A Systematic Review of Cost-Effectiveness Studies of Interventions with a Personalized Nutrition Component in Adults

Objectives: Important links between dietary patterns and diseases have been widely applied to establish nutrition in- terventions. However, knowledge about between-person heterogeneity regarding the bene ﬁ ts of nutrition intervention can be used to personalize the intervention and thereby improve health outcomes and ef ﬁ ciency. We performed a systematic review of cost-effectiveness analyses (CEAs) of interventions with a personalized nutrition (PN) component to assess their methodology and ﬁ ndings. Methods: A systematic search (March 2019) was performed in 5 databases: EMBASE, Medline Ovid, Web of Science, Cochrane CENTRAL, and Google Scholar. CEAs involving interventions in adults with a PN component were included; CEAs focusing on clinical nutrition or undernutrition were excluded. The CHEERS checklist was used to assess the quality of CEAs. Results: We identi ﬁ ed 49 eligible studies among 1792 unique records. Substantial variation in methodology was found. Most studies (91%) focused only on psychological concepts of PN such as behavior and preferences. Thirty-four CEAs were trial-based, 13 were modeling studies, and 4 studies were both trial- and model-based. Thirty-two studies used quality- adjusted life-year as an outcome measure. Different time horizons, comparators, and modeling assumptions were applied, leading to differences in costs/quality-adjusted life-years. Twenty-eight CEAs (49%) concluded that the intervention was cost-effective, and 75% of the incremental cost-utility ratios were cost-effective given a willingness-to-pay threshold of $50 000 per quality-adjusted life-year. Conclusions: Interventions with PN components are often evaluated using various types of models. However, most PN in- terventions have been considered cost-effective. More studies should examine the cost-effectiveness of PN interventions that combine psychological and biological concepts of personalization. systematic review. VALUE HEALTH. 2021; - ( - ): - – Introduction


Introduction
There are well-established links between poor dietary patterns, representing a complex set of highly correlated dietary exposures 1 and an increased risk of different diseases. 2,3 Obesity may be an intermediate outcome of these links, 4 since obesity often leads to diet-related diseases such as type 2 diabetes, heart disease, stroke, and cancer. 2 In other cases, poor dietary patterns can arise from other problems (eg, hip fracture), which may lead to malnutrition and possibly result in disorders such as functional disability and impaired cognitive function. 5 In this regard, dietbased prevention of obesity and malnutrition can help to reduce the frequency of various diseases, improve health outcomes, and reduce economic burden. 6 This knowledge has led to the development of many nutrition interventions based on population averages. However, although these nutrition interventions might have an acceptable average overall effectiveness (ie, population level), they often have poor individual-level effectiveness. 3,7 Studies have shown this might be caused by inter-individual variability of metabolic responses to specific diets and food components that affect health. 8,9 Knowledge about an individual's response could lead to a personalized intervention to maximize the potential health benefits of these diets and food components. 9 Various personalized nutrition (PN) interventions, which can be defined as an approach that uses information on individual characteristics to develop targeted nutritional advice, products, or services, 2 have been developed and assessed. For example, the Food4Me study found that internet-delivered personalized advice produced larger and more appropriate changes in dietary behavior than a conventional (one-size-fits-all) approach. 10 However, policy decisions must be guided by their ability to improve health outcomes and their cost-effectiveness, 11 given the ever-present tension between effectiveness and financial constraints. 12 In fact, various cost-effectiveness analyses (CEAs) of nutrition interventions have been published, and systematic reviews of these CEAs have been conducted. 11,13,14 However, these reviews often focused on specific diseases or interventions (eg, salt reduction 14 ). To our knowledge, no review has ever focused specifically on PN. Therefore, we reviewed and critically appraised CEAs of personalized interventions with a nutrition component in adults by describing and assessing their methodology, findings, and quality. This can support policy decisions around PN. 2,12 In addition, this review can help to design and improve future CEAs of PN interventions.

Literature Search
The approach in this review was based on a series of 3 articles describing methodological guidelines for systematic reviews of CEAs. 12,15,16 The term CEA was used as an overarching term for full economic evaluations such as CEA and cost-utility analysis (CUA Two authors (MMJG, WKR) independently reviewed titles and abstracts of all articles (including CEAs found via screening systematic reviews) to determine which ones met the eligibility criteria. Interrater agreement about the eligibility for full-text review was then assessed and found to be moderate (Cohen's kappa: 0.498). 17,18 Any disagreement not resolved by discussion resulted in full-text screening. Full-text versions of the articles were then examined to determine which ones met all eligibility criteria. This was done primarily by the first author (MMJG) using a detailed list of criteria, and any doubt was discussed with a second reviewer (WKR).

