Simple diagnostics for common diseases

A new combination of single infrared light measurement and machine learning can be used to detect metabolic disorders and high blood pressure

Some common diseases could be easier and quicker to diagnose in future. A team from the Max Planck Institute of Quantum Optics, the Ludwig Maximilian University of Munich and the Helmholtz Zentrum München has demonstrated in a representative study that infrared light measurements of blood plasma when combined with machine learning can be used to detect various metabolic disorders such as type-2 diabetes and high blood pressure. The method also detected prediabetes, a precursor stage of diabetes that other diagnostic methods often overlook. Until now, separate tests have been required to detect the various diseases. As a single measurement is now all that is needed, which only requires a bit of blood and a few minutes to measure, the method enables comprehensive health screening of the population, for example for the early detection of diseases.

According to the Robert Koch Institute (RKI), around nine per cent of adults in Germany are thought to suffer from diabetes, more specifically type-2 diabetes, which mainly occurs in adults over 40. However, only around seven per cent were diagnosed with the disease. This means that around 1.3 million people are unaware that they have diabetes, which significantly increases their risk of cardiovascular disease, among other things. Even more individuals in Germany are unaware that they suffer from high blood pressure and therefore have an increased risk of a heart attack. According to a study by the RKI, in 2014/15 a good 30 per cent of adults, i.e. around 22 million people, knew that they had the disease. However, around five million people suffer from unrecognised high blood pressure. Routine screening of the population could therefore save millions of people from the consequences of diabetes or high blood pressure. "Such population-wide studies will be easier to conduct in future thanks to our method", says Mihaela Žigman, who heads a research group at the Max Planck Institute of Quantum Optics and Ludwig Maximilian University and is responsible for the current study.

Žigman's team has developed an integrated approach to detecting the biochemical changes in blood plasma associated with the various diseases by measuring them with an infrared light.  While the spectroscopic technology itself is not new, what is innovative in the study is the combination of infrared spectroscopy with machine learning, as well as how these technologies were applied in a large-scale populational study. The approach is based on the principle that the biochemical composition of the blood changes as people develop diseases. These molecular changes in the blood result in differences in the infrared spectrum of the blood plasma, creating a so called infrared molecular fingerprint. To identify changes in a person's health, Tarek Eissa, PhD-student at the Max Planck Institute of Quantum Optics,  has trained a machine learning algorithm to recognise the infrared molecular fingerprints corresponding to certain diseases such as type-2 diabetes, prediabetes, high blood pressure and elevated blood lipid levels. The researchers can also diagnose metabolic syndrome and its precursors. Metabolic syndrome comprises different changes in health status, for example high blood pressure, elevated blood lipid levels  or insulin resistance, which indicates developing diabetes. People with metabolic syndrome have an increased risk of developing cardiovascular diseases and bowel or liver cancer, for example.

Diagnostics with high precision

In a study involving around 5200 blood samples from almost 3200 test subjects, the team from the Max Planck Institute of Quantum Optics, Ludwig Maximilian University and the Helmholtz Centre Munich has now investigated how reliably their method detects the various diseases. To this end, they analysed the participants' blood plasma not only with infrared light, but also with the current standard diagnostics for each of the diseases. The study has been published in the journal Cell Reports Medicine. According to the study, type-2 diabetes and elevated blood lipid levels can be identified with around 95 per cent accuracy using the infrared fingerprint. The metabolic syndrome was recognised by the method with almost 90 percent probability. For high blood pressure and prediabetes, the sensitivity is about 75 per cent.

The researchers analysed two samples from over 2000 participants in the study, which they had taken six to seven years apart. Of these, more than 200 developed metabolic syndrome between the two measurements. The scientists used this to evaluate their algorithm to predict whether a person will develop metabolic syndrome in the next six and a half years based on a blood sample. In fact, the algorithm succeeded with a 77 per cent hit rate.

An infrared fingerprint of health

Another result of the current study is also relevant from a medical point of view: " Our algorithm could also single out individuals who were healthy in respect of the investigated diseases and remained healthy over the investigated years", says Mihaela Žigman. "Many conventional diagnostic methods, on the other hand, often produce false results because the values that indicate a disease are often based on measurements of individual molecules or individual biomarkers only." In an earlier study published in eLife, a team led by Mihaela Žigman had already shown that lung, breast, prostate and bladder cancer can be easily and cost-effectively detected using their respective infrared fingerprints and machine learning.

Further studies by independent research groups are required before the method developed by the team from Garching and Munich can be used clinically. In addition, an industrial partner must be identified to develop a practical device that integrates the spectrometer with the machine learning algorithm and ensure it meets the stringent certification criteria for medical devices. "We are convinced that we can significantly simplify the diagnosis of many diseases with an infrared fingerprint", says Mihaela Žigman. "In medicine, there is great interest in simple diagnostic tools for comprehensive screening. However, our method still needs to establish itself within the healthcare sector."