BE-a-PAL: A cluster-randomized trial of algorithm-based default palliative care referral among patients with advanced cancer.

Abstract

12002

Background: Patients with advanced solid malignancies often experience poor quality of life and aggressive end-of-life care. Early specialist palliative care (PC) can improve these outcomes. However, most patients do not receive a PC referral before death, with clinician inertia and difficulty identifying high-risk patients being barriers to initiating PC referrals. Methods: This was a 2-arm pragmatic cluster-randomized clinical trial. Eligible patients had stage 3 or 4 lung or non-colorectal gastrointestinal cancer. An automated electronic health record (EHR) algorithm, adapted from NCCN Palliative Care prognostic or psychosocial risk factors, assigned each patient a score from 0-20; high-risk patients with scores ≥2 (if stage III disease) or ≥1 (stage IV) were eligible. We randomized 15 clinics in a large community oncology network, stratifying randomization based on patient volume. In the intervention arm, oncologists received weekly default EHR notifications prompting specialty PC referral for high-risk patients. If oncologists did not opt out, a coordinator introduced specialty PC to patients using a standard script and offered to schedule a PC visit. In the control arm, oncologists referred to PC at their discretion. Adjusted Cox proportional hazards models with clustered standard errors assessed the primary outcome of completed PC visit at 12 weeks. Clustered logistic regression models assessed intervention impacts on change in quality of life (measured using PAL-14) from baseline to 9 weeks and intensive end-of-life care (no hospice enrollment prior to death, chemotherapy receipt within 14 days of death). To address acceptability of the intervention among clinicians, we conducted semi-structured interviews with 12 clinicians post-trial. Results: Among 562 patients (296 intervention; 266 control), mean age was 68.5, 79.5% were White, 48.8% were female, and 77.0% had lung cancer. Mean risk score was similar for intervention and control patients (3.0 vs. 3.2). In the intervention arm, 89% of clinicians allowed PC referrals and 79% of patients agreed to PC visits. Compared to control, the intervention resulted in higher rates of completed PC visits (46.6% vs. 11.3%, adjusted odds ratio 5.4, 95% CI 3.2 to 9.2). Among 179 decedents, compared to control, the intervention decreased end-of-life chemotherapy (6.5% vs. 16.1%, p = 0.06). There was no difference in quality of life or hospice among decedents. In interviews, clinicians viewed algorithm criteria as appropriate and the nurse coordinator as a resource to introduce PC to patients. Perceived barriers included staffing limitations and inappropriateness for PC due to low symptom burden or stable disease. Conclusions: In a large community oncology network, algorithm-based default PC referrals were acceptable to clinicians and led to > 3-fold increase in specialty PC and decreased end-of-life chemotherapy. Clinical trial information: NCT05590962 .