Ben-Gurion University in the Negev answered the question with two studies that suggest using machine learning for common laboratory tests. These studies were recently presented at the Medical Informatics Europe (MIE) conference in Nice, France.
How many times have you come to the family doctor with complaints about various sensations or as part of a periodic examination and received a referral for blood tests? It turns out that the amount of laboratory tests performed in Israel is significantly greater compared to any other western country. Although these tests are available and allow for quick and consistent information regarding patients, they are often tested according to a method that only detects results that deviate from the norm.
Researchers from Ben-Gurion University of the Negev found that machine learning will make it possible to locate abnormal results that may indicate a possible disease. This, by locating and comparing all the results together and comparing them to the entire population, and not just by referring to the abnormal result. This method enables the identification and detection of disease risk as well as a health condition that requires follow-up. So even if each of the results of the various tests are within the normal range, there is still a certain combination that can indicate an abnormal result.
Furthermore, the researchers even built a model that predicts biological age by the results of common blood tests and in it they were able to show that people with a biological age lower than the chronological age are healthier. Using the results of laboratory tests taken from the human biological sample bank of the British Biobank, approximately 500,000 subjects aged 82-37 were examined, with the help of which the researchers built a model for predicting biological age.
The model was able to closely predict the true age of the specimen with an average error of more or less 6 years. The researchers examined the health of subjects who had a discrepancy between their biological age and their actual age and showed that subjects for whom the model predicted an age younger than their actual age were found to be healthier than expected, as they had fewer diagnoses, fewer surgeries and a lower incidence of specific diseases compared to an age-matched control group.
However, for subjects who were predicted to be older than their chronological age, there were no significant differences in the number of diagnoses, the number of surgeries and specific diseases compared to an age-matched control group. The researchers showed that people with abnormal results at the global level have a higher chance of being hospitalized and getting sick in general.
The results of the studies were confirmed at Mount Sinai Hospital in New York. There, the researchers looked at about 100,000 periodic doctor visits (wellness visits) during which laboratory tests were taken. The results were similar to those the researchers found in the British database.
There are currently tests for biological age estimation by sequencing the DNA ends or by testing epigenetic markers, but they are expensive, complicated, time-consuming and require dedicated sample collection, so the majority of the population will not perform this test, certainly not frequently. A computational model based on laboratory tests such as the one that the researchers propose makes it possible to obtain a personal health status picture for each applicant, at a low cost and in a relatively short time, and even as an additional result of the laboratory tests that are done following various indications.
"To determine a health condition is no small matter", he noted Dr. Nadav Rapoport (pictured) from the Department of Software and Information Systems Engineering and the lead researcher in both studies. "We look both at the specific level of each disease and at data independent of the disease, while deepening and examining several different results. The information we produce will allow everyone to know their health status in order to improve or maintain their lifestyle." (Photo by Danny Machlis)
The research group included: students Bar Ezra and Lin Peretz from Ben-Gurion University of the Negev, as well as the partners Shreyas Havaldar & Benjamin Glicksberg from Mount Sinai Hospital in New York.
The studies were funded by Ben-Gurion University of the Negev and the Data Science Center and were published in the journals:
More of the topic in Hayadan: