-

Systematic Literature Review

Contents lists available at sciencedirect.com
Journal homepage: www.elsevier.com/locate/jval

The Net Benefit of Personalized Medicine: A Systematic Literature Review

and Regression Analysis
Heleen Vellekoop, MSc, Matthijs Versteegh, PhD, Simone Huygens, PhD, Isaac Corro Ramos, PhD, László Szilberhorn, PhD,
Tamás Zelei, PhD, Balázs Nagy, PhD, Apostolos Tsiachristas, PhD, Rositsa Koleva-Kolarova, PhD, Sarah Wordsworth, PhD,
Maureen Rutten-van Mölken, PhD, on behalf of the HEcoPerMed consortium

A B S T R A C T

Objectives: Amidst conflicting expectations about the benefits of personalized medicine (PM) and the potentially high
implementation costs, we reviewed the available evidence on the cost-effectiveness of PM relative to non-PM.
Methods: We conducted a systematic literature review of economic evaluations of PM and extracted data, including incre-
mental quality-adjusted life-years (DQALYs) and incremental costs (Dcosts). DQALYs and Dcosts were combined with
estimates of national cost-effectiveness thresholds to calculate incremental net monetary benefit (DNMB). Regression
analyses were performed with these variables as dependent variables and PM intervention characteristics as independent
variables. Random intercepts were used to cluster studies according to country.
Results: Of 4774 studies reviewed, 128 were selected, providing cost-effectiveness data for 279 PM interventions. Most studies
were set in the United States (48%) and the United Kingdom (16%) and adopted a healthcare perspective (82%). Cancer
treatments (60%) and pharmaceutical interventions (72%) occurred frequently. Prognostic tests (19%) and tests to identify
(non)responders (37%) were least and most common, respectively. Industry sponsorship occurred in 32%. Median DQALYs,
Dcosts, and DNMB per individual were 0.03, Int$575, and Int$18, respectively. We found large heterogeneity in cost-
effectiveness. Regression analysis showed that gene therapies were associated with higher DQALYs than other
interventions. PM interventions for neoplasms brought higher DNMB than PM interventions for other conditions.
Nonetheless, average DNMB in the ‘neoplasm’ group was found to be negative.
Conclusions: PM brings improvements in health but often at a high cost, resulting in 0 to negative DNMB on average. Pricing
policies may be needed to reduce the costs of interventions with negative DNMB.

Keywords: cost-effectiveness, net benefit, precision medicine, personalized medicine, QALY, test, threshold.

VALUE HEALTH. 2022; -(-):-–-

Introduction between health benefits and costs of individual PM interventions

coming onto the market, there is limited knowledge about the
Personalized medicine (PM), a term often used to describe average net benefit of PM.
innovative healthcare interventions that enable improved patient Previous reviews have found that cost savings are relatively
stratification (generally through genetic or genomic testing), may rare; most economic evaluations of PM show increased costs and
improve health outcomes (eg, by preventing adverse drug re- higher health benefits.3-5 They have also found large heteroge-
actions in “slow metabolizer” patients) and reduce healthcare neity in study methods and cost-effectiveness outcomes,
costs (eg, by preventing the prescription of treatments to patients sometimes even across different evaluations of the same inter-
who do not benefit from it). Therefore, PM has been subject to vention.6-8 The previous reviews focused on incremental cost-
high hopes and expectations. Nevertheless, there are concerns effectiveness ratios (ICERs) or binary “cost-effective yes/no”
about the potentially high costs of (implementing) PM. Among judgments to assess cost-effectiveness. In this study, we aim to
these are worries about the costs of larger-scale gene testing and build upon previous research by investigating the net monetary
concerns about the steep pricing of some of the PM interventions benefit (NMB) of PM interventions instead of their ICERs and by
that have come onto the market in recent years.1,2 Although many performing regression analyses in which we explore the hetero-
countries perform economic evaluations to assess the balance geneity in the cost-effectiveness of PM interventions.

1098-3015/Copyright ª 2022, International Society for Pharmacoeconomics and Outcomes Research, Inc. Published by Elsevier Inc. This is an open access article
under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
2 VALUE IN HEALTH - 2022

