Identification of Uncontrolled Symptoms in Cancer Patients Using Natural Language Processing.

CONTEXT: For patients with cancer, uncontrolled pain and other symptoms are the leading cause of unplanned hospitalizations. Early access to specialty palliative care (PC) is effective to reduce symptom burden, but more efficient approaches are needed for rapid identification and referral. Information on symptom burden largely exists in free-text notes, limiting its utility as a trigger for best practice alerts or automated referrals. OBJECTIVES: To evaluate whether natural language processing (NLP) can be used to identify uncontrolled symptoms (pain, dyspnea, or nausea/vomiting) in the electronic health record (EHR) among hospitalized cancer patients with advanced disease. METHODS: The dataset included 1,644 hospitalization encounters for cancer patients admitted from 1/2017 -6/2019. We randomly sampled 296 encounters, which included 15,580 clinical notes. We manually reviewed the notes and recorded symptom severity. The primary endpoint was an indicator for whether a symptom was labeled as "controlled" (none, mild, not reported) or as "uncontrolled" (moderate or severe). We randomly split the data into training and test sets and used the Random Forest algorithm to evaluate final model performance. RESULTS: Our models predicted presence of an uncontrolled symptom with the following performance: pain with 61% accuracy, 69% sensitivity, and 46% specificity (F1: 69.5); nausea/vomiting with 68% accuracy, 21% sensitivity, and 90% specificity (F1: 29.4); and dyspnea with 80% accuracy, 22% sensitivity, and 88% specificity (F1: 21.1). CONCLUSION: This study demonstrated initial feasibility of using NLP to identify hospitalized cancer patients with uncontrolled symptoms. Further model development is needed before these algorithms could be implemented to trigger early access to PC.

Measuring Goal-Concordant Care in Palliative Care Research.

>Goal-concordant care is a priority outcome for palliative care research, yet the field lacks consensus on optimal methods for measurement. We sought to 1) categorize methods used to measure goal-concordant care, and 2) discuss strengths and limitations of each method using empirical examples from palliative care research. We categorized measurement methods for goal-concordant care. We identified empirical examples of each method to illustrate the strengths, limitations, and applicability of each method to relevant study designs. We defined four methods used to measure goal-concordant care: 1) Patient- or Caregiver-Reported, 2) Caregiver-Reported After Death, 3) Concordance in Longitudinal Data, and 4) Population-Level Indicators. Patient or caregiver-reported goal-concordant care draws on strengths of patient-reported outcomes, and can be captured for multiple aspects of treatment; these methods are subject to recall bias or family-proxy bias. Concordance in longitudinal data is optimal when a treatment preference can be specifically and temporally linked to actual treatment; the method is limited to common life-sustaining treatment choices and validity may be affected by temporal variation between preference and treatment. Population-level indicators allow pragmatic research to include large populations; its primary limitation is the assumption that preferences held by a majority of persons should correspond to patterns of actual treatment in similar populations. Methods used to measure goal-concordant care have distinct strengths and limitations, and methods should be selected based on research question and study design. Existing methods could be improved, yet a future gold standard is unlikely to suit all research designs.

How well do documented goals-of-care discussions for patients with stage IV cancer reflect communication best practices?

BACKGROUND: Written clinical communication regarding patients' disease understanding and values may facilitate goal-concordant care, yet little is known about the quality of electronic health record (EHR) documentation. We sought to (1) describe frequency of communication best practices in EHR-documented goals-of-care discussions, and (2) assess whether templated notes improve quality of documentation. METHODS: Researchers pulled text of EHR-documented goals-of-care discussions for hospitalized patients with Stage IV cancer from admission to 60-days follow-up. Text was included when in a single encounter the clinician addressed: (a) prognosis and/or illness understanding; and (b) goals and/or treatment options. Researchers qualitatively coded text based on guidelines for communication best practices, and noted if an EHR template was used. RESULTS: Forty-two percent (206/492) of patients had EHR-documented goals-of-care discussions. Text frequently described communication of cancer progression (89%), though rarely included prognosis (22%). Text often included patients' goals and values (83%), and at least on specific treatment decision (82%). Communication about treatments was included for 98% of patients; common examples included cancer treatment (62%), hospice (62%), resuscitation (51%), or intensive care (38%). Clinicians documented making recommendations for 40% of patients. Text addressing patient emotional and spiritual concerns was uncommon (15%). Compared to free text, use of a template was associated with increased documentation of goals and values (80% vs. 61%, p < 0.01), but not other best practices. CONCLUSION: Insights from the study can be used to guide future training and research to study and improve the quality of documentation about goal of care, and its impact on goal-concordant care.

