ELSEVIER

Contents lists available at ScienceDirect

Clinica Chimica Acta

journal homepage: www.elsevier.com/locate/cca

ifce

CLINICA CHIMICA ACTA

Steroid profiling in adrenal disease

Danni Mu ª,1, Dandan Suna,1, Xia Qian ª, Xiaoli Ma a, Ling Qiu a,b,”, Xinqi Cheng a,”, Songlin Yu ª,

Check for updates

a Department of Laboratory Medicine, Peking Union Medical College Hospital, Peking Union Medical College & Chinese Academy of Medical Science, Beijing 100730, China

b State Key Laboratory of Complex Severe and Rare Diseases, Peking Union Medical College Hospital, Chinese Academy of Medical Science and Peking Union Medical College, Beijing, China

ARTICLE INFO

Keywords: Steroid profiling Mass spectrometry Adrenal diseases Machine learning Steroid metabolomics

ABSTRACT

The measurement of steroid hormones in blood and urine, which reflects steroid biosynthesis and metabolism, has been recognized as a valuable tool for identifying and distinguishing steroidogenic disorders. The application of mass spectrometry enables the reliable and simultaneous analysis of large panels of steroids, ushering in a new era for diagnosing adrenal diseases. However, the interpretation of complex hormone results necessitates the expertise and experience of skilled clinicians. In this scenario, machine learning techniques are gaining world- wide attention within healthcare fields. The clinical values of combining mass spectrometry-based steroid profiles analysis with machine learning models, also known as steroid metabolomics, have been investigated for identifying and discriminating adrenal disorders such as adrenocortical carcinomas, adrenocortical adenomas, and congenital adrenal hyperplasia. This promising approach is expected to lead to enhanced clinical decision- making in the field of adrenal diseases. This review will focus on the clinical performances of steroid profiling, which is measured using mass spectrometry and analyzed by machine learning techniques, in the realm of decision-making for adrenal diseases.

1. Introduction

Adrenal tumors are relatively common neoplasms that can be detected using abdominal imaging techniques, which have gained popularity in recent years. The majority of the tumors are incidentalo- mas, with a prevalence ranging from 1.05% to 8.7% in autopsy series involving large patient cohorts [1-3]. The etiology of adrenal inci- dentaloma is heterogeneous, encompassing tumors originating from the adrenal cortex, medulla, as well as metastatic lesions. While adrenal incidentalomas are typically defined as nonfunctional adrenocortical adenomas (ACAs) incidentally discovered during imaging unrelated to adrenal diseases (constituting approximately 71%-84% cases) [4-6], it is crucial to conduct specific examinations and implement appropriate clinical management strategies in certain scenarios where cortisol, aldosterone, catecholamine-secreting tumors are present [6,7]. Adre- nocortical carcinomas (ACCs) are rare yet highly aggressive, entailing prompt diagnosis and appropriate treatment to prevent tumor progres- sion. Another category of adrenal diseases comprises a group of

autosomal recessive disorders known as congenital adrenal hyperplasia (CAH), resulting from enzyme deficiencies in the adrenal steroidogen- esis pathway [8]. The diagnosis of adrenal diseases requires compre- hensive hormone measurements and detailed imaging tests. Imaging techniques can precisely locate and qualitatively characterize the mass and assess peripheral organ and vascular damage. Currently, radiolog- ical examination, such as computed tomography, serves as the primary method for screening adrenal tumors; however, repeated and prolonged exposure to radiation can impose psychological and economic burdens on patients [7]. Therefore, the utilization of steroid maps is considered a promising tool to facilitate rapid diagnosis and prompt treatment, thereby eliminating the need for radiological examinations or invasive procedures on patients [9].

Steroid hormones, which are derived from cholesterol, are mainly synthesized in the adrenal cortex and can be categorized into three distinct classes: mineralocorticoids, glucocorticoids, and androgens. This biosynthetic process involves two pivotal enzymes known as cy- tochrome P450 enzymes (CYPs) and hydroxysteroid dehydrogenases

* Corresponding authors.

E-mail addresses: qiul@pumch.cn (L. Qiu), chengxq@pumch.cn (X. Cheng), yusonglinpku@163.com (S. Yu).

1 These authors have equal contribution to this work.

https://doi.org/10.1016/j.cca.2023.117749

Fig. 1. The schematic diagram illustrating adrenal steroidogenesis.

cholesterol

CYP11A1

SIAR ADX

(17a-Hydroxylase) CYP17A1 PDR

“OH

(17,20-lyase) CYP17A1 PDR

SULT2A1 PAPSS2

Pregnenolone (Preg)

17a-Hydroxypregnanolone (17OHPreg)

Dehydroepiandrosterone (DHEA)

Dehydroepiandrosterone sulfate (DHEAS)

HSD3B2

(17a-Hydroxylase) CYP17A1 PDR

HSD3B2

(17,20-lyase) CYP17A1

PDR

aromatase

Progesterone (Prog)

17a-Hydroxyprogesterone (17OHProg)

Androstenedione (A4)

Estrone (E1)

CYP21A2 PDR

CYP21A2

AKR1C3

PDR

HSD17B3

HSD17B2/4

(21-Hydroxylase)

(21-Deoxycrotisol)

aromatase

11-Deoxycorticosterone (DOC)

Testosterone (T)

Estradiol (E2)

(11B-Hydroxylase) CYP11B2 ADX

11-Deoxycortisol (S)

-Hydroxytestosterone (11-OHT)

SRD5A1/2

(18-Oxocrotisol)

“OH

CYP11B1 ADX

Corticosterone (B)

11-Ketotestosterone (11KetoT)

Estratriol (E3)

(18-Hydroxylase)

CYP11B2 ADX

(18-Hydroxycrotisol)

18-Hydroxycorticosterone (18OHB)

Dihydrotestosterone (DHT)

(18-Oxidase)

CYP11B2 ADX

Cortisol (F)

HSD11B2

HSD11B1

H6PDH

Aldoesterone (Aldo)

Cortisone (E)

(HSDs). Fig. 1 provides a schematic diagram illustrating adrenal ste- roidogenesis. These hormones exert essential impacts on various phys- iological processes such as water and salt balance, metabolism and stress response, sexual differentiation and reproduction. Given that adrenal disorders are characterized by dysfunction or dysregulation of ste- roidogenesis, the measurement of terminal products (e.g., aldosterone, cortisol and testosterone), along with certain precursors and their me- tabolites is essential for supporting the identification and diagnosis of adrenal tumors as well as other disorders associated with steroids. Ac- cording to the Clinical Practice Guideline of the European Society of Endocrinology, it is recommended that every patient with an adrenal incidentaloma undergo assessment for symptoms and signs of steroid hormone excess [7]. For instance, a serum cortisol level ≤ 50 nmol/L after a 1 mg overnight dexamethasone suppression test (DST) can be utilized to exclude cortisol excess. The measurement of plasma aldo- sterone level and plasma renin activity ratio serves as an indicator for screening primary aldosteronism (PA). Hormonal work-ups including

DST cortisol, dehydroepiandrosterone sulfate (DHEAS), testosterone, aldosterone and renin are mandatory steps for initial screening for ACC [10,11]. Additionally, for infants and children with classic CAH due to 21-hydroxylase deficiency (21OHD), the measurement of 17-hydroxy- progesterone (17OHP) holds significant value [12]. Evaluation of ste- roid profiles is imperative for patients detected with adrenal masses or requiring screening for adrenal diseases.

2. Usefulness of mass spectrometry in hormone diseases

Initially, due to the rapid increase in clinical demand for steroid assays, immunoassay became widely used in clinical laboratories because of its sensitivity, convenience, low cost and availability of commercial kits. However, immunoassays for steroids have inherent limitations such as lack of specificity and poor consistency and stan- dardization among laboratories [13]. In clinical applications, immuno- assays can significantly overestimate urinary free cortisol levels due to

cross-reactivity of cortisol metabolites that have similar molecular structures [14,15]. Furthermore, accurately detecting low levels of testosterone in samples from women and children using immunoassays is challenging. Conversely, mass spectrometry is recommended as a specific and sensitive technique for determining testosterone [16].

The analysis of steroid hormones using chromatography/mass spectrometry has a historical background dating back to the 1960s [17]. However, during the mid-1960s, gas chromatography/mass spectrom- etry (GC/MS) was only capable of detecting typical steroids with straightforward structures such as androstanes and 17-deoxypregnane. GC/MS had been extensively employed in clinical urine steroid anal- ysis and targeted serum steroid analysis until the 1970s [18]. In 1987, high-performance liquid chromatography/mass spectrometry (HPLC/ MS), which proved suitable for both free and conjugated hormones, was first utilized for steroid analysis [19]. Since 2002, tandem mass spec- trometry has been commercially used for clinical steroid analysis. The advantages of mass spectrometry-based steroid determination lie in its ability to simultaneously detect multiple analytes using small sample sizes while maintaining high specificity as well as sensitively compara- ble to the best immunoassay methods.

According to an article published in the Journal of Clinical Endo- crinology and Metabolism in 2013 [20], manuscripts describing sex steroid hormone tests must use MS-based assays; otherwise, they would be rejected. This requirement is expected to extend to other adrenal steroids and vitamin D testing in the near future. A clinical practice guideline released in 2018 [12] recommends using liquid chromatog- raphy tandem mass spectrometry (LC-MS/MS) for measuring 17-OHP and other adrenal steroid hormones as a secondary test to confirm positive cases identified through initial radioimmunoassay screening. This approach helps prevent false positives resulting from fetal steroid cross-contamination. Furthermore, international evidence-based guide- lines for assessing and treating polycystic ovarian syndrome (PCOS) suggest employing LC-MS/MS for measuring testosterone or free testosterone levels in patients with PCOS [21].

Currently, LC-MS/MS is increasingly being employed as a highly specific analytical tool in clinical laboratories for the measurement of steroid profiles in plasma or serum. This technique allows for simulta- neous measurement of an increasing number of steroid hormones in a single injection. However, due to the complicated relevance between the clinically available steroids and the prediction or classification of ad- renal disorders, interpreting test results remains challenging and re- quires expertise from clinicians. In this context, steroid metabolomics offers a promising approach by combining MS-based steroid profiling with computational machine learning (ML)-based analysis of datasets to enhance identification and management of adrenal diseases using het- erogeneous steroids data.

3. Time to use machine learning in the medical decision-making

Computational systems and tools that emulate the cognitive func- tions of the human brain fall within the domain of artificial intelligence (AI), which can facilitate problem-solving, reasoning, pattern recogni- tion, and knowledge acquisition process [22]. Machine learning (ML) techniques represent the most commonly used forms of AI in healthcare settings as they recognize previously undiscovered patterns and re- lationships among various features in datasets through computational analysis [23]. In brief, the ML process encompasses data input, computerized analysis, and prediction of output values within an acceptable range of accuracy. ML approaches encompass supervised, unsupervised, and reinforcement learning methods. With the develop- ment of computer science and exponential growth of medical data, ML has garnered global attention and momentum in the healthcare fields. This is particularly crucial as biological data often exhibit complexity, noise, and limited understanding, posing significant challenges for extracting valuable insights using conventional analytical approaches [24]. ML techniques can effectively reduce high-dimensional data

features to their most relevant variables. Consequently, they yield more interpretable outcomes that greatly facilitate diagnosis, stratification, and prognosis within the medical context [25]. For instance, the appli- cation of ML algorithms in analyzing proteomics data shows great promise in diagnosing various diseases, including both cancers and non- cancerous conditions [26-31]. Furthermore, employing machine learning approaches to analyze multi-omics data encompassing geno- mics, transcriptomics, and glycomics can offer a comprehensive un- derstanding of physiology and pathology, and assist the diagnosis, prognosis, and therapy management for various diseases [32]. Comprehensive review articles on these subjects are readily available through databases [33-37].

