Mortality outcomes in hospitalized oncology patients after rapid response team activation

Cancer is the second leading cause of death in the United States, exceeded only by heart disease.¹ Despite the overall decline in cancer death rates from 2000 through 2014, physicians struggle to accurately predict disease progression and mortality in patients with cancer who are within 6 months of death.^2-8 This prognostic uncertainty makes clinical decision making difficult for patients, families, and health care providers. On a health care system level, an insight into end-of-life prognostication could also have substantial financial implications. In 2013, $74 billion was spent on cancer-related health care in the United States.⁹ Studies have shown that from 5% to 6% of Medicare beneficiaries with cancer consumed up to 30% of the annual Medicare payments, with a staggering 78% of costs being from acute care in the final 30 days of life.¹⁰

Rapid response teams (RRTs) were first introduced in 1995 and are now widely used at many hospitals to identify and provide critical care at the bedside of deteriorating patients outside of the intensive care unit (ICU) to prevent morbidity and mortality.^11-15 Although not the original aim, RRTs are commonly activated on patients at the end of life and have therefore come to play an important role in end-of-life care.^11,16 RRT activation in the oncology population is of special interest because the activation may predict higher inpatient mortality.¹⁷ In addition, RRT activation can serve as a sentinel event that fosters discussion on goals of care, change in code status, and initiation of palliative care or hospice use, particularly when also accompanied by an upgrade in level of care.^11,18 As such, the ability to predict mortality after an RRT event, both inpatient and at 100 days after the event, could be of great help in deciding whether to pursue further treatments or, alternatively, palliative or hospice care.

To that end, the purpose of this study was to identify baseline patient characteristics, causes of deterioration leading to the RRT event, and vital signs and laboratory abnormalities in the peri-RRT period – the 24-hour periods preceding and following the time of the RRT event – that are associated with increased mortality, both inpatient and at 100 days after RRT activation. By choosing this acutely decompensated population, the knowledge gained may be able to guide improved advance care and end-of-life planning for terminally ill cancer patients.
 

Methods and materials

A retrospective study was performed at a single, 900+ bed academic center in the northeastern United States during a 2-year study period from October 2014 through November 2016. The Institutional Review Board at Thomas Jefferson University Hospital in Philadelphia, Pennsylvania, reviewed and approved the study.

Through our institution’s RRT database, all consecutive RRT activations during the study period involving hospitalized oncology patients were reviewed. We included patients 18 years or older with a cancer diagnosis, including solid tumor and hematologic malignancy, as well as those who were status post–bone marrow transplantation (BMT), who required rapid response activation while hospitalized at our institution. We excluded patients who activated rapid response while they were in the ICU, including the BMT unit, those on the surgical floors, and those with RRT activation at other hospitals before transfer to our institution. Data for both in-hospital mortality as well as 100-day mortality for all admitted oncology patients was obtained from a separate electronic health record database at our institution from a similar time period.

Our goal was to identify patient characteristics, reasons for the RRT activation, and vital sign and laboratory abnormalities in the peri-RRT period that were associated with increased mortality, both inpatient and at 100 days after RRT activation. Our institution’s RRT database and electronic health records were accessed for data collection. Primary outcome variables for this study were inpatient and 100-day mortality post-RRT activation. We investigated the following predictor variables: age, sex, cancer diagnosis, code status at the time of RRT activation, duration from hospital admission to RRT event, length of hospital stay, time of the day the RRT event occurred (daytime vs nighttime), change in level of care (telemetry upgrade and ICU transfer), previous ICU treatment during the same hospital stay, hospice discharge, reasons cited for the RRT event (increased work of breathing, hypotension, tachyarrhythmia, change in mental status, stroke, gastrointestinal bleed, and seizure), peri-RRT lactate level, international normalized ratio (INR), hemoglobin, positive blood cultures, peri-RRT blood product administration, and scores for systemic inflammatory response syndrome (SIRS) and quick sequential organ failure assessment (qSOFA) in the 24 hours preceding the RRT activation. The SIRS includes abnormal temperature (>38°C or <36°C), heart rate of >90 bpm, increased respiratory rate of >20 times/min, and abnormal white blood cell count (>12,000 cells/mm³, <4,000/mm³, or >10% bands). Its score ranges from 0 to 4, based on the number of SIRS criteria documented. The qSOFA includes hypotension (systolic blood pressure of ≤100 mmHg), increased respiratory rate of ≥22 times/min, and altered mentation and ranges from 0 to 3 based on the number of qSOFA score documented.

Descriptive statistics were generated, and we then conducted bivariate analysis using chi-square tests or Fisher exact tests for categorical variables and simple logistic regression for continuous variables. Multivariable logistic regression models were performed to identify predictors of inpatient and 100-day mortality. Regression models were fit separately for subsets defined by the type of cancer diagnosis. Variables with P < .2 were included in the models, and backward selection method was performed, keeping variables with P < .2. The results are presented as odds ratios (OR) and 95% confidence intervals (CI). C-statistics were used to measure goodness of fit for the models. A c-statistic value of 0.5 indicates the model is not better than random chance; a value higher than 0.7 indicates moderate accuracy, whereas a value higher than 0.8 indicates strong accuracy. P < .05 was considered significant. All analyses were conducted using SAS version 9.4 (SAS Institute Inc, Cary, NC).