Systematic Review and Meta‐Analysis of Health State Utility Values for Osteoarthritis‐Related Conditions

Health state utility values (HSUVs) are a key input in health economic modeling, but HSUVs of people with osteoarthritis (OA)–related conditions have not been systematically reviewed and meta‐analyzed. Our objective was to systematically review and meta‐analyze the HSUVs for people with OA.


INTRODUCTION
Osteoarthritis (OA) is one of the most common chronic joint diseases. It mostly affects knees, hips, and small joints of hands. OA is characterized by joint pain, stiffness, swelling, loss of function, and disability, which in turn negatively impact individuals' health-related quality of life (HRQoL) (1) and incur a substantial socioeconomic burden (2,3). Currently, there is no cure for OA, but many treatments and approaches, including lifestyle, medications, injections, and surgery, are available to help relieve disease syptoms.
Health state utility values (HSUVs) are typically used to reflect HRQoL and to calculate quality-adjusted life years, a preferred measure of clinical effectiveness in cost utility (CUA)/clinical effectiveness analyses (CEA) (4). HSUVs measure the strength of a preference for a particular health state, represented as a number between 0 (death) and 1 (optimal health). Health states worse than death may exist, with negative HSUVs assigned (5). HSUVs can be obtained through several methods (6). Direct methods ask individuals to describe and assess health states and place weights on them, using valuation techniques such as the standard gamble, time trade-off, and rating scales (6). Indirect methods involve the use of preference-based multi-attribute utility instruments (MAUIs), where patients answer questions relating to multiple dimensions of their current health state, and the responses are then scored using a value set obtained from respective general populations. Commonly used MAUIs include the EuroQoL 5-dimension (EQ-5D), the Health Utility Index, the Short Form 6-dimension (SF-6D), and the Assessment of Quality of Life (AQoL) instruments (7). Finally, mapping techniques are used to transform nonpreferenced-based HRQoL measures into HSUVs.
As the stated preference data for a set of health states for an appropriate population are not always available, HSUVs obtained from the literature are widely used in economic evaluations (4). These HSUV estimates may differ from each other due to several factors, including differences in the utility elicitation techniques, MAUIs, the choice of respondent, sample size, and the quality of studies (4). With an increasingly growing literature of HSUVs, the selection of which values to use in economic evaluations becomes challenging. The correct choice of HSUVs is important to accurately calculate quality-adjusted life years and other CUA outcomes. To obtain the best estimate for a decision-analytic model from the literature, the methods of identification of the data should be systematic and transparent. To date, there is no systematic review and meta-analysis that summarizes estimates of OA-related HSUVs and evaluates the extent of differences between various subgroups of patients based on affected OA joint sites, treatments, and utility measurements. Our systematic review and meta-analysis aimed to generate a database of OA-related HSUVs to address this gap.

MATERIALS AND METHODS
Protocol registration. The study protocol was registered on April 17, 2019 at PROSPERO (#CRD42019129408). Our systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (8).
Literature search. Based on previous recommendations (9), 4 databases were searched from their inception up to March 2019: Embase, Health Technology Assessment database, Medline, and Scopus. This search was supplemented by hand searching the bibliography lists of all included articles and relevant reviews. The search strategy was developed in consultation with co-authors based on the previous literature (10,11 Screening criteria. Title/abstract screening and full-text screening were conducted in Covidence (12) (an online systematic review program to manage and facilitate the selection of studies) by 2 reviewers (TZ and HA) independently based on predefined criteria. Any disagreements were discussed between the 2 reviewers, and a third reviewer (AJP) was consulted in cases of no consensus. Studies were included if they involved humans, reported OA-related HSUV estimates (excluding those based on mapping techniques), and were published in English, Chinese, or German. Conference abstracts were included when adequate data were available for extraction. If the OA patients were part of a broader study population, we included studies reporting on a cohort with ≥80% OA representation. Health economic modeling studies based on HSUVs reported elsewhere and those based on systematic reviews or meta-analyses were excluded. Review reports, books, and case reports were excluded.