Methods Despite the importance of cost-effectiveness thresholds in cost-

effectiveness analysis, limited research has been conducted
Systematic Literature Review regarding the appropriate threshold value in each country. Indeed, in
many countries, the standard cost-effectiveness thresholds applied
We performed a systematic literature review aiming to identify during HTAs are based on little to no data. Part of the reason for the
all published economic evaluations of PM between 2009 and 2019. limited research into cost-effectiveness thresholds may be that there
PM was defined as “a medical model that bases therapeutic choice exists conceptual disagreement about what the threshold represents
on the result of gene profiling or aims to correct pathogenic gene and how it should be calculated.12,13 The 2 main views are that the
mutations,” based on a study by Hatz et al.7 Given the rapid pace threshold should reflect (1) society’s willingness-to-pay for increases
of innovation in PM and changes in its costs over time, studies in health (v) and (2) the opportunity cost of healthcare spending (k).
from before the 2009 cutoff were expected to be less insightful We opted to use k thresholds, because we were able to find national
about the current value of PM. estimates for all countries included in our data set (apart from
The review was conducted according to the Preferred Report- Taiwan), whereas v estimates were not available for all countries. In
ing Items for Systematic Reviews and Meta-Analyses guidelines.9 addition, v thresholds might not always be appropriate, especially in
On March 13, 2019, databases Embase, Google Scholar, Medline studies with a healthcare perspective, given that society’s willingness
Ovid, and Web of Science were searched. An additional search for to pay for health benefits may not align with available (healthcare)
gray literature was performed on May 16, 2019, and included the resources. We explicitly chose not to use the thresholds countries
Centre for Reviews and Dissemination and EconLit databases and have historically used in our base case analysis. This was partly
the reimbursement dossier sections on the websites of the Na- because several countries do not make use of a specific, single
tional Institute for Health and Care Excellence and the Institute for threshold and partly because there are likely inconsistencies in how
Clinical and Economic Review. different countries arrived at their thresholds.
A total of 3 groups of search terms were used, combined with Values for national k thresholds were mostly taken from a 2016
the Boolean operator AND: “economic evaluation”; “modelling”; study by Woods et al,14 which estimates k thresholds for 183
“personalized medicine” (see Appendix 1 in Supplemental Mate- countries. Whenever country-specific studies were available, the
rials found at https://doi.org/10.1016/j.jval.2022.01.006). Studies estimates from these studies were used. See Appendix 2 in Sup-
were included if they fell within our definition of PM, presented a plemental Materials found at https://doi.org/10.1016/j.jval.2022.
cost-effectiveness model, provided patient-level cost and quality- 01.006 for an overview of the national threshold values.
adjusted life-year (QALY) outcomes, provided the cost year (so
that cost outcomes could be inflated to 2020), extrapolated out-
comes beyond short-term clinical trial data, and described an Exploring Heterogeneity
existing (ie, nonhypothetical) intervention. Studies also had to Regression analysis was conducted to explore the heterogeneity
compare a PM intervention with a non-PM intervention, given in the reported cost-effectiveness of PM in the included studies,
that PM versus PM comparisons would not allow for the assess- aiming to identify characteristics of PM that may be associated with
ment of the added value of PM. higher (or lower) health benefits, costs, and NMB. Generalized
Various data items were collected from the included studies, linear mixed models were used, with random intercepts at country
using a Microsoft Excel–based data extraction form. Study details level to account for national differences in prescribed HTA methods
were recorded (first author, year of publication, country, cur- (such as discount rates for future health benefits and costs),
rency), as well as information about the health technology healthcare systems, and epidemiology. Restricted maximum likeli-
assessment (HTA) methods used (perspective, time horizon, dis- hood estimation was applied instead of maximum likelihood esti-
count rates, cost-effectiveness threshold). Details about the in- mation to avoid bias in the variance estimation. Analysis of variance
terventions under evaluation (description of the intervention, (ANOVA) was used to assess the benefit of a random intercept at
description of the comparator, disease class according to the In- country level. “Variance explained” was assessed using conditional
ternational Classification of Diseases, Tenth Revision [ICD-10]), R2 (following the definitions in Nakagawa and Schielzeth15) and
and cost-effectiveness outcomes (incremental costs [Dcosts], in- compared with the adjusted R2 of a simple linear model with the
cremental QALYs [DQALYs], ICERs) were also captured. If studies same specification but without the random intercept. A total of 3
evaluated multiple interventions or comparators, all 2-way com- separate models were specified, with DQALYs, Dcosts, or DNMB as
parisons between PM and non-PM interventions were recorded. the dependent variables, and all included the same independent
variables: “purpose of test,” “type of treatment,” “gene therapy,”
Assessing Cost-Effectiveness “industry sponsorship,” and “disease classification.”
The independent variable “purpose of test” was based on a
NMB was used as the measure of cost-effectiveness. Although previous literature review of economic evaluations of PM, which
the ICER measure is widely used to measure cost-effectiveness, identified broad categories that tests could be classed into and
NMB is better suited for ranking large numbers of interventions found possible differences in (median) cost-effectiveness between
(eg, because of issues around interpreting negative ICERs) and for the categories.7 The categories were testing to (1) screen for a
assessing the magnitude to which an intervention is more (or less) disease or a genetic marker in an asymptomatic population (eg,
cost-effective than another one.10 Therefore, the NMB measure genetic testing for LDL receptor mutations in relatives of patients
was deemed more appropriate for our study. with familial hypercholesterolemia), (2) gain information about
The incremental NMB (DNMB) of each intervention i was disease prognosis (eg, OncotypeDX), (3) identify likely (non)re-
calculated with the formula DNMBij ¼ Dhij  kj 2 Dcij , where sponders to treatment (eg, testing for NTRK gene fusions so that
Dhij = DQALYs for intervention i in country j, kj = cost-effectiveness TRK inhibitors can be provided to cancer patients who test posi-
threshold in country j, and Dcij = Dcosts for intervention i in tive), and (4) identify patients who may experience adverse drug
country j. Dcosts were inflated to 2020 prices using country- reactions (eg, CYP2D6 testing to optimize pharmacotherapy).
specific inflation rates and converted to purchasing power parity The variable “type of treatment” indicates for each interven-
using conversion factors from the World Bank Global Economic tion whether the treatment is pharmaceutical, non-
Monitor.11 pharmaceutical, or a combination of both (eg, gene-expression
 -- 3

