Clinicians frequently need a concise, objective measure of overall illness severity drawn from routinely collected data. The laboratory-based frailty index (FI-Lab), defined as the proportion of abnormal laboratory test results, has precedent in geriatric research. This study aimed to determine an optimal, automated FI-Lab construction for acute care and to evaluate its validity as a measure of latent health status across the adult life span in diverse samples.
The investigation was a retrospective multicohort study using three large datasets. Emergency department encounters came from MIMIC-IV-ED (Boston, USA; 2011–2019) and King’s College Hospital (KCH; London, UK; 2017–2020). A community cohort was represented by UK Biobank (UK; baseline data 2006–2010). Source counts reported were 227,736 visits (113,032 patients) in MIMIC-IV-ED; 152,305 visits (96,843 patients) in KCH; and 492,703 participants in UK Biobank.
The authors evaluated multiple FI-Lab configurations by varying how laboratory tests were selected, the total number of tests included, and minimum test thresholds required to compute an index for a given encounter. One configuration highlighted in the source used 25 commonly ordered tests with a requirement that at least 15 be present for calculation. The FI-Lab is calculated as the proportion of included tests that are abnormal, producing a continuous score reflective of accumulated abnormalities.
The primary outcome used to assess construct validity was 1-year all-cause mortality. The authors also examined outcomes such as hospital admission and readmission. Performance of the FI-Lab was compared against chronological age and a commonly used deterioration score in acute care, the National Early Warning Score 2 (NEWS2).
An FI-Lab composed of 25 commonly ordered laboratory tests (minimum 15) demonstrated hazard ratios for 1-year mortality that approached those for chronological age and exceeded those associated with NEWS2 in the source cohorts. In combined models reported in the article, hazard ratios for FI-Lab per standard deviation increase were substantial across datasets. The FI-Lab achieved associations with mortality, admission, and readmission that were comparable with age and superior to NEWS2 according to the study.
Sensitivity analyses explored performance across age groups, sexes, ethnicities, and different model specifications to identify potential blind spots. The FI-Lab’s performance was consistent across these subject groupings and healthcare settings, indicating robustness in diverse populations and age ranges examined in the datasets.
The study found that discrimination of the FI-Lab—its ability to distinguish outcomes—plateaued when the index incorporated approximately 20–40 tests. This suggests a pragmatic core set of commonly ordered laboratory tests is sufficient to capture the prognostic signal without requiring exhaustive panels.
As acknowledged in the source, this was a retrospective study. The authors state that future work should prospectively evaluate how real-time access to the FI-Lab affects everyday clinical decision-making and patient care, potentially via randomized controlled trials. Details such as implementation logistics, clinician workflows, and direct impact on outcomes were not assessed in this retrospective analysis.
The FI-Lab offers an automated, scalable measure of patient vulnerability that can be derived from routinely collected laboratory results without additional testing or manual data entry. Because it tracks accumulated physiological abnormalities, it may serve as a pragmatic complement to clinical judgement across the adult life span. The source also suggests the approach could be extended to other routinely collected measures (for example, vital signs) that can be classified as normal or abnormal.
The article reports that UK Biobank and MIMIC-IV-ED data are available under their respective access processes, while KCH data are not publicly available and require approved researcher access under ethics governance. All analysis code for the FI-Lab is reported as available on GitHub and archived on Zenodo per the source. Funding sources and competing interests are disclosed in the original article; the source notes a range of funders and declared competing interests among authors. The authors emphasize that funders did not influence study design, data collection, analysis, or manuscript preparation.