A systematic review of the methodological quality of economic evaluations in genetic screening and testing for monogenic disorders

Abstract

Purpose:

Understanding the value of genetic screening and testing for monogenic disorders requires high-quality, methodologically robust economic evaluations. This systematic review sought to assess the methodological quality among such studies and examined opportunities for improvement.

Methods:

We searched PubMed, Cochrane, Embase, and Web of Science for economic evaluations of genetic screening/testing (2013–2019). Methodological rigor and adherence to best practices were systematically assessed using the British Medical Journal checklist.

Results:

Across the 47 identified studies, there were substantial variations in modeling approaches, reporting detail, and sophistication. Models ranged from simple decision trees to individual-level microsimulations that compared between 2 and >20 alternative interventions. Many studies failed to report sufficient detail to enable replication or did not justify modeling assumptions, especially for costing methods and utility values. Meta-analyses, systematic reviews, or calibration were rarely used to derive parameter estimates. Nearly all studies conducted some sensitivity analysis, and more sophisticated studies implemented probabilistic sensitivity/uncertainty analysis, threshold analysis, and value of information analysis.

Conclusion:

We describe a heterogeneous body of work and present recommendations and exemplar studies across the methodological domains of (1) perspective, scope, and parameter selection; (2) use of uncertainty/sensitivity analyses; and (3) reporting transparency for improvement in the economic evaluation of genetic screening/testing.

Introduction

Genetic screening and testing for monogenic diseases can be used to establish a definitive molecular diagnosis in symptomatic patients, identify increased risk of disease in presymptomatic individuals, provide information about prognosis or management of rare disorders, identify other at-risk family members, and guide reproductive planning. If used appropriately, such analysis has the potential to reduce morbidity and mortality through disease prevention or early intervention, targeted treatment, and avoidance of inappropriate or ineffective treatment. However, genetic analysis and indicated downstream care for people who test positive can be costly for both the health care system and the patient. Despite being rare (the most common affecting less than 1% of the population), molecular diagnosis of monogenic conditions can be highly useful from a clinical perspective. Currently, diagnostic genetic testing is recommended only to those meeting specific clinical criteria or after other clinical tests have failed to give a definite diagnosis. It may be cost-effective to identify and care for patients with monogenic conditions before symptoms manifest, especially for conditions with effective interventions that could improve clinical outcomes. Researchers are assessing the value of screening for clinically useful monogenic conditions within a broader population. Economic evaluations—including cost-consequence analyses, cost-benefit analyses, cost-utility analyses (CUAs), and cost-effectiveness analyses (CEA)¹—are critical for assessing the potential value of genetic screening/testing for specific applications. Over the last 2 decades, the number of such evaluations has increased rapidly.^2,3 Yet, the speed with which economic evaluations have been produced may be outpacing the field’s ability to disseminate and widely adopt best practices as well as identify gaps where best practices have not been adopted.

High-quality methodological approaches to economic evaluations are essential for the appropriate interpretation and implementation of study findings. Despite the recent publication of several methodological recommendations for CEAs in genetic medicine, study quality across disease areas has not been systematically reviewed.^4–6 Importantly, there are methodological challenges unique to economic evaluations of clinical genetic screening and testing for monogenic disorders that deserve specific attention.⁷ Compared with other medical interventions that have more routinely been subjects of economic evaluations (eg, pharmacoeconomics), the methodological tendencies of economic evaluators of genetic screening and testing programs may still be in formation.

This qualitative systematic review characterizes the methodological quality of recent economic evaluations of genetic screening and testing for monogenic disorders, spanning from birth to diagnosis. Throughout this review, we use the term genetic testing when referring to a clinical diagnostic setting in which a patient is at increased risk for a genetic disorder because of their personal and/or family history; we use genetic screening when the individual being screened is not known to be symptomatic of, nor at substantially increased risk for, such a condition. We emphasized this distinction given that both the differing resources are demanded of and the health outcomes that may be experienced through either strategy and the field’s interest in evaluating screening programs. See Appendix 1 for more detail. The goal of this review is to improve the methodological quality of future economic evaluations to guide implementation of such studies. Where consistent methodological limitations were identified, we have provided recommendations and exemplar models.

Materials and Methods

Search strategy

This systematic review identified economic evaluations of genetic screening and testing for monogenic disorders, focusing on assays that seek to establish (or refine) a genetic risk or diagnosis. Included studies incorporated costs and health outcomes downstream from genetic testing and diagnosis. The review was registered with the International Prospective Register of Systematic Reviews on July 2, 2019 (record number CRD42019141086). Studies that did not include complete economic evaluations (“the comparative analysis of alternative courses of action in terms of both their costs and consequences”¹) or considered no health outcomes beyond diagnostic yield were excluded.^8,9 Studies of common variants and polygenic risk scores for complex diseases were excluded as were studies of somatic variant or gene expression in tumors.^10,11 Pharmacogenetic screening, which is defined as testing for genetic variants primarily related to adverse reactions to drugs or drug metabolism, was excluded.¹² Genetic testing/screening specifically related to reproductive planning (preconception or prenatal) was excluded.¹³ Systematic reviews and commentaries were also excluded.¹⁴ Additional search strategy details are included in Appendix 2.

Code development

Qualitative codes reflecting methodological features of evaluations were developed using a top-down and inductive approach. Initial codes were adopted from the 1996 checklist developed by Drummond and Jefferson for the British Medical Journal (BMJ) (hereafter BMJ checklist),¹⁵ along with features highlighted in similar systematic reviews.^16–19 Although more recent checklists have been developed as guides for authors,²⁰ the BMJ checklist was chosen given its emphasis on quality assessment by reviewers, its use in recent reviews of genetic evaluations,^19,21 and its wide-spread use among similar systematic reviews published from 2010 to 2018.²² A full list of the codes and summary statement templates used can be found in Appendix Table 3.

BMJ checklist and qualitative assessment beyond the BMJ checklist

We used the 35 BMJ checklist items (hereafter items) to assess included studies. Items were classified as not met (0), partially met (1), fully met (2), or not relevant (N/R). If relevant information was not contained in the primary publication or supplemental materials, but an appropriate citation was listed, we classified that item as not available (N/A). A detailed rubric was developed for each checklist item (Appendix 4 and Appendix Table 4). Average quality scores were calculated for each question by summing the 0s, 1s, and 2s each article received across all studies and then dividing that sum by the number of items for which 0s, 1s, and 2s were possible (excluding N/R and N/A).

Additional items were created to track, in more detail, important article features that the BMJ checklist did not directly address but have been recommended in other authoritative guidelines (Appendix Table 5).^3,7,22–29 These features did not contribute to average checklist scores. During analysis, we grouped these additional items, along with select BMJ checklist items that we wished to highlight in more detail, into 3 distinct methodological constructs: perspective, scope, and parameter selection; the use of sensitivity/uncertainty analyses; and reporting transparency.

Review process

Article coding and assessment began with a primary coder who applied qualitative codes and assessed items. Next, a secondary coder received the already coded articles from primary coders and cross-examined articles to ensure codes were appropriately applied. Secondary coders independently assessed all 35 BMJ items and were blind to the assessment given by primary coders. Conflicts were discussed and resolved between 2 reviewers (K.J., I.G., K.H., K.H.L.).

Results

Study characteristics

Of the 5727 records identified through database searches, 47 studies met the inclusion criteria (Figure 1). Table 1 shows the main features of the 47 articles included in this review along with each article’s average quality assessment. Three genetic conditions constituted nearly half of all studies: Lynch syndrome (n = 10), familial hypercholesterolemia (n = 7), and hereditary breast and ovarian cancer (n = 14). A smaller set of studies considered maturity-onset diabetes of the young (n = 3) thrombophilia (n = 2), or multiple conditions (n = 2), and undiagnosed pediatric disorders (n = 4). The setting of most studies was in the United States (n = 11), the United Kingdom (n = 9), and Australia (n = 9), with smaller numbers also conducted in Germany (n = 4), both the United States and the United Kingdom (n = 3), the Netherlands (n = 3), and elsewhere (Spain: n = 2; Poland: n = 1; Norway: n = 1; Malaysia: n = 1; Italy: n = 2; Taiwan, n = 1; Singapore: n = 1).

