Considerations for applying emerging technologies in paediatric laboratory medicine

Neonatal hypoglycaemia

The first days of life are an important time in neonatal development and adaptation to extrauterine life. The transition from intrauterine homeostasis to self-regulated homeostasis exposes the neonate to potential risks of hypoglycaemia and haemolysis and presents an increased metabolic load to the neonatal liver. Neonatal hypoglycaemia and hyperbilirubinaemia are the most common conditions encountered in the perinatal period. Early recognition is essential, as hypoglycaemia and or hyperbilirubinaemia may cause long term damage to the developing neonate.

The definition of the actual value of hypoglycaemia that correlates with short or long-term complications remains controversial [4, 5]. Recommendations have been developed to identify potential risk factors and to initiate treatment for prevention of hypoglycaemia in this population. Unanswered questions remain [6]. Conventional point-of-care testing (POCT) analysers developed for the adult diabetic population are not well-suited for the neonatal population. The neonatal haematocrit is generally higher in value, and the individual red cells have higher mean corpuscular volume, which results in variability in glucose measurements. In addition, glucometers manifest poor analytical sensitivity at low glucose concentrations [7].

Continuous glucose monitoring (CGM) devices measure interstitial glucose concentrations, and these devices have been introduced for diabetic management in both adult and paediatric patient populations. During the last five years, this technology has been adapted to monitor neonatal glucose concentrations, allowing some of these areas of clinical uncertainty to be studied [8, 9]. The Glucose in Well Babies (GLOW) study has used interstitial glucose sensing electrodes to monitor glucose concentrations during the first four days of life, in parallel with routine laboratory-based methods. The goals were to study early neonatal physiology and to monitor for significant physiological changes [8]. The observation that by day 4, the transitional phase had resolved yielded a better understanding of neonatal glucose physiology.

The use of CGM technology has also been investigated to support management of neonates at risk for persistent hypoglycaemia due to congenital hyperinsulinemia [9, 10]. It has shown that premature and term neonates may be exposed to clinically silent periods of hypoglycaemia, which may or may not cause long term damage [9]. Studies also show that coupling these CGM sensors to closed loop systems which deliver glucose and/or insulin could potentially support adherence to tighter glucose limits and prevent hypoglycaemic episode [11, 12]. The limitations of CGM in this physiologically challenging population must be recognized [9]. In addition, regulatory and licensing issues must be addressed before this technology can be accepted for routine use in this fragile population.

Neonatal hyperbilirubinemia

Recognition of neonatal jaundice and accurate assessment of bilirubin levels remain an ongoing and global clinical issue, especially in low- and middle-income countries which do not have access to appropriate diagnostic technologies. Development of non-invasive near patient solutions have addressed some of these issues. However, these technologies are still lacking in the analytical performance characteristics required to support critical patient management decisions such as phototherapy and exchange transfusions, compared to traditional laboratory-based assessment [13, 14].

Transcutaneous bilirubinometers (TcB) is an established poct technology using direct spectrophotometry which has offered substantial support for screening and identification of early jaundice over the past decade. These meters remain expensive and have limitations due to measurement inaccuracy for serum bilirubin levels >250 umol/L [13, 14]. Emerging technologies such as biosensors show promise but are still in the pre-clinical phase [13]. Trials will determine whether bilirubin concentrations in the skin and sclera can be reliably quantified by Smartphone camera image capture, coupled to software or machine learning (ML) image processing algorithms, [13]. Blood-based POCT technologies based on lateral flow separation, biosensors or light sensors linked to smart phones offer the potential to provide an accessible, quick, cost-effective solutions for lower to middle income countries [13, 14]. Quality programs are essential to ensure operational reliability when these technologies are used in different environments, and to ensure reproducibility against reference methods.

The identification of risks associated with neonatal jaundice such as haemolysis, prematurity, post-natal bruising, macrosomia, and maternal aspects is also essential. Educational programs have been developed to empower parents to identify infants at risk [14, 15]. Risk-based ML models are in development to identify infants who would benefit from phototherapy to prevent early exposure to clinically significant bilirubin concentrations [16]. Others are using metabolomic pathways to predict the length of time required for effective PT [17].

Congenital adrenal hyperplasia

Congenital adrenal hyperplasia (CAH) encompasses a group of autosomal recessive disorders characterized by the adrenal cortex’s impaired ability to synthesize cortisol from cholesterol. Deficiency in the 21-hydroxylase (21-OHD) is due to mutations in the CYP21A2 gene and accounts for more than 95 % of CAH cases, resulting in the defective conversion of 17-hydroxyprogesterone (17OHP) to 11-deoxycortisol [18]. Overall incidence of the classic form of 21-OHD is 1 in 15,000 live births as determined by neonatal screening studies [19].

