Real‐world performance of point‐of‐care metre, blood gas and laboratory glucose methods in Australian and American hospital inpatient cohorts

Abstract Aims Point‐of‐care (‘POC’) glucose metres are ubiquitous in hospitals, with results guiding clinical glucose management. Inpatient accuracy can diverge from controlled evaluations and may alter decision making. We therefore aim to characterise real‐world inpatient glucose metre performance against gold standard methods. Methods We assess POC metre, blood gas and laboratory glucose measures across two large, retrospective, adult, inpatient cohorts (Australian quaternary centre and US MIMIC‐IV dataset). Glucose ‘pairs’ were defined as measurements by different modalities occurring within 15 min, with sensitivity analysis within 5 min. Agreement was evaluated using robust regression, mean absolute relative difference, and error‐grid analyses. Stratified analyses assessed the effect of haemoglobin and pO 2 . Results Blood gas closely paralleled laboratory glucose, supporting its use as a practical comparator. POC glucose metres in both datasets exhibited systematic bias: positive at lower glucose and negative at higher glucose, with greatest variability at extremes. Narrowing the pairing window modestly improved overall error in the Australian dataset but still did not meet ISO or FDA numerical targets. For both datasets, on Parkes/Consensus Error Grids, >98% of pairs fell within low‐risk zones (A/B). On the DTS grid, zone‐A percentages were 90.7% in the Australian cohort and 75.7% in MIMIC‐IV. Low haemoglobin amplified negative bias at higher glucose, while pO 2 strata showed no consistent impact. Conclusions In routine inpatient care, POC glucose metres show systematic bias, amplified by anaemia. Our results support local validation of glucose metre performance and identification of high‐risk results to prompt confirmatory testing.