A Randomized Trial of Shared Decision-Making in Code Status Discussions

Becker. NEJM Evidence 2025; doi:10.1056/EVIDoa2400422

Clinical Question

• In hospitalized adults requiring code status discussions, does a shared decision-making approach (including a checklist and decision aid) compared to usual care increase the proportion of patients choosing a DNR status?

Background

• In-hospital cardiac arrest outcomes are poor (<20% survival; ~40% with neurologic deficits)
• Patients frequently overestimate benefits of CPR, resulting in unrealistic “full code” preferences
• Goals of care (GOC) conversations are essential but often poorly conducted
• SDM has potential to better align treatments with patient values but has been underutilized in GOC discussion

Design

• Pragmatic cluster-randomized controlled trial
• Residents randomized (to avoid contamination); patients assigned per resident
• Stratified by hospital site using block randomization (blocks of 4–6)
• Open-label (residents and patients aware) but outcome assessors blinded for surveys
• 25 hospitalised patients and 15 clinicians involved in trial design
• Informed consent obtained from patients usually at the time of administering follow up questionnaire
• Power calculation:
  • Based on a DNR rate of 30% in control group, at least 174 residents (cluster) and 2610 patients (15/resident) would provide 80% power with an alpha of 0.05 and an intra-cluster correlation of 0.5 to detect a 50% increase in frequency of DNR (from 30 to 45%)
• Registered at clinicaltrials.gov

Setting

• 6 Swiss teaching hospitals
• June 2019 – April 2023

Population

• Inclusion:
  • Adults ≥18 years
  • Admitted to general medical wards
  • Requiring code status discussion
  • Able to communicate in German/French/English (with interpreter if needed)
• Exclusion:
  • Cognitive impairment (e.g. dementia, delirium)
  • Severe hearing loss or language barrier
  • Predicted futility (GO-FAR ≥14 or CFS ≥7)
    • A GO-FAR of 14 or more equates to a predicted survival with minimal neurological disability of < 1.7%
• Numbers:
  • 214/220 residents agreed to participate > 8 post randomisation exclusions leading to 106 in intervention group and 100 in usual care group
  • Intervention group: 1370 patients
  • Usual care group: 1293 patients
• Comparing baseline characteristics of intervention vs. control group
  • Age: 68.1 vs 67.9
  • Female Sex: 43.8% vs 46.2%
  • Marital Status:
    • Married/in relationship: 59.3% vs 55.7%
    • Widowed: 15.1% vs 18.0%
  • Children: ~72% in both groups
  • Citizenship: Majority Swiss (~79%) in both groups
  • Religious Affiliation: Evenly distributed across groups
  • Principle diagnosis on admission: similar between groups
  • Employment: ~72% retired in both groups
  • Comorbidities:
    • Mean Charlson Comorbidity Index Score: 4.7 (both groups)
  • NEWS2: 2.0 vs 1.9
  • GO-FAR Score: –1.9 vs –2.0
  • Frailty Score: 3.6 vs 3.6
  • Residents:
    • Age: 30.6 vs 31.0
    • Female: 50.9% vs 55.0%
    • Primary Language German: 86.7% vs 74.0%
    • Work Experience: 3.4 vs 3.7 years

Intervention

• Provision of a decision aid and checklist
  • The decision aid provided images of a mechanically ventilated patient in ICU, a patient receiving CPR and graph depicting outcomes of CPR
  • Checklist based on acronym CLEAR:
    • clinician–patient engagement; learn and inform; explore patient preferences; assess and document; and review advanced directives
• Resident Training Workshop
  • Duration: 1 hour (single session).
  • Content:
    • Didactic teaching on shared decision-making (SDM) principles.
    • Prognostic data review: CPR survival rates (<20%), neurologic outcomes (40% deficits).
    • Simulated patient practice using the CLEAR checklist.
• Supervised SDM discussions (at least 3)

Control

• 1-hour session focused on general communication (active listening, structuring)
  • It did not cover principles of shared decision making
• No provision of checklist or decision aids

