Formula-dependent LDL-C threshold classification and reporting uncertainty in the UK Biobank

Calculated low-density lipoprotein cholesterol (LDL-C) is widely reported, but formula-dependent estimation differences may alter classification near decision thresholds. We assessed formula-dependent LDL-C classification relative to direct LDL-C in UK Biobank and examined cardiovascular outcomes descriptively.

We analyzed 397,490 participants after quality-control exclusions and exclusion of those with missing direct LDL-C or triglycerides ≥4.52 mmol/L (≥400 mg/dL). Friedewald, Sampson-NIH, and Martin-Hopkins estimates were compared with the UK Biobank direct LDL-C comparator at a 2.60 mmol/L (100 mg/dL) method-comparison threshold. Analyses assessed threshold discordance, triglyceride strata, apolipoprotein B (apoB)/non-high-density lipoprotein cholesterol (non-HDL-C) phenotypes, threshold-window sensitivity analyses, and narrower atherosclerotic cardiovascular disease (ASCVD) endpoints.

Total threshold discordance was 5.96 % for Friedewald, 3.91 % for Sampson-NIH, and 3.23 % for Martin-Hopkins. Discordance increased with triglycerides. At 2.26-4.51 mmol/L (200-399 mg/dL), Friedewald had the largest negative calculated-minus-comparator difference and discordant low frequency (median difference -0.379 mmol/L; 12.9 %), whereas Martin-Hopkins had smaller differences (median -0.119 mmol/L; 4.3 %). Broad cardiovascular composite and narrower ASCVD endpoint models did not support binary threshold discordance as a stable adverse prognostic marker. Discordant low participants had higher triglycerides, body mass index (BMI), diabetes prevalence, and cholesterol-lowering medication use, but lower apoB and non-HDL-C.

LDL-C formula selection affected classification, especially in triglyceride-rich samples near 2.60 mmol/L (100 mg/dL). Calculated LDL-C may warrant interpretation as an estimate with formula-related uncertainty near decision thresholds; non-HDL-C and, where available, apoB may provide complementary information.