This review specifically focuses on the diagnosis of adrenal disorders using steroid profiling, where supervised ML techniques are widely utilized. In this context, the computer is provided with disease-related features such as the concentration of multiple steroid hormones and a ‘benchmark’ dataset that has been expertly labelled based on adrenal disease diagnoses. The ultimate objective is to establish correlations between these two aspects. The developed model can subsequently be applied to novel, unlabeled datasets for accurate prediction of diagnoses or outcomes based on input features with relatively high sensitivity and specificity. In the realm of unsupervised learning, algorithms utilize unlabeled data to accomplish tasks such as grouping data samples (clusters) and reducing the dimensionality of complex dataset. Although no published studies have been retrieved that employ unsupervised ML specifically for steroid profiling in identifying or subtyping adrenal disorders, it is worth noting that unsupervised learning methods hold promise in clustering and dimensionality reduction of intricate steroid hormones alongside other omics features. These techniques could significantly contribute to the classification and management of com- plex diseases like PA.

To develop an ML model that meets the requirements of clinical performance, several considerations need to be taken into account. Firstly, it is essential to partition the available data into training, vali- dation, and testing sets. K-fold cross-validation is commonly used to divide the training set into k evenly sized partitions and compare the performance of each segment to select optimal hyperparameters. Addi- tionally, selecting a suitable ML algorithm for specific data types is crucial. The type of data categories and the number of data samples are two factors influencing algorithm selection. For instance, when dealing with limited data volume, manual features and well-regularized classi- fiers (e.g., hierarchical Bayesian models) are needed. The larger the volume of data, the algorithm will find it easier to identify distinguishing features among data points and make accurate predictions [24]. How- ever, a fundamental challenge in developing ML methods is overfitting, which results in poor predictive performance on external validation sets. Overfitting can occur when models have excessive parameters or when training continues beyond learning the true relationship between vari- ables [38]. To mitigate this issue, cross-validation and regularization techniques are employed. Furthermore, due to ML algorithms’ inherent complexity (referred to as ‘black boxes’), researchers face difficulties in uncovering underlying mechanisms or influential factors by interpreting model outputs. It is recommended that clinicians familiarize themselves with basic concepts and potential applications of ML techniques to effectively integrate these valuable tools into modern medicine [39].

4. Application of MS and ML in adrenal tumors

States of steroidogenesis serve as valuable indicators of adrenal pa- thology; however, the interpretation of steroid profiling requires medi- cal and technical expertise of clinicians who can make a subjective diagnostic judgment based on multiple discrete steroid measurements in conjunction with other medical evidence. For individuals lacking the ability to draw reliable conclusion from complex medical information, ML algorithms offer potential in developing supportive systems for clinical decision-making. Briefly, relevant studies were searched using

Table 1 Recent articles on the combination of steroid profiling with machine learning in the diagnosis of adrenal diseases.
Author/ year	Country	Number of centers	Aim	Sample type	Sample number	ML methods	Steroid profiling	Mass spectrometry	Diagnostic accuracy
Arlt/2011 [48] Kotłowska/ 2017 [72]	UK, Netherlands, Germany, France, Italy, Poland	7 1	To evaluate the diagnostic value of a 32-steroid panel for the detection of ACC from ACA. To discriminate CS and possible subclinical hypercortisolism in patients diagnosed with NFA.	24-h urine 24-h urine	ACA = 102, ACC =45 Negative controls = 37, CS (excluding iatrogenic CS) = 16, NFA =25	GMLVQ LDA	An; Etio; DHEA; 16x-OH-DHEA; 5-PT; 5- PD; THA; 5a-THA; THB; 5a-THB; Tetrahydroaldosterone; THDOC; 5a- THDOC; PD; 3x5x-17HP; 17HP; PT; PTONE; THS; F; 66-OH-F; THF; 5a-THF; a-cortol; B-cortol; 116-OH-An; 11₿-OHEtio; E; THE; a-cortolone; B-cortolone; 11-oxo Etio. An; Etio; 11-Keto-An; 11-Hydroxy-An; 11- Hydroxy-etio; Androstenetriol; PT; THS; 11- Keto-PT; 5-PT; THE; THA; THB; allo-THB; THF; allo-THF; «-Cortol; a-Cortolone; B-Cortolone.	GC/MS: Agilent 5973 GC/MS: A single- quadrupole Finnigan MAT SSQ 710 mass spectrometer (Bremen, Germany); a Hewlett Packard 5890 GC; SolGel-1 ms capillary column (SGE Analytical Science, Ringwood, Australia); Carlo Erba 8000 TOP GC.	GMLVQ analysis identified a subset of nine steroids that performed best in differentiating ACA from ACC; Employing all 32 steroids: sensitivity = specificity = 90% (AUC = 0.97); Using only the nine most differentiating markers sensitivity = specificity = 88% (AUC = 0.96). Overall classification rate for the obtained functions was equal to approximately 90%; CS group: the correct classification rate was 100%, adrenal incidentaloma: the classification rates was 80%
Wilkes/ 2018 [40]	UK	1 2	To investigated the utility of various ML algorithms for the automated biochemical interpretation of urine steroid profiles.	24-h urine plasma	1314 profiles	RF; Weighted- subspace RF; Extreme gradient boosted tree.	16x-hydroxypregnenolone; THB; THA; 6x- hydroxy-THE; 16x-18-dihydroxv- dehydroepiandrosterone; 1-hydroxv-20ß- cortolone; B-cortolone; 16ß-hydroxy- dehydroepiandrosterone; 6-hydroxy-20ß- cortolone; 6-hydroxy-20a-cortolone; Hexahydro-11-deoxycortisol; 11ß-	GC/MS LC-MS/MS: AB Sciex QTRAP 5500 triple quadrupole mass spectrometer; an Acquity uPLC system (Waters Corporation, Millford, MA, USA); a Phenomenex Kinetex C18 column.	The best performing binary classifier could predict the interpretation of No significant abnormality and? Abnormal profiles with a mean area under the ROC
Masjkur/ 2019 [77]	Germany, Switzerland	1 2	To assess whether plasma steroid profiles might assist diagnosis of SC.	24-h urine plasma	AC = 21, SC = 35, CS disease was excluded, but with hypercortisolism (EX) = 152, hypertensive and normotensive volunteers = 277	LR, Discriminant analysis	hydroxyaetiocholanolone; Allo-THB; 5-PD; An-5x; Aetiocholanolone-56; THS; 16x-OH- DHEA; 5-PT; 11ß-OH-An; Androstenetriol; a-cortolone; An; PT; DHEA; 5a-THF; 17-HP; THF; THE; 11-oxo-PT, etc. Aldo; F; S; 21-deoxycortisol; B; DOC; 18- oxocortisol; 18-hydroxycortisol; E; Prog; 17OHP; Preg; A4; DHEA; DHEAS.		curve of 0.955 (95% CI,0.949-0.961); The best performing multiclass classifier could predict the individual abnormal profile interpretation with a mean balanced accuracy of 0.873 (0.86-0.880). The steroid panel provided discrimination of patients in the AC, EX and SC groups with areas under ROC curves of 0.9962, 0.9904 and 0.9901, respectively.
Schweitzer/ 2019 [50]	Germany	1	To evaluate the diagnostic value of a 15-steroid plasma	plasma	ACA = 66, ACC = 42	Logistic regression modeling	Aldo; A4; S; DOC; 21-deoxycortisol; F; E; B; DHEA; DHEAS; Dihydrotestosterone; Estradiol; Prog; 17OHP; T	LC-MS/MS: a Sciex 6500 + QTRAP (SCIEX, Framingham, USA)	The 6-steroid panel had AUCs of 0.95 and 0.94 for male and female patients (continued on next page)

Table 1 (continued)
Author/ year	Country	Number of centers	Aim	Sample type	Sample number	ML methods	Steroid profiling	Mass spectrometry	Diagnostic accuracy
			panel in a large series of ACA and ACC.					MS-system; an Agilent 1290 HPLC-system.	in the diagnosis of adrenal malignancy.
Chortis/ 2020 [51]	UK, Netherlands, Germany, US,	14	To evaluate the performance of urine	24-h urine	developed disease recurrence = 32, clinically and radiologically disease-free for ≥ 3 years =39	RF	An; Etio; 11-OH-An; DHEA; 16x-OH- DHEA; 5-PT; 5-PD; THA; 5x-THA; Tetraydrocorticosterone; 5x-THB; 3a,5ß- tetrahydroaldosterone; Tetrahydrodeoxycorticosterone; PD; 3x5a-17HP; 17-HP; PT; PTONE; THS	GC/MS: Agilent 5975	The accuracy in the detection of ACC recurrence is 85%,
	France, Italy,		steroid metabolomics			RF			The accuracy in the detection of ACC recurrence is 85%,
	Portugal,		for postoperative						AUROC:0.89 (95%CI
	Croatia, Republic of Ireland,		recurrence detection after microscopically complete (R0)						0.86-0.91), sensitivity = specificity = 81%. High-volume recurrences were more likely to be detected by RF classifier (sensitivity 60% [95% CI 36%-80%] vs. 42% [95% CI 19%-68%] in low- volume recurrences).
	Poland, Greece,		recurrence detection after microscopically complete (R0)
	Switzerland,		resection of ACC.
Eisenhofer/ 2020 [59]	Germany, Italy, Poland	4	To classify patients with PA, particularly for patients with unilateral adenomas	plasma	patients with hypertension = 201, PA = 273 (bilateral = 134, unilateral = 139 [KCNJ5^WT = 81, KCNJ5^MUT =58])	RF; SVM; LDA; LR	Aldo; 18-oxocortisol; 18- hydroxycortisol; F; E; S; 21-deoxycortisol; B; DOC; Prog; 17OHP; Preg; A4; DHEA; DHEAS	LC-MS/MS: AB Sciex QTRAP 5500 triple quadrupole mass spectrometer; an Acquity uPLC system; a Phenomenex Kinetex C18 column.	Using RF to identify patients with PA, diagnostic sensitivity = 69% (95% CI,68%-71%)

			due to pathogenic KCNJ5 sequence variants.						and specificity = 94%
			due to pathogenic KCNJ5 sequence variants.						(95% CI, 93%-94%). The external validation series yielded sensitivity of 85% and specificity of 100%. Using SVM to identify patients with APAs due to KCNJ5 variants, diagnostic sensitivity = 85% (95% CI, 81%-88%) and specificity = 97% (95% CI, 97%-98%). The external validation series yielded sensitivity of 100% and specificity of 98%.
Bancos/ 2020 [49]	Brazil, Croatia,	21	To validate a urine steroid metabolomics	24-h urine	ACC = 98, other malignant tumors = 65, ACA = 1767, other benign tumors =87	GMLVQ	An, Etio, 11ß-OH-An, DHEA, 5-PT, 5-PD, PD, 17-HP, PT, THS, F, 11ß-OHEtio, E, THE, B-cortolone	LC-MS/MS: Waters Xevo mass spectrometer; an acquity uPLC system; HSS T3 column	As a single test, urine steroid metabolomics had a high positive predictive
	France,		To validate a urine steroid metabolomics	24-h urine
	Germany,		approach for the
	Greece, Ireland,		diagnosis of ACC.						value 34.6% (95% CI
	Italy, the Netherlands,								28.6%-41.0%) in patients with high risk of ACC.
	Norway, Poland,								When the three tests
	Portugal, Serbia, UK, USA								(tumor diameter, positive
	Portugal, Serbia, UK, USA								imaging characteristics, and urine steroid

metabolomics) were combined, 106 (5.3%) participants had the result maximum tumor diameter of 4 cm or larger, positive imaging characteristics (with the 20 HU cutoff), and urine steroid

