Background

In 2018, the United States spent $3.6 trillion on health care, accounting for 17.7% of Gross Domestic Product (GDP) [1]. As a percentage of GDP, U.S. health care spending is nearly double the Organisation for Economic Cooperation and Development average, yet this does not translate into superior health outcomes [2]. Moreover, approximately 25% or more of health care spending is considered waste [3].

Researchers, health care managers, and policy makers thus have increasingly begun to focus on improving value in health care. Value is commonly conceptualized as outcomes or quality divided by costs, although the precise definition varies depending on the clinical context and stakeholder perspective. Along with this focus, large-scale initiatives to improve value have emerged. For example, the Affordable Care Act created incentives for improving quality while reducing costs through Medicare Shared Savings and bundled payment programs [4]. In addition, the American Board of Internal Medicine’s Choosing Wisely initiative generated lists of low value services across a range of specialties with the aim of reducing use of services that have little or no benefit to patients [5].

Though research on the quality and cost of care has been performed for decades, emphasis on value is more recent. Researchers seeking to understand barriers and facilitators of high value care have hypothesized that a wide range of factors—including financial incentives, delivery system structure, geography, demographics, medical education, and patient involvement—contribute to health care value and must be optimized to promote high value care [6]. In addition, extensive work has been done to identify and decrease utilization of low value services. Yet empirical studies analyzing drivers of health care value are relatively sparse. We therefore performed a scoping review to identify potential factors driving value in health care.

Understanding factors associated with health care value requires examination of drivers at multiple levels, including system-level factors (such as policies or insurance structures), hospital-level factors, physician- or practice-level factors, and patient-level factors. This review synthesizes existing literature exploring factors associated with value across these levels to begin building an understanding of how we can address disparities in value and, ultimately, create high value health care systems.

Methods

We conducted a scoping review to identify potential drivers of value in health care. Scoping review methodology, as opposed to systematic review methodology, was selected for this study given that our aim was to provide an overview of a broad range of factors as opposed to gathering evidence in favor of a given treatment, for example [7, 8]. This review followed the five core stages of scoping reviews outlined in the methodological framework by Arksey and O’Malley [9].

Search strategy

With the assistance of a research librarian, we searched PubMed, Embase, PolicyFile, and Google Scholar to identify articles using the terms “value,” “health care,” “low value care,” and “high value care,” in their titles. Boolean operators “AND” and “OR” were used to combine search terms where relevant. The search strategy was limited to English-language publications. No limitations were placed on publication date or study design during the search. Google Scholar hits were limited to the first 100 articles. Scopus citation tracking was also used to identify the top 20 cited articles based on the search terms “value” and “health care.” Searches were performed between February and March 2020. The complete search strategy is contained in Appendix 1. Reference lists of studies selected for inclusion after full-text screening were scanned for potentially relevant articles that may have been missed in initial searches. Grey literature was used to provide context for this review but was not used in analyses given the already broad scope of review.

Study selection

Empirical studies that explored potential factors associated with value were eligible for inclusion. There were no limitations placed on the population studied. We defined the primary concept of interest as the value of healthcare. Only articles that used value as the dependent variable were eligible for inclusion. We accepted a broad range of definitions of value for the purposes of this review but required that the operational definition used a combined measure of cost and outcomes/quality or used a set of services considered to be low or high value (which implicitly includes these terms). Examples of cost measures can include expenditures for an episode of care or specific service, or extend to include downstream costs related to an initial episode of care or service. Examples of operational definitions of quality or outcomes, on the other hand, include patient experience, quality of life measures, and life-years lost or gained. Defining value based on a set of services considered to be low value relies on these services having been shown to provide little to no clinical benefit, thus creating an exceedingly high cost-benefit ratio. Studies exploring a broad range of independent variables and their associations with value were considered. No limitations were placed on context for this review, as value is important across all care settings and environments. Opinion-based articles and literature reviews were excluded from analysis. Studies that were not available in full text through NYU libraries were excluded from analysis. Studies about interventions or implementation of interventions to improve value were excluded. A table with full inclusion and exclusion criteria is presented in Appendix 2.

