Health state utility values associated with advanced gastric, oesophageal, or gastro-oesophageal junction adenocarcinoma: a systematic review

Abstract

Objectives:

To systematically identify utility values associated with advanced gastric cancer (GC), oesophageal cancer (OC), or gastro-oesophageal junction (GEJ) cancer. Utility values relating to health states are an essential component for cost-utility analysis (CUA).

Methods:

MEDLINE, Embase, Cochrane Library, and EconLit databases were reviewed for relevant studies using a pre-defined search strategy. Studies eligible for inclusion reported health state utility values (HSUVs) using direct (standard gamble [SG] and time-trade-off [TTO]) and indirect (such as EuroQol 5D [EQ-5D], short-form 6D [SF-6D], and the 15-dimensional instrument [15D]) methods for patients with advanced GC, OC, or GEJ cancer.

Results:

A total of 539 unique publications were identified, of which eight met the inclusion criteria (GC, n = 2; mixed population [gastrointestinal cancers], n = 4; OC, n = 2). The most commonly used instrument to estimate HSUVs was the EQ-5D (n = 7). Utilities were also estimated using the SF-6D and the 15D in the same study (n = 1). Direct elicitation methods included the TTO (n = 2) and SG (n = 1). Across the eight identified publications, health states and study populations were heterogeneous and sample sizes were limited.

Limitations:

This review, as with all summaries of this nature, is only as robust as the data derived from the identified studies. The systematic review process does not resolve any design issues or biases associated with the original studies.

Conclusions:

Limited data estimate HSUVs in patients with advanced GC, OC, or GEJ cancer. Utilities for advanced GC alone and advanced OC alone were reported in only two publications for each cancer type. No publications considered advanced GEJ utilities alone, and four publications considered utilities for a mixed population of gastrointestinal cancer types. Comparisons are confounded by heterogeneity across the identified publications. Further research into HSUVs associated with advanced GC and OC is required to improve the evidence available for use in CUAs.

Keywords::

• Cancer
• Gastric
• Oesophageal
• Gastro-oesophageal junction
• Health state utility values

Introduction

Assessing the comparative value of interventions is an important element of the decision-making process for health technology assessment (HTA) bodies. One method that may be utilized in the comparative value assessment of an intervention is a cost-utility analysis (CUA). An important element of CUA is the health state utility value (HSUV), which is a measure used to define and value health states of interest for economic modeling. A preference for the way in which HSUVs are derived for CUAs has been expressed by some HTA bodies, such as the National Institute for Health and Care Excellence (NICE; England and Wales), the Scottish Medicine Consortium (SMC; Scotland), and the Pharmaceutical Benefits Advisory Committee (PBAC; Australia). In contrast, HTA bodies in countries such as South Korea do not specify a preference1^,2. Therefore, the evidence requirements of individual appraising bodies clearly vary2.

One method of communicating the value of an intervention to decision-makers is through CUAs, which frequently report cost per quality adjusted life year. The quality adjusted life year is a standard measure of both longevity and health-related quality-of-life (HRQoL) and is not disease-specific, facilitating comparisons across treatments and conditions3. To calculate quality adjusted life years for CUA, a system is required that evaluates HRQoL over time4. A means of valuing HRQoL is the HSUV, which is a single score that represents an individual’s preference for a specific health-state on a scale of 0 (death) to 1 (perfect health)5^,6. These HSUVs are then multiplied by the time spent in that state to estimate quality adjusted life years5^,7. Methods of deriving HSUVs can include direct elicitation techniques (e.g., standard gamble [SG] and time-trade-off [TTO]) or indirect valuation methods using generic health status instruments (e.g., the EuroQol 5 dimensions [EQ-5D], short-form 6D [SF-6D], and 15D)6. A variety of utility-based methods can be used when conducting a study. However, HSUVs are often not collected as primary data and, to perform CUAs, it may be necessary to map a disease-specific instrument to a generic single index measure3, although this practice remains controversial and has not been validated for many disease settings.

The collection of appropriate HSUVs is important in oncology, where the aims of therapy are often to manage symptoms, improve quality-of-life, and/or prolong overall survival and progression-free survival (PFS)5^,7. In the economic modeling of cancer interventions, health states are typically based on whether a patient experiences disease progression8. Therefore, trial outcomes, such as tumor response, PFS, and overall survival are important to capture so that these data can be used to define distinct health states through which a patient transitions over the time horizon of the economic model. It may also be necessary to model health states that reflect the toxic effects of treatment (intervention-specific adverse events such as nausea and fatigue) as well as outcomes (such as recurrence) in adjuvant setting5. Health states may differ in terms of treatment costs and HRQoL8 so it is necessary to select appropriate HSUVs that reflect preferences for cancer treatments to better inform decision-makers regarding approval and reimbursement of a particular intervention.

