Improving Readmission Risk Prediction Algorithms for Sepsis

We read with great interest the invited review titled "Reducing Readmission for Sepsis by Improving Risk Prediction Algorithms" by Renard and colleagues.1 This article highlights the high burden of readmissions among sepsis survivors—reported as nearing 30%2—and the need for improved prediction tools in this vulnerable population.The authors provide a comprehensive overview of current risk prediction models, noting that many are relatively simple—often using only 5 to 10 variables— and rely primarily on clinical variables such as length of stay, severity of illness, comorbidities, recent health care utilization, and discharge laboratory values.3–5 Although these clinical factors are important, we agree with the authors that incorporating social determinants of health (SDOH) is critical to accurately capturing the full range of risks faced by sepsis survivors.Nonclinical factors such as socioeconomic status, area of residence, and access to care significantly influence patients' ability to recover and manage long-term complications following sepsis.6 However, as we have observed in our own work, the availability and quality of SDOH data in electronic health records (EHRs) remain limited. Although some EHR systems have begun collecting SDOH domains (eg, transportation needs, financial resource strain, housing instability, and food insecurity), the fields are often incomplete or underutilized. These data are often collected early during hospital admission, when rapport with patients and caregivers is still being established, potentially contributing to missing or inaccurate responses.To improve the completeness and utility of SDOH data, we suggest considering the timing and framing of data collection. Delaying sensitive questions until later in the hospital stay, or clearly explaining how the information will be used to enhance care, may improve both data quality and patient engagement. Additionally, increasing clinicians' awareness of the impact of SDOH on patient outcomes may further support more complete and accurate data collection.We also believe that risk prediction models should account for postdischarge factors. In addition to SDOH, factors such as physical functional status, cognitive impairment, psychological distress, living arrangements, and receipt of post–acute care services should be considered. These community-living factors are critical to understanding the risks that sepsis survivors face during transitions in care and may further improve the predictive accuracy and relevance of risk models.7Improving readmission prediction for sepsis survivors requires not only better algorithms but also better data—and a more comprehensive view of the patient experience beyond the hospital walls.