All studies identified through database searches were exported and uploaded into Covidence online software. Duplicates were automatically removed. An initial review of approximately 15 articles was conducted with all 3 authors to ensure agreement on inclusion criteria. One author (S.L.) then screened the remaining records by title and abstract. Potentially relevant articles were included in the full text review. Any articles of questionable eligibility were discussed by all three authors in the initial screening phase or during full text review. The final decision to include or exclude these publications was determined by consensus. Studies meeting all inclusion criteria after full-text review were kept for data extraction. Reference lists of included studies were hand-searched for any articles that may have been missed in initial searches.

Data extraction and quality appraisal

A data extraction template was created in Microsoft Excel. Elements extracted from eligible articles included basic publication information, study design, population, definition of value (stated and/or operational), statistical methods, main results, conclusions, and limitations. The Critical Appraisal Skills Programme (CASP) quality appraisal tool for cohort studies was used to assess methodological quality of included articles [10]. All articles best fit the CASP cohort studies tool. One author (S.L.) completed all data extraction and quality appraisal.

Data analysis

Findings from studies were synthesized using narrative methods, characterizing studies based on the independent variables (i.e. drivers of value) studied. We organized these drivers into four domains, based on the Institute of Medicine’s delivery system framework: the economic environment (insurance characteristics), the organizational environment (health system and practice characteristics), the care team (physician and other care provider characteristics) and the individual patient (patient-level characteristics) [11]. Zotero was used to manage citations for all articles included in the manuscript. Writing of the review followed PRISMA guidelines for scoping reviews [12].

Results

Study selection

After removing duplicates, a total of 1750 publications were identified through initial searches. An additional 15 publications were identified through hand searching reference lists of articles selected for inclusion. Following screening by title and abstract, 68 were included in the full-text review. Ultimately, 22 studies were found to meet inclusion criteria. The three most common reasons for excluding articles after full text review were that the article did not use value as the dependent variable, focused on implementation of interventions to improve value, or used an ineligible study design (e.g. literature review, opinion-based article). A PRISMA flow-diagram is presented in Fig. 1 to illustrate the study selection process [13].

Fig. 1
figure 1

PRISMA Flow Diagram of Study Selection Process [10]

Characteristics of included studies

Seventeen of the 22 studies exploring factors associated with value in health care were retrospective cohort or cross-sectional studies [14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30]. Three studies used a difference-in-difference or interrupted time series approach [31,32,33] and two used simulation modelling methods [34, 35]. Eighteen studies focused on low value care, two on both low and high value, [23, 29] and two on overall value [27, 35]. Among the included studies, 17 were set in the US, 3 in Canada, [16, 22, 24] 1 in England, [33] and 1 in Australia [34]. Table 1 provides a summary of the methods and major findings of included studies.

Table 1 Summary of Included Studies

The 20 studies that explored factors associated with low value used receipt of low value services, as identified using claims data, as the outcome variables. Of these 20 studies, three also looked at costs associated with the specified low value services as an outcome [15, 25, 31]. In contrast to those studies, Weeks et al., invented a composite measure of value to use as the primary outcome variable. Defining value generally as quality divided by expenditures, the authors deconstructed quality into four component pieces which can be summarized as patient experience, care processes, care outcomes, and safety [27]. Data for each of these components was drawn from Hospital Compare. The authors standardized these measures, multiplied the component pieces together, then divided by total expenditures for the episode of care in order to get a measure of value [27]. On the other hand, Braithwaite et al., defined value as incremental benefits over incremental costs and examined incremental changes in life expectancy and costs as the outcome variables [35].

Quality appraisal

Overall, the quality of included studies was high (Appendix 3). All studies had clearly stated objectives. All of the low value studies used large, representative datasets for cohort selection, and claims-based measures to assess receipt of low value services. Weeks et al. used Hospital Compare data, which allowed for examination of a broader definition of value but lacked some granularity given its aggregation of data [27]. Braithwaite et al. used a validated computer simulation of the entire US health care system [35].