profiling to help diagnose cancers of unclear origin, with subse- Table 1. Descriptive statistics of the included studies.
quent surgical and pharmaceutical treatment). The variable was
included based on debate around the affordability and cost-
Category Number of
effectiveness of expensive pharmaceuticals in PM. interventions
Literature and initial descriptive analysis showed that genetic (percent
therapies tend to be outliers, with sizably higher incremental of total)
health benefits and costs than other PM interventions. Therefore, Country
the dichotomous variable “gene therapy” was added to avoid ge- Canada 10 (4)
netic therapies skewing the results for the other variables. China 15 (5)
We included “industry sponsorship” as a dichotomous variable Germany 15 (5)
to investigate any differences in the reported cost-effectiveness The Netherlands 14 (5)
between industry-sponsored and nonindustry-sponsored studies UK 44 (16)
US 135 (48)
given that previous studies have found that (cost-)effectiveness
Other* 42 (15)
outcomes tend to be more favorable in industry-sponsored studies
than in studies by publicly funded and independent research Perspective
organisations.5,16,17 Healthcare 229 (82)
Societal 50 (18)
Finally, we included a dichotomous variable “disease classifi-
cation,” which could take on the value “neoplasm” and “non- Disease class
neoplasm”, depending on whether the intervention was used to Diseases of the circulatory system 54 (19)
Endocrine, nutritional, 11 (4)
treat neoplasms (ie, cancer) or other conditions. This variable ac-
or metabolic diseases
counts for the predictive value of belonging to the largest set of
Mental, behavioral, or 9 (3)
studies found in the literature review. The variable did not further neurodevelopmental disorders
specify the “other” category to avoid multicollinearity with the Neoplasms 167 (60)
other predictor variables. Other† 40 (14)
Purpose of test
Sensitivity Analysis Screening 58 (21)
In the base case, all studies received equal weight in the Info prognosis 54 (19)
Identify responders 103 (37)
regression analysis, only studies with a healthcare perspective
Identify adverse drug reactions 64 (23)
were included and DNMB was calculated using country-specific k
thresholds. Sensitivity analyses were performed where (1) studies Type of treatment
Pharmaceutical 201 (72)
were weighted according to their quality score in the Tufts Cost-
Nonpharmaceutical 70 (25)
Effectiveness Analysis Registry, (2) DNMB was calculated based
Combination 8 (3)
on the cost-effectiveness threshold stated by the authors of each
study, and (3) studies with a societal perspective were also Gene therapy
Gene therapy 11 (4)
included. We also repeated the base case analysis in a subsample
No gene therapy 268 (96)
including only interventions for neoplasms. Finally, because of
ambiguity in the definition of genetic therepies, an additional Industry sponsorship
analysis was performed in which the classification of a specific Industry sponsorship 90 (32)
No industry sponsorship 189 (68)
intervention (Spinraza to treat spinal muscular atrophy) was
UK indicates United Kingdom; US, United States.
changed from ‘genetic’ to ‘other’.
*Included in “Other” are Australia (3 interventions assessed), Austria (3), France
(5), Hong Kong (1), Italy (3), Japan (5), Malaysia (2), New Zealand (2), Puerto
Rico (1), Singapore (4), Slovenia (1), South Korea (2), Spain (3), Sweden (1),
Results Switzerland (2), Taiwan (3), and Thailand (5).
†
Included in “Other” are adverse drug reactions (7), diseases of the digestive
system (3), diseases of the immune system (6), diseases of the musculoskeletal
Study Sample system or connective tissue (6), diseases of the nervous system (5), diseases of
the respiratory system (2), and diseases of the visual system (2).
The systematic search rendered 4774 articles, whose abstracts
were screened. Full-text articles were read for 615 studies, of
which 128 met the inclusion criteria and were included for data
analysis (see Appendices 3-4 in Supplemental Materials found at evaluated interventions were in the “neoplasms” category (60%).
https://doi.org/10.1016/j.jval.2022.01.006 for the inclusion flow- Pharmaceutical treatments allocated based on markers in the tu-
chart and an overview of included studies). The included articles mor DNA were common in this category. Other frequently occur-
provided cost-effectiveness outcomes for 279 PM interventions. ring interventions were gene assays providing risk assessments
regarding the aggressiveness of tumors and screening in-
Characteristics of Studies, Interventions, and Methods terventions aiming to identify individuals at risk of developing
The distribution of interventions across countries, perspective, cancer. Common interventions in the “diseases of the circulatory
disease classes, types of test, types of treatment, and types of system” category (19%) were pharmacogenomic testing before
funding is presented in Table 1. A total of 23 countries were anticoagulation therapy and genetic screening to identify patients
included in the data set, although most interventions were eval- at risk of various heart conditions. Interventions in the “endocrine,
uated in the United States and the United Kingdom (48% and 16%, nutritional, or metabolic” diseases category (4%) focused on
respectively). All included countries are upper-middle- or high- screening for familial hypercholesterolemia and maturity onset
income economies according to the World Bank country classifi- diabetes of the young, whereas interventions in the “mental,
cation,18 meaning no economic evaluations of PM were identified behavioral, or neurodevelopmental disorders” category (3%)
for lower-middle- or low-income economies. The healthcare mostly involved pharmacogenomic testing before starting
perspective was the most common perspective (81%). Most antidepressants.
4 VALUE IN HEALTH - 2022

Notably, 21% of the evaluated interventions fell in the suggesting large QALY gains for gene therapies. This may be
“screening” category, 19% in “info prognosis,” 37% in “identify re- because most of the gene therapies included in the review focus
sponders,” and 23% in “identify adverse drug reactions”; 72% of on early onset conditions with high morbidity and mortality. The
the evaluated interventions were pharmaceuticals, 25% non- conditional R2 of the mixed model for QALYs was 0.47 compared
pharmaceutical, and 3% a combination of both. Non- with an adjusted R2 of 0.46 of a simple linear model without
pharmaceutical interventions consisted mostly of gene tests to random intercepts at country level. ANOVA therefore showed that
determine the appropriate screening interval (eg, increased the use of a random intercept did not improve goodness of fit.
screening frequency for patients at increased risk of hypertrophic For Dcosts of PM versus non-PM, “gene therapy” and “non-
cardiomyopathy or colorectal cancer) and gene tests to determine neoplasms” have a 95% CI that does not cross 0. The regression
whether surgery is necessary (eg, preventive surgery for patients coefficient of 1179 540 for “gene therapy” implies that, on average,
with BRCA mutations). Only 4% of evaluated interventions were the Dcost for gene therapies is 1179 540 higher than for PM in-
gene therapies, of which 6 were in “neoplasms” and 5 in “non- terventions that are not gene therapies. Similarly, the regression
neoplasms.” Moreover, 32% of interventions were evaluated in coefficient of 386 325 implies that for PM interventions in “non-
industry-sponsored studies. neoplasm” the Dcost is 386 325 higher than for interventions in
“neoplasm.” Conditional R2 of the mixed model for costs was 0.66
Estimated Cost-Effectiveness compared with an adjusted R2 of 0.33 of a simple linear model
without random intercepts at country level. Hence, ANOVA
The median amount of DQALYs of PM interventions relative to showed that using a random intercept improved goodness of fit of
their non-PM comparators was 0.03, whereas the mean was 0.26. the model.
As can be seen in Table 2, most DQALY values are just above 0, Finally, in the DNMB model of PM versus non-PM, the
with 0.00 DQALY and 0.16 DQALY at the 25th and 75th percentile, regression coefficients for “gene therapy” has a negative regres-
respectively. Nonetheless, several interventions had larger bene- sion coefficient, with the 95% CI again not crossing 0. The coeffi-
fits. Sixteen interventions (6%) rendered . 1 DQALY. cient suggests that, on average, gene therapies bring Int$868 759
Median Dcosts were Int$575, whereas mean Dcosts were less net benefit compared with non-PM interventions, despite
Int$99 777. Figure 1 shows that a small number of interventions offering higher QALY gains. This implies that the costs associated
have notably higher Dcosts than the rest. to gene therapies are higher than the monetary value of the QALY
Median DNMB across the included interventions was Int$18, gains, leading to a net loss. The coefficient for “non-neoplasm”
and mean DNMB was Int$277 072. DNMB centers around 0, with a also does not cross 0 and implies that PM interventions for con-
value of Int$22665 at the first quantile and Int$3538 at the third ditions other than neoplasms render Int$380 950 less DNMB than
quantile. As can be seen in Figure 1, extreme negative values are PM interventions in “neoplasm”. In line with the findings from the
more common than extreme positive values for DNMB. None- regression analysis, the average DNMB in the observed data was
theless, there are also some positive outliers, with a maximum Int$21 287417 (median Int$2343 379) for gene therapies,
DNMB of Int$406 277. Details of the interventions that are among whereas it was Int$227 394 (median Int$49) for the other in-
the top 5% in terms of DNMB are presented in Table 3.19-30 terventions. Average DNMB was Int$21161 (median Int$2426) for
In Figure 2, median DNMB per country (ie, based on all PM neoplasms and Int$2190 260 (median Int$164) for other in-
interventions evaluated in the country) is plotted against the na- terventions. Conditional R2 of the mixed model for NMB was 0.53
tional k threshold and the (median) author-reported threshold. compared with an adjusted R2 of 0.23 of a simple linear model
We see that median DNMB of PM is generally close to 0, regardless without random intercepts at country level. ANOVA showed the
of the national threshold. This implies that any QALY gains of PM model goodness of fit improved by using a random intercept.
interventions tend to be counterbalanced by their costs to the
healthcare system. Sensitivity Analysis
Results for the sensitivity analysis can be found in Appendix 5
Heterogeneity in Cost-Effectiveness
in Supplemental Materials found at https://doi.org/10.1016/j.jval.2
Our data set is inherently heterogeneous, because of the in- 022.01.006. The size and direction of regression coefficients tend
clusion of studies from many countries and many authors, each to be similar between the analyses using base case assumptions
with slight methodological differences. This variation is reflected and analyses using the alternative assumptions. In all analyses,
in wide confidence intervals (CIs) (Table 4). Nevertheless, the gene therapies are associated with significantly higher QALY gains,
mean values predicted by the models are close to the observed and PM interventions for neoplasms are associated with lower
values. costs and higher DNMB. Within the neoplasm subsample (n =
In Table 4, the regression results for the model with DQALYs of 167), conclusions are similar to those reported in the main anal-
PM versus non-PM as the dependent variable are presented first. ysis, although the 95% CI for “gene therapy” now crosses 0.
The regression coefficient for the gene therapy variable is the only Reclassification of genetic therapy Spinraza to nongene therapy
coefficient for which 0 is not included in the 95% CI. The coeffi- increased the coefficient of “gene therapy” for QALYs to 4.5. After
cient of 3.22 is much larger than any of the other coefficients, the reclassification, the CI for “gene therapy” crossed 0 in the cost

