Using nudges with electronic health records systems to improve advance care planning: a systematic review

Abstract

Background

Researchers are increasingly studying nudges as a theoretical approach to promote behavior change. With the advancement of technology, nudges with electronic health records (EHR) systems become a new way to encourage healthcare professionals to perform advance care planning (ACP). This review aims to summarize nudges with EHR interventions and assess their effects on ACP outcomes.

Methods

A comprehensive search was conducted in Scopus, CINAHL, Web of Science, PubMed, and Embase for randomized controlled trials (RCTs) published up to December 2024. Studies were included if they employed nudges developed within the EHR environment in healthcare setting and reported at least one objective measure of ACP. Two reviewers screened the titles, abstracts, and full texts and extracted data. Risk of bias was assessed using the Cochrane Risk of Bias Tool 2.0. A narrative synthesis was conducted for the data.

Results

Ten reports (nine studies), including 27,556 participants, met all of the inclusion criteria. Nudge interventions included both EHR-delivered nudges and non-EHR nudges. Using the MINDSPACE framework, the EHR-delivered nudges were categorized as priming (n = 10), salience/affect (n = 1), and default (n = 1). The non-EHR nudges were classified as priming (n = 5), salience/affect (n = 2), and norms and messenger (n = 1). Narrative synthesis showed consistently positive effects on ACP documentation, serious illness conversations documentation, prognosis communication, advance directive completion, and goals-of-care documentation. Most studies reported statistically significant improvements. In contrast, the effects of interventions on end-of-life outcomes were inconsistent and largely non-significant.

Conclusion

Overall, findings suggest that EHR-integrated nudge strategies may improve documentation and communication practices related to ACP in patients with serious illness. However, their impact on downstream clinical outcomes remains uncertain. Due to the limited number of studies and high heterogeneity, further research is needed to validate these findings.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12904-025-01820-4.

Introduction

Advance care planning (ACP) refers to a process that supports people, regardless of their age or medical condition, in considering and expressing their personal values, life objectives, and choices regarding future healthcare decisions [1]. Serious illness conversations (SIC), a specialized form of ACP, involve structured dialogues among clinicians, patients, and their families to clarify prognostic understanding, care priorities, and treatment objectives [2]. Unlike ACP, SIC does not center on immediate or future medical decisions but rather on understanding the impact of illnesses that significantly affect patients’ lives [3]. Goals of care conversations (GOC), in contrast, define the broader objectives of medical treatment based on a patient’s values and clinical circumstances, guiding decisions about specific interventions or their limitations [4]. These discussions are oriented toward real-time medical decisions for patients with active health conditions [3]. According to O’Shea et al., ACP establishes values and preferences, SIC clarifies prognosis and priorities, and GOC translates these insights into specific medical decisions, forming a continuum of patient-centered care discussions [5]. This suggests that ACP has broadened its scope to also include SIC and GOC, in addition to its traditional focus [6–8]. Early engagement in these conversations is associated with improved outcomes for both patients and caregivers, including increased alignment of care with patient preferences, reduced overtreatment, and enhanced emotional well-being [9–11].

Despite their well-documented benefits, implementing ACP in routine clinical practice remains challenging [12]. The unpredictable trajectory of many serious illnesses, combined with physicians’ tendency to overestimate patient survival, often delays the timely identification of individuals who would benefit from these discussions [13–15]. In addition, some clinicians may avoid initiating ACP due to concerns about diminishing hope or confronting death prematurely, potentially reinforcing patients’ optimism bias and contributing to social norms that stigmatize end-of-life conversations [16–19]. Moreover, providers’ decisions are influenced by deeply embedded habits and heuristics, shaped by time pressures, cognitive load, and default system behaviors [20–22]. These behavioral patterns can undermine efforts to change practice, regardless of knowledge or intent [22].

Behavioral economics offers promising strategies to overcome these barriers. In behavioural economics, “nudge” describes the choice architecture that converts people’s behaviour in a foreseeable manner without forbidding any choices or significantly modifying their economic incentives [23, 24]. The Institute for Government has created the MINDSPACE framework to facilitate the use of nudge approaches in policymaking in the UK. This framework outlines nine categories of interventions that are believed to exert the strongest influence on automatic cognitive processes, including messenger, incentives, norms, defaults, salience, priming, affect, commitment, and ego [25]. Recent literature has adapted this framework to nudge strategies in digital health interventions [26]. Hashemi et al. [27] further categorized nudges into two broad types based on their delivery mode: EHR-based nudges, which are incorporated into health information systems (e.g., EHR), and non-EHR-based nudges, which are delivered via other channels (e.g., letters, emails, or SMS).

