EHR Symptom Data Improve Readmission Risk Prediction After AML Induction

In a retrospective study published in JCO Clinical Cancer Informatics, adding comorbidity data and symptom frequency from clinical notes to a standard clinical model improved prediction of unplanned readmission or death within 30 days after induction chemotherapy for acute myeloid leukemia (AML).¹

Of 305 adults treated at a Midwestern academic hospital, 39.7% had an unplanned readmission or died within 30 days of discharge from their initial induction hospitalization. Fever and infection accounted for more than half of documented readmission reasons.

Study Design

Investigators from the University of Iowa and the University of Florida analyzed 1,111 inpatient encounters from adults with AML treated between August 2006 and December 2021. The cohort included patients with newly diagnosed and relapsed/refractory disease. Patients with acute promyelocytic leukemia, stem cell transplant recipients, and patients with confirmed chronic myeloid leukemia were excluded.¹

Patients had a mean age of 64.6 years and a mean induction stay of 25.8 days. Most were male (58.7%) and White (91.2%). Lower-intensity regimens, such as azacitidine, decitabine, or venetoclax, were given to 43.3% of patients.¹

Readmission and Mortality Findings

Within 30 days of discharge, 67 patients (22%) had at least 1 unplanned readmission and 67 (22%) died. Of the 121 patients with an adverse event, 54 were readmitted only, 54 died only, and 13 experienced both.¹

The 67 readmitted patients had 79 readmission encounters with 99 documented reasons:

• Fever/infection: 53.5%
• Metabolic/GI/renal complications, such as diarrhea, vomiting, hypercalcemia, and acute kidney injury: 16.2%

• Pain/discomfort: 14.1%
• Neurologic symptoms: 7.1%
• Disease progression or treatment-related complications: 5.1%
• Cardiopulmonary issues and bleeding: 2% each

The 22% readmission rate is lower than the 27% to 30% seen in prior reports. The authors said the gap may reflect their stricter exclusion of planned admissions rather than a true difference in risk.¹

Predictors of Adverse Events

The base model included age, sex, treatment era, and chemotherapy intensity, and had an area under the curve (AUC) of 0.67. Adding comorbidities and NLP-derived symptom frequency raised the AUC to 0.74. That change was not statistically significant (P = .061). None of the 9 laboratory tests evaluated were selected for the models.¹

In the final model, 3 factors were independently associated with adverse events:

• Solid neoplasm history. Odds ratio (OR), 2.34 (95% CI, 1.02-5.38).
• Cardiopulmonary symptoms documented per day, such as edema, cough, and shortness of breath. OR, 1.16 (95% CI, 1.07-1.26).
• Age combined with chemotherapy intensity. Each 5-year increase in age raised the odds of an adverse event by 22% with high-intensity chemotherapy (OR, 1.22; 95% CI, 1.03-1.45). With low-intensity chemotherapy, each 5-year increase lowered the odds by 27% (OR, 0.73; 95% CI, 0.58-0.93).

Treatment in 2018 or later was associated with higher odds of an adverse event in the base model (OR, 1.93). It lost significance after adjustment for comorbidities. The authors suggested that venetoclax-based combinations may have extended treatment eligibility to older, less fit patients with more comorbidities.¹

Female sex was significant in the earlier models but not the final model. The authors noted that female patients with cancer tend to report higher symptom burden, which cardiopulmonary symptom frequency may have captured.¹

Takeaways for Oncology Nurses

The authors noted that many complications can be detected before discharge. They cited antimicrobial prophylaxis and structured outpatient monitoring as areas for prevention.¹ The findings point to several nursing priorities:

• Infection preparedness. Fever and infection drove most readmissions, and the authors noted that infection remained the leading cause even after neutrophil recovery. Discharge teaching should cover fever thresholds and whom to call.
• Symptom documentation. Daily frequency of charted cardiopulmonary symptoms predicted risk. Consistent, specific documentation of edema, cough, and dyspnea feeds the data these tools rely on.
• Age and regimen together. Older patients receiving intensive induction carried higher risk and may warrant closer post-discharge follow-up.
• Oncologic history. A prior solid tumor more than doubled the odds of an adverse event and belongs in discharge risk assessment.
• GI, metabolic, and renal complications. These were the second most common reason for readmission. Patient teaching should cover hydration and when to report diarrhea or vomiting.

Reference

1. Chae S, Harb A, Sutamtewagul G, et al. Early unplanned readmissions and mortality after induction chemotherapy in AML. JCO Clin Cancer Inform. 2026;10:e2600018. doi:10.1200/CCI-26-00018