Estimating 1-Year Mortality for High-Risk Primary Care Patients Using the "Surprise" Question

The Physician Surprise Question in the Emergency Department: prospective cohort study

OBJECTIVES

This study aims to test the ability of the surprise question (SQ), when asked to emergency physicians (EPs), to predict in-hospital mortality among adults admitted to an emergency room (ER).

METHODS

This prospective cohort study at an academic medical centre included consecutive patients 18 years or older who received care in the ER and were subsequently admitted to the hospital from 20 April 2018 to 20 October 2018. EPs were required to answer the SQ for all patients who were being admitted to hospital. The primary outcome was in-hospital mortality.

RESULTS

The cohort included 725 adults (mean (SD) age, 60 (17) years, 51% men) from 58 128 emergency department (ED) visits. The mortality rates were 20.6% for 30-day all-cause in-hospital mortality and 23.6% for in-hospital mortality. The diagnostic test characteristics of the SQ have a sensitivity of 53.7% and specificity of 87.1%, and a relative risk of 4.02 (95% CI 3.15 to 5.13), p<0.01). The positive and negative predictive values were 57% and 86%, respectively; the positive likelihood ratio was 4.1 and negative likelihood ratio was 0.53; and the accuracy was 79.2%.

CONCLUSIONS

We found that asking the SQ to EPs may be a useful tool to identify patients in the ED with a high risk of in-hospital mortality.

Would you be surprised? Prospective multicenter study of the Surprise Question as a screening tool to predict mortality in trauma patients

BACKGROUND

The Surprise Question (SQ) ("Would I be surprised if the patient died within the next year?") is a validated tool used to identify patients with limited life expectancy. Because it may have potential to expedite palliative care interventions per American College of Surgeons Trauma Quality Improvement Program Palliative Care Best Practices Guidelines, we sought to determine if trauma team members could use the SQ to accurately predict 1-year mortality in trauma patients.

METHODS

A multicenter, prospective, cohort study collected data (August 2020 to February 2021) on trauma team members' responses to the SQ at 24 hours from admission. One-year mortality was obtained via social security death index records. Positive/negative predictive values and accuracy were calculated overall, by provider role and by patient age.

RESULTS

Ten Level I/II centers enrolled 1,172 patients (87.9% blunt). The median age was 57 years (interquartile range, 36-74 years), and the median Injury Severity Score was 10 (interquartile range, 5-14 years). Overall 1-year mortality was 13.3%. Positive predictive value was low (30.5%) regardless of role. Mortality prediction minimally improved as age increased (positive predictive value highest between 65 and 74 years old, 34.5%) but consistently trended to overprediction of death, even in younger patients.

CONCLUSION

Trauma team members' ability to forecast 1-year mortality using the SQ at 24 hours appears limited perhaps because of overestimation of injury effects, preinjury conditions, and/or team bias. This has implications for the Trauma Quality Improvement Program Guidelines and suggests that more research is needed to determine the optimal time to screen trauma patients with the SQ.

LEVEL OF EVIDENCE

Prognostic and Epidemiological; Level III.

A comparison of palliative care and rapid emergency screening (P-CaRES) tool, broad and narrow criteria, and surprise questions to predict survival of older emergency department patients

BACKGROUND

Palliative care is a form of medical care designed to enhance the quality of life of patients with life-threatening conditions. This study was conducted to compare the accuracy of predicted survival the 1 and 3-month survival rate of Broad and narrow criteria, Surprise questions (SQ), and Palliative Care and Rapid Emergency Screening (P-CaRES) after admission to the emergency department (ED).

METHODS

This prospective cohort study was conducted at an urban teaching hospital in Thailand. Patients aged ≥ 65 years admitted to the ED were classified according to their emergency severity index (ESI) (Level: 1-3). We collected data on SQ, P-CaRES, and broad and narrow criteria. A survival data of participants were collected at 1 and 3 months after admission to the ED. The survival rate was calculated using the Kaplan-Meier and log-rank tests.

RESULTS

A total of 269 patients completed the study. P-CaRES positive and P-CaRES negative patients had 1-month survival rates of 81% and 94.8%, respectively (P = 0.37), and at 3-month survival rates of 70.7% and 90.1%, respectively (P < 0.001). SQ (not surprised) had a 1-month survival rate of 79.3%, while SQ (surprised) had a 97% survival rate (P = 0.01), and SQ (not surprised) had a 75.4% survival rate at 3-months, while SQ (surprised) had a 96.3% survival rate (P = 0.01). Broad and narrow criteria that were positive and negative had 1-month survival rates of 88.1% and 92.5%, respectively (P = 0.71), while those that were positive and negative had 3-month survival rates of 78.6% and 87.2%, respectively (P = 0.19). The hazard ratio (HR) of SQ (not surprised) at 1 month was 3.22( 95%CI:1.16-8.89). The HR at 3 months of P-CaRES (positive) was 3.31 with a 95% confidence interval (CI): 1.74 - 6.27, while the HR for SQ (not surprise) was 7.33, 95% CI: 3.03-19.79; however, broad and narrow criteria had an HR of 1.78, 95% CI:0.84-3.77.

CONCLUSIONS

Among older adults who visited the ED, the SQ were good prognosis tools for predicting 1 and 3-month survival, and P-CaRES were good prognostic tools for predicting 3-month survival.

Machine Learning for Lung Cancer Diagnosis, Treatment, and Prognosis

The recent development of imaging and sequencing technologies enables systematic advances in the clinical study of lung cancer. Meanwhile, the human mind is limited in effectively handling and fully utilizing the accumulation of such enormous amounts of data. Machine learning-based approaches play a critical role in integrating and analyzing these large and complex datasets, which have extensively characterized lung cancer through the use of different perspectives from these accrued data. In this review, we provide an overview of machine learning-based approaches that strengthen the varying aspects of lung cancer diagnosis and therapy, including early detection, auxiliary diagnosis, prognosis prediction, and immunotherapy practice. Moreover, we highlight the challenges and opportunities for future applications of machine learning in lung cancer.

Who Gets (and Who Should Get) a Serious Illness Conversation in the Hospital? An Analysis of Readmission Risk Score in an Electronic Health Record

Serious Illness Conversations (SICs) explore patients' prognostic awareness, hopes, and worries, and can help establish priorities for their care during and after hospitalization. While identifying patients who benefit from an SIC remains a challenge, this task may be facilitated by use of validated prediction scores available in most commercial electronic health records (EHRs), such as Epic's Readmission Risk Score (RRS). We identified the RRS on admission for all hospital encounters from October 2018 to August 2019 and measured the area under the receiver operating characteristic (AUROC) curve to determine whether RRS could accurately discriminate post discharge 6-month mortality. For encounters with standardized SIC documentation matched in a 1:3 ratio to controls by sex and age (±5 years), we constructed a multivariable, paired logistic regression model and measured the odds of SIC documentation per every 10% absolute increase in RRS. RRS was predictive of 6-month mortality with acceptable discrimination (AUROC .71) and was significantly associated with SIC documentation (adjusted OR 1.42, 95% CI 1.24-1.63). An RRS >28% used to identify patients with post discharge 6-month mortality had a high specificity (89.0%) and negative predictive value (NPV) (97.0%), but low sensitivity (25.2%) and positive predictive value (PPV) (7.9%). RRS may serve as a practical EHR-based screen to exclude patients not requiring an SIC, thereby leaving a smaller cohort to be further evaluated for SIC needs using other validated tools and clinical assessment.

The utility of the surprise question: A useful tool for identifying patients nearing the last phase of life? A systematic review and meta-analysis

BACKGROUND

The surprise question is widely used to identify patients nearing the last phase of life. Potential differences in accuracy between timeframe, patient subgroups and type of healthcare professionals answering the surprise question have been suggested. Recent studies might give new insights.