The widespread adoption of EHR systems has created new opportunities for embedding nudge interventions into clinicians’ workflows [28]. Huber et al. [29] systematically reviewed interventions incorporating EHR components to improve ACP documentation and found that most were associated with increased rates of ACP documentation. However, the majority of included studies were observational. Lemon et al. [30] further evaluated the role of electronic medical records in facilitating advance directives (AD), reporting potential improvements but noting that most studies had a high risk of bias and were also observational in design. In addition, two systematic reviews highlighted the potential of behavior change strategies to enhance clinician engagement with ACP. Schichtel et al. [31] identified that reminder systems and similar interventions significantly improved ACP uptake. In a subsequent review [32], prompts/cues were significantly associated with improved clinician behavior related to ACP. Collectively, these findings suggest that while EHR tools hold promise, their effectiveness may be maximized when combined with behaviorally changed interventions. Therefore, this review aims to systematically evaluate the effectiveness of EHR-delivered nudges in promoting ACP, as tested in randomized controlled trials, identify the underlying behavioral mechanisms, and inform future digital health strategies grounded in behavioral science.

Methods

This systematic study followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) criteria [26]. PROSPERO (CRD42024500716) has registered this review.

Information sources and search strategy

Up to December 2024, each published paper was thoroughly reviewed in the following five databases: Scopus, CINAHL, Web of Science, PubMed, and Embase. T.Z. and R.Q.W. searched the five databases for the phrases “Electronic Health Records” and “Advance Care Planning” using free text terms, medical topic headings, and Boolean operations. To broaden the search, we did not include nudge as a search term. Supplementary Material 1 describes the search approach. We also checked the reference lists of the included publications for additional research.

Eligibility criteria

We outlined the inclusion criteria as follows: (1) Population: The study included adult patients aged 18 and above. Some of these patients had serious illnesses, such as advanced, incurable stage 4 diseases, and may have received palliative care. (2) Intervention: Nudge interventions via EHR to improve ACP. (3) Control: Usual care. (4) Outcomes: The study reported outcomes related to ACP, such as GOC discussions, SIC documentation, advance directives, and prognosis documentation. (5) Setting: healthcare settings. (6) Design: randomized controlled trials (RCT).

We outlined the following exclusion criteria: (1) Studies that did not use a randomized controlled trial design, including quasi-experimental studies, observational studies, qualitative studies, conference abstracts, letters, comments, reviews, and protocols. (2) Patients under the age of 18. (3) Patients are users of the electronic system. (4) Measurements do not correlate with ACP.

Selection process

All retrieved records were imported into the systematic review management platform Covidence (http://www.covidence.org). The predefined inclusion and exclusion criteria were applied in Covidence, which automatically removed duplicate records and highlighted potentially ineligible studies. T.Z. and R.Q.W. independently screened the titles and abstracts in the first round based on the predefined inclusion and exclusion criteria. Subsequently, both reviewers independently performed full-text screening in the second round. Any discrepancies between reviewers were resolved through discussion with A.T.Z.

Data collection process

Data extraction was performed using a standardized form that included the following variables: author, year of publication, country, study design, setting, the nudger, the nudged, whether the intervention was multicomponent (Y/N), description of the intervention, EHR-delivered nudges, non-EHR nudges, control condition, primary outcomes, and effect sizes with 95% confidence intervals (95% CI). Each report was manually extracted by T.Z. and J.X.W. Disagreements were resolved through discussion with A.T.Z.

Study risk of bias assessment

We assessed the methodological quality of the included studies using the Cochrane Risk of Bias Tool 2.0. T.Z. and J.X.W. examined various aspects to assess the quality of the research, including randomization bias, deviation from the intended intervention, missing outcome data, outcome measurements, and the selection of reported outcomes. The overall risk-of-bias score categorizes studies into “low,” “high,” and “some problems.” We addressed discrepancies and reached an agreement. The study risk of bias assessment was visualized using R software, with graphical outputs summarizing the domains of bias across included studies.

