Two-Step Bedside Frailty Screen Spared Assessments in Liver Disease Study

A nurse-administered, two-step frailty-screening pathway identified all cases of liver frailty index-defined frailty in a single-center cohort of hospitalized patients with chronic liver disease, according to a study that simulated how often the complete assessment would be needed. The pathway used two rapid bedside measures—five timed chair stands and grip strength—and referred patients who screened positive on either measure for a full liver frailty index assessment. In the 604-patient analysis set, the either-positive rule had sensitivity of 1.000, with a 95% confidence interval of 0.968 to 1.000, and negative predictive value of 1.000, with a 95% confidence interval of 0.990 to 1.000. Only 228 patients, or 37.7%, would have proceeded to the complete index, so the simulation indicated that 62.3% of full assessments could have been spared rather than demonstrating realized savings in practice.

The work was designed as a pathway-efficiency study rather than as a validation of an independent diagnostic instrument. That distinction is essential because the two first-stage measures are themselves components of the liver frailty index, or LFI, which served as the reference standard for defining frailty. This shared structure creates incorporation bias, meaning the first-stage tests and the reference standard are not fully independent. The investigators therefore framed the results as triage performance within the cohort, not as evidence that the rapid screen is a stand-alone diagnostic test, is clinically safe, or should immediately replace routine assessment. The study was conducted at a tertiary hospital in southern China from August 2021 through August 2025. All 606 enrolled participants were inpatients, and the analysis did not include outpatients.

Among all 606 patients, the LFI classified 100, or 16.5%, as robust, 393, or 64.9%, as pre-frail, and 113, or 18.6%, as frail. Frailty was defined as an LFI score of 4.5 or higher, pre-frailty as 3.2 to less than 4.5, and robustness as a score below 3.2. The main pathway analysis included 604 patients, comprising 112 frail and 492 non-frail patients, after two patients with missing grip-strength data were excluded. A separate complete-component analysis included 600 patients. The first-stage thresholds were selected from the full analytic sample by maximizing the Youden index against the frailty threshold. A patient screened positive if five chair stands took at least 16.55 seconds or grip strength was 23.42 kilograms or less. Under a prespecified safety rule, bedridden patients who could not attempt the chair-stand test were automatically classified as screen-positive so that inability to perform the test triggered further evaluation.

When evaluated as individual first-stage screens, timed chair stands and grip strength each missed some patients who were classified as frail by the complete index. Chair stands had an area under the receiver operating characteristic curve of 0.982, with a 95% confidence interval of 0.970 to 0.991, and at the selected threshold had sensitivity of 0.918 and specificity of 0.955; nine frail patients were missed. Grip strength had an AUC of 0.920, with a 95% confidence interval of 0.895 to 0.943, sensitivity of 0.955 and specificity of 0.791; five frail patients were missed. Standing balance performed poorly as a stand-alone gate, with an AUC of 0.627, a 95% confidence interval of 0.583 to 0.674, and sensitivity of 0.336 at its selected cutoff. At that cutoff, it missed 73 of 110 frail patients in the component dataset. The investigators therefore reserved balance testing for the complete second-stage LFI assessment instead of using it in the initial triage step.

The combined rule’s zero-miss result reflected complementary capture by the two rapid tests. Of the 112 frail patients, 98 were positive on both chair stands and grip strength, five were positive only on chair stands, and nine were positive only on grip strength. None was negative on both measures. In other words, every patient missed by one component was captured by the other, and a positive result on either test was sufficient for referral. In 2,000 bootstrap resamples, the median sensitivity of the combined rule was 1.000, with both the 2.5th and 97.5th percentiles also at 1.000, and no frail cases were missed in 100% of resamples. An exploratory grid search examined joint cutoffs under a sensitivity constraint of 1.00 and found that the maximum attainable specificity was 0.770, close to the 0.764 of the main strategy. Within this dataset, that result suggests the selected thresholds were near the feasible frontier, although it cannot remove the structural problem of overlap with the reference standard.