Pre-Post Evaluation of Collaborative Oncology Palliative Care for Patients With Stage IV Cancer.

CONTEXT: The Collaborative Care Model improves care processes and outcomes but has never been tested for palliative care. OBJECTIVES: To develop and evaluate a model of collaborative oncology palliative care for Stage IV cancer. METHODS: We conducted a pre-post evaluation of Collaborative Oncology Palliative Care (CO-Pal), enrolling patients with Stage IV lung, breast or genitourinary cancers and acute illness hospitalization. CO-Pal has 4 components: 1) oncologist communication skills training; 2) patient tracking; 3) palliative care needs assessment; and 4) care coordination stratified by high vs. low palliative care need. Health record reviews from hospital admission through 60 days provided data on outcomes - goals-of-care discussions (primary outcome), advance care planning, symptom treatment, specialty palliative care and hospice use, and hospital transfers. RESULTS: We enrolled 256 patients (n = 114 pre and n = 142 post-intervention); 60-day mortality was 32%. Comparing patients pre vs post-intervention, CO-Pal did not increase overall goals-of-care discussions, but did increase advance care planning (48% vs 63%, P = 0.021) and hospice use (19% vs 31%, P = 0.034). CO-Pal did not impact symptom treatment, overall treatment plans, or 60-day hospital transfers. During the intervention phase, high-need vs low-need patients had more goals-of-care discussions (60% vs. 15%, P < 0.001) and more use of specialty palliative care (64% vs 22%, P < 0.001) and hospice (44% vs 16%, P < 0.001). CONCLUSION: Collaborative oncology palliative care is efficient and feasible. While it did not increase overall goals-of-care discussions, it was effective to increase overall advance care planning and hospice use for patients with Stage IV cancer.

Mortality Risk for Patients With Stage IV Cancer and Acute Illness Hospitalization.

CONTEXT: Cancer prognosis data often come from clinical trials which exclude patients with acute illness. OBJECTIVES: For patients with Stage IV cancer and acute illness hospitalization to 1) describe predictors of 60-day mortality and 2) compare documented decision-making for survivors and decedents. METHODS: Investigators studied a consecutive prospective cohort of patients with Stage IV cancer and acute illness hospitalization. Structured health record and obituary reviews provided data on 60-day mortality (outcome), demographics, health status, and treatment; logistic regression models identified mortality predictors. RESULTS: Four hundred ninety-two patients with Stage IV cancer and acute illness hospitalization had median age of 60.2 (51% female, 38% minority race/ethnicity); 156 (32%) died within 60 days, and median survival for decedents was 28 days. Nutritional insufficiency (odds rato [OR] 1.83), serum albumin (OR 2.15 per 1.0 g/dL), and hospital days (OR 1.04) were associated with mortality; age, gender, race, cancer, and acute illness type were not predictive. On admission, 79% of patients had orders indicating Full Code. During 60-day follow-up, 42% of patients discussed goals of care. Documented goals of care discussions were more common for decedents than survivors (70% vs. 28%, P < 0.001), as were orders for do not resuscitate/do not intubate (68% vs. 24%, P < 0.001), stopping cancer-directed therapy (29% vs. 10%, P < 0.001), specialty Palliative Care (79% vs. 44%, P < 0.001), and Hospice (68% vs. 14%, P < 0.001). CONCLUSION: Acute illness hospitalization is a sentinel event in Stage IV cancer. Short-term mortality is high; nutritional decline increases risk. For patients with Stage IV cancer, acute illness hospitalization should trigger goals of care discussions.

