Randomized Experiments to Reduce Overuse of Healthcare: A Scoping Review

Abstract

Objective:

Healthcare overuse is pervasive in countries with advanced healthcare delivery systems. We hypothesize that effective interventions to reduce low-value care that target patients or clinicians are mediated by psychological and cognitive processes that change behaviors, and that interventions targeting these processes are varied. Thus, we performed a scoping review of experimental studies of interventions, including the interventions’ objectives and characteristics, to reduce low-value care that targeted psychological and cognitive processes.

Methods:

We systematically searched databases for experimental studies of interventions to change cognitive orientations and affective states in the setting of healthcare overuse. Outcomes included observed overuse or a stated intention to use services. We used existing frameworks for behavior change and mechanisms of change to categorize the interventions and the mediating processes.

Results:

Twenty-seven articles met inclusion criteria. Sixteen studied the provision of information to patients or clinicians, with most providing cost information. Six studies used educational interventions including provision of feedback about individual practice. Studies rarely used counseling, behavioral nudges, persuasion, and rewards. Mechanisms for behavior change included gain in knowledge or confidence and motivation by social norms.

Conclusions:

In this scoping review, we found few experiments testing interventions that directly target psychological and cognitive processes of patients or clinicians to reduce low-value care. Most studies provided information to patients or clinicians without measuring or considering mediating factors towards behavior change. These findings highlight the need for process-driven experimental designs, including trials of behavioral nudges and persuasive language involving a trusting patient-clinician relationship, to identify effective interventions to reduce low-value care.

The overuse of healthcare remains pervasive in the U.S. and in other countries with advanced healthcare delivery systems.¹ Overuse of healthcare, or the provision of low- or no-value care, is consistently identified as contributing to high costs in the U.S. healthcare system.² Low-value care often refers to clinical services that provide little benefit to patients in specific clinical scenarios relative to their cost, potential risks, and effectiveness in diagnosis or treatment. More importantly, these wasteful services can be physically, psychologically, or financially harmful to patients.^3–5 Many studies have explored potential determinants of low-value care,^6,7 yet there are few interventions targeting these drivers. Many interventions targeting low-value healthcare have used quality improvement and process improvement methods to reduce these practices. These interventions are typically hyper-local and target a specific overused process, such as those identified as discussion opportunities by the Choosing Wisely Initiative, a decade-long campaign to promote conversations between clinicians and patients regarding unnecessary tests, treatments, and procedures.⁸ Local interventions often do not readily generalize to other settings, and quality improvement interventions have had varying impact on reducing low-value care.^9,10 In contrast, some proposed interventions use population-directed or policy interventions to align incentives to encourage high-value care or remove opportunities to use low-value services. Supportive evidence is sparse, though, that interventions like partial capitation, bundled payments, and other value-based payment structures have reduced low-value care delivery.¹¹

Colla proposed attending to both demand-side and supply-side mechanisms to reduce low-value care such as through incentives to change clinician behavior or through patient education to alter their demand for services.¹² Many demand-side mechanisms such as cost-sharing and supply-side mechanisms such as prior authorization, can reduce both low-value and high-value care.^13,14 Multicomponent interventions addressing patient and clinician roles in overuse may hold particular potential to reduce low-value care.¹⁵ We were interested in learning which behavior change interventions to impact low-value care have been experimentally tested. For example, providing information to patients about the risks associated with antibiotic treatment of acute sinusitis may decrease demand for such therapies. We hypothesized that the impact of these interventions, which target either patients or clinicians, is mediated by psychological and cognitive processes that alter individual behavior or decision-making. Thus, we performed a scoping review of experimental studies of interventions to reduce low-value care that targeted psychological and cognitive processes of patients or clinicians.¹⁶ We describe the interventions’ objectives and characteristics and classify the interventions, building on existing frameworks. We anticipated that the evidence of effectiveness for any given intervention would be sparse and of low strength and that a scoping review would be appropriate for this goal.

Methods

The protocol for this review was registered in PROSPERO (ID# CRD42023396499). This review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA).^17,18 This study did not require institutional review board review.

Eligibility Criteria

We sought articles that tested a) cognitive orientations, schema, biases and processing styles (e.g., constraint mindsets) and b) affective information or other temporarily activated states (e.g., emotions) in the setting of using healthcare or in response to the presentation of a vignette. We included only those studies that evaluated an intervention’s impact on the use of low-value healthcare services in observed or hypothetical settings. We excluded studies that were published in languages other than English, did not study human subjects, or were non-experimental.

