Provision of a medical risk assessment for nonclassical 21-hydroxylase deficiency
Patent Information
- Application Number
- EP2022737940
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2022-04-21
- Publication Date
- 2025-02-26
AI Technical Summary
Current methods for diagnosing non-classical 21-hydroxylase deficiency are limited by the need for specific timing of steroid measurements during the follicular phase and lack of validation on independent cohorts, making them unreliable and inconvenient.
An automated system that processes measurements of physiological steroids like 21-deoxycortisol, 21-deoxycorticosterone, 11-p-hydroxyandrostenedione, 17-OH-progesterone, androstenedione, corticosterone, and testosterone to calculate a 21-hydroxylase deficiency score, allowing for reliable assessment during any phase of the menstrual cycle and incorporating machine learning for robustness.
Enables reliable and robust assessment of non-classical 21-hydroxylase deficiency regardless of the menstrual cycle phase, providing a reliable clinical decision support tool by using a combination of steroids and machine learning for scoring.
Smart Images
Figure 1.1
Abstract
Description
Description TITLE: PROVISION OF A MEDICAL RISK ASSESSMENT OF NON-CLASSICAL FORM 21-HYDROXYLASE DEFICIENCY
[0001] The present invention relates to the field of nonclassical 21-hydroxylase deficiency (NC21 OHD). More specifically, the present invention relates to a system, method, and computer program for providing a medical risk assessment for nonclassical 21-hydroxylase deficiency (NC21 OHD), and a method for assessing the medical risk of nonclassical 21-hydroxylase deficiency (NC21 OHD).
[0002] The article "Androgen excess and diagnostic steroid biomarkers for nonclassic 21-hydroxylase deficiency without cosyntropin stimulation" by TURCU et al., published in the European Journal of Endocrinology, 183, 63-71, in 2020, aimed to identify steroids that would allow for a reliable diagnosis of nonclassical 21-hydroxylase deficiency (NC21OHD). To this end, different steroids were measured for each woman in a cohort, and, using logistic regression modeling, three steroids were identified as particularly discriminating in combination: 21-deoxycortisol, 17-OH-progesterone, and corticosterone.
[0003] The article emphasizes that the measurements must be taken at the beginning of the follicular phase of the menstrual cycle, which is difficult to implement in some cases.
[0004] Furthermore, the model consisting of the combination of the three steroids mentioned above has not been validated on an independent cohort, so the robustness of this model can be questioned.
[0005] The invention aims to address all or part of these problems.
[0006] To this end, an automated system for processing measurements of physiological parameters is proposed for the provision of a medical risk assessment, characterized in that it comprises: a module for receiving measurements of physiological parameters from a woman to be assessed; and a calculation module for a non-classical form of 21-hydroxylase deficiency score in the woman being assessed, following a model that provides the score from the measurements received; physiological parameters including at least the following steroids: 21-deoxycortisol, 21-deoxycorticosterone; 11 p-hydroxyandrostenedione; 17-OH-progesterone; androstenedione; corticosterone; and testosterone.
[0007] Surprisingly, the use of the steroids described above allows for a reliable and robust assessment of non-classical form 21-hydroxylase deficiency (NC21 OHD), even when measurements are taken during the luteal phase of the menstrual cycle. Thus, the selected steroids eliminate dependence on the timing of measurements.
[0008] Optionally, the physiological parameters also include at least one of the following steroids: pregnenolone; dehydroepiandrosterone; cortisol; cortisone; 11-deoxycorticosterone; 17-OH-pregnenolone; aldosterone; and progesterone.
[0009] Optionally, the physiological parameters also include only steroids.
[0010] Optionally, the model giving the non-classical form 21-hydroxylase deficiency score is also pre-trained by machine learning.
[0011] Optionally, the model giving the non-classical form 21-hydroxylase deficiency score includes a linear combination of measurements of physiological parameters.
[0012] Optionally, at least one of the steroids was also measured during a luteal phase of a menstrual cycle of the woman being evaluated.
[0013] Optionally, the automated system also includes: a module for comparing the non-classical form 21-hydroxylase deficiency score with a predefined threshold, for example 0.5; and a calculation module, when the non-classical form 21-hydroxylase deficiency score is on a predefined side of the threshold, for a polycystic ovarian disease score of the woman to be assessed, following a model giving the polycystic ovarian disease score from at least some of the measurements received.
[0014] Optionally, the polycystic ovarian disease score is also given from at least the measurements of the steroids: 1 i p-hydroxyandrostenedione and androstenedione, preferably from additional measurements of the steroids: testosterone and corticosterone.
[0015] A method for obtaining a model for an automated system according to the invention is also proposed, comprising: for each of several women in a cohort, obtaining measurements of the woman's physiological parameters and a status indicating that this woman either has the non-classical form of 21-hydroxylase deficiency or does not have this deficiency; and determining the model, by training the model with a training algorithm, from the measurements of the physiological parameters of the women in the cohort.
[0016] Optionally, all steroids were measured during a follicular phase of a menstrual cycle.