Does Receipt of Recommended Elements of Palliative Care Precede In-Hospital Death or Hospice Referral?

CONTEXT: Palliative care aligns treatments with patients' values and improves quality of life, yet whether receipt of recommended elements of palliative care is associated with end-of-life outcomes is understudied. OBJECTIVES: To assess whether recommended elements of palliative care (pain and symptom management, goals of care, and spiritual care) precede in-hospital death and hospice referral and whether delivery by specialty palliative care affects that relationship. METHODS: We conducted structured chart reviews for decedents with late-stage cancer, dementia, and chronic kidney disease with a hospital admission during the six months preceding death. Measures included receipt of recommended elements of palliative care delivered by any clinician and specialty palliative care consult. We assessed associations between recommended elements of palliative care and in-hospital death and hospice referral using multivariable Poisson regression models. RESULTS: Of 402 decedents, 67 (16.7%) died in hospital, and 168 (41.8%) had hospice referral. Among elements of palliative care, only goals-of-care discussion was associated with in-hospital death (incidence rate ratio [IRR] 1.37; 95% CI 1.01-1.84) and hospice referral (IRR 1.85; 95% CI 1.31-2.61). Specialty palliative care consult was associated with a lower likelihood of in-hospital death (IRR 0.57; 95% CI 0.44-0.73) and a higher likelihood of hospice referral (IRR 1.45; 95% CI 1.12-1.89) compared with no consult. CONCLUSION: Goals-of-care discussions by different types of clinicians commonly precede end-of-life care in hospital or hospice. However, engagement with specialty palliative care reduced in-hospital death and increased hospice referral. Understanding the causal pathways of goals-of-care discussions may help build primary palliative care interventions to support patients near the end of life.

Elements of Palliative Care in the Last 6 Months of Life: Frequency, Predictors, and Timing.

IMPORTANCE: Persons living with serious illness often need skilled symptom management, communication, and spiritual support. Palliative care addresses these needs and may be delivered by either specialists or clinicians trained in other fields. It is important to understand core elements of palliative care to best provide patient-centered care. OBJECTIVE: To describe frequency, predictors, and timing of core elements of palliative care during the last 6 months of life. DESIGN: Retrospective chart review. SETTING: Inpatient academic medical center. PARTICIPANTS: Decedents with cancer, dementia, or chronic kidney disease (CKD) admitted during the 6 months preceding death. EXPOSURES: We identified receipt and timing of core elements of palliative care: pain and symptom management, goals of care, spiritual care; and specialty palliative care utilization; hospital encounters; demographics; and comorbid diagnoses. We ran Poisson regression models to assess whether diagnosis or hospital encounters were associated with core elements of palliative care. RESULTS: Among 402 decedents, the mean (SD) number of appropriately screened and treated symptoms was 2.9 (1.7)/10. Among 76.1% with documented goals of care, 58.0% had a primary goal of comfort; 55.0% had documented spiritual care. In multivariable models, compared with decedents with cancer, those with dementia or CKD were less likely to have pain and symptom management (respectively, 31% (incidence rate ratio [IRR], 0.69; 95% CI, 0.56-0.85) and 17% (IRR, 0.83; CI, 0.71-0.97)). There was a median of 3 days (IQR, 0-173) between transition to a goal of comfort and death, and a median of 12 days (IQR, 5-47) between hospice referral and death. CONCLUSIONS AND RELEVANCE: Although a high proportion of patients received elements of palliative care, transitions to a goal of comfort or hospice happened very near death. Palliative care delivery can be improved by systematizing existing mechanisms, including prompts for earlier goals-of-care discussion, symptom screening, and spiritual care, and by building collaboration between primary and specialty palliative care services.

Electronic Health Record Phenotypes for Identifying Patients with Late-Stage Disease: a Method for Research and Clinical Application.