Literature Search

We developed search strategies appropriate for PubMed Medline, Embase, PsychINFO, and EconLit (see Supplement Table 1, Supplemental Digital Content 1) and searched these databases from March 2023 to May 2023. The search strategy prioritized MeSH headings in PubMed where available. We included terms to reflect healthcare utilization, overuse, efficiency, and expenditures and limited this search to studies that used experimental designs including experimental surveys, incentivized experiments, or randomized controlled trials (RCTs) to test decision-making processes. We developed the search strategy iteratively by targeting specific articles that we classified as being appropriate for inclusion. We conducted a second search with modestly revised terms to include discrete choice experiments and conjoint analyses. Titles and abstracts were imported into the Covidence software (Melbourne, VIC).

Study Selection

Paired investigators reviewed the abstracts identified by the search. If either reviewer thought the abstract appropriate, it was included for further review. Next, paired investigators reviewed the complete text of the articles and applied inclusion criteria as follow: involvement of adult subjects, a demonstrated or implied affective response and alteration of cognition evoked by the intervention, presence of a comparator, and inclusion of either patient- or clinician-reported intent to use an overused healthcare service or their observed use of these services as an outcome. We excluded trials of clinical decision support interventions or decision aids. We also excluded trials of interventions aimed at reducing use of services at end-of-life and medication deprescribing studies. If there were disagreements about inclusion, the entire study team reviewed and discussed the article and came to consensus.

Data extraction

We extracted key data from the included studies into tables. We focused on the studies’ experimental designs, characteristics of the interventions, primary and secondary outcomes, and any hypothesized mediators of the outcomes that were explicitly stated in the articles. A single reviewer extracted data and the accuracy was verified by a second reviewer. Disagreements were resolved by discussion among the full team.

Data Synthesis

We summarized the study characteristics and qualitatively synthesized the information about interventions and the likely mediators of effects. We determined whether these mediators were stated by the studies’ authors or inferred. Through discussion, the team identified unifying themes across the body of literature. We generated a flowchart that drew upon a behavior change framework by Michie et al¹⁹ and an implementation science framework for mechanisms of change by Lewis et al²⁰ to categorize the interventions.

Risk of Bias and Strength of Evidence Assessment

We did not assess the risk of bias in the included studies as this is less relevant to a scoping review. Accordingly, we also did not grade the strength of evidence as this review did not aim to summarize the effectiveness of these disparate interventions.

Results

Results of Search

The initial search yielded 1816 titles after removal of duplicates. Most articles were then excluded as they were non-experimental or did not study low-value care. Twenty-seven articles were appropriate for inclusion (see Figure and Supplement Table 2, Supplemental Digital Content 1). Most articles were published between 2010 and 2020, with half of these conducted in the U.S (Table 1). Fourteen (52%) experiments targeted clinicians and 13 (48%) targeted patients. Fourteen (52%) studies observed use of services as the outcome and 13 (48%) evaluated the intended usage of services in hypothetical scenarios. The clinical services and setting of care varied widely. The experiments tested interventions (or stimuli) that we classified into the following categories: provision of information, education (which includes feedback), counseling, nudges, persuasion, and use of penalties or rewards (Table 2 and Supplement Figure, Supplemental Digital Content 1). Of the 27 included studies, more than half used an intervention of providing information to patients or clinicians.

Figure.

Search Results of Randomized Experiments Reducing Overuse of Healthcare

Table 1.

Characteristics of 27 Included Randomized Experiments to Reduce Overuse of Healthcare

Characteristics	Number of studies (%)
Decade
1980s	1 (3.7)
1990s	0 (0)
2000s	3 (11)
2010s	19 (70)
2020s	4 (15)
Country
United States	14 (52)
Canada	3 (11)
Australia	5 (19)
Spain	1 (3.7)
Germany	1 (3.7)
Netherlands	1 (3.7)
Portugal	1 (3.7)
Iran	1 (3.7)
Target
Clinicians	14 (52)
Patients	13 (48)
Outcome
Hypothetical	13 (48)
Observed	14 (52)
Clinical Service Category
Imaging	5 (19)
Medications	5 (19)
Screening	4 (15)
Surgical vs. conservative management	5 (19)
High-intensity settings or use	3 (11)
Multiple categories	2 (7.4)
Other	3 (11)a
Setting
Inpatient	5 (19)
Primary care	11 (41)
Outpatient specialty care	3 (11)
Population-based	8 (30)

^aThe “Other” category included a study that assessed patient choice of hospital and two studies that did not specify the service(s)

Table 2.