Table 2. Quantiles DQALYs, Dcosts, and DNMB.

Variables Min 5% 25% 50% 75% 95% Max Mean

DQALYs 20.76 20.10 0.00 0.03 0.16 1.08 11.8 0.26
Dcosts (2020 Int$) 234 062 27233 2338 575 3233 282 080 8 095 744 99 777
DNMB (2020 Int$) 27 997 236 291 832 22665 18 3538 21 615 406 277 277 072
Dcost indicates incremental cost; DNMB, incremental net monetary benefit; DQALY, incremental quality-adjusted life-year; Max, maximum; Min, minimum.
 -- 5

Figure 1. Boxplots DQALY, Dcost, DNMB.

1.0
0.8
0.6

ΔQALY
0.4
0.2
0.0

250 000
200 000
Δcosts (Int$)

150 000
100 000
50 000
0

25 000

0
ΔNMB (Int$)

–25 000

–50 000

–75 000

For each boxplot, the bottom and top 5% of the distribution were excluded from the figure. DQALY indicates incremental quality-adjusted life-year; Dcost, incremental
cost; DNMB, incremental NMB.

model. The coefficient for “gene therapy” in the NMB model was Gene therapies were found to offer high QALY gains in all an-
Int$777 987. The average DNMB for gene therapies remained alyses. Gene therapies were also found to be associated with high
negative after the reclassification and was Int$2356 016. costs and had an average DNMB of Int$21 287417 (or
Int$2356 016 after reclassifying Spinraza to nongene therapy). It
has been argued that the QALY insufficiently captures the value of
Discussion gene therapies and that additional value elements, such as the
value of a cure (ie, nonhealth-related welfare benefits, such as
Interpretation Results being able to do more future-planning), should also be considered.
Although there may indeed be benefits beyond QALYs to being
The median and mean DQALYs found in this study (0.03 and completely cured of a condition, our findings suggest that, on
0.26, respectively) are comparable with the QALY gains found by a average, the monetary value of these additional benefits would
literature review of cost-utility analyses for all types of healthcare, have to be Int$21 287417 per patient for the DNMB to no longer
which identified a median QALY increase of 0.06 (mean 0.31).31 be negative, which arguably is implausibly high, especially in the
We found that 6% of interventions had a QALY increase of . 1, light of a recent study by Reed et al32 in which much lower figures
while this was 8% in the previous study. This suggests that the are found for the “value of hope”. The genetic therapies included
health benefits of PM tend to be similar to (or possibly slightly in our analysis were treatments for spinal muscle atrophy (Spin-
lower than) the health benefits of other (new) healthcare raza, Zolgensma) and loss of vision because of inherited retinal
interventions. dystrophy (Luxturna), and CAR-T cell therapies. Spinraza, an
The median DNMB of PM interventions was close to 0 (Int$18), antisense oligonucleotide, is referred to as a “genetic therapy,” yet
implying that the health benefits rendered by PM interventions oligonucleotides are not classified as a “gene therapy” by the Food
are counterbalanced by the increased costs associated with the and Drug Administration or as advanced therapy medicinal
interventions. It could be that PM interventions are associated products by European Medicines Agency.33 After reclassifying
with higher costs than non-PM interventions because of the Spinraza to a nongene therapy, the outcomes for the DQALY and
additional testing needed for the improved stratification in PM. Dcost models were largely the same. However, the coefficient for
Nevertheless, the costs of testing were often not—or only “gene therapy” in the DNMB model changed from a negative to a
partially—considered in the studies included in our review. positive value. Given that the “gene therapy” coefficient is
6 VALUE IN HEALTH - 2022

Table 3. Details for interventions among top 5% of DNMB.