AIM

To determine the accuracy of the surprise question in predicting death, differentiating by timeframe, patient subgroup and by type of healthcare professional.

DESIGN

Systematic review and meta-analysis.

DATA SOURCES

Electronic databases PubMed, Embase, Cochrane Library, Scopus, Web of Science and CINAHL were searched from inception till 22nd January 2021. Studies were eligible if they used the surprise question prospectively and assessed mortality. Sensitivity, specificity, negative predictive value, positive predictive value and c-statistic were calculated.

RESULTS

Fifty-nine studies met the inclusion criteria, including 88.268 assessments. The meta-analysis resulted in an estimated sensitivity of 71.4% (95% CI [66.3-76.4]) and specificity of 74.0% (95% CI [69.3-78.6]). The negative predictive value varied from 98.0% (95% CI [97.7-98.3]) to 88.6% (95% CI [87.1-90.0]) with a mortality rate of 5% and 25% respectively. The positive predictive value varied from 12.6% (95% CI [11.0-14.2]) with a mortality rate of 5% to 47.8% (95% CI [44.2-51.3]) with a mortality rate of 25%. Seven studies provided detailed information on different healthcare professionals answering the surprise question.

CONCLUSION

We found overall reasonable test characteristics for the surprise question. Additionally, this study showed notable differences in performance within patient subgroups. However, we did not find an indication of notable differences between timeframe and healthcare professionals.

Health care professionals' perceptions of factors influencing the process of identifying patients for serious illness conversations: A qualitative study

BACKGROUND

The Serious Illness Care Programme enables patients to receive care that is in accordance with their priorities. However, despite clarity about palliative care needs, many barriers to and difficulties in identifying patients for serious illness conversations remain.

AIM

To explore healthcare professionals' perceptions about factors influencing the process of identifying patients for serious illness conversations.

DESIGN

Qualitative design. A thematic analysis of observations and semi-structured interviews was used.

SETTING/PARTICIPANTS

Twelve observations at team meetings in which physicians and nurses discussed the process of identifying the patients for serious illness conversations were conducted at eight different clinics in two hospitals. Semi-structured interviews were conducted with three physicians and two nurses from five clinics.

RESULTS

Identifying the right patient and doing so at the right time were key to identifying patients for serious illness conversations. The continuity of relations and continuity over time could facilitate the identification process, while attitudes towards death and its relation to hope could hinder the process.

CONCLUSIONS

The process of identifying patients for serious illness conversations is complex and may not be captured only by generic tools such as the surprise question. It is crucial to address existential and ethical obstacles that can hinder the identification of patients for serious illness conversations.

Video Images about Decisions for Ethical Outcomes in Kidney Disease (VIDEO-KD): the study protocol for a multi-centre randomised controlled trial

INTRODUCTION

Older patients with advanced chronic kidney disease (CKD) often are inadequately prepared to make informed decisions about treatments including dialysis and cardiopulmonary resuscitation. Further, evidence shows that patients with advanced CKD do not commonly engage in advance care planning (ACP), may suffer from poor quality of life, and may be exposed to end-of-life care that is not concordant with their goals. We aim to study the effectiveness of a video intervention on ACP, treatment preferences and other patient-reported outcomes.

METHODS AND ANALYSIS

The Video Images about Decisions for Ethical Outcomes in Kidney Disease trial is a multi-centre randomised controlled trial that will test the effectiveness of an intervention that includes a CKD-related video decision aid followed by recording personal video declarations about goals of care and treatment preferences in older adults with advancing CKD. We aim to enrol 600 patients over 5 years at 10 sites.

ETHICS AND DISSEMINATION

Regulatory and ethical aspects of this trial include a single Institutional Review Board mechanism for approval, data use agreements among sites, and a Data Safety and Monitoring Board. We intend to disseminate findings at national meetings and publish our results.

TRIAL REGISTRATION NUMBER

NCT04347629.

Prognostic value of the Surprise Question for one-year mortality in older patients: a prospective multicenter study in acute geriatric and cardiology units

OBJECTIVES

To determine the prognostic value of the Surprise Question (SQ) in older persons.

METHODS

A multicenter prospective study, including patients aged 75 years or older admitted to acute geriatric (AGU) or cardiology unit (CU). The SQ was answered by the treating physician. Patients or relatives were contacted after 1 year to determine 1-year survival. Logistic regression was used to explore parameters associated with SQ. Summary ROC curves were constructed to obtain the pooled values of sensitivity and specificity based on a bivariate model.

RESULTS

The SQ was positive (death within 1 year is no surprise) in 34.7% AGU and 33.3% CU patients (p = 0.773). Parameters associated with a positive SQ were more severe comorbidity, worse functionality, significant weight loss, refractory symptoms and the request for palliative care by patient or family. One-year mortality was, respectively, 24.9% and 20.2% for patients hospitalized on AGU and CU (p = 0.319). There was no difference in sensitivity or specificity, respectively, 64% and 77% (AUC 0.635) for AGU versus 63% and 76% (AUC 0.758) for CU (p = 0.870). A positive SQ is associated with a significant shorter time until death (HR 5.425 (95% CI 3.332-8.834), p < 0.001) independently from the ward.

CONCLUSION

The Surprise Question is moderately accurate to predict 1-year mortality in older persons hospitalized on acute geriatric and cardiologic units.

How Well Does the Surprise Question Predict 1-year Mortality for Patients Admitted with COPD?

BACKGROUND

Patients with chronic obstructive pulmonary disease (COPD) often receive burdensome care at end-of-life (EOL) and infrequently complete advance care planning (ACP). The surprise question (SQ) is a prognostic tool that may facilitate ACP.

OBJECTIVE

To assess how well the SQ predicts mortality and prompts ACP for COPD patients.

DESIGN

Retrospective cohort study.

SUBJECTS

Patients admitted to the hospital for an acute exacerbation of COPD between July 2015 and September 2018.

MAIN MEASURES

Emergency department (ED) and inpatient clinicians answered, "Would you be surprised if this patient died in the next 30 days (ED)/one year (inpatient)?" The primary outcome measure was the accuracy of the SQ in predicting 30-day and 1-year mortality. The secondary outcome was the correlation between SQ and ACP (palliative care consultation, documented goals-of-care conversation, change in code status, or completion of ACP document).

KEY RESULTS

The 30-day SQ had a high specificity but low sensitivity for predicting 30-day mortality: sensitivity 12%, specificity 95%, PPV 11%, and NPV 96%. The 1-year SQ demonstrated better accuracy for predicting 1-year mortality: sensitivity 47%, specificity 75%, PPV 35%, and NPV 83%. After multivariable adjustment for age, sex, and prior 6-month admissions, 1-year SQ+ responses were associated with greater odds of 1-year mortality (OR 2.38, 95% CI 1.39-4.08) versus SQ-. One-year SQ+ patients were more likely to have a goals-of-care conversation (25% vs. 11%, p < 0.01) and complete an advance directive or POLST (46% vs. 23%, p < 0.01). After multivariable adjustment, SQ+ responses to the 1-year SQ were associated with greater odds of ACP receipt (OR 2.67, 95% CI 1.64-4.36).

CONCLUSIONS

The 1-year surprise question may be an effective component of prognostication and advance care planning for COPD patients in the inpatient setting.

