Comparability of 11 different equations for estimating LDL cholesterol on different analysers.

OBJECTIVES: Low-density lipoprotein cholesterol (LDL-C) estimation is critical for risk classification, prevention and treatment of atherosclerotic cardiovascular disease (ASCVD). Predictive equations and direct LDL-C are used. We investigated the comparability between the Martin/Hopkins, Sampson, Friedewald and eight other predictive equations on two analysers, to determine whether the equation or analyser influences predicted LDL-C result. METHODS: In two unpaired datasets, 9,995 lipid profiles were analysed by the Abbott Architect and 4,782 by the Roche Cobas analysers. Non-parametric statistics and Bland Altman plots were used to compare LDL-C. RESULTS: On the Abbott analyser; the Martin/Hopkins, Sampson and Friedewald LDL-C were comparable (median bias ≤1.8%) over a range of 1-4.9 mmol/L. On the Roche platform, Martin/Hopkins LDL-C was comparable to Friedewald (median bias 0.3%) but not to Sampson LDL-C (median bias 25%). In patients with LDL-C <1.8 mmol/L and triglycerides (TG) ≤1.7 mmol/L, predicted LDL-C using Abbott reagents was similar between Martin/Hopkins, Sampson and Friedewald equations but not comparable using Roche reagents. Abbott reagents classified 10-20% of patients in the 1.0-1.8 mmol/L range (Martin/Hopkins 13.4%; Sampson 14.5%; Friedewald 16%; direct LDL-C 13.2%). Roche reagents classified 11-30% in the 1.0-1.8 mmol/L range (Martin/Hopkins 23%; Sampson 11%; Friedewald 25%; direct LDL-C 17%). CONCLUSIONS: Performance of predictive equations is influenced by the choice of analyser for total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C) and TG. Replacement of the Friedewald equation with Martin/Hopkins estimation to improve quality of LDL-C results can be safely implemented across analysers, whereas caution is advised regarding the Sampson equation.

Critical evaluation of equations for serum osmolality: Proposals for effective clinical utility.

BACKGROUND: Many studies have assessed the predictive accuracy of serum osmolality equations. Different approaches for selecting a usable equation were compared using thirty published equations and patient data from a regional hospital laboratory. METHODS: Laboratory records were extracted with same-sample results for measured serum osmolality, sodium, potassium, urea and glucose analysed in a regional hospital laboratory between 1/1/2017-31/12/2018. Differences were analysed using Passing-Bablok and difference (Bland-Altman) analysis. Three approaches were compared: the shotgun approach, adjusting for bias, and deriving a novel equation using multivariate analysis. The criteria for success included bias ≤0.7%, a 230 - 400 mOsm/kg range, and osmolal gap (OG) 95% reference limits within ±10 mOsm/kg. RESULTS: The majority of equations produced proportionally negative-biased results. The shotgun approach identified two equations (EQ19, EQ6) with bias ≤0.7% but unworkable OG reference limits. The bias adjustment approach produced several equations with bias ≤ 0.7% and OG reference limits within or equivalent to ±10 mOsm/kg. A novel equation generated by us (1.89Na+ + 1.71 K+ + 1.08 Urea + 1.08 Glucose + 13.7) improved with the adjustment of bias and was not superior to the adjusted published equations. CONCLUSION: Few published equations are immediately usable. Adjustment of bias derives several usable equations of which the best had OG ranges <20 mOsm/kg. We conclude that adjustment of bias can generate equations of equal or superior performance to that of novel equations.