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram of search strategy, including inclusion and exclusion criteria and the number of articles maintained or removed at every step. w/o, without.

Table 1.

Summary characteristics of included studies

Study	Syndrome/Genetic Condition of Interest	Country	Population	Intervention of Interest	Comparison	Health Outcomes Considered
Catchpool et al41	Cardiomyopathy	Australia	Unaffected 18-year-old first-degree relatives of patients with DCM	Testing for monogenic disease variants	Clinical surveillance alone	Clinically unaffected, preclinical/ (MDCM), DCM, and death
Ademi et al40	FH	Australia	Relatives of patients with FH	Genetic testing combined with LDL-C testing	No screening of relatives	CVD
Chen and Hay45	FH	US	People with family history or indications of FH using statins	Genetic screening and lipidbased screening with statin adherence	Lipid-based screening alone	CVD event/stroke, which served as summary category for MI, stroke, and angina. Three health states were considered: Pre-CVD, CVD event/stroke, and death
Crosland et al30	FH	UK	Potential FH cases identified in primary care databases and their relatives	Testing using an FH genetic panel	No case identification and no cascade testing	Stable angina, unstable angina, MI, TIA, stroke, heart failure, peripheral artery disease, cardiovascular mortality, and noncardiac mortality
Kerr et al62	FH	UK	Adult relatives of those with monogenic FH	Testing for variants in LDLR, APOB, or PCSK9	No cascade testing	Stable angina, unstable angina, MI, TIA, stroke, CHD death, non-CHD death, poststable angina, unstable angina, post-unstable angina, and poststroke.
Lázaro et al44	FH	Spain	High cholesterol in children and adults identified in primary care	Testing for FH pathogenic variants, followed by cascade screening	No genetic testing	Coronary event, modeled as a single event but which encompassed any of the following: MI, angina pectoris, percutaneous coronary intervention, or coronary artery bypass grafting.
McKay at el32	Familial hypercholesterolemia	UK	One-year-olds to 2-year-olds	Universal screening of FH (using cholesterol and/or genetic screening)	No universal screening (ongoing cluster testing)	Well (entry state), stable angina, poststable angina, unstable angina, MI, post-MI, transient ischemic attack, post-transient ischemic attack, stroke, poststroke, CHD death, non-CHD CVD death, and non-CVD death.
Pelczarska et al63	FH	Poland	Six-year-olds, first job takers, or individuals after an acute coronary syndrome event (all followed by cascade screening)	Screening for FH	No screening	Any CVD, which served as summary category for CHD, angina pectoris, heart failure, stroke, and MI. Four states were possible: general, CVD, post-CVD, and death.
Asphaug and Melberg33	Hereditary breast and ovarian cancer	Norway	Patients with breast cancer aged <60 years (and first-degree female relatives if positive)	Testing for pathogenic variants in a 7-gene or a 14-gene panel	BRCA1/2 screening	Breast cancer, ovarian cancer
Eccleston et al56	Hereditary breast and ovarian cancer	UK	All women with epithelial ovarian cancer	Testing for germline BRCA variants (for the benefit of first- and second-degree relatives)	No germline genetic screening	Breast cancer, ovarian cancer
Hoskins et al57	Hereditary breast and ovarian cancer	Canada	All women with epithelial ovarian cancer	Testing for germline BRCA variants (for the benefit of first- and second-degree relatives)	No germline genetic screening	Ovarian cancer
Kemp et al64	Hereditary breast and ovarian cancer	UK	Female and male patients with an expected 10% chance of pathogenic variants (early-onset breast cancer or family history indication of hereditary breast and ovarian cancer)	Testing for pathogenic variants using a 9-gene panel	No germline genetic screening	Breast cancer, ovarian cancer
Kwon et al65	Hereditary breast and ovarian cancer	Canada	First-degree relatives of women with ovarian cancer	Testing for pathogenic variants in BRCA1/2	No genetic screening	Breast cancer, ovarian cancer
Li et al66	Hereditary breast and ovarian cancer	US	Asymptomatic 40- (or 50-) year-old women with family history of breast or ovarian cancer	Testing for pathogenic variants in a 7-gene panel of breast cancer–associated genes	Only screening BRCA1/2	Breast cancer, ovarian cancer
Lim et al67	Hereditary breast and ovarian cancer	Malaysia	Female patients with breast cancer in a low/middle income country setting (Malaysia)	Screening for pathogenic variants in BRCA1/2	Routine clinical surveillance without genetic testing	Breast cancer, ovarian cancer
Manchanda et al68	Hereditary breast and ovarian cancer	US and UK	All women	Screening for pathogenic variants in BRCA1/BRCA2/RAD51C/RAD51D/BRIP1/PALB2	BRCA1/2 testing only in women who meet family/personal history criteria	Breast cancer, ovarian cancer
Manchanda et al39	Hereditary breast and ovarian cancer	UK	Ashkenazi Jewish women aged >30 years	Screening for specific BRCA founder variants (2.5% pathogenic variant prevalence)	Testing just those who meet personal/family history criteria (9.4% pathogenic variant prevalence)	Breast cancer, ovarian cancer
Manchanda et al69	Hereditary breast and ovarian cancer	US and UK	Women with 1, 2, 3, or 4 Ashkenazi Jewish grandparents	Testing for pathogenic variants in BRCA 1/2 (1.1%, 1.6%, 2.0%, and 2.5% pathogenic variant prevalence, respectively)	Testing just those who meet family/personal history criteria (9.4% pathogenic variant prevalence)	Breast cancer, ovarian cancer
Müller et al54	Hereditary breast and ovarian cancer	Germany	Women aged 35 years with family history indications (>10% risk)	Testing for variants in BRCA1/2	No genetic testing	Breast cancer, ovarian cancer
Kwon et al65	Hereditary breast and ovarian cancer	US and UK	SJ women aged 30 yeas	Screen for the SJ BRCA1 founder variants	BRCA1/2 testing just those who meet family/personal history criteria	Breast cancer, ovarian cancer, cardiac events
Tuffaha et al70	Hereditary breast and ovarian cancer	Australia	Forty-year-old female patients with breast cancer with >10% risk of BRCA variants (and first- and second-degree relatives if positive)	Screen for pathogenic BRCA variants	No BRCA screening	Breast cancer, ovarian cancer
Neusser et al28	Hereditary breast and ovarian cancer	Germany	Women in Germany, aged 25–65, with relatives with confirmed pathogenic variants in BRCA1/2 or another moderate risk gene. The model starts with 2509 women, and new women enter the model each year, for a total of 47,659 after 10 years.	Increased demand (90% genetic test uptake) for screening for variants in BRCA1/2	Current rates of genetic testing (9% genetic test uptake)	Breast cancer, ovarian cancer
Graaff et al37	Hereditary hemochromatosis	Australia	Males aged 30 years males and females aged 45 years of northern European ancestry	Screen for HFE C282Y variant homozygosity	Cascade or incidental screening	Four different hemochromatosis categories were possible, each of which represented an assortment of distinct health outcomes. Category 3 included early symptoms (eg, arthritis, fatigue, lethargy) and category 4 included organ damage (eg, liver cirrhosis, hepatocellular carcinoma, heart disease, type 2 diabetes)
Barzi et al71	Lynch syndrome	US	General population	Twenty different diagnostic algorithms that include predictive models, MSI, IHC, BRAF, and germline DNA testing for Lynch syndrome	No screening	At risk for CRC, curable CRC, noncurable CRC, curable gynecologic cancers, noncurable gynecologic cancer, curable other cancer (not CRC or gynecologic), noncurable other cancers, death.