Automated model versus treating physician for predicting survival time of patients with metastatic cancer

OBJECTIVE

Being able to predict a patient's life expectancy can help doctors and patients prioritize treatments and supportive care. For predicting life expectancy, physicians have been shown to outperform traditional models that use only a few predictor variables. It is possible that a machine learning model that uses many predictor variables and diverse data sources from the electronic medical record can improve on physicians' performance. For patients with metastatic cancer, we compared accuracy of life expectancy predictions by the treating physician, a machine learning model, and a traditional model.

MATERIALS AND METHODS

A machine learning model was trained using 14 600 metastatic cancer patients' data to predict each patient's distribution of survival time. Data sources included note text, laboratory values, and vital signs. From 2015-2016, 899 patients receiving radiotherapy for metastatic cancer were enrolled in a study in which their radiation oncologist estimated life expectancy. Survival predictions were also made by the machine learning model and a traditional model using only performance status. Performance was assessed with area under the curve for 1-year survival and calibration plots.

RESULTS

The radiotherapy study included 1190 treatment courses in 899 patients. A total of 879 treatment courses in 685 patients were included in this analysis. Median overall survival was 11.7 months. Physicians, machine learning model, and traditional model had area under the curve for 1-year survival of 0.72 (95% CI 0.63-0.81), 0.77 (0.73-0.81), and 0.68 (0.65-0.71), respectively.

CONCLUSIONS

The machine learning model's predictions were more accurate than those of the treating physician or a traditional model.

Disparities in Palliative Care Utilization Among Hospitalized People With Huntington Disease: A National Cross-Sectional Study

BACKGROUND

People with Huntington's disease (HD) often become institutionalized and more frequently die away from the home setting. The reasons behind differences in end-of-life care are poorly understood. Less than 5% of people with HD report utilization of palliative care (PC) or hospice services, regardless of the lack of curative therapies for this neurodegenerative disease. It is unknown what factors are associated with in-patient specialty PC consultation in this population and how PC might be related to discharge disposition.

OBJECTIVES

To determine what HD-specific (e.g., psychosis) and serious illness-specific factors (e.g., resuscitation preferences) are associated with PC encounters in people with HD and explore how PC encounters are associated with discharge disposition.

DESIGN

We analyzed factors associated with PC consultation for people with HD using discharge data from the National Inpatient Sample and the Nationwide Inpatient Sample (NIS), Healthcare Cost and Utilization Project (HCUP), Agency for Healthcare Research and Quality. An anonymized, cross-sectional, and stratified sample of 20% of United States hospitalizations from 2007 through 2014 were included using ICD-9 codes.

RESULTS

8521 patients with HD were admitted to the hospital. Of those, 321 (3.8%) received specialty PC. Payer type, (specifically private insurer or other insurer as compared to Medicare), income, (specifically the top quartile as compared to the bottom quartile), mortality risk, D.N.R., aspiration pneumonia, and depression were significantly associated with PC in a multivariate model. Among those who received PC, the odds ratio (OR) of discharge to a facility was 0.43 (95% CI, 0.32-0.58), whereas the OR of discharge to home with services was 2.25 (95% CI 1.57-3.23), even after adjusting for possible confounders.

CONCLUSIONS

Among patients with HD, economic factors, depression, and serious illness-specific factors were associated with PC, and PC was associated with discharge disposition. These findings have implications for the adaptation of inpatient PC models to meet the needs of persons with HD.

Augmented intelligence to predict 30-day mortality in patients with cancer

Aim: An augmented intelligence tool to predict short-term mortality risk among patients with cancer could help identify those in need of actionable interventions or palliative care services. Patients & methods: An algorithm to predict 30-day mortality risk was developed using socioeconomic and clinical data from patients in a large community hematology/oncology practice. Patients were scored weekly; algorithm performance was assessed using dates of death in patients' electronic health records. Results: For patients scored as highest risk for 30-day mortality, the event rate was 4.9% (vs 0.7% in patients scored as low risk; a 7.4-times greater risk). Conclusion: The development and validation of a decision tool to accurately identify patients with cancer who are at risk for short-term mortality is feasible.

Identifying patients who could benefit from palliative care by making use of the general practice information system: the Surprise Question versus the SPICT

OBJECTIVE

We compared the performance of two tools to help general practitioners (GPs) identify patients in need of palliative care: the Surprise Question (SQ) and the Supportive and Palliative Care Indicators Tool (SPICT).

METHODS

Prospective cohort study in two general practices in the Netherlands with a size of 3640 patients. At the start of the study the GPs selected patients by heart using the SQ. The SPICT was translated into a digital search in electronic patient records. The GPs then selected patients from the list thus created. Afterwards the GPs were interviewed about their experiences. The following year a record was kept of all the patients deceased in both practices. We analysed the characteristics of the patients selected and the deceased. We calculated the performance characteristics concerning predicting 1-year mortality.

RESULTS

The sensitivity of the SQ was 50%, of the SPICT 57%; the specificity 99% and 98%. When analysing the deceased (n = 36), 10 died relatively suddenly and arguably could not be identified. Leaving out these 10, the sensitivity of the SQ became 69%, of the SPICT 81%. The GPs found the performance of the digital search quite time consuming.

CONCLUSION

The SPICT seems to be better in identifying patients in need of palliative care than the SQ. It is also more time consuming than the SQ. However, as the digital search can be performed more easily after it has been done for the first time, initial investments can repay themselves.

Reducing Disparities in the Quality of Palliative Care for Older African Americans through Improved Advance Care Planning: Study Design and Protocol

Advance care planning (ACP) improves end-of-life care for patients and their caregivers. However, only one-third of adults have participated in ACP and rates are substantially lower among African Americans than among whites. Importantly, ACP improves many domains of care where there are racial disparities in outcomes, including receipt of goal-concordant care, hospice use, and provider communication. Yet, few studies have examined the effectiveness of ACP interventions among African Americans. The objectives of reducing disparities in the quality of palliative care for older African Americans through improved advance care planning (EQUAL ACP) are as follows: to compare the effectiveness of two interventions in (1) increasing ACP among African Americans and whites and (2) reducing racial disparities in both ACP and end-of-life care; and to examine whether racial concordance of the interventionist and patient is associated with ACP. EQUAL ACP is a longitudinal, multisite, cluster randomized trial and a qualitative study describing the ACP experience of participants. The study will include 800 adults ≥65 years of age (half African American and half white) from 10 primary care clinics in the South. Eligible patients have a serious illness (advanced cancer, heart failure, lung disease, etc.), disability in activities of daily living, or recent hospitalization. Patients are followed for one year and participate in either a patient-guided, self-management ACP approach, including a Five Wishes form or structured ACP with Respecting Choices First Steps. The primary outcome is formal or informal ACP-completion of advance directives, documented discussions with clinicians, and other written or verbal communication with surrogate decision makers about care preferences. Secondary outcomes assessed through after-death interviews with surrogates of patients who die during the study include receipt of goal-concordant care, health services use in the last year of life, and satisfaction with care. EQUAL ACP is the first large study to assess which strategies are most effective at both increasing rates of ACP and promoting equitable palliative care outcomes for seriously ill African Americans.

Identification of patients with potential palliative care needs: A systematic review of screening tools in primary care

BACKGROUND

Despite increasing evidence of the benefits of early access to palliative care, many patients do not receive palliative care in a timely manner. A systematic approach in primary care can facilitate earlier identification of patients with potential palliative care needs and prompt further assessment.

AIM

To identify existing screening tools for identification of patients with advanced progressive diseases who are likely to have palliative care needs in primary healthcare and evaluate their accuracy.

DESIGN

Systematic review (PROSPERO registration number CRD42019111568).

DATA SOURCES

Cochrane, MEDLINE, Embase and CINAHL were searched from inception to March 2019.