Synthesis methods

Given the heterogeneity in intervention types, populations, and outcome measures, a meta-analysis was not feasible. Therefore, we conducted a narrative synthesis. Studies were grouped by outcome domains, including ACP documentation, SIC documentation, GOC documentation, prognosis documentation, and completion of advance directives. We used vote counting based on the direction of effect to summarize whether individual studies favored the intervention or not. The direction of effect was determined by comparing outcome values between intervention and control groups. In addition, we extracted effect estimates and confidence intervals and assessed whether the reported effects were statistically significant (P < 0.05).

Results

Study selection

As shown in Fig. 1, a total of 2,680 records were identified through database searches, including PubMed (n = 631), CINAHL (n = 333), Embase (n = 532), Scopus (n = 744), and Web of Science (n = 441). After removing 1,377 duplicate records and 954 records automatically flagged as ineligible by screening tools, 349 records remained for title and abstract screening. Of these, 335 records were excluded for being irrelevant, and 14 full-text articles were retrieved and assessed for eligibility.

Fig. 1.

PRISMA flow diagram

Among the 14 full-text articles, 6 records were excluded for the following reasons: not randomized controlled trials (RCTs) (n = 2), control groups that included EHR nudges (n = 2), feasibility study without a control group (n = 1), and outcomes not related to ACP (n = 1). Additionally, 6 more records were identified through citation searching. After full-text screening, 2 records were included, while 4 records were excluded because they did not involve EHR-based nudges (n = 2) or had outcomes unrelated to ACP (n = 2). Ultimately, 9 studies (10 reports) were included in the review [33–41].

Study characteristics

All 10 RCTs took place in the United States. The included studies were published between 1998 and 2024. The reports were conducted in diverse healthcare settings, including medical oncology clinics (n = 3), cancer institutes (n = 2), primary care practices (n = 1), and large academic/public hospitals (n = 4). There were stepped-wedge randomized clinical trials (n = 3), randomized clinical trials (n = 4), pilot study (n = 2), and cluster randomized clinical trial (n = 1). The individuals delivering the interventions (“nudgers”) varied across studies. These included general clinicians (n = 3); specialized providers such as oncology clinicians (n = 3) and nurse navigators (n = 1); and multidisciplinary teams, which consisted of attending physicians and advanced practice clinicians (n = 2), or attending physicians and rapid response teams (n = 1). (see Table 1 for details).

Table 1.