Two articles considered to be of lower quality were those by Koehlmoos et al., and Pendrith et al., as these studies did not adjust for potential confounding variables [20, 24]. The remaining 20 studies adjusted for potential confounders and/or conducted sensitivity analyses. In general, most studies included identified patient-level characteristics as the main potential confounders.

In terms of generalizability of the studies, given the data sources and study designs, it is likely that the findings would apply to the local populations studied. It is unclear, however, if a subset of services considered to be low value is representative of all low value care, and thus if the results are generalizable to all low value care. It is also unclear if results would be applicable to different populations or health systems.

Insurance and health system characteristics

Six studies examined associations between insurance characteristics and receipt of low value services [17, 19, 20, 29, 31, 32]. One study examined associations between different value-based insurance designs and incremental changes in life expectancy and cost [35]. One study examined effects of a national efficiency savings program on low value service utilization [33]. Of the studies exploring associations between insurance characteristics and low value service utilization, five studies showed no clear pattern of significant associations between insurance type and low value care [17, 19, 20, 29, 31]. Three of these studies compared low value service utilization in commercial versus public insurance, [17, 19, 29] one compared direct versus purchased care, [20] and one compared those in a Consumer Directed Health Plan versus traditional plan [31]. Barnett et al. found similar rates of low value care in Medicaid, uninsured, and privately uninsured populations [29]. However, they also found that Medicaid patients were less likely than privately insured patients to receive high value services when pooling estimates of high value service utilization whereas uninsured were more likely to receive high value services compared to their privately insured counterparts. The authors speculated that the latter finding may have been due to unmeasured confounding in a highly select population [29].

In contrast to studies that found no associations between insurance type and value, Schwartz et al. found that patients enrolled in a Pioneer Accountable Care Organization (ACO) Medicare plan saw a differential reduction in the use of low value service (− 0.8 services/100 beneficiaries; 95% CI: − 1.2,-0.4) in the first year of the contract compared to those who remained on a traditional Medicare plan [32]. Accountable Care Organizations are groups of healthcare providers that jointly contract with an insurer (typically, Medicare) to provide coordinated care in exchange for a share of any savings. Some also share in any excess costs. In a computer simulation study, Braithwaite et al. found that value-based insurance designs reducing cost-sharing for high value services and increasing cost-sharing for low value services were associated with a 0.24–0.44 gain in life years without increasing spending [35]. Finally, Coronini-Cronberg et al. found mixed effects on low value service utilization across commissioning organizations from a national efficiency savings policy applied to the National Health Service [33].

Hospital characteristics

Two studies included focused on hospital characteristics associated with value [27, 34]. Badgery-Parker et al. used a multilevel modelling design to test the effects of hospital and region on receipt of low value care. They found that hospital is a more significant contributor to low value care than region [34]. Weeks et al. looked at hospital characteristics such as census, beds, number of operations, accreditation, and for-profit status. They found that higher value was associated with higher census, more beds, and more operations [27].

Physician and practice characteristics

Seven studies focused on physician and/or practice characteristics [14,15,16, 21, 23, 28, 30]. Two additional studies explored physician characteristics in addition to patient characteristics, [18, 22] and one study explored physician characteristics in addition to insurance characteristics [29]. Physician characteristics associated with reduced use of low value services include more recent medical school graduation, allopathic training, smaller patient panel, and being female [15, 16]. Mafi et al. also found that Advanced Practice Clinicians (APCs) had comparable low value service utilization to physicians [21] and Barnett et al. found that safety-net and non-safety net providers had comparable low value service utilization [29].