Data extraction. A predefined Microsoft Excel spreadsheet was piloted to extract data from 20% of studies by the first author (TZ). Adjustments and improvements were made to the initial spreadsheet where necessary, and the improved spreadsheet was then used to extract data independently by TZ and HA. Discrepancies were resolved by consensus, and an additional reviewer (AJP) was consulted to reach an agreement in cases of no consensus. The following data were extracted: authors' names, year of publication, study setting, study design (e.g., trial, observational), sample size, characteristics of the patients (e.g., age, sex, body mass index), affected OA joint sites, treatment type, utility elicitation method, the health states considered, and the reported HSUVs (mean AE SD/SE, 95% confidence intervals [95% CIs], median, minimum/maximum, quartile) (see Supplementary Table 1 Meta-analyses. Based on data availability, the selection of studies for meta-analyses included studies related to knee, hip, and mixed OA (including a variety of OA patients without specifying their affected OA joint site), and studies of core intervention, medication, intraarticular injection, and primary surgery treatments. We followed OA management guidelines (13) to group the included interventions under 1 of these 4 categories of treatment. The core intervention category included exercise, weight management, and education/programs related to exercise and weight management. Medications included all drugs used to decrease pain and improve function in patients with OA. Intraarticular injections included corticosteroids, viscosupplements, and blood-derived products. Finally, primary surgery Observational studies that did not include delivery of an intervention were excluded from the metaanalysis. HSUVs were summarized by key OA affected joint sites (knee, hip, and mixed OA) for baseline (pretreatment) and at the most commonly available posttreatment time points (i.e., 3, 6, 12, and 24 months). When more than 1 HSUV study was based on the same data, the study with the highest number of participants was included in the meta-analyses. Subgroup metaanalyses by utility elicitation methods were also conducted, where possible. The meta-analyses were programmed in Stata software, version 15.1, using the "metan" command that required mean and SD/SE as meta-analytical inputs (14). Therefore, when the mean values and SD/SE were not reported, we used 95% CIs, median, minimum/maximum, first quartile, and third quartile values to estimate these parameters (15)(16)(17). HSUVs at baseline in observational studies and in both control and intervention groups of trials were pooled (termed pretreatment HSUVs). Posttreatment HSUVs were calculated by pooling HSUVs from longitudinal observational studies of interventions and intervention arms of trials (including active treatment groups but not control groups), for each time point. Heterogeneity among the pooled studies was assessed using the I 2 statistic (where an I 2 ≥ 50% indicated substantial heterogeneity) (17). To account for within-study and between-study heterogeneity, random-effects models were estimated.

RESULTS
Eligible studies. Initially 7,621 potential references were identified ( Figure 1). After we removed duplicates (n = 4,358), 3,263 were left for title and abstract screening. We excluded 2,593 during title and abstract screening, leaving 670 for the fulltext assessment. Of those, 522 were excluded due to not meeting the inclusion criteria ( Figure 1). Three additional studies  . Flow chart results of study search based on Preferred Reporting Items for Systematic Reviews and Meta-Analyses methodology. The exclusions by osteoarthritis (OA) joint sites and treatment type were because of the small numbers of studies in these joint sites and treatments, which meant that meta-analysis was not feasible. Eight exclusions by OA joint sites involved 2 shoulder and 6 hand OA-related studies; 41 exclusions by treatment type involved studies of massage, foot insoles, brace, mud therapy, balneotherapy, spa therapy, revision surgery, and observational studies that did not focus on any treatment. HTA = Health Assessment Technology database; HSUV = health state utility value.
identified through reference hand-searching were subsequently included, resulting in a final total of 151 (including 7 abstracts) being included in the systematic review. Eighty-eight of these studies were included in meta-analyses (including 4 conference abstracts).
Results of systematic review. The majority of included studies (n = 131, 87%) were published after 2010 (Figure 2A). More than half (n = 86, 57%) were conducted in Europe, followed by Asia (n = 20, 13%) and the Americas (n = 16, 11%). Four studies focused on Australians with OA, 1 study was conducted in multiple countries, and 24 studies did not report the study setting ( Figure 2B). Fifty-eight included studies (38%) were trials, 65 (43%) were observational studies of interventions, and 28 (19%) were observational studies that did not have an intervention component.
Twelve (17%) reporting the cross-sectional HSUVs of knee OA did not focus on any specific treatment (Table 1).
Of the 54 hip OA-related studies, the majority (n = 46, 85%) focused on surgical treatments. Two studies (4%) focused on core interventions, 1 investigated balneotherapy, and 5 (9%) reporting the cross-sectional HSUVs of hip OA did not focus on any specific treatment. There were no studies reporting the HSUVs related to hip OA medication and injection treatments (Table 1).
Two shoulder OA-related studies focused on surgical treatments. Among 6 hand OA-related studies, 2 reported the crosssectional HSUVs of hand OA populations, and 1 study each focused on spa, mud, a core intervention, and surgery treatment.