(continued on next page)

Table 1 (continued)
Author/ year	Country	Number of centers	Aim	Sample type	Sample number	ML methods	Steroid profiling	Mass spectrometry	Diagnostic accuracy
Ku/2021 [52] Ye/2022 [83]	Korea China	16 1	To discriminate adrenal tumors, including NFA, CS and primary PA. To distinguish and subtype CAH (11BOHD, 17OHD, and 21OHD).	serum plasma	NFA =73, CS =30, PA = 40 (the perspective application cohort) 11BOHD =11, 17OHD = 19, 21OHD = 132, other = 94	DT; RF; XGBoost linear logistic regression model	THE; 18-hydroxycortisol; DHEA; DHEAS; 20x-dihydrocortisol; 66-OH-F; E; THF; S; Pregnenolone sulfate; 17x- hydroxypregnenolone; B; A4; F; Allo-THF Preg; 17-hydropregnenolone; 17OHP; S; F; E; Prog; DOC; B; Aldo; A4; T; Estradiol.	LC-MS/MS: an LCMS- 8050 triple- quadrupole-mass spectrometer (Shimadzu Corp.); a Nexera UHPLC system (Shimadzu Corp., Kyoto, Japan); a Hypersil Gold C18 column. LC-MS/MS: Analyst (V1.6.1, ABSciex); a Phenomenex Kinetex C18 column.	metabolomics indicating high risk of ACC, for which the PPV was 76.4% (95% CI 67.2%-84.1%). The balanced accuracies of DT, RF, and XGBoost for classifying each type were 78%, 96%, and 97%. For diagnosing CS: DT ( AUC:0.776 [95% CI 0.685-0.867]), RF (AUC:0.925 [95% CI 0.861-0.988]), XGBoost (AUC:0.911[95% CI 0.847-0.976]). For diagnosing PA: DT ( AUC:0.838 [95% CI 0.765-0.911]), RF (AUC:0.933 [95% CI 0.880-0.985]), XGBoost (AUC:0.881 [95% CI 0.815-0.946]). For 11BOHD: sensitivity = 0.900 (95% CI 0.541-0.995), specificity = 0.992(95% CI 0.968-0.999) and AUC = 0.994 (95% CI, 0.983-1 0.000); for 17OHD: sensitivity = 0.826 (95% CI 0.605-0.943), specificity = 1.000 (95% CI 0.980-1.000) and AUC = 0.993 (95% CI, 0.985-1 0.000); for 21OHD: sensitivity = 0.976 (95% CI
Ku/2021 [52] Ye/2022 [83]	Korea China	16 1		serum plasma		DT; RF; XGBoost linear logistic regression model			0.927-0.994), specificity = 0.931 (95% CI 0.869-0.966) and AUC = 0.979 (95% CI, 0.964-0.994).
Agnani/ 2022 [84]	France	1	To distinguish children with premature pubarche (PP) from those with NCCAH.	serum	NCCAH =13, PP =84	OPLS-DA	S; 21-deoxycortisol; 11₿- hydroxyandrostenedione; A4; T; 17OHP; 17a-hydroxypregnenolone; DHEA; Aldo; E; F; B; DOC; 21-deoxycorticosterone; Prog; Preg.	LC-MS/MS: a Triple Quad 6500 (ABSciex, Foster City, CA); a Shimadzu Nexera XR system (Shimazu France, Marne la Vallee, France); a Coreshell C18 column (Phenomenex, Le Pecq, France).	The prediction score was accurate (100%) at differentiating PP from NCCAH.

Table 1 (continued)
Author/ year	Country	Number of centers	Aim	Sample type	Sample number	ML methods	Steroid profiling	Mass spectrometry	Diagnostic accuracy
Reel/2022 [67] Bachelot/ 2023 [85]	France,	11	To distinguish	plasma,	PA =113, PPGL = 88,	DT, Naive Bayes,	Plasma steroid profile: Aldo, B, DOC, Prog,	LC-MS/MS: for urine:	Balanced accuracy for
	Germany, UK,		different endocrine	24-h	CS = 41, PHT =112, and Normotensive	K-nearest	Preg, E, F, S, 17OHP, A4, DHEA, DHEAS,	Waters Xevo mass	discrimination of EHT and
	Switzerland,		hypertension (EHT)	urine	CS = 41, PHT =112, and Normotensive	neighbours,	21-Deoxycortisol, 18-0xocortisol, 18-	spectrometer; an	PHT: plasma steroids
	Italy, Poland,		subtypes (including		Volunteers (NV) =	LogitBoost,	Hydroxycortisol;	acquity uPLC system; a HSS T3 column; for serum: an AB Sciex QTRAP 5500 triple quadrupole mass spectrometer; an acquity uPLC system; a Phenomenex Kinetex C18 column. LC-MS/MS: Triple	using LogitBoost = 72.1%;
	Ireland, Australia	2	PA, PPGL, and CS) from PHT To distinguish	serum	133 (test set) NC21OHD	Logistic Model Tree, SL, RF, and Sequential Minimal Optimisation OPLS-DA	urine steroid profile: An, Etio, 11ß-OH-An, DHEA, 5-PT, 5-PD, PD, 17-HP, PT, THS, F, 116-OHEtio, E, THE, B-cortolone 21-deoxycortisol; 11₿-		urine steroids using RF = 83.1%; multi-omics features (including plasma miRNAs, plasma catechol O-methylated metabolites, plasma steroids, urinary steroid metabolites, and plasma small metabolites) using SL = 87.9%. Discrimination of PA and PHT: plasma steroids using SL = 83.0%; urine steroids using RF = 80.2%; multi-omics features using SL = 90.4%. Discrimination of PPGL and PHT: plasma steroids using RF = 60.7%; urine steroids using SL = 79.8%; multi-omics features using LogitBoost = 96.4%. Discrimination of CS and PHT: plasma steroids using SL = 89.3%; urine steroids using LogitBoost = 89.3%; multi-omics features using SL = 91.7%. The constructed model

	France
			NC21OHD from polycystic ovary syndrome (PCOS)		= 17, Controls = 72, PCOS = 266		hydroxyandrostenedione; A4; T; 17OHP; 17-OH-pregnenolone; DHEA; Aldo; E; F; B; DOC; 21-deoxycorticosterone; Prog; Preg.	Quad 6500; a Shimadzu Nexera XR system; a Coreshell C18 column.	successfully excluded all 17 NC21OHDs (sensitivity and specificity of 100%).

*Abbreviations: GC-MS, gas chromatography-mass spectrometry; LC-MS/MS: liquid chromatography tandem mass spectrometry; uPLC: ultra performance liquid chromatography; ML: machine learning; 11BOHD, 11ß- hydroxylase deficiency; 170HD, 17a-hydroxylase/17,20-lyase deficiency; 210HD, 21a-hydroxylase deficiency; NFA, nonfunctioning adenoma; CS, Cushing’s syndrome; PA, primary aldosteronism; ACA, adrenocortical adenomas; ACC, adrenocortical carcinomas; SC,subclinical Cushing syndrome; AC,adrenal Cushing syndrome; PHT, primary hypertension; NCCAH, nonclassical congenital adrenal hyperplasia; NC21OHD, nonclassic 21- hydroxylasedeficiency; EHT, endocrine hypertension; PPGL, pheochromocytoma/catecholamine-producing paraganglioma;

RF: the random forest model; XGBoost: extreme gradient boost; LDA: Linear discriminant analysis; SVM: Support vector machine; DT: decision tree; OPLS-DA: Orthogonal partial least square discriminant analysis; GMLVQ: generalized matrix learning vector quantization; LR: Logistic regression; SL: Simple Logistic;

An: Androsterone; 116-OH-An: 116-hydroxyandrosterone; Etio: Etiocholanolone; DHEA: Dehydroepiandrosterone; DHEAS: Dehydroepiandrosterone-sulfate; F: Cortisol; E: Cortisone; 5-PT: Pregnenetriol; 5-PD: Pre- gnenediol; PD: Pregnanediol; 17HP: 17-hydroxypregnanolone; PT: Pregnanetriol; S: 11-Deoxycortisol; THS: Tetrahydro-11-deoxycortisol; 116-OHEtio: 116-hydroxyetiocholanolone; THE: Tetrahydrocortisone; 16x-OH- DHEA: 16x-hydroxy-dehydroepiandrosterone; THA: Tetrahydro-11-dehydrocorticosterone; THB: Tetrahydrocorticosterone; DOC: 11-deoxycorticosterone; THDOC: Tetrahydro-11-deoxycorticosterone; 3x5x-17HP: 3a, 5a-17-hydroxypregnanolone; THF: Tetrahydrocortisol; PTONE: Pregnanetriolone; 66-OH-F: 66-hydroxycortisol; 17OHP: 17-hydroxyprogesterone; T: Testosterone; Preg: Pregnenolone; Prog: Progesterone; B: Cortico- sterone; Aldo: Aldosterone; A4: Androstenedione.

