Improving Readmission Risk Prediction Algorithms for Sepsis

We sincerely thank the letter authors for their thoughtful response to our review, “Reducing Readmission for Sepsis by Improving Risk Prediction Algorithms.”1 We agree that social determinants of health (SDOH) play a critical role in shaping outcomes among sepsis survivors and are often underrepresented in our current risk prediction models. Their comments reinforce and extend our central argument: that current models inadequately capture the multifactorial risks faced by sepsis survivors and that incorporating both SDOH and postdischarge factors is essential for improving predictive accuracy and equity.2–5As highlighted in their letter, the timing and framing of SDOH data collection are central to improving its completeness and utility.6 We agree that early collection (often embedded in admission workflows) may compromise both accuracy and acceptability.7 Patients with sepsis are often cognitively impaired, hemodynamically unstable, or emotionally overwhelmed upon admission.8 Sensitive questions regarding housing, finances, or social support may not be appropriate at this stage, and patients’ responses may be incomplete or unreliable. Moreover, the clinical team may not yet have built sufficient rapport to elicit trust in sharing personal information.Deferring SDOH screening to a later phase of hospitalization, when patients are more medically stable and relationships with care teams have begun to form, may improve patients’ receptivity and engagement.7 Equally important is how SDOH questions are framed. Patients may be more likely to respond honestly when they understand that the information is not simply being collected as a bureaucratic task but will be used to connect them to tangible resources, enhance discharge planning, and reduce their risk for readmission. Framing SDOH collection as a tool for individualized care repositions it from surveillance to service and reflects core principles of respect for autonomy, transparency, and ethical patient engagement.We also agree that current risk prediction tools often overlook key postdischarge and community-living factors.7 Physical function, cognitive impairment, psychological distress, housing stability, and access to home- and community-based services all strongly influence recovery after sepsis.9 Both SDOH and postdischarge factors reflect not just risks but possible remediable barriers. Considering these domains in readmission models may improve predictive accuracy and help staff anticipate possible care deficits.We are encouraged by the shared vision outlined in the letter and agree that the path forward requires multilevel strategies: enhancing SDOH data quality, integrating postdischarge indicators, validating models across diverse populations, and ensuring transparency in how risk is defined and operationalized.