Study Interventions and Cognitive and Psychological Mechanisms that are Mediators of Behavior Change

Intervention	Stated Mechanisms	Inferred Mechanisms
Provision of information	Skills or knowledge23,25,27,29,31,33,37,41, altruism33, social norms27	Knowledge21,22,24,26,28,30,32,38,46
Education	Trust34, skills or knowledge42	Knowledge35,36,40,43, self-efficacy35
Counseling	Knowledge39, confidence39	-
Nudges	-	Knowledge44
Persuasion	Social norms45	Altruism45
Penalty/Rewards	Gain in money47	Social norms47

Provision of Information about Costs

Nine studies tested sharing cost-related information.^21–29 Four of the nine studies displayed costs to clinicians in the electronic health record (EHR).^21–24 In one, hospitalist physicians were randomized to see the cost of diagnostic cardiovascular tests, with a primary outcome of change in the proportion of imaging stress tests.²¹ In another, clinicians were randomized within their practice locations to see the median prices of imaging or procedures if done within the accountable care organization or to see prices both within and outside of the accountable care organization,²² with an outcome of test ordering rates. Sedrak and colleagues randomized a set of laboratory test requests to display the Medicare allowable fees or not, and assessed the number of tests ordered per patient-day.²³ Tamblyn and colleagues randomized primary care physicians to see alerts in the EHR about patient out-of-pocket costs for antihypertensive medicines and assessed prescribing patterns.²⁴

Another three studies evaluated the presentation of costs to clinicians in hypothetical scenarios.^25,27,28 Rudy and colleagues presented charge data for diagnostic tests to internal medicine residents.²⁷ The authors assessed the appropriateness of the tests and compared individual and group decisions on test ordering attributable to the availability of charge data. Ubel and colleagues presented one of three hypothetical scenarios to primary care physicians: familiar cancer screening scenarios without cost-effectiveness information, familiar scenarios with cost-effectiveness information, and unfamiliar scenarios with cost-effectiveness information; they then assessed their screening recommendations.²⁸ Gimbel and colleagues assessed physicians’ initial imaging choice for hypothetical cases and then randomized the physicians to see costs or not and assessed changes in imaging choice.²⁵ Two studies displayed cost information to patients.^26,29 Kortlever and colleagues randomized new and returning patients with nontraumatic upper extremity diagnoses to review scenarios describing societal costs for carpal tunnel release procedures.²⁶ The authors then assessed preferences for surgery or splinting. Greene and colleagues conducted two online experiments with employees of a company.²⁹ The authors randomized participants to view comparative data on four hospitals: participants saw one of five versions with equivalent cost data but different displays of readmissions data.

Provision of Information about Risks and Benefits

Four articles tested the impact of presenting information about the risks and benefits of a medical intervention on use of overused services.^30–33 In a survey of first-year female psychology students, Phillips and colleagues examined the influence of the “cancer effect” on young women’s responses to overdiagnosis in cervical screening.³⁰ They varied the information presented, either about benefits and harms of cervical cancer screening or a fictious non-cancer screening, with or without overdiagnosis information and assessed screening intentions, perceived risks, and decisional conflict. Miller and colleagues focused on women’s decisions about birth after a caesarean section.³¹ Using a convenience sample of women who had used maternity services, they learned how three factors – selectivity of risk information (selective vs non-selective), format of risk information (absolute vs relative), and role in decision making (choice vs compliance) – affected hypothetical childbirth preference and perceptions of risk, information adequacy, and involvement. Torrens and colleagues conducted an RCT with patients recruited at their first clinical visit after being diagnosed with rotator cuff tears.³² Patients were randomly assigned to read information about benefits and risks of rotator cuff surgery and indicated whether they might accept or decline surgery in the hypothetical scenario provided. One article tested an appeal to altruism by providing information on the risks and benefits of a medical intervention on themselves and others.³³ In the study, Riggs and colleagues presented a vignette to volunteer patients about three overused healthcare services (e.g., sinus imaging).³³ In some versions of the vignette, the text included the physician appealing to patients’ sense of altruism to limit their use of these services and in the other vignettes they did not.