*Bold fonts: markers with the highest discriminatory power.

terms ([“Adrenal Gland Diseases” OR “adrenocortical adenomas” OR “adrenocortical carcinomas” OR “primary hyperaldosteronism” OR “Cushing’s syndrome” OR “congenital adrenal hyperplasia”] AND [“mass spectrometry” OR “liquid chromatography tandem mass spec- trometry” OR “steroid profiling”] AND [“machine learning” OR “artifi- cial intelligence” OR “steroid metabolomics”] through electronic databases including Medline, EMBASE, Science Citation Index Expanded from inception to May 31, 2023. Recent articles analyzing the clinical performance of steroid metabolomics have been listed in Table 1. In a proof-of-concept study [40], supervised ML algorithms, namely random forest (RF), weighted-subspace random forest (WSRF), and extreme gradient boosted tree (XGBT) algorithms, were trained using a dataset of 4,619 cases to facilitate the interpretation of urine steroid profiles. The binary classifiers performed by each algorithm achieved mean AUC values ranging from 0.940 to 0.955 for predicting the interpretation of ‘No significant abnormality’ and ‘?Abnormal’ (refers to all forms of adrenocortical tumor) profiles. For multiclass classifiers predicting the interpretation of ‘No significant abnormality’, ‘?Adrenal suppression’, ‘? Adrenal tumor’, ‘?Cushing’s’ ‘?21-OH CAH’ and ‘?5x-reductase inhibi- tion’ profiles, ML algorithms demonstrated mean balanced accuracies ranging from 0.835 to 0.873. The performance of ML algorithms in predicting the interpretation of experienced practitioners indicates their potential for application in automating the interpretation of urine ste- roid profiles within clinical practice. However, further studies incorpo- rating gold standard diagnostic outcome data are necessary to comprehensively assess the efficacy and accuracy of ML algorithms.

4.1. Combination of MS and ML in adrenocortical carcinoma (ACC)

ACC is a rare and highly aggressive malignant tumor, with an esti- mated annual incidence of 1-2 cases per million [41,42]. The mean 5- year survival rate ranging from 16% to 47%, dropping to 5-10% in metastatic diseases [7]. Adrenocortical hormone excesses are the major symptoms and signs in approximately 45%-70% of patients with ACC [43,44]. Hypercortisolism is the predominant presentation in 50%-80% of those patients, while adrenal androgens overproduction occurs in 40%-60%. Estrogen and aldosterone secretion are relatively infrequent occurrences [10,45]. Initial evaluation should encompass patient his- tory assessment along with hormonal evaluation to establish or exclude the diagnosis of hormone excess. Determining a state of hyper- cortisolism can be achieved through measurements such as adrenocor- ticotropic hormone (ACTH) and cortisol levels obtained from an early morning blood draw, midnight salivary cortisol levels, 24-hour urine free cortisol, and 1 mg DST. The diagnosis of aldosterone excess is based on assessing plasma renin activity and serum aldosterone levels. Addi- tionally, it is recommended to measure DHEAS and testosterone levels [10]. Nonetheless, distinguishing between benign and malignant lesions (i.e., ACA vs. ACC) remains a major diagnostic challenge in incidenta- lomas cases. An inaccurate diagnosis can expose patients with ACA to unnecessary surgical risks and further medical burdens. Approximately one third of ACC patients do not exhibit apparent symptoms or signs of hormone overproduction, which may be explained by a dedifferentiated and incomplete pattern of steroidogenic enzyme expression, leading to relatively inefficient steroid production and increased steroid precursors.

In a study evaluating the distribution differences of serum steroid panel in patients with suspected adrenal cancer [46], it was observed that all enrolled 10 patients with ACC exhibited a significant increase in 11-deoxycortisol levels. While some ACC patients showed increased levels of other steroid markers (including androstenedione, DHEAS, cortisol, pregnenolone, 17-hydroxypregnenolone, corticosterone, 17OHP, 11-deoxycorticosterone, and cortisone), not all patients demonstrated this trend. Another cohort [47] investigating the urinary steroid profiling of 152 patients (27 with ACC) to differentiate between ACC and other adrenal disorders found that tetrahydro-11-deoxycortisol (THS) achieved a sensitivity of 100 % and specificity of 99 % when using

a cut-off value of 2.35 pmol/24 h. THS is a metabolite of 11-deoxycorti- sol excreted in urine, and its increased production indicates relative deficient function or reduced activity of 11ß -hydroxylase in patients with ACC. However, due to the complexity of steroid hormones and their precursors and metabolites in the metabolic pathway, the results inter- pretation can be challenging, necessitating professional expertise. Therefore, ML techniques combined with steroid profiles can serve as a promising tool for simplifying clinical diagnosis of ACC, particularly in non-specialized centers.

The first study, published in 2011, utilized a ML algorithm to assist in the differential diagnosis of ACC and benign adrenal tumors [48]. GC/ MS was employed to measure a total of 32 distinct adrenal-derived steroids in 24-h urine samples from 102 patients with ACA and 45 pa- tients with ACC. Through generalized matrix learning vector quantiza- tion (GMLVQ) analysis, nine most discriminative steroids markers (THS, pregnenetriol, pregnenediol [5-PD], pregnanetriol, tetrahydro-11- deoxycorticosterone [THDOC], 5a-tetrahydro-11-dehydrocorticoster- one [5xTHA], Etiocholanolone [Etio], 5x-tetrahydrocortisol [5xTHF], and pregnanediol) were identified. Receiver-operating characteristics (ROC) analysis of GMLVQ results demonstrated a sensitivity and speci- ficity of 90% (area under the curve [AUC] 0.97) employing all 32 ste- roids for discriminating between benign and malignant adrenal tumors. These findings unequivocally demonstrate that urine steroid metab- olomics holds significant potential in discriminating between benign and malignant adrenal tumors, thereby highlighting its clinical relevance.

The urine steroid profiling results were interpreted using the same ML algorithm, GMLVQ, in a prospective multi-center study conducted in 2020 [49]. The authors assessed the diagnostic accuracy of tumor diameter imaging (with a cut-off of 4 cm), imaging characteristics (positive vs. negative), and urine steroid metabolomics, separately and in combination, in patients newly detected with adrenal masses. Notably, urine steroid metabolomics achieved high accuracy with an AUC of 94.6% (95% CI 92.2-96.9) for all patients and a positive pre- dictive value (PPV) of 34.6% (95% CI 28.6-41.0) specifically for pa- tients at high risk of ACC. When combined with tumor diameter, only 4 out of 98 patients with ACC were dismissed; similarly, when combined with imaging characteristics, only 3 were dismissed. These results indicated the necessity of employing a triple-test strategy to effectively identify cases of ACC.

Plasma steroid metabolomics also proved to be a valuable diagnostic tool for ACC, as demonstrated by a retrospective cohort study [50] recruiting 42 patients with ACC and 66 with ACA. The plasma levels of 11-deoxycorticosterone, progesterone, 17OHP, 11-deoxycortisol, dehy- droepiandrosterone (DHEA), DHEAS and estradiol were significantly elevated in ACC. Logistic regression modeling revealed that AUC values for male and female patients were calculated to be 0.95 and 0.94 respectively when including six specific markers. These results demon- strated the significance of combining steroid profiling with ML methods in facilitating accurate diagnosis of ACC. Further studies should include a comparison between serum/plasma steroids and urine steroids, as well as their combination, using ML techniques to discriminate between malignant and benign tumors. Additionally, for the generalization of this metabolomics tool, it is necessary to conduct retrospective cohorts utilizing extensive information from clinical chemistry laboratories and prospective studies that recruit a larger number of patients with gold standard diagnostic outcome.

Close surveillance of patients after operation through regular cross- sectional imaging for several years is essential to enable prompt inter- vention in event of recurrence, given the high rates of recurrence. However, regular CT scans are associated with substantial medical ex- penses, repeated radiation exposure, and frequent diagnostic ambiguity during the early stages of recurrent or metastatic disease. Chortis et al. [51] assessed the clinical performance of urine steroid profiles in detecting postoperative recurrence after microscopically complete (R0) resection of ACC. The RF analysis identified THS as the single most

important steroid metabolite for differentiating post-recurrence urine samples from those provided by non-recurred patients, followed by THDOC, 5-PD and Etio. The accuracy of differentiation reached 85%, with an AUC of 0.89 (95% CI 0.86-0.91; sensitivity = specificity = 81%). These findings suggest that employing a ML-based algorithm to analyze steroid profiling data holds promise as a noninvasive and radiation-free tool for recurrence surveillance.

4.2. Combination of MS and ML in metabolic subtyping of secreting adenoma

Although only approximately 15% of adrenal incidentalomas exhibit hormone secretion, it is crucial to employ multiple steps including hormone tests and imaging for stratification and diagnosis in all patients in the clinical circumstances. Simplified approaches are necessary to accurately identify the existence of CS or PA through a single blood sampling, thereby reducing medical costs and alleviating psychological burdens on patients. In a prospective multi-center study [52] serum steroid profiling was conducted on single blood samples obtained from 73 patients with nonfunctioning adenoma (NFA), 30 with CS, and 40 with PA in order to categorize the subtypes of adrenal tumors. Compared to the NFA group, the CS group exhibited increased levels of 11-deoxy- cortisol, decreased levels of DHEA and DHEAS, while the PA group demonstrated increased levels of 18-hydroxycortisol (18-OHF) and decreased levels of tetrahydrocortisone (THE). When comparing the CS and PA groups, higher levels of THE, 20x-dihydrocortisol, THF, and 6ß- hydroxycortisol (66-OHF) were observed in the CS group. Conversely, the PA group exhibited higher levels of 18-OHF, DHEA, and DHEAS. To differentiate adrenal tumor types, diagnostic models including decision tree (DT), RF, and extreme gradient boost (XGBoost) were used. The overall accuracies for classification were 78% (95% CI 71%-85%) for DT analysis, 96% (95% CI 91%-98%) for RF analysis, and 97% (95% CI 92%-99%) for XGBoost analysis. In diagnosing CS cases, both XGBoost and RF demonstrated significantly higher AUCs compared to DT (0.911 and 0.925 vs. 0.776). For PA diagnosis, the AUC calculated by RF was superior to that obtained from the other two approaches (0.933 vs. 0.838 and 0.881). By combining steroid profiling with ML algorithms, a simplified yet highly accurate method for subtyping adrenal tumors was developed. However, the limited number of participants in this cohort may contribute to the inconsistency of AUC results calculated by the three ML methods.

4.2.1. Combination of MS and ML in primary hyperaldosteronism

PA is the leading cause of secondary hypertension, accounting for approximately 5% of hypertension cases in primary care and 10-20% of cases referred to specialist care [53-55]. Unilateral aldosterone- producing adenoma (APA) and bilateral adrenal hyperplasia (BAH) are the major causes of autonomous aldosterone production in patients with PA. Accurate diagnosis and differentiation between unilateral and bilateral forms are crucial for effective therapeutic management. The reliable identification of subtypes heavily relies on adrenal venous sampling (AVS), which enables determination of the source of excess aldosterone. However, AVS poses challenges due to its complexity, associated costs, and potential masking effect caused by concomitant cortisol overproduction in unilateral APA detection. Measurements of aldosterone and cortisol through AVS are a standard practice for differentiating between APA and BAH. Notably, the median ratios of 11- deoxycortisol, 17-OHP, pregnenolone, androstenedione, and DHEA in adrenal venous/peripheral venous samples were reported to be signifi- cantly higher than those for cortisol, providing a potentially viable approach for subtyping patients with PA [56]. Another study demon- strated that patients with APA exhibited significantly higher peripheral venous plasma levels of 18-oxocortisol while experiencing lower levels of cortisol, corticosterone, and DHEA compared to those with BAH [57].

