Machine learning to improve false-positive results in the Dutch newborn screening for congenital hypothyroidism

OBJECTIVE

The Dutch Congenital hypothyroidism (CH) Newborn Screening (NBS) algorithm for thyroidal and central congenital hypothyroidism (CH-T and CH-C, respectively) is primarily based on determination of thyroxine (T4) concentrations in dried blood spots, followed by thyroid-stimulating hormone (TSH) and thyroxine-binding globulin (TBG) measurements enabling detection of both CH-T and CH-C, with a positive predictive value (PPV) of 21%. A calculated T4/TBG ratio serves as an indirect measure for free T4. The aim of this study is to investigate whether machine learning techniques can help to improve the PPV of the algorithm without missing the positive cases that should have been detected with the current algorithm.

DESIGN & METHODS

NBS data and parameters of CH patients and false-positive referrals in the period 2007-2017 and of a healthy reference population were included in the study. A random forest model was trained and tested using a stratified split and improved using synthetic minority oversampling technique (SMOTE). NBS data of 4668 newborns were included, containing 458 CH-T and 82 CH-C patients, 2332 false-positive referrals and 1670 healthy newborns.

RESULTS

Variables determining identification of CH were (in order of importance) TSH, T4/TBG ratio, gestational age, TBG, T4 and age at NBS sampling. In a Receiver-Operating Characteristic (ROC) analysis on the test set, current sensitivity could be maintained, while increasing the PPV to 26%.

CONCLUSIONS

Machine learning techniques have the potential to improve the PPV of the Dutch CH NBS. However, improved detection of currently missed cases is only possible with new, better predictors of especially CH-C and a better registration and inclusion of these cases in future models.

Second-tier Testing for 21-Hydroxylase Deficiency in the Netherlands: A Newborn Screening Pilot Study

CONTEXT

Newborn screening (NBS) for classic congenital adrenal hyperplasia (CAH) consists of 17-hydroxyprogesterone (17-OHP) measurement with gestational age-adjusted cutoffs. A second heel puncture (HP) is performed in newborns with inconclusive results to reduce false positives.

OBJECTIVE

We assessed the accuracy and turnaround time of the current CAH NBS algorithm in comparison with alternative algorithms by performing a second-tier 21-deoxycortisol (21-DF) pilot study.

METHODS

Dried blood spots (DBS) of newborns with inconclusive and positive 17-OHP (immunoassay) first HP results were sent from regional NBS laboratories to the Amsterdam UMC Endocrine Laboratory. In 2017-2019, 21-DF concentrations were analyzed by LC-MS/MS in parallel with routine NBS. Diagnoses were confirmed by mutation analysis.

RESULTS

A total of 328 DBS were analyzed; 37 newborns had confirmed classic CAH, 33 were false-positive and 258 were categorized as negative in the second HP following the current algorithm. With second-tier testing, all 37 confirmed CAH had elevated 21-DF, while all 33 false positives and 253/258 second-HP negatives had undetectable 21-DF. The elevated 21-DF of the other 5 newborns may be NBS false negatives or second-tier false positives. Adding the second-tier results to inconclusive first HPs reduced the number of false positives to 11 and prevented all 286 second HPs. Adding the second tier to both positive and inconclusive first HPs eliminated all false positives but delayed referral for 31 CAH patients (1-4 days).

CONCLUSION

Application of the second-tier 21-DF measurement to inconclusive first HPs improved our CAH NBS by reducing false positives, abolishing the second HP, and thereby shortening referral time.

Improving the Dutch Newborn Screening for Central Congenital Hypothyroidism by Using 95% Reference Intervals for Thyroxine-Binding Globulin

INTRODUCTION

Newborn screening (NBS) for congenital hypothyroidism (CH) in the Netherlands consists of thyroxine (T4), thyroid-stimulating hormone (TSH), and T4-binding globulin (TBG) measurements to detect thyroidal CH and central CH (CH-C). CH-C is detected by T4 or a calculated T4/TBG ratio, which serves as an indirect measure of free T4. TSH and TBG are only measured in the lowest 20 and 5% of daily T4 values, respectively. A recent evaluation of the Dutch NBS for CH showed that the T4 and T4/TBG ratio contribute to the detection of CH-C but also lead to a low positive predictive value (PPV). Dried blood spot (DBS) reference intervals (RIs) are currently unknown and may contribute to improvement of our NBS algorithm.

MATERIALS AND METHODS

RIs of T4, TSH, TBG, and the T4/TBG ratio were determined according to Clinical & Laboratory Standards Institute guidelines in heel puncture cards from routine NBS in both sexes and at the common NBS sampling ages. Scatter plots were used to compare the healthy reference population to previously published data of CH-C patients and false positives.