Investigating both physician and practice variation in low value service utilization, Schwartz et al. found that physician characteristics account for relatively little variation in provision of low value services while organizations contribute substantially more to variation in low value service provision [28] Another study by Schwartz et al. demonstrated a standard deviation of 10 low value services per 100 beneficiaries between provider organizations, further supporting that low value service utilization varies substantially across provider organizations [30]. Other articles identified more specific practice characteristics that may contribute to variation in rates of low value service utilization. For example, fee-for-service practices had higher rates of low value service utilization than capitated practices, [16] community health centers had higher rates of high value care and lower rates of low value care than private practices, [23] and hospital-based practices had higher use of low value services than community-based practices [14]. In addition, two studies showed higher specialist to primary care provider ratios were associated with greater low value care utilization [18, 22].

Patient characteristics

Three studies explored patient characteristics associated with low value care [24,25,26] in addition to the two studies that explored both physician and patient characteristics [18, 22].. Two studies found that being female was associated with higher rates of low value care utilization compared to being male [24, 25]. Two studies showed that black and Hispanic patients had received low value services more frequently [18, 26]. In contrast, Reid et al. found decreased low value spending among black beneficiaries [25]. Four studies showed that higher income was associated with increased rates of low value care utilization [18, 22, 24, 25]. Three studies showed that region was associated with low value care utilization, although patterns for this varied [22, 24, 25].

Discussion

In this review examining factors associated with health care value, we identified 22 studies exploring a range of system- (including insurance), hospital-, provider-, and patient-level characteristics and their associations with value. Across the 22 studies included, almost all [20] focused on utilization of low value services as the dependent variable. Although included studies showed mixed results, several more consistent findings emerged, many of which are reminiscent of findings from past research on factors associated with health care quality and factors associated with health care spending.

One overarching finding from studies included in this review is that incentive schemes may be important drivers of low value service utilization, but only when structured certain ways. For example, in the included articles, neither consumer-directed health plans nor public vs private insurance were associated with low value service utilization, but ACO enrollment was associated with reduced utilization of low value services and value-based insurance design was predicted to increase life expectancy while keeping expenditures constant [32, 35]. ACOs are designed with a primary aim of improving value, and these findings suggest that such a structure may in fact be helpful in doing so. ACO and Alternative Quality Contract enrollment has also been shown to be associated with reduced spending and higher quality, although associations with quality have been less consistently shown [36,37,38,39].

This review also illustrated considerable variation in value by region, hospital, and provider type [22, 24, 25, 27, 28, 30, 34]. Similarly, previously published research has shown regional variation in quality and spending [40,41,42] as well as variation in readmission rates by hospital [43]. In many cases, the associations found with value were similar to those found with quality. For instance, much like Weeks et al. found that higher volume hospitals tended to have higher value, Birkmeyer et al. showed that higher hospital volume was associated with lower surgical mortality—an important element of hospital quality [44]. However, it is important to note that value and quality are distinct. Weeks et al., for instance, found that 50% of the hospitals in the highest or lowest value quintiles were not in the highest or lowest quality quintiles, respectively [27].

Additionally, this review underscored that greater intensity of care (defined as higher ratio of specialists to primary care providers, hospital-based, and fee-for-service payment structure) was associated with higher utilization of low value services [14, 16, 18, 22]. This trend is similarly reflected in quality and cost literature. For example, Wennberg et al. found that greater intensity of hospital care was associated with lower quality care and lower patient experience ratings [45]. In terms of cost, Barnett et al., found that hospital-based physician networks were associated with higher costs and higher intensity of care [46].

Studies included in this review also highlighted physician characteristics associated with value, showing that recent medical school graduation and allopathic training was associated with lower utilization of low value services [16, 18]. It is plausible that this finding reflects changing medical school cultures at allopathic institutions, which are now placing a greater emphasis on value. Tsugawa et al. found that patients treated by older physicians had higher 30-day mortality rates compared to those treated by younger physicians; this difference was not observed in high volume providers nor when comparing readmission rates [47]. They also found slightly higher costs in older physicians [47]. In contrast, Mehrotra et al. found that more recent medical school graduates tended to have higher cost profiles [48]. This discrepancy could reflect several factors including change over time in medical education, lack of concordance between overall spending and spending on low value services, or, given that one of the articles on physician characteristics associated with low value service utilization was set in Canada, there could also be differences across countries. Additionally, Schwartz et al. found that provider organizations had much larger effects on low value service utilization than individual physician characteristics, providing further support for the role of value culture and potentially explaining some of this uncertainty [28].

