Search

EP-4736185-A1 - PREDICTIVE ASSESSMENT OF HEALTH RISK PARAMETERS REGARDING GESTATIONAL DIABETES MELLITUS AND PREECLAMPSIA

EP4736185A1EP 4736185 A1EP4736185 A1EP 4736185A1EP-4736185-A1

Abstract

The invention provides a computer-implemented method for predictive assessment of a health risk parameter, such as a risk for diabetes, in particular gestational diabetes mellitus, and/or preeclampsia, for a woman based on at least one pregnancy-related parameter, comprising the following steps: providing an assessment of the health risk parameter using a first machine-learning model (1) if the pregnancy-related parameter fulfils a first condition; providing an assessment of the health risk parameter using a second machine-learning model (2) if the pregnancy-related parameter fulfils a second condition; and making available the health risk parameter.

Inventors

  • JOHNSON, MARK THOMAS
  • ASVADI, SIMA
  • PALERO, Jonathan Alambra
  • WU, YANQI
  • LONG, XI

Assignees

  • Koninklijke Philips N.V.

Dates

Publication Date
20260506
Application Date
20240621

Claims (11)

  1. 1. A computer-implemented method for predictive assessment of a health risk parameter, such as a risk for diabetes, in particular gestational diabetes mellitus, and/or pre-eclampsia, for a woman based on at least one pregnancy-related parameter, comprising the following steps: providing an assessment of the health risk parameter using a first machine-learning model (1) if the pregnancy-related parameter fulfils a first condition; providing an assessment of the health risk parameter using a second machine-learning model (2) if the pregnancy-related parameter fulfils a second condition; and making available the health risk parameter, wherein the first machine-learning model (1) and the second machine-learning model (2) are deployed on separate cohorts, and wherein the first machine-learning model (1) and the second machine-learning model (2) have been trained on separate cohorts.
  2. 2. The method of claim 1, wherein the first and second condition are of such nature that they mutually exclude each other.
  3. 3. The method of claim 1 or 2, wherein the pregnancy-related parameter comprises a number of pregnancies already carