary file 4. Studies excluded after full-text review). Of note, 28 potentially eligible studies were excluded as they did not report HRQoL measures in sufficient detail to be mapped to utility values.

The seven studies included: one economic evaluation using data from a randomised clinical trial [38], two observational studies gathering HRQoL measures [39, 40], and four studies that used data derived from randomised clinical trials (reported data of entire sample) to validate a HRQoL tool [41] or describe HSUVs of patient cohorts [42‐44]. Characteristics of the included studies are presented in Table 3. Six studies included newly diagnosed patients [38, 40‐44], and one included people at any stage of the illness [39]. The combined total sample was 2,606 people, with ages ranging from median 61 (IQR 49–69) [40] to mean 76 years (range 41–91) [43]. One study included a mixed sample of people with acute leukaemias of various types, and reported data on people with AML separately [39].

The HSUVs reported in the included studies are displayed in Table 3 and mapped to health states in Table 4. Four studies measured HSUVs via the EQ-5D [40, 41, 43‐45] and one via the earlier EuroQol (no further description provided) [38]. Two studies reported European Organization for Research and Treatment of Cancer Quality-of-Life Questionnaire Core 30 (EORTC-QLQ-C30) data [39, 42]. The lowest HSUV (0.535) was reported for the health state of intensive therapy (day 2 after chemotherapy) and the highest HSUV (0.834) for people in controlled remission (6 months post-treatment) [39]. A ‘General AML state’ was used to categorise HSUVs where the health state of the sample was unclear. Data from two studies were mapped to in the ‘general AML state’: (1) Mamalo and colleagues reported data from people who were newly diagnosed or relapsed (with no description of treatment) ; and (2) Groot and colleagues reported data from people at home after hospitalisation (after daunomycin cytosine arabinoside) with the time frame not described [38]. Values ranged from 0.68 to 0.81 in this state (Table 3). No studies were found reporting HSUVs for low intensity therapy or supportive care, four reported HSUVs for newly diagnosed disease, five studies reported on the health states of controlled remission, and relapsed or refractory disease (see Table 4). Of note, utility values reported at baseline (day 1 or prior to chemotherapy) were categorised as newly diagnosed disease as they did not reflect treatment health states.

The only studies in the review which included convertible HRQoL measures administered the EORTC-QLQ-C30 [39, 42]. To allow for conversion, data on the individual domains (i.e., physical functioning, pain, global health status) within the EORTC-QLQ-C30 had to be reported, as opposed to an overall summary score. A published algorithm by Kim et al. (2012) in a mixed cancer population receiving chemotherapy was used to map EORTC-QLQ-C30 scores to EQ-5D utility values [46]. This algorithm was selected as it was developed in a population with similarities to the review population [37] and led to clinically plausible utility values (comparable to other utility values for the same health state). Calculations are shown in Supplementary file 3. Of note, no variance was available when HSUVs were converted via algorithms.

Quality assessment of included full studies and abstracts using updated guidance from the NICE Decision Support Unit Technical Support Document 9 [32] is displayed in Table 5. The authors’ overall assessment of quality (as determined in Table 5) is included in Table 4, to provide guidance on the level of confidence in each of the HSUVs. All studies had relatively large sample sizes (n = 212–576). Three studies were subjectively deemed by the authorship team to have low quality, relating to low response rates (6–52%, 15% [40]) and poorly described health states [38‐40]. One study was considered high quality due to the high response rate and broad selection criteria, and HSUVs were reported on a clearly defined health state. Of note, only two studies reported data from the broader population [39, 40], rather than drawing from clinical trials with stringent eligibility criteria. Where there were insufficient data to determine methodological quality of abstracts, related publications were sourced that reported relevant information in full (noted with reference in Table 5). The three abstracts that were included in the review [42‐44] were of moderate quality with clearly defined health states and good response rates, but had stringent selection criteria with data were drawn from RCTs.

Quality assessment of included full studies using the relevant CASP checklist is presented in Supplementary 2. Assessment of abstracts was not performed due to inadequate information reported in abstract format. Of the full text studies, Groot and colleagues’ [38] cost-effectiveness study had significant limitations due to unclear research question, lack of sensitivity analysis, and limitations in generalizability of findings to other settings. Mamolo et al. [40] and Lennmyr et al. [39] were limited by lack of information on the specific characteristics of participants, and consideration of how such confounding could bias study results. Piepert and colleagues [41] study scored consistently well in quality assessment, with only limitation relating to unclear application of results to a broader population (outside of a RCT).