[0017] An automated method for processing physiological parameter measurements to provide a medical risk assessment is also proposed, characterized in that it comprises: receiving physiological parameter measurements from a woman to be assessed; and using a model to provide a non-classical form 21-hydroxylase deficiency score for the woman to be assessed, based on the received measurements; the physiological parameters comprising at least the following steroids: 21-deoxycortisol; 21-deoxycorticosterone; 11 p-hydroxyandrostenedione; 17-OH-progesterone; androstenedione; corticosterone; and testosterone.
[0018] Optionally, the physiological parameters also include at least one of the following steroids: pregnenolone; dehydroepiandrosterone; cortisol; cortisone; 11-deoxycorticosterone; 17-OH-pregnenolone; aldosterone; and progesterone.
[0019] Optionally, the physiological parameters include only steroids.
[0020] Optionally, the model giving the non-classical form 21-hydroxylase deficiency score is also pre-trained by machine learning.
[0021] Optionally, the model giving the non-classical form 21-hydroxylase deficiency score includes a linear combination of measurements of physiological parameters.
[0022] Optionally, at least one of the steroids was also measured during a luteal phase of a menstrual cycle of the woman being evaluated.
[0023] Optionally, the procedure also includes: a comparison of the non-classical form 21-hydroxylase deficiency score with a predefined threshold, for example 0.5; and a calculation, when the non-classical form 21-hydroxylase deficiency score is on a predefined side of the threshold, of a polycystic ovarian disease score for the woman being assessed, following a model that gives the polycystic ovarian disease score from at least some of the measurements received.
[0024] Optionally, the polycystic ovarian disease score is also given from at least the measurements of the steroids: 1 i p-hydroxyandrostenedione and androstenedione, preferably from additional measurements of the steroids: testosterone and corticosterone.
[0025] Also proposed is a computer program downloadable from a communication network and / or recorded on a computer-readable medium, characterized in that it includes instructions for the execution of a process according to the invention, when said computer program is executed on a computer.
[0026] A method for assessing the medical risk of non-classical 21-hydroxylase deficiency in a woman is also proposed, comprising: the collection of physiological parameters of the woman to be evaluated, the physiological parameters including at least the following steroids: 21-deoxycortisol, 21-deoxycorticosterone, 11-p-hydroxyandrostenedione, 17-OH-progesterone, androstenedione, corticosterone, and testosterone; the calculation of a non-classical form 21-hydroxylase deficiency score by inputting these physiological parameters into an automated system according to the invention; and the comparison of this score with a threshold, the woman to be evaluated being assessed as being at risk of having non-classical form 21-hydroxylase deficiency if this score is below the threshold.
[0027] Optionally, at least one of the steroids is measured during a luteal phase of a menstrual cycle of the woman being evaluated.
[0028] The invention will be better understood with the aid of the following description, given by way of example only and with reference to the accompanying drawings, in which: Figure 1 is a simplified view of an automated data processing system according to the invention for providing a medical risk assessment, in particular for non-classical form 21-hydroxylase deficiency (NC21OHD), based on the physiological parameters of a woman to be assessed; Figure 2 is a block diagram illustrating the steps of a method according to the invention for data processing to provide a medical risk assessment, in particular for non-classical form 21-hydroxylase deficiency (NC21OHD); Figure 3 shows several views of the system of Figure 1 displaying information relating to the assessment carried out in several scenarios; Figure 4 is a block diagram illustrating the steps of a method for obtaining a model M NC2IOHD assessment of medical risk of non-classical form 21-hydroxylase deficiency (NC21OHD) and an M model PC bone of medical risk assessment of polycystic ovarian disease, also known as polycystic ovary syndrome (“polycystic ovarian syndrome” in English, also designated by the acronym PCOS), Figure 5 is a score plot resulting from an orthogonal partial least squares discriminant analysis (OPLS-DA) performed on a primary development cohort to obtain the MNC2IOHD model of the system in Figure 1. Figure 6 is a score plot after applying the MNC2IOHD model to a validation cohort. Figure 7 shows the score plot from Figure 6, without the measurements from the validation cohort of women with polycystic ovary syndrome (PCOS), and distinguishing between measurements from control women and measurements from women with non-classical 21-hydroxylase deficiency (NC21 OHD). Figure 8 shows the score plot from Figure 7, with the measurements from the validation cohort of women with polycystic ovary syndrome (PCOS). Figure 9 shows the score plot from Figure 1. 6,By isolating women for whom at least one physiological parameter was measured in the luteal phase of their menstrual cycle, Figure 10 is a variable importance in projection graph (VIP) of steroids used by the MNC2IOHD model, Figure 11 is a variable importance in projection graph of steroids used by the MPCOS model, and Figure 12 is an odds graph resulting from a partial least squares discriminant analysis (OPLS-DA) performed on a so-called annex development cohort, to obtain the M model, PC bone-
[0029] With reference to Figure 1, an example of an automated data processing system 100, according to the invention, will now be described. More specifically, as will become apparent later, the system 100 is designed to provide a medical risk assessment for a woman with non-classical form 21-hydroxylase deficiency (NC21 OHD). This assessment is, for example, provided in the form of an SNC2IOHD score for non-classical form 21-hydroxylase deficiency (NC21 OHD). Thus, the system 100 provides clinical decision support, for example, for the practitioner.
