A nomogram based on metabolic profiling to discriminate lung cancer among patients with lung nodules.

OBJECTIVE

To develop a nomogram that discriminates lung cancer from benign lung nodules through metabolic profiling.

METHODS

This was a retrospective cohort study that recruited 848 participants who were randomized into training and validation sets at a 7:3 ratio. Clinical characteristics and metabolic profiles were retrieved. Variables in the training set with statistically significant differences were selected for further least absolute shrinkage and selection operator (LASSO) regression. The nomogram was built from 13 variables identified by stepwise regression analysis. Receiver operating characteristic, calibration curve, and decision curve analyses were conducted to evaluate the performance of the nomogram by internal validation.

RESULTS

Thirteen variables were selected through LASSO regression to build the nomogram: age, sex, ornithine, tyrosine, glutamine, valine, serine, asparagine, arginine, methylmalonylcarnitine, tetradecenoylcarnitine, 3-hydroxyisovaleryl carnitine/2-methyl-3-hydroxybutyrylcarnitine, and hydroxybutyrylcarnitine. The nomogram had good discrimination for the training set, with an area under the curve of 0.836 (95% confidence interval: 0.830-0.890). Moreover, the calibration curve with 1000 bootstrap resamples showed that the predicted value coincided well with the actual value. Decision curve analysis described a net benefit superior to baseline within the threshold probability range of 15% to 93%.

CONCLUSIONS

The nomogram constructed from metabolic profiling accurately predicted risk of lung cancer.