This review highlights a substantial lack of published studies with easily accessible and comprehensively reported data on HSUVs for older people with AML that can be used in cost-utility analyses, particularly for health states related to treatment. The only study that reported on treatment-related health states (intensive therapy) reported utilities for an outdated chemotherapy regimen that no longer reflects current practice for frontline induction therapy of people with AML [47]. Further, two studies reporting HSUVs had insufficient data to determine the associated health state, limiting the usefulness for economic evaluation [38, 40]. A key finding is that no HSUVs were found for health states of low intensity therapy or supportive care. Rather most studies collected data within clinical trials of new drug treatments. While HRQoL has been reported for patients for low intensity therapy or supportive care health states, this gap in the HSUV literature limits cost effectiveness evaluations which include less aggressive therapies or usual care as a comparator, despite this potentially valid ‘next best’ treatment option. Quality assessment scores can be used to guide selection of the most reliable and appropriate HSUVs where multiple exist in a single health state. Other aspects to consider when choosing HSUVs are focused on relevance assessment criteria including: comparability between study and model, appropriate instrument used to describe health states, HSUV taken elicited directly from patient, and appropriate valuation of changes in HRQoL (i.e., general population using choice-based methods) [32].

The HSUVs reported in the included studies appear to be clinically relevant with values reflecting the illness trajectory of people with AML. The HSUVs for newly diagnosed and intensive therapy were lower, in line with the increased symptom burden and complications related to the underlying disease and chemotherapy experienced in these health states (i.e., fatigue, shortness of breath, bleeding, infections, nausea) [48]. Utility values during controlled remission were highest, reflecting a time of clinical stability with fewer symptoms and complications, and improved HRQoL [48]. Relapsed or refractory disease HSUVs declined over time as people drew closer to the end of life and experienced increasing symptoms and complications related to advancing disease (i.e., bone marrow suppression, pain, fatigue) [49].

The HSUVs included in this review were predominantly from EuroQol-generated instruments. There have been concerns expressed that these instruments may not be sensitive enough to capture all relevant domains of health for older people with AML [50, 51]. While EQ-5D is the most frequently used preference-based, generic method of obtaining utility values, there is mixed evidence regarding the responsiveness of the tool to detect meaningful change in all conditions [52]. Consequently, condition-specific measures that cover more cancer-relevant domains such as cognitive function [41] and fatigue have been used alongside the EuroQol tools [50]. Although the EQ-5D has been demonstrated to be appropriate for measuring change in HRQoL in patients treated with a palliative approach, the tool may not capture changes in symptom burden and holistic impacts of disease, which are important to people near the end of life [51, 53, 54]. The EORTC-QLQ-C30 captures more cancer-relevant domains of health and has been shown to be valid and reliable in people with advanced cancer [55] and is commonly used for older people with AML [29].

The EORTC-QLQ-C30 has traditionally been the most common HRQoL measure used for people with AML in observational and intervention based research [29]. During screening, many potentially eligible studies were excluded as their reporting of EORTC-QLQ-C30 data did not allow for conversion to HSUVs (where data on the individual domains were required rather than an overall summary score). Additionally, reporting of EQ-5D Visual Analogue Scale data alone were not able to be used as a utility value. Two high quality international clinical trials that collected HRQoL measures as a secondary endpoint were excluded from the review for this reason [56, 57]. Although HRQoL data were reported to address their study research questions, it was not able to be extracted as a HSUV or converted to a HSUV if it was reported as a mean difference, hazard ratio, for a single domain (i.e., fatigue, as a global health score, or as a summary score. Furthermore, a recent cost-effectiveness analysis of older people with AML included unpublished utility values sourced from a pharmaceutical company, which were not publicly available and had not undergone peer review. This highlights a potential gap for researchers when seeking accessible HSUVs to be used in economic evaluation. This could be due to a possible lack of awareness in the research community around the value of, and how to report HRQoL data that can be used in economic evaluation. Additionally, challenges may exist getting HSUV data published in appropriate journals.