Arlt et al. [58] discovered a significant increase in tetrahy- droaldosterone and total glucocorticoid metabolites in 24-hour urinary

excretion among patients with PA compared to controls. However, this study did not establish a diagnostic model for identifying these patients. Subsequently, Eisenhofer et al. [59] employed ML techniques combined with plasma steroid profiling to identify and classify patients with PA (273 patients [134 with BAH and 139 with unilateral APA]) among a cohort of 632 individuals with hypertension. Their findings revealed that aldosterone, 18-oxocortisol, 18-hydroxycortisol and four other steroids exhibited superior performance in accurately identifying and classifying patients with PA. The diagnostic sensitivity and specificity for predicting PA in a single step were 69% and 94%, respectively, using the RF model.

Patients with unilateral APAs caused by pathogenic sequence vari- ants of KCNJ5 were found to exhibit distinct plasma steroid profiles, characterized by increased levels of 18-oxocortisol and 18-hydroxycor- tisol, compared to individuals with non-KCNJ5 mutations and those in the wild-type group [60]; while another study comparing urinary glucocorticoid excretion in patients with KCNJ5 mutations did not show significant differences when compared to individuals with non-KCNJ5 mutations and those with wild type [58]. By utilizing a support vector machine model, plasma steroid profiling achieved diagnostic sensitivity and specificity of 85% and 97%, respectively, for subtyping patients with unilateral APA due to KCNJ5 variants [59]. These findings suggest that combining steroid profiling with ML models can assist diagnosing patients with PA through a single test, thereby facilitating appropriate surgical intervention particularly for those presenting with unilateral adenomas due to pathogenic KCNJ5 variants.

As one of the leading risk factors for disease and disability world- wide, arterial hypertension encompasses primary hypertension (PHT) and secondary hypertension [61,62]. Identification of secondary forms of hypertension is key for targeted management and reduction of car- diovascular complications. Among them, endocrine hypertension (EHT), is caused by excess hormone production leading to increased blood pressure, such as PA, pheochromocytoma/functional paraganglioma (PPGL), and CS [63,64]. Retrospective analyses of PHT and EHT patients from a European multicenter study (ENSAT-HT) [65,66] showed that targeted metabolomics including acylcarnitines, amino acids, glycer- ophospholipids, and sphingolipids, had promising results in profiling cardiovascular diseases and endocrine conditions associated with hy- pertension. However, steroid profiles were not incorporated in these ML models. Subsequently, Reel et al. further [67] developed a ML pipeline utilizing multi-omics data, including 16 plasma steroid biomarkers and 27 urinary steroid metabolites, to differentiate PHT from secondary hypertension. The results revealed that the RF classifier using 57 multi- omics features achieved a balanced accuracy of approximately 92% (sensitivity 88%, specificity 96%, AUC 0.95) to distinguish one out of the EHT (including PPGL, PA and CS) and PHT on an unseen test set. The sensitivities and specificities to discriminate PA, PPGL and CS from PHT were 95% and 86% using Simple Logistic (SL), 93% and 100% using LogitBoost, 83% and 100% using SL, respectively. Of note, the clinical accuracies of plasma and urine steroids alone in distinguishing CS from PHT were found to be comparably high when compared to the multi- omics features. However, the relatively small sample size of patients with CS might impact the generalizability of these findings. Despite limitations in obtaining multi-omics data in routine clinical practice due to constraints on sample volume or specific quality measures, this study presents an innovative approach for predicting distinct subtypes of arterial hypertension using multi-omics data.

4.2.2. Combination of MS and ML in Cushing’s syndrome

CS is characterized by hypercortisolism and the presence of central obesity, hypertension, among other syndromes. Its estimated prevalence is 40 cases per million people, with an annual incidence ranging from 0.7 to 2.4 cases per million [68,69]. Standard diagnostic approaches include the overnight low-dose DST, 24-h urinary free cortisol level, and late-night salivary cortisol measurement [68]. However, these methods can be challenging and time-consuming. Therefore, more sensitive and

simplified diagnostic techniques for CS are required [68,70,71]. In the study conducted by Kotłowska et al. [72], urinary steroid metabolomics of 16 patients with CS, 25 with adrenal incidentaloma, and 37 controls were subjected to linear discriminant analysis to identify potential biomarkers responsible for differentiation. Six hormones, namely Etio, THE, THA, tetrahydrocorticosterone, THF, and a-cortol, were selected to generate discriminant functions. The overall classification accuracy achieved 89.7% in discriminating among CS, incidentaloma and con- trols groups. Although the limited number of participants restricts the generalization of these results, these findings provide evidence sup- porting the selection of urinary steroid biomarkers through ML analysis for distinguishing between CS and other entities. Additionally, to iden- tify the cause of hypercortisolism, plasma steroid profiles were analyzed for the classification of pituitary, ectopic, and adrenal subtypes of CS. The results showed that by utilizing 10 selected steroids (11-deoxy- cortisol, cortisol, cortisone, corticosterone, 11-deoxycorticosterone, androstenedione, 18-oxocortisol, DHEA, DHEAS, and aldosterone), subjects with and without different CS subtypes could be discriminated as effectively as with salivary and urinary free cortisol, DST, and plasma ACTH levels [73]. These findings suggest that urine and plasma steroid profiles have the potential to serve as a single-test alternative for iden- tifying and distinguishing patients with hypercortisolism. However, given the complexity of steroid biomarkers available in clinical practice, the application of ML techniques is anticipated to facilitate the estab- lishment of robust models for screening purposes.

Subclinical hypercortisolism (SH), also known as subclinical Cush- ing’s Syndrome (SCS) or mild autonomous cortisol secretion, refers to autonomous secretion of cortisol in patients who do not exhibit the typical signs and symptoms of hypercortisolism [7,74]. The optimal strategy for identifying and managing SH remains uncertain [75]. A previous study [76] demonstrated that patients with SH showed lower basal and 1-24 ACTH-stimulated levels of two androgens, namely DHEA and androstenedione, compared to those with non-secreting adenomas and controls. In females with SH, testosterone levels were found to be lower. The sensitivity and specificity for predicting SH were 71% and 76% for DHEA, respectively, while for androstenedione they were found to be 69% and 61%, respectively. Another study [77] identified a lower level of DHEAS and higher levels of 11-deoxycortisol and 11-deoxycor- ticosterone in patients with SH. These results indicated the significance of serum steroid profiling in distinguishing SH from patients with adrenocortical adenomas. Moreover, when combined with ML algo- rithms, urinary steroid profiles also demonstrated potential for early diagnosis of CS and SH in patients detected with adrenal incidentalomas [72]. However, further research is needed to enroll patients diagnosed with SH based on specified criteria, in order to develop a ML model utilizing steroid profiles for screening patients with SH among those presenting adrenal incidentalomas.

4.3. Combination of MS and ML in congenital adrenal hyperplasia

Congenital adrenal hyperplasia (CAH) encompasses a cluster of autosomal recessive hereditary diseases resulting from deficiencies in key enzymes involved in the biosynthesis of adrenal corticosteroids, leading to inadequate synthesis of downstream products and accumu- lation of upstream substances. Under normal circumstances, cortisol inhibits the secretion of ACTH; however, when cortisol synthesis is impaired within the adrenal glands, there is an augmented release of ACTH by the pituitary gland, ultimately causing adrenal cortical hy- perplasia. The clinical manifestations primarily include adrenal cortical insufficiency, disturbances in water and salt metabolism, and gonadal dysplasia [78]. CAH is classified into different types based on the enzyme deficiencies, including 21-hydroxylase deficiency (21OHD) caused by mutations in CYP21A2, 11ß-hydroxylase deficiency (11BOHD) caused by mutations in CYP11B1, 36-hydroxysteroid dehy- drogenase deficiency (3(HSD) caused by mutations in HSD3B2, 17a- hydroxylase deficiency/17,20 lyase deficiency (17«OHD) caused by

mutations in CYP17A1, and cytochrome P450 oxidoreductase deficiency (PORD) caused by mutations in POR [79]. The most prevalent type in clinical practice is 21OHD, accounting for approximately 90%-95% of cases. The prevalence of 11ß-OHD ranges between 5% and 8%, while the other subtypes are rare [80]. 21-hydroxylase catalyzes the conversion of 17-OHP to 11-deoxycortisol and progesterone to 11-deoxycorticoster- one, which are precursors for cortisol and aldosterone, respectively. In terms of clinical phenotype severity, 21OHD can be categorized into classic and non-classic forms. Based on newborn screening data, the classic form is a rare disease affecting 1 in every 14,000 to 18,000 infants globally [78]. The incidence rate of non-classic 21OHD ranges from about 1 in every 200 to 1 in every 1000 infants. Patients with the non- classic form typically retain about half of the enzyme activity and generally do not exhibit evident clinical symptoms during early stages until childhood or adolescence [8,81]. However, boys may develop precocious puberty while girls may manifest symptoms such as hirsut- ism, irregular menstruation or infertility.

Hormone imbalances in patients with CAH can cause abnormalities in gonadal development and an increased risk of metabolic syndrome. During the newborn period, some patients may undergo life-threatening incidents due to adrenal crisis [82]. Therefore, early identification and treatment of CAH are crucial for effectively preventing long-term com- plications, improving prognosis and quality of life, and reducing the high fatality rate associated with adrenal crisis. The diagnosis of 21OHD is confirmed based on serum levels of 17-OHP. Immunoassay was pre- viously used as first-tier screening tool to measure 17-OHP in dried blood spots; however, it had a high false positive rate. According to the clinical practice guideline published by the Endocrine Society in 2018, measuring 17-OHP and other adrenal steroid hormones (e.g., 21-deoxy- cortisol, androstenedione) using LC-MS/MS is recommended as a sec- ondary screening tool to increase the positive predictive value [12]. However, the current diagnosis of CAH often involves a tedious and uncertain process, especially for other forms of CAH such as 17xOHD and 11BOHD, which still lack well-defined diagnostic criteria and guidelines. In light of these circumstances, ML technologies provide new possibilities in terms of result interpretation and early screening for CAH.

A study [83] analyzed 13 plasma steroid profiles from a prospective cohort consisting of 256 patients (11BOHD = 11, 170HD = 19, 210HD = 132, and 94 patients excluded as CAH). The study utilized the cascade logistic regression model (named the “Steroidogenesis Score”) to distinguish 3 most common subtypes of CAH (11BOHD, 17OHD and 21OHD). Remarkably high diagnostic accuracies were achieved for all subtypes (AUC 0.994 [ 95% CI 0.983-1.000] for 11BOHD; 0.993 [95% CI 0.985-1.000] for 17OHD; 0.979 [95% CI 0.964-0.994] for 210HD, respectively). Sensitivities ranged from 0.826 to 0.976 while specific- ities ranged from 0.931 to 1.000 in differentiating between these sub- types. For non-classic 21OHD patients within the pooled cohort (n = 86), the model had significantly higher sensitivity compared to basal measurement of 17OHP (0.973 vs. 0.840), and was not inferior when compared against to measurement of basal and ACTH-stimulated 170HP (0.973 vs 0.947). These findings demonstrated that “Steroido- genesis Score” developed using the ML algorithm exhibited high accu- racy in classifying CAH, thereby indicating its potential as a valuable tool for precise classification.