RESULTS

From 4,127 unique articles screened, 25 reported the use or development of 10 screening tools. Most tools use prediction of death and/or deterioration as a proxy for the identification of people with potential palliative care needs. The tools are based on a wide range of general and disease-specific indicators. The accuracy of five tools was assessed in eight studies; these tools differed significantly in their ability to identify patients with potential palliative care needs with sensitivity ranging from 3% to 94% and specificity ranging from 26% to 99%.

CONCLUSION

The ability of current screening tools to identify patients with advanced progressive diseases who are likely to have palliative care needs in primary care is limited. Further research is needed to identify standardised screening processes that are based not only on predicting mortality and deterioration but also on anticipating the palliative care needs and predicting the rate and course of functional decline. This would prompt a comprehensive assessment to identify and meet their needs on time.

Implementing Serious Illness Communication Processes in Primary Care: A Qualitative Study

PURPOSE

Primary care clinicians face barriers to engaging patients in conversations about prognosis, values, and goals ("serious illness conversations"). We introduced a structured, multi-component intervention, the Serious Illness Care Program (SICP), to facilitate conversations in the primary care setting. We present findings of a qualitative study to explore practical aspects of program implementation.

METHODS

We conducted semi-structured interviews of participating primary care physicians, nurse care coordinators, and social workers and coded transcripts to assess the activities used to integrate SICP into the workflow.

RESULTS

We conducted interviews with 14 of 46 clinicians from 6 primary care clinics, stopping with thematic saturation. Qualitative analysis revealed major themes around activities in the timing of the conversation (before, during, and after) and overarching insights about the program. Clinicians used a variety of strategies to adapt program components while preserving key program goals, including processes to generate accountability to ensure that conversations happen in busy clinical workflows. The interviews revealed changes to clinicians' mindset and norms, such as the recognition of the need to start conversations earlier in the illness course and the use of more expansive models of prognostic communication that address function and quality of life. Data also revealed indicators of sustainable behavior change and the spread of communication practices to patients outside the intended program scope.

CONCLUSION

SICP served as a framework for primary care clinicians to integrate serious illness communication into routine care. The shifts in processes employed by inter-professional clinicians revealed comprehensive models for prognostic communication and creative workflows to ensure that patients with complex illnesses had proactive, longitudinal, and patient-centered serious illness conversations and care planning.

A systematic intervention to improve serious illness communication in primary care: Effect on expenses at the end of life

BACKGROUND

At a population level, conversations between clinicians and seriously ill patients exploring patients' goals and values can drive high-value healthcare, improving patient outcomes and reducing spending.

METHODS

We examined the impact of a quality improvement intervention to drive better communication on total medical expenses in a high-risk care management program. We present our analysis of secondary expense outcomes from a prospective implementation trial of the Serious Illness Care Program, which includes clinician training, coaching, tools, and system interventions. We included patients who died between January 2014 and September 2016 who were selected for serious illness conversations, using the "Surprise Question," as part of implementation of the program in fourteen primary care clinics.

RESULTS

We evaluated 124 patients and observed no differences in total medical expenses between intervention and comparison clinic patients. When comparing patients in intervention clinics who did and did not have conversations, we observed lower average monthly expenses over the last 6 ($6297 vs. $8,876, p = 0.0363) and 3 months ($7263 vs. $11,406, p = 0.0237) of life for patients who had conversations.

CONCLUSIONS

Possible savings observed in this study are similar in magnitude to previous studies in advance care planning and specialty palliative care but occur earlier in the disease course and in the context of documented conversations and a comprehensive, interprofessional case management program.

IMPLICATIONS

Programs designed to drive more, earlier, and better serious illness communication hold the potential to reduce costs.

LEVEL OF EVIDENCE

Prospectively designed trial, non-randomized sample, analysis of secondary outcomes.

Derivation and validation of a prognostic model to predict mortality in patients with advanced chronic kidney disease

BACKGROUND

Guiding patients with advanced chronic kidney disease (CKD) through advance care planning about future treatment obliges an assessment of prognosis. A patient-specific integrated model to predict mortality could inform shared decision-making for patients with CKD.

METHODS

Patients with Stages 4 and 5 CKD from Massachusetts (749) and West Virginia (437) were prospectively evaluated for clinical parameters, functional status [Karnofsky Performance Score (KPS)] and their provider's response to the Surprise Question (SQ). A predictive model for 12-month mortality was derived with the Massachusetts cohort and then validated externally on the West Virginia cohort. Logistic regression was used to create the model, and the c-statistic and Hosmer-Lemeshow statistic were used to assess model discrimination and calibration, respectively.

RESULTS

In the derivation cohort, the SQ, KPS and age were most predictive of 12-month mortality with odds ratios (ORs) [95% confidence interval (CI)] of 3.29 (1.87-5.78) for a 'No' response to the SQ, 2.09 (95% CI 1.19-3.66) for fair KPS and 1.41 (95% CI 1.15-1.74) per 10-year increase in age. The c-statistic for the 12-month mortality model for the derivation cohort was 0.80 (95% CI 0.75-0.84) and for the validation cohort was 0.74 (95% CI 0.66-0.83).

CONCLUSIONS

Our integrated prognostic model for 12-month mortality in patients with advanced CKD had good discrimination and calibration. This model provides prognostic information to aid nephrologists in identifying and counseling advanced CKD patients with poor prognosis who are facing the decision to initiate dialysis or pursue medical management without dialysis.

Integrating machine-generated mortality estimates and behavioral nudges to promote serious illness conversations for cancer patients: Design and methods for a stepped-wedge cluster randomized controlled trial

INTRODUCTION

Patients with cancer often receive care that is not aligned with their personal values and goals. Serious illness conversations (SICs) between clinicians and patients can help increase a patient's understanding of their prognosis, goals and values.

METHODS AND ANALYSIS

In this study, we describe the design of a stepped-wedge cluster randomized trial to evaluate the impact of an intervention that employs machine learning-based prognostic algorithms and behavioral nudges to prompt oncologists to have SICs with patients at high risk of short-term mortality. Data are collected on documented SICs, documented advance care planning discussions, and end-of-life care utilization (emergency room and inpatient admissions, chemotherapy and hospice utilization) for patients of all enrolled clinicians.

CONCLUSION

This trial represents a novel application of machine-generated mortality predictions combined with behavioral nudges in the routine care of outpatients with cancer. Findings from the trial may inform strategies to encourage early serious illness conversations and the application of mortality risk predictions in clinical settings.

TRIAL REGISTRATION

Clinicaltrials.gov Identifier: NCT03984773.

Prognostic Indices for Advance Care Planning in Primary Care: A Scoping Review

BACKGROUND

Patient identification is an important step for advance care planning (ACP) discussions.

OBJECTIVES

We conducted a scoping review to identify prognostic indices potentially useful for initiating ACP.

METHODS

We included studies that developed and/or validated a multivariable prognostic index for all-cause mortality between 6 months and 5 years in community-dwelling adults. PubMed was searched in October 2018 for articles meeting our search criteria. If a systematic review was identified from the search, we checked for additional eligible articles in its references. We abstracted data on population studied, discrimination, calibration, where to find the index, and variables included. Each index was further assessed for clinical usability.

RESULTS

We identified 18 articles with a total of 17 unique prognostic indices after screening 9154 titles. The majority of indices (88%) had c-statistics greater than or equal to 0.70. Only 1 index was externally validated. Ten indices, 8 developed in the United States and 2 in the United Kingdom, were considered clinically usable.

CONCLUSION

Of the 17 unique prognostic indices, 10 may be useful for implementation in the primary care setting to identify patients who may benefit from ACP discussions. An index classified as "clinically usable" may not be easy to use because of a large number of variables that are not routinely collected and the need to program the index into the electronic medical record.

