Identifying drivers of early readmission in cancer patients to improve care and outcomes.

Abstract

e23295 Background: Cancer patients are at increased risk for hospital readmission, which adversely affects outcomes and burdens the healthcare system. In a prior quality improvement study, we identified factors linked to potentially avoidable 30-day readmissions among adult cancer patients, including symptomatic presentation at admission, failure to thrive, absence of a post-discharge caregiver, and lack of outpatient follow-up. The current study compares cancer patients with unplanned 30-day readmissions to those without to further identify opportunities for prevention and improved care. Methods: A retrospective chart review was performed at Stony Brook University Hospital between June 30, 2021 and July 31, 2022. Adult patients with active malignancy were included. Patients followed with non-SBUH cancer specialists or discharged after July 1, 2022 were excluded. Demographics, cancer characteristics, hospital course, and post-discharge factors were analyzed. Independent predictors of readmission were evaluated using a multivariable Cox proportional hazards model. Results: Of 468 patients identified, 99 were readmitted (21.2%), 85 were not readmitted (18.2%), and 284 were excluded. Patients with hematologic malignancies were readmitted more often than those with solid tumors (73% vs 23%, p<0.001; HR = 2.26). Patients with designated health care proxies (HCP) had a higher readmission rate (63% vs 31%, p<0.001). There was no significant difference in the percentage of patients reporting pain on initial admission between readmitted (37.4%) and non-readmitted (27%) patients (p=0.183). However, patients prescribed a new pain medication at discharge had a lower readmission rate (47% vs 69%, p=0.003; HR = 0.58). Readmitted patients had a longer length of stay (10.8 days vs. 6.2 days, p=0.0013) with each additional day of hospitalization associated with a 2.4% increase in the HR for readmission. Risk of readmission predicted by the electronic medical record’s (Cerner) proprietary risk model demonstrated a positive predictive value of 77% and negative predictive value of 50%. Among patients flagged by the EMR to be at high risk, only 33% were readmitted (p<0.001). Code status and caregiver availability were not associated with readmission. Conclusions: Identification of factors associated with readmission can identify patients at risk for deterioration and guide targeted interventions to improve outcomes. Improved pain management may be a readily available option to reduce hospital readmission risk for cancer patients. Current EMR readmission risk assessment models are not validated for cancer patients and fail to identify many patients at higher risk for readmission. A risk assessment tool validated in cancer patients and their unique characteristics is needed. Identifying both risk and protective factors for readmission would optimize care for cancer patients and reduce healthcare costs.