DNMB Intervention Comparator Country

406 277 CAR-T-cell therapy (tisagenlecleucel) for pediatric Clofarabine US

patients with relapsed or refractory B-cell acute
lymphoblastic leukemia
265 848 The addition of gene-expression profiling to Therapy choice based on usual care (including US
usual care to identify the tissue of origin and IHC, blood tests, physical examination)
guide therapy choice in
adults with metastatic cancer of uncertain origin
208 766 CAR-T-cell therapy (tisagenlecleucel) for pediatric Blinatumomab; clofarabine, cyclophosphamide, US
patients with relapsed or refractory B-cell acute and etoposide combination therapy; and
lymphoblastic leukemia clofarabine monotherapy
95 175 Gene profiling (mutational load-based) in Periodic endoscopic surveillance, ablative US
patients with nondysplastic Barrett’s esophagus. therapy when high grade dysplasia or
Only patients with high esophageal adenocarcinoma are detected
risk score receive ablative therapy
78 312 Gene-expression profiling (MammaprintTM) in All patients receive adjuvant chemotherapy NL
patients with node-negative, estrogen receptor-
positive breast cancer.
Only patients with high risk score receive
adjuvant chemotherapy
62 705 Gene profiling (mutational load-based) in Periodic endoscopic surveillance, ablative US
patients with nondysplastic Barrett’s esophagus. therapy when high grade dysplasia or
All patients who test positive esophageal adenocarcinoma is found
for mutational load (high- or low-risk) receive
ablative therapy
48 518 Gene profiling in children with neonatal Insulin treatment for all patients US
diabetes. Patients switch from insulin to
sulfonylurea when mutations in the KCNJ11 or
ABCC8 genes are found
41 885 Gene profiling to identify long-QT syndrome in b-blockers for all first-degree relatives US
relatives of index patients.
Relatives who test positive are treated with
b-blockers
37 277 Gene profiling (CPGxTM) in treatment-resistant Therapy choice based on treatment history, US
patients with major depressive disorder to guide physical examination, lab tests that are part of
therapy choice usual care
35 604 Gene profiling (assessing IL-28B genotype) of Pegylated interferon and ribavirin (standard US
patients with chronic hepatitis C to therapy) for all
identify patients who would benefit from the
addition of
protease inhibitors to standard therapy
35 424 Gene profiling (CYP2C19) in patients with acute All patients receive prasugrel US
coronary syndrome. Patients with
CYP2C19*2 received prasugrel, all other patients
clopidogrel
26 891 Gene profiling (CYP2C19) in patients with acute All patients receive clopidogrel US
coronary syndrome. Patients with
CYP2C19*2 received prasugrel, all other patients
clopidogrel
24 294 Gene profiling (loss of heterozygosity-based) in All patients return for follow-ups every 6 months. CA
patients with low-grade oral dysplasia. Patients Patients are referred for surgery when cancer is
with a low-risk found
score return for follow-up every 5 years,
patients with an intermediate risk score every 2
years. Patients with a high risk score are referred
for surgery
24 052 Gene profiling (IDGx) in patients with major Therapy choice based on usual care, in which US
depressive disorder who are treatment-naïve or nonresponders to treatment iteratively try
whose depression is alternative options
inadequately controlled to guide therapy choice

Dcost indicates incremental cost; DNMB, incremental net monetary benefit; DQALY, incremental quality-adjusted life-year; ADR, adverse drug reaction; CA, Canada; CAR,
chimeric antigen receptor; IHC, immunohistochemistry; IL-28B, interleukin-28B; NL, The Netherlands; Pharm, Pharmaceutical; Ref, reference; US, United States.
 -- 7

Table 3. Continued

Perspective Disease class Purpose of Type of Gene Industry DQALY DCosts Ref
test treatment therapy sponsor
Healthcare Neoplasms Identify (non) Pharm Yes No 7.18 346 163 19
responders

Healthcare Neoplasms Identify (non) Pharm No Yes 2.65 11 863 20

responders

Healthcare Neoplasms Identify (non) Pharm Yes Yes 5.17 333 033 21
responders

Healthcare Diseases of Info Nonpharm No Yes 0.86 25260 22

the digestive prognosis
system

Healthcare Neoplasms Info Pharm No No 1.20 212 170 23

prognosis

Healthcare Diseases Info Nonpharm No Yes 0.54 26115 22

of the digestive prognosis
system

Societal Endocrine, Identify (non) Pharm No No 0.32 214 983 24

nutritional, or responders
metabolic
diseases
Societal Diseases of Screen Pharm No No 0.41 1081 25
the circulatory asymptomatic
system

Societal Mental, Identify (non) Pharm No Yes 0.32 24161 26

behavioral, or responders
neurodevelopmental
disorders
Societal Certain infectious Identify (non) Pharm No No 0.54 20 986 27
or parasitic diseases responders

Healthcare Diseases of the Identify ADR Pharm No No 0.01 234 062 28

circulatory system

Healthcare Diseases of the Identify ADR Pharm No No 0.12 214 629 28

circulatory system

Healthcare Diseases of the Info prognosis Non-pharm No No 0.64 27467 29

digestive system

Societal Mental, behavioral, or Identify (non) Pharm No Yes 0.17 26237 30

neurodevelopmental responders
disorders
8 VALUE IN HEALTH - 2022

Figure 2. Median DNMB plotted against cost-effectiveness threshold, per country.

DNMB indicates incremental net monetary benefit; CN, China; DE, Germany; JP, Japan; NL, New Zealand; SG, Singapore; TH, Thailand; UK, United Kingdom; US, United
States.