Medication regimen complexity on hospital admission in patients with advanced chronic conditions in need of palliative care

Objectives

To evaluate characteristics of the medication complexity, risk factors associated with high medication complexity and their clinical consequences in patients with advanced chronic conditions.

Methods

A 10-month cross-sectional study was performed in an acute-hospital care Geriatric Unit. Patients with advanced chronic conditions were identified by the NECPAL test. Medication complexity was established using the Medication Regimen Complexity Index (MRCI) tool. Demographic, pharmacological and clinical patient data were collected with the objective of determining risk factors related to high medication complexity. Measured clinical outcomes were hospital length of stay, destination on hospital discharge, in-hospital mortality and 2-year survival.

Results

Two hundred and thirty-five patients (mean age 86.8, SD 5.37; 65.5% female) were recruited. MRCI's mean score was 38 points (SD 16.54, rank: 2.00-98.50), with 57.9% of patients with high medication complexity (MRCI >35 points).

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

Importance

Machine learning algorithms could identify patients with cancer who are at risk of short-term mortality. However, it is unclear how different machine learning algorithms compare and whether they could prompt clinicians to have timely conversations about treatment and end-of-life preferences.

Objectives

To develop, validate, and compare machine learning algorithms that use structured electronic health record data before a clinic visit to predict mortality among patients with cancer.

Design, Setting, and Participants

Cohort study of 26 525 adult patients who had outpatient oncology or hematology/oncology encounters at a large academic cancer center and 10 affiliated community practices between February 1, 2016, and July 1, 2016. Patients were not required to receive cancer-directed treatment. Patients were observed for up to 500 days after the encounter. Data analysis took place between October 1, 2018, and September 1, 2019.

Exposures

Logistic regression, gradient boosting, and random forest algorithms.

Main Outcomes and Measures

Primary outcome was 180-day mortality from the index encounter; secondary outcome was 500-day mortality from the index encounter.

Results

Among 26 525 patients in the analysis, 1065 (4.0%) died within 180 days of the index encounter. Among those who died, the mean age was 67.3 (95% CI, 66.5-68.0) years, and 500 (47.0%) were women. Among those who were alive at 180 days, the mean age was 61.3 (95% CI, 61.1-61.5) years, and 15 922 (62.5%) were women. The population was randomly partitioned into training (18 567 [70.0%]) and validation (7958 [30.0%]) cohorts at the patient level, and a randomly selected encounter was included in either the training or validation set. At a prespecified alert rate of 0.02, positive predictive values were higher for the random forest (51.3%) and gradient boosting (49.4%) algorithms compared with the logistic regression algorithm (44.7%). There was no significant difference in discrimination among the random forest (area under the receiver operating characteristic curve [AUC], 0.88; 95% CI, 0.86-0.89), gradient boosting (AUC, 0.87; 95% CI, 0.85-0.89), and logistic regression (AUC, 0.86; 95% CI, 0.84-0.88) models (P for comparison = .02). In the random forest model, observed 180-day mortality was 51.3% (95% CI, 43.6%-58.8%) in the high-risk group vs 3.4% (95% CI, 3.0%-3.8%) in the low-risk group; at 500 days, observed mortality was 64.4% (95% CI, 56.7%-71.4%) in the high-risk group and 7.6% (7.0%-8.2%) in the low-risk group. In a survey of 15 oncology clinicians with a 52.1% response rate, 100 of 171 patients (58.8%) who had been flagged as having high risk by the gradient boosting algorithm were deemed appropriate for a conversation about treatment and end-of-life preferences in the upcoming week.

Conclusions and Relevance

In this cohort study, machine learning algorithms based on structured electronic health record data accurately identified patients with cancer at risk of short-term mortality. When the gradient boosting algorithm was applied in real time, clinicians believed that most patients who had been identified as having high risk were appropriate for a timely conversation about treatment and end-of-life preferences.

Association of Emergency Clinicians' Assessment of Mortality Risk With Actual 1-Month Mortality Among Older Adults Admitted to the Hospital

IMPORTANCE

The accuracy of mortality assessment by emergency clinicians is unknown and may affect subsequent medical decision-making.

OBJECTIVE

To determine the association of the question, "Would you be surprised if your patient died in the next one month?" (known as the surprise question) asked of emergency clinicians with actual 1-month mortality among undifferentiated older adults who visited the emergency department (ED).

DESIGN, SETTING, AND PARTICIPANTS

This prospective cohort study at a single academic medical center in Portland, Maine, included consecutive patients 65 years or older who received care in the ED and were subsequently admitted to the hospital from January 1, 2014, to December 31, 2015. Data analyses were conducted from January 2018 to March 2019.

EXPOSURES

Treating emergency clinicians were required to answer the surprise question, "Would you be surprised if your patient died in the next one month?" in the electronic medical record when placing a bed request for all patients who were being admitted to the hospital.

MAIN OUTCOMES AND MEASURES

The primary outcome was mortality at 1 month, assessed from the National Death Index. The secondary outcomes included accuracies of responses by both emergency clinicians and admitting internal medicine clinicians to the surprise question in identifying older patients with high 6-month and 12-month mortality.

RESULTS

The full cohort included 10 737 older adults (mean [SD] age, 75.9 [8.8] years; 5532 [52%] women; 10 157 [94.6%] white) in 16 223 visits treated in the ED and admitted to the hospital. There were 5132 patients (31.6%) with a Charlson Comorbidity Index score of 2 or more. Mortality rates were 8.3% at 1 month, 17.2% at 6 months, and 22.5% at 12 months. Emergency clinicians stated that they would not be surprised if the patient died in the next month for 2104 patients (19.6%). In multivariable analysis controlling for age, sex, race, admission diagnosis, and comorbid conditions, the odds of death at 1 month were higher in patients for whom clinicians answered that they would not be surprised if the patient died in the next 1 month compared with patients for whom clinicians answered that they would be surprised if the patient died in the next 1 month (odds ratio, 2.4 [95% CI, 2.2-2.7]; P < .001). However, the diagnostic test characteristics of the surprise question were poor (sensitivity, 20%; specificity, 93%; positive predictive value, 43%; negative predictive value, 82%; accuracy, 78%; area under the receiver operating curve of the multivariable model, 0.73 [95% CI, 0.72-0.74; P < .001]).

CONCLUSIONS AND RELEVANCE

This study found that asking the surprise question of emergency clinicians may be a valuable tool to identify older patients in the ED with a high risk of 1-month mortality. The effect of implementing the surprise question to improve population-level health care for older adults in the ED who are seriously ill remains to be seen.

Would You Be Surprised if This Patient Died in the Next 12 Months? Using the Surprise Question to Increase Palliative Care Consults From the Emergency Department

BACKGROUND

There is a growing movement to increase palliative care consults from the emergency department (ED) to reduce healthcare costs and improve quality of life. The surprise question is a screening tool that emergency medicine physicians may be able to use towards achieving this goal.

OBJECTIVE

The objectives of this study were to increase awareness of hospice and palliative care medicine (HPM) among emergency medicine (EM) providers and to evaluate whether this heightened awareness increased palliative care consults among participating emergency medicine providers.

METHODS

We conducted an anonymous convenience sample survey and two educational interventions about HPM including the surprise question among emergency medicine resident and attending physicians at a large urban public academic quaternary care center from July to November 2018. A report of palliative care consults ordered between August 1, 2017 and January 1, 2019 was generated from the electronic health records used by the hospital. The number of palliative care consults made before and after the educational intervention was compared.