Diseases of the upper gastrointestinal (GI) tract have a significant burden on morbidity and mortality. Globally, gastric cancer (GC) is the fifth most common malignancy (with 952,000 new cases in 2012 and a 5-year prevalence of 1,538,000) and is the third leading cause of worldwide cancer mortality (723 000 deaths)9. It has been estimated that ∼50% of patients present with advanced incurable GC, although it should be considered that this value will vary depending on the literature source and geographical location (due to variations in cancer screening)10. Oesophageal cancer (OC) is the eighth most common malignancy worldwide (456 000 new cases in 2012) and the sixth most common cause of mortality from cancer (400 000 deaths)11. The epidemiology of gastro-oesophageal junction (GEJ) cancer is less clear as historically there was no standard classification and many cases of OC and GC were misclassified as GEJ cancer and vice versa12. Furthermore, according to the International Classification of Diseases, tenth revision (ICD-10) published by the World Health Organization (WHO), GEJ cancer is coded as a malignant neoplasm of the stomach13. In a recent analysis of data from the National Cancer Institute’s Surveillance Epidemiology, and End Results (SEER) Program, it was suggested that the incidence of GEJ cancer in the US had increased from 1.22 cases per 100,000 (from 1973–1978) to 1.94 cases per 100,000 (from 2003–2008)12. The global incidence of GC, OC, and GEJ cancer is higher in males than it is in females9^,12. However, the incidence of GC and OC varies dramatically depending on geographical location, with the majority of cases occurring in developing countries and a large proportion of the global total occurring in East Asia9^,11. Geographical patterns regarding age-standardized rates of incidence and mortality are similar between OC and GC.

In patients with early stages of GC, surgery and neoadjuvant and/or adjuvant chemotherapy is usually adopted as the treatment approach14. However, many patients relapse despite treatment, and a proportion of patients present with advanced GC (50%)10. Palliative chemotherapy incorporating fluoropyrimidine is typically the primary first-line treatment strategy in advanced GC15–17 and other GCs18 (including metastatic OC and GEJ cancer)16^,17^,19. In patients with progressive disease despite chemotherapy, treatment options are limited, and historically there has been much debate and lack of consensus surrounding the standard of care in the second-line setting due to a lack of clinical evidence20. There is now mounting evidence to support the use of second-line therapies in advanced GC20^,21.

Rather than restricting searches of HSUVs for one specific cancer type, the current systematic review considers HSUVs required for modeling advanced GC, OC, and GEJ cancer. The rationale for this approach was that the enrollment of a mixed population of patients with upper GI cancers (including advanced GC, OC, and GEJ cancer) has been permitted in several studies and, historically, misclassification of such cancer types has occurred12. Furthermore, it was considered that studies in a mixed population of patients with upper GI cancers may be appropriate to provide proxy HSUVs for advanced GC, OC, and GEJ cancer in the absence of specific HSUVs.

It is important to understand the impact of the disease on patients’ health utility as treatment strategies emerge to address the burden associated with advanced GC, OC, and GEJ cancer, because decision-makers have a need to assess the relative value of new therapies. However, to conduct CUAs, HSUVs are needed for each health state that is modeled. Therefore, the objective of the current systematic review was to identify HSUVs in the published literature relating to advanced GC, OC, or GEJ cancer.

Methods

The current systematic review was conducted in accordance with a pre-defined protocol and PRISMA guidelines22. Search strings were used to review MEDLINE, Embase, Cochrane Library, and EconLit electronic databases (conducted September 25, 2013; Appendices 1–4). No date or language restrictions were imposed on the searches. Hand searches were conducted of the grey literature (including meetings/congresses of interest from 2010 onwards, and bibliographies of included articles) and for HTA data from three bodies known to require utility data as part of CUA and are publically available in English: NICE, SMC, and PBAC (Appendix 5). These HTA bodies were considered in more detail because they have stricter criteria for CUAs than many other HTA agencies and their assessments are publically available in English. The Cost-Effectiveness Analysis (CEA) registry, EQ-5D website, and Research Papers in Economics (RePEc) were also searched using terms including ‘gastric’ and ‘oesophageal/esophageal’.

Key eligibility criteria (Appendix 6) were the following: studies including adult patients with advanced GC, GEJ, or OC; studies reporting HSUVs associated with these cancers (including randomized controlled trials and observational studies); and publications stating mean or median HSUVs and a standard method of utility assessment. Studies eligible for inclusion also included those reporting HSUVs using direct (SG and TTO) and indirect methods (such as EQ-5D, SF-36, SF-12, SF-6D, HUI2, HUI3, and 15D). Publications reporting on the mapping of condition-specific measures into preference-based instruments (such as the EQ-5D) were also included. There were no restrictions regarding intervention or comparator.

Citations yielded by the searches were imported into a database and initially assessed based on title and abstract. Full publications of potentially relevant citations were examined using the pre-defined eligibility criteria and verified by a second reviewer. Disputes were resolved via discussion with a third party until a consensus was reached. A pre-determined data extraction table, designed in Microsoft Excel (Microsoft Corporation, Redmond, WA), was used to capture the following information (if reported): population, study type, patient characteristics, HSUVs, utility scales, study interventions, and follow-up time. There are no standard methods of reporting the quality of utility studies; therefore, quality assessment was based on criteria from a recognized HTA body (NICE), which rated studies on factors including sample size, respondent selection and recruitment, survey response rates, and extent of follow-up and missing data23.

Details of the search strings, inclusion/exclusion criteria, and extracted studies and their results (including quality assessment) can be found in the appendix.

Results

Study selection and overview

A diagram illustrating the flow of studies through the eligibility review is provided in Figure 1. A total of 539 unique publications were identified, of which 18 were considered potentially relevant based on title and abstract. After review of the full publication, the final dataset included eight references that met the inclusion criteria6^,24–30. A list of the eight included and 10 excluded publications is provided in Appendices 7 and 8. Quality assessment of the eight included studies is provided in Appendix 9.