Chen et al72	Lynch syndrome	Italy	First-degree relatives of patients with known pathogenic MMR variants	Screening using genetic testing with intensive surveillance	No genetic testing with intensive surveillance for all first-degree relatives	Colon and endometrium cancers
Tuffaha et al70	Lynch syndrome	Taiwan	Patients with newly-diagnosed CRC (and relatives if positive)	Four different diagnostic strategies which include a combination of IHC, BRAF, MSI, and germline DNA testing	Routine FIT screening for a minority of the population	CRC
Gallego et al73	Lynch syndrome	US	Patients referred to the medical genetics clinic for CRC and polyposis syndrome evaluation	Testing using next-generation sequencing	Sequential evaluation for Lynch syndrome recommended by current guidelines	CRC
Gansen et al36	Lynch syndrome	Germany	Patients with newly-diagnosed CRC (and their first-degree relatives)	Twenty-one different diagnostic algorithms that include Revised Bethesda and Amsterdam II criteria, MSI, IHC, BRAF, and germline DNA testing	No screening	Well, CRC, metachronous CRC, well after cancer, and death (cancer stages were classified as 1–4)
Goverde et al74	Lynch syndrome	Netherlands	Patients aged ≤70 years with EC (and relatives if positive)	Testing for LS using a combination of MSI, IHC, and germline DNA analysis	Testing in patients aged ≤50 years with EC	CRC and EC
Leenen et al55	Lynch syndrome	Netherlands	All patients aged ≤70 years with CRC (and relatives if positive)	Testing for LS using MSI, IHC, and MLH1 hypermethylation followed by germline testing	Testing all patients aged ≤50 or ≤60 years with CRC	Presumably CRC, although details of Life Year Gained estimates are unclear
Severin et al75	Lynch syndrome	Germany	Patients with newly-diagnosed CRC and their first-degree relatives	Twenty-one different diagnostic algorithms that include Revised Bethesda and Amsterdam II criteria, MSI, IHC, BRAF, and germline DNA testing for Lynch syndrome	No screening	Well, CRC, metachronous CRC, well after cancer, and death (cancer stages were classified as 1–4)
Snowsill et al31	Lynch syndrome	UK	Patients with newly-diagnosed CRC and their biological relatives	Nine different diagnostic algorithms that include MSI, IHC, BRAF V600E, MLH1 promotor methylation testing and germline DNA testing	No testing	CRC and EC
Snowsill et al38	Lynch syndrome	UK	Individuals (aged <50 years) with newly-diagnosed early-onset CRC (not metachronous CRC) and their relatives	Nine different diagnostic algorithms that include Amsterdam II criteria, MSI, IHC, BRAF, and germline DNA testing for LS	No testing	CRC, metachronous CRC, EC, death from prophylactic hysterectomy
Johnson et al76	MODY	Australia	Children presenting with diabetes	Testing for MODY using targeted massively parallel sequencing testing	Ad hoc testing for MODY using Sanger sequencing on clinical grounds	Nephropathy, neuropathy, retinopathy, CVD, severe hypoglycemia, diabetic ketoacidosis, end-stage renal disease, cardiovascular events, or other (non–diabetesrelated)
Naylor et al51	MODY	US	Patients aged 15 to 40 years with newly-diagnosed T2D	Testing for HNF1A-, HNF4A-, and GCK MODY	No testing	Blindness, renal failure, amputation, CHD, MI, congestive heart failure, and stroke
van Nguyen et al77	MODY	US	Patients with diabetes diagnosed before age 45 years	Testing using algorithm driven MODY testing (GAD antibodies—antibody testing followed by 16 gene panel)	No testing	No complications associated with MODY or T2D are considered
Bennette et al78	Multiple conditions: hereditary breast and ovarian cancer, Lynch syndrome, FH, hypertrophic/DCM, long QT syndrome, ARVD, malignant hyperthermia susceptibility	US	Three distinct patient populations (those with cardiomyopathy, those with CRC, or healthy individuals)	Returning incidental findings from next-generation genome sequencing	Not returning incidental findings	Malignant hyperthermia event, sudden cardiac death, heart failure, stroke (potentially others from borrowed CEA models)
Zhang et al35	Multiple conditions: hereditary breast and ovarian cancer, Lynch syndrome; carrier testing for cystic fibrosis, spinal muscular atrophy, fragile X syndrome	Australia	All adults aged 18 to 25 years	Screening for cancer risk and carrier status	No screening	Breast, ovarian cancer, cancer, and CRC
Ngeow et al79	Other cancers: CS	US	CS-like patients	PTEN CC score as a clinical risk calculator to identity for PTEN germline testing	No use of PTEN germline testing	Breast, endometrial, kidney, and thyroid cancer
Compagni et al80	Other cancers: NF1	US	Pediatric patients with suspected NF1 (1.3% risk of legius) or suspected NF1 with cafe-au-lait spots (2.95% risk of legius)	Screening for pathogenic variants in SPRED1 to rule out NF1	No genetic testing, depending on age at genetic testing	None
Rubio-Terrés et al53	Thrombophilia	Italy	Women aged 15 to 45 years at risk for VTE who are seeking oral contraception	Testing for genetic risk factors	Either a battery of biochemical tests or no testing	Disease sequelae associated with pulmonary embolism (recurrent VTE events, hemorrhage due to warfarin, MI, stroke, and pulmonary hypertension) and deep vein thrombosis (recurrent VTE events, hemorrhage due to warfarin, MI, stroke, and post-thrombotic syndrome).
Farnaes et al23	Thrombophilia	Spain	Patients with VTE	Testing using a 12-gene panel (Thrombo inCode)	Testing only factor V Leiden and prothrombin G20210A	Deep vein thrombosis, pulmonary embolism, bleeding caused by warfarin
Compagni et al80	Undiagnosed pediatric disorders: multiple clinical conditions possible, including those of the following systems: neurological, hepatic, cardiac, hematological, gastrointestinal, endocrine/biochemical, musculoskeletal, pulmonary	US	Acutely-ill infants	Rapid ES	Standard genetic testing	A wide variety of health outcomes associated with the rare clinical conditions for each infant, including among others seizures, severe cholestasis, respiratory distress and metabolic acidosis, hyperinsulinemia. Health outcomes were not modeled but rather reported on the basis of retrospective analysis of individual patient trajectories.
Rubio-Terrés et al53	Undiagnosed pediatric disorders: multiple structural malformations and/or unexplained developmental delay/intellectual disability (specific conditions not reported)	Singapore	Children with developmental delay	ES	Standard care (chromosome microarray)	Not explicitly reported nor modeled
Vrijenhoek et al25	Undiagnosed pediatric disorders: neurodevelopmental delay	Netherlands	Infants with intellectual disabilities	ES	No ES	Specific health outcomes were not reported
Schofield et al81	Undiagnosed pediatric disorders: suspected monogenic disorders	Australia	Infants with suspected monogenic disorders	ES	Standard diagnostic pathway with singlegene and multigene panel tests and complex/invasive tests	Specific health disutilities are provided for each infant included in the analysis (see their Supplemental Table 1). Select health outcomes projected included but were not limited to severe mental retardation and severe cerebral palsy.
Stark et al34	Undiagnosed pediatric disorders: suspected monogenic disorders	Australia	Infants with suspected monogenic disorders	ES	Standard diagnostic pathway with singlegene and multigene panel tests and complex/invasive tests	Projected health outcomes in the absence of treatment included the following for the 4 infants for whom ES diagnosis resulted in a change in disease management (specific outcomes were unique for each infant): alternating hemiplegia, hyperkalemia, progressive deterioration (probably fatal), and continued need for blood transfusions.