RESULTS

After the first educational intervention centered on the surprise question, palliative care consults from the ED increased from an average of 2.25 per month (range 0 to 8, SD: 2.38) to 12.67 per month (range 9 to 19, SD: 4.01, p < .001).

CONCLUSION

Educating EM physicians about the surprise question can increase the number of palliative care consults from the ED, thereby potentially improving patient care and decreasing costs by avoiding unwanted healthcare interventions.

Prioritizing Primary Care Patients for a Communication Intervention Using the "Surprise Question": a Prospective Cohort Study

BACKGROUND

Communication about priorities and goals improves the value of care for patients with serious illnesses. Resource constraints necessitate targeting interventions to patients who need them most.

OBJECTIVE

To evaluate the effectiveness of a clinician screening tool to identify patients for a communication intervention.

DESIGN

Prospective cohort study.

SETTING

Primary care clinics in Boston, MA.

PARTICIPANTS

Primary care physicians (PCPs) and nurse care coordinators (RNCCs) identified patients at high risk of dying by answering the Surprise Question (SQ): "Would you be surprised if this patient died in the next 2 years?"

MEASUREMENTS

Performance of the SQ for predicting mortality, measured by the area under receiver operating curve (AUC), sensitivity, specificity, and likelihood ratios.

RESULTS

Sensitivity of PCP response to the SQ at 2 years was 79.4% and specificity 68.6%; for RNCCs, sensitivity was 52.6% and specificity 80.6%. In univariate regression, the odds of 2-year mortality for patients identified as high risk by PCPs were 8.4 times higher than those predicted to be at low risk (95% CI 5.7-12.4, AUC 0.74) and 4.6 for RNCCs (3.4-6.2, AUC 0.67). In multivariate analysis, both PCP and RNCC prediction of high risk of death remained associated with the odds of 2-year mortality.

LIMITATIONS

This study was conducted in the context of a high-risk care management program, including an initial screening process and training, both of which affect the generalizability of the results.

CONCLUSION

When used in combination with a high-risk algorithm, the 2-year version of the SQ captured the majority of patients who died, demonstrating better than expected performance as a screening tool for a serious illness communication intervention in a heterogeneous primary care population.

Development and Validation of a Deep Learning Algorithm for Mortality Prediction in Selecting Patients With Dementia for Earlier Palliative Care Interventions

Importance

Early palliative care interventions drive high-value care but currently are underused. Health care professionals face challenges in identifying patients who may benefit from palliative care.

Objective

To develop a deep learning algorithm using longitudinal electronic health records to predict mortality risk as a proxy indicator for identifying patients with dementia who may benefit from palliative care.

Design, Setting, and Participants

In this retrospective cohort study, 6-month, 1-year, and 2-year mortality prediction models with recurrent neural networks used patient demographic information and topics generated from clinical notes within Partners HealthCare System, an integrated health care delivery system in Boston, Massachusetts. This study included 26 921 adult patients with dementia who visited the health care system from January 1, 2011, through December 31, 2017. The models were trained using a data set of 24 229 patients and validated using another data set of 2692 patients. Data were analyzed from September 18, 2018, to May 15, 2019.

Main Outcomes and Measures

The area under the receiver operating characteristic curve (AUC) for 6-month and 1- and 2-year mortality prediction models and the factors contributing to the predictions.

Results

The study cohort included 26 921 patients (16 263 women [60.4%]; mean [SD] age, 74.6 [13.5] years). For the 24 229 patients in the training data set, mean (SD) age was 74.8 (13.2) years and 14 632 (60.4%) were women. For the 2692 patients in the validation data set, mean (SD) age was 75.0 (12.6) years and 1631 (60.6%) were women. The 6-month model reached an AUC of 0.978 (95% CI, 0.977-0.978); the 1-year model, 0.956 (95% CI, 0.955-0.956); and the 2-year model, 0.943 (95% CI, 0.942-0.944). The top-ranked latent topics associated with 6-month and 1- and 2-year mortality in patients with dementia include palliative and end-of-life care, cognitive function, delirium, testing of cholesterol levels, cancer, pain, use of health care services, arthritis, nutritional status, skin care, family meeting, shock, respiratory failure, and swallowing function.

Conclusions and Relevance

A deep learning algorithm based on patient demographic information and longitudinal clinical notes appeared to show promising results in predicting mortality among patients with dementia in different time frames. Further research is necessary to determine the feasibility of applying this algorithm in clinical settings for identifying unmet palliative care needs earlier.

Earlier identification of seriously ill patients: an implementation case series

OBJECTIVE

To describe the strategies used by a collection of healthcare systems to apply different methods of identifying seriously ill patients for a targeted palliative care intervention to improve communication around goals and values.

METHODS

We present an implementation case series describing the experiences, challenges and best practices in applying patient selection strategies across multiple healthcare systems implementing the Serious Illness Care Program (SICP).

RESULTS

Five sites across the USA and England described their individual experiences implementing patient selection as part of the SICP. They employed a combination of clinician screens (such as the 'Surprise Question'), disease-specific criteria, existing registries or algorithms as a starting point. Notably, each describes adaptation and evolution of their patient selection methodology over time, with several sites moving towards using more advanced machine learning-based analytical approaches.

CONCLUSIONS

Involving clinical and programme staff to choose a simple initial method for patient identification is the ideal starting place for selecting patients for palliative care interventions. However, improving and refining methods over time is important and we need ongoing research into better patient selection methodologies that move beyond mortality prediction and instead focus on identifying seriously ill patients-those with poor quality of life, worsening functional status and medical care that is negatively impacting their families.

Space for intuition - the 'Surprise'-Question in haemato-oncology: Qualitative analysis of experiences and perceptions of haemato-oncologists

BACKGROUND

Early integration of palliative care can improve outcomes for people with cancer and non-cancer diagnoses. However, prediction of survival for individuals is challenging, in particular in patients with haematological malignancies who are known to have limited access to palliative care. The 'Surprise'-Question can be used to facilitate referral to palliative care.

AIM

To explore experiences, views and perceptions of haemato-oncologists on the use of the 'Surprise'-Question in the haemato-oncology outpatients clinics of a university hospital in Germany.

DESIGN

A qualitative study using individual semi-structured interviews transcribed verbatim and analysed thematically based on the framework approach.

SETTING/PARTICIPANTS

The study took place at the haemato-oncology outpatient clinic and the bone marrow transplantation outpatient clinic of a university hospital. Nine haemato-oncologists participated in qualitative interviews.

RESULTS

Thematic analysis identified 4 themes and 11 subthemes: (1) meaning and relevance of the 'Surprise'-Question; (2) feasibility; (3) the concept of 'surprise' and (4) personal aspects of prognostication. A key function of the 'Surprise'-Question was to stimulate intuition and promote patient-centred goals of care by initiating a process of pause → reflection → change of perspective. It was easy and quick to use, but required time and communication skills to act on. Participants' training in palliative care enhanced their willingness to use the 'Surprise'-Question.

CONCLUSION

Irrespective of its use in prognostication, the 'Surprise'-Question is a valuable tool to facilitate consideration of patient-centred goals and promote holistic care in haemato-oncology. However, prognostic uncertainty, lack of time and communication skills are barriers for integration into daily practice. Further research should involve haematology patients to integrate their needs and preferences.

The Surprise Question Can Be Used to Identify Heart Failure Patients in the Emergency Department Who Would Benefit From Palliative Care

CONTEXT

Heart failure (HF) is associated with symptom exacerbations and risk of mortality after an emergency department (ED) visit. Although emergency physicians (EPs) treat symptoms of HF, often the opportunity to connect with palliative care is missed. The "surprise question" (SQ) "Would you be surprised if this patient died in the next 12 months?" is a simple tool to identify patients at risk for 12-month mortality.

