A growing number of older adults in the United States who are near the end of life are admitted to intensive care units (ICUs) with acute critical illness and receive invasive medical treatments. The use of prolonged life-sustaining treatments such as mechanical ventilation for older adults near death is common and increasing over time. However, most older adults report end-of-life care preferences to prioritize their comfort and quality of life near death over burdensome life-sustaining treatments. To make medical treatment decisions that are aligned with end-of-life-care preferences, patients and their surrogates require meaningful prognostic information in the setting of an acute illness. However, traditional prognostic tools for critically ill patients have focused on estimating patients' risk of death. Patients and surrogates often struggle to interpret or use mortality estimates for decision making, and the dichotomous outcome of mortality fails to describe the longitudinal trajectory of a patient's illness. We hypothesize that a novel tool that incorporates longitudinal prognostic information and a more descriptive account of potential outcomes will help patients, surrogates, and ICU physicians make treatment decisions that are aligned with end-of-life care preferences. The central goals of this NRSA Individual Fellowship are to (1) facilitate the candidate's career development into an independent, successful health services researcher in critical care medicine and (2) address the disparity between end-of-life medical care that is delivered in the United States and patients' goals values, and preferences. The candidate and her mentors have designed a training plan to accomplish these goals that includes two specific research objectives. First, we aim to create a novel, longitudinal predictive model for critically ill adults that characterizes the expected trajectory of medical treatments over time and a description of the patient's likely outcome. To accomplish our first objective, we will apply the innovative tools of predictive analytics from the field of engineering (i.e. robust machine learning) to a large EHR-derived clinical database of all admissions to an adult medical ICU over 3 years (N ? 3,000 patients). The predictive model will characterize discrete, temporal patterns of patients' exposure to medical treatments during an ICU stay (?illness trajectories?) and describe the outcomes of patients who follow each trajectory. Second, we plan to use the output of our predictive model to design a user-centered prognostic tool for clinical use. Using both real and hypothetical output from our predictive model, we will engage critically ill patients (if able to participate), their surrogate decision-makers, and critical care clinicians to design and iteratively revise a prognostic tool that presents longitudinal data. The stakeholders will participate in qualitative interviews and focus groups to provide feedback about the content, format, and usability of the tool for treatment decisions in the ICU. This study will lead to the prototype of a novel, patient-centered prognostic tool that is designed to better align treatment decisions in the ICU with patients' end-of-life care preferences.
Many critically ill patients and their families are confronted with a decision between life-prolonging but burdensome treatments or comfort-focused medical care that may lead to an earlier death. These decisions can be very difficult when there is uncertainty about whether the patient can recover and what the patient's life will be like during the illness and recovery. To help with these difficult decisions, we propose a study to design a new tool to help patients, families, and ICU clinicians predict the range of possible outcomes for an individual patient.