ARVD, arrhythmogenic right ventricular cardiomyopathy; CC, Cleveland Clinic; CEA, cost-effectiveness analysis; CHD, coronary heart disease; CRC, Colorectal cancer; CS, Cowden syndrome; CVD, cardiovascular disease; DCM, dilated cardiomyopathy; EC, endometrial cancer; ES, exome sequencing; FH, Familial hypercholesterolemia; FIT, fecal immunochemical test; IHC, immunohistochemistry test; LDL, low density lipoprotein; LS, Lynch syndrome; MDCM, mild dilated cardiomyopathy; MI, myocardial infarction; MODY, Maturity-onset diabetes of the young; MSI, microsatellite instability; NF1, neurofibromatosis type 1; SJ, Sephardic Jewish; T2D, type 2 diabetes; TIA, transient ischaemic attacks; UK, United Kingdom; US, United States; VTE, venous thromboembolism.

Table 2 shows the major model characteristics across all studies. Most studies used the combination of a decision tree with a Markov model (n = 17), although several studies used either just a decision tree (n = 11) or just a Markov model (n = 6). Five studies employed some form of individual-level simulation (eg, microsimulation). Less than half of all studies (n = 18) compared only 1 alternative to usual care, which often consisted of the standard-of-care genetic or clinical testing/screening protocol. Most studies conducted CUAs (ie, health outcomes are expressed in utility measures such as quality-adjusted life-year [QALYs] or disability-adjusted life-year [DALYs]) (n = 26), with several conducting both CUAs and CEAs (ie, health outcomes are expressed in clinical measures such as total diagnoses or deaths) (n = 10) and a limited number (n = 6) conducting cost-consequence analyses. Three studies incorporated societal costs, and the rest were strictly from either the health care sector or payer perspective.

Table 2.

Primary modeling characteristics of included studies

Study	Type of Evaluation	Perspective	Discounting	Time Horizon	Model Type	Costing Method	Sensitivity Analyses Conducted	Forms of Analysis Presentation
Catchpool et al41	CUA	Health care system (Australian Government)	5% costs and outcomes	Lifetime	Decision tree and Markov model	Gross	PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Ademi et al40	CEA and CUA	Health care system (Australian Government)	5% costs and outcomes	10 years	Decision tree and Markov model	Gross	PSA, 1/2-way deterministic analysis	CE plane/scatter plot, CEAC
Chen and Hay45	CUA	Societal and health care system combined	3% costs and outcomes	Lifetime	Decision tree and Markov model	Gross	Threshold analysis, PSA, 1/2-way deterministic analysis	CEAC, Tornado Diagram
Crosland et al30	CUA	Health care system (UK NHS)	3.5% costs and outcomes	Lifetime	Decision tree and Markov model	Micro	Threshold Analysis, PSA, 1/2-way deterministic analysis	CE plane/scatter plot, CEAC
Kerr et al62	CUA	Health care system (UK NHS)	3.5% costs and outcomes	30 years	Markov model	Micro	1/2-way deterministic analysis	None
Lázaro et al44	CEA and CUA	Health care system (Spanish National Health System) and Societal	3% costs and outcomes	10 years	Decision tree	Gross	Scenario analysis, 1/2-way deterministic analysis	CE Frontier
McKay et al32	CUA	Health care system (UK NHS)	3.5% costs and outcomes	Lifetime (limited to 100 years)	Decision tree and Markov model	Micro	Threshold analysis, PSA, 1/2-Way deterministic analysis	CEAC
Pelczarska et al63	CEA and CUA	Health care system (Polish Government)	5% costs, 3.5% outcomes	Lifetime	Decision tree and Markov model	Gross	One/two-way deterministic analysis	None
Asphaug and Melberg33	CUA	Health care sector	4% costs and outcomes	Lifetime (limited to 100 years)	Patient-level microsimulation with memory	Micro	PSA	CEAC
Eccleston et al56	CUA	Health care system (UK NHS)	3.5% costs and outcomes	50 years	Patient-level microsimulation with memory	Gross	Scenario analysis, PSA, 1/2-way deterministic analysis	CE plane/scatter plot, CEAC
Hoskins et al57	CUA	Canadian health care system perspective	1.5% costs and outcomes	50 years	Patient-level microsimulation with memory	Gross	Scenario analysis, PSA, 1/2-way deterministic analysis	None
Kemp et al64	CUA	Health care system (UK NHS)	3.5% costs and outcomes	50 years	Patient-level microsimulation with memory	Gross	PSA	CEAC
Kwon et al52	CUA	Health care system (Canadian Government)	3% costs and outcomes	50 years	Decision tree and Markov model	Gross	Threshold analysis, scenario analysis, 1/2-way deterministic analysis	None
Li et al66	CEA and CUA	Health care payer	3.5% costs and outcomes (strictly for QALYs, not life-years)	Lifetime (limited to 100 years)	Decision tree and Markov model	Gross	PSA, 1/2-way deterministic analysis	CE plane/scatter plot, CEAC, Tornado diagram
Lim et al67	CEA and CUA	Health care system (Malaysian Government)	3% costs and outcomes	Lifetime	Decision tree and Markov model	Gross	Scenario analysis, PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Manchanda et al68	CEA and CUA	Health care system (US and UK)	3.5% costs and outcomes	Lifetime (to age 83 based on life tables)	Decision tree	Gross	Scenario analysis, PSA, 1/2-way deterministic analysis	CEAC
Manchanda et al39	CUA	Health care system (UK NHS)	3.5% cost and outcomes	Lifetime (to age 83 based on life tables)	Decision tree	Gross	Scenario analysis, PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Manchanda et al69	CEA and CUA	Health care system (US and UK)	3.5% costs and outcomes	Lifetime (to age 83 based on life tables)	Decision tree	Gross	Scenario analysis, PSA	CEAC
Müller et al54	CEA and CUA	Health care payer (German Statutory Health Insurance)	3% costs and outcomes	65 years	Decision tree and Markov model	Gross	PSA, 1/2-way deterministic analysis	CEAC
Patel et al82	CEA and CUA	Health care payer	3.5% costs and outcomes	Lifetime (up until 83 and 82 years for UK and US women, respectively)	Markov model	Gross	Scenario analysis, PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Tuffaha et al70	CUA	Health care system (Australian Government)	5% costs and outcomes	Lifetime (limited to 90 years)	Decision tree and Markov model	Gross	PSA, 1/2-way deterministic analysis	None
Neusser et al28	CCA	Health care payer (German Statutory Health Insurance)	3% costs	10 years	Markov model	Gross	None	None
Graaff et al37	CUA	Health care system (Australian Government)	5% costs and outcomes	Lifetime	Markov model	Micro	PSA, 1/2-way deterministic analysis	CEAC
Barzi et al71	CEA	Societal (no clear societal costs) and health care system combined	3% costs and outcomes	Whichever comes first: the death, age 80 years, or 50 years of follow-up.	Decision tree followed by Markov -based individual patient simulation	Micro	Scenario analysis, 1/2-way deterministic analysis	None
Bonfanti et al26	CCA	Not stated (assumed health care system)	Discounted at the 2012 level	10 years	Informal epidemiological model	Micro	None	None
Chen et al72	CEA	Health care system (MOHW) of the Taiwan government)	3% costs and outcomes	Lifetime	Decision tree and Markov model	Gross	PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Gallego et al73	CUA and CEA (exclusively CUA in sensitivity analyses)	Not stated (assumed health care payer)	3% (unclear how applied)	Lifetime	Decision tree	Gross	Scenario analysis, PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Gansen et al36	CEA	Health care payer (German Statutory Health Insurance)	3% costs and outcomes	120 years	Decision tree and Markov model	Micro	Scenario analysis, PSA	CE Frontier
Goverde et al74	CEA	Not stated	3% costs and outcomes	Not stated (presumably lifetime)	Decision tree	Micro	1/2-way deterministic analysis	Tornado diagram
Leenen et al55	CEA	Health care sector	3% costs and outcomes	Lifetime	Decision tree	Micro	None	Tornado diagram
Severin et al75	CEA	Health care payer (German Statutory Health Insurance)	3% costs and outcomes	Lifetime (limited to 120 years)	Decision tree and Markov model	Micro	Scenario analysis, PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Snowsill et al31	CUA	Health care system (UK NHS) and Personal Social Service	3.5% costs and outcomes (strictly for QALYs, not life-years)	Lifetime (limited to 100 years)	Decision tree and individual patient simulation	Micro	Scenario analysis	CE Frontier
Snowsill et al38	CUA	Health care system (UK NHS)	3.5% costs and outcomes	Lifetime (limited to 100 years)	Decision tree and individual patient simulation	Micro	Scenario analysis, 1/2-way deterministic analysis	CE Frontier, Tornado diagram
Johnson et al76	CUA	Health care system (Australian Government)	3% costs and outcomes	10 years and 30 years	Decision tree and Markov model	Gross	1/2-way deterministic Analysis	Tornado diagram
Naylor et al51	CUA	Health care system	3% costs and outcomes	Lifetime	Decision tree followed by Markov-based individual patient simulation	Gross	Threshold analysis, scenario analysis, 1/2-way deterministic analysis	Tornado diagram
van Nguyen et al77	CUA	Health care payer	3.5% costs and outcomes	30 years	Decision tree	Gross	Threshold analysis, PSA, 1/2-way deterministic analysis	CEAC, Tornado diagram
Bennette et al78	CUA	Health care system	3% costs and outcomes	Lifetime	Decision tree and Markov model	Gross (costs based on prior CEAs)	Threshold analysis, scenario analysis, PSA, 1/2-way deterministic analysis	CEAC
Zhang et al35	CUA	Health care system	3% costs and outcomes	Lifetime	Decision tree	Gross	Scenario Analysis, PSA, 1/2-way deterministic analysis	CE plane/scatter plot
Ngeow et al79	CUA	Societal and health care system combined	3% costs and outcomes	Lifetime	Decision tree and Markov model	Gross	Scenario analysis, PSA	CEAC, Tornado diagram
Muram et al27	CCA	Health care payer	3% costs and outcomes	17 years (18 months old - 18 years old)	Markov model and individual patient simulation	Gross	None	None
Compagni et al80	CUA	Health care system (Italian National Health System)	3.5% costs and benefits	Lifetime	Decision tree	Micro	Scenario analysis, PSA, 1/2-way deterministic analysis	CE plane/scatter plot, CEAC, Tornado diagram
Rubio-Terrés et al53	CUA	Health care system (UK NHS)	3.5% costs and outcomes	35 years	Decision tree	Gross	Threshold analysis, scenario analysis, PSA, 1/2-way deterministic analysis	CE plane/scatter plot, Tornado diagram
Farnaes et al23	CCA	Health care system	N/A	Various for different infants	N/A	Gross	None	None
Hayeems et al24	CCA	Health care system	N/A	On average, 15 months after diagnostic results (standard care or GS) were reported.	Linear mixed effects model	Gross	None	None
Vrijenhoek et al25	CCA	Health care system	N/A	The length of follow-up was, on average, 240 days after ES and 922 days before ES.	None	Micro	None	None
Schofield et al81	CUA	Not stated	5% (unclear how applied)	20 years	Decision tree	Gross	1/2-way deterministic analysis	None
Stark et al34	CUA	Health care system	None stated	Mediation duration of follow-up: 473 days (interquartile range: 411–650)	No formal model (individual prospective cohort)	Gross	PSA	CE plane/scatter plot