OBJECTIVES

The objective of this study was to assess the accuracy of the SQ when used by EPs to assess patients with HF.

METHODS

We conducted a prospective cohort study in which clinicians applied the SQ to patients presenting to the ED with symptoms of HF. Chart review and review of death records were completed. The primary outcome was accuracy of the surprise question to predict 12-month mortality. A univariate analysis for potential predictors of 12-month mortality was performed.

RESULTS

During the study period, 199 patients were identified, and complete data were available for 97% of observations (n = 193). The one-year mortality was 29%. EPs reported that "they would not be surprised" if the patient died within the next 12 months in 53% of cases. 42.7% of these patients died within 12 months compared to 13.3% in the "would be surprised" group. There was a strong association with death in the "not surprised" group (odds ratio 4.85, 95% CI 2.34-9.98, P < 0.0001). The sensitivity, specificity, positive predictive value, and negative predictive value of the SQ were 78.6%, 56.9%, 42.7%, and 86.7%, respectively, with c-statistic = 0.68.

CONCLUSION

The SQ screening tool can assist ED providers in identifying HF patients that would benefit from early palliative care involvement.

Using Big Data and Predictive Analytics to Determine Patient Risk in Oncology

Big data and predictive analytics have immense potential to improve risk stratification, particularly in data-rich fields like oncology. This article reviews the literature published on use cases and challenges in applying predictive analytics to improve risk stratification in oncology. We characterized evidence-based use cases of predictive analytics in oncology into three distinct fields: (1) population health management, (2) radiomics, and (3) pathology. We then highlight promising future use cases of predictive analytics in clinical decision support and genomic risk stratification. We conclude by describing challenges in the future applications of big data in oncology, namely (1) difficulties in acquisition of comprehensive data and endpoints, (2) the lack of prospective validation of predictive tools, and (3) the risk of automating bias in observational datasets. If such challenges can be overcome, computational techniques for clinical risk stratification will in short order improve clinical risk stratification for patients with cancer.

Predicting one-year mortality in heart failure using the 'Surprise Question': a prospective pilot study

BACKGROUND

The Surprise Question: 'would you be surprised if this patient were to die within the next year?' has been shown to predict mortality in patients with chronic kidney disease and cancer. This prospective study aimed to determine whether the Surprise Question could identify heart failure patients with a prognosis of less than 1 year, and whether the Surprise Question can be used by different healthcare professionals.

METHODS AND RESULTS

Overall, 129 consecutive patients admitted with decompensated heart failure were included. Doctors and nurses were asked to provide a 'surprised' or 'not surprised' response to the Surprise Question for each patient. Patients were followed up until death or 1 year following study inclusion. The sensitivity, specificity, positive predictive value and negative predictive value of the Surprise Question were assessed. Cox regression was used to determine covariates significantly associated with survival. The Surprise Question showed excellent sensitivity (0.85) and negative predictive value (0.88) but only fair specificity (0.59) and positive predictive value (0.52) when asked of cardiologists. There were similar levels of accuracy between doctors and specialist nurses. The Surprise Question was significantly associated with all-cause mortality in multivariate regression analysis (hazard ratio 2.8, 95% confidence interval 1.0-7.9, P = 0.046).

CONCLUSION

This study demonstrates that the Surprise Question can identify heart failure patients within the last year of life. Despite over-classification of patients into the 'not surprised' category, the Surprise Question identified nearly all patients who were within the last year of life, whilst also accurately identifying those unlikely to die.

Predictive modeling of U.S. health care spending in late life

That one-quarter of Medicare spending in the United States occurs in the last year of life is commonly interpreted as waste. But this interpretation presumes knowledge of who will die and when. Here we analyze how spending is distributed by predicted mortality, based on a machine-learning model of annual mortality risk built using Medicare claims. Death is highly unpredictable. Less than 5% of spending is accounted for by individuals with predicted mortality above 50%. The simple fact that we spend more on the sick-both on those who recover and those who die-accounts for 30 to 50% of the concentration of spending on the dead. Our results suggest that spending on the ex post dead does not necessarily mean that we spend on the ex ante "hopeless."

Derivation and Validation of a Prognostic Model to Predict 6-Month Mortality in an Intensive Care Unit Population

RATIONALE

Identification of terminally ill patients in the intensive care unit (ICU) would facilitate decision making and timely palliative care.

OBJECTIVES

To develop and validate a patient-specific integrated prognostic model to predict 6-month mortality in medical ICU patients.

METHODS

A longitudinal prospective cohort study of temporally split samples of 1,049 consecutive medical ICU patients in a tertiary care hospital was performed. For each patient, we collected demographic data, Acute Physiology and Chronic Health Evaluation III score, Charlson comorbidity index, intensivist response to a surprise question (SQ; "Would I be surprised if this patient died in the next 6 months?") on admission, and vital status at 6 months.

RESULTS

Between November 2013 and May 2015, derivation and validation cohorts of 500 and 549 consecutive patients were studied to develop a multivariate logistic regression model. In the multivariate logistic regression model, Charlson comorbidity index (P = 0.033), Acute Physiology and Chronic Health Evaluation III score (P < 0.001), and SQ response (P < 0.001) were predictors of vital status at 6 months. The odds of dying within 6 months were significantly higher when the SQ was answered "no" than when it was answered "yes" (odds ratio, 7.29; P < 0.001). The c-statistic for the derivation and validation cohorts were 0.832 (95% confidence interval, 0.795-0.870) and 0.84 (95% confidence interval, 0.806-0.875), respectively.

CONCLUSIONS

Our integrated prognostic model, which includes the SQ, has strong discrimination and calibration to predict 6-month mortality in medical ICU patients. This model can aid clinicians in identifying ICU patients who may benefit from the integration of palliative care into their treatment.

Performance of the Surprise Question Compared to Prediction Models in Hemodialysis Patients: A Prospective Study

BACKGROUND

The surprise question (SQ) ("Would you be surprised if this patient were still alive in 6 or 12 months?") is used as a mortality prognostication tool in hemodialysis (HD) patients. We compared the performance of the SQ with that of prediction models (PMs) for 6- and 12-month mortality prediction.

METHODS

Demographic, clinical, laboratory, and dialysis treatment indicators were used to model 6- and 12-month mortality probability in a HD patients training cohort (n = 6,633) using generalized linear models (GLMs). A total of 10 nephrologists from 5 HD clinics responded to the SQ in 215 patients followed prospectively for 12 months. The performance of PM was evaluated in the validation (n = 6,634) and SQ cohorts (n = 215) using the areas under receiver operating characteristics curves. We compared sensitivities and specificities of PM and SQ.

RESULTS

The PM and SQ cohorts comprised 13,267 (mean age 61 years, 55% men, 54% whites) and 215 (mean age 62 years, 59% men, 50% whites) patients, respectively. During the 12-month follow-up, 1,313 patients died in the prediction model cohort and 22 in the SQ cohort. For 6-month mortality prediction, the GLM had areas under the curve of 0.77 in the validation cohort and 0.77 in the SQ cohort. As for 12-month mortality, areas under the curve were 0.77 and 0.80 in the validation and SQ cohorts, respectively. The 6- and 12-month PMs had sensitivities of 0.62 (95% CI 0.35-0.88) and 0.75 (95% CI 0.56-0.94), respectively. The 6- and 12-month SQ sensitivities were 0.23 (95% CI 0.002-0.46) and 0.35 (95% CI 0.14-0.56), respectively.

CONCLUSION

PMs exhibit superior sensitivity compared to the SQ for mortality prognostication in HD patients.