[0030] The system 100 includes, firstly, a human-machine interface 102 comprising at least one data input device and at least one data presentation device, for example, a visual presentation device such as a screen. For example, the human-machine interface 102 includes a touchscreen that serves as both the data input and data presentation device. For example, the system 100 is a tablet computer or a smartphone.
[0031] System 100 further includes a data receiving module 104, designed to receive measurements of physiological parameters of the woman being evaluated. These physiological parameters include at least the following steroids: 21-deoxycortisol (21DF), 21-deoxycorticosterone (21DB), 11-p-hydroxyandrostenedione (11bOHA), 17-OH-progesterone (17OHP), androstenedione (A), corticosterone (B), and testosterone (T). Preferably, the physiological parameters also include at least one of the following steroids: pregnenolone (Preg), dehydroepiandrosterone (DHEA), cortisol (F), cortisone (E), 11-deoxycorticosterone (DOC), 17-OH-pregnenolone (17-OHPreg), aldosterone (ALDO), and progesterone (P). Preferably, the physiological parameters should include only steroids.
[0032] In the example in Figure 1, the measurements of the fifteen steroids listed previously are received by the data reception module 104, and noted p1...p15.
[0033] For example, the data receiving module 104 is designed to cooperate with the human / machine interface 102 to receive measurements p1 .. ,p15. For example, a user of system 100 can enter measurements p1 ...p15 manually via the human / machine interface 102. Alternatively, all or part of the measurements p1...p15 can be received from a remote device (not shown), such as a computer, via a wireless or wired connection.
[0034] The physiological parameters form a multidimensional space (of dimension fifteen when all the steroids indicated previously are used) of which each set of measurements p1...p15 is a point.
[0035] System 100 also includes a module 106 for calculating the SNC2IOHD score of the woman being assessed. This calculation module 106 is designed to calculate the SNC2IOHD score according to an MNC2IOHD model, for example, one previously trained using machine learning.
[0036] For example, the MNC2IOHD model is a linear combination, so the SNC2IOHD score is a linear combination of the received measurements p1...p15. In this linear combination, the measurements p1...p15 are respectively assigned weights w1...w15, and the weighted measurements p1...p15 are summed to obtain the SNC2IOHD score. Thus, in this case, the SNC2IOHD score is given by the following formula: [Math. 1] NC21OHD = En=l wn ■ pn + E where N is the number of physiological parameters measured (for example fifteen in the example in Figure 1), and s is a constant representing a residual. This constant can be zero.
[0037] As will be explained in more detail later, machine learning involves determining the w1...w15 weights, allowing the SNC2IOHD score to be representative of the risk of non-classical form 21-hydroxylase deficiency (NC21 OHD) of the woman being assessed.
[0038] Machine learning can be performed outside of the automated system 100. In this case, the data reception module 104 can also be designed to receive an update containing the weights w1 .. ,w15. This update is, for example, received from a remote device (not shown), such as a computer, via a wireless or wired connection.
[0039] Alternatively, the MNC2IOHD model could incorporate a neural network.
[0040] In general, the MNC2IOHD model implements a function including parameters and prior machine learning of the MNC2IOHD model includes the determination of these parameters allowing the SNC2IOHD score to be representative of the risk of non-classical form 21-hydroxylase deficiency (NC21 OHD).
[0041] Preferably, the M model N C2IOHD is pre-trained so that the SNC2IOHD score falls within a predefined range, for example between 0 and 1.
[0042] In addition, system 100 includes, for example, a module 108 for comparing the SNC2IOHD score with a threshold T NA predefined C2IOHD score, for example 0.5, is used. When the SNC2IOHD score is on one side of the predefined threshold (above or well below) the TNC2IOHD threshold, the woman being assessed is considered to have a high risk of non-classical 21-hydroxylase deficiency (NC21 OHD) and a low risk on the other side. Thus, the result of this comparison provides a DNC2IOHD assessment of the Risk of non-classical deformed 21-hydroxylase deficiency (NC21OHD). For example, the DNC2IOHD risk assessment is binary and indicates either a high risk of non-classical deformed 21-hydroxylase deficiency (NC21OHD) when the SNC2IOHD score is above the TNC2IOHD threshold, or a low risk of non-classical deformed 21-hydroxylase deficiency (NC21OHD) when the SNC2IOHD score is below the TNC2IOHD threshold, or vice versa. Alternatively, the DNC2IOHD risk assessment could be more gradual, for example, further indicating a very low risk when the SNC2IOHD score is well below the TNC2IOHD threshold. N C2IOHD (e.g., less than 0.25) and / or a very high risk when the SNC2IOHD score is significantly above the T threshold N C2IOHD (for example greater than 0.75).