CCA, const-consequence analysis; CEA, cost-effectiveness analysis; CEAC, cost-effectiveness acceptability curve; CE, cost-effectiveness frontier; CE Plane, cost-effectiveness plane; CUA, cost-utility analysis; DMC, dilated cardiomyopathy; GS, genome sequencing; MOHW, The Ministry of Health and Welfare; N/A, not available; NHS, National Health Service; PSA, probabilistic sensitivity analysis; QALY, quality-adjusted life-year; UK, United Kingdom; US, United States.

BMJ checklist assessment

Some basic items from the BMJ checklist were fully met by nearly all studies, including “The research question is stated” (average score [AS]: 2) and “The primary outcome measure(s) for the economic evaluation are clearly stated” (AS: 1.98). Conversely, several checklist items consistently were not met or partially met by all studies, including “Quantities of resources are reported separately from their unit costs” (AS: 0.87) and “Details of the subjects from whom valuations were obtained are given (AS: 1.10). Some items were consistently addressed by citing external sources but without an overview of the source material (N/A), such as “Methods to value health states and other benefits are stated.” Several of the cost-consequence analyses received an above average number of N/R assessments. ASs for each variable are presented in Table 3. Although comparative assessment of studies is not the primary focus of our analysis and the BMJ checklist is not intended to produce a quantitative assessment, the distribution of score and the average score for each article is presented in Appendix Table 6 and Appendix Table 7.

Table 3.

BMJ checklist scores across all items

BMJ Checklist Item	Total 2s	Total 1s	Total 0s	Total N/Rs	Total N/As	Average Scorea
The research question is stated	47	0	0	0	0	2.00
The economic importance of the research question is stated	25	13	9	0	0	1.34
The viewpoint(s) of the analysis are clearly stated and justified	31	11	5	0	0	1.55
The rationale for choosing the alternative programs or interventions compared is stated	44	3	0	0	0	1.94
The alternatives being compared are clearly described	42	5	0	0	0	1.89
The form of economic evaluation used is stated	39	3	0	5	0	1.93
The choice of form of economic evaluation is justified in relation to the questions addressed	43	0	0	4	0	2.00
The source(s) of effectiveness estimates used are stated	43	1	0	1	2	1.98
Details of the design and results of effectiveness study are given (if based on a single study)	16	2	0	28	1	1.89
Details of the method of synthesis or meta-analysis of estimates are given (if based on an overview of a number of effectiveness studies)	10	10	6	20	1	1.15
The primary outcome measure(s) for the economic evaluation are clearly stated	46	1	0	0	0	1.98
Methods to value health states and other benefits are stated	13	2	13	11	8	1.00
Details of the subjects from whom valuations were obtained are given	14	5	11	10	7	1.10
Productivity changes (if included) are reported separately	1	0	4	42	0	0.40
The relevance of productivity changes to the study question is discussed	1	2	2	42	0	0.80
Quantities of resources are reported separately from their unit costs	15	9	21	0	2	0.87
Methods for the estimation of quantities and unit costs are described	26	12	4	0	5	1.52
Currency and price data are recorded (year, currency of costs, break into key components)	35	7	4	0	1	1.67
Details of currency of price adjustments for inflation or currency conversion are given	25	3	17	1	1	1.18
Details of any model used are given	38	4	1	1	3	1.86
The choice of model used and the key parameters on which it is based are justified	40	3	0	2	2	1.93
Time horizon of costs and benefits is stated	36	6	2	3	0	1.77
The discount rate(s) is stated	42	1	1	3	0	1.93
The choice of rate(s) is justified	22	4	18	3	0	1.09
An explanation is given if costs or benefits are not discounted	0	0	2	45	0	0.00
Details of statistical tests and confidence intervals are given for stochastic data	8	1	32	5	1	0.41
The approach to sensitivity analysis is given	34	8	1	4	0	1.77
The choice of variables for sensitivity analysis is justified	25	5	11	5	1	1.34
The ranges over which the variables are varied are stated	36	2	3	5	1	1.80
Relevant alternatives are compared	38	3	5	1	0	1.72
Incremental analysis is reported	43	0	1	3	0	1.95
Major outcomes are presented in a disaggregated as well as aggregated form	43	0	1	3	0	1.95
The answer to the study question is given	47	0	0	0	0	2.00
Conclusions follow from the data reported	47	0	0	0	0	2.00
Conclusions are accompanied by the appropriate caveats	41	6	0	0	0	1.87

N/A, not applicable; N/R, not relevant.