Relatedly, three articles tested formats for presenting information about risks and benefits on overuse of services.^34–36 Kang and colleagues used an experimental survey and tested the impact of varied presentation of content to patients, with a vignette, on their stated treatment preferences.³⁴ The authors recruited individuals just prior to advanced imaging and presented them with vignettes about renal masses; they varied whether the information about malignant potential was descriptive only or descriptive with quantitative information in a pictograph.³⁴ They assessed the patients’ concern, estimation of risk of a serious adverse outcome, and the treatment option the patient would most likely choose after viewing the information. A more recent controlled trial, by Sharma and colleagues in a teaching hospital in Australia, targeted patients visiting the emergency department by presenting information via digital posters about low back pain imaging.³⁵ The authors assessed utilization of low back pain imaging during the period in which the posters were displayed compared to not displayed and collected patient-reported outcomes in a subset of exposed patients. Finally, McCaffery and colleagues randomly assigned a community sample of women to a hypothetical scenario involving a diagnosis of ductal carcinoma in situ described as abnormal cells first and then described as pre-invasive breast cancer cells, or the two descriptions in the reverse order.³⁶ Participants’ level of concern and treatment preferences were assessed after each scenario.

Counseling

Hadizadeh-Talasaz and colleagues randomized pregnant women with a previous cesarean section to receive, in addition to routine care, a counseling session moderated by a midwife and a gynecologist based on a “three-talk model” of shared decision making (SDM).³⁷ The authors assessed decisional conflict and regret with validated scales, as well as rates of vaginal birth versus repeat cesarean section, with repeat cesarean section considered to be an overused service.

Education

Six of the included studies were trials of educating patients or clinicians in order to impact utilization of healthcare services.^38–43 Rubin and colleagues assessed the effectiveness of sending letters to staff members about the rate of inappropriate red blood cell transfusions in their institutions.³⁸ This study tested a letter only intervention and a letter plus visit intervention. The authors measured the proportion of red blood cell transfusions deemed inappropriate. This personalized education is sometimes referred to as feedback. Another feedback trial was done by Sacarny and Colleagues who randomized clinicians to receive either a placebo letter or 3 peer-comparison letters stating that their quetiapine prescribing volume was high relative to their peers and that they would be hearing from the Centers for Medicare and Medicaid Services.³⁹ They assessed prescribing practices after receipt of the letter. Another trial, targeting patients, was conducted more than 40 years ago: Vickery and colleagues randomized members of a U.S. health plan to receive self-care information delivered, at a minimum, through the mail, and maximally through a multicomponent telephone and nurse-led intervention.⁴⁰ They tested whether provision of self-care information reduced ambulatory services use, which the authors considered to be overused. Sanders and colleagues assessed the effectiveness of educating clinicians about implementing SDM. The structured program involved training general practitioners to communicate well with patients with nonspecific low back pain and offer them positive reinforcement about their treatment choices.⁴¹ Two trials by Legarè and colleagues – one a pilot and the other a larger trial – evaluated the impact of education about SDM techniques on antibiotic use in acute respiratory infections.^42,43 The program involved an online tutorial and an on-site interactive workshop for clinicians. It was expected that if clinicians learned skills of SDM in this clinical setting that antibiotic use would decrease.

Nudges

We found only a single study testing nudges to change behavior. Martins and colleagues randomized the displays of EHR menus of diagnostic and laboratory test options to family physicians.⁴⁴ The test menus were created based on evidence-based guidelines. The authors assessed the number of tests ordered among the intervention groups.

Persuasion

In one study, Strough and colleagues asked members of an online panel to either make or predict a healthcare decision for one of five recipients: themselves, an older loved one, a younger loved one, an older acquaintance, or a younger acquaintance.⁴⁵ They measured panel members’ hypothetical choices between relatively safe lower-risk treatments with a good chance of yielding mild health improvements versus high-risk treatments that offered a moderate chance of substantial health improvements. In another study, Janssen and colleagues conducted a discrete choice experiment involving participants from a national online survey panel in which they varied hypothetical patients’ life expectancy, age, quality of life, and the physician’s recommendation.⁴⁶ Based on combinations of these attributes, they asked respondents whether they would choose screening with either colonoscopy or prostate-specific antigen or mammography. These are considered examples of persuasion because they involve a process of communication to change behaviors.

Penalties and Rewards

One study by Hennig-Schmidt and colleagues tested the opportunity for profit by having medical students respond to a computerized survey as if they were practicing clinicians.⁴⁷ The authors exposed the participants to scenarios reflecting a capitated health system and a fee-for-service environment and asked them to make decisions about utilization of services under conditions where they had varying opportunities to financially profit from the decision.

