Uncovering Barriers to Screening for Distress in Patients With Cancer via Machine Learning

12002 Background: Most patients with cancer die without a documented serious illness conversation (SIC) about prognosis and goals. Interventions that increase SICs between oncology clinicians and patients may improve goal-concordant care and end-of-life outcomes. Methods: In this stepped-wedge cluster randomized trial (NCT03984773), we tested the effect of an intervention delivering machine learning-based mortality estimates with behavioral nudges to oncologists to increase SICs among patients with cancer. The clinician-focused intervention consisted of 1) weekly emails providing individual SIC performance feedback (number of SICs in the past month) and peer comparisons; 2) a list of patients scheduled for the next week with a ≥10% predicted risk of 6 month mortality by a validated machine learning prognostic algorithm, and 3) automated opt-out text prompts on the patient’s appointment day to consider an SIC. Eight medical oncology clinics were randomized to receive the intervention in a stepped-wedge fashion every four weeks for a total of 16 weeks. Medical oncology clinicians were included if they were trained to use the SIC Guide (Ariadne Labs, Boston MA). Patients were included if they had an outpatient encounter with an eligible clinician between June 17 and November 1, 2019. The primary outcome was the percent of patient encounters with a documented SIC. Intention to treat analyses adjusted for clinic and wedge fixed effects and clustered at the oncologist level. Results: The sample consisted of 78 clinicians and 14,607 patients. The mean age of patients was 61.7 years, 55.7% were female, 70.4% were white, and 19.6% were black. The percent of patient encounters with an SIC was 1.2% (106/8536) during the pre-intervention period and 4.0% (401/10,152) during the intervention period. In intention to treat adjusted analyses, the intervention led to a significant increase in SICs (adjusted odds ratio, 3.7; 95% CI, 2.5 to 5.4, P value < 0.0001). Conclusions: An intervention consisting of machine learning mortality estimates and behavioral nudges to oncology clinicians increased SICs by three-fold over 16 weeks, a significant difference.This is one of the first studies evaluating a machine learning-based behavioral intervention to improve serious illness communication in oncology. Secondary analyses (completed April 2020) will clarify whether this intervention leads to a sustained increase in SIC rates and improves goal-concordant care and end-of-life outcomes. Clinical trial information: NCT03984773 .

Download Full-text

Augmenting machine learning algorithms to predict mortality using patient-reported outcomes in oncology.

Journal of Clinical Oncology ◽

10.1200/jco.2021.39.15_suppl.1510 ◽

2021 ◽

Vol 39 (15_suppl) ◽

pp. 1510-1510

Author(s):

Ravi Bharat Parikh ◽

Jill Schnall ◽

Manqing Liu ◽

Peter Edward Gabriel ◽

Corey Chivers ◽

...

Keyword(s):

Quality Of Life ◽

Machine Learning ◽

Mortality Risk ◽

Patient Reported Outcomes ◽

Machine Learning Algorithms ◽

Measurement Information ◽

Patients With Cancer ◽

Patient Reported ◽

Mortality Table

1510 Background: Machine learning (ML) algorithms based on electronic health record (EHR) data have been shown to accurately predict mortality risk among patients with cancer, with areas under the curve (AUC) generally greater than 0.80. While patient-reported outcomes (PROs) may also predict mortality among patients with cancer, it is unclear whether routinely-collected PROs improve the predictive performance of EHR-based ML algorithms. Methods: This cohort study included 8600 patients with cancer who had an outpatient encounter at one of 18 medical oncology practices in a large academic health system between July 1st, 2019 and January 1st, 2020. 4692 (54.9%) patients completed assessments of symptoms, performance status, and quality of life from the PRO version of the Common Terminology Criteria for Adverse Events and the Patient-Reported Outcomes Measurement Information System Global v.1.2 scales. We hypothesized that ML models predicting 180-day all-cause mortality based on EHR + PRO data would improve AUC compared to ML models based on EHR data alone. We assessed univariate and adjusted associations between each PRO and 180-day mortality. To train the EHR-only model, we fit a Least Absolute Shrinkage and Selection Operator (LASSO) regression using 192 EHR demographic, comorbidity, and laboratory variables. To train the EHR + PRO model, we used a two-phase approach to fit a model using EHR data for all patients and PRO data for those who completed assessments. To test our hypothesis, we compared the bootstrapped AUC, area under the precision-recall curve (AUPRC), and sensitivity at a 20% risk threshold for both models. Results: 464 (5.4%) patients died within 180 days of the encounter. Decreased quality of life, functional status, and appetite were associated with greater 180-day mortality (Table). Compared to the EHR-only model, the EHR + PRO model significantly improved AUC (0.86 [95% CI 0.85-0.86] vs. 0.80 [95% CI 0.80-0.81]), AUPRC (0.40 [95% CI 0.37-0.42] vs. 0.30 [95% CI 0.28-0.32]), and sensitivity (0.45 [95% CI 0.42-0.48] vs. 0.33 [95% CI 0.30-0.35]). Conclusions: Routinely collected PROs augment EHR-based ML mortality risk algorithms. ML algorithms based on EHR and PRO data may facilitate earlier supportive care for patients with cancer. Association of PROs with 180-day mortality.[Table: see text]

Download Full-text

Machine Learning Approaches to Predict 6-Month Mortality Among Patients With Cancer

JAMA Network Open ◽

10.1001/jamanetworkopen.2019.15997 ◽

2019 ◽

Vol 2 (10) ◽

pp. e1915997 ◽

Cited By ~ 22

Author(s):

Ravi B. Parikh ◽

Christopher Manz ◽

Corey Chivers ◽

Susan Harkness Regli ◽

Jennifer Braun ◽

...