^aAverage quality scores were calculated for each question by summing the 1s and 2s each article received across all studies then dividing that sum by the number of items for which 0s, 1s, and 2s were possible.

Assessment of key methodological constructs

Perspective, scope, and parameter selection

For studies that based effectiveness estimates for preventive interventions on several sources (n = 25), roughly a third (n = 7) presented a thorough evidence synthesis, which outlined how they identified the parameters used in their analysis. A systematic literature review was conducted and a formal meta-analysis was completed for important variables in only 4 articles (Appendix Table 5).^30–33

A total of 14 articles either conducted microcosting or referenced previous microcosting analyses, whereas the rest opted for a macrocosting approach. Furthermore, several studies (n = 7) adopted costing information from other, similar CEAs without justifying the primary source of the costing data.

Of studies with a clearly stated perspective, all presented at least a health care payer or health care system perspective. Three articles also included components of a societal perspective; 2 of these studies incorporated lost labor productivity costs into overall costs and 1 conducted 2 separate analyses from either the health care sector or societal perspective. No studies incorporated nonmedical benefits of genetic screening or testing, such as the personal utility of nonactionable genetic information or psychological benefits of negative test results. Studies that only examined carrier screening were excluded from the review, although 2 studies either incorporated costs associated with the use of assisted reproductive technology by parents after a child’s genetic diagnosis or DALYs averted by decisions to avoid having children with genetic disease.^34,35 One study included a discussion of the privacy implications of familial cascade testing,³⁶ although privacy costs were not incorporated into their model.

No studies calibrated their model using real-world data. Two articles attempted some form of internal or external model validation, although this was not conducted to inform model parameterization but rather to confirm that model outcomes aligned with data used within the model and external values (eg, known prevalence of disease).^33,37

Use of sensitivity/uncertainty analyses

Although all evaluations considered in this review conducted sensitivity analyses, the depth, breadth, and presentation of analyses varied widely. Most studies (n = 33) conducted some form of 1-way or 2-way deterministic sensitivity analyses, and 19 of such studies presented the results in the form of a tornado diagram. Among the 29 studies that included a probabilistic sensitivity analysis (PSA), 9 displayed PSA results in incremental cost-effectiveness ratio (ICER) scatter plots, 23 presented cost-effectiveness acceptability curves, but only 3 presented uncertainty intervals for primary estimates. A total of 21 studies conducted at least 1 scenario analysis, and 8 studies conducted at least 1 threshold analysis. Only 2 value of information analyses were conducted, which included an expected value of perfect information analysis and an expected value of partial perfect information analysis for specific parameter groups (eg, treatment costs, probability of cancer recurrence) (Table 2).

Reporting transparency

Both the study question and answer to the study question were clear in all papers. Only 2 studies did not clearly report the discount rate for their analysis, although many studies (n = 19) did not provide a proper justification of why their specific rate was selected. Similarly, most papers (n = 39) clearly articulated the year and price information of their cost units but only about half (n = 24) reported whether or how these prices had been adjusted for inflation or currency conversion.

All articles presented both disaggregated outcomes (such as total QALYs gained or total health care costs) as well as final ICER calculations. However, less than half of the studies (n = 23) based the population size on a real-world population. Only 1 article disaggregated intervention costs into specific categories unique to genetic screening, and 9 studies disaggregated costing results on the basis of the generic source of costs such as genetic sequencing, disease prevention, and disease treatment.

For studies that reported results in the form of either QALYs or DALYs (n = 37), about half (n = 16) presented the valuation method or study by which their utility values were generated and slightly more than half (n = 19) reported the population from whom these values were generated.

Of the 19 studies that reported the results of 1-way sensitivity analyses in the form of a tornado diagram, 9 had figures that did not indicate the direction of the associations between each parameter and the ICER. It was also unclear for several studies (n = 10) why certain variables were ultimately selected to be included in deterministic sensitivity analyses (such as tornado diagrams) and not others.

Discussion

Overview of major findings

This systematic review analyzed the methodological quality of 47 recent economic evaluations of genetic screening or testing for monogenic disorders across disease arenas. There was substantial variation in model sophistication and reporting quality. Most articles satisfied basic criteria for their presentation of parameter values, model design, and results as well as their implementation and interpretation of sensitivity/uncertainty analyses. A few studies achieved higher levels of sophistication or quality and can serve as exemplars for future work.^{31,33,35,38–41}

Recommendations for future evaluations and exemplar cases

Although uniformity of evaluation design and reporting should not come at the cost of analytical flexibility, the heterogeneity of quality assessed in our review suggests the importance of further training to develop high-quality economic evaluations of genetic screening/testing. Scholars are encouraged to reference 1 or more of the guidelines that have been published over the past 20 years; these guidelines demonstrate near consensus on the key elements of an economic evaluation.^20,42 Within the last 5 years, several textbooks have also been published on the proper methodological approach to economic evaluation.^3,7,29

Informed by our assessment and considering authoritative sources, we make several recommendations for future economic evaluations of genetic testing/screening. Our recommendations focus on 3 arenas that consistently caused difficulty in the articles considered in our review (parameter selection, use of sensitivity/uncertainty analyses, and reporting transparency). Table 4 summarizes this discussion along with several exemplar cases from our review that are provided to demonstrate recommended practices.

Table 4.