[0043] System 100 also includes a calculation module 110, which, when the SNC2IOHD score is on the predefined side of the TNC2IOHD threshold indicating a low risk of non-classical 21-hydroxylase deficiency (NC21 OHD), calculates a Spcos score for polycystic ovary syndrome (PCOS) in the woman being assessed, based on at least some of the p1...p15 measurements, for example, all of them, or even just all of them. This calculation module 110 is designed to calculate the Spcos score according to an MPCOS model, for example, one previously trained using machine learning.
[0044] For example, the MPCOS model is a linear combination, so the Spcos score is a linear combination of the p1...p15 measurements used. In this linear combination, the p1...p15 measurements used are respectively assigned weights w'1...w'15, and the weighted p1...p15 measurements are summed to obtain the Spcos score. Thus, in this case, the SNC2IOHD score is given by the following formula: [Math. 2] Spcos = Zn=i w 'n ■ pn + E' where N is the number of measured physiological parameters used (for example fifteen in the example in Figure 1), and s' is a constant representing a residual. This constant can be zero.
[0045] As will be explained in more detail later, machine learning involves determining the weights w'1...w'15 allowing the Spcos score to be representative of the risk that the woman being assessed has polycystic ovary syndrome (PCOS).
[0046] Machine learning can be performed outside the automated system 100. In this case, the data reception module 104 can also be designed to receive an update containing the weights w'1 ...w'15. This update is, for example, received from a remote device (not shown), such as a computer, via a wireless or wired connection.
[0047] Alternatively, the MPCOS model could include a neural network.
[0048] In general, the M model PC os implements a function including parameters and prior machine learning of the MPCOS model includes the determination of these parameters allowing the SPCOS score to be representative of the risk of having polycystic ovary syndrome (PCOS).
[0049] Preferably, the M model PC os is pre-trained so that the Spcos score falls within a predefined range, for example between 0 and 1.
[0050] Furthermore, the 100 system includes, for example, a module (112) for comparing the Spcos score with a predefined TPCOS threshold, such as 0.5. When the Spcos score is on one predefined side (above or well below) the TPCOS threshold, the woman being assessed is considered to have a high risk of polycystic ovary syndrome (PCOS), and a low risk on the other side. The result of this comparison provides a DPCOS risk assessment for PCOS. For example, the DPCOS risk assessment is binary and indicates either a low risk of PCOS when the Spcos score is above the TPCOS threshold, or a high risk of PCOS when the Spcos score is below the TPCOS threshold, or vice versa. Alternatively, the DPCOS risk assessment could be more gradual, for example, by also indicating a very low risk when the Spcos score is significantly below the T threshold.PC bone (e.g., less than 0.25) and / or a very high risk when the SPCOS score is significantly above the T threshold PC bone (for example greater than 0.75).
[0051] System 100 also includes an information presentation module 116 designed to display the SNC2IOHD score and / or the DNC2IOHD risk assessment to a System 100 user. Presentation module 116 can also be designed to display the Spcos score and / or the DPCOS risk assessment.
[0052] Presentation module 116, for example, is designed to cooperate with human-machine interface 102. For example, presentation module 116 is designed to trigger the display of information on the human / machine interface 102, for example on the touch screen when such a screen is used.
[0053] Furthermore, the system 100 includes, for example, a computer system comprising a processing unit 118 (such as one or more microprocessors) and a main memory 120 coupled to the processing unit 118. A computer program 124, containing computer program instructions, is intended to be loaded, either in its entirety or sequentially in pieces, into the main memory 120. The loaded instructions can then be executed by the processing unit 118. For example, the system 100 also includes a mass storage device 122 in which the computer program 124 is stored before being loaded into the main memory 120.
[0054] Thus, the modules described above are, for example, implemented in the computer program as software modules.
[0055] For example, particularly when system 100 is a smartphone, computer program 124 may be in the form of an application (in English "APP"), for example available in the Apple (registered trademark) and / or Android (registered trademark) app store.
[0056] Alternatively, all or part of the modules could be implemented as hardware modules, i.e. as an electronic circuit, for example micro-wired, not involving a computer program.
[0057] With reference to Figure 2, an example 200 of an automated data processing method, within the framework of the example of system 100 in Figure 1, will now be described.
[0058] During a step 202, the receiving module 104 receives the measurements p1...p15.
[0059] During step 204, the calculation module 106 implementing the MNC2IOHD model calculates the SNC2IOHD score from the received measurements p1 ...p15.
[0060] During step 206, the comparison module 108 compares the SNC2IOHD score to the predefined threshold and determines the risk assessment D N C2IOHD-
[0061] During step 208, if the SNC2IOHD score is on the side of the TNC2IOHD threshold indicating a low risk of non-classical form 21-hydroxylase deficiency (NC21OHD), for example if the SNC2IOHD score is less than 0.5, the calculation module 110 implementing the MPCOS model calculates the Spcos score from at least some of the p1 ...p15 measurements received.