Characteristics of included studies

Author name, year of publication, study design, country	Setting	The nudger	Number of healthcare	The nudged	Number of patients	Control group	Multicomponent strategy (yes (Y)/no (N) ); Description of intervention	EHR-delivered nudges in the intervention	Other nudges in the intervention
Gensheimer 2024, SWRCT, USA	8 medical oncology clinics: 5 disease clinics at our main campus and 3 outreach general oncology clinic	Attending Physicians and Advanced Practice Clinicians	55	Adult patients with cancer	1825	No automated emails highlighting high-priority patients and no direct involvement of care coaches. All clinicians occasionally received emails from the cancer center leadership encouraging them to have and document serious illness conversations.	Y; Clinicians received automated weekly emails highlighting high-priority patients and were asked to document prognoses for them. Care coaches contacted these patients to conduct the remainder of the conversation	Priming Nudge: An EHR-based machine learning model identified high-risk patients.	Salience/Affect Nudge: Emails highlighted patients with ≤ 2 years predicted survival and no prior documentation, to increase attention.
Curtis 2023, RCT, USA	3 US hospitals within 1 health care system, including a university, county, and community hospital	Attending physician, resident physicians, and advanced practice clinicians	Unreported	Hospitalized adults who were either ≥ 55 y of age with any of the chronic illnesses used by the Dartmouth Atlas project to study end-of-life care or ≥ 80 y of age	2512	Usual care	Y; Automated methods examined structured EHR data to identify code status, prior life-sustaining treatment orders and directives. This info was included in a 1-page Jumpstart Guide sent via secure email and pager message to the primary hospital team.	Priming Nudge: Automatedly examine prior EHR notes to identify care planning documentation, and include this information on Jumpstart Guides to inform discussions.	Priming Nudge: The 1-page Jumpstart Guide was formally sent via email to the entire primary hospital team, complete with recommended language and patient-specific information.
Manz 2023, SWRCT, USA Manz 2020, SWRCT, USA	9 medical oncology clinics (8 subspecialty oncology and 1 general oncology clinics) within a large academic health system in Pennsylvania	Oncology clinicians	156	Adult patients with cancer	20,506	Usual care	Y; Identifying high-risk patients using a machine learning algorithm to predict 6-month mortality. Weekly emails to clinicians comparing their SIC rates against peers, weekly lists of high-risk patients, and opt-out text messages to prompt SICs before encounters with high-risk patients.	Priming Nudge: EHR-based machine learning algorithm flagged high-risk patients Salience/Affect Nudge: Dashboard highlighted patients with high 180-day mortality and no recent SIC. Default Nudge: A checkbox to opt out of reminder texts is set on the dashboard, using opt-out (as opposed to opt-in) as the default.	Norms and Messenger Nudge: Weekly emails with individualized performance feedback and peer comparison graphs
Gabbard 2023, RCT, USA	Primary care practices	Nurse navigator	Unreported	Patients aged ≥ 65 years with multimorbidity plus cognitive/physical impairments and/or frailty.	759	Usual care	Y; A nurse navigator–led ACP pathway combined with a health care professional–facing EHR interface (ACPWise)	Priming Nudge: ACPWise was embedded directly within the EHR workflow to prompt clinicians to engage in ACP during patient encounters. Telephone-based ACP discussions conducted by nurse navigators were auto-populated into clinicians’ notes to pre-activate their awareness.	/
Lee 2022, RCT, USA	hospitals in an academic health care system in Seattle, Washington	Clinicians	128	Hospitalized adults aged 65 years or older with chronic life-limiting illness and markers of frailty, or aged 80 years or older.	150	Usual care	Y; Identifying seriously ill patients using a metrics program, completing a survey on care preferences and barriers, using NLP/ML to identify prior care discussions and directives, and delivering a “Jumpstart Guides” to facilitate goals-of-care discussions with the patient or family.	Priming Nudge: Use of NLP to identify care planning documentation in the EHR and inclusion of this information in a concise, personalized Jumpstart Guides.	Salience/Affect Nudge: The Jumpstart Guides summarizes patients’ survey responses on preferences and goals, which are personally meaningful. Priming Nudge: Deliver Jumpstart Guides to primary team via email, in-person if possible, and provide patient/family version.
Paladino 2019, CRCT, USA	Dana-Farber Cancer Institute	oncology clinicians	91	18 years or older, receiving ongoing oncology care at the center, the answer to the surprise question is no.	161	Usual care	Y; Combined structured communication tools (SICG), clinician training, and systematic reminders to promote serious illness conversations, supported by patient-facing materials (preconversation letters and a Family Guide) to facilitate engagement.	Priming Nudge: A structured EMR documentation template aligned with the Serious Illness Conversation Guide (SICG) served as a behavioral cue for clinicians.	Priming Nudge: Clinicians received email reminders and copies of the SICG one day prior to scheduled outpatient visits
Pollak 2019, Pilot RCT, USA	Duke Cancer Institute	hospitalists	15	(1) metastatic cancer, (2) dementia, (3) admission from a long-term care facility or skilled nursing facility, and (4) chronic illness (defined as congestive heart failure, chronic obstructive pulmonary disease, or end-stage renal disease) combined with either prior ICU stay in last six months or age > 85 years.	428	Usual care	Y; Combined electronic health record alerts (in-basket messages and chart-opening banners) prompted goals-of-care conversations, supplemented by physician communication coaching (SUPER framework with recorded encounter feedback)	Priming Nudge: EHR notifications highlighted patients meeting criteria for goals-of-care (GOC) conversations via Inbasket messages and a visible banner on the summary screen upon first chart access.	Priming Nudge: Pocket cards with the SUPER mnemonic were distributed to facilitate communication strategies.
Picker 2017, Pilot RCT, USA	Barnes-Jewish Hospital	Attending physician and rapid response teams	Unreported	Adult patients in inpatient general medicine units	206	Usual care	Y; Real-time EWS alerts triggering scripted palliative care discussions between the RRT nurse and the primary medical team, prompting structured conversations about ACP.	Priming Nudge: Real-time EWS alerts embedded in the EHR triggered immediate responses from Rapid Response Team.	Priming Nudge: Standardized scripted recommendations (developed with input from palliative care specialists) guided attending physicians in discussing ACP.
Dexter 1998, RCT, USA	an urban public hospital	Primary care physicians	147	patients aged 75 or older or aged 50–75 with serious underlying diseases	1009	No reminders	N; Computer-generated reminders that recommended discussion of one or both of two types of advance directives.	Priming Nudge: Computer-generated reminders that recommended discussion of one or both of two types of advance directives.	/

The reports targeted diverse patient populations, including oncology patients (n = 4) and non-oncology patients (n = 6). The non-oncology studies primarily focused on three populations: (1) older hospitalized adults (typically aged ≥ 65) with multiple chronic conditions, functional or cognitive impairments, and/or frailty (n = 3); (2) patients at high risk for poor outcomes, including those aged ≥ 75–80 or aged 50–75 with serious underlying illnesses, or individuals with markers of clinical decline such as dementia, prior ICU stays, or residence in long-term care facilities (n = 2); and (3) general medicine inpatients, regardless of age or diagnosis, who were hospitalized in inpatient internal medicine unit (n = 1).