Agnani et al [84] developed a scoring tool based on orthogonal partial least squares discriminant analysis, which involved the analysis of serum steroid profiling and clinical parameters in 97 patients (13 with non-classic CAH and 84 with premature pubarche). This tool accurately distinguished between children with premature pubarche and those with non-classic CAH. From sixteen steroid hormones, three most discriminative steroid markers (21-deoxycorticosterone [21-DB], 17- OHP, and 21-deoxycortisol [21-DF]) were identified. The implementa- tion of this scoring tool has the potential to replace ACTH in the dif- ferential diagnosis of non-classic CAH within pediatric populations.

The orthogonal partial least squares discriminant analysis model was

Fig. 2. An overview of the development in ML algorithms for the purpose of screening and diagnosing adrenal disorders.

100

GC-MS LC-MS/MS

Time

Supervised methods

Serum/plasma steroid profiling

GMLVQ

Comparison of actual and model predicted values

Urine steroid profiling

Linear discriminant analysis Support vector machine

Values

Decision tree Random forest Extreme gradient boosted tree

Logistic regression modeling

XGBoost Naive Bayes

Samples

Disease labels

K-nearest neighbours

Choose data set

Select model

Train model

Validate model

Test and application

Diagnosis

Differentiation

Prognosis

Treatment management

ACC

VS ACA

ACC

ACA

CAH

ACC

NFA

CAH

PCOS

also employed in another study [85] to distinguish non-classic 21OHD from PCOS by analyzing the results of fifteen serum steroid hormones in a cohort of 355 patients (17with non-classic 21OHD, 266 with PCOS and 72 controls). The model demonstrated sensitivity and specificity of 100%. Among the various steroid hormones analyzed, four (namely, 21- DF, 21-DB, 11ß-hydroxyandrostenedione, and 17-OHP) were identified as the most discriminative markers for classifying disorders associated with hyperandrogenism. The results offer novel insights into the biochemical alterations underlying hyperandrogenic conditions.

5. Conclusions and prospects

Nowadays, the availability of multidimensional data in both infor- mation systems of medical centers and public databases has facilitated the exploration of ML technologies in medicine. Additionally, the democratization of ML algorithms and the popularization of high- performance computers have significantly reduced barriers for research teams worldwide [24]. As depicted in Fig. 2, recent advance- ments in utilizing ML algorithm for screening and diagnosing adrenal disorders have positioned this technique as a promising tool in clinical practice. It enables more precise and personalized diagnoses through a single blood drawing with reduced time and costs. However, this field is still nascent, necessitating larger patient cohorts and multidimensional data from multiple centers to develop robust ML models with adequate predictive ability. Furthermore, the investigation of prognosis and therapeutic management in adrenal patients through the integration of steroid profiles with ML models represents a compelling area for further exploration.

Funding

This research was funded by the Beijing Natural Science Foundation (7232115) and the National High Level Hospital Clinical Research Funding (2022-PUMCH-A-138).

CRediT authorship contribution statement

Danni Mu: Writing - original draft. Dandan Sun: Data curation. Xia Qian: Visualization. Xiaoli Ma: Validation. Ling Qiu: Supervision.

Xinqi Cheng: Writing- Reviewing and Editing. Songlin Yu: Conceptu- alization, Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

No data was used for the research described in the article.

References

[1] M. Sherlock, A. Scarsbrook, A. Abbas, S. Fraser, P. Limumpornpetch, R. Dineen, P. M. Stewart, Adrenal Incidentaloma, Endocr. Rev 41 (6) (2020) 775-820.

[2] J.P. Kokko, T.C. Brown, M.M. Berman, Adrenal adenoma and hypertension, Lancet 1 (7488) (1967) 468-470.

[3] H. Hedeland, G. Ostberg, B. Hökfelt, On the prevalence of adrenocortical adenomas in an autopsy material in relation to hypertension and diabetes, Acta. Med. Scand 184 (3) (1968) 211-214.

[4] L. Barzon, N. Sonino, F. Fallo, G. Palu, M. Boscaro, Prevalence and natural history of adrenal incidentalomas, Eur. J. Endocrinol 149 (4) (2003) 273-285.

[5] R.T. Kloos, M.D. Gross, I.R. Francis, M. Korobkin, B. Shapiro, Incidentally discovered adrenal masses, Endocr. Rev 16 (4) (1995) 460-484.

[6] F. Mantero, M. Terzolo, G. Arnaldi, G. Osella, A.M. Masini, A. Alì, M. Giovagnetti, G. Opocher, A. Angeli, A survey on adrenal incidentaloma in Italy. Study Group on Adrenal Tumors of the Italian Society of Endocrinology, J. Clin. Endocrinol. Metab 85 (2) (2000) 637-644.

[7] M. Fassnacht, W. Arlt, I. Bancos, H. Dralle, J. Newell-Price, A. Sahdev, A. Tabarin, M. Terzolo, S. Tsagarakis, O.M. Dekkers, Management of adrenal incidentalomas: European Society of Endocrinology Clinical Practice Guideline in collaboration with the European Network for the Study of Adrenal Tumors, Eur. J. Endocrinol 175 (2) (2016) G1-g34.

[8] D. El-Maouche, W. Arlt, D.P. Merke, Congenital adrenal hyperplasia, Lancet 390 (10108) (2017) 2194-2210.

[9] C. Rossi, I. Cicalini, S. Verrocchio, G. Di Dalmazi, L. Federici, I. Bucci, The Potential of Steroid Profiling by Mass Spectrometry in the Management of Adrenocortical Carcinoma, Biomedicines 8 (9) (2020).

[10] T. Else, A.C. Kim, A. Sabolch, V.M. Raymond, A. Kandathil, E.M. Caoili, S. Jolly, B. S. Miller, T.J. Giordano, G.D. Hammer, Adrenocortical carcinoma, Endocr. Rev 35 (2) (2014) 282-326.

[11] M. Fassnacht, B. Allolio, Clinical management of adrenocortical carcinoma, Best. Pract. Res. Clin. Endocrinol. Metab 23 (2) (2009) 273-289.

[12] P.W. Speiser, W. Arlt, R.J. Auchus, L.S. Baskin, G.S. Conway, D.P. Merke, H.F. L. Meyer-Bahlburg, W.L. Miller, M.H. Murad, S.E. Oberfield, P.C. White, Congenital Adrenal Hyperplasia Due to Steroid 21-Hydroxylase Deficiency: An Endocrine Society Clinical Practice Guideline, J. Clin. Endocrinol. Metab 103 (11) (2018) 4043-4088.

[13] A.E. Taylor, B. Keevil, I.T. Huhtaniemi, Mass spectrometry and immunoassay: how to measure steroid hormones today and tomorrow, Eur. J. Endocrinol 173 (2) (2015) D1-D.

[14] L. Wood, D.H. Ducroq, H.L. Fraser, S. Gillingwater, C. Evans, A.J. Pickett, D. W. Rees, R. John, A. Turkes, Measurement of urinary free cortisol by tandem mass spectrometry and comparison with results obtained by gas chromatography-mass spectrometry and two commercial immunoassays, Ann. Clin. Biochem 45 (Pt 4) (2008) 380-388.

[15] L. Bianchi, B. Campi, M.R. Sessa, G. De Marco, E. Ferrarini, R. Zucchi, C. Marcocci, P. Vitti, L. Manetti, A. Saba, P. Agretti, Measurement of urinary free cortisol by LC- MS-MS: adoption of a literature reference range and comparison with our current immunometric method, J. Endocrinol. Invest 42 (11) (2019) 1299-1305.

[16] W. Rosner, R.J. Auchus, R. Azziz, P.M. Sluss, H. Raff, Position statement: Utility, limitations, and pitfalls in measuring testosterone: an Endocrine Society position statement, J. Clin. Endocrinol. Metab 92 (2) (2007) 405-413.

[17] P. Eneroth, K. Hellstroem, R. Ryhage, Identification and quantification of neutral fecal steroids by gas-liquid chromatography and mass spectrometry: studies of human excretion during two dietary regimens, J. Lipid. Res 5 (1964) 245-262.

[18] C. Shackleton, Clinical steroid mass spectrometry: a 45-year history culminating in HPLC-MS/MS becoming an essential tool for patient diagnosis, J. Steroid. Biochem. Mol. Biol 121 (3-5) (2010) 481-490.

[19] D.J. Liberato, A.L. Yergey, N. Esteban, C.E. Gomez-Sanchez, C.H. Shackleton, Thermospray HPLC/MS: a new mass spectrometric technique for the profiling of steroids, J. Steroid. Biochem 27 (1-3) (1987) 61-70.

[20] D.J. Handelsman, L. Wartofsky, Requirement for mass spectrometry sex steroid assays in the Journal of Clinical Endocrinology and Metabolism, J. Clin. Endocrinol. Metab 98 (10) (2013) 3971-3973.

[21] H.J. Teede, M.L. Misso, M.F. Costello, A. Dokras, J. Laven, L. Moran, T. Piltonen, R. J. Norman, Recommendations from the international evidence-based guideline for the assessment and management of polycystic ovary syndrome, Fertil. Steril 110 (3) (2018) 364-379.

[22] S. Agatonovic-Kustrin, R. Beresford, Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research, J. Pharm. Biomed. Anal 22 (5) (2000) 717-727.

[23] K.H. Yu, A.L. Beam, I.S. Kohane, Artificial intelligence in healthcare, Nat. Biomed. Eng 2 (10) (2018) 719-731.

[24] M.A. Myszczynska, P.N. Ojamies, A.M.B. Lacoste, D. Neil, A. Saffari, R. Mead, G. M. Hautbergue, J.D. Holbrook, L. Ferraiuolo, Applications of machine learning to diagnosis and treatment of neurodegenerative diseases, Nat. Rev. Neurol 16 (8) (2020) 440-456.

[25] M.D. Ritchie, L.W. Hahn, N. Roodi, L.R. Bailey, W.D. Dupont, F.F. Parl, J.H. Moore, Multifactor-dimensionality reduction reveals high-order interactions among estrogen-metabolism genes in sporadic breast cancer, Am. J. Hum. Genet 69 (1) (2001) 138-147.

[26] F. Hosseinzadeh, A.H. Kayvanjoo, M. Ebrahimi, B. Goliaei, Prediction of lung tumor types based on protein attributes by machine learning algorithms, Springerplus 2 (1) (2013) 238.

[27] P. Jeyananthan, K.M.D.D. Bandara, Y.G.A. Nayanajith, Protein data in the identification and stage prediction of bronchopulmonary dysplasia on preterm infants: a machine learning study, Int. J. Informat. Technol (2023).

[28] D. Das, J. Ito, T. Kadowaki, K. Tsuda, An interpretable machine learning model for diagnosis of Alzheimer’s disease, PeerJ 7 (2019) e6543.

[29] J. Hindson, Proteomics and machine-learning models for alcohol-related liver disease biomarkers, Nat. Rev. Gastroenterol. Hepatol 19 (8) (2022) 488.

[30] M. Mou, Z. Pan, M. Lu, H. Sun, Y. Wang, Y. Luo, F. Zhu, Application of Machine Learning in Spatial Proteomics, J. Chem. Inf. Model 62 (23) (2022) 5875-5895.