Figure 1. PRISMA flow diagram. HRQoL, health-related quality-of-life; QoL, quality-of-life.

Three HTA documents were also identified during the systematic review31–33 but were secondary sources for the utilities described; these data had already been captured from the primary sources identified in the current systematic review and, therefore, are not considered in this section.

Four of the eligible publications reported data from observational studies6^,28–30 and four from randomized controlled trials24–27. HSUVs for the following populations were reported in the identified publications: patients with locally advanced/metastatic GC24; non-metastatic GC6; inoperable OC28; local, regional, or metastatic OC30; advanced/metastatic or locally recurrent (with ≥1 measurable lymph node) GC or GEJ cancer25; locally advanced or metastatic GI cancer27^,29; and inoperable metastatic (and/or poor medical condition with progressive dysphagia) OC or GEJ cancer26. The utilities identified were associated with the following treatment status: post-chemotherapy in patients who were previously chemotherapy naïve (or population included few patients who were pre-treated [9%])24^,25; pre-treated with chemotherapy and/or surgery6^,28; mixed status (pre-treated with chemotherapy [and/or surgery]) or chemotherapy naïve)24^,30; best supportive care27; and unclear26^,29. Further details of the included publications are presented in Table 1.

Table 1. Health state utilities reported in relevant publications.

Study, year (country)	Cancer type	Population	Health state	Instrument	Mean [median] HSUV (SD) or [CILL, CIUL] of HSUV
Aultman and Urspruch24, 2010 (24 countries)	GC	Patients with HER2+ locally advanced/metastatic GC (n = 584)	PFS while on therapy	EQ-5D	0.73 [0.71, 0.75] • IHC 3+ or FISH positive population: 0.7282 • Sub-group of patients (IHC 2+/FISH positive or IHC 3+): 0.7292
Kontodimopoulos et al.6, 2009 (Greece)	GC	Patients had previously attended 2–4 chemotherapy sessions (≥20 days previously), and had undergone surgery (n = 48)	Baseline: pre-treated patients attending hospital for chemotherapy (i.e., considered as currently receiving chemotherapy)	EQ-5D SF-6D 15D	0.550 (0.307) 0.606 (0.094) 0.685 (0.166)
McNamee et al.28, 2004 (UK)	OC	Recruited from a database of patients who had received curative treatment for OC at a specialist UK center (n = 56)	Health state 1^a Health state 2^a	TTO SG TTO SG	0.66 [0.50, 0.81] 0.78 [0.66, 0.89] 0.45 [0.31, 0.60] 0.49 [0.35, 0.63]
			Health state 3^a	TTO	0.35 [0.21, 0.50]
				SG	0.27 [0.15, 0.40]
			Health state 4^a	TTO	0.25 [0.13, 0.38]
				SG	0.20 [0.08, 0.31]
			Health state 5^a	TTO	0.08 [0.00, 0.17]
				SG	0.08 [0.00, 0.17]
			Treatment 1^a	TTO	0.64 [0.50, 0.78]
				SG	0.60 [0.47, 0.74]
			Treatment 2^a	TTO	0.54 [0.40, 0.69]
				SG	0.61 [0.48, 0.74]
			Treatment 3^a	TTO	0.62 [0.50, 0.76]
				SG	0.59 [0.45, 0.74]
Wildi et al.30, 2004 (US)	OC	• Patients with OC (n = 50) • Pre-treated (n = 13): palliative chemotherapy (n = 6), non-curative surgery (n = 1), or both (n = 6) • Treated after interview (n = 37) • EUS and CT staged: localized (n = 11), regional (n = 24), or metastatic (n = 12) cancer; non-detectable tumor after palliative care (n = 3) • In situ (n = 3), localized (nn = 11), regional (n = 24), metastatic (n = 12) • Dysphagia score: • 0 (n = 16), 1 (n = 18), 2 (n = 5), 3 (n = 8), 4 (n = 3)	Patient’s own health status (in situ to metastatic) Societal perspective (rated by the patient for a theoretical person with same stage and prognosis): (localized to metastatic) Patient’s own health status (in situ to metastatic) Patient’s own health status: dysphagia scores (0–4)	TTO EQ-5D TTO	• In situ: 0.99 (0) • Localized: 0.80 (0.30) • Regional: 0.54 (0.39) • Metastatic: 0.52 (0.31) • Localized: 0.77 (0.24) • Regional: 0.46 (0.35) • Metastatic: 0.15 (0.21) • In situ: 0.93 (0.12) • Localized: 0.60 (0.29) • Regional: 0.71 (0.21) • Metastatic: 0.69 (0.35) • Dysphagia 0: 0.86 (0.23) • Dysphagia 1: 0.62 (0.36) • Dysphagia 2: 0.29 (0.25) • Dysphagia 3: 0.51 (0.37) • Dysphagia 4: 0.25 (0.17)
				EQ-5D	• Dysphagia 0: 0.82 (0.17) • Dysphagia 1: 0.67 (0.30) • Dysphagia 2: 0.80 (0.19) • Dysphagia 3: 0.60 (0.22) • Dysphagia 4: 0.39 (0.27)
Curran et al.25, 2009 (NR)	MP: GC, GEJ	Patients had histologically confirmed metastatic adenocarcinoma of the stomach or esophagogastric junction, with measurable or evaluable metastatic disease, or locally recurrent disease (n = 333)	Post-baseline IF^b Post-baseline CF^c	EQ-5D	0.76 (0.23) 0.66 (0.27)
McMillan et al.27, 1999 (UK)	MP: advanced colorectal; OC, GC, pancreatic, cholangiocarcinoma	Patients with histologically proven, locally advanced or metastatic gastrointestinal cancer, with more than 5% weight loss who were receiving supportive care only, and had a life expectancy of at least 2 months and no moderate/severe dysphagia (n = 73)	Baseline utility: weight-losing gastrointestinal cancer patients	EQ-5D	• Megestrol acetate/placebo group: [0.630; range = −0.095–1.000] • Megestrol acetate/ibuprofen group: [0.689; range = −0.261–1.000]
O’Gorman et al.29, 1998 (UK)	MP: colorectal, GC, OC, pancreatic	Outpatients, advanced or metastatic gastrointestinal disease (n = 119)	Weight-stable patients	EQ-5D	[0.85]
			Weight-losing patients		[0.52]
Homs et al.26, 2004 (The Netherlands)	MP: Advanced OC or esophagogastric junction	Patients with inoperable cancer of the esophagus or esophagogastric junction due to metastatic disease and/or a poor medical condition with progressive dysphagia (n = 209)	Baseline utility for patients with inoperable cancer of the esophagus or esophagogastric junction due to metastatic disease and/or a poor medical condition with progressive dysphagia, before randomization	EQ-5D EQ-5D: Post-intervention: total group: changes in EQ-5D during follow-up (per 0.5 year follow-up)	63 (Change: −21) [−27 to −16]