Risk of bias in studies

As shown in Fig. 2, among the ten included reports, six reports were assessed as having low risk of bias, two reports were rated as having some concerns, and two studies were judged to be at high risk of bias. The main sources of bias in the reports with concerns or high risk included inadequate reporting of allocation concealment methods, lack of blinding of participants and personnel due to the behavioral nature of the intervention, and absence of outcome assessor blinding. These methodological limitations may have affected the internal validity of the findings in those reports.

Fig. 2.

Risk of Bias Assessment

Characteristics of nudge interventions

The interventions were categorized into multicomponent strategies (n = 9) and single-strategy approaches (n = 1). Based on the involvement of EHR, the interventions were further classified as EHR-delivered nudges and non-EHR nudges. Using the MINDSPACE framework, the EHR-based nudges were grouped into three categories: priming nudges (n = 10), salience/affect nudges (n = 1), and default nudges (n = 1) (see Table 1 for details).

Among the priming nudges identified in the included reports were the use of machine learning algorithms to identify high-risk patients (n = 3), automated extraction of care planning documents from EHR notes (n = 2), integration of structured documentation templates (n = 1), visible EHR alerts to prompt clinician action (n = 2), real-time EWS alerts embedded in the EHR that triggered immediate responses (n = 1), and implementation of the ACPWise tool (n = 1). In general, these EHR-based nudges were designed to present relevant patient information and deliver behavioral cues at critical decision points, increasing g the salience of ACP opportunities during clinical interactions. The salience/affect nudge within the EHR category involved a dashboard feature that highlighted patients with a high 180-day mortality risk and no recent SIC. The default nudge utilized an opt-out mechanism on the dashboard, where a checkbox allowed clinicians to opt out of reminder texts, thereby setting the default choice as participation rather than non-participation.

Non-EHR nudges, categorized using the MINDSPACE framework, included three types: priming (n = 5), salience/affect (n = 2), and norms/messenger (n = 1). Priming nudges involved tools like Jumpstart Guides (via email or in-person), email reminders with SICG copies, SUPER mnemonic pocket cards for communication, and scripted recommendations for physicians to discuss ACP. Salience/affect nudges used attention-grabbing emails highlighting high-priority patients and personalized care goals. Norms/messenger nudges involved weekly performance feedback emails with peer comparison data. These non-EHR nudges often complement EHR-based nudges to reinforce behavior change through multiple channels. The control group included usual care (n = 7), no automated emails highlighting high-priority patients and no direct involvement of care coaches (n = 1), no receiving of the electronic health record alerts or communication coaching (n = 1), and no reminders (n = 1) (see Table 1 for details).

Outcomes and measurements

We categorized ACP documentation outcomes into five types: ACP documentation (n = 4), serious illness conversation documentation (n = 3), advance directives (n = 3), goals-of-care discussions documentation (n = 3), and prognosis documentation (n = 2). Additionally, we reviewed exploratory outcomes related to end-of-life care including ICU-related outcomes (n = 3), palliative care consultations (n = 3), and hospital mortality (n = 3). A summary of results and metrics is provided in Table 2.

Table 2.