BACKGROUND: Systematic identification of patients allows researchers and clinicians to test new models of care delivery. EHR phenotypes-structured algorithms based on clinical indicators from EHRs-can aid in such identification. OBJECTIVE: To develop EHR phenotypes to identify decedents with stage 4 solid-tumor cancer or stage 4-5 chronic kidney disease (CKD). DESIGN: We developed two EHR phenotypes. Each phenotype included International Classification of Diseases (ICD)-9 and ICD-10 codes. We used natural language processing (NLP) to further specify stage 4 cancer, and lab values for CKD. SUBJECTS: Decedents with cancer or CKD who had been admitted to an academic medical center in the last 6 months of life and died August 26, 2017-December 31, 2017. MAIN MEASURE: We calculated positive predictive values (PPV), false discovery rates (FDR), false negative rates (FNR), and sensitivity. Phenotypes were validated by a comparison with manual chart review. We also compared the EHR phenotype results to those admitted to the oncology and nephrology inpatient services. KEY RESULTS: The EHR phenotypes identified 271 decedents with cancer, of whom 186 had stage 4 disease; of 192 decedents with CKD, 89 had stage 4-5 disease. The EHR phenotype for stage 4 cancer had a PPV of 68.6%, FDR of 31.4%, FNR of 0.5%, and 99.5% sensitivity. The EHR phenotype for stage 4-5 CKD had a PPV of 46.4%, FDR of 53.7%, FNR of 0.0%, and 100% sensitivity. CONCLUSIONS: EHR phenotypes efficiently identified patients who died with late-stage cancer or CKD. Future EHR phenotypes can prioritize specificity over sensitivity, and incorporate stratification of high- and low-palliative care need. EHR phenotypes are a promising method for identifying patients for research and clinical purposes, including equitable distribution of specialty palliative care.

Concordance between Goals of Care and Treatment Decisions for Persons with Dementia.

BACKGROUND: Nursing home (NH) residents with dementia experience high rates of intensive treatment near the end of life. Limited research examines whether treatment is concordant with goals of care (GOC). OBJECTIVES: We analyzed data from the GOC trial to describe family decision makers' preferred GOC and perceptions of goal-concordant care for NH residents with late-stage dementia We compared subsequent treatment orders when families chose a primary goal of comfort versus other goals. DESIGN: We performed a secondary analysis of data from baseline and 9-month family decision-maker interviews and chart reviews. SETTING AND PARTICIPANTS: A total of 302 dyads of NH residents and family decision makers in 22 North Carolina NHs were enrolled. MEASUREMENTS: In baseline and follow-up interviews, families reported on their and NH staff's primary GOC, and perceived prognosis and goal-concordant care. Chart reviews provided data on treatment orders, hospital transfers, and hospice, which were compared after selection of a primary goal of comfort versus other goals. RESULTS: Family chose comfort as the primary goal for 66% of residents at baseline, and for nearly 80% by 9 months or death. At baseline, 49% perceived concordance with NH staff on the primary goal, and 69% at follow-up. In multivariate models, choice of comfort as the primary goal, versus other goals, was associated with half as many hospital transfers (0.11 vs. 0.25/90 person-days, confidence interval [-0.2 to -0.01]), but not with hospice or treatment orders. CONCLUSIONS: Most families chose comfort as the primary GOC. Further research is needed to translate this preference into comfort-focused treatment plans for late-stage dementia. Clinicaltrials.gov : NCT01565642 (3/26/12).

A Novel Screening Method to Identify Late-Stage Dementia Patients for Palliative Care Research and Practice.