The clinical presentation of CAH can range from severe to mild forms, depending on the degree of enzyme deficiency. Historically, infants with classic CAH were categorized into “salt-wasting” and “simple-virilizing” forms. However, these terms are now considered less useful as there is considerable overlap in phenotype, e.g. all patients lose salt and girls with either type have atypical genitalia. Therefore, patients should be categorized into classic and non-classic CAH only. General genotype-phenotype correlation exists but is complex [20]. The “salt-wasting” form is the most severe, usually caused by deletions or mutations with null enzyme activity. The “simple-virilizing” form is less severe, with 1–2 % enzyme activity due to specific point mutations. The non-classic form is the least severe, retaining 5–20 % activity [21]. Compound heterozygotes have the phenotype of the less severe mutation, while heterozygotes may have mild biochemical abnormalities but no significant disorder [22].

Diagnosis of CAH is typically made at birth through mandated neonatal screening. In many countries, screening begins with the measurement of 17OHP in a dried blood spot obtained from a heel-stick. The diagnosis of classic CAH is based on significantly elevated levels of 17OHP. Most affected infants have concentrations greater than 105–150 nmol/L [18]. Confirmatory and additional testing may be necessary in cases where results are equivocal. False-positive results are common in premature infants, and reference intervals for screening are often adjusted based on weight and gestational age. False-negative results can be up to 22 %, especially in girls [23, 24]. It can also occur due to maternal antenatal glucocorticoid use. Utility of 21-deoxycortisol by LC-MS/MS with appropriate isomeric separation has recently been recommended as the best practice for identifying classical 21-OHD CAH [25, 26]. Treatment of classic CAH involves lifelong glucocorticoid replacement therapy to compensate for the deficient hormone production. The specific treatment approach may vary depending on the age of the patient. In addition, expanded carrier screening for CAH has been proposed which may aid in utero treatment of affected foetus [27, 28].

Sickle cell anaemia

Sickle cell disease (SCD) is an autosomal recessive hemoglobinopathy caused by the presence of abnormal haemoglobin (Hb) S. HbS polymerizes with deoxygenation, leading to rigid red blood cells that are prone to lysis and interaction with leucocytes, platelets, and the vascular endothelium. The main manifestations of SCD are anaemia (due to shortened half-life of erythrocytes) and painful vaso-occlusive crises that impair circulation and cause poor tissue oxygenation. Some of the signs and symptoms experienced by SCD patients (acute chest pain, stroke, and priapism) constitute medical emergencies. In addition to these acute manifestations, chronic sequelae result from the combination of haemolytic anaemia and functional organ impairments following vaso-occlusive crises (VOC). For example, functional asplenia caused by splenic infarctions renders patients susceptible to infections by encapsulated bacteria (e.g. Streptococcus pneumoniae, Neisseria meningitidis, and others). The clinical expression of SCD varies widely from one patient to another: some patients suffer multiple episodes of acute chest pain syndrome and VOCs beginning early in life, whereas others develop very few complications. A third patient category presents with cerebrovascular signs and symptoms and are at high risk of stroke in childhood or adolescence [29, 30].

Over the past decade, many countries established newborn screening programs to identify SCD using mass spectroscopy of dried blood spots. Additionally, successful pilots of non-invasive prenatal diagnosis for at-risk pregnancies have used targeted massively parallel sequencing assay of foetal cell-free DNA from maternal plasma. This approach has no requirement for paternal or proband samples. This technique significantly improved the detection rates of at-risk pregnancies from <50 to >98 % [31, 32].

Genomics research could accelerate progress for SCD patients in three ways. First, investigation of hereditability of the strongest known modifier of SCD clinical expression, Hemoglobin F (HgbF) among genetically highly heterogenous patient populations is important, along with the identification of novel therapeutic targets for HbF induction. Second, the search for RNA therapies is a high priority, using microRNA to target HbF protein production by binding to cellular transcription machinery, or by directly mediating production of HbF or HbA through introduction of therapeutic messenger RNA. Global research efforts are underway to uncover genomic keys to unlock SCD therapeutics. Third, research to identify currently unknown genetic risk factors for SCD cardiovascular complications could contribute to mortality risk reduction. Encouraging results using next-generation sequencing (NGS), whole genome sequencing and/or whole exome sequencing have identified genes responsible for more severe expression of disease features e.g. APOL1 G1 (rs73885319/rs6090145) and G2 (rs71785313) variants, responsible for severe renal insufficiency in patients with SCD [33], [34], [35].