Summary of results and metrics

Author, Year	Outcome	Effect Size (95% CI)	P-value
Advance Care Planning documentation
Gensheimer, 2024	the proportion of ACP documentation at the visit level	OR 14.2 (8.9 to 22.6)	< 0.001
Manz, 2020	Advanced care planning encounters for All patients	aRD 3.0 (2.1–4.1) pp	0.001
Manz, 2020	Advanced care planning encounters for High-risk patients	aRD 11.7 (8.4–15.7) pp	< 0.001
Gabbard, 2023	ACP documentation	OR 20.7 (11.6–36.9)	< 0.001
Pollak, 2019	ACP documentation	OR 2.01 (0.96–4.21)	0.052
Advance directives
Gabbard, 2023	Advance directives	OR 3.0 (2.0-4.5)	< 0.001
Picker, 2017	Advanced directives	RR 2.4 (1.5–4.0)	< 0.001
Dexter, 1998	Advance directives	aOR 7.0 (2.9 to 17)	< 0.001
Goals-of-care discussions documentation
Curtis, 2023	documented goals-of-care discussions	aRD 4.1% (0.4–7.8%)	0.03
Lee, 2022	documented goals-of-care discussions	RR 2.67 (1.10–6.44)	0.04
Pollak, 2019	goals of care addressed in discharge summary	OR 0.97 (0.65–1.43)	0.861
Serious illness conversation documentation
Paladino, 2019	serious illness conversation documentation	96% vs. 79%	0.005
Manz, 2023	Serious Illness Conversations encounters rate for All encounters	aOR 2.09 (1.53–2.87)	< 0.001
Manz, 2023	Serious Illness Conversations encounters rate for High-risk patients	aOR 2.62 (1.84–3.72)	< 0.001
Manz, 2023	Serious Illness Conversations encounters rate for Non-high-risk patients	aOR 2.07 (1.52–2.82)	< 0.001
Manz, 2023	Serious Illness Conversations encounters rate for Deceased patients	aOR 2.44 (1.35–4.42)	< 0.001
Manz, 2020	Serious illness encounters for All patients	aRD 3.3 (2.3 to 4.5) pp	< 0.001
Manz, 2020	Serious illness encounters for High-risk patients	aRD 11.6 (8.2–15.5) pp	< 0.001
Prognosis documentation
Gensheimer, 2024	The proportion of visits with prognosis documentation	OR 4.3 (2.2 to 8.2)	< 0.001
Paladino, 2019	Documentation of discussions about prognosis	91% vs. 48%	< 0.001
ICU-related outcomes
Curtis, 2023	ICU care	aRD − 1.0% (− 4.4–2.5%)	0.58
Manz, 2023	ICU admission	aOR 1.04 (0.54 to 1.98)	0.92
Picker, 2017	ICU transfer	RR 0.5 (0.2–0.8)	0.009
Palliative care consultations
Curtis, 2023	Palliative care consultations	aRD 0.001% (− 1.6–1.8%)	0.91
Pollak, 2019	Palliative care consultations	OR 0.85 (0.51 to 1.43)	0.549
Picker, 2017	Palliative care consultations	RR 1.3 (0.5–3.6)	0.595
Hospital mortality
Manz, 2023	Hospital mortality	aOR 1.72 (0.7 to 4.24)	0.24
Pollak, 2019	Hospital mortality	OR 2.1 (0.43 to 10.22)	0.33
Picker, 2017	Hospital mortality	RR 1.2 (0.6–2.6)	0.635

Effects on ACP Documentation

Four reports evaluated ACP documentation as an outcome. Two reports assessed the proportion of patients with documented ACP, while the other two analyzed ACP documentation rate at the visit level, allowing multiple observations per patient. Across the four reports, the direction of effect was consistently positive, with all interventions associated with increased ACP documentation compared to usual care. Specifically, Gensheimer et al. [36] reported a significant increase in ACP documentation rates from 1.8% to 7.7% (P < 0.001) through a machine learning–driven intervention targeting high-risk cancer patients. Manz et al. (2020) [38] similarly observed improvements in the overall patient population (from 1.9% to 4.9%, P = 0.001). Another study using a nurse navigator–led ACP pathway with EHR interface showed a significant absolute increase (42.2% vs. 3.7%, P < 0.001) [35]. and Pollak et al. [41] found higher rates in the intervention group (12% vs. 6%), though not statistically significant (P = 0.052). Overall, three of four studies reported significant improvements, indicating that EHR-based nudge interventions effectively enhance ACP documentation.

Effects on serious illness conversations Documentation

Three reports reported SIC documentation as an outcome. In both Manz et al. (2020, 2023) [38, 39], a machine learning–based intervention incorporating behavioral nudges for clinicians significantly increased the frequency of SICs during patient encounters (P < 0.001). Paladino et al. [40] demonstrated improvements across multiple SIC-related outcomes, including frequency (96% vs. 79%, P = 0.005) and accessibility (61% vs. 11% documentation in accessible fields of the EHR, P < 0.001). Using vote counting by direction of effect, both reports showed positive effects favoring the interventions. Furthermore, all comparisons reached statistical significance (P < 0.05), supporting the conclusion that EHR-facilitated nudging interventions positively influence SIC documentation.

