Creating and Validating a Predictive Model for Suitability of Hospital at Home for Patients With Solid-Tumor Malignancies

Abstract

QUESTION ASKED: Can structured electronic health record (EHR) data be used to predict which unplanned hospital admissions of patients with solid tumor ma-lignancies are potentially suitable for a substitutive hospital-at-home (HaH) program? SUMMARY ANSWER: Over a quarter of admissions to an oncology unit at an academic, urban cancer hospital by patients with solid tumor malignancies fit our definition of potential suitability for HaH. A predictive model created for identifying potentially suitable admissions had moderate discrimination (c-statistics of 0.71 [95% CI, 0.68 to 0.75] and 0.63 [0.59 to 0.67] in derivation and validation cohorts).) to identify unplanned admissions potentially suitable for HaH. We considered an admission potentially suitable if the entire episode could theoretically have occurred in a home setting. We determined this by creating a list of exclusionary events or interventions that were considered incompatible with safe or feasible care at home (escalation of care, rapid response evaluation, in-hospital death, telemetry, surgical procedure, consultation to a procedural service , advanced imaging, transfusion, restraints, and nasogastric tube placement). Admissions that did not involve these items were deemed potentially suitable. We used multivariable logistic regression to find factors associated with potential suitability using structured EHR data available at the time of admission decision (admission source, patient demographics, vital signs, laboratory test results, comorbidities, admission and active cancer diagnoses, and recent hospital utilization). WHAT WE FOUND: Of 3,322 admissions analyzed, we found that over a quarter of admissions (n 5 905, 27.2%) were potentially suitable for HaH. Thirteen factors were associated with potential suitability (ad-mission source; temperature and respiratory rate at presentation; hemoglobin; breast cancer, GI cancer, or malignancy of secondary or ill-defined origin; admission for genitourinary, musculoskeletal, or neurologic symptoms, intestinal obstruction or ileus, or evaluation of secondary malignancy; and emergency department visit in prior 90 days) with model c-statistics of 0.71 (95% CI, 0.68 to 0.75) and 0.63 (0.59 to 0.67) in the derivation and validation cohorts. BIAS, CONFOUNDING FACTORS, DRAWBACKS: Our study used EHR data from a single site; variability in protocols for admissions, orders, diagnostics, consultations, and diagnosis coding between institutions may limit gen-eralizability. Furthermore, our definition of potential suitability was based on the presence or absence of specific documented events or interventions and did not account for other factors that are important for determining the proper setting of care, such as nursing needs, home safety, and functional status. REAL-LIFE IMPLICATIONS: A notable proportion of patients with solid tumor malignancies can potentially be cared for using home-based care as a substitution for hospitalization. A predictive model may help supplement clinical judgment for identifying potentially suitable admissions. Evidence from other settings has suggested that patients receiving inpatient-level care at home may have fewer exposures to iatrogenic complications and avoid high financial costs associated with hospitalization. HaH may thus be a viable part of a strategy for providing high-value care and reducing emergency department and hospital use in the care of patients with cancer. abstract PURPOSE Hospital at home (HaH) is a means of providing inpatient-level care at home. Selection of admissions potentially suitable for HaH in oncology is not well studied. We sought to create a predictive model for identifying admissions of patients with cancer, specifically solid-tumor malignancies, potentially suitable for HaH. METHODS In this observational study, we analyzed admissions of patients with solid-tumor malignancies and unplanned admissions (January 1, 2015, to June 12, 2019) at an academic, urban cancer hospital. Potential suitability for HaH was the primary outcome. Admissions were considered potentially suitable if they did not involve escalation of care, rapid response evaluation, in-hospital death, telemetry, surgical procedure, consultation to a procedural service, advanced imaging, transfusion, restraints, and nasogastric tube placement. Admission source, patient demographics, vital signs, laboratory test results, comorbidities, admission and active cancer diagnoses, and recent hospital utilization were included as candidate variables in a multivariable logistic regression model. RESULTS Of 3,322 admissions, 905 (27.2%) patients were potentially suitable for HaH. After variable selection in the derivation cohort (n 5 1,097), thirteen factors predicted potential suitability: admission source; temperature and respiratory rate at presentation; hemoglobin; breast cancer, GI cancer, or malignancy of secondary or ill-defined origin; admission for genitourinary, musculoskeletal, or neurologic symptoms, intestinal obstruction or ileus, or evaluation of secondary malignancy; and emergency department visit in prior 90 days. Model c-statistics were 0.71 (95% CI, 0.68 to 0.75) and 0.63 (0.59 to 0.67) in the derivation and validation (n 5 1,095) cohorts. CONCLUSION Hospital admissions of patients potentially suitable for HaH may be identifiable using data available at admission.