Metabolomic analysis using liquid chromatography-mass spectrometry could allow for characterization of the endogenous and exogenous effects of potential new treatments. Studies of SCD patients’ red cell metabolome revealed important metabolic abnormalities of glutathione and nitric oxide biosynthetic pathways. Altered concentrations of the metabolites serving as substrates in these cycles was observed. Low levels of L-phenylamine and L-tyrosine, essential sources for multiple neurotransmitters biosynthesis and neurobehavioral health, were evident. The complex pathophysiology of SCD makes unlikely that a single therapeutic agent will prevent or reverse all SCD complications [36].

The discovery of clustered, regularly interspaced short palindromic repeats (CRISPR) and CRISPR-associated protein 9 (Cas9) systems has revolutionized gene therapy (GT) by enabling precise gene targeting. Gene therapy using CRISPR-Cas9 has emerged as a potential treatment for SCD, by correcting the mutation in patients’ cells. One of the significant challenges in utilizing CRISPR-Cas9 for treating SCD is ensuring that the system specifically targets only the mutated gene, without affecting other genes in the genome. The recognition of myeloid malignancies in SCD patients post-GT underscores the need to better understand clonal selection of HSCs after GT and find ways to avoid clonal haematopoiesis [37, 38].

Microassays

The development of novel techniques which provide definitive analysis on a minimal volume of sampled blood (<100 μL or less) offers enormous potential reduction in blood loss for pediatric populations. To date, the promise of multiple analyses performed on a single microsample has been explored with limited success. A variety of platforms have been used in these investigations, including nano fluidics, spectroscopy, polymerase chain reaction (PCR) analyses and multi-omics. The minimal volume specifications, ease of transport and storage, stability and biobank compatibility are well-established advantages of dried blood spots collected for newborn screening. Dried blood spots were recently reported to allow for combined genomic/metabolomic analyses, including four genomic, three epigenomic, four proteomic and 11 metabolomic studies. The strategy of high throughput untargeted assays followed by targeted analyses, as demonstrated in this study, is a promising path for biomarker discovery. Wider use of these techniques, applied to dried blood spots, could facilitate large population-based studies to advance the understanding of neonatal disease [39].

One emerging area of innovation is the use of molecularly imprinted polymer-quantum dot materials (MIPS) as components of optical sensors used for targeted applications [40]. MIPS are synthesized with precisely designed recognition sites to selectively bind a target molecule. MIPS are combined with quantum dots (QDs), which are fluorescent semiconductor nanomaterials with unique chemical and electrical properties – a new type of fluorescent probe. Fluorescent probes have sensitivity for low limits of detection and are combined with MIPS which have high selectivity. These MIP-QDs sensors produce a quantified optical change during binding. The intensity of fluorescence quenching is used for quantification. Synthetic strategies are developed and refined to improve performance. The chosen synthetic strategy divides MIP-QDs into several categories based on chemical composition. To date, silica-based MIP-QDs are most often reported; there are also organic MIP-QDs, and hybrid inorganic-organic MIP-QDs. Regardless of synthetic type, organic imprinted polymers are often employed to capture the target because their composition can be precisely designed.

Sensor microdevices comprise another class of novel technology and can be designed to use microfluidic test strips for the point of care [41]. Sample fluid is transported pump-free, through capillary action on paper. Distance-based paper analytical devices (dPADs) use a signal readout that is a naked-eye distance measurement of color change in an analyte – reagent reaction channel. The distance length is proportional to analyte concentration. Sample and reagent requirements are only approximately 1 μL. This measurement can be performed by untrained individuals, is low cost and provides rapid results.

For under-resourced settings, traditional analytical methods require highly trained personnel, sample processing, labor costs, recurring costs for proprietary and often expensive reagents, and capital expenditures for equipment. MIP-QDs optical sensors offer simplicity and are potentially adaptable to a range of target analytes and sample matrices. dPADs offer design simplicity and do not require highly trained testing personnel. Oversight of novel technologies will vary by local governance and site, but shared ethical principles would require assessment of patient risk-benefit to guide adoption and protect patients.

