Algorithm can identify people who have been mis- or overdiagnosed

Disease and treatment 4. apr 2021 3 min Postdoc Isabella Friis Jørgensen Written by Kristian Sjøgren

For some diseases, many of the people suspected of having them are mis- or overdiagnosed. A new algorithm can help physicians identify people who may not have been correctly diagnosed.

The healthcare systems in Denmark and many other countries are challenged by excessive mis- or overdiagnosis. For example, some people with symptoms of respiratory disease are diagnosed with chronic obstructive pulmonary disease (COPD) when they actually have lung cancer.

An algorithm developed in Denmark appears to be able to identify patients at risk of mis- or overdiagnosis.

By reviewing the typical disease trajectory for people with, for example, COPD, the algorithm can identify people who do not fit the typical disease trajectory, so their doctor has an opportunity to check the diagnosis again.

“Worldwide, an estimated 5–60% of the people with COPD are misdiagnosed, and tools to ensure more accurate diagnosis are therefore required. This applies not only to COPD but also to many other diseases,” explains the researcher behind the project, Isabella Friis Jørgensen, a Postdoctoral Fellow from the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen.

The research has been published in NPJ Digital Medicine.

Algorithm maps the typical disease trajectory

Isabella Friis Jørgensen’s research originates from Søren Brunak’s research group at the University of Copenhagen.

We have previously written about Søren Brunak’s research in this article. Essentially, researchers at the University of Copenhagen have developed algorithms that characterize the development of a disease based on the illnesses that typically precede a diagnosis with a specific disease.

The researchers use big data from medical records to generate the trajectory that characterize each disease.

For example, metabolic syndrome often precedes being diagnosed with type 2 diabetes, and irregular menstruation frequently precedes being diagnosed with breast cancer.

“In the project, we show that we can identify people with different or unusual disease trajectories, which may indicate that they have been mis- or overdiagnosed,” says Isabella Friis Jørgensen.

Some do not follow the typical disease trajectory

Specifically, the researchers used COPD as an example of how algorithms and medical data can identify people who have potentially been mis- or overdiagnosed.

Pneumonia, other respiratory diseases or cardiovascular disease often precede or accompany COPD, but some people do not have these typical comorbidities but have nevertheless been diagnosed with COPD.

“Most people follow the typical trajectory of the disease, but some do not. We can identify them by examining their trajectory,” explains Isabella Friis Jørgensen.

People misdiagnosed with COPD presumably have lung cancer

The researchers identified 284,154 people who had been diagnosed with COPD over 21 years; 42,459 did not have the standard comorbidities, and of these, the researchers more closely examined the records of 9,597 of them.

Of these, the researchers assessed that 2,185 had been misdiagnosed because they did not have the typical disease trajectory for COPD.

Ten percent were eventually also diagnosed with lung cancer, and Isabella Friis Jørgensen suspects that a much larger proportion probably had lung cancer but were only diagnosed with COPD.

This conclusion was supported by the laboratory test values, including white blood cells, C-reactive protein and haemoglobin, in which the potentially misdiagnosed people had values closer to those of people with lung cancer than the average values of people with COPD.

“The numbers looked more like lung cancer than COPD. We therefore hypothesize that most of these people should have been diagnosed with lung cancer but instead were diagnosed with COPD,” says Isabella Friis Jørgensen.

Misdiagnosed people die sooner

The further examination of the trajectories supported the conclusion of mis- or overdiagnosis.

Survival analysis showed that, among the people diagnosed with COPD, those who had a typical COPD trajectory lived longer than those who lacked its classical signs.

According to Isabella Friis Jørgensen, this result was initially surprising because the researchers thought that people with COPD and many other comorbidities would die sooner than people with COPD with no comorbidities.

However, the reverse was the case.

“These people were probably diagnosed with COPD despite having lung cancer and therefore did not receive the right treatment. This is probably why they died sooner than those correctly diagnosed with COPD,” explains Isabella Friis Jørgensen.

Helping doctors diagnose correctly

Isabella Friis Jørgensen says that the algorithm used to identify the people mis- and overdiagnosed with COPD can be applied to data from people with various diseases to identify in real time those who do not follow the typical disease trajectory.

In these cases, the doctor might take a second look at the diagnosis and perhaps perform confirmatory examinations.

The researchers found that only 4% of those diagnosed with COPD had undergone a lung function test to confirm the diagnosis.

“We hope that we can develop the algorithm to become an integral part of doctors’ clinical work, so that they can look up a patient in the system, which then indicates whether the disease trajectory is atypical. In this case, the doctor could ensure that the correct diagnostic tests are performed and that the patient is examined for other similar diseases,” says Isabella Friis Jørgensen.

Time-ordered comorbidity correlations identify patients at risk of mis- and overdiagnosis” has been published in NPJ Digital Medicine. The authors are employed at the Novo Nordisk Foundation Center for Protein Research, University of Copenhagen.

The Brunak Group aims for understanding multi-morbidity disease progression patterns and their relation to treatment events. The group integrates hete...

© All rights reserved, Sciencenews 2020