Automated Multisource Electronic Frailty Index in Acute Ischemic Stroke: Development and Clinical Utility
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Frailty is common in acute ischemic stroke (AIS) and predicts poor outcomes, but is not routinely captured in acute stroke care. Manual frailty tools are difficult to apply consistently in busy inpatient settings, while existing electronic frailty indices (eFIs) often rely on limited data modalities. We developed a scalable pre-stroke electronic frailty index (eFI) using multisource electronic medical record (EMR) data and evaluated its clinical utility.
Methods
We conducted a retrospective cohort study of AIS admissions to Singapore General Hospital from July 1, 2024, to January 31, 2025. A fully automated pipeline derived an eFI from EMR data over a 3-year lookback period, incorporating ICD-10 codes, vital signs, anthropometry, laboratory results, medications, and free-text documentation processed using artificial intelligence–augmented extraction of predefined, clinically interpretable deficits. Candidate variables were screened using a validated 10-step frailty index framework and refined by multidisciplinary expert consensus.
Results
Among 501 AIS cases, the pipeline generated 75 candidate variables and a final 33-variable eFI, with scores derived for 492 cases (98.2%). Frail patients had greater premorbid disability, higher stroke severity, longer hospitalization, greater rehabilitation use, worse discharge disability, higher 30-day readmission, and higher cumulative post-discharge mortality. In multivariable analyses adjusted for age, sex, NIHSS, premorbid mRS, and reperfusion therapy, each 0.1-unit increase in eFI was associated with mortality beyond 90 days after discharge (adjusted HR, 1.47; 95% CI, 1.19–1.81), 30-day readmission (adjusted OR, 1.91; 95% CI, 1.43–2.59), longer hospital stay (β, 2.8 days; 95% CI, 1.4–4.2), and discharge to inpatient rehabilitation rather than home (adjusted RRR, 1.66; 95% CI, 1.27–2.16).
Conclusions
In a well-documented acute stroke service supported by comprehensive longitudinal EMR data, automated multisource eFI derivation was feasible and clinically informative in AIS, capturing baseline vulnerability beyond conventional stroke measures and supporting frailty-informed risk stratification and discharge planning.