Noninvasive testing and alternative matrices

There is increased appreciation of the impact of pain on the developing infant and child. The most common invasive procedure in early infancy is a capillary puncture, and in children, venipuncture. The exploration of applying novel technologies to use alternative sample matrices for pain-free diagnostic procedures has followed [42].

Noninvasive collection reduces or eliminates pain associated with phlebotomy for children and offers the potential of moving testing closer to the patient. With noninvasive collection, sample procurement and analytic testing can, and has, moved into the home. Alternative matrices which can be noninvasively collected and analyzed in children include sweat, saliva, tears, hair, nails, urine, and meconium (in newborns) [42, 43].

An early recognition that analysis of elevated chloride content in sweat could diagnose cystic fibrosis led to the use of pilocarpine iontophoresis. Iontophoresis uses application of pilocarpine to the skin and electrical stimulation to increase sweat production. For many years the sweat test was, and continues to be, the gold standard for cystic fibrosis testing [44]. It can cause transient skin irritation, and very rarely can cause burns, but is needleless.

Saliva has proved to be a highly useful matrix, collected either by passive drooling or after an oral sodium chloride rinse. Saliva has long been a matrix of choice for accurate and reliable HIV testing. The use of alternative, accessible and noninvasive matrices such as saliva is enormously helpful for pediatric and adolescent patients with needle phobia, for employee safety and for cost reduction in resource-poor areas. More recently, a powerful emerging technology was developed for detection of subclinical malaria in saliva [45]. These subclinical cases cannot be diagnosed by rapid diagnostic tests or microscopy because of low density of organisms. Ongoing malaria parasite transmission is principally attributed to low-density subclinical infections not reliably detected by available rapid diagnostic tests (RDTs) or microscopy. First, proteomic analysis was performed by LC/MS-MS on saliva samples from 364 children in Cameroon and Zambia with subclinical malaria, which identified a candidate protein biomarker present in these subclinical cases. This was followed by designing a lateral-flow immunoassay rapid diagnostic test targeting this protein marker in saliva samples of P. falciparum gametocyte carriers. The sensitivity was 83 % compared to multiplex PCR and 79 % compared to microscopy.

Meconium is another alternative matrix, entirely unique to the fetus. It is the fetal waste product, slowly accumulated over the second and third trimesters of gestation. Meconium analysis represents a retrospective evaluation of fetal in utero exposures and is a highly informative and valuable diagnostic tool [42]. Once meconium is deposited, this matrix represents a stable record of antenatal exposures. Complete evacuation of meconium is achieved within the first 50–150 h after birth. Therefore, meconium collection is ideally within the first 72 h or so of postnatal life [46]. After that time the matrix is a transitional one, changing from meconium to feces. An exception to this guidance is made for preterm, extremely low birth weight infants, in whom the evacuation of meconium is substantially delayed. In these patients, the median age at first stool is 3 days, with 90 % of infants evacuating meconium by 12 postnatal days [47]. A principal application for meconium analysis is testing for maternal substance abuse. If testing is positive, this can result in engagement of child protective services (depending on the approach of the locale, context and clinical assessments). Meconium can also be used for assessment of prenatal exposure to environmental toxins [42].

There is increasing recognition of the impact of environmental toxins on the developing fetus. An alternative matrix of note for monitoring prenatal exposure to methylmercury and other heavy metals is maternal hair analysis. Hair is a historical record of concentrations of transportable species in plasma. The fetus is extremely sensitive to neurotoxic effects on the developing central nervous system. Hair is analyzed as a surrogate for fetal tissue (brain) concentrations [48]. In remote settings around the world, there is increasing recognition of the effect of toxic environmental exposures on developing infants and children. This technology is applied in novel ways to identify and quantify children’s risk. For example, hair analysis in children living near a tungsten-molybdenum mine in Russia [49], and adolescents living near a petrochemical plant in Italy [50] were recently shown to have substantial heavy metal exposures by hair analysis. Hair is best analyzed by careful washing protocols to remove interferents and external contaminants, homogenization and extraction, followed by screening testing and – if presumptive positive-a confirmatory test. Methods range from rapid immunoassays to inductively coupled plasma-mass spectrometry (ICP-MS) [51]. Properly cleaned hair is an ideal testing matrix in several respects: noninvasive collection, easy and inexpensive transport and storage with stable specimen integrity.

Children’s fingernail clippings can also be used for a variety of investigations. For example, a systematic review of prenatal exposure assessments for four drugs (nicotine, cannabis, alcohol and opioids) using nail clippings from pregnant individuals or mother-child dyads found this matrix to be acceptable. A dose-response relationship between exposure and analyte concentration in nail clippings was evident. Advantages include minimal invasiveness, broad exposure time period allowing for infrequent sampling and lower analytic detection level compared with hair; disadvantages are contamination risk and insufficient sample quantity risk [52].