Effects on prognosis Documentation

Two trials reported prognosis documentation as an outcome. Gensheimer et al. [36] utilized a machine learning model with care coaches to enhance prognosis documentation, resulting in a 2.9% increase compared to 1.1% (P < 0.001). Similarly, Paladino et al. [40] improved documentation rates (91% vs. 48%, P = 0.005). Both studies showed significant improvements, highlighting that EHR-integrated nudge strategies effectively enhance prognosis documentation.

Effects on advance directives

Three reports reported completion rates of advance directives as an outcome. A nurse navigator–led ACP pathway combined with an EHR interface significantly increased the completion of legal ACP forms compared to usual care (24.3% vs. 10.0%, P < 0.001) [35]. An Early Warning System intervention targeting high-risk patients for palliative care discussions also led to significantly higher rates of advance directive documentation prior to discharge (37.1% vs. 15.4%, P < 0.001) [42]. In Dexter et al., computer-generated reminders prompting physicians to initiate advance directive discussions resulted in significantly higher completion rates (15% vs. 4%, P < 0.001) [34]. All three reports demonstrated statistically significant improvements, and the direction of effect consistently favored the interventions, indicating that structured, EHR-integrated nudge strategies positively impact the completion of advance directives.

Effects on goals-of-care discussions Documentation

Three reports evaluated the completion rates of documented goals-of-care (GOC) discussions. All reports implemented clinician- or patient-facing communication interventions designed to prompt or facilitate these discussions, though intervention components varied. The study by Curtis et al. [33] significantly increased documentation rates (34.5% vs. 30.4%, P = 0.03). Lee et al. [37] used an intervention informed by questionnaires and EHR data, resulting in a significant increase in documentation (21% vs. 8%, P = 0.04). In contrast, Pollak et al. [41], which employed EHR alerts and communication coaching for hospitalists, reported no significant effect on documentation frequency (P = 0.861). Using vote counting by direction of effect, two of the three reports demonstrated a favorable effect of the intervention on GOC documentation. By statistical significance, two reports reached P < 0.05, while one showed no effect. This suggests that EHR-based nudges can improve GOC documentation of goals-of-care discussions, though their effectiveness may vary depending on intervention design and implementation.

Effects on end-of-life care outcomes

Interventions showed inconsistent and largely non-significant effects on ICU-related outcomes, including ICU care, ICU admission, and ICU transfer. Among the included reports, only Picker et al. [42] reported a statistically significant reduction in ICU transfers (P = 0.009). All other comparisons, including those evaluating palliative care consultations and hospital mortality, reported P values greater than 0.05, suggesting no statistically significant effect of the interventions on these outcomes.

Discussion

This systematic review is the first to focus on the effectiveness of nudges delivered via EHR in promoting ACP. Given the limited number of randomized controlled trials included, the high heterogeneity, and the moderate to high risk of bias present in the four included trials, the findings of this systematic review should be interpreted with caution. We found that EHR-delivered nudges significantly increased ACP documentation. However, these nudges did not demonstrate significant improvements in end-of-life care metrics, including ICU-related outcomes, palliative care consultations, and mortality. This is consistent with the results reported by Nguyen et al. [43], which suggest that EHR nudges are effective in improving process-oriented or behavioral outcomes, such as documentation practices. However, their impact on other dimensions of health care quality remains unclear. The absence of significant improvements in end-of-life care metrics is not entirely surprising, as these interventions are primarily intended to prompt earlier ACP discussions and evaluate changes in clinician behavior, rather than directly improving patient outcomes [44]. Although EHR nudges are more effective in increasing ACP documentation, we could also be mindful that improvements in documentation or procedural compliance may not necessarily translate into meaningful changes in patient outcomes.

In this review, the EHR-delivered nudges using the MINDSPACE framework were grouped into priming nudges, salience/affect nudges, and default nudges. This meant that the types of interventions were very different, which made it challenging to combine the results and come to a clear conclusion about the most important aspects of EHR nudges. In EHR-delivered nudges, it is worth noting that three reports (covering two studies) used machine learning to proactively identify high-risk patients through mortality predictions, demonstrating potential in altering clinician behavior [36, 38, 39]. This is consistent with a review by Reddy et al., which indicates that artificial intelligence, including machine learning and natural language processing, can be used to support decision-making by supportive and palliative care clinicians, reduce manual workload [45]. However, the current models still present a “black box” problem: clinicians cannot view the rationale behind the predictions or indicate their level of agreement [15]. As the field of ACP increasingly adopts artificial intelligence models to support interventions, future efforts could focus on showing clinicians the basis for predictions and incorporating feedback mechanisms.