RESULTS

Analyses of 1,670 heel puncture cards showed small differences between subgroups and led to the formulation of total sample DBS RIs for T4 (56-118 nmol/L), TSH (<2.6 mIU/L), TBG (116-271 nmol/L), and the T4/TBG ratio (>20). 46% of false-positive referrals based on T4 alone had a TBG below the RI, indicating preventable referral due to partial TBG deficiency. One case of CH-C also had partial TBG deficiency (TBG 59 and T4 12 nmol/L blood).

DISCUSSION/CONCLUSION

Established DBS RIs provided possibilities to improve the PPV of the Dutch CH NBS algorithm. We conclude that by taking partial TBG deficiency into account, approximately half of T4 false-positive referrals may be prevented while maintaining NBS sensitivity at the current level.

Critical evaluation of the newborn screening for congenital hypothyroidism in the Netherlands

OBJECTIVE

Congenital hypothyroidism (CH) is defined as thyroid hormone deficiency at birth due to disorders of the thyroid gland (thyroidal CH, CH-T), or the hypothalamus or pituitary (central CH, CH-C). The Dutch Newborn Screening (NBS) strategy is primarily based on determination of thyroxine (T4) concentrations in dried blood spots followed, if necessary, by thyroid-stimulating hormone (TSH) and thyroxine-binding globulin (TBG) measurement enabling detection of both CH-T and CH-C. A calculated T4/TBG ratio serves as an indirect measure for free T4. A T4/TBG ratio ≤ 17 in a second heel puncture is suggestive of CH-C.

DESIGN AND METHODS

In the present study, we evaluated 11 years of Dutch CH NBS using a database of referred cases by assessing the contribution of each criterion in the unique stepwise T4-TSH-TBG NBS algorithm.

RESULTS

Between 2007 and the end of 2017, 1 963 465 newborns were screened in the Netherlands. Use of the stepwise algorithm led to 3044 referrals and the identification of 612 CH cases, consisting of 496 CH-T, 86 CH-C, and 30 CH of unknown origin diagnoses. We detected 62.8% of CH-C cases by the T4/TBG ratio in the second heel puncture. The positive predictive value (PPV) of the stepwise T4-TSH-TBG NBS algorithm was 21.0%.

CONCLUSION

This evaluation shows that the Dutch stepwise T4-TSH-TBG NBS algorithm with a calculated T4/TBG ratio is of great value for the detection of both CH-T and CH-C in the Netherlands, at the cost of a lower PPV compared to TSH-based NBS strategies.

Evaluation of 11 years of newborn screening for maple syrup urine disease in the Netherlands and a systematic review of the literature: Strategies for optimization

Maple syrup urine disease (MSUD) leads to severe neurological deterioration unless diagnosed early and treated immediately. We have evaluated the effectiveness of 11 years of MSUD newborn screening (NBS) in the Netherlands (screening >72 hours, referral if both total leucine (Xle) and valine ≥400 μmol/L blood) and have explored possibilities for improvement by combining our data with a systematic literature review and data from Collaborative Laboratory Integrated Reports (CLIR). Dutch MSUD NBS characteristics and accuracy were determined. The hypothetical referral numbers in the Dutch population of additional screening markers suggested by CLIR were calculated. In a systematic review, articles reporting NBS leucine concentrations of confirmed patients were included. Our data showed that NBS of 1 963 465 newborns identified 4 MSUD patients and led to 118 false-positive referrals (PPV 3.28%; incidence 1:491 000 newborns). In literature, leucine is the preferred NBS parameter. Total leucine (Xle) concentrations (mass-spectrometry) of 53 detected and 8 false-negative patients (sampling age within 25 hours in 3 patients) reported in literature ranged from 288 to 3376 (median 900) and 42 to 325 (median 209) μmol/L blood respectively. CLIR showed increasing Xle concentrations with sampling age and early NBS sampling and milder variant MSUD phenotypes with (nearly) normal biochemical profiles are causes of false-negative NBS results. We evaluated the effect of additional screening markers and established the Xle/phenylalanine ratio as a promising additional marker ratio for increasing the PPV, while maintaining high sensitivity in the Dutch MSUD NBS.

Recommendations for newborn screening for galactokinase deficiency: A systematic review and evaluation of Dutch newborn screening data

INTRODUCTION

Galactokinase (GALK) deficiency causes cataract leading to severe developmental consequences unless treated early. Because of the easy prevention and rapid reversibility of cataract with treatment, the Dutch Health Council advised to include GALK deficiency in the Dutch newborn screening program. The aim of this study is to establish the optimal screening method and cut-off value (COV) for GALK deficiency screening by performing a systematic review of the literature of screening strategies and total galactose (TGAL) values and by evaluating TGAL values in the first week of life in a cohort of screened newborns in the Netherlands.

METHODS

Systematic literature search strategies in OVID MEDLINE and OVID EMBASE were developed and study selection, data collection and analyses were performed by two independent investigators. A range of TGAL values measured by the Quantase Neonatal Total Galactose screening assay in a cohort of Dutch newborns in 2007 was evaluated.