^aPatient interview data and clinician experience were used to create five health states, describing five grades of inoperable esophageal cancer from mild (grade 1) to severe (grade 5).

^bIrinotecan 80 mg/m² over a 30-min i.v. infusion, followed by FA 500 mg/m² as a 2-h i.v. infusion, immediately followed by 5-FU 2000 mg/m² over a 22-h i.v. infusion, day 1 every week for 6 weeks followed by a 1-week rest.

^cCisplatin 100 mg/m² as a 1–3-h i.v. infusion, day 1, followed by 5-FU 1000 mg/m²/day over a 24-h i.v. infusion, days 1–5, and every 4 weeks.

15D, 15-dimensional; CF, cisplatin with 5-FU; CILL, confidence interval, lower limit; CIUL, confidence interval, upper limit; CT, computerized tomography; EQ-5D, EuroQoL 5 dimensions; EUS, endoscopic ultrasound; FISH, fluorescence in situ hybridization; GEJ, gastro-esophageal junction; GC, gastric cancer; HER2+, human epidermal growth factor receptor 2 positive; HSUV, health-state utility value; IF, irinotecan, folinic acid and 5-fluorouracil (5-FU); IHC, immunohistochemisty; MP, mixed population; NR, not reported; OC, esophageal cancer; PFS, progression-free survival, SD, standard deviation; SF-6D, short-form 6 dimensions; SG, standard gamble; TTO, time trade-off.

The EQ-5D was the most common instrument used to estimate HSUVs in the eligible studies (seven out of eight studies)6^,24–27^,29^,30. Other instruments reported in the identified publications included the SF-6D and 15D6. Direct elicitation techniques included the TTO28^,30 and SG28. Utilities for the following health states were derived using the EQ-5D: PFS post-chemotherapy in patients with advanced GC (0.73)24; post-chemotherapy in patients with GC (0.550)6; post-chemotherapy in patients with advanced GC or GEJ cancer (0.66–0.76)25; weight loss in patients with advanced GI cancer (0.630–0.689 dependent on treatment)27; weight stable (0.85) and weight losing (0.52) patients with advanced GI cancer29; patients with in situ, local, regional, or metastatic OC (0.60–0.93, depending on stage)30; dysphagia (score, 0–4) in patients with OC (0.39–0.82, depending on score)30; and inoperable advanced OC or GEJ (0.63)26. The TTO method was used to elicit the following utilities: health states in patients who had received curative treatment for OC (0.08–0.66, depending on health state)28; patients with local, regional, or metastatic OC (0.52–0.80, depending on stage)30; dysphagia (score 0–4) in patients with OC (0.25–0.86, depending on score)30; and societal valuation of health states in local, regional, or metastatic OC (0.15–0.77)30. The following HSUV was derived using the SG method: inoperable advanced OC (0.08–0.78, depending on health state)28.

Health state utility values associated with gastric cancer

There were two publications that reported HSUVs exclusively for GC (Table 1)6^,24. Aultman and Urspruch24 described the HSUVs for PFS derived from the open-label, international, phase III, randomized controlled Trastuzumab for GAstric cancer (ToGA) trial34. ToGA was undertaken across 24 countries in Asia, Central and South America, and Europe34. Adult patients with human epidermal growth factor receptor 2 oncogene (HER2+) locally advanced/metastatic GC were randomized to receive trastuzumab + chemotherapy (mean age = 59.4 years [SD = 10.8 years]; males: 77%) or chemotherapy alone (mean age = 58.5 years [SD = 11.2 years]; males: 75%)34. Only 9% of patients had previously received chemotherapy34. The EQ-5D questionnaire was completed by patients at baseline and every 3 weeks (i.e., at the end of a treatment cycle) until disease progression, death, or loss to follow-up and was used to derive a HSUV for PFS (excluding the time point at which disease progression occurred)24^,34. An HSUV of 0.73 (95% confidence interval = 0.71–0.75) was generated for PFS in advanced HER-2+ GC and was considered to be a robust estimate due to the large number of patients included in ToGA (n = 584) (Table 1)24.