Keyword(s):

Machine Learning ◽

Learning Approaches ◽

Patients With Cancer

Download Full-text

Screening for Distress and Unmet Needs in Patients With Cancer: Review and Recommendations

Journal of Clinical Oncology ◽

10.1200/jco.2011.39.5509 ◽

2012 ◽

Vol 30 (11) ◽

pp. 1160-1177 ◽

Cited By ~ 298

Author(s):

Linda E. Carlson ◽

Amy Waller ◽

Alex J. Mitchell

Keyword(s):

Quality Of Life ◽

Psychometric Properties ◽

Unmet Needs ◽

Assessment Tools ◽

Screening Tools ◽

Distress Screening ◽

Screening Programs ◽

Patients With Cancer ◽

Screening For Distress

Purpose This review summarizes the need for and process of screening for distress and assessing unmet needs of patients with cancer as well as the possible benefits of implementing screening. Methods Three areas of the relevant literature were reviewed and summarized using structured literature searches: psychometric properties of commonly used distress screening tools, psychometric properties of relevant unmet needs assessment tools, and implementation of distress screening programs that assessed patient-reported outcomes (PROs). Results Distress and unmet needs are common problems in cancer settings, and programs that routinely screen for and treat distress are feasible, particularly when staff are supported and links with specialist psychosocial services exist. Many distress screening and unmet need tools have been subject to preliminary validation, but few have been compared head to head in independent centers and in different stages of cancer. Research investigating the overall effectiveness of screening for distress in terms of improved recognition and treatment of distress and associated problems is not yet conclusive, but screening seems to improve communication between patients and clinicians and may enhance psychosocial referrals. Direct effects on quality of life are uncertain, but screening may help improve discussion of quality-of-life issues. Conclusion Involving all stakeholders and frontline clinicians when planning screening for distress programs is recommended. Training frontline staff to deliver screening programs is crucial, and continuing to rigorously evaluate outcomes, including PROs, process of care, referrals, and economic costs and benefits is essential.

Download Full-text

TAVR and cancer: machine learning-augmented propensity score mortality and cost analysis in over 30 million patients

Cardio-Oncology ◽

10.1186/s40959-021-00111-0 ◽

2021 ◽

Vol 7 (1) ◽

Author(s):

Dominique J. Monlezun ◽

Logan Hostetter ◽

Prakash Balan ◽

Nicolas Palaskas ◽

Juan Lopez-Mattei ◽

...

Keyword(s):

Machine Learning ◽

Aortic Valve ◽

Propensity Score ◽

Hospital Admissions ◽

Treatment Options ◽

The United States ◽

Healthcare Research ◽

United States Department ◽

Patients With Cancer ◽

Multivariable Regression

Abstract Introduction Cardiovascular disease (CVD) and cancer are the top mortality causes globally, yet little is known about how the diagnosis of cancer affects treatment options in patients with hemodynamically compromising aortic stenosis (AS). Patients with cancer often are excluded from aortic valve replacement (AVR) trials including trials with transcatheter AVR (TAVR) and surgical AVR (SAVR). This study looks at how cancer may influence treatment options and assesses the outcome of patients with cancer who undergo SAVR or TAVR intervention. Additionally, we sought to quantitate and compare both clinical and cost outcomes for patients with and without cancer. Methods This population-based case-control study uses the most recent year available National Inpatient Sample (NIS (2016) from the United States Department of Health and Human Services’ Agency for Healthcare Research and Quality (AHRQ). Machine learning augmented propensity score adjusted multivariable regression was conducted based on the likelihood of undergoing TAVR versus medical management (MM) and TAVR versus SAVR with model optimization supported by backward propagation neural network machine learning. Results Of the 30,195,722 total hospital admissions, 39,254 (0.13%) TAVRs were performed, with significantly fewer performed in patients with versus without cancer even in those of comparable age and mortality risk (23.82% versus 76.18%, p < 0.001) despite having similar hospital and procedural mortality. Multivariable regression in patients with cancer demonstrated that mortality was similar for TAVR, MM, and SAVR, though LOS and cost was significantly lower for TAVR versus MM and comparable for TAVR versus SAVR. Patients with prostate cancer constituted the largest primary cancer among TAVR patients including those with metastatic disease. There were no significant race or geographic disparities for TAVR mortality. Discussion Comparison of aortic valve intervention in patients with and without cancer suggests that interventions are underutilized in the cancer population. This study suggests that patients with cancer including those with metastasis have similar inpatient outcomes to patients without cancer. Further, patients who have symptomatic AS and those with higher risk aortic valve disease should be offered the benefit of intervention. Modern techniques have reduced intervention-related adverse events, provided improved quality of life, and appear to be cost effective; these advantages should not necessarily be denied to patients with co-existing cancer.

Download Full-text