The "Surprise Question" Asked of Emergency Physicians May Predict 12-Month Mortality among Older Emergency Department Patients

BACKGROUND

Identification of older adults with serious illness (life expectancy less than one year) who may benefit from serious illness conversations or other palliative care interventions in the emergency department (ED) is difficult.

OBJECTIVES

To assess the performance of the "surprise question (SQ)" asked of emergency physicians to predict 12-month mortality.

DESIGN

We asked attending emergency physician "Would you be surprised whether this patient died in the next 12 months?" regarding patients ≥65 years old that they had cared for that shift. We prospectively obtained death records from Massachusetts Department of Health Vital Records.

SETTING

An urban, university-affiliated ED.

MEASUREMENT

Twelve-month mortality.

RESULTS

We approached 38 physicians to answer the SQ, and 86% participated. The mean age of our cohort was 76 years, 51% were male, and 45% had at least one serious illness. Out of 207 patients, the physicians stated that they "would not be surprised" if the patient died in the next 12 months for 102 of the patients (49%); 44 of the 207 patients (21%) died within 12 months. The SQ demonstrated sensitivity of 77%, specificity of 56%, positive predictive value of 32%, and negative predictive value of 90%. When combined with other predictors, the model sorted the patient who lived from the patient who died correctly 72% of the time (c-statistic = 0.72).

CONCLUSION

Use of the SQ by emergency physicians may predict 12-month mortality in older ED patients and may help emergency physicians identify older adults in need of palliative care interventions.

Potentially inappropriate medication at hospital admission in patients with palliative care needs

Background Potentially inappropriate medications (PIMs) are common in older patients with polypharmacy, and are related to negative clinical results. Little information is available on the characteristics and consequences of PIMs in patients with advanced chronic conditions and palliative care needs. Objective To evaluate, for this population: (i) the prevalence of PIMs; (ii) the possible risk factors associated with its onset; and (iii) the related clinical consequences. Setting Acute-hospital care Geriatric Unit (AGU) in County of Osona, Spain. Method Ten-month prospective cross-sectional study. Patients with palliative care needs were identified according to the NECPAL CCOMS-ICO^® test. Upon hospital admission, a multidisciplinary team consisting of a pharmacist and two AGU physicians determined the PIMs of the routine chronic medication of the patients. Sociodemographic and pharmacological data were collected with the objective of determining possible risk factors related to the existence of PIMs. Main outcome measure Prevalence and type of PIMs according to STOPP version 2 and MAI criteria at the time of hospital admission. Furthermore, days of hospital admission, destination at hospital discharge and survival analysis at 12 months related to PIMs were evaluated. Results Two hundred thirty-five patients (mean age 86.80, SD 5.37; 65.50% women) were recruited. According to the STOPP criteria, 88.50% of patients had ≥1 criterion (mainly 'indication of medication', followed by those that affect the nervous system and psychotropic drugs and risk drugs in people suffering from falls), and according to the MAI tool, 97.40% of the patients had some criterion related to inappropriate medication (mainly, duration of therapy). The following conditions were identified as risk factors for the existence of PIMs: insomnia, anxiety-depressive disorder, falls, pain, excessive polypharmacy and therapeutic complexity. There were no differences among patients in days of hospital stay, discharge's destination or survival at 12 months, regardless of the tool used. Conclusion The presence of PIMs is high in patients requiring palliative care. Some potentially modifiable risk factors such as the pharmacological ones are associated with a greater presence of inappropriate medication. The presence of PIMs does not affect this population in terms of mortality.

The "surprise question" for predicting death in seriously ill patients: a systematic review and meta-analysis

BACKGROUND

The surprise question - "Would I be surprised if this patient died in the next 12 months?" - has been used to identify patients at high risk of death who might benefit from palliative care services. Our objective was to systematically review the performance characteristics of the surprise question in predicting death.

METHODS

We searched multiple electronic databases from inception to 2016 to identify studies that prospectively screened patients with the surprise question and reported on death at 6 to 18 months. We constructed models of hierarchical summary receiver operating characteristics (sROCs) to determine prognostic performance.

RESULTS

Sixteen studies (17 cohorts, 11 621 patients) met the selection criteria. For the outcome of death at 6 to 18 months, the pooled prognostic characteristics were sensitivity 67.0% (95% confidence interval [CI] 55.7%-76.7%), specificity 80.2% (73.3%-85.6%), positive likelihood ratio 3.4 (95% CI 2.8-4.1), negative likelihood ratio 0.41 (95% CI 0.32-0.54), positive predictive value 37.1% (95% CI 30.2%-44.6%) and negative predictive value 93.1% (95% CI 91.0%-94.8%). The surprise question had worse discrimination in patients with noncancer illness (area under sROC curve 0.77 [95% CI 0.73-0.81]) than in patients with cancer (area under sROC curve 0.83 [95% CI 0.79-0.87; p = 0.02 for difference]). Most studies had a moderate to high risk of bias, often because they had a low or unknown participation rate or had missing data.

INTERPRETATION

The surprise question performs poorly to modestly as a predictive tool for death, with worse performance in noncancer illness. Further studies are needed to develop accurate tools to identify patients with palliative care needs and to assess the surprise question for this purpose.

Using the Surprise Question To Identify Those with Unmet Palliative Care Needs in Emergency and Inpatient Settings: What Do Clinicians Think?

BACKGROUND

The surprise question (SQ), "Would you be surprised if this patient died within the next year?" is effective in identifying end-stage renal disease and cancer patients at high risk of death and therefore potentially unmet palliative care needs. Following implementation of the SQ in our acute care setting, we sought to explore hospital-based providers' perceptions of the tool.

OBJECTIVES

To evaluate (1) providers' perceptions regarding the feasibility of SQ use in emergency and inpatient settings, (2) clinician perceptions regarding the utility of the SQ, and (3) barriers to SQ use.

DESIGN

A cross-sectional survey of medical providers following addition of the SQ to the electronic record for all patients admitted to a tertiary care hospital.

RESULTS

A total of 111/203 (55%) providers participated: 48/57 (84%) emergency physicians (EPs) and 63/146 (43%) inpatient providers (IPs). Most reported no difficulty using the SQ. Modest numbers in both groups reported that the SQ influenced care delivery (EPs 37%, IPs 42%) as well as goals of care (EPs 45%, IPs 52%). At least some advance care planning discussions were prompted by the SQ (EPs 45%, IPs 58%). Team discussions were influenced by SQ use for more than half of each group. Most respondents (55%) expressed some concern that their SQ responses could be inaccurate.

CONCLUSIONS

In this setting, clinicians indicated that use of the SQ is feasible, acceptable, and useful in facilitating advance care planning discussions among teams, patients, and families. Many reported that SQ use influenced goals of care, but concern regarding accuracy was a barrier. Additional research examining SQ accuracy and predictive ability is warranted.

Hospice in heart failure: why, when, and what then?

Hospice is a model of care for patients nearing the end of their lives that emphasizes symptom management, quality of life (QOL), and support of the patient and caregiving family through the death of the patient and the family's bereavement. It is associated with high patient and caregiver satisfaction and appears to not shorten lifespan for appropriately referred patients. Patients with advanced heart failure are being referred to hospice care more often than in the past, but the majority of deaths occur without this benefit. Hospice care in the USA is defined by the Medicare Hospice Benefit and associated regulations. Hospice is appropriate for patients with an expected survival prognosis of 6 months or less, and multiple predictive factors and tools are available to assist in prognostication. Management of symptoms and specific drug therapy options are discussed. For many patients, deactivation of electronic cardiac devices is appropriate when the goals of care are comfort and QOL. Ongoing collaboration of the referring physician with the hospice agency and staff offers opportunities for seamless and quality care.