In the second publication, Kontodimopoulos et al.6 used ordinary least squares regression to predict EQ-5D (preference elicited using the TTO method), SF-6D, and 15D utilities from disease-specific European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ-C30) scale scores6. A common data-set was used to estimate utilities and included patients with non-metastatic GC who were attending a Greek hospital for further chemotherapy. All patients had previously attended two-to-four chemotherapy sessions and had undergone prior surgery (n = 48). The instruments were administered before the current chemotherapy session. Four age groups (40–49, 50–59, 60–69, and ≥70 years) for each gender group were represented6. The mapping algorithms applied were used in the context of GC treated with chemotherapy6. The following mean utilities were estimated: EQ-5D, 0.550 (SD = 0.307); SF-6D, 0.606 (SD = 0.094); 15D, 0.685 (SD = 0.166) (Table 1). It is of note that no known mapping attempts using a condition-specific instrument had been published for the15D before the Kontodimopoulos et al.6 study. The study demonstrated that EORTC QLQ-C30 could be used to predict EQ-5D, SF-6D, and 15D utilities6.

Health state utility values associated with esophageal cancer

Two publications reported HSUVs exclusively for OC (Table 1)28^,30. The population of interest in the study by McNamee et al.28 was recruited from a UK database of adult patients (mean age = 68 years [SD = 7.6 years]; males: 70%) who had received potentially curative surgery for OC (n = 56). The authors compared health state and treatment utilities derived using the chained TTO and SG methods. Using a chained approach means that health states are compared with other less-than-perfect health states and with perfect/good health or death28. Patients were randomized to be interviewed using either the TTO or SG and were told that health and treatment states applied for 12 months, followed by death. Five disease-specific health states were developed (based on clinician and patient input) to describe grades of inoperable OC (from mild [grade 1] to severe [grade 5]). In addition, three treatment descriptions were created that took into account health and non-health related attributes (hospital trips, nights in hospital, and pain following treatment). Details of the health state and treatment descriptions were provided in Appendix 1 of the McNamee et al.28 article. The EQ-5D instrument was also administered and, although the study sample was described as having poorer health than the general population, single HSUVs were not reported. Mean utilities for health state 5 to health state 1 ranged from 0.08–0.66 for TTO and 0.08–0.78 for SG, demonstrating that HSUVs were higher for better (vs worse) health states28. No statistically significant differences were found between mean TTO and SG HSUVs, whereas statistically significant differences were observed between states within each measure, indicating that changes were detectable using these methods28. No significant differences between or within mean TTO and SG treatment values were reported (HSUV: TTO, ranged from 0.54–0.64; SG, ranged from 0.59–0.61)28 (Table 1). The study demonstrated that chained TTO and SG methods were suitable to derive HSUVs for inoperable OC, although it should be noted that the sample size was small28 and the health states rated were hypothetical conditions and not based on actual patient health or well-being.

The study by Wildi et al.30 was the first that employed a well-validated method to derive HSUV estimates for patients with malignant dysphagia (n = 50; mean age = 63.58 years; range = 46–83 years; males: 84%). The sample population included American patients with non-detectable tumor after palliative care (n = 3), and localized (n = 11), regional (n = 24), and metastatic (n = 12) disease30. The proportion of patients with baseline dysphagia scores ranging from 0–4 are provided in Table 1. A total of 13 patients had previously received treatment (palliative chemotherapy [n = 6], non-curative surgery [n = 1], or both [n = 6]). Patients were interviewed before therapy, and mean EQ-5D and TTO utilities were derived for a patient’s current health state. Utilities were also reported for hypothetical scenarios associated with staging (societal perspective; localized to metastatic disease) using the TTO method30. Mean TTO utilities ranged from 0.99–0.52 (non-detectable tumor to metastatic disease) when rated according to a patient’s own disease vs 0.77–0.15 (localized to metastatic disease) from a societal perspective (Table 1). Patients consistently rated their own TTO HSUVs higher than that of a theoretical matched patient, although this was only significant for metastatic disease30. Although TTO HSUVs appeared to decrease with progression and the EQ-5D HSUV for localized disease was numerically lower than that of more advanced disease (regional and metastatic) (Table 1), there was no significant difference across measures according to stage of disease30. In general, EQ-5D and TTO HSUVs decreased with increasing severity of dysphagia (Table 1). Significant differences were observed regarding the relationship between mean HSUVs and the degree of dysphagia (EQ-5D, p = 0.0384; TTO, p = 0.0025), indicating that both instruments were capable of detecting differences between states. The authors stated that dysphagia appeared to be a major determinant of HSUVs30, although more specifically it seems that severity of dysphagia is a driver of HSUVs (increasing severity was associated with decreasing HSUVs).