[31] K. Tsukita, H. Sakamaki-Tsukita, S. Kaiser, L. Zhang, M. Messa, P. Serrano- Fernandez, R. Takahashi, High-Throughput CSF Proteomics and Machine Learning to Identify Proteomic Signatures for Parkinson Disease Development and Progression, Neurology 101 (14) (2023) e1434-e1447.

[32] Y. Yang, L. Xu, L. Sun, P. Zhang, S.S. Farid, Machine learning application in personalised lung cancer recurrence and survivability prediction, Comput Struct, Biotechnol. J 20 (2022) 1811-1820.

[33] K. Kourou, T.P. Exarchos, K.P. Exarchos, M.V. Karamouzis, D.I. Fotiadis, Machine learning applications in cancer prognosis and prediction, Comput Struct, Biotechnol. J 13 (2015) 8-17.

[34] B. Wen, W.F. Zeng, Y. Liao, Z. Shi, S.R. Savage, W. Jiang, B. Zhang, Deep Learning in Proteomics, Proteomics 20 (21-22) (2020) e1900335.

[35] K. Swanson, E. Wu, A. Zhang, A.A. Alizadeh, J. Zou, From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment, Cell 186 (8) (2023) 1772-1791.

[36] J.F. McCarthy, K.A. Marx, P.E. Hoffman, A.G. Gee, P. O’Neil, M.L. Ujwal, J. Hotchkiss, Applications of machine learning and high-dimensional visualization in cancer detection, diagnosis, and management, Ann. N. Y. Acad. Sci 1020 (2004) 239-262.

[37] R. Li, L. Li, Y. Xu, J. Yang, Machine learning meets omics: applications and perspectives, Brief. Bioinform 23 (1) (2022).

[38] J.G. Greener, S.M. Kandathil, L. Moffat, D.T. Jones, A guide to machine learning for biologists, Nat. Rev. Mol. Cell. Biol 23 (1) (2022) 40-55.

[39] G.S. Handelman, H.K. Kok, R.V. Chandra, A.H. Razavi, M.J. Lee, H. Asadi, eDoctor: machine learning and the future of medicine, J. Intern. Med 284 (6) (2018) 603-619.

[40] E.H. Wilkes, G. Rumsby, G.M. Woodward, Using machine learning to aid the interpretation of urine steroid profiles, Clinical. Chemistry 64 (11) (2018) 1586-1595.

[41] L. Ng, J.M. Libertino, Adrenocortical carcinoma: diagnosis, evaluation and treatment, J. Urol 169 (1) (2003) 5-11.

[42] B. Allolio, M. Fassnacht, Clinical review: Adrenocortical carcinoma: clinical update, J. Clin. Endocrinol. Metab 91 (6) (2006) 2027-2037.

[43] J.P. Luton, S. Cerdas, L. Billaud, G. Thomas, B. Guilhaume, X. Bertagna, M. H. Laudat, A. Louvel, Y. Chapuis, P. Blondeau, et al., Clinical features of adrenocortical carcinoma, prognostic factors, and the effect of mitotane therapy, N. Engl. J. Med 322 (17) (1990) 1195-1201.

[44] G. Abiven, J. Coste, L. Groussin, P. Anract, F. Tissier, P. Legmann, B. Dousset, X. Bertagna, J. Bertherat, Clinical and biological features in the prognosis of adrenocortical cancer: poor outcome of cortisol-secreting tumors in a series of 202 consecutive patients, J. Clin. Endocrinol. Metab 91 (7) (2006) 2650-2655.

[45] T.M. Seccia, A. Fassina, G.G. Nussdorfer, A.C. Pessina, G.P. Rossi, Aldosterone- producing adrenocortical carcinoma: an unusual cause of Conn’s syndrome with an ominous clinical course, Endocr. Relat. Cancer 12 (1) (2005) 149-159.

[46] D.R. Taylor, L. Ghataore, L. Couchman, R.P. Vincent, B. Whitelaw, D. Lewis, S. Diaz-Cano, G. Galata, K.M. Schulte, S. Aylwin, N.F. Taylor, A 13-Steroid Serum Panel Based on LC-MS/MS: Use in Detection of Adrenocortical Carcinoma, Clin. Chem 63 (12) (2017) 1836-1846.

[47] T.M. Kerkhofs, M.N. Kerstens, I.P. Kema, T.P. Willems, H.R. Haak, Diagnostic Value of Urinary Steroid Profiling in the Evaluation of Adrenal Tumors, Horm. Cancer 6 (4) (2015) 168-175.

[48] W. Arlt, M. Biehl, A.E. Taylor, S. Hahner, R. Libé, B.A. Hughes, P. Schneider, D. J. Smith, H. Stiekema, N. Krone, E. Porfiri, G. Opocher, J. Bertherat, F. Mantero, B. Allolio, M. Terzolo, P. Nightingale, C.H.L. Shackleton, X. Bertagna, M. Fassnacht, P.M. Stewart, Urine steroid metabolomics as a biomarker tool for detecting malignancy in adrenal tumors, J. Clin. Endocrinol. Metabol. 96 (12) (2011) 3775-3784.

[49] I. Bancos, A.E. Taylor, V. Chortis, A.J. Sitch, C. Jenkinson, C.J. Davidge-Pitts, K. Lang, S. Tsagarakis, M. Macech, A. Riester, T. Deutschbein, I.D. Pupovac, T. Kienitz, A. Prete, T.G. Papathomas, L.C. Gilligan, C. Bancos, G. Reimondo, M. Haissaguerre, L. Marina, M.A. Grytaas, A. Sajwani, K. Langton, H.E. Ivison, C.H. L. Shackleton, D. Erickson, M. Asia, S. Palimeri, A. Kondracka, A. Spyroglou, C.

L. Ronchi, B. Simunov, D.A. Delivanis, R.P. Sutcliffe, I. Tsirou, T. Bednarczuk,

M. Reincke, S. Burger-Stritt, R.A. Feelders, L. Canu, H.R. Haak, G. Eisenhofer, M.

C. Dennedy, G.A. Ueland, M. Ivovic, A. Tabarin, M. Terzolo, M. Quinkler, D. Kastelan, M. Fassnacht, F. Beuschlein, U. Ambroziak, D.A. Vassiliadi, M. W. O’Reilly, W.F. Young Jr., M. Biehl, J.J. Deeks, W. Arlt, Urine steroid metabolomics for the differential diagnosis of adrenal incidentalomas in the EURINE-ACT study: a prospective test validation study, Lancet. Diabetes. Endocrinol 8 (9) (2020) 773-781.

[50] S. Schweitzer, M. Kunz, M. Kurlbaum, J. Vey, S. Kendl, T. Deutschbein, S. Hahner, M. Fassnacht, T. Dandekar, M. Kroiss, Plasma steroid metabolome profiling for the diagnosis of adrenocortical carcinoma, Eur. J. Endocrinol 180 (2) (2019) 117-125.

[51] V. Chortis, I. Bancos, T. Nijman, L.C. Gilligan, A.E. Taylor, C.L. Ronchi, M. W. O’Reilly, J. Schreiner, M. Asia, A. Riester, P. Perotti, R. Libé, M. Quinkler, L. Canu, I. Paiva, M.J. Bugalho, D. Kastelan, M.C. Dennedy, M. Sherlock, U. Ambroziak, D. Vassiliadi, J. Bertherat, F. Beuschlein, M. Fassnacht, J.J. Deeks, M. Biehl, W. Arlt, Urine Steroid Metabolomics as a Novel Tool for Detection of Recurrent Adrenocortical Carcinoma, J. Clin. Endocrinol. Metabol. 105 (3) (2020).

[52] E.J. Ku, C. Lee, J. Shim, S. Lee, K.A. Kim, S.W. Kim, Y. Rhee, H.J. Kim, J.S. Lim, C. H. Chung, S.W. Chun, S.J. Yoo, O.H. Ryu, H.C. Cho, A.R. Hong, C.H. Ahn, J.H. Kim, M.H. Choi, Metabolic Subtyping of Adrenal Tumors: Prospective Multi-Center Cohort Study in Korea, Endocrinol. Metab. (Seoul) 36 (5) (2021) 1131-1141.

[53] S. Monticone, J. Burrello, D. Tizzani, C. Bertello, A. Viola, F. Buffolo, L. Gabetti, G. Mengozzi, T.A. Williams, F. Rabbia, F. Veglio, P. Mulatero, Prevalence and Clinical Manifestations of Primary Aldosteronism Encountered in Primary Care Practice, J. Am. Coll. Cardiol 69 (14) (2017) 1811-1820.

[54] A. Hannemann, H. Wallaschofski, Prevalence of primary aldosteronism in patient’s cohorts and in population-based studies-a review of the current literature, Horm. Metab. Res 44 (3) (2012) 157-162.

[55] J.M. Brown, M. Siddiqui, D.A. Calhoun, R.M. Carey, P.N. Hopkins, G.H. Williams, A. Vaidya, The Unrecognized Prevalence of Primary Aldosteronism: A Cross- sectional Study, Ann. Intern. Med 173 (1) (2020) 10-20.

[56] M. Peitzsch, T. Dekkers, M. Haase, F.C. Sweep, I. Quack, G. Antoch, G. Siegert, J. W. Lenders, J. Deinum, H.S. Willenberg, G. Eisenhofer, An LC-MS/MS method for steroid profiling during adrenal venous sampling for investigation of primary aldosteronism, J. Steroid. Biochem. Mol. Biol 145 (2015) 75-84.

[57] G. Eisenhofer, T. Dekkers, M. Peitzsch, A.S. Dietz, M. Bidlingmaier, M. Treitl, T. A. Williams, S.R. Bornstein, M. Haase, L.C. Rump, H.S. Willenberg, F. Beuschlein, J. Deinum, J.W. Lenders, M. Reincke, Mass Spectrometry-Based Adrenal and Peripheral Venous Steroid Profiling for Subtyping Primary Aldosteronism, Clin. Chem 62 (3) (2016) 514-524.

[58] W. Arlt, K. Lang, A.J. Sitch, A.S. Dietz, Y. Rhayem, I. Bancos, A. Feuchtinger, V. Chortis, L.C. Gilligan, P. Ludwig, A. Riester, E. Asbach, B.A. Hughes, D. M. O’Neil, M. Bidlingmaier, J.W. Tomlinson, Z.K. Hassan-Smith, D.A. Rees, C. Adolf, S. Hahner, M. Quinkler, T. Dekkers, J. Deinum, M. Biehl, B.G. Keevil, C. H. Shackleton, J.J. Deeks, A.K. Walch, F. Beuschlein, M. Reincke, Steroid metabolome analysis reveals prevalent glucocorticoid excess in primary aldosteronism, JCI. Insight 2 (8) (2017).

[59] G. Eisenhofer, C. Durán, C.V. Cannistraci, M. Peitzsch, T.A. Williams, A. Riester, J. Burrello, F. Buffolo, A. Prejbisz, F. Beuschlein, A. Januszewicz, P. Mulatero, J.W. M. Lenders, M. Reincke, Use of Steroid Profiling Combined With Machine Learning for Identification and Subtype Classification in Primary Aldosteronism, JAMA. Netw. Open 3 (9) (2020) e2016209.