Internal split-sample analyses reproduced the same zero-miss pattern. In a temporal validation, cutoffs were derived from the first 400 enrolled patients, including 86 frail patients, and then frozen for the subsequent 204 patients, including 26 frail patients. The later group had sensitivity and negative predictive value of 1.000, specificity of 0.803, a positivity rate of 29.9%, and no missed frailty cases. Applying the full-sample cutoffs to that temporal validation group also produced no missed cases, with 30.9% screening positive. In both directions of an odd-even split, each 302-patient subset also had no missed cases when its own derived thresholds were used. The independently derived cutoffs were similar across subsets: 16.9 to 17.7 seconds for chair stands and 23.2 to 23.4 kilograms for grip strength. These analyses support the within-cohort stability of the pathway, but they are not external validation because all subsets came from the same hospital and recruitment period.

Clinical assessments outside the LFI showed a stepwise gradient across three pathway-defined groups: 376 screen-negative patients, 116 patients who were screen-positive but LFI-negative, and 112 confirmed frail patients. Activities-of-daily-living limitation increased from 25.0% to 34.5% to 75.0% across those groups. High fall risk rose from 26.9% to 43.1% to 70.5%, while nutritional risk increased from 15.4% to 30.2% to 42.9%. Depression and anxiety were available for 455 patients, with denominators of 279, 84 and 92 in the respective groups. Depression was recorded in 21.9%, 22.6% and 33.7% of those patients, while anxiety occurred in 13.3%, 11.9% and 25.0%. Median LFI values also rose from 3.57 to 4.13 to 4.93. These concurrent clinical measures support the interpretation that the three strata differed in clinical burden, but they do not independently validate the LFI definition of frailty.

The pathway also has a clearly defined boundary: it targeted frailty at an LFI score of 4.5 or higher, not the full spectrum of pre-frailty. Among the screen-negative patients, 74.7%, or 281 of 376, were classified as pre-frail, and the highest LFI score among them was 4.45. When the secondary reference threshold of 3.2 or higher was used to combine frailty and pre-frailty, the sensitivity of the either-positive rule fell to 0.442. A negative first-stage result therefore meant that the patient was not classified as frail under the study’s reference standard, not that the patient lacked meaningful functional vulnerability. The study’s boundary analysis supports treating the first stage as a triage process that defers more comprehensive assessment for some patients, rather than as evidence that all clinical concerns had been ruled out.

Several limitations constrain the findings. The single-center, cross-sectional design cannot establish prediction of future outcomes or any longitudinal benefit. No dedicated blinding procedure was documented, and assessors may have known patients’ clinical status. Records did not fully trace the grip-strength dynamometer’s brand, model, nominal resolution or calibration history. Depression and anxiety instruments were not standardized by name, and psychological assessments were available only for a subset. Women numbered 117, and the pooled cutoffs produced a particularly high positivity rate of 78.6% in women, suggesting that sex-specific thresholds may warrant separate evaluation. Exploratory analyses also found lower estimated assessment savings among older patients and those with more severe disease. Liver-disease etiology, fibrosis stage and reason for admission were not systematically retained, limiting subgroup analyses.

Most importantly, the apparent sensitivity and negative predictive value of 1.000 cannot be interpreted as independent diagnostic accuracy because the first-stage measures were embedded in the reference standard. The results show that the simulated pathway had no observed misses among this cohort’s LFI-defined frail patients and could have directed the full index to fewer than four in ten patients. They do not yet establish independent accuracy, clinical safety, improved patient outcomes or time savings in actual nursing workflows. Before routine adoption, the frozen thresholds require prospective, multicenter external validation in independent populations, including larger and more clinically diverse samples of women and patients with chronic liver disease. Implementation studies would also be needed to measure nursing time, workflow burden and patient outcomes.

Subject of Research: Internal evaluation of a two-step bedside frailty-screening pathway in hospitalized adults with chronic liver disease

Article Title: Two‐Step Sequential Screening for Frailty in Patients With Chronic Liver Disease: A Triage Efficiency Study of Liver Frailty Index Components

Article References: Wang, J., Ouyang, S., Zheng, L., Jiang, S., He, N., & Li, L. (2026). Two‐Step Sequential Screening for Frailty in Patients With Chronic Liver Disease: A Triage Efficiency Study of Liver Frailty Index Components. Nursing Open, 13(10), Article e70915. https://doi.org/10.1002/nop2.70915

Image Credits: AI Generated

DOI: 10.1002/nop2.70915

Keywords: frailty screening, chronic liver disease, liver frailty index, nurse-led care, triage efficiency, bedside assessment, grip strength, chair stands, internal validation, clinical workflow, Step, Sequential