A Learning Health Systems Approach to Frailty Prediction Using EHR Data
Abstract
Research Objective
To predict occurrence of preventable patient safety events using only data elements that are almost universally supported in EHR systems.
Study Design
We drew a random sample was drawn from the EHR records of adult patients not restricted to any particular age range or diagnosis and therefore representative of a real-world health system population. This sample was subdivided into a developmental and a test subset. We used the Rockwood Deficit Accumulation method to construct a frailty index from ICD10 codes, lab value-flags, and vital signs on a per-visit basis with a rolling two-year time-window (EFI, electronic frailty index). The Cox proportional hazard model was used to model days elapsed from a randomly selected visit to the first occurrence of the selected outcomes (which are also based on routinely available EHR data elements). In contrast to previous studies, we treated EFI as a time-varying predictor with multiple follow-ups per patient, which is more realistic than relying on one static time-point. We used a representative sample of the adult patient population rather than limiting it to older individuals and found EFI to be a useful metric even at relatively young ages.
Population Studied
We accessed EHR data for 14,844 patients randomly sampled from our academic health center's data warehouse which supports our ACT/SHRINE node and is regularly updated from our institution's EHR system.
Principal Findings
We found that this electronic frailty index was robustly predictive of ED utilization, hospitalization, discharge from hospital to SNF/ICF, hospital readmission, all-cause mortality, and complications including in-hospital trauma and infections.
Conclusions
Though currently frailty indexes are most used in geriatrics and gerontology, we find evidence that age-related declines follow a lifelong trajectory that is observable in patient charts even in younger patients (18–45).
Implications for Policy or Practice
Frailty predicts poor patient outcomes and healthcare services utilization. Accurate assessment of frailty can inform clinical management decisions and assist with anticipating healthcare resource utilization. Deficit accumulation indexes have a much simpler algorithm than most other frailty scores (e.g. HCC-RAF). Because they capture a large number of conditions, they are robust against missing data and variations between disparate data sources. To further reduce barriers to implementation and accelerate the evolution of this tool, we are making the source code freely available under and open source license. Our next goal will be a side-by-side comparison of HCC-RAF and EFI for predicting the outcomes presented here as well as cost of care. If EFI can be shown to out-perform CMS risk scores, the implications for value-based care will be significant.
Primary Funding Source
National Institutes of Health.