sensitive to single data points, it appears that data are too scarce The regression coefficient for pharmaceutical interventions
to draw definitive conclusions on the NMB of gene therapies. was positive in the DQALY and costs models and negative in the
PM interventions in neoplasms were shown to have lower DNMB model. This could mean that although PM pharmaceuticals
Dcosts and higher DNMB. Indeed, average DNMB was higher in the have higher health gains than nonpharmaceuticals, PM pharma-
“neoplasm” group than in the “non-neoplasm” group (Int$21161 ceuticals come at a higher cost than nonpharmaceuticals, causing
vs. Int$2190,260). However, median DNMB is more similar across lower net value (DNMB). The higher cost for PM pharmaceuticals
the groups, with median DNMB even being a little lower for compared with PM nonpharmaceuticals could be because of high
neoplasms (Int$-426 for “neoplasm” and Int$164 for “non- prices charged by pharmaceutical companies. Nevertheless, the
neoplasm”). This suggests large heterogeneity in the “non- higher cost for pharmaceuticals could also be caused by other
neoplasm” group and precludes a simple explanation of why in- factors, such as the nature of the treated diseases.
terventions for neoplasms might render more DNMB. The number The positive coefficient for “industry sponsorship” in the
of interventions per “non-neoplasm” condition are too low to DNMB could mean that industry-sponsored studies are more
perform condition-specific regression analysis. Different research likely to have positive cost-effectiveness outcomes, which is in
methods may be needed to further explore the heterogeneity in line with previous studies.5,16,17 This might be because industry-
the benefit that interventions for “other” conditions bring. The sponsored studies are often for interventions about to be
addition of a random intercept at country level improved good- evaluated for reimbursement and focused on a single promising
ness of fit for the Dcost and DNMB models but not for the DQALY intervention, whereas nonindustry-sponsored studies may take a
model, suggesting that the estimation of QALYs is less affected by wider approach and include several interventions in the evalua-
differences between countries than the estimation of costs and tion, some of which perhaps less promising. Additionally, authors
NMB. of industry-sponsored reimbursement studies may have an
The CIs around the regression coefficients other than “gene incentive to limit the included costs (eg, include only testing costs
therapy” and “neoplasm” are wide and cross 0 for all above- for patients who tested positive as opposed to testing costs for the
mentioned categories. Therefore, no definitive conclusions entire population that received testing) or otherwise make model
regarding the association between the different categories and assumptions to improve cost-effectiveness outcomes. Finally, the
dependent variable DNMB can be drawn. Nonetheless, we offer issue of publication bias, whereby only studies with positive re-
possible explanations for our findings below. sults are published, may be more prominent for industry- than for
First, a potential explanation for the positive coefficient for nonindustry-sponsored research.
“identify ADR” could be that many of the interventions included in
this category aim to better stratify patients to existing treatments
Strengths and Limitations
instead of to new treatments (eg, by offering reduced warfarin
dosing to patients with gene mutations that are associated with Previous studies investigating the cost-effectiveness of PM
increased sensitivity to warfarin). Conversely, many interventions focused on ICERs and descriptive analyses.3-8 Our study expands
in the “identify responders” category stratify toward new treat- on these studies, by focusing on NMB as opposed to ICERs, given
ments. New treatments are generally still patented and may be that NMB has been argued to be more appropriate for comparing
costly, especially when the target population is small, which has a large numbers of interventions. Our study also adds to the liter-
negative effect on the interventions’ DNMB. Nonetheless, in the ature by presenting regression analyses, for each of the variables
future, the benefit from new innovations may increase. Decisions DQALYs, Dcosts, and DNMB separately. Our study builds on the
on whether to invest in products that currently bring more DNMB work of Hatz et al7 in particular, by incorporating the types of test
to the healthcare system or in products that may bring more they identified into our analysis.
DNMB in the future depend on value judgments on the extent to Our definition of PM focuses on genetic and genomic test-
which current versus future (uncertain) QALYs should be treatment combinations. We acknowledge that alternative in-
prioritized. terpretations of “personalized medicine” exist. Some understand
 -- 9

Table 4. Regression results (N = 229).

Variable Category Regression coefficient 95% confidence interval t-value

Dependent variable: DQALYs
Intercept 0.02 [-0.27 to 0.32] 0.16
Purpose of test* Info prognosis 0.10 [20.27 to 0.47] 0.55
Identify responders 0.22 [20.13 to 0.56] 1.23
Identify ADR 20.25 [20.63 to 0.14] 21.25
Type of treatment† Pharmaceutical 0.02 [20.27 to 0.31] 0.14
Combination 0.12 [20.58 to 0.82] 0.33
Gene therapy Gene therapy 3.22 [2.69 to 3.75] 12.0
Sponsorship Industry 20.15 [20.38 to 0.08] 21.31
Disease classification‡ Non-neoplasm 0.25 [20.05 to 0.55] 1.64
Dependent variable: Dcosts
Intercept 2163 055 [2474 256 to 148 146] 21.04
Purpose of test* Info prognosis 137 639 [2190 300 to 465 578] 0.83
Identify responders 239 144 [285 158 to 563 446] 1.46
Identify ADR 2183 129 [2524 414 to 158 156] 21.06
Type of treatment† Pharmaceutical 7226 [2254 120 to 268 571] 0.06
Combination 296 374 [2687 344 to 494 595] 20.32
Gene therapy Gene therapy 1 179 540 [732 527 to 1 626 554] 5.25
Sponsorship Industry 292 400 [2292 338 to 107 539] 20.92
Disease classification‡ Non-neoplasm 386 325 [122 244 to 650 407] 2.91
Dependent variable: DNMB
Intercept 152 210 [2144 118 to 448 539] 1.02
Purpose of test* Info prognosis 2126 431 [2445 368 to 192 505] 20.78
Identify responders 2221 146 [2535 623 to 93 331] 21.39
Identify ADR 176 913 [2156 155 to 509 981] 1.06
†
Type of treatment Pharmaceutical 3479 [2251 023 to 257 981] 0.03
Combination 99 635 [2475 897 to 675 166] 0.34
Gene therapy Gene therapy 2868 759 [21 307 289 to 2430 229] 23.94
Sponsorship Industry 92 109 [2103 308 to 287 527] 0.94
Disease classification‡ Non-neoplasm 2380 950 [2638 867 to 2123 032] 22.94

For values in bold, the 95% confidence interval does not cross 0.
Dcost indicates incremental cost; DNMB, incremental net monetary benefit; DQALY, incremental quality-adjusted life-year; ADR, adverse drug reaction.
*Reference category is “screening.”
†
Reference category is “nonpharmaceutical interventions.”
‡
Reference category is “neoplasms.”

the “personalized” aspect as an increased focus on patient pref- positive,” some expressed the health benefit “per patient tested.”
erences, and some include decision making based on patients’ The health benefit “per patient tested” may be different from the
phenotypes in their definition of PM.34,35 Our decision to focus on health benefit “per patient who tested positive.” For example,
patients’ genotypes was based on a study by Schleidgen et al,34 in when genetic testing has to be applied to a large number of pa-
which a systematic literature review was conducted to understand tients to detect a few patients with a rare mutation that de-
how the term “personalized medicine” is used in scientific prac- termines eligibility for a particular treatment, the average health
tice. The study finds that most scientific articles on PM focus on benefit per patient tested (and treated) may be diluted compared
the use of gene information in medical decision making. Indeed, to the health benefit per patient who tested positive. Nonetheless,
Schleidgen et al34 argue that decision making based on phenotype if outcomes across patients with different test results are expected
(eg, weight, sex) and patient preferences is part of traditional to be similar, health benefit “per patient tested” and “per patient
healthcare. Including these in the definition of PM would blur the who tested positive” may be similar. For example, patients with
distinction between PM and traditional healthcare. Nonetheless, certain variants in the VKORC1 and CYP2C9 genes require lower
we acknowledge that this study provides an incomplete overview warfarin dosing. Once warfarin dosages have been adjusted
of the net benefit of PM to those who hold a wider definition of appropriately, health outcomes for patients who test positive for
PM. these gene variants may be similar to health outcomes for patients
Although many authors expressed the incremental health who test negative for the gene variants and receive normal dosing.
benefit of a test-treatment combination “per patient who tested Studies of cascade screening often add the health benefit of family
10 VALUE IN HEALTH - 2022

