Predict a patient’s risk of developing COPD using a machine learning application

By Bob Yirka, Medical Xpress

ML-based COPD GWAS Manhattan plot with DeepNull. We performed an ML-based COPD GWAS. Here we used the same set of covariates as in Figure 4, with one additional covariate provided by DeepNull. DeepNull model predicts ML-based COPD using age, gender, genotype sequence, and FEV₁/FVC as input. An additional DeepNull covariate is the DeepNull model prediction of ML-based COPD. DeepNull learns a function (that is, linear or nonlinear) to predict ML-based COPD via age, gender, genotype sequence, and FEV.₁/FVC as input. Therefore, this analysis is similar to ML-based COPD GWAS conditioned on FEV.₁/FVC Now instead of assuming that FEV₁/FVC has a linear relationship with ML-based COPD, and DeepNull handles age, gender, and FEV cases.₁/FVC can have a nonlinear relationship with ML-based COPD. p-values were obtained from BOLT-LMM using a two-tailed test. Dashed green lines are genome-wide significant levels (P < 5 × 10 − 8). credit: natural genetics (2023). DOI: 10.1038/s41588-023-01372-4

A team of medical researchers, engineers, and computer scientists at multiple institutions across the United States has discovered that machine learning technology can help doctors predict which patients are at risk of developing COPD.Their study was reported in the journal natural geneticsthe group trained a deep learning network using patient spirogram data to predict the development of COPD.

COPD is the third leading cause of death worldwide. The term describes a number of obstructive lung diseases such as asthma, bronchitis, and emphysema. Previous studies have shown that the earlier COPD is treated, the more likely it is that treatment will be available and the progression will be slower. That’s why medical scientists have been working hard to find new ways to identify patients who are most at risk. In this new effort, the research group applied machine learning to the task.

Researchers trained a deep convolutional neural network to recognize the difference between people with COPD and those without. Data to teach the system were obtained from patient medical records, potential diagnostic classification systems, and spirograms. Spirograms are created by administering spirometry to patients, where they breathe into a tube-like device that is connected to a machine that calculates lung strength.

Once the system was able to distinguish between healthy and COPD lungs, the team added responsibility score data compiled over the years to help identify early-stage COPD in patients. She then ran the system on data (including spirograms) from her 325,000 patients in the UK Biobank. We also provided risk data from participants in several other healthcare-related initiatives. They found that they could train the system to detect a patient’s very early signs of her COPD.

The team concludes by suggesting that their system could be readily used to screen patients for COPD by inputting spirogram data. They also note that it could be used in new research efforts aimed at understanding more fully how COPD starts in the lungs and why it progresses so rapidly.

For more information:
Justin Cosentino et al. Inference of chronic obstructive pulmonary disease by deep learning on raw spirograms identifies new loci and improves risk models. natural genetics (2023). DOI: 10.1038/s41588-023-01372-4

Journal information:
natural genetics