Inadequate and mixed reporting of HRQoL data is a widespread issue associated with considerable research waste [58, 59]. The 30 excluded studies due to insufficiently reported HRQoL data could have filled important gaps in knowledge for treatment-related health states in this review (these studies have been noted in Supplementary file 3). Contacting authors from individual studies and seeking access to data is time consuming, often results in no reply (or bounced back emails), and can lead to delays and challenges regarding data sharing [60‐62]. Ideally, HSUV data should be easily accessible in publications or open access data repositories. Furthermore, it is argued that appropriate and meaningful HRQoL measures are often not adequately measured in haematology clinical trials, leading to missed opportunity to collect and report valuable data. Greater awareness is needed so that variable reporting of data does not limit the full impact of research work, and lead to redundant, biased, underpowered and repetitive studies [58, 59]. Clinical trials in haematology should prioritise HRQoL as an important outcome that reflects the patient’s experience, and can be used to conduct economic analysis.

Researchers need appropriate and validated published algorithms to enable mapping of HRQoL measures to HSUVs. In this review, only studies with currently available algorithms were included, limiting studies reporting HRQoL data in potentially relevant populations. For the included studies, careful selection of a published algorithm was required to ensure clinically plausible HSUVs were calculated. Initially a published algorithm developed on a sample of women with locally advanced breast cancer with relatively good health was proposed for conversion of the scores to HSUVs [63, 64]. The algorithm developed in this population had previously been used to map HRQoL to utility values in people with AML and the population in the study of older AML patients [39] had similar clinical characteristics to the population used to generate the algorithm [63, 64]. This is recommended as a guiding factor when selecting an algorithm for mapping [37]. However, this initial algorithm returned utility values that were lower than clinically expected or plausible. Therefore, an alternative published algorithm by Kim and colleagues in a mixed cancer population receiving chemotherapy was used to map utility values which produced clinically plausible results [46]. Ideally, primary HSUV data should be collected and reported to avoid such issues and potential inaccuracies in data. Where this is not feasible, appropriate algorithms must be available to convert HRQoL data.

This review expands on similar previous systematic reviews of people with AML [8, 21, 29] by including six subsequently published studies and abstracts, enabling researchers to efficiently access up-to-date HSUVs to be used in cost-utility analyses. The main focus of the review by Stauder and colleagues was to understand which patient reported outcomes are used to measure HRQoL, therefore individual studies were not discussed or referenced (only meta-data presented), limiting usefulness for economic evaluation [29]. The 2018 review by Forsythe and colleagues focused on HSUVs derived from primary and secondary data of people with AML of any age [21].

The few studies included in this review provide valuable HSUVs that can be used in economic evaluations of interventions for older people with AML. However, the quality of the included studies varied, potentially limiting the usefulness of certain utilities; one study was outdated [38], another had a large amount of missing data [39], and insufficient details were provided to allow mapping of a HSUV to a health state [40]. The low number of included studies should be a driver for future research of older people with AML. The results demonstrate that more research is needed reporting HSUVs of older people with AML, particularly in treatment-related health states. Due to paucity of data, and low quality research, high quality studies reporting HSUVs for the following health states are desperately needed: (1) intensive therapy; (2) low intensive therapy; (3) and supportive care. Further high quality research reporting HSUVs for health states of controlled remission, and relapsed or refractory disease are also required. Additionally, future studies of HSUVs for older people with AML should be conducted in real-world settings rather than clinical trials to promote generalisability of the utility values.

Intensive therapy, low intensity therapy, and supportive care for people with AML are all resource intensive interventions [12, 13]. Major drivers of costs include hospitalisation and medication [14, 15]. Hospitalisation costs are likely to reduce over the coming decade as more treatments are offered in out-patient settings [14]. However, medication costs are expected to increase with the continued development of novel drugs [14]. Greater numbers of older people with AML who are unsuitable for intensive therapy will be eligible to undergo expensive novel targeted therapies that may prolong their life [14], increasing the costs of treatment of AML [13, 14]. It is vital that the value of such therapies relative to other options is rigorously evaluated to inform resource allocation decisions.

A strength of this review was the broad search strategy conducted across multiple databases, with detailed search terms, including studies of any date. The review was conducted with rigorous methods according to published guidelines. Additionally, the inclusion of HRQoL measures that can be converted to utility values with algorithms, reduced the likelihood of missing studies with relevant data. Forwards and backwards citation tracking, and cross referencing with previous reviews reduced any potential gaps in the search results. A limitation of this review was that only studies in English were included. It is unlikely many publications were missed as most (98%) scientific journals are published in English [65].