Lastly, human machine interfaces (HMIs) are being designed to capture the biophysical status of an individual – they have “biological perception” [53]. Fingertips are the site at which investigators chose to engineer the HMI. The fingertip functions as identity authentication. Circulating biomarker molecules move onto the fingertip skin, in sweat. A thin hydrogel-based electrochemical sensor is employed as a medium for analyte sampling at the fingertip. The wait time between touch and sensor activation is 30 s. The influx of analytes travels from the fingertip into the hydrogel surface and sensor. The amperometric sensor was designed to detect the alcohol-free state vs. the recent intake of alcohol state for a driving activation application. The voltametric sensor was designed to detect the acetaminophen-free state vs. the state of circulating acetaminophen. This latter sensor was for a medication dispensation system to avoid overdosage of the potentially hepatotoxic drug.

Next generation sequencing

In recent years the cost of NGS, also known as massively parallel sequencing has decreased and NGS-based testing has now become the main methodology in clinical molecular diagnostic laboratories. Genomic medicine is no longer the future of medicine but has rather become an integral method in routine testing and is rapidly revolutionising the healthcare system. NGS methods are high throughput in orders of magnitude and allow for sequencing of the entire genome at a low cost and within a few days. There are several different NGS platforms using different sequencing technologies.

Most clinical laboratories use NGS assays that are designed to detect a specific set of genes and are referred to as gene panels. These assays can have as few as one gene or as many as hundreds of genes. For inherited diseases, NGS methods are now routinely being used for several different disorders such as cardiac syndromes, epilepsy, connective tissue, hereditary hearing loss, immune deficiencies, bone marrow failure syndromes, renal disorders, and inborn errors of metabolism [54]. NGS gene panel assays are also used for the evaluation of aneuploidy in foetuses by prenatal testing using foetal cell-free DNA, collected from a simple maternal blood phlebotomy sample, commonly referred to as NIPT (discussed in the example of sickle cell anaemia above).

In cancer, targeted panels are used for cancer predisposition syndromes as well as somatic cancer. For the first time, large gene panel NGS assays are providing an opportunity to effectively evaluate changes between cancer and normal tissues and significantly impact therapeutic choices in paediatric cancer patients [55].

Whole exome sequencing (WES) has also been used as a clinical test, particularly in paediatrics for the diagnosis of rare genetic disorders. The assay is targeting only the protein coding genes of the genome (∼1 % of the entire genome). Whole exome sequencing has been shown to provide a diagnosis in patients with a rare disease for which specific targeted panels do not exist.

In the last two years, whole genome sequencing (WGS) has emerged as a more comprehensive genetic test and is being integrated into health care systems internationally. WGS assays are designed to sequence the entire genome which includes both coding and non-coding regions of the genome (the introns between the exons of a gene and the sequences between the genes) and at this time, can diagnose >6,000 genetic disorders. WGS has the additional advantage that all types of variants in the genome can be detected, and thereby avoid the need for other technologies such as chromosomal microarray analysis.

As a result of using WGS, several new genes have been discovered and clinical phenotypes have been expanded. There is an increasing amount of evidence that supports the role of WGS in paediatric settings with respect to both diagnostic yield and time to diagnosis (primary outcome measures) as well as cost-effectiveness (secondary outcome). Particularly high yields have been seen in cohorts of patients with severe intellectual disability (50 % of patients). The applications and potential uses of WGS are broad and can address many different clinical questions. Importantly, two randomized controlled trials with rapid turnaround time have demonstrated increased diagnostic yields in neonatal and paediatric intensive care settings [56], [57], [58], [59].

Metagenomic assays are another emerging application of NGS in paediatric laboratory medicine for the investigation of pyrexia of unknown origin [60]. Under this approach hospitalised children with prolonged fever but without definitive microbial aetiology, despite routine microbiological work up are subject to non-targeted sequencing of body fluids, including blood. Consequently, the anti-microbial regime is adjusted based on whether genetic microbial signatures are detected that may explain the clinical presentation of the patient. The metagenomic approach may also be useful for detecting novel pathogens as well as routine surveillance and outbreak investigation [61]. Nonetheless, the application of this technique should be carefully considered within the local diagnostic workflow and disease prevalence to ensure cost-effective service delivery [62].