Data Extraction/Analyses
Data extraction was initially done by one author (MMJG) and checked by a second author (WKR). General features of the studies that might influence economic outcomes (eg, intervention characteristics including definitions) were collected as well as economic findings themselves (eg, incremental cost-effectiveness ratio and incremental cost-utility ratio (ICUR)). Summary tables and figures of these characteristics were created, and each intervention was matched to a PN concept. Previous literature defined the conceptual basis for PN; specifically, personalization can be based on the analysis of current eating habits, behavior, preferences, barriers, and objectives ("psychological concept") or on the biological evidence of differential responses to foods/nutrients (ie, biomarkers, genotype, and microbiota) ("biological concept"). 2,19 Conclusions of the authors regarding the cost-effectiveness of the intervention were collected and arranged into 4 categories: "yes" (cost-effective), "no" (not cost-effective), "sometimes" (only cost-effective in some subgroups), and "no conclusion." Total costs and ICURs were inflated to 2019 costs using the country-specific Consumer Price Index 20 and converted to Unites States dollars (US$) using the purchasing power parity. 21 If the cost year of the study was not specified, it was assumed to be the year of publication. To determine whether an intervention would be considered cost-effective, ICURs were compared with 2 willingness-to-pay (WTP) thresholds (values in US$ per qualityadjusted life-years (QALY)): $20 000 (close to the thresholds of £20 000 ($25 937 22 ) used in United Kingdom and V20 000 ($23 680 22 ) in The Netherlands for interventions targeting diseases with a low disease burden 23 ) and $50 000 (widely used in the United States). The incremental net monetary benefit (iNMB) was calculated by valuing incremental QALYs in monetary values using both thresholds. Furthermore, we examined possible relationships between the results (QALYs and costs) and general features (ie, population, intervention, choice of comparator) and modeling choices (ie, time horizon, perspective, discount rate, number of health states, intermediate outcomes, and assumptions regarding intervention effects).

Quality Assessment
The quality of all studies was assessed using the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist, 24 which is preferred when modeling studies are included. 16 This checklist consists of 24 items, subdivided into 6 categories: (1) title and abstract; (2) introduction; (3) methods; (4) results; (5) discussion; and (6) other. There are 3 possible answers for each item: fulfilled, not fulfilled, and not applicable.

Methodology of the CEAs
Nineteen studies 27

Relationship Between Study Characteristics, Methods, and Results
Examination of the relationship between study features and economic outcomes yielded a number of noteworthy findings. First, interventions that were considered cost-effective according to the authors showed incremental QALYs that varied from 0.0090 48 to 0.7714 68 and costs varying from $-4877 30 to $7369 30 (iNMB(l = $50 000) mean: $5769) ( Table 1). In contrast, interventions considered not cost-effective by the authors showed incremental QALYs varying from -0.0340 48 to 0.0200 54 and costs from $-1087 60 to $2026 49 (iNMB(l = $50 000) mean: $-940).
Second, variation in incremental costs, QALYs, and iNMB is seen between the PN concepts (Fig. 3B). The highest mean iNMB (l = $50 000) was found in the integrated approach ($13 366), followed by the psychological concept ($4443) and the biological concept ($13) (Table 1). Third, a wide variation in incremental costs and QALYs is found within the DPP and DPS interventions, despite their comparable PICOs (Fig. 3C). For example, 2 main outliers were found in the DPP CUAs; 1 study was associated with relatively high costs ($10 242) and low QALY gain (0.034) ($299 424 per QALY, iNMB (l = $50 000) $-8531), 28 and the other outlier reported costs of $-4877 and QALY gain of 0.4500 (iNMB (l = $50 000) $27 377). 30 The relationship between costs and QALY results of DPP and DPS CUAs and various study characteristics, including methodology, was explored. First, some differences in PICOs of DPS studies might explain differences in outcomes (see Appendix 3 in Supplemental Materials found at https://doi.org/10.1016/j.jval.202 0.12.006); slightly different populations were studied in different countries (eg, Switzerland 36 and the UK 41 ). Moreover, different comparators were used, but no clear pattern related to outcomes was observed here. Second, longer time horizons were associated with more QALY gain. Third, we found that an assumed prolonged effect of DPS intervention 80 (for 20 years) causes higher QALY gain compared to waning or no lasting effect. Fourth, 1 study did not consider the DPS intervention impact on hypertension, hypercholesterolemia, and CVD and reported lower QALYs than other CUAs. 41    correctly. Most problems in reporting were found in statement 18 related to study parameters (n = 26 not fulfilled) and in reporting heterogeneity of cost-effectiveness results across different subgroups/patient populations (statement 21); 13 studies 28,30,48,60,78,31,[34][35][36][37][38][39]47 reported this appropriately. Appendix 9 in Supplemental Materials found at https://doi.org/10.1016/j. jval.2020.12.006 provides information about the quality per study.