Mechanisms

Guided by our flowchart (see Supplement Figure, Supplemental Digital Content 1), we considered the mechanisms by which the stimuli changed overuse behavior in these trials. We anticipated that we might see stimuli affecting patients’ or clinicians’ motivation to change their overuse of healthcare (such as by altering trust, rapport, autonomy, social norms) or that we would see stimuli affecting patients’ or clinicians’ capability to make behavioral changes (such as through improved knowledge, skills, understanding, self-efficacy, confidence, or competence). We also expected that we might see studies of stimuli targeting extrinsic motivators of clinician behavior, specifically gains in power, status, or money. We found that most of the studies did not explicitly describe the mechanisms through which the interventions should impact behavior change. Table 2 highlights the few studies that explicitly described the mechanisms.

Discussion

Interventions to reduce healthcare overuse need to elicit behavior changes in patients and clinicians. Patients seek services and clinicians order services; either behavior could be altered to reduce low-value care. This scoping review focused on interventions that target psychological and cognitive processes of patients and clinicians to change behaviors. The interventions varied by the stimulus (e.g., provision of information, or education), their target (i.e., physicians, patients), and the specific low-value clinical service (e.g., imaging, medications, cancer screening) of interest. Most interventions included in this scoping review were concentrated among a few implementation strategies: provision of information about costs or about risks and benefits. Other strategies, such as provision of rewards or penalties, use of behavioral nudges, or persuasion were minimally represented in the included studies. In keeping with behavior change frameworks, we anticipated that the tested interventions would change the motivations or capabilities of the patients or clinicians, thereby altering their perceptions of the benefits and harms of the healthcare service and result in behavior change.¹⁹ Ultimately, we found relatively few experimental studies that evaluated the mechanisms of these interventions.

To classify these interventions, we built upon work by Michie et al¹⁹ and Lewis et al.²⁰ We adopted certain intervention functions described by Mitchie et al, such as persuasion and incentivization, while we further decomposed other intervention functions from their “wheel” such as education to distinguish between important intervention characteristics. These frameworks motivated us to consider how external stimuli may alter the perception of benefits and harms of specific services and thereby change utilization. For example, patients are frequently unaware of services that are low value.¹⁰ Thus, patient education may be an appropriate intervention to improve knowledge. Prior studies have underscored the importance of communication and longitudinal relationships between patients and their primary care providers in reducing healthcare overuse,^6,10,48 suggesting that interventions that improve trust may reduce healthcare overuse. Interventions utilizing such strategies may be a direction for future research, although these may be weak interventions for behavior change.⁴⁹ The rich taxonomy for behavior change by Kok and colleagues suggests that there are many other mechanisms to reduce overuse of healthcare.⁵⁰

Although there is a rich literature describing overuse of healthcare services and the drivers of overuse, we found that there are few studies testing interventions to reduce these practices and they have not been systematically summarized. Our work supports the framework developed by Morgan and colleagues, who suggested that patient-clinician decision-making and the associated culture, attitudes, beliefs, experiences, and practice environments are the central nexus through which drivers of overuse of healthcare operate.² This scoping review supports this framework as we can describe cognitive and psychological mechanisms for which there may be some evidence and that may serve as targets for future studies.

This scoping review has limitations. First, the review was intentionally limited to studies of interventions that were conducted experimentally. While this restriction limits the generalizability of our findings, it allows us to highlight the studies that are likely to have high internal validity. Second, while we drew upon existing behavioral change frameworks and relied on extensive discussion to identify a strategy to categorize the interventions, this list of cognitive and psychological process mediators cannot be considered exhaustive. Third, because we focused on the types of interventions in these studies, we did not appraise the risk of bias in these studies nor the strength of evidence for any given intervention. Fourth, while we included studies of interventions aiming to reduce low-value care, we excluded studies examining end-of-life care and medication deprescribing due to their specialized literature and discomfort with classifying end of life care as low-value care.

There are interventions to reduce low-value care that do not target cognitive and psychological factors of clinicians and patients; there is a separate body of literature composed of experimental studies evaluating structural changes in the healthcare delivery system. These include efforts to promote high-value care through payer coverage policy or alternative payment models including bundled payment programs or implementation of accountable care organizations.⁵¹ The review of these interventions was outside the scope of this project.

Conclusion

In this scoping review, we found that there were few experiments testing interventions that directly target the psychological and cognitive processes of clinicians or patients to reduce low-value care. Most of the studies tested the provision of information to patients or clinicians, which is not uniformly an impactful intervention to change behaviors. These findings highlight the need for experimental designs that attend to the underlying mechanisms of behavior change to support more effective interventions to reduce low value care. The evidence gaps suggest that these might be trials of behavioral nudges or trials of persuasive language in the setting of a trusting patient-clinician relationship.

Data access and responsibility:

All authors had full access to all of the data in this study and take responsibility for the integrity of the data and the accuracy of the data analysis.