[60] T.A. Williams, M. Peitzsch, A.S. Dietz, T. Dekkers, M. Bidlingmaier, A. Riester, M. Treitl, Y. Rhayem, F. Beuschlein, J.W. Lenders, J. Deinum, G. Eisenhofer, M. Reincke, Genotype-Specific Steroid Profiles Associated With Aldosterone- Producing Adenomas, Hypertension 67 (1) (2016) 139-145.

[61] K. Rahimi, C.A. Emdin, S. MacMahon, The epidemiology of blood pressure and its worldwide management, Circ. Res 116 (6) (2015) 925-936.

[62] K.T. Mills, J.D. Bundy, T.N. Kelly, J.E. Reed, P.M. Kearney, K. Reynolds, J. Chen, J. He, Global Disparities of Hypertension Prevalence and Control: A Systematic Analysis of Population-Based Studies From 90 Countries, Circulation 134 (6) (2016) 441-450.

[63] S.F. Rimoldi, U. Scherrer, F.H. Messerli, Secondary arterial hypertension: when, who, and how to screen? Eur. Heart. J 35 (19) (2014) 1245-1254.

[64] J.B. de Freminville, L. Amar, M. Azizi, J. Mallart-Riancho, Endocrine causes of hypertension: literature review and practical approach, Hypertens. Res 46 (12) (2023) 2679-2692.

[65] Z. Erlic, P. Reel, S. Reel, L. Amar, A. Pecori, C.K. Larsen, M. Tetti, C. Pamporaki, C. Prehn, J. Adamski, A. Prejbisz, F. Ceccato, C. Scaroni, M. Kroiss, M.C. Dennedy, J. Deinum, K. Langton, P. Mulatero, M. Reincke, L. Lenzini, A.P. Gimenez- Roqueplo, G. Assié, A. Blanchard, M.C. Zennaro, E. Jefferson, F. Beuschlein, Targeted Metabolomics as a Tool in Discriminating Endocrine From Primary Hypertension, J. Clin. Endocrinol. Metab 106 (4) (2021) 1111-1128.

[66] S. Reel, P.S. Reel, Z. Erlic, L. Amar, A. Pecori, C.K. Larsen, M. Tetti, C. Pamporaki, C. Prehn, J. Adamski, A. Prejbisz, F. Ceccato, C. Scaroni, M. Kroiss, M.C. Dennedy, J. Deinum, G. Eisenhofer, K. Langton, P. Mulatero, M. Reincke, G.P. Rossi, L. Lenzini, E. Davies, A.P. Gimenez-Roqueplo, G. Assié, A. Blanchard, M. C. Zennaro, F. Beuschlein, E.R. Jefferson, Predicting Hypertension Subtypes with Machine Learning Using Targeted Metabolites and Their Ratios, Metabolites 12 (8) (2022).

[67] P.S. Reel, S. Reel, J.C. van Kralingen, K. Langton, K. Lang, Z. Erlic, C.K. Larsen, L. Amar, C. Pamporaki, P. Mulatero, A. Blanchard, M. Kabat, S. Robertson, S. M. Mackenzie, A.E. Taylor, M. Peitzsch, F. Ceccato, C. Scaroni, M. Reincke, M. Kroiss, M.C. Dennedy, A. Pecori, S. Monticone, J. Deinum, G.P. Rossi, L. Lenzini, J.D. McClure, T. Nind, A. Riddell, A. Stell, C. Cole, I. Sudano, C. Prehn, J. Adamski, A.P. Gimenez-Roqueplo, G. Assié, W. Arlt, F. Beuschlein, G. Eisenhofer, E. Davies, M.C. Zennaro, E. Jefferson, Machine learning for classification of hypertension subtypes using multi-omics: A multi-centre, retrospective, data-driven study, EBioMedicine 84 (2022) 104276.

[68] L.K. Nieman, B.M. Biller, J.W. Findling, J. Newell-Price, M.O. Savage, P. M. Stewart, V.M. Montori, The diagnosis of Cushing’s syndrome: an Endocrine Society Clinical Practice Guideline, J. Clin. Endocrinol. Metab 93 (5) (2008) 1526-1540.

[69] R. Pivonello, A.M. Isidori, M.C. De Martino, J. Newell-Price, B.M. Biller, A. Colao, Complications of Cushing’s syndrome: state of the art, Lancet. Diabetes. Endocrinol 4 (7) (2016) 611-629.

[70] G. Arnaldi, A. Angeli, A.B. Atkinson, X. Bertagna, F. Cavagnini, G.P. Chrousos, G. A. Fava, J.W. Findling, R.C. Gaillard, A.B. Grossman, B. Kola, A. Lacroix,

T. Mancini, F. Mantero, J. Newell-Price, L.K. Nieman, N. Sonino, M.L. Vance, A. Giustina, M. Boscaro, Diagnosis and complications of Cushing’s syndrome: a consensus statement, J. Clin. Endocrinol. Metab 88 (12) (2003) 5593-5602.

[71] R. Pivonello, M. De Leo, A. Cozzolino, A. Colao, The Treatment of Cushing’s Disease, Endocr. Rev 36 (4) (2015) 385-486.

[72] A. Kotłowska, T. Puzyn, K. Sworczak, P. Stepnowski, P. Szefer, Metabolomic Biomarkers in Urine of Cushing’s Syndrome Patients, Int. J. Mol. Sci 18 (2) (2017).

[73] G. Eisenhofer, J. Masjkur, M. Peitzsch, G. Di Dalmazi, M. Bidlingmaier, M. Grüber, J. Fazel, A. Osswald, F. Beuschlein, M. Reincke, Plasma Steroid Metabolome Profiling for Diagnosis and Subtyping Patients with Cushing Syndrome, Clin. Chem 64 (3) (2018) 586-596.

[74] W.F. Young Jr., Clinical practice. The incidentally discovered adrenal mass, N. Engl. J. Med 356 (6) (2007) 601-610.

[75] M. De Leo, A. Cozzolino, A. Colao, R. Pivonello, Subclinical Cushing’s syndrome, Best. Pract. Res. Clin. Endocrinol. Metab 26 (4) (2012) 497-505.

[76] G. Di Dalmazi, F. Fanelli, M. Mezzullo, E. Casadio, E. Rinaldi, S. Garelli, E. Giampalma, C. Mosconi, R. Golfieri, V. Vicennati, U. Pagotto, R. Pasquali, Steroid Profiling by LC-MS/MS in Nonsecreting and Subclinical Cortisol-Secreting Adrenocortical Adenomas, J. Clin. Endocrinol. Metab 100 (9) (2015) 3529-3538.

[77] J. Masjkur, M. Gruber, M. Peitzsch, D. Kaden, G. Di Dalmazi, M. Bidlingmaier, S. Zopp, K. Langton, J. Fazel, F. Beuschlein, S.R. Bornstein, M. Reincke, G. Eisenhofer, Plasma Steroid Profiles in Subclinical Compared With Overt Adrenal Cushing Syndrome, J. Clin. Endocrinol. Metab 104 (10) (2019) 4331-4340.

[78] H.L. Claahsen-van der Grinten, P.W. Speiser, S.F. Ahmed, W. Arlt, R.J. Auchus, H. Falhammar, C.E. Flück, L. Guasti, A. Huebner, B.B.M. Kortmann, N. Krone, D. P. Merke, W.L. Miller, A. Nordenström, N. Reisch, D.E. Sandberg, N. Stikkelbroeck, P. Touraine, A. Utari, S.A. Wudy, P.C. White, Congenital adrenal hyperplasia- current insights in pathophysiology, diagnostics, and management, Endocr. Rev 43 (1) (2022) 91-159.

[79] M.K. Auer, A. Nordenström, S. Lajic, N. Reisch, Congenital adrenal hyperplasia, Lancet 401 (10372) (2023) 227-244.

[80] A. Khattab, S. Haider, A. Kumar, S. Dhawan, D. Alam, R. Romero, J. Burns, D. Li, J. Estatico, S. Rahi, S. Fatima, A. Alzahrani, M. Hafez, N. Musa, M. Razzghy Azar, N. Khaloul, M. Gribaa, A. Saad, I.B. Charfeddine, B. Bilharinho de Mendonça, A. Belgorosky, K. Dumic, M. Dumic, J. Aisenberg, N. Kandemir, A. Alikasifoglu, A. Ozon, N. Gonc, T. Cheng, U. Kuhnle-Krahl, M. Cappa, P.M. Holterhus, M. A. Nour, D. Pacaud, A. Holtzman, S. Li, M. Zaidi, T. Yuen, M.I. New, Clinical, genetic, and structural basis of congenital adrenal hyperplasia due to 11ß- hydroxylase deficiency, Proc. Natl. Acad. Sci. USA 114 (10) (2017) E1933-E1940.

[81] P. Kamenický, A. Blanchard, A. Lamaziere, C. Piedvache, B. Donadille, L. Duranteau, H. Bry, J.F. Gautier, S. Salenave, M.L. Raffin-Sanson, S. Genc, L. Pietri, S. Christin-Maitre, J. Thomas, A. Lorthioir, M. Azizi, P. Chanson, Y. Le Bouc, S. Brailly-Tabard, J. Young, Cortisol and Aldosterone Responses to Hypoglycemia and Na Depletion in Women With Non-Classic 21-Hydroxylase Deficiency, J. Clin. Endocrinol. Metab 105 (1) (2020).

[82] D.P. Merke, R.J. Auchus, Congenital Adrenal Hyperplasia Due to 21-Hydroxylase Deficiency, N. Engl. J. Med 383 (13) (2020) 1248-1261.

[83] L. Ye, Z. Zhao, H. Ren, W. Wang, W. Zhou, S. Zheng, R. Han, J. Zhang, H. Li, Z. Wan, C. Tang, S. Sun, W. Wang, G. Ning, A Multiclassifier System to Identify and Subtype Congenital Adrenal Hyperplasia Based on Circulating Steroid Hormones, J. Clin. Endocrinol. Metab 107 (8) (2022) e3304-e3312.

[84] H. Agnani, G. Bachelot, T. Eguether, B. Ribault, J. Fiet, Y. Le Bouc, I. Netchine, M. Houang, A. Lamazière, A proof of concept of a machine learning algorithm to predict late-onset 21-hydroxylase deficiency in children with premature pubic hair, J. Steroid. Biochem. Mol. Biol 220 (2022) 106085.

[85] G. Bachelot, A. Bachelot, M. Bonnier, J.E. Salem, D. Farabos, S. Trabado, C. Dupont, P. Kamenicky, M. Houang, J. Fiet, Y. Le Bouc, J. Young, A. Lamazière, Combining metabolomics and machine learning models as a tool to distinguish non-classic 21-hydroxylase deficiency from polycystic ovary syndrome without adrenocorticotropic hormone testing, Hum. Reprod 38 (2) (2023) 266-276.

Quartz 4

Explorer

38169194