The Reliable, Automatic Classification of Neonates in First-Tier MALDI-MS Screening for Sickle Cell Disease

Marven El Osta; Pierre Naubourg; Olivier Grunewald; Gilles Renom; Patrick Ducoroy; Jean Marc Périni

doi:10.3390/ijns5030031

The Reliable, Automatic Classification of Neonates in First-Tier MALDI-MS Screening for Sickle Cell Disease

Int J Neonatal Screen. 2019 Aug 31;5(3):31. doi: 10.3390/ijns5030031. eCollection 2019 Sep.

Authors

Marven El Osta¹, Pierre Naubourg¹, Olivier Grunewald², Gilles Renom², Patrick Ducoroy¹, Jean Marc Périni²

Affiliations

¹ Biomaneo, 22B boulevard Winston Churchill, F-21000 Dijon, France.
² Newborn Screening Laboratory, Biology and Pathology Center, Lille University Medical Centre, F-59000 Lille, France.

Abstract

Previous research has shown that a MALDI-MS technique can be used to screen for sickle cell disease (SCD), and that a system combining automated sample preparation, MALDI-MS analysis and classification software is a relevant approach for first-line, high-throughput SCD screening. In order to achieve a high-throughput "plug and play" approach while detecting "non-standard" profiles that might prompt the misclassification of a sample, we have incorporated various sets of alerts into the decision support software. These included "biological alert" indicators of a newborn's clinical status (e. g., detecting samples with no or low HbA), and "technical alerts" indicators for the most common non-standard profiles, i.e., those which might otherwise lead to sample misclassification. We evaluated these alerts by applying them to two datasets (produced by different laboratories). Despite the random generation of abnormal spectra by one-off technical faults or due to the nature and quality of the samples, the use of alerts fully secured the process of automatic sample classification. Firstly, cases of β-thalassemia were detected. Secondly, after a visual check on the tagged profiles and reanalysis of the corresponding biological samples, all the samples were correctly reclassified without prompting further alerts. All of the samples for which the results were not tagged were well classified (i.e., sensitivity and specificity = 1). The alerts were mainly designed for detecting false-negative classifications; all the FAS samples misclassified by the software as FA (a false negative) were marked with an alert. The implementation of alerts in the NeoScreening^® Laboratory Information Management System's decision support software opens up perspectives for the safe, reliable, automated classification of samples, with a visual check solely on abnormal results or samples. It should now be possible to evaluate the combination of the NeoSickle^® analytical solution and the NeoScreening^® Laboratory Information Management System in a real-life, prospective study of first-line SCD screening.

Keywords: Laboratory Information Management System (LIMS); MALDI-MS; automatic classification; high-throughput SCD screening; sickle cell disease; software; β-thalassemia.