Of the 30 mixed OA-related studies, 14 (47%) focused on core interventions, and 12 (40%) reported the cross-sectional HSUVs of an OA population without specifying any treatment type. Two studies focused on surgical treatments, 1 focused on medication, and 1 focused on spa therapy (Table 1).

DISCUSSION
This is the first wide-ranging systematic review of OA-related HSUVs and meta-analyses on HSUVs for people with different OA-affected joint sites before and after various treatments. Our systematic review identified important areas where the current evidence is lacking, namely underrepresented geographical locations and ethnicities, affected OA joint sites, treatment options, and HSUVs based on more sensitive MAUIs. Our meta-analyses provide an HSUV database for alternative pre-and post-OA treatments that could offer a variety of HSUV inputs for future costutility models of OA-related conditions. HSUVs associated with 4 key treatment categories (core interventions, medication, injection, and surgery) often differed, as expected, pre-and posttreatment. Furthermore, we found significant inter-MAUI differences in the mean HSUVs, which is as expected from alternative descriptive systems and utility algorithms. Therefore this review provides important information that could be used by health economists and policy makers to determine the cost-effectiveness of various OA treatments and long-term disease outcomes using modeling techniques.
Our systematic review identified numerous gaps in the data on OA-related HSUVs, including geographical locations and ethnicities, affected OA joint sites, treatment options, and HSUVs based on more sensitive MAUIs. We found that more than half of included studies (57%) were conducted in Europe, and none in Africa. Because HSUVs should ideally be based on local population preferences, the generalizability of our results to underrepresented populations (e.g., African and Asian) may therefore be limited. Seventy-six percent of included studies focused on knee and hip OA, while other joint sites (e.g., shoulder and hand) attracted limited attention. While these results align well with the higher clinical impact, prevalence, and societal burden of knee and hip OA (18)(19)(20), the increasing prevalence and disease burden of hand and shoulder OA as a result of population aging (21,22) mandate further primary studies investigating the HSUVs of these joint sites.
The HSUVs that we have meta-analyzed differed as expected between alternative OA joint sites, treatments, HSUVs measures, and time points. We found a mean HSUV difference of +0.09 units in patients with knee OA using core interventions between baseline and 3-month postintervention, and this difference exceeds the minimal clinically important difference for all the MAUIs reported in previous studies (from +0.04 units [EQ-5D] to 0.08 units [AQoL-8D]) (23)(24)(25)(26)(27). Our findings are consistent with the randomized controlled trial (RCT) evidence showing the short-lived effects of knee OA core interventions (28,29). Other possible explanations include the limited number of core intervention studies with a follow-up a period of >3 months (and hence, wider 95% CIs for our 6-month and 1-year posttreatment HSUVs), and a likely reduction in the core intervention adherence in the long-term (30,31).
Most studies of knee OA medication treatments (83.3%) had relatively shorter follow-up periods (3 months), with only 1 study with a follow-up period of >3months. Consistent with RCT evidence of effectiveness of medication treatments (32,33), the pooled HSUV of studies with follow-up at 3 months postmedication treatment was significantly higher than the pooled HSUV of studies with baseline measures. As we did not have enough data on long-term HSUVs in patients using OA medications, we leave this question on the agenda for future research when long-term data become available. We found similar HSUVs at baseline and 1-year follow-up for knee OA injection treatments. However, these results should be carefully interpreted and used in economic modeling, as they are derived from only a limited number of studies (n = 5 at baseline and n = 1 at 1-year follow-up). HSUVs of knee OA patients recorded the largest difference (+0.25 units) between baseline and 1-year postprimary surgery, and it remained relatively stable to 2-years postprimary surgery. These findings are once again consistent with the previous evidence of the effectiveness of knee surgery, suggesting that HSUVs record a significant improvement within 1 year of knee surgery, and this change in HSUVs is sustained for years (34).
Surgery was the most common treatment in hip OA HSUV studies (85%). HSUVs in patients with primary hip OA surgery were significantly higher at 6 months postsurgery than at baseline and remained improved over the long term. The difference between pooled HSUV before and after surgery over 1 year was smaller in knee OA primary surgery (+0.25 units) than hip OA (+0.31 units). These findings align well with previous research (35) advocating a relatively higher efficacy of hip OA joint surgery. Only 2 studies (both based on the EQ-5D) investigated HSUVs in patients with hip OA core intervention, which aligned well with the previous findings of the dearth of studies measuring the HSUVs in patients using hip OA core interventions (36,37). No studies on hip OA medication and injection treatments were identified in our review as expected (38,39); thus, no meta-analysis for these treatments was possible. We recommend future studies to investigate HSUVs in patients using medications and injections, subject to the availability of better long-term observed data.