Review-informed recommendations across methodological constructs

Methodological Construct	Identified Challenge	Emphasized Recommendation	Exemplar Studies Identified in Systematic Review
Perspective, scope, and parameter selection	As the variety of genetic testing/screening interventions expand (eg, full-gene sequencing, multi-gene panels, exome or genome sequencing), it is difficult to track the accuracy of these interventions (eg, sensitivity and specificity)	For parameter values that are especially influential or uncertain, conduct systematic reviews (with or without meta-analyses, depending on the consensus of the review); provide justifications for variations in parameter values when consensus is not available. For parameter values that are likely to change in different environments, base estimates on available evidence and justify choices.	In the context of familial hypercholesteremia, Crosland et al30 conducted a systematic review to determine the diagnostic accuracy of the Simon Broome and Dutch Lipid Clinic Network clinical assessment tools (incidentally, the review also determined the absence of information available to inform uptake probabilities) See also: McKay et al32 and Asphaug and Melberg33
	Estimating the costs of implementing a new genetic screening or testing intervention in practice or the ongoing costs such as training or clinical decision support systems that need to be maintained over time to support intervention	Conduct microcosting to estimate the varied sources of cost and categories of cost within the intervention, especially for analyses centered on changes in the way resources are delivered within a specific program or diagnostic odyssey.	Asphaug and Melberg33 used a departmental micro-costing analysis to estimate the cost of materials and equipment as well as direct labor, indirect labor, overhead, capital, and maintenance services for all scenarios included in the model. See also: Crosland et al30; Snowsill et al31,38; Compagni et al80; Vrijenhoek et al25
	When implementing genetic analyses, cost-effectiveness may not be clear for all stakeholder perspectives. It is challenging to appropriately capture all potential benefits from genetic analyses (eg, secondary findings or nonhealth-related personal or reproductive utility) across these perspectives, including distinguishing benefits from screening and cascade testing	To ensure all relevant impacts of the intervention have been considered from all appropriate perspectives (eg, health care and societal as distinct), refer to the Impact Inventory (developed by the Second Panel on Cost-Effectiveness in Health and Medicine)	Lázaro et al44 showed that family cascade testing was shown to be cost-effective (ie, compared with usual care, the additional cost of testing was considered worthwhile given the additional benefits brought) when using the health care sector perspective and dominant (ie, screening was both less costly and more effective than usual care) when using the societal perspective, primarily because of the days off work that testing prevented. See also: Asphaug and Melberg33
Use of sensitivity/uncertainty analyses	The cost of genetic screening and testing interventions is constantly being updated (often becoming cheaper), which has dynamic implications for the cost-effectiveness of such interventions	Conduct threshold analyses to interrogate key parameters that may change in response to policy decisions, programmatic design, or other exogenous factors, such as the cost of genetic screening necessary for an intervention to be cost-effective.	Naylor et al51 conducted a threshold analysis to predict the minimum prevalence of pathogenic variants for MODY at which screening would become cost-saving. Rubio-Terrés et al53 found that the cost of the new genetic tool Thrombo inCode (GEN inCode) would need to fall substantially for it to be cost-effectively used to screen for risk of venous thromboembolism in Spain. See also: Kwon et al52
	Appropriately accounting for potential uncertainty of information, such as the population prevalence of genetic variants or variants of uncertain significance	Conduct value of information analyses to quantify the value of investing in research activities that generate additional evidence that lessens parameter uncertainty.	Asphaug and Melberg33 conducted an EVPPI for select parameter groups (including relative cancer risk, pathogenic variant prevalence, cost of cancer treatment, utility weights), which estimated the net monetary benefit from the removal of uncertainty around parameter values. The authors determined that gaining certainty about the relative cancer risk associated with specific pathogenic variants and the cost of breast cancer treatment had the highest per-person EVPPI. This analysis prompted the authors to advocate for variant-specific prevalence data, which would allow for within-gene stratification in models.
	Genetic testing/screening interventions may be improved by several adaptations to the screening algorithm (eg, which subpopulations to target) or investments in outreach (eg, additional assistance to contact relatives of index cases), for which the cost-effectiveness is unclear and will need to be studied further	Conduct scenario analyses to learn about the relationship between such choices and estimated incremental cost-effectiveness	For instance, Gansen et al36 used scenario analyses to consider whether intensified outreach for cascade testing is cost-effective. For a detailed description of how scenario analyses were used across Familial hypercholesterolemia studies, see Appendix 8 See also: Snowsill et al31,38; McKay et al32; Chen and Hay45
Reporting transparency	Genetic testing/screening interventions may lead to nonhealth-related changes in utility resulting from new awareness of having a condition or health-related changes in utility not commonly described in the literature	Identify valuation studies (ie, studies attempting to assess the utility of distinct health states) among those with your genetic condition of interest, or those that closely parallel that condition; clearly articulate the target populations in which and valuations methods by which the studies were conducted to derive health state utility values.	When presenting the utility values selected for individuals with breast or ovarian cancer, Müller et al54 clearly articulated the populations in which valuation studies were conducted (women with a present pathogenic variant/breast cancer or women from a healthy reference group), the valuation methods used across different studies (TTO or SG), and the reason for ultimately preferring one set of studies over another (SG more accurately reflected health-related quality of life compared with TTO, per their analysis).
	The costs of genetic testing/screening programs are constantly evolving, often at a different pace than other medical goods	Specify inflation or cost adjustments to medical commodities that have increased or decreased in price relative to the rest of the industry.	Gansen et al36 identified medical costs that had been updated and how they were updated (using consumer price indices and purchasing power parity) since a publication of results using the same model 4 years prior, including the impact of new classification of tests relevant to Lynch syndrome (though the specific classification was not mentioned).
	Genetic testing/screening interventions are often composed of several distinct activities that all demand varying resources costs, such as genetic counseling and clinical genetics, phlebotomy and ordering, and sequencing, analysis, and interpretation	When modeling and reporting the costs of the interventions, disaggregate intervention costs into specific categories unique to the genetic condition or disaggregate generic sources of cost into relevant categories for the testing or screening program.	Ademi et al40 helpfully disaggregated intervention costs into specific categories unique to the genetic condition: disease costs, intervention costs, and screening and imaging (although the specific item costs attributed to each category is not clear); if genetic testing/screening costs were to substantially change after their publication, readers would be more able to account for those changes and recalculate cost-effectiveness outcomes, thereby preserving the value of the original evaluation. See also: Leenen et al55; Eccleston et al56; Hoskins et al57

EVPPI, expected value of partial perfect information; MODY, maturity-onset diabetes of the young; SG, standard gamble; TTO, time trade-off.

Perspective, scope, and parameter selection

A central challenge in conducting any economic evaluation is employing expert judgment and the evidence synthesis needed to select or estimate parameter values for the model. A formal systematic review with or without meta-analysis should be attempted for parameter values that are especially influential, uncertain, or likely to change in different environments (eg, as a consequence of policy decisions).^3,30,32,33

Most economic evaluations of genetic testing/screening take a simplistic view of genetic analysis costs, often ignoring costs of implementation and patent outreach. For more realistic integration of the costs incurred by genetic testing/screening, microcosting is recommended.^7,43 Microcosting is especially important for analyses centered on changes in the way resources are delivered within a specific program or diagnostic odyssey, which is often the case for innovative genetic medicine programs.^29,33 Micro-costing may not be suitable for studies primarily concerned with nationally aggregated or long-run costs, and the importance of various subcomponents may depend on the perspective.

Economic evaluations of genetic testing/screening should evaluate value across relevant stakeholders, including but not limited to payer and societal perspectives. Genetic analyses are unusual in that they have implications not just for the individual being tested but also for family members, who may or may not be covered by the same payer. For settings without a single payer, including family members in models requires careful consideration of how and even whether cascade testing is relevant in a payer-perspective analysis. Moreover, it has been strongly recommended that economic evaluations report 2 standard reference case perspectives: 1 from the health care payer perspective (ie, formal health care sector costs borne by third-party payers or paid for out-of-pocket by patients) and, in parallel, 1 from the societal perspective (ie, including patient/family time costs involved in receiving an intervention and for self-management).³ Presenting a reference case from a particular third-party payer (eg, the federal government, a single health care system, or a particular insurance company) can also be warranted, although care should be taken to consider whether the covered population is stable, especially when benefits may lag many years behind initial investments (eg, crossing Medicaid and Medicare programs or attrition from insurance plans). Presenting both analytical perspectives in tandem clarifies how value may vary substantially among key stakeholders.⁴⁴

To account for the balance between the burden of screening and recovered productivity, future studies should refer to the Impact Inventory developed by the Second Panel on Cost-Effectiveness in Health and Medicine to guide which costs should be considered from either perspective.⁵ This resource was used by only 1 study in this review.³³ For the specific context of genomic screening programs, Fragoulakis et al⁷ outlined several direct costs (eg, health care payer costs) and indirect costs (eg, patient productivity loss and family expenses) that may also serve as a useful guide. Examples of indirect costs modeled in reviewed studies included work productivity loss because of illness⁴⁴ and physician visits.⁴⁵ Screening for genetic conditions in the general population may also lead to nonhealth-related changes in utility resulting from new awareness of having a condition that either requires additional interaction with the health care system or cannot be addressed medically.

Model calibration is a process used in economic evaluations to improve the accuracy of parameters that cannot be directly measured, leveraging available data that can be matched with the model.⁴⁶ Calibration efficiently searches the space of plausible parameter values to find those that optimize the model’s fit to real-world data.⁴⁷ Calibration is not always necessary but should be used when it can reduce the amount of parameter uncertainty in the model, especially for the most influential or actionable model parameters.⁴⁸

Authors should pay special attention to test performance in the model. True clinical sensitivity is extremely difficult to measure for most conditions, and categories of possible test results vary between diagnostic testing, family cascade testing, and population screening.⁴⁹ The probability of further interaction between the health care system and patient will differ based on how these categories are reported. Evaluators should ensure that their modeling of test results accurately reflects both what is known about the clinical sensitivity and specificity of the genetic test and how that knowledge is translated into diagnostic protocols, which may vary across sites of implementation.