Management common to both groups

• Code status discussions conducted by assigned residents
• If admitted overnight then discussion had the following day during regular working hours
• Data collected at baseline, post-discussion, and 30-day follow-up
• Follow-up surveys conducted by blinded assessor

Outcome

• Primary outcome: DNR Preference in case of cardiac arrest after discussion
  • Intervention: 685 (50%) vs Usual Care: 481 (37.2%)
  • Adjusted Risk Ratio 1.37 (95% CI 1.25 – 1.50)
• Secondary outcomes:
• Comparing intervention vs. control group
  • No significant difference in
    • Patient’s concerns or fears (disturbance caused by discussion, fear of an actual cardiac arrest or life threatening disease)
    • Length of hospital stay
    • All cause 30-day mortality
    • Physician’s overall satisfaction with code status discussion (VAS, 0 to 10): 7.7 vs 7.6
  • Significantly greater in intervention group
    • Patient’s involvement in decision making (SDM-Q-9: 0 to 100): 76.6 vs 58.6
    • Perception of being put under pressure (VAS, 0 to 10): 0.7 vs 0.4
    • Perceived transparency of discussion (VAS 0 to 10): 8.9 vs 8.5
    • 30-day DNR status: 48.0 vs 36.7%
  • Significantly less in intervention group
    • Documented preference for mechanical ventilation in case of deterioration: 65.7 vs 72.0%
    • Documented preference for ICU admission in case of deterioration: 79.1% vs 82.5%
    • Decisional Conflict Score (0 – 100): 14.4 vs 21.8
      • Adjusted Difference: −7.06 (95% CI −9.43 to −4.68)
    • Decisional Conflict Score > 25: 21.3% vs 36.7%

Authors’ Conclusions

• A structured SDM approach significantly increased DNR preferences and improved decisional quality (reduced uncertainty, improved knowledge)

Strengths

• Low rates of missing data
• Pragmatic Cluster-RCT Design
  • Real-world setting across 6 hospitals
  • Avoided contamination by randomizing residents (not patients)
• High Clinical Relevance
  • Targeted patients with meaningful CPR potential (excluded GO-FAR ≥14/CFS ≥7)
  • Measured both choices (DNR rates) and decisional quality (conflict, knowledge)
• Robust Intervention
  • CLEAR checklist + visual aids co-developed with patients
  • Fidelity checks (observed discussions + feedback)
• Supportive longer-term data (30 day) improves robustness

Weaknesses

• Generalizability Gaps
  • Swiss teaching hospitals only
    • May not apply to community/non-teaching settings or other cultures
  • Excluded non-German/English/French speakers
• Potential Biases
  • Open-label design (patients/residents knew their group)
  • Hawthorne effect (observed residents may have altered behaviour)
• Visual Aid Limitations
  • Used sanitised images
  • Unclear if more graphic visuals would improve decisions
• Long-Term Impact Unknown
  • No data on actual CPR attempts/survival
• Short time frame (24 hours) between discussion and questionnaire
  • Unclear if patients had a longer time to reflect then these may be different
• Conducted over 3 weeks
  • Unclear if performance by residents would continue when temporally more removed from training session

The Bottom Line

• An approach for shared decision-making that included the discussion of expected outcomes had a significant influence on the code status of medical patients, with a higher preference for DNR code status, and was associated with less early uncertainty around the decision
• This model is adaptable for ICU and ward use
  • A training workshop and visual aids may improve shared decision making discussions for both patients and clinicians – particularly for new clinicians navigating challenging GOC discussions

External Links

• article A Randomized Trial of Shared Decision-Making in Code Status Discussions
• editorial Goals-of-Care Conversations — What Is the Goal?
• further reading: Becker et al  Association of Communication Interventions to Discuss Code Status With Patient Decisions for Do-Not-Resuscitate Orders
• further reading: Jackson et al Navigating and Communicating about Serious Illness and End of Life
• further reading: Aronson Beyond Code Status

Metadata

Summary author: Conor McDonald
Summary date: 15 June 2025
Peer-review editor: George Walker

Picture by: L Bauer / Pexels