members that are identified after an initial index patient tests doi: https://doi.org/10.1016/j.jval.2022.01.006
positive to the health benefit of the index patient, resulting in a Author Affiliations: Institute for Medical Technology Assessment, Eras-
single QALY value “per index patient.”36 A single value combining mus University Rotterdam, Rotterdam, The Netherlands (Vellekoop, Ver-
the health gains of multiple patients is likely higher than the steegh, Huygens, Corro Ramos, Rutten-van Mölken); Syreon Research
benefit “per patient who tested positive.” Therefore, there may be Institute, Budapest, Hungary (Szilberhorn, Zelei, Nagy); Health Economics
some inconsistency in what is captured in the DQALY values that Research Centre, University of Oxford, Oxford, UK (Tsiachristas, Koleva-
Kolarova, Wordsworth); Erasmus School of Health Policy & Management,
we extracted from the selected studies.
Erasmus University Rotterdam, Rotterdam, The Netherlands (Rutten-van
Indeed, our sample has a high level of heterogeneity as a result Mölken).
of the wide scope of our literature search and inconsistency in
methods across studies. Therefore, the results of our regression Correspondence: Heleen Vellekoop, Institute for Medical Technology
Assessment, P.O. Box 1738, 3000 DR Rotterdam, The Netherlands. Email:
analysis are uncertain. Although we attempted to account for
vellekoop@imta.eur.nl
country-level differences by clustering studies according to
country, more streamlining of methods across studies may be Author Contributions: Concept and design: Vellekoop, Versteegh, Huy-
needed to reduce uncertainty around the cost-effectiveness of PM. gens, Rutten-van Mölken
Acquisition of data: Vellekoop, Huygens, Rutten-van Mölken
Finally, the countries included in our review were mostly high-
Analysis and interpretation of data: Vellekoop, Versteegh, Huygens, Ramos,
income countries (and a few higher to middle income). Therefore, Szilberhorn, Zelei, Nagy, Tsiachristas, Koleva-Kolarova, Wordsworth, Rut-
our results do not necessarily apply to low- and (lower-) middle- ten-van Mölken
income countries. Although this was caused by the limited eco- Drafting of the manuscript: Vellekoop, Versteegh, Huygens, Rutten-van
nomic evaluations of PM currently available for lower-income Mölken
countries, we acknowledge the importance of more evidence Critical revision of the paper for important intellectual content: Vellekoop,
Versteegh, Huygens, Ramos, Szilberhorn, Zelei, Nagy, Tsiachristas, Koleva-
generation for lower-income countries, given that insights into
Kolarova, Wordsworth, Rutten-van Mölken
which interventions offer the highest added value are of critical Statistical analysis: Vellekoop, Versteegh, Ramos
importance in settings with highly constrained healthcare Obtaining funding: Versteegh, Nagy, Tsiachristas, Wordsworth, Rutten-van
budgets. Mölken
Administrative, technical, or logistic support: Vellekoop, Versteegh, Huygens,
Implications Ramos, Szilberhorn, Zelei, Koleva-Kolarova
Supervision: Versteegh, Huygens, Rutten-van Mölken
Our study results show modest health benefits for PM versus
Conflict of Interest Disclosures: All authors reported receiving grants
non-PM interventions and a median DNMB of close to 0. The
from European Union Horizon 2020 research and innovation program
available evidence therefore seems to contrast the high anticipa- grant number 824997 during the conduct of the study. No other disclo-
tions that have surrounded PM over recent years. Nonetheless, sures were reported.
there are several PM interventions with (very) positive DNMB. It
Funding/Support: The HEcoPerMed project has received funding from the
appears that the term “personalized medicine” may be too general
European Union’s Horizon 2020 research and innovation program under
because it conceals sizable differences in the net benefit of grant agreement number 824997.
different PM interventions. A more precise division into sub-
categories of PM may be needed to uncover the most promising Role of the Funder/Sponsor: The funder had no role in the design and
conduct of the study; collection, management, analysis, and interpretation
areas for further investment.
of the data; preparation, review, or approval of the manuscript; and de-
Despite the general tendency of PM interventions toward a cision to submit the manuscript for publication.
DNMB of 0, we identified various interventions with (very)
negative DNMB, among which several gene therapies. National
and international pricing policies may be needed to reduce the
costs of these interventions and ensure that societies are not faced
REFERENCES
with negative added value when implementing them. 1. O’Sullivan BP, Orenstein DM, Milla CE. Pricing for orphan drugs: will the
market bear what society cannot? JAMA. 2013;310(13):1343–1344.
2. Experts in Chronic Myeloid Leukemia. The price of drugs for chronic myeloid
Conclusions leukemia (CML) is a reflection of the unsustainable prices of cancer drugs:
from the perspective of a large group of CML experts. Blood.
2013;121(22):4439–4442.
This study has provided evidence that PM leads to additional 3. D’Andrea E, Marzuillo C, Pelone F, De Vito C, Villari P. Genetic testing and
health gains compared with non-PM but its costs tend to result in economic evaluations: a systematic review of the literature. Epidemiol Prev.
0 to negative DNMB. Gene therapies offer high QALY gains and 2015;39(4 suppl 1):45–50.
4. Plumpton CO, Roberts D, Pirmohamed M, Hughes DA. A systematic review of
render negative net monetary benefit on average, though data economic evaluations of pharmacogenetic testing for prevention of adverse
scarcity prohibits drawing firm conclusions on their added value. drug reactions. Pharmacoeconomics. 2016;34(8):771–793.
For PM interventions with negative DNMB, the benefit to society 5. Berm EJJ, Looff Mde, Wilffert B, et al. Economic evaluations of pharmacoge-
netic and pharmacogenomic screening tests: a systematic review. Second
may be increased if ways can be found to reduce costs. update of the literature. PLoS One. 2016;11(1):e0146262.
6. Faruque F, Noh H, Hussain A, Neuberger E, Onukwugha E. Economic value of
pharmacogenetic testing for cancer drugs with clinically relevant drug-gene
Supplemental Materials associations: a systematic literature review. J Manag Care Spec Pharm.
2019;25(2):260–271.
7. Hatz MHM, Schremser K, Rogowski WH. Is individualized medicine more
Supplementary data associated with this article can be found in the cost-effective? A systematic review. Pharmacoeconomics. 2014;32(5):
online version at https://doi.org/10.1016/j.jval.2022.01.006. 443–455.
8. Beaulieu M, de Denus S, Lachaine J. Systematic review of pharmacoeconomic
studies of pharmacogenomic tests. Pharmacogenomics. 2010;11(11):
1573–1590.
Article and Author Information 9. Liberati A, Altman DG, Tetzlaff J, et al. The PRISMA statement for reporting
systematic reviews and meta-analyses of studies that evaluate healthcare
Accepted for Publication: January 9, 2022 interventions: explanation and elaboration. BMJ. 2009;339:b2700.
10. Paulden M. Calculating and interpreting ICERs and net benefit. Pharmacoe-
Published Online: xxxx conomics. 2020;38(8):785–807.
 -- 11