Higher income and female sex were associated with lower value care in several studies included in this review. Interestingly, opposite results have been found with studies of quality and cost individually. Okunrintemi et al. found that low income patients reported lower quality of care, greater difficulty accessing care, and worse experiences of care compared to high income patients [49]. Epstein et al. also found that patients of a lower socioeconomic status had longer hospital stays and higher charges compared to those with a higher socioeconomic status [50]. Bird et al. showed that women tended to receive higher quality of care than men except in areas related to cardiovascular disease and adverse drug-disease interactions [51]. These differences again highlight the difference between low value service utilization and provision of high quality care.

Importantly, studies included in this review overwhelmingly focused on low value, highlighting current limitations in data and metrics and raising the question of whether patterns of low value service utilization are reflective of value more generally. Although reducing utilization of low value services may seem to be a logical approach to improving value, it is likely insufficient to create a high value health care system. As described in this review, there was a paradoxical finding that high volume was associated with high value, whereas greater intensity of care was associated with higher utilization of low value services. This seeming contradiction provides evidence of a rift between what is deemed high value (including excellent patient outcomes) and the mere absence of low value service utilization. Moreover, studies have repeatedly found that efforts to influence utilization tend to change rates of both appropriate and inappropriate care [52, 53].

By contrast, only four articles in this review explored factors associated with high value (or value) in health care, possibly because high value is often conflated with high quality. In fact, Oronce et al. and Barnett et al. operationalized high value using lists of services that are frequently considered to be quality process measures. But high value and high quality are not the same construct. A health system providing high quality care at very high cost is likely not high value; a system providing both high quality care and excessive low value services is also not high value. These tradeoffs are explicitly addressed in cost effectiveness literature, but this review illustrates that they have had limited use to date when identifying “high value” care.

Strengths and limitations

This scoping review provides an overview of the literature exploring factors associated with value in health care. It is the first to bring together this body of literature as a step towards understanding and utilizing the complex interplay of system-, hospital-, physician-, and patient-level characteristics driving value.

This review has several limitations which are important to address. First, as a scoping review, this relied on title searches to identify relevant articles. Thus, it is possible that relevant studies were missed given the constraints of the search. However, we searched reference lists of included studies to identify additional articles. Second, studies included explored a range of factors that may affect value, with value having been defined using different lists of low value services and different levels of sensitivity and specificity. Consequently, study findings did not elucidate a consistent factor driving value, and comparability and implications of results must be interpreted with caution. Third, given that included studies were primarily observational, we cannot establish causality. Fourth, given the heterogeneity of studies included, it was not possible to conduct a meta-analysis or combine the findings in any quantitative manner. Finally, although three researchers were involved in the planning of the study and any questions that arose regarding inclusion and exclusion criteria, only one reviewer selected articles for inclusion and appraised the quality, making it prone to researcher bias. Pre-determined criteria were used for article selection and standardized templates were used for data extraction in an effort to mitigate this bias, but we acknowledge that this is nonetheless a major limitation of this study.

Conclusion

A complex interplay of system-, hospital-, provider-, and patient-level factors appear to be driving value in health care. In this review, we found evidence supporting the role of insurance incentive schemes, intensity of care, and culture as key drivers of health care value. Although there exists some overlap between factors driving quality of care and factors driving health care value, we found they are not identical, and thus the two constructs must be considered as distinct entities. Moreover, drivers of high value care and drivers of low value care appear to differ. This is especially important to consider when developing interventions to improve value in health care and establishing standards for defining and measuring value. To effectively improve health care value, developing interventions to address drivers across multiple levels of care as well as novel measurement strategies to capture both high and low value practices will be essential.