[0062] During step 210, the comparison module 110 compares the Spcos score to the TPCOS threshold and determines the DPCOS risk assessment.
[0063] During step 212, the information presentation module 116 presents the SNC2IOHD score and / or the DNC2IOHD risk assessment, as well as, where applicable, the SPCOS score and / or the D risk assessment PC bone.
[0064] Referring to Figure 3, during step 212, presentation module 116, for example, displays a sentence representing the DNC2IOHD risk assessment, such as "high NC21OHD risk" or "low NC21 OHD risk", depending on the risk assessment D NC2IOHD - For example, presentation module 116 displays the SNC2IOHD score and, preferably, the predefined threshold. This threshold is 0.5 in the example shown in Figure 3. Furthermore, when the SNC2IOHD score is close to the TNC2IOHD threshold indicating a low risk of non-classical form 21-hydroxylase deficiency (NC21OHD), presentation module 116 displays a sentence representing the DPCOS risk assessment for polycystic ovary syndrome (PCOS), such as "high PCOS risk" or "low PCOS risk," depending on the DPCOS risk assessment. Again, for example, presentation module 116 displays the Spcos score and, preferably, the TPCOS threshold.
[0065] With reference to Figure 4, a method 400 for obtaining the MNC2IOHD and MPCOS models for an automated system such as that in Figure 1 will now be described.
[0066] In step 402, for each of several women in a so-called main development cohort, p1...p15 measurements of the woman's physiological parameters are obtained, along with that woman's non-classical form 21-hydroxylase deficiency (NC21OHD) status. For example, measurements are taken, for all women in the main development cohort, during the follicular phase of the menstrual cycle. The physiological parameters include at least the following steroids: 21-deoxycortisol (21DF), 21-deoxycorticosterone (21DB), 11-hydroxyandrostenedione (11bOHA), 17-OH-progesterone (17OHP), androstenedione (A), corticosterone (B), and testosterone (T). Preferably, the physiological parameters also include at least one of the following steroids: pregnenolone (Preg), dehydroepiandrosterone (DHEA), cortisol (F), cortisone (E), 11-deoxycorticosterone (DOC), 17-OH-pregnenolone (17-OHPreg), aldosterone (ALDO), and progesterone (P). Preferably, the physiological parameters should include only steroids.
[0067] The status is binary and indicates either a non-classical form of 21-hydroxylase (NC21OHD) deficiency or the absence of this deficiency. For proper operation of process 400, the status of at least one woman indicates that she has a non-classical form of 21-hydroxylase (NC21OHD) deficiency, and the status of at least one other woman indicates the absence of this deficiency.
[0068] For some of the women in the main development cohort, measurements of certain steroids may be missing.
[0069] Furthermore, during step 402, p1...p15 measurements of physiological parameters are obtained for each of several women in a validation cohort, along with a non-classical form of 21-hydroxylase deficiency (NC21 OHD) status for that woman. Preferably, a polycystic ovary syndrome (PCOS) status is also obtained for each woman in the validation cohort.
[0070] In a 404 step, the measurements of the main development cohort are analyzed to identify statistically outliers for one or more women in the cohort. Furthermore, each woman with statistically outliers is removed from the main development cohort. To perform this data analysis, the measurements are first centered and reduced, for example. Principal component analysis (PCA) can then be used to identify one or more sets of statistically outliers.
[0071] During step 406, the M model N C2IOHD is built using machine learning from measurements of physiological parameters for women in the main development cohort. The goal of machine learning is for the model M NC2IOHD provides an SNC2IOHD score that reliably predicts the status of women in the primary development cohort and, subsequently, of a new woman whose steroid measurements are provided to the MNC2IOHD model. Generally, a training algorithm is used for this purpose. This algorithm is designed to adjust parameters of the MNC2IOHD model based on training data. input and output data, so that the MNC2IOHD model provides known output data from known input data.
[0072] For example, the MNC2IOHD model is a linear combination of measurements of physiological parameters and machine learning is designed to determine weights of this linear combination, so that the linear combination can reliably predict the statuses of women in the main development cohort.
[0073] For example, a partial least squares discriminant analysis (or OPLS-DA, from the English "Orthogonal Partial Least Squares - Discriminant Analysis") is used.
[0074] Indeed, measuring steroid levels in women is a difficult task, so measurements can generally only be obtained for a small number of women, for example, a few dozen to a few hundred. Furthermore, some measurements may be missing. In addition, it is reasonable to assume that steroids are inherently multicollinear. Thus, OPLS-DA analysis seems particularly appropriate because it is effective in cases of multicollinearity, even in a small cohort. Moreover, it is robust, meaning it can be performed even with some missing measurements.