Health state utility values derived for mixed populations

Four publications described HSUVs in populations with a variety of GI cancer types including GC, OC, and/or GEJ cancer25–27^,29. Curran et al.25 reported chemotherapy-naïve patients with advanced GC or GEJ cancer (metastatic or locally recurrent disease). Demographic data were not reported, but further details of the phase II and III parts of the study have been reported elsewhere35^,36. Based on data from the phase II study, it is assumed that the population were derived from 13 European countries, Israel, Lebanon, Turkey, and South Africa36, although this was not stated in the identified publication25. Patients were randomized to receive irinotecan, folic acid, and 5-fluorouracil (5-FU) (IF group) (n = 170) or cisplatin with 5-FU (CF group) (n = 163)25. Median age was 58 years (range = 29–76 years) in the IF group and 59 years (range = 28–77 years) in the CF group, with males comprising 73.5% and 66.3% of the sample, respectively, as reported previously35. The aim of the study by Curran et al.25 was to compare HRQoL between the two treatment groups, among other assessments, the EQ-5D was administered every 8 weeks until disease progression and then every 3 months until death. Post-baseline mean EQ-5D HSUVs were reported for the IF and CF groups (0.76 [SD = 0.23] and 0.66 [SD = 0.27], respectively) (Table 1)25. Although the HSUVs reflect the state of chemotherapy-naïve patients after treatment, values are presented separately for each regimen and, therefore, are intervention specific25.

A prospective randomized study reported by McMillian et al.27 evaluated weight loss in patients with GI cancer. The study population included patients with locally advanced or metastatic GI cancer, >5% weight loss, a life expectancy of ≥2 months, and no moderate or severe dysphagia. Patients were randomized to receive megestrol acetate/ibuprofen (n = 35 [including GC, n = 5; OC, n = 2]; median age = 69 years; range = 52–88 years; males: 63%) or megestrol acetate/placebo (n = 38 [including GC, n = 6; OC, n = 2]; median age = 72 years; range = 50–90; males: 55%) for 12 weeks. The study was designed to compare the two treatment arms with respect to improved weight gain and HRQoL in patients losing weight. Median baseline EQ-5D HSUVs were reported per group (megestrol acetate/placebo group: 0.630; range = −0.095–1.000; megestrol acetate/ibuprofen group: 0.689; range = −0.261–1.000) (Table 1)27. The study provides EQ-5D HSUVs that may be appropriate for modeling weight loss in patients with GI cancer. However, the HSUVs were derived from a limited study population.

The impact of weight loss associated with GI cancer was also considered by O’Gorman et al.29. The study included outpatients (n = 119) with locally advanced and/or metastatic disease and no apparent issues with the intake or absorption of food. Patients in this study had colorectal, esophageal, gastric, or pancreatic cancer29. Patients were described as weight-stable (n = 22; median age = 70 years; range = 49–84 years; males: 45%) or weight-losing (n = 97; median age = 67 years; range = 44–84 years; males: 64%), and among other quality-of-life assessments, median EQ-5D HSUVs were reported. There was a significant difference between median EQ-5D HSUVs for patients who were weight-stable (0.85; range = 0.03–1.00 years) and weight losing (0.52; range = −0.26–1.00) (p < 0.001), indicating that weight loss was associated with lower HSUVs29. Although the authors state that the study represents only a “snapshot” view29, the reported HSUVs may be useful in the modeling of weight loss in patients with GI cancer.

Homs et al.26 assessed HRQoL after palliative treatment in patients from the Netherlands with inoperable OC or GEJ cancer due to metastatic disease and/or progressive dysphagia. Patients in the study were randomized to receive a metal stent (n = 108; mean age = 69 years [SD = 11 years]; males: 80%) or single-dose brachytherapy (n = 101; mean age = 69 years [SD = 13 years]; males: 75%)26. The EQ-5D was administered before treatment and at 1, 3, 6, 9, and 12 months after therapy. The mean EQ-5D HSUV at baseline was 0.63 (reported as 63, where 100 was described as “best”) and decreased by 21 (−21; 95% CI = −27, −16) per 0.5 years of follow-up (Table 1)26. The authors described the use of the EQ-5D as “advisable” should cost-effectiveness analysis be required26.

Discussion

A systematic literature review was conducted to assess available HSUVs associated with advanced GC, OC, and GEJ cancer. Publications reporting on a variety of advanced GI carcinomas, including the cancer of interest, were also included. Despite the broad search criteria, there was a paucity of HSUVs related to advanced GC, OC, or GEJ cancer, and only eight publications were considered eligible for inclusion6^,24–30. HSUVs for advanced GC alone and OC alone were reported in only two publications for each cancer type. No publications considered GEJ cancer utilities alone, and four publications considered utilities for a mixed population of GI cancers without reporting HSUVs separately by tumor type. The generalizability of OC or other GI cancer HSUVs to an advanced GC population is unknown, limiting the ability to apply data from one GI tumor type to another.

A key limitation of the majority of studies (six of eight) was the small sample size (<150 patients in a treatment arm) used to estimate HSUVs6^,26–30. Only two publications included >300 patients24^,25. Studies with small sample numbers (in relation to the numbers required for the chosen study design) may be open to uncertainty and consequently may not be robust sources of HSUVs.

The identified studies were heterogeneous regarding the populations, health states, cancer stage, and treatments considered, limiting the ability to make cross-study comparisons. The age, gender, and treatment of participants varied, and HSUVs (medians and means) ranged from 0.08–0.85, depending on the population and health state considered. Several studies were conducted in only one country6^,26–30. The suitability of utilities for CUA depends on similarities between the source population and that to be modeled and the body to which the CUA is to be submitted.