For the non-EHR nudges using the MINDSPACE framework, the categories were priming, salience/affect, norms, and messenger. These findings are similar to those of reviews for clinicians, which have shown that behavior change techniques are very effective in encouraging clinicians to implement ACP [32]. In addition, in the Norms and Messenger Nudge interventions, peer comparisons acted as a motivational driver by leveraging clinicians’ intrinsic competitiveness. Clinicians were provided with performance feedback on their conversation rates relative to their peers, which established a form of social norm reinforcement [46].

Furthermore, our study found that the majority of nudgers were physicians, who played a prominent role as key facilitators in the intervention. However, even when physicians know which patients should be engaged in ACP conversations, their time is often extremely limited [47]. Many physicians have reported that they lack the time to conduct ACP discussions with their patients [48]. In this review, one included study empowered care coaches to initiate conversations with patients in cooperation with physicians. Existing research indicates that structured ACP education delivered by lay health workers, when combined with clinician-led interventions, enhances oncology ACP documentation more effectively than clinician interventions alone [49]. A review of social workers’ role in ACP (n = 31 studies) revealed geographic concentration, with 26 studies conducted in the United States and the remaining five in South Korea, Singapore, and Israel [50]. It suggests that such models may be context-dependent and particularly applicable in countries like the U.S. This model still requires more in-depth research in regional and cultural contexts.

Limitations of the research

This review has several limitations. First, the EHR-based nudges varied significantly across studies, making it difficult to identify the key factors that contribute to their effectiveness or how specific designs might influence effect sizes. Second, we included only RCTs published in English, excluding cohort and quasi-experimental studies. This may have left out valuable research, limiting the generalizability of our findings.

Finally, our search strategy primarily focused on studies involving clinician-facing nudges with EHR interventions to promote ACP. Consequently, we excluded studies in which patients were the primary users of electronic systems (e.g., patient portals or mobile applications). Two prior reviews specifically examined patient-facing ACP web-based tools while excluding studies targeting healthcare professionals [51, 52]. Although this exclusion aligned with our goal of evaluating clinician workflow-embedded nudges, it may have omitted relevant studies on patient-initiated ACP behaviors.

Recommendations for clinical practice

In translating nudges into clinical practice, it is important not only to prompt earlier ACP discussions through digital systems but also to foster a longitudinal communication process throughout the continuum of care. This approach may better support sustained engagement with ACP. Although this review does not primarily focus on nudge strategies directed at patients, future interventions could be more effective by identifying patients earlier in their illness and involving them more actively in the process, rather than concentrating solely on the clinicians. In addition, clinical teams may benefit from more systematic support, including targeted training, dedicated time for discussion, and collaboration with non-physician team members.

Conclusions

Overall, findings suggest that EHR-integrated nudge strategies may improve documentation and communication practices related to ACP in patients with serious illness. However, their impacts on downstream clinical outcomes remain uncertain. Due to the limited number of studies and high heterogeneity, further research is warranted to validate these findings.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

EHR

Electronic Health Records

ACP

Advance Care Planning

RCTs

Randomized Controlled Trials

GOC

Goals of Care

SIC

Serious Illness Conversations

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-analyses

OR

Odds Ratios

CI

Confidence Intervals

aOR

adjusted Odds Ratio

aRD

adjusted Risk Difference

PP

Percentage Points

Author contributions

T.Z., M.L., R.Y.C.K. and R.L.D. conceived the study idea, designed the research framework, and supervised the overall process. T.Z. and R.Q.W. conducted the literature search and screening based on predefined inclusion and exclusion criteria.T.Z. and J.X.W. were responsible for data extraction, quality assessment, and risk of bias evaluation. R.Q.W. and A.T.Z. performed the statistical analysis and contributed to the interpretation of the results. M.L., R.Y.C.K. and R.L.D. contributed to the critical review of the manuscript, provided expert insights, and revised the manuscript. All authors approved the final version of the manuscript.

Funding

This study was supported by the National Natural Science Foundation of China [grant numbers 72274235]; the Guizhou Provincial Science and Technology Program [grant number Qiankehe Platform Talent-CXTD [2023]028]; and the Zunyi Science and Technology Program [grant number Zun Shi Ke He HZ Zi (2022) 338].

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.