[0075] As is known in itself, by denoting X as a matrix grouping the measurements of physiological parameters for the women in the main development cohort and Y as a vector grouping the statuses of these women, the OPLS-DA analysis decomposes the matrices X and Y as follows: [Math. 3] X = TP T + E x Y = UQ T + E Y where T and U are scores matrices of matrices X and Y respectively, P and Q are loading matrices of matrices X and Y respectively, and Ex and EY are residuals respectively, which the decomposition seeks to make as small as possible.
[0076] The weights of the linear combination, denoted wn with n = 1...N where N is the number of physiological parameters, are then the components of a vector W defined as follows: [Math. 4] T = XW* W* being the adjoint vector (also called the transconjugate) of the vector W.
[0077] The MNC2IOHD model is therefore the linear combination of measurements of physiological parameters, with the weights corresponding respectively to the components wn of the vector W.
[0078] During a 408 step, the MNC2IOHD model can be evaluated, in order to be validated or invalidated.
[0079] The evaluation can be an internal evaluation (i.e., based on the main development cohort) and / or an external evaluation (i.e., based on the validation cohort).
[0080] Internal evaluation may use, for example, one or more of the following: the coefficient of determination R 2 , the permutation test, the prediction coefficient Q 2 and the average accuracy (translation of the English term "accuracy", i.e. the proportion of correct predictions made by the MNC2IOHD model).
[0081] External evaluation uses, for example, the receiver operating characteristic (ROC) curve of the validation cohort. For instance, the area under the curve (AUC) of the ROC curve can be compared to a predefined threshold. The mean precision can also be used for external evaluation, based on the validation cohort (validation precision).
[0082] In step 410, the MPCOS model is built using machine learning from measurements of physiological parameters p1...p15 for women in a so-called ancillary development cohort. This ancillary development cohort might consist of control women from the validation cohort and women with polycystic ovary syndrome (PCOS). The goal of machine learning is for the MPCOS model to provide a Spcos score that reliably predicts the status of women in the ancillary development cohort and, subsequently, of a new woman whose steroid measurements are provided to the MPCOS model. Generally, a training algorithm is used for this purpose. This algorithm is designed to adjust parameters of the MPCOS model based on training data containing input data and output data, so that the MPCOS model provides, from known input data, known output data.
[0083] For example, the MPCOS model is a linear combination of measurements of physiological parameters and machine learning is designed to determine weights of this linear combination, so that the linear combination can reliably predict the statuses of women in the ancillary development cohort.
[0084] For example, a partial least squares discriminant analysis (or OPLS-DA, from the English "Orthogonal Partial Least Squares - Discriminant Analysis") is used.
[0085] During step 412, the M model PC The system can be evaluated to validate or invalidate it. For example, cross-validation from the ancillary development cohort can be performed concurrently with machine learning.
[0086] An example of the implementation of process 400 will now be described.
[0087] Step 402: Obtaining measurements
[0088] In the example described, the primary development cohort comprises 38 women, including 19 with non-classical 21-hydroxylase deficiency (NC21OHD) and 19 control women, meaning they have neither this deficiency nor polycystic ovary syndrome (PCOS). The validation cohort, in the example described, comprises 355 women, including 17 with non-classical 21-hydroxylase deficiency (NC21OHD), 266 women with polycystic ovary syndrome (PCOS), and 72 control women, meaning they have neither of these conditions.
[0089] Furthermore, in the example described, the fifteen steroids detailed previously were measured in each woman in the primary development cohort during the follicular phase of their menstrual cycle. These same fifteen steroids were measured in each woman in the validation cohort during the luteal phase of their menstrual cycle for some women and during the follicular phase for the others.
[0090] Step 404: Statistical verification of measurements
[0091] In the example described, no statistically outliers were found in the measures of the main development cohort.
[0092] Step 406: Obtaining the model
[0093] The main development cohort is used to determine the MNC2IOHD model from measurements of the fifteen steroids. In the example described, OPLS-DA analysis is used, so the resulting MNC2IOHD model is a linear combination of the measurements of the fifteen steroids. Figure 5 is a odds graph of the measurements of women in the main development cohort. As can be seen, the measurements of women with non-classical form 21-hydroxylase deficiency (NC21OHD) (dark gray dots) and the measurements of control women (light gray dots) form two distinct groups, respectively to the left and right of the central vertical line.
[0094] Step 408: Validation of the MNC2IOHD model
[0095] Figure 6 is a odds graph of measurements of women in the main development cohort (dark gray dots and light gray dots, as in Figure 5) and measurements of women in the validation cohort (black stars), without distinction for the latter between control women, women with non-classical deformed 21-hydroxylase deficiency (NC21OHD) and women with polycystic ovary syndrome (PCOS).
[0096] Figure 7 is similar to Figure 6, without the women with polycystic ovary syndrome (PCOS) and distinguishing control women (light grey stars) from women with non-classical deformed 21-hydroxylase deficiency (NC21 OHD) (dark grey stars).
[0097] Figure 8 is similar to Figure 7, without the control women and women with non-classical form 21-hydroxylase deficiency (NC21 OHD), but with women with polycystic ovary syndrome (PCOS) (black stars).