Although adoption of genome sequencing is challenging, it is anticipated that WGS will become a first-tier genetic test in children with complex clinical presentations. This has the potential to reduce the time to diagnosis, the number of tests performed and the overall cost.

Exosome analysis

Exosomes are nano-sized (30–150 nm in diameter) extracellular vesicles that are secreted from cells upon fusion of multivesicular bodies and the plasma membrane [63, 64]. They can be found circulating in peripheral blood and in various body fluids including urine, saliva, breast milk, cerebrospinal fluids, semen, amniotic fluid, and ascites. Exosomes can cross the blood brain barrier bidirectionally and are involved in the modulation of the barrier itself [65]. Exosomes are extra-cellular vehicles that carry rich biomolecules such as proteins, lipids, and genetic materials such as DNA, micro-RNA, and non-coding RNAs. These biomolecules can provide information about the origin and status of the tissue and are involved in certain cell-to-cell and microenvironment regulations. Exosomes are considered as both emerging diagnostic and therapeutic targe.

Exosomes have been studied in several paediatric conditions including childhood obesity [64], paediatric sleep apnoea [66], and brain tumours such as glioma [67, 68]. In childhood obesity, the exosomal micro-RNA and long non-coding RNA may be involved in gene regulations related to the pathophysiological processes of adipocyte biology that may contribute to development of obesity and related metabolic and inflammatory disorders [57]. Obstructive sleep apnoea is a relatively common condition in children, is commonly caused by enlarged upper airway lymphoid tissues and is worsened by coexisting obesity. Obstructive sleep apnoea is associated with adverse cardiometabolic, neurobehavioural and cognitive functions. Recently, biofunction of exosomes have been hypothesised to differ between children with obstructive sleep apnoea with and without neurocognitive alterations and is mediated by disruption of the blood brain barrier by the exosomes, leading ultimately to brain tissue dysfunction and loss [58]. At the same time, exosomal micro-RNA profiles are differentially expressed between healthy and tumour cells and may play a role in tumorigenesis and malignant features through a signalling pathway [67, 68]. By contrast, cellular micro-RNA did not differ between cells from the healthy tissue and tumour cells.

Traditionally, the isolation of exosome requires significant pre-analytical processing such as ultra-centrifugation. This process may be time consuming and requires specialised equipment. More recently, significant advancements have been made in the isolation and detection of exosomes using biosensors that allows direct on-chip isolation of exosomes and high dimension (multiplexed) direct detection of target biomolecules such as RNA [69, 70]. These technologies may be exploited as novel tools for understanding the pathophysiology of diseases, diagnostic testing, and treatment monitoring. Importantly, exosomes, as a novel class of biomarker, opens the possibility of non-invasive diagnostics for tissues that are currently difficult to access, such as the brain.

Machine learning

Machine learning (ML), a type of artificial intelligence, is increasingly employed in all areas of laboratory medicine. In paediatric laboratory medicine, it is being used to analyse large data sets to identify patterns of analytes in clinical disease and to prioritise investigations [71], [72], [73], [74], [75]. ML models are used to predict outcomes based on interpreting a set of data against a previously learned data set. The task is completed in a significantly shorter time than if performed by a human operator. ML does not negate the responsibilities of laboratory professionals to provide analytical and clinical interpretations of data. It can support workload prioritisation; or identification of the best test to perform when resources are limited [71], [72], [73], [74], [75]. In inherited metabolic disease diagnosis, ML algorithms have been developed to interpret plasma amino acids profiles. In this way, ML can support laboratories with large workloads, but limited staff or experience to interpret them [72]. Computer vision, a sub-set of ML, has been used to develop a model to accurately identify unsuitable newborn blood spot screening specimens based on spot diameter measurements. When compared with the human eye of the punch operator, these measurements improved the quality of spots analysed [73].

ML models can also be used to refine a diagnostic model to predict which biomarkers, risk factors, clinical signs and symptoms have the most significance [74, 75]. Neonatal early-onset sepsis is an area where ML algorithms have been used to identify those biomarkers such as C-reactive protein and white blood cell counts as key factors that should be included in management models, especially in pre-selected individuals who may been started on prophylactic antibiotics [74]. A similar approach has also been used to develop an online appendicitis prediction tool based on several factors allowing a more personalised approach to a child’s care [75]. However, as with all emerging technologies the process requires refinement over time. Additional validation, access to additional diverse and larger data sets, especially for rare conditions, will help refine the models and further improve their performance [71], [72], [73], [74], [75].