11. Global economic monitor. The World Bank. https://datacatalog.worldbank. 24. Greeley SAW, John PM, Winn AN, et al. The cost-effectiveness of personalized
org/dataset/global-economic-monitor. Accessed December 3, 2020. genetic medicine: the case of genetic testing in neonatal diabetes. Diabetes
12. Vallejo-Torres L, García-Lorenzo B, Castilla I, et al. On the estimation of the Care. 2011;34(3):622–627.
cost-effectiveness threshold: why, what, how? Value Health. 2016;19(5): 25. Perez MV, Kumarasamy NA, Owens DK, Wang PJ, Hlatky MA.
558–566. Cost-effectiveness of genetic testing in family members of patients
13. Cameron D, Ubels J, Norström F. On what basis are medical cost-effectiveness with long-qt syndrome. Circ Cardiovasc Qual Outcomes. 2011;4(1):76–84.
thresholds set? Clashing opinions and an absence of data: a systematic re- 26. Hornberger J, Li Q, Quinn B. Cost-effectiveness of combinatorial pharmaco-
view. Glob Health Action. 2018;11(1):1447828. genomic testing for treatment-resistant major depressive disorder patients.
14. Woods B, Revill P, Sculpher M, Claxton K. Country-level cost-effectiveness Am J Manag Care. 2015;21(6):e357–e365.
thresholds: initial estimates and the need for further research. Value 27. Liu S, Cipriano LE, Holodniy M, Owens DK, Goldhaber-Fiebert JD. New pro-
Health. 2016;19(8):929–935. tease inhibitors for the treatment of chronic hepatitis C: a cost-effectiveness
15. Nakagawa S, Schielzeth H. A general and simple method for obtaining R2 analysis. Ann Intern Med. 2012;156(4):279–290.
from generalized linear mixed-effects models. Methods Ecol Evol. 28. Lala A, Berger JS, Sharma G, Hochman JS, Scott Braithwaite R, Ladapo JA.
2013;4(2):133–142. Genetic testing in patients with acute coronary syndrome undergoing
16. Lundh A, Lexchin J, Mintzes B, Schroll JB, Bero L. Industry sponsorship and percutaneous coronary intervention: a cost-effectiveness analysis. J Thromb
research outcome: systematic review with meta-analysis. Intensive Care Med. Haemost. 2013;11(1):81–91.
2018;44(10):1603–1612. 29. Cromwell I, Regier DA, Peacock SJ, Poh CF. Cost-effectiveness analysis of using
17. Kim DD, Silver MC, Kunst N, Cohen JT, Ollendorf DA, Neumann PJ. Perspective loss of heterozygosity to manage premalignant oral dysplasia in British
and costing in cost-effectiveness analysis, 1974-2018. Pharmacoeconomics. Columbia, Canada. Oncologist. 2016;21(9):1099–1106.
2020;38(10):1135–1145. 30. Groessl EJ, Tally SR, Hillery N, Maciel A, Garces JA. Cost-effectiveness of a
18. The world by income. The World Bank. https://datahelpdesk.worldbank.org/ pharmacogenetic test to guide treatment for major depressive disorder.
knowledgebase/articles/906519-world-bank-country-and-lending-groups. J Manag Care Spec Pharm. 2018;24(8):726–734.
Accessed August 12, 2021. 31. Wisløff T, Hagen G, Hamidi V, Movik E, Klemp M, Olsen JA. Estimating QALY
19. Whittington MD, McQueen RB, Ollendorf DA, et al. Long-term survival and gains in applied studies: a review of cost-utility analyses published in 2010.
value of chimeric antigen receptor T-cell therapy for pediatric patients with Pharmacoeconomics. 2014;32(4):367–375.
relapsed or refractory leukemia. JAMA Pediatr. 2018;172(12):1161–1168. 32. Reed SD, Yang JC, Gonzalez JM, Johnson FR. Quantifying value of hope. Value
20. Hornberger J, Degtiar I, Gutierrez H, et al. Cost-effectiveness of gene- Health. 2021;24(10):1511–1519.
expression profiling for tumor-site origin. Value Health. 2013;16(1):46–56. 33. Chilcott EM, Muiruri EW, Hirst TC, Yáñez-Muñoz RJ. Systematic review and
21. Lin JK, Lerman BJ, Barnes JI, et al. Cost effectiveness of chimeric antigen re- meta-analysis determining the benefits of in vivo genetic therapy in spinal
ceptor T-cell therapy in relapsed or refractory pediatric B-cell acute muscular atrophy rodent models [published online October 6, 2021]. Gene
lymphoblastic leukemia. J Clin Oncol. 2018;36(32):3192–3202. Ther. https://doi.org/10.1038/s41434-021-00292-4.
22. Das A, Callenberg KM, Styn MA, Jackson SA. Endoscopic ablation is a cost- 34. Schleidgen S, Klingler C, Bertram T, Rogowski WH, Marckmann G. What is
effective cancer preventative therapy in patients with Barrett’s esophagus personalized medicine: sharpening a vague term based on a systematic
who have elevated genomic instability. Endosc Int Open. 2016;4(5):E549– literature review. BMC Med Ethics. 2013;14(1):55.
E559. 35. Pokorska-Bocci A, Stewart A, Sagoo GS, Hall A, Kroese M, Burton H.
23. Retel VP, Joore MA, Knauer M, Linn SC, Hauptmann M, van Harten WH. Cost- “Personalized medicine”: what’s in a name? Per Med. 2014;11(2):197–210.
effectiveness of the 70-gene signature versus St. Gallen guidelines and 36. Vellekoop H, Huygens S, Versteegh M, et al. Guidance for the harmonisation
Adjuvant Online for early breast cancer. Eur J Cancer. 2010;46(8): and improvement of economic evaluations of personalised medicine. Phar-
1382–1391. macoeconomics. 2021;39(7):771–788.