Discussion
This systematic literature review was done to synthesize and critically appraise CEAs of PN interventions. We identified 53 CEAs of interventions with a PN component in adults. Interventions were based mostly on the psychological concept of PN (48 studies), 1 study 47 on the biological concept and 4 studies 59,68,78,88 on the integrated approach. Approximately half of the authors concluded that an intervention with a PN component was costeffective (49%). Of the interventions that reported a QALY gain, 55% were cost-effective according to the lowest assessed threshold $20 000, increasing to 75% based on a threshold of $50 000. Moreover, studies that used an integrated approach showed the highest iNMB based on both $50 000 and $20 000 thresholds.
Wide variation in methodology of the CEAs in this review was found. First, variation is seen in terminology/definitions of PN and in the conceptualization of the terms. For example, Sherwood et al 34 used "individualized" to describe individual counseling sessions with goal-setting and individual feedback, whereas Olsen et al 38 only used "individualized" to describe individualized counseling sessions. Furthermore, the duration of the personalized component used in the interventions varied. For example, 2 studies used the term "personalized" but varied the duration of the interventions; participants receiving 1 intervention could expect to have 4 counseling sessions on personalizing snacks, 53 whereas participants receiving a different intervention received personalized messages via the internet when needed. 33 Future research could examine how the different terms used in PN relate to cost-effectiveness.
Second, different comparators and number of comparators are used in studies, resulting in different cost-effectiveness outcomes. While the "best" comparator is study-dependent, 1 comparator might be insufficient in some cases. For example, if usual care is used as a comparator to assess a PN intervention, a second comparator could be a similar nutrition intervention but without the personalized component. By adding this third arm, researchers would be able to see not only the effect of the intervention (when Standards. The 24 statements of the checklist are shown on the y-axis. The frequencies of each category are shown on the x-axis. Three categories were used: Fulfilled (study scored well on this statement), Not fulfilled (study scored poorly) and Not Applicable (ie, the statement was not applicable for a study). The total number of studies included was 49 since the article of Dalziel  compared to usual care), but also the effect of a specific personalized component. Additional research regarding the best choice of comparator when studying PN interventions is needed. Third, different cost perspectives were used; choice is mainly depending on the resident country of the population. Two CEAs found in this review used the perspective of an individual, 28,76 which might be considered when assessing the costeffectiveness of PN interventions since individuals will likely have to pay for at least part of the extra costs; the actual amount would be country-and intervention-dependent. However, these 2 CEAs did not include all costs related to this perspective. This is very similar to what Bilvick Tai et al 90 reported in their systematic review. They not only found a paucity of CEAs using a patient perspective but also observed that studies that used this perspective did not fully explore the true patient costs.
Fourth, variation was observed in time horizons, and many CEAs used time horizons that are probably too short to capture all important effects of PN interventions on outcomes and costs. That is, CEAs with a short follow-up would not observe any long-term benefits of behavioral change and would therefore show less favorable results than ones with a longer follow-up. 80,91 Furthermore, nutrition often has a preventive effect, in which benefits take longer timespans to develop. 92 One CEA from this review supports this and showed a decrease in ICURs when time horizons increase (per QALY gained: £113 905 ($238 856) (year 1) -£5825 ($12 215) (year 15)). 41 Moreover, from DPP/DPS studies it was observed that longer time horizons were associated with more QALY gain. It is therefore recommended to use longer time horizons and/or to include both trial and model data to investigate the full impact of PN. While well-designed trials can help to establish short-term (cost-)effectiveness of interventions, modeling beyond that point may be unavoidable to estimate the intervention's overall cost-effectiveness.
It is debatable what the best modeling approach for PN interventions beyond the trial can be. Nutrition economics requires a holistic approach because of the complexity of food and its interactions with multiple interdependent processes, 92 and yet there is no systematic approach to assess the health impact of (personalized) nutrition. 93 Therefore, there is still much variation in models for PN (even those with comparable PICOs, eg, DPP/DPS interventions), resulting in avoidable variation in estimated costs and QALYs. Some suggestions specific for nutrition interventions could be made for models, such as linking identified markers in trials to longer-term outcomes. 92 For example, Eddy et al 28 linked LDL cholesterol to a reduction in long-term CVD risk. More research is needed to define good PN modeling approaches.
Variation in QALYs was observed between populations. The smallest QALY gain was observed in the malnourished population. Since all studies found in this population were done in elderly, this might explain the lower QALY gain compared to younger populations. These findings are in line with an earlier review that reported that studies in elderly found no differences in quality of life between intervention and control treatments. 94 Additionally, variations in health economic outcomes between the different PN concepts were found, in which most promising outcomes were found by the integrated approach. However, only a few CEAs with different methodologies evaluated the integrated approach. Nevertheless, there are different reasons to suspect that an integrated approach will be most cost-effective. First, this review found a lowest iNMB in CEAs with an integrated approach. Second, previous studies in the nutrition field have mentioned that an integration of biological and psychological characteristics is the optimal approach. 2,19,95 An example of an intervention with an effective integrated approach is Food4Me, which has shown greater improvement in dietary behavior. 10,96 Moreover, CEAs of integrated approaches in different disease areas often tend to have positive results, such as improved cost-effectiveness of the integrated care management versus the standard care of advanced chronic obstructive pulmonary disease. 97 This integrated approach of PN deserves further investigation.

