AI/ML: It’s Utilization in In Vitro Diagnostics

By Senior Medical Device/Regulatory Analysts

Share this article!

Who knew in 1930 that statistical methods being developed for scientific research application would spearhead AI/ML (artificial intelligence/machine learning) pattern recognition predictive modeling in in vitro diagnostics (IVD) some 80 years later? The logistic regression model is an example of classical statistical thinking that discovers relationships between one laboratory diagnostic variable and multiple other diagnostic variables in patients’ clinical activities. (1) The rapid development of effective COVID-19 testing is an example of the implementation of logistic regression modeling in IVDs, revolutionizing the assessment of disease.

When world governments implemented efforts to control the spread of severe acute respiratory syndrome (SARS)-coronavirus (CoV)-2, dubbed clinically as COVID-19 disease, doctors quickly discovered that previously reliable molecular tests using plasma and whole blood samples did not deliver results fast enough to curtail the spread of the infection. (2) Instead, microfluidics, samples derived from nose and cheek swabs, taken by point-of-care rapid diagnostic devices offered real-time results to physicians for faster treatment and effective quarantining. The pathology biosensors in rapid diagnostic IVDs categorize these devices as “smart diagnostics” capable of detecting thousands of proteins per sample and identifying disease-specific profiles. (3) Smart diagnostics results become part of a smart diagnostic platform as they are uploaded to an electronic medical record system (EMR). (4)

The EMR creates an ever-growing repository of individual patient data that is analyzed by artificially intelligent (AI) software which uses machine learning (ML) techniques for pattern recognition and disease prediction. (4) The European Commission considers software used to diagnose and monitor disease progression and regression to have “autonomous decision-making” skill synonymous to human performance. To help manufacturers discern whether their software is classified as a medical device, the European Commission published an example chart of AI software as a medical device (SaMD). See Figure 1 below. (5)

Pattern recognition is established when an EMR system provides common disease variables that improve the pre-test probability for disease. Specificity for sepsis diagnosis is reported to be >85% with the use of ML algorithms that analyze 15 defined variables encompassing both laboratory and clinical indicators of organ dysfunction related to sepsis. (6) In Figure 2 (cited as Fig. 3 in the literature), Tran et al. delineates machine learning algorithms used to analyze laboratory medicine data. (4)

Machine Learning in Laboratory Medicine

Figure 2: Machine Learning Techniques

In addition to facilitating disease diagnosis and identifying treatment options, “the European AI-in-healthcare market is estimated to grow with a compound annual growth rate (CAGR) of 35.45% during 2020-2028”. (5) This accelerated growth rate is in part due to the linear development of SaMD. The following diagram, Figure 3, demonstrates how the interface of the SaMD ecosystem is already focused in patient-centered clinical environments. (5)

AI/ML Regulations

The Medical Device Regulation (EU) 2017/745 and In Vitro Diagnostic Devices Regulation (EU) 2017/746 classify AI/ML clinical software and deep learning algorithms as Class II medium risk medical devices. Implied in the very name, “learning algorithms” are unique medical devices which inherently develop and evolve as more real-world data (such as that which comes with patient medical records) become available and are incorporated into diagnostic projections of disease. Unique to this sizable data mine is that not only is it subject to performance approval, but it must meet data privacy and cybersecurity standards in order to obtain ethical use approval as well. Furthermore, manufacturers need to ensure that AI tools in use do not generate unethical outcomes due to biased or skewed datasets. (5)

The U.S. Food and Drug Administration (FDA) has issued several guidance documents for manufacturers of AI-driven clinical tools, including a guideline for the regulatory submission of stand-alone software-based medical devices. The guideline defines a mobile app as “a software application that can be executed (run) on a mobile platform (i.e., a handheld commercial off-the-shelf computing platform, with or without wireless connectivity), or a web-based software application that is tailored to a mobile platform but is executed on a server.” It also ensures that manufacturers develop safety requirements for their software-based medical devices. (7) In 2021, the FDA authorized the first machine-learning device that “identifies certain biomarkers that may be indicative of SARS-CoV-2 infection as well as other hypercoagulable conditions (such as sepsis or cancer) or hyper-inflammatory states (such as severe allergic reactions), in asymptomatic individuals over the age of 5.” (8)

Allow Nerac to help you obtain a CE mark for your AI/ML technology.

How Can Nerac Help?

Complying with the new IVDR requirements may seem like a daunting task, particularly for manufacturers who could previously self-certify, but partnering with Nerac can help you navigate the process and successfully achieve compliance.  Nerac has a knowledgeable team of analysts/medical writers with expertise in clinical literature searches and targeted literature reviews to support regulatory submissions.  The Nerac team assists IVD manufacturers with identifying and analyzing clinical data from the literature as it pertains to state of the art, scientific validity, and device performance, using an approach that has been refined over many years supporting manufacturers with EU medical device submissions. Call us at 860.872.7000 or contact us here to learn more!


  1. Richardson A, Signor BM, Lidbury BA, Badrick T. Clinical chemistry in higher dimensions: Machine-learning and enhanced prediction from routine clinical chemistry data. Clinical Biochemistry. 2016;49(16–17):1213–20. doi:10.1016/j.clinbiochem.2016.07.013
  2. Johansson MA, Quandelacy TM, Kada S, Prasad PV, Steele M, Brooks JT, et al. SARS-COV-2 transmission from people without COVID-19 symptoms. JAMA Network Open. 2021;4(1). doi:10.1001/jamanetworkopen.2020.35057
  3. McRae MP, Rajsri KS, Alcorn TM, McDevitt JT. Smart diagnostics: Combining artificial intelligence and in vitro diagnostics. Sensors. 2022;22(17):6355. doi:10.3390/s22176355
  4. Tran NK, Albahra S, May L, Waldman S, Crabtree S, Bainbridge S, et al. Evolving applications of Artificial Intelligence and machine learning in infectious diseases testing. Clinical Chemistry. 2021;68(1):125–33. doi:10.1093/clinchem/hvab239
  5. Izsak K, Terrier A. Artificial Intelligence-based software as a medical device [Internet]. European Commission; 2020 [cited 2023 Jun 13]. Available from:
  6. Goh KH, Wang L, Yeow AY, Poh H, Li K, Yeow JJ, et al. Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare. Nature Communications. 2021;12(1). doi:10.1038/s41467-021-20910-4
  7. Geller J. Food and Drug Administration Issues Final Guidance on Mobile Medical Applications. Journal of Clinical Engineering. 2014;39(1):4–7. doi:10.1097/jce.0000000000000001
  8. Commissioner O of the. Coronavirus (COVID-19) update: FDA authorizes First Machine Learning-based screening device to identify certain biomarkers that may indicate COVID-19 infection [Internet]. FDA; [cited 2023 Jun 12]. Available from:

Get Our Newsletter

Sign up for monthly digest articles to keep up to date with Nerac

Select Your Area(s) of Interest *

Topics of Interest

you might also like