CONTEXT: Investigators need novel methods for timely identification of patients with serious illness to test or implement new palliative care models. OBJECTIVES: The study's aim was to develop an electronic health record (EHR) phenotype to identify patients with late-stage dementia for a clinical trial of palliative care consultation. METHODS: We developed a computerized method to identify patients with dementia on hospital admission. Within a data warehouse derived from the hospital's EHR, we used search terms of age, admission date, and ICD-9 and ICD-10 diagnosis codes to create an EHR dementia phenotype, followed by brief medical record review to confirm late-stage dementia. We calculated positive predictive value, false discovery rate, and false negative rate of this novel screening method. RESULTS: The EHR phenotype screening method had a positive predictive value of 76.3% for dementia patients and 24.5% for late-stage dementia patients; a false discovery rate of 23.7% for dementia patients and 75.5% for late-stage dementia patients compared to physician assessment. The sensitivity of this screening method was 59.7% to identify hospitalized patients with dementia. Daily screening-including confirmatory chart reviews-averaged 20 minutes and was more feasible, efficient, and more complete than manual screening. CONCLUSION: A novel method using an EHR phenotype plus brief medical record review is effective to identify hospitalized patients with late-stage dementia. In health care systems with similar clinical data warehouses, this method may be applied to serious illness populations to improve enrollment in clinical trials of palliative care or to facilitate access to palliative care services.

Strategies to support recruitment of patients with life-limiting illness for research: the Palliative Care Research Cooperative Group.

CONTEXT: The Palliative Care Research Cooperative Group (PCRC) is the first clinical trials cooperative for palliative care in the U.S. OBJECTIVES: To describe barriers and strategies for recruitment during the inaugural PCRC clinical trial. METHODS: The parent study was a multisite randomized controlled trial enrolling adults with life expectancy anticipated to be one to six months, randomized to discontinue statins (intervention) vs. to continue on statins (control). To study recruitment best practices, we conducted semistructured interviews with 18 site principal investigators (PIs) and clinical research coordinators (CRCs) and reviewed recruitment rates. Interviews covered three topics: 1) successful strategies for recruitment, 2) barriers to recruitment, and 3) optimal roles of the PI and CRC. RESULTS: All eligible site PIs and CRCs completed interviews and provided data on statin protocol recruitment. The parent study completed recruitment of 381 patients. Site enrollment ranged from 1 to 109 participants, with an average of 25 enrolled per site. Five major barriers included difficulty locating eligible patients, severity of illness, family and provider protectiveness, seeking patients in multiple settings, and lack of resources for recruitment activities. Five effective recruitment strategies included systematic screening of patient lists, thoughtful messaging to make research relevant, flexible protocols to accommodate patients' needs, support from clinical champions, and the additional resources of a trials cooperative group. CONCLUSION: The recruitment experience from the multisite PCRC yields new insights into methods for effective recruitment to palliative care clinical trials. These results will inform training materials for the PCRC and may assist other investigators in the field.

Measuring palliative care quality for seriously ill hospitalized patients.

OBJECTIVE: Hospice and palliative care providers need ways to measure and improve care processes. We tested feasibility, usability, reliability, and validity of Prepare, Embrace, Attend, Communicate, Empower (PEACE) quality measures for palliative care. METHODS: Trained research nurses abstracted data from medical records to generate quality measures for a random sample of 460 seriously ill patients without, and 102 patients with, specialty palliative care (SPC) services. RESULTS: Patient age ranged from 16 to 99 years, 50% were women, and 24% were African American. Of 34 PEACE quality measures, 17 were feasible for hospital palliative care. Inter-rater reliability was high (κ>0.80) for all but two quality measures. Face validity was endorsed by clinical service leaders, and construct validity was established by higher scores for patients receiving SPC. Comprehensive palliative care assessment was completed for only 10% of seriously ill hospitalized patients, compared with 56% of patients with SPC (p<0.001). Patients with moderate or severe pain were more likely to have a clinical assessment with SPC (67% versus 42%, p=0.002). Patients with SPC more often received attention for their emotional and spiritual needs (64% versus 40%, p<0.001) and documentation of preferences for life-sustaining treatments (91% versus 59%, p>0.001). Usability was endorsed by service leaders, who initiated two practice improvement projects. CONCLUSION: PEACE quality measures are feasible and reliable, and may be useful to examine and improve the quality of palliative care for seriously ill hospitalized patients as well as for patients in hospice. Research is needed to test measures for actionability and responsiveness to intervention.