An android app “Apolipoprotein B Calculator” calculates Apolipoprotein B using regression analysis and neural network – Using the friedewald equation is the same as directly measured low-density lipoprotein cholesterol and better at low-density lipoprotein levels

Background: Apolipoprotein B (Apo B) is an important predictor of the risk of atherosclerotic cardiovascular disease over and above low-density lipoprotein cholesterol (LDL-C), especially in statin-treated patients. Assays of Apo B are not available widely. Objectives: The objective of this study is to derive the Apo B from a lipid profile, using a regression equation and a neural network and compare the results, to compare LDL-C measured by direct assay and a Friedewald equation derived LDL-C in their efficacy to predict Apo B, to determine the effect of lower levels of LDL-C on the prediction models, and to develop an android app “Apo B Calculator” which will calculate the Apo B and also give the predictive accuracy of the result. Methodology: Eight hundred and eighty-five persons were split into a training set and a validation set. Both the regression equation and neural network methods were applied on the training set of 442 patients and the best regression equation and neural network predictive model for Apo B were derived. This was then applied on the validation set of 443 patients to test the R2 of the models. Results: The regression equation Apo B = 25.199 + 0.266 (LDL) + 0.062 (triglycerides level [TGL]) + 0.248 (non-high-density lipoprotein cholesterol) was the best predictor of Apo B when directly measured LDL-C was used. The predictive accuracy of the neural network was lesser than the regression equation (75% vs. 87.4% at 95% confidence interval [CI]). The regression equation for the Friedewald equation derived LDL-C was Apo B = 25.077 + 0.528 friedewald equation (F. LDL) +0.138 (TGL) and was comparable with the neural network (86.4% vs. 85% at 95% CI). The overall efficacy of both the direct assay and Friedewald equation-derived LDL-C were nearly the same (87.4% vs. 86.4% at 95% CI). There was a linear decline in the predictive accuracy of both methods at diminishing LDL-C levels. At lower levels of LDL-C (<70 mg/dl), the accuracy of the Friedewald equation derived LDL-C was a better predictor of Apo B (70% vs. 59.8%). With this data, we developed an android app “Apo B Calculator” which will calculate the Apo B from a directly measured or Friedewald equation derived LDL-C. The app will also mention the predictive accuracy of the results. Conclusions: The regression equation derived from directly measured LDL-C and Friedewald equation derived LDL-C, and the neural network using the Friedewald equation showed near similar efficacy in predicting the Apo B value (87.4%, 86.4%, and 85%). A regression equation using a Friedewald formula is a better predictor of Apo B at LDL-C levels <70 mg/dl. The app “Apo B Calculator” can predict the Apo B from both directly measured and Friedewald equation derived LDL-C and give the predictive accuracy for the method – This will help the clinician to know the Apo B and also the predictive accuracy of such calculated value.