Limitations
First, since our literature search was restricted to CEAs published in English-language journals, it may have missed CEAs reported elsewhere. Second, some bias in our review might have arisen through inclusion of poor-quality CEAs. Nevertheless, assessing quality of the CEAs was important for revealing improvements for future CEAs, such as better reporting on study parameters. Third, our results could have been influenced by publication bias, since interventions that are found to be costeffective are more likely to be published. 98 Fourth, heterogeneities in methodology and the limited number of CEAs that studied the integrated or biological concept made it difficult to draw stable conclusions about the cost-effectiveness of these concepts; more CEAs are therefore needed.

Future Research
In addition to the suggestions for future research already given above, another question to consider is how much people are willing and able to spend on PN. This review calculated iNMBs with 2 different WTP thresholds, but there is no specific cut-off point defined in the literature for PN. 54 A study by Corso et al 99 found that treatment is preferred above prevention by society, which might imply that the WTP might be greater for a comparable treatment rather than for prevention-oriented PN. Since costs of these interventions are often (partly) borne by the user, WTP studies of PN interventions could give perspectives on potential consumer behavior for 2 reasons. First, a WTP will indicate the willingness of the user to make the required behavioral change and how much the user expects to benefit from PN. Second, these studies show policy makers how much demand might vary between different social classes and indicate how demand for PN varies depending on the level of public subsidy applied. However, to date, it seems there has been only 1 published WTP study in this area. 100 Moreover, multiple criteria decision analysis might be considered for future research, because there are many factors besides cost-effectiveness that affect the value of PN. 42,[101][102][103] Personal preferences might be relevant as well, and particularly for dietrelated interventions, since food-and all activities related to food-has a profound role in a person's life. Therefore, any assessment of the merits of PN strategies should consider preferences.

Conclusions
Heterogeneity exists in the methodology of CEAs done in the field of PN, including variation in definitions and its conceptualization, PICOs, and modeling approaches. This leads to differences in health economic outcomes. Nevertheless, PN interventions tend to be cost-effective compared to usual care and drug-related treatments with WTP thresholds of $20 000 and $50 000. This suggests that many PN interventions may offer good value for money. Moreover, this review found that an integration of PN concepts may yield the greatest iNMB. Future CEAs should improve their methods to support later implementation and reimbursement decisions.