NEW YORK – UK diagnostics firm Global Access Diagnostics has developed a method of predicting flare-ups of chronic obstructive pulmonary disease (COPD) up to a week earlier using artificial intelligence combined with a lateral flow test.
In a study published Wednesday in ERJ Open Research with researchers from the University of Leicester, the team found that their method, called Headstart, predicted a flare-up approximately one week before any symptoms occurred.
The research team first began their work by sampling patients both when their COPD was exacerbated and when they were stable and identified "a group of mediators in the urine that could separate those who were having an attack versus the stable visit," said Chris Brightling, a clinical professor of respiratory medicine at the university and one of the paper's lead authors.
Global Access Diagnostics — formerly known as Mologic — then developed a lateral flow test and an optical reader to detect the presence or absence of the biomarkers and provide a semiquantitative assessment for each of the analytes. The optical reader was connected to a smartphone, which then connected to a "central hub" that stored the data, Brightling said.
The researchers collected data every day for six months to build a patient's "profile" and determine their baseline measurements of each analyte. They then used an artificial neural network based on the data to identify the patterns that predicted an exacerbation.
In its original retrospective study, the researchers selected 35 analytes they thought would be relevant for COPD but eventually narrowed the pool to 10 that best distinguished between stable and exacerbation states after analyzing the cohort, Brightling said. Those analytes are associated with inflammation and COPD exacerbations and aren't necessarily novel biomarkers on their own, he noted. "The combination of them and then the use of them in an AI model" are what's novel.
The 10 biomarkers are neutrophil gelatinase-associated lipocalin (NGAL), tissue inhibitor of metalloproteinase 1 and 2 (TIMP1, TIMP2), C-reactive protein (CRP), fibrinogen, club cell protein 16 (CC16), formyl-methionine-leucine-phenylalanine (fMLP), alpha-1 antitrypsin (A1AT), beta-2 microglobulin (B2M), and retinol binding protein 4 (RBP4).
The 10-biomarker panel was used to differentiate between stable and exacerbation states, but when developing the neural network model, researchers narrowed the panel to five biomarkers for prediction — NGAL, TIMP1, CRP, fibrinogen, and CC16.
The panel was able to distinguish between exacerbation and stable states in the discovery study with an area under the curve of 0.84 and in the validation study with an area under the curve of 0.81. The neural network model predicted an exacerbation within 13 days with an area under the curve of 0.89 and identified an exacerbation seven days before clinical diagnosis.
The AI-based model requires about a month's worth of data to start to learn what is normal for an individual patient, but once it has determined a baseline, it can return results in minutes. The model is designed to provide a "traffic light system" that indicates the likelihood of a flare based on the test's results. Repeating the test the next day then provides greater confidence about whether the result is a "one-off blip" or if the patient is heading toward an exacerbation, Brightling said.
According to Gita Parekh, head of respiratory at Global Access Diagnostics, a urine sample is applied to the wick of the disposable test cassette and immediately inserted into the optical reader, which interprets the intensities of the test lines after about 10 minutes. Using an app developed by the company, the user sends the data from the reader to the cloud, where it is fed into the AI-based algorithm to "unscramble the overall biomarker message found in the urine," Parekh said. The result is then returned to the patient as either a green, amber, or red result, with red indicating they should contact a healthcare provider.
Symptoms of COPD include cough, breathlessness, wheezing, sputum production, and sputum discoloration, and worsening of those symptoms for two or more days indicates an exacerbation, Brightling said. However, "by the time someone actually … presents with an exacerbation, there's not the same opportunity to then be able to change the trajectory of that event."
Having an earlier signal allows clinicians to intervene with a more specific therapy or recommendation, he said.
In the ERJ study, the researchers noted that at the onset of an exacerbation, blood eosinophils and C-reactive protein can be used to direct oral corticosteroid and antibiotic therapy, but these tools "do not provide a reliable risk prediction of the timing of an onset of an exacerbation."
The researchers also wrote that "serial assessment of inflammation using frequent blood testing is unlikely to be acceptable to patients and therefore a less invasive alternative is required."
Brightling said he sees the test being used in patients at higher risk of frequent COPD attacks, noting that it could be used at home with results linked to a central service monitored by a healthcare professional. If a patient's results indicate they're at risk, the clinician may contact the patient and recommend they avoid triggers, such as not going outside on days with high levels of air pollution, or optimize their therapies, such as steroids. Patients could also be tested for specific viruses that may cause a flare or further tested for inflammation to work out whether to start a new therapy, he added.
Parekh noted that any user would need to test frequently at home so the algorithm can learn and track their biomarker profiles and recognize changes from their baselines.
The test could also be modified for other lung diseases with exacerbations, such as asthma, if appropriate biomarkers were found in urine samples, Brightling said. It could also be applied to other conditions outside the lungs that have a molecular signature in the urine.
Global Access Diagnostics plans to commercialize Headstart and has patent protected the combination of biomarkers, while the algorithm remains a "trade secret," she said. Parekh added that the company is exploring commercialization in both the UK and the US, but she noted that the company will need to conduct further clinical studies to generate the evidence needed to meet the requirements for Europe's In Vitro Diagnostic Regulation and US Food and Drug Administration clearance.
After the test was developed, a preliminary trial was conducted in 2018, but the company was unable to complete its validation trial as planned due to the COVID-19 pandemic.
The firm is planning a clinical validation trial and an intervention trial to demonstrate that the test will improve clinical outcomes and will have a cost benefit. It also is partnering with clinical research groups in the US and UK to support the clinical validation trials for regulatory approval and clinical adoption.
Brightling noted that it would be useful to perform further clinical trials to know how frequently measurements need to be taken and how the test performs in different settings.