A structured shared decision-making approach to code status increased DNR selections and reduced decisional conflict without raising patient anxiety.

Background

In-hospital cardiac arrest has low survival and high risk of neurologic impairment. Patients often overestimate CPR benefits and choose full-code status. Whether a structured shared decision-making (SDM) approach, including prognostic information and decision aids, affects code status choices and decision quality was unknown.

Patients

• Setting: Six Swiss teaching hospitals (general medical wards).
• Participants: 2663 adult medical inpatients (mean age 68; 45% female) cared for by 206 resident physicians.
• Key exclusions: Cognitive impairment (e.g., dementia, delirium), severe mental health or hearing impairment, language barrier without interpreter, prior enrollment, and patients with very low likelihood of meaningful recovery from arrest (GO-FAR ≥14) or severe frailty (Clinical Frailty Scale ≥7).
• Baseline comorbidity: Charlson Comorbidity Index mean 4.7; NEWS2 mean ~2.

Intervention

• Resident-level training (1-hour workshop) focused on SDM for code status, including realistic CPR outcomes.
• Use of a structured SDM checklist (CLEAR: engage, learn/inform, explore preferences, assess/document, review directives) and a patient decision aid (visuals of CPR/ICU and outcome graphs).
• On-ward observation with feedback/coaching for ≥3 discussions to reinforce fidelity.

Control

• General communication workshop of equal duration (structuring information, responding to emotions).
• No SDM training, checklist, decision aids, or coached observations.

Outcome

• Primary: Proportion choosing do-not-resuscitate (DNR) for cardiac arrest (same-day documentation).
• Key secondary: Decisional Conflict Scale (0–100; lower is better).
• Other prespecified outcomes: Preferences for mechanical ventilation and ICU in non-arrest deterioration; patient knowledge (0–6), SDM involvement (SDM-Q-9; 0–100), perceived communication quality and pressure, physician satisfaction/time; 30-day chart outcomes (DNR status, ICU admission, readmission, mortality, CPR attempts, code-status violations).

Study Design

• Pragmatic, multicenter, cluster-randomized controlled trial with randomization at the resident physician level (block randomization).
• Intention-to-treat analysis using binomial regression with log link; robust standard errors accounting for clustering; adjusted for age, sex, principal diagnosis, and center.
• ClinicalTrials.gov: NCT03872154.

Level of Evidence

Level I (pragmatic cluster-randomized controlled trial).

Follow up period

• Primary and process outcomes: at index hospitalization (same day; patient interviews within 24 hours).
• Clinical and status outcomes: 30 days after discharge (chart review).

Results

Primary outcome

• DNR selection (same day): 50.0% (685/1370) intervention vs 37.2% (481/1293) usual care; RR 1.37 (95% CI, 1.25–1.50).
• Absolute risk difference: +12.8 percentage points.
• NNT: 8 patients need the SDM approach for one additional DNR selection.

Secondary outcomes

• Decisional conflict (DCS, 0–100): Mean 14.4 vs 21.8; adjusted difference −7.06 (95% CI, −9.43 to −4.68).
  • High decisional conflict (≥25): 21.3% vs 36.7%; RR 0.65 (95% CI, 0.54–0.78).
  • Absolute risk difference: −15.4 percentage points.
  • NNT: 7 to prevent one case of high decisional conflict.
• Patient knowledge about resuscitation (0–6): 4.0 vs 3.2; adjusted difference 0.77 (95% CI, 0.59–0.94).
• Patient involvement (SDM-Q-9, 0–100): 76.6 vs 58.6; adjusted difference 17.2 (95% CI, 13.7–20.8).
• Preference for mechanical ventilation (non-arrest deterioration): 65.7% vs 72.0%; RR 0.91 (95% CI, 0.86–0.96).
  • Absolute risk difference: −6.3 percentage points.
  • NNT: 16 to prevent one patient preferring mechanical ventilation.
• Preference for ICU admission (non-arrest deterioration): 79.1% vs 82.5%; RR 0.96 (95% CI, 0.92–0.99).
  • Absolute risk difference: −3.4 percentage points.
  • NNT: 30 to prevent one patient preferring ICU admission.
• Patient-reported experience:
  • Perceived explanation quality (0–10): 7.7 vs 5.3; adjusted difference 2.3 (95% CI, 1.78–2.81).
  • Perceived pressure (0–10): 0.7 vs 0.4; adjusted difference 0.31 (95% CI, 0.13–0.48).
  • Anxiety/fear metrics: Similar between groups.
• Physician-reported outcomes: Overall satisfaction similar; more time spent on discussions in intervention arm.

30-day outcomes

• DNR documented at 30 days: 48.0% vs 36.7%; RR 1.26 (95% CI, 1.14–1.38).
  • Absolute risk difference: +11.3 percentage points.
  • NNT: 9 for one additional DNR status at 30 days.
• ICU admission, readmission, length of stay, mortality: Similar between groups.
• In-hospital CPR attempts and code-status violations: Rare and similar.

Limitations

• Conducted in six Swiss teaching hospitals; generalizability to non-teaching settings, other countries, or non-medical inpatient populations is uncertain.
• Not powered to detect differences in clinical outcomes (e.g., mortality, CPR events).
• Possible Hawthorne effect due to observed discussions for fidelity.
• No detailed content analysis of conversations to isolate mechanisms.
• Emotional outcomes assessed with brief scales; no decision-regret measure.
• Most participants were White; race/ethnicity not collected.

Funding

Swiss National Science Foundation (10001C_192850/1) and the Swiss Society of General Internal Medicine.

Citation

Becker C, Gross S, Beck K, Amacher SA, Vincent A, Mueller J, et al. A Randomized Trial of Shared Decision-Making in Code Status Discussions. NEJM Evidence. 2025;4(5). Published April 22, 2025. DOI: 10.1056/EVIDoa2400422. ClinicalTrials.gov: NCT03872154.