Even within the same tumor types, several health states and assessment tools were found in the identified publications. Some considered HSUVs associated with PFS24, but excluded patients pre-treated for metastatic disease; the number of participants who had received previous chemotherapy was low (9%)34. The applicability of these data to pre-treated patients is unknown. In other cases, study participants were assessed post-chemotherapy6^,25, but were either chemotherapy-naïve25 or pre-treated6 and HSUVs were reported at post-treatment25 or baseline6. Furthermore, HSUVs were also reported separately for each treatment regimen in one publication and, as a result, represented intervention-specific health states25. The specificity of these data to unique settings and treatments makes the generalizability of these data highly uncertain. HSUVs for specific health states associated with weight loss27^,29 and dysphagia30 were also identified in the literature search, but it is unlikely that these would be suitable for use for advanced GI cancer CUAs unless the modeling of these particular conditions was required. Due to the mixed populations used in most of the identified studies, it would be difficult to assume the same data would remain consistent for any one disease site alone. The stage of cancer also varied across publications; for example, some studies included localized disease only6 or metastatic26 disease only, whereas, in others, patients with metastatic and localized lesions were combined24^,27^,29^,30.

When conducting CUAs for a HTA, it is necessary to fulfil the criteria of the assessing body. For example, NICE has a preference for health status valued using the EQ-5D1^,3, and the SF-6D is considered suitable for use in sensitivity analyses3. However, the SMC believe that the use of the EQ-5D to the exclusion of any other valid generic utility measures would be inappropriate4. The EQ-5D and SF-6D (in addition to the Health Utilities Index) are considered suitable utility indexes by PBAC, whereas the use of the 15D requires particular justification37. In seven of the eight identified publications, the EQ-5D was used to estimate HSUVs6^,24–27^,29^,30 and other reported instruments were the SF-6D and 15D6. Based on NICE, SMC, and PBAC criteria, these publications use appropriate instruments for deriving utilities (either for use in a base case or sensitivity analyses). The SMC also accepts direct measurement of utilities for appropriate disease/condition health states using TTO or SG4. Two publications used TTO28^,30 and SG28, although the utilities derived were based on health states specific to OC (such as dysphagia), which would not be the most appropriate for modeling GC. Regardless of the instrument used, justification to the HTA body should be provided. The use of non-standard and unjustified approaches to quality adjusted life year measurement would be criticized, and potentially not accepted, by NICE, the SMC, or PBAC.

The EQ-5D is a preference-based HSUV instrument that includes five dimensions (mobility, self-care, usual activities, pain/discomfort, and anxiety/depression), with three severity levels for each dimension (no problems, some or moderate problems, extreme problems). It can be used across disease settings but was not designed specifically for cancer and is, therefore, considered an indirect valuation method for the derivation of HSUVs in cancer. In some cases, the use of a disease-specific measure (such as the EORTC QLQ-C30) may be more appropriate to capture the impact of cancer3 because the dimensions of the EQ-5D may not capture relevant health effects. The EORTC QLQ-C30 instrument includes five functional scales (physical, role, emotional, cognitive, and social), a global health item, three symptom scales (fatigue, nausea/vomiting, and pain), and single symptom items (dyspnoea, appetite loss, constipation, diarrhea, and financial difficulties). It is clear that the EORTC QLQ-C30 more comprehensively captures items of relevance to the quality-of-life of a patient with cancer when compared with the EQ-5D. However, cancer-specific quality-of-life scales cannot be used in CUAs submitted to a HTA body without conversion to a preference-based HSUV, such as the EQ-5D, via the use of statistical mapping functions3. This mapping function has not yet been fully validated in the setting of specific GI cancers. One of the identified publications used mapping algorithms to derive EQ–5D utilities from the EORTC QLQ-C306. The mapping algorithms were used in the context of GC treated with chemotherapy, but the authors stated that they could be applied to patients with other cancers and similar EORTC QLQ-C30 scores6. Although mapping may be necessary in some circumstances, well-designed explicit methods are required for HTA bodies4 and uncertainty is increased. The direct use of the EQ-5D remains the preferred choice for HTA bodies such as NICE3.

The relevance of HSUVs to the CUA and to the HTA body to which the economic model will be submitted is equally as important as the quality of HSUVs used in the model23. Based on NICE, SMC, and PBAC criteria, only two publications identified in this literature review were considered appropriate to inform HSUVs for HTA24^,25. In the case of Aultman and Urspruch24, the population was representative of advanced GC, HSUVs were derived from a large sample, and values had been used in previous HTA submissions31^,33. However, details on the age and gender of study participants were not supplied but were instead cited from Bang et al.34. Furthermore, patients were not pre-treated and the majority of the ToGA population were Asian24^,34, limiting the generalizability of these data to non-first-line treatment settings in a European population. Sensitivity analyses would be required to demonstrate the robustness of HSUV estimates. For example, patients with either advanced GC or GEJ cancer were included in the study by Curran et al.25; however, the HSUVs were intervention-specific and were evaluated after the completion of first-line treatment and may not be applicable to other settings. Therefore, it is suggested that Curran et al.25 should be used as a sensitivity analysis in CUAs that do not use the same setting or intervention.