Use of sensitivity/uncertainty analyses

Beyond reporting outcomes of a base case, economic evaluations of genetic testing/screening should identify and consider the impact of stochastic, parameter, and/or structural uncertainty as well as patient heterogeneity. Analyses should distinguish between variability in inputs that may affect outcomes (sensitivity analysis) and uncertainty in model inputs that may alter the uncertainty of model conclusions (uncertainty analysis).^29,50 As with all economic evaluations, it is challenging to estimate the collective impact that uncertainty within individual parameters will have on the uncertainty of overall model outcomes. We strongly recommend studies to conduct a probabilistic uncertainty analysis (PUA) and to use the PUA results to clearly report the degree of uncertainty of estimates for primary outcomes of interest (ie, CIs). Given their likely dramatic impact on model outcomes, we recommend studies to consider the following parameters within their PUA: pathogenic variant prevalence (which depends on the target population and clinical scenario), probability of referral to genetic counseling and genetic testing uptake, likelihood of clinical outcomes (based on penetrance and expressivity of the condition), uptake/adherence and efficacy of interventions in symptomatic and presymptomatic individuals, morbidity and mortality in the absence of a genetic diagnosis, and cost of genetic analysis, implementation of interventions, and care used as part of post-result clinical interventions.

Studies should incorporate threshold analyses (a type of sensitivity analysis) to interrogate key parameters that may change in response to policy decisions, programmatic design, or other exogenous factors. Threshold analyses identify the minimum or maximum value for a given parameter that results in the intervention meeting willingness-to-pay thresholds. In the context of genetic testing/screening, this may fruitfully be applied to parameters such as the prevalence of the pathogenic variant being screened, with the assumption that programs could be developed to target populations with a critical prevalence rate (eg, those with a clinical history suggestive of genetic disease). Threshold analyses could also determine the minimum rate of uptake for accepting genetic testing/screening or prophylactic interventions for screening to become cost-effective.^51,52 It is widely appreciated that genetic laboratory costs have reduced over the past decade, and there is speculation over whether testing prices will continue to fall or may even increase if testing companies capture greater control of markets. This value should also be strongly considered for threshold analyses.⁵³

As with any novel intervention, many parameters necessary to evaluate genetic testing/screening are fixed but unknown or uncertain. Some of these parameters, such as the prevalence of pathogenic variants in populations, could be studied using biobank or cohort studies and epidemiological research methods. Value of information analyses should be conducted to quantify the value of investing in research activities that generate additional evidence that lessens parameter uncertainty.³ This type of analysis informs what research is most valuable—essential information for researchers and funders.

Scenario analyses should be used to estimate structural uncertainty or to compare different intervention approaches in a model. In the context of genetic testing/screening, they could be used to consider alternative scenarios in which more energy is dedicated to certain subpopulations or the diagnostic pathway is slightly modified for these subpopulations.^31,36,38 The consistent incorporation of scenario analyses will not only make models more informative (by calling attention to particularly uncertain or variable parameters) but also improve methodological rigor because authors are forced to critically think about the specific questions that their model must be designed to address.

Reporting transparency

The amount of content necessary to properly present an economic evaluation is often too much to fit in a single manuscript, prompting evaluators to reference secondary literature.²² When referencing secondary literature (especially for parameter estimation), summary information should be available within the main manuscript or appendix for readers to understand the context and methods behind the results produced from that literature.⁵⁴

When price transformations are necessary—either between different years or between different currencies—authors must be clear what year was used as the benchmark and what exchange rate was employed for the transformation. When adjusting for inflation, authors should use inflation rates unique to the medical industry.²⁹ When relevant, inflation or cost adjustments should be specific to medical commodities that have increased or decreased in price relative to the rest of the industry (eg, when a patented drug becomes available in generic forms).³⁶ Evaluators should also clearly identify when monetary amounts included in the model reflect price or cost estimates; we recommend accounting for all the associated costs of a medical good or activity.²⁹ The cost of the same genetic analysis may also vary considerably depending on the equipment used, throughput level, and sequencing method; we recommend clearly identifying the subcomponents of costs associated with the genetic analysis.

We strongly recommend disaggregating the outcome of CEAs into total costs and total effectiveness. Disaggregation is especially useful if the size of the model population is reported and corresponds to a real-world population rate. This allows for population-wide health and economic impacts (eg, a budgetary impact analysis) to be reported in addition to per-person cost and effectiveness. When expressing the total costs associated with any screening strategy, it is also recommended that authors report both total costs as well as costs disaggregated into relevant categories. This categorization provides a clear depiction of which aspect of genetic testing/screening is responsible for incremental cost differences. Detecting sources of incremental variation is especially important for a field such as genetics in which materials and activities are rapidly changing costs.^40,54–57

Alignment with similar systematic reviews

Several recent systematic reviews of economic evaluations of genetic screening have been conducted for either specific populations or a more limited set of medical conditions. Although previous reviews primarily covered older studies, were limited to specific genetic conditions, and were not as comprehensive as our own regarding methodological assessment, these reviews have identified many of the same limitations in economic evaluations as those identified by our review. These include emphasizing the health care payer or health care system costing perspectives over societal perspectives,^58–61 dependence on macrocosting strategies and adopting costing estimates from other similar studies,¹⁷ and limited or opaque use of complex sensitivity analyses.^19,58 The performance of our articles as measured by the BMJ checklist is also consistent with 2 recent reviews of economic evaluations of genetic testing that employed the BMJ checklist.^19,21 Both these reviews found that most studies failed to provide a rigorous description of how costs were derived, provided no description for how disparate sources were synthesized to establish effectiveness estimates, failed to appropriately adjust price or currency or report such adjustment, and had limited description of the valuation methods by which utility weights were calculated or characterizations of the population from which they were derived.

Study limitations

There are several limitations to this systematic review. First, our assessment mechanism gave equal weight to all items, implying that all items were of equal ease to achieve and of equal importance to the methodological quality of an article when important inequalities likely exist across both dimensions. To account for this limitation, we have focused our discussion on those items that we believe to be of greater importance to overall quality and have provided recommendations to facilitate ease of achievement. Second, this review does not consider the influence methodological limitations may have on the primary or secondary outcomes of studies. For instance, an opaque presentation of parameter derivation may complicate a reader’s ability to interrogate the integrity of a model, although these parameters may ultimately be the most appropriate leaving results unbiased. On the contrary, the lack of a PSA may indirectly hide the fact that primary outcomes are widely variable and cannot be interpreted with high confidence. Future research should consider which methodological features of an article may have the largest influence on outcomes. Third, there is an abundance of methodological detail that went beyond the scope of this review, such as how well the structure of the model reflected the actual decision nodes within the health care system under study and whether a comprehensive selection of alternative strategies was considered for each model. This level of granularity is best suited for reviews with a much more limited scope than the one we conducted.

Conclusion

Economic evaluation of genetic medicine has been recently accelerating. Our review considered the methodological quality of such studies and showed that, with notable exceptions, many studies fell short across several key methodological criteria. Improvements in these arenas highlighted earlier would enhance the extent to which outcomes can be understood, translated, and faithfully replicated. Renewed attention to the methodological design of future economic evaluations of genetic testing/screening is warranted. Future economic evaluations in this space should adhere to established guidelines and may benefit from considering the specific recommendations and exemplar articles identified in this review.

Data Availability

All articles included in this review are accessible online, and the search terms used to query these articles can be found in Appendix 2.