[0098] As can be seen in Figure 6, Figure 7 and Figure 8, the MNC2IOHD model obtained makes it possible to distinguish women with non-classical form 21-hydroxylase deficiency (NC21 OHD) from not only control women, but also from women with polycystic ovary syndrome (PCOS).
[0099] Figure 9 is a odds ratio graph of measurements for control women and women with polycystic ovary syndrome (PCOS) in the validation cohort, for whom steroids were measured during the luteal phase of their cycle. menstrual. As can be seen, despite the luteal phase measurements, they are well recognized as being at low risk of non-classical form 21-hydroxylase deficiency (NC21OHD).
[0100] Surprisingly, reliable results were obtained even for women whose measurements were taken during the luteal phase of their menstrual cycle. Therefore, the selected steroids eliminate the need to rely on the timing of measurements.
[0101] Figure 10 is a variable importance in projection (VIP) graph of the fifteen steroids in the M model N C2IOHD-
[0102] As can be seen in this figure, the seven most important steroids are: 21-deoxycortisol (21DF), 21-deoxycorticosterone (21DB), 11(3-hydroxyandrostenedione (11bOHA), 17-OH-progesterone (17-OHP), androstenedione (A), corticosterone (B), and testosterone (T). It has even been found that using these seven steroids, without the eight others listed previously, allows for a fairly reliable assessment, even with measurements taken during the luteal phase.
[0103] It would be beneficial if new women could join the core development cohort to refine the MNC2IOHD model. This would allow for a new OPLS-DA analysis (such as that implemented, for example, in the steps of process 400 detailed above) to be performed with the expanded development cohort, thus increasing the number of new women, in order to obtain updated weights for the MNC2IOHD model. These updated weights can then be transferred to the systems 100 for further updating.
[0104] Step 410: Obtaining the MPCQS model
[0105] The ancillary development cohort consists of 266 women with polycystic ovary syndrome (PCOS) and 72 control women from the validation cohort. This ancillary development cohort is used to obtain the M model. PC bone.
[0106] Figure 11 is a variable importance in projection graph (also known as VIP) of the fifteen steroids in the MPCOS model.
[0107] As can be seen in this figure, the two most important steroids are: androstenedione (A) and 11 p-hydroxyandrostenedione (11 bOHA), possibly supplemented by pregnenolone (Preg) and 21-deoxycorticosterone (21 DB).
[0108] As can be seen in Figure 12, the MPCOS model obtained makes it possible to distinguish women with polycystic ovarian disease (PCOS) (circles) from control women (stars).
[0109] In conclusion, it is clear that a system such as the one described above allows for a reliable assessment of the risk of non-classical form 21-hydroxylase deficiency (NC21OHD).
[0110] It should be noted that the invention is not limited to the embodiments described above. Indeed, it will be apparent to those skilled in the art that various modifications can be made to the embodiments described above, in light of the information just disclosed to them.
[0111] In particular, the M model PC os could be used independently of the MNC2IOHD model to provide an assessment of a medical risk of polycystic ovary syndrome (PCOS), for example in cases where a medical risk of non-classical form 21-hydroxylase deficiency (NC21 OHD) of the woman to be assessed has been ruled out by a method other than the use of the MNC2IOHD model.
[0112] In the detailed presentation of the invention given above, the terms used shall not be interpreted as limiting the invention to the embodiments set forth in this description, but shall be interpreted to include all equivalents which can be foreseen by a person skilled in the art by applying their general knowledge to the implementation of the teaching which has just been disclosed to them.
Claims
Demands [1] An automated system (100) for processing measurements of physiological parameters to provide a medical risk assessment, characterized in that it comprises: a module (104) for receiving measurements (p1...p15) of physiological parameters of a woman to be assessed; and a module (106) for calculating a score (SNC2IOHD) of non-classical form 21-hydroxylase deficiency of the woman to be assessed, according to a model that gives the score (SNC2IOHD) of non-classical form 21-hydroxylase deficiency from the measurements (p1...p15) received; the physiological parameters comprising at least the following steroids: 21-deoxycortisol (21 DF); 21 -deoxycorticosterone (21 DB); 11 p-hydroxyandrostenedione (11 bOHA); 17-OH-progesterone (17OHP); androstenedione (A); corticosterone (B); and testosterone (T). [2] Automated system (100) according to claim 1, wherein the physiological parameters further comprise at least one of the following steroids: pregnenolone (Preg); dehydroepiandrosterone (DHEA); cortisol (F); cortisone (E); 11-deoxycorticosterone (DOC); 17-OH-pregnenolone (17OHPreg); aldosterone (ALDO); and progesterone (P). [3] Automated system (100) according to claim 1 or 2, wherein the physiological parameters include only steroids. [4] Automated system (100) according to any one of claims 1 to 3, wherein the model giving the score (SNC2IOHD) of non-classical form 21-hydroxylase deficiency is pre-trained by machine learning. [5] Automated system (100) according to claim 4, wherein the model giving the score (SNC2IOHD) of non-classical form 21-hydroxylase deficiency comprises a linear