The HSUVs for mixed OA core interventions showed the same trend observed for knee OA, with a significant difference (+0.10 units) between baseline and 3-month postintervention HSUVs. This finding aligns with the existing findings of short-term benefits associated with OA core interventions (29). A small number of studies of medication treatment (n = 1) for mixed OA did not allow us to generate HSUV estimates of before and after medication treatments for use in health-economic modeling. Future primary HSUV studies in this area should therefore be imperative in bridging this evidence gap.
The EQ-5D was the most commonly used MAUI in the included studies (79%), with little to no representation from other more detailed MAUIs (e.g., AQoL-8D) that can more fully capture and assess the complex physical and psychosocial health aspects of OA patients (23,40). Our MAUI-specific subgroup analysis revealed significant differences between HSUVs based on alternative MAUIs (EQ-5D and SF-6D, for example), which is as expected from the MAUIs that are far from identical in terms of their descriptive systems and measurement scales (41). As the key objective of our review was not to explore the extent of agreement between alternative MAUIs, we leave the head-to-head comparison of HSUVs obtained through alternative MAUIs on the agenda for future research. Moreover, there is no consensus on the choice of MAUI to be used in measuring HSUVs of patients with OA (41,42). Future research should also endeavor to identify MAUIs that could be preferentially recommended for OA patients.
When the baseline HSUVs for various treatments were compared, the mean baseline HSUVs for patients with knee and hip OA using core interventions were significantly higher than those using surgery treatments, which is likely to be due to the specified selection criteria for RCTs. Due to the recommended stepwise approach for OA treatments (43), patients are more likely to receive core interventions at earlier stages of their OA (with better HRQoL) and surgery treatments at more severe stages of their OA (with relatively worse HRQoL), which can also explain this pattern. This result reinforces the need to use different HSUVs in modeling for treatments used at different disease stages.
The strength of this study is that this is the first comprehensive review and meta-analysis of all types of OA-related preference-based HSUVs by OA-affected joint sites, OA treatments, and utility elicitation methods. The study provided an HSUV database for alternative pre-and post-OA treatments that could offer a variety of HSUV inputs to future cost-utility models of OA-related conditions and identified important areas where evidence gaps exist in these estimates to inform future research needs. Our study has several limitations. It is important that the differences in HSUVs at different time points are not interpreted as true pre-/postchange or as direct evidence of intervention effectiveness, as the data do not examine differences in change in HSUVs between controls and intervention groups over time, and the data included in pooling at each time point come from different studies. Heterogeneity of the included studies due to the differences in terms of their study design, settings, and HSUV elicitation techniques can affect the interpretation of generated HSUVs. While we have conducted subgroup analyses where possible to highlight some possible sources of heterogeneity, we had limited capacity to explain and account for all sources of heterogeneity. The random-effects model in our meta-analyses aims to account for heterogeneity but may have consequences for the precision of model estimates (44). Therefore, in modeling, as well as the pooled mean sensitivity analyses, consideration of the potential imprecision of our estimates is important.
A further limitation is that due to the paucity of available studies, we could not conduct meta-analyses for all treatments of hip, knee, and other OA joint sites, nor could we group the treatment types in a more detailed way or perform meta-regression to account for >1 potential effect-modifying variable at a time. Quite a few potentially eligible CUA/CEA reports could not be included as they did not adequately report the required HSUVs (preand/or posttreatment) (45,46), despite clear reporting guidelines that recommend these be reported (47,48). We recommend that future CUA/CEA studies refer to these guidelines to help improve the availability of this important data. Also, the exploration of longterm HSUVs of patients using different OA treatments was mostly not possible. Finally, due to the paucity of data, we could not generate the estimates of HSUVs associated with alternative therapy adherence levels and medication adverse event types.
Our systematic review found that studies of OA-related HSUVs are of wide variety and differ from each other in terms of their setting, design, focused OA joint sites, utility measurement technique, generalizability, and other factors. The HSUVs that we have generated will be useful in conducting future health economic modeling for people experiencing various OA-related conditions. Our results should, however, be interpreted with caution as being derived from a relatively small number of heterogeneous studies. More research is needed to investigate changes in HSUVs of OA patients for longer follow-up periods.