Predictive Analytics and Behavioral Nudges to Improve Palliative Care in Advanced Cancer

2.1 ADVANCED CANCER BURDEN Over half of patients with advanced cancer report moderate-to-severe symptom burden and poor quality of life - both of which are associated with up to 70% lower overall survival.1-3 Despite heavy symptom burden, 40% of patients with advanced cancer receive aggressive end-of-life care, including chemotherapy and lack of hospice referral close to death, that is not concordant with patient goals.4 Suboptimal symptom management, poor communication about expected treatment benefit, and lack of attention to patient goals and wishes near the end of life contribute to these gaps.5 2.2 PALLIATIVE CARE IMPROVES QUALITY OF LIFE \& SYMPTOMS Palliative care is a medical specialty focused on providing relief from the symptoms and stress of serious illnesses such as cancer and is available in inpatient, outpatient, and community-based settings.6 Outpatient palliative care is available at 98% of NCI-designated cancer centers and 63% of non-NCI centers.7 Early outpatient palliative care concurrent with cancer-directed treatment improves quality of life, reduces symptom burden, and decreases rates of aggressive end-of-life care.8,9 Since 2017, the American Society of Clinical Oncology has recommended specialty outpatient palliative care consultation for patients within 8 weeks of advanced cancer diagnosis.10 During the COVID-19 pandemic, other organizations have called for earlier palliative care to ensure that high-risk cancer care meets patients' goals.11,12 Despite such guidelines, nearly two-thirds of patients with advanced cancer do not receive palliative care prior to death.4 Delayed or missed outpatient palliative care referrals are a major contributor to aggressive end-of-life care.8 2.3 PALLIATIVE CARE RARELY USED NCCN-based indications for palliative care referral include limited prognosis and prognostic risk factors, such as uncontrolled symptoms or poor performance status.13 Better awareness of mortality risk may inform clinicians' decisions around palliative care referral and prompt goal-concordant cancer care.14 However, oncologists correctly identify only 20% of patients with advanced cancer who will die in one year and overestimate prognosis for 70% of patients.15,16 Furthermore, existing palliative care triggers ignore patient- and cancer-specific heterogeneity in important variables such as laboratories and comorbidities.17 2.4 IMPROVE SHORT-TERM MORTALITY PREDICTION Advances in electronic health record (EHR) infrastructure and predictive analytics allow accurate and automated identification of patients with cancer at risk of short-term mortality. We have trained and deployed EHR-based predictive algorithms with better performance (c-statistic \>0.80; sensitivity \>60%) than traditional prognostic aids into routine oncology practice in order to identify patients who may benefit from early palliative care and advance care planning.18,19 At Tennessee Oncology, a rules-based automated EHR algorithm based on 14 components derived from 2021 NCCN guidelines (Exhibit 1) accurately identifies patients at risk of 180-day-month mortality.20 This algorithm has been incorporated in pilot studies, and has generated weekly reports of high-risk patients who may benefit from timely palliative care referral. There is an urgent need to implement strategies based on algorithm-based triggers to increase early outpatient palliative care among patients with advanced cancer. 2.5 PALLIATIVE CARE UNDERUTILIZED Two-thirds of patients with advanced cancer do not receive palliative care prior to dying. Furthermore, clinicians underutilize palliative care, usually initiating referrals only 2 months before death. Lack of standardized referral and screening criteria for outpatient palliative care contributes to underutilization. This is particularly true for Black and Hispanic populations, for whom palliative care referrals are 50% lower compared to White populations. 2.6 PALLIATIVE CARE BIASES Status quo bias, which predisposes clinicians to continue current practice even if not the optimal option, may lead to delayed or missed palliative care referrals. Additionally, optimism bias, the cognitive bias that causes clinicians to believe that their own patients are at lesser risk of negative outcomes, may cause clinicians to underestimate a patient's mortality risk, thus delaying palliative care referral. Finally, overconfidence bias, the propensity to overestimate one's desired behaviors when it is not objectively reasonable, may lead clinicians to incorrectly believe they are initiating similar or more palliative care referrals than their peers. 2.7 PALLIATIVE CARE CONSTRAINTS Despite increasing availability in tertiary cancer care settings, specialty palliative care is sparsely available in community oncology practices - where 75% of patients receive their primary oncologic care. Furthermore, while the number of patients with cancer eligible for palliative care is expected to grow by 20% in the upcoming decade, there will be a shortage of 18,000 palliative care specialty physicians, particularly in the outpatient setting. Because of these capacity constraints, it is crucial to identify scalable strategies to automatically identify high-risk patients with advanced cancer in order to initiate timely outpatient palliative care referrals. 2.8 PALLIATIVE CARE UTILIZATION IMPROVEMENTS Overcoming suboptimal clinician decision-making biases is key to increasing palliative care referrals. Principles from behavioral economics can inform "nudges" that change how clinicians receive information and make choices such as palliative care referral. Default, opt-out nudges that make the optimal choice the path of least resistance can mitigate clinicians' status quo bias. Reframing clinicians' prognoses via "triggered" identification of high-risk patients may combat optimism bias. These strategies are associated with 10-25 absolute percentage-point increases in guideline-based practices such as statin prescribing and transition from brand to generic drugs. However, to our knowledge no published randomized trials have used behavioral strategies to improve palliative care utilization in advanced cancer. Given rising demand for palliative care with constrained supply across the United States oncology care system, our contribution will be significant because it will leverage scalable automated predictive algorithms with a behaviorally informed intervention to increase palliative care utilization among high- risk patients with advanced cancer. This intervention is expected to create a feasible, adaptable, and acceptable process in a community oncology setting that increases palliative care utilization earlier in the advanced cancer disease trajectory. The main objective is to evaluate the impact of an intervention consisting of default algorithm-based referrals, compared to usual practice, on outpatient palliative care visits and quality of end-of-life care among patients with advanced cancer. The investigators hypothesize that this intervention will increase palliative care visits by 10 percentage points and decrease aggressive end-of-life utilization by 15 percentage points, relative to usual practice