Background: As the approach to low-density lipoprotein cholesterol (LDL-C) lowering becomes increasingly intensive, accurate assessment of LDL-C at very low levels warrants closer attention in individualized clinical efficacy and safety evaluation. We aimed to assess the accuracy of LDL-C estimation at very low levels by the Friedewald equation, the de facto clinical standard, and compare its accuracy with a novel, big data-derived LDL-C estimate. Methods: In 191,333 individuals with Friedewald LDL-C<70mg/dL, we compared the accuracy of Friedewald and novel LDL-C values in relation to direct measurements by Vertical Auto Profile ultracentrifugation. We examined differences (estimate minus ultracentrifugation) and classification according to levels initiating additional safety precautions per clinical practice guidelines. Results: Friedewald values were less than ultracentrifugation measurement, with a median difference (25th to 75th percentile) of -2.4 (-7.4 to 0.6) at 50-69mg/dL, -7.0 (-16.2 to -1.2) at 25-39mg/dL, and -29.0 (-37.4 to -19.6) at<15mg/dL. The respective values by novel estimation were -0.1 (-1.5 to 1.3), -1.1 (-2.5 to 0.3), and -2.7 (-4.9 to 0.0) mg/dL. Among those with Friedewald LDL-C<15, 15 to<25, and 25 to<40mg/dL, the classification was discordantly low in 94.9%, 82.6%, and 59.9% of individuals as compared with 48.3%, 42.4%, and 22.4% by novel estimation. Conclusions: Estimation of even lower LDL-C values (by Friedewald and novel methods) is even more inaccurate. More often than not, a Friedewald value<40mg/dL is underestimated, which translates into unnecessary safety alarms that could be reduced in half by estimation using our novel method.