RESULTS

Eight publications were included in the systematic review. All four studies describing screening strategies used TGAL as the primary screening marker combined with galactose-1-phosphate uridyltransferase (GALT) measurement that is used for classical galactosemia screening. TGAL COVs of 2200 μmol/L, 1665 μmol/L and 1110 μmol/L blood resulted in positive predictive values (PPV) of 100%, 82% and 10% respectively. TGAL values measured in the newborn period were reported for 39 GALK deficiency patients with individual values ranging from 3963 to 8159 μmol/L blood and 2 group values with mean 8892 μmol/L blood (SD ± 5243) and 4856 μmol/L blood (SD ± 461). Dutch newborn screening data of 72,786 newborns from 2007 provided a median TGAL value of 110 μmol/L blood with a range of 30-2431 μmol/L blood.

CONCLUSION

Based on TGAL values measured in GALK deficiency patients reported in the literature and TGAL measurements in the Dutch cohort by newborn screening we suggest to perform the GALK screening with TGAL as a primary marker with a COV of 2500 μmol/L blood, combined with GALT enzyme activity measurement as used in the classical galactosemia screening, to ensure detection of GALK deficiency patients and minimize false positive referrals.

Metformin-related colonic glucose uptake; potential role for increasing glucose disposal?--A retrospective analysis of (18)F-FDG uptake in the colon on PET-CT

AIM

The use of metformin has been associated with diffusely increased colonic (18)F-fluorodeoxyglucose ((18)F-FDG) uptake. Interestingly, metformin use is associated with moderate weight loss. It could be hypothesized that increased colonic glucose disposal is related to this weight loss. It is unknown whether other factors influence (18)F-FDG uptake in the colon. The aim of this study was to retrospectively assess independent determinants of colonic (18)F-FDG uptake.

METHODS

We retrospectively analysed 270 (18)F-FDG PET-CTs which were made for diagnostic purposes. Colonic (18)F-FDG uptake was assessed using a 4-point scale using the liver as a reference (1; lower, 2; similar, 3; moderately higher than hepatic activity, 4; intense diffuse increased uptake). Determinants of (18)F-FDG uptake in the colon were assessed using forward logistic regression (i.e., grade 1&2 vs 3&4).

RESULTS

The patients had a mean age of 60.2±14.8 years, a BMI of 25.8±5.2kg/m(2) and 52% were female. Most patients had a grade 2 (44%) or grade 3 (39%) (18)F-FDG uptake in the colon. Diabetes mellitus type 2 was observed in 14% of the patients. In total, 5% of the patients used insulin, 12% used metformin and 5% used sulfonylurea derivatives (SU). While there seemed to be an effect of SU on (18)F-FDG uptake in the ileum [OR 3.6 (95% CI: 1.3-33.1), p=0.03], metformin was the only drug associated with (18)F-FDG uptake for both the whole colon [OR 10.0 (95% CI: 2.9-34.7), p<0.001] and all individual segments.

CONCLUSION

Metformin use is an independent determinant of increased colonic (18)F-FDG uptake, suggesting a potential role for increasing colonic glucose disposal.

The Brown-Vialetto-Van Laere and Fazio Londe syndrome revisited: natural history, genetics, treatment and future perspectives

The Brown-Vialetto-Van Laere syndrome is a rare neurological disorder which may present at all ages with sensorineural deafness, bulbar palsy and respiratory compromise. Fazio-Londe syndrome is considered to be the same disease entity. Recently it was demonstrated that in some patients the disease is caused by mutations in the SLC52A3 gene which encodes the intestinal (hRFT2) riboflavin transporter. In these patients riboflavin deficiency is the cause of the BVVL/FL syndrome and supplementation of riboflavin proved a life saving treatment. Mutations in the SLC52A2 gene and the SLC52A1 (GPR172B) gene, coding for human riboflavin transporters hRFT3 and hRFT1 have been associated with the BVVL syndrome as well. We performed a review of the literature, with emphasis on the natural history and the effects of treatment in these patients. A total of 35 publications were traced reporting on the clinical presentation of 74 patients who presented before age 18. The most prevalent symptoms were bulbar palsy, hearing loss, facial weakness and respiratory compromise. Death was reported in 28 of the 61 untreated patients, with a very low survival in patients presenting before age 4. All 13 patients who were treated with riboflavin survived, with a strong clinical improvement after days to months of treatment in eight patients. Three patients demonstrated a stable clinical course and treatment was stopped early in two patients. Abnormalities in plasma flavin levels and/or plasma acylcarnitine profiles were observed in some but not in all patients, and also patients with normal plasma flavin levels and acylcarnitine profiles demonstrated a striking clinical improvement on riboflavin supplementation. It is now clear that proper diagnosis requires mutation analysis of all three transporter genes and treatment should be started immediately without first awaiting results of molecular analysis. Clinical improvement may be rapid or gradual over a period of more than 12 months.