The current systematic review follows the recommendations of Papaioannou et al.1 for identifying suitable HSUVs. For example, it was recommended that HSUV reviews should be examined to determine which health states are required for modeling and what type of data reimbursement agencies prefer1. The scope of the current systematic review was broad and HSUVs derived by various instruments were sought. The eligibility criteria were not limited to a specific utility-based instrument. Furthermore, HSUVs were sought from observational studies and randomized controlled studies, as recommended by Papaioannou et al.1.

Furthermore, the sources of HSUVs identified in the current systematic review have also been used in previous GC-related submissions to HTA bodies (NICE and PBAC)31–33. These documents (NICE submissions/Evidence Review Group [ERG] reports and an Australian public summary document) were identified as secondary sources of utilities during the review31–33. The ERG report for trastuzumab31 states that utility values for PFS (0.7292) were estimated from the EQ-5D collected in the ToGA trial34, this is the same HSUV reported by Aultman and Urspruch24, which was identified as a source of HSUVs in the current systematic review. For progressive disease, a HSUV of 0.577 was taken from a previous NICE evaluation of a treatment for GI stromal tumors (as EQ-5D post-progression was not collected in ToGA), which supports the findings of the systematic review regarding the lack of specific HSUV data for progression in advanced GC, OC, or GEJ cancer. The ERG also identified Curran et al.25 as a more recent chemotherapy study in GC, but stated that it was not identified by the manufacturer of trastuzumab and was not considered in sensitivity analyses31. It was determined by the ERG32 that the ToGA estimate was similar to that of Curran et al.25. In a second ERG report32, considering capecitabine in advanced GC, the ToGA trial was again used to determine PFS utilities (0.73). Finally, in an Australian public summary document33 considering trastuzumab in advanced adenocarcinoma of the stomach or GEJ (termed GC), HSUVs were taken from the ToGA trial34 and Curran et al.25.

Economic models typically investigate a treatment pathway until a specific point (such as progression or death) and, therefore, require utility values for a series of health states1. Few studies that reported HSUVs specifically in advanced GC, OC, or GEJ cancer were identified in the current literature review; it may be necessary to use proxy data for any of these diseases to model certain health states at various stages of the treatment pathway. For example, HSUVs for progressive disease were not identified; therefore, proxy data from other diseases would need to be used.

A limitation of the current systematic review, as with all summaries of this type, is that the data presented are only as robust as that of the primary studies. For example, problems in the design of the identified studies and biases from these sources are not resolved through the systematic review process. Therefore, an element of caution has to be exercised when considering the findings. Despite the limitations, the current systematic review presents the best known available evidence at this time and indicates areas for additional research to facilitate the modeling of advanced GC, OC, or GEJ cancer. For example, although HSUVs were identified specific to PFS, the context in which they would be implemented in an economic model must be considered. When HSUVs for survival data are not available, data from the progression-free interval may be a reasonable proxy. However, the relationship between PFS and overall survival has to be quantified for the disease under consideration and, ideally, sensitivity analyses should be used to explore any uncertainty8. Larger studies are required in GC, OC, and GEJ cancer because a limitation of the studies identified in this search was their small sample size, adding to uncertainty surrounding the HSUVs. In addition, not all studies were designed to assess HSUVs; rather, these data were reported incidentally with little discussion of how they were derived. There is also a need for HSUVs for progressive disease and for specific chemotherapy exposures (e.g., pre-treated or never exposed). Although additional studies have been published since the timeframe of the current systematic review38, there is still a paucity of information. Further research regarding utilities’ HSUVs associated with different stages of GC is required in order to allow modeling without proxy data.

Conclusion

Advanced GC, OC, and GEJ cancer remain an area of high unmet need; however, demonstrating the cost-effectiveness of treatments for this disease is challenging due to a paucity of appropriate HSUVs. Quality-of-life in many clinical trials is measured using instruments that are not suitable for deriving HSUVs for economic modeling. The current systematic review identified limited data for HSUVs that specifically relate to advanced GC and may be suitable for economic modeling. Furthermore, limited data were available specifically for advanced OC or GEJ cancer, which may have been used to provide proxy HSUVs for advanced GC, although the generalizability of OC or other GI cancer HSUVs to an advanced GC population is unknown. It is important to assess HSUVs derived from the literature according to the quality and relevance of the data to the disease modeled and to the body appraising the CUA. However, heterogeneity across patient and study characteristics in the eligible publications confounded the identification of appropriate HSUVs. The current systematic review provides sources of HSUV data that may be used as a proxy in economic models in the absence of more robust and specific data for advanced GC, OC, and GEJ cancer. The review also indicates areas where further research is required, such as the need for large studies looking at HSUVs associated with different stages of cancer (including progressive disease) and specific chemotherapy exposures. Furthermore, the relationship between overall survival and PFS in advanced upper GI cancers requires quantification to determine whether PFS is a suitable proxy for survival in an economic model. Such research will improve the evidence base available for CUAs and facilitate the evaluation and reimbursement of treatments, particularly for patients with progression despite chemotherapy.

Transparency

Declaration of funding

Funding for this research was provided by Eli Lilly and Company, Indianapolis.

Declaration of financial/other relationships

GCC, LMH, K-LT, UK, NR, DN, and AML are employees of Eli Lilly and Company. DTK and SAM were paid consultants to Eli Lilly and Company in conjunction with the preparation of this manuscript.

Supplemental data

Supplemental data is available for this article.

Acknowledgments

We thank Rachael Baker-Searle for medical writing support.