combination of measurements of physiological parameters. [6] Automated system (100) according to any one of claims 1 to 5, wherein at least one of the steroids was measured during a luteal phase of a menstrual cycle of the woman to be evaluated. [7] Automated system (100) according to any one of claims 1 to 6, further comprising: a module (108) for comparing the (SNC2IOHD) score of non-classical form 21-hydroxylase deficiency with a predefined threshold, for example 0.5; and a module (110) for calculating, when the (SNC2IOHD) score of non-classical form 21-hydroxylase deficiency is on a predefined side of the threshold, a (Spcos) score of polycystic ovary of the woman to be evaluated, according to a model giving the (Spcos) score of polycystic ovary from at least a part of the measurements (p1...p15) received. [8] Automated system (100) according to claim 7, wherein the polycystic ovarian disease (PCOS) score is given from at least steroid measurements: 1 i p-hydroxyandrostenedione (11 bOHA) and androstenedione (A), preferably from additional measurements of steroids: testosterone (T) and corticosterone (B). [9] A method (400) for obtaining a model for an automated system (100) according to any one of claims 1 to 8, comprising: for each of several women in a cohort, obtaining (402) measurements of the woman's physiological parameters and a status indicating that this woman either has the non-classical form of 21-hydroxylase deficiency or does not have this deficiency; and determining (408) the model, by training the model with a training algorithm, from the measurements of the physiological parameters of the women in the cohort. [10] Method (400) according to claim 9, wherein all steroids were measured during a follicular phase of a menstrual cycle. [11] A method (200) for the automated processing of physiological parameter measurements to provide a medical risk assessment, characterized in that it comprises: receiving measurements (p1...p15) of physiological parameters from a woman to be assessed; and using (204) a model to provide a score (S) of non-classical form 21-hydroxylase deficiency for the woman to be assessed, based on the received measurements (p1...p15); the physiological parameters comprising at least the following steroids: 21-deoxycortisol (21 DF); 21 -deoxycorticosterone (21 DB); 11 p-hydroxyandrostenedione (11 bOHA); 17-OH-progesterone (17OHP); androstenedione (A); corticosterone (B); and testosterone (T). [12] Method (200) according to claim 11, wherein the physiological parameters further comprise at least one of the following steroids: pregnenolone (Preg); dehydroepiandrosterone (DHEA); cortisol (F); cortisone (E); 11-deoxycorticosterone (DOC); 17-OH-pregnenolone (17OHPreg); aldosterone (ALDO); and progesterone (P). [13] Method (200) according to claim 11 or 12, wherein the physiological parameters comprise only steroids. [14] Method (200) according to any one of claims 11 to 13, wherein the model giving the score (SNC2IOHD) of non-classical form 21-hydroxylase deficiency is pre-trained by machine learning. [15] Method (200) according to any one of claims 11 to 14, wherein the model giving the non-classical form 21-hydroxylase deficiency score comprises a linear combination of measurements of physiological parameters. [16] Method (200) according to any one of claims 11 to 15, wherein at least one of the steroids was measured during a luteal phase of a menstrual cycle of the woman to be evaluated. [17] A method (200) according to any one of claims 11 to 16, further comprising: a comparison of the (SNC2IOHD) score of non-classical form 21-hydroxylase deficiency with a predefined threshold, for example 0.5; and a calculation, when the (SNC2IOHD) score of non-classical form 21-hydroxylase deficiency is on a predefined side of the threshold, of a (Spcos) score of polycystic ovary of the woman to be assessed, following a model giving the (Spcos) score of polycystic ovary from at least a part of the measurements (p1 ...p15) received. [18] A method according to claim 17, wherein the polycystic ovarian syndrome (PCOS) score is given from at least the measurements of the steroids: 110-hydroxyandrostenedione (11 bOHA) and androstenedione (A), preferably from further measurements of the steroids: testosterone (T) and corticosterone (B). [19] Computer program (124) downloadable from a communication network and / or stored on a computer-readable medium, characterized in that it includes instructions for carrying out a process according to any one of claims 11 to 18, when said computer program is executed on a computer. [20] Method for assessing the medical risk of non-classical form 21-hydroxylase deficiency in a woman to be assessed, comprising: collecting physiological parameters of the woman to be assessed, the physiological parameters including at least the following steroids: 21-deoxycortisol (21DF), 21-deoxycorticosterone (21DB), 110-hydroxyandrostenedione (11bOHA), 17-OH-progesterone (17OHP), androstenedione (A), corticosterone (B), and testosterone (T); calculating a score (S) of non-classical form 21-hydroxylase deficiency by entering these physiological parameters into an automated system according to any one of claims 1 to 6; and comparing this score (S) with a threshold, the woman to be assessed being assessed as being at risk of having non-classical form 21-hydroxylase deficiency if this score is below the threshold. [21] Method according to claim 20, wherein at least one of the steroids is measured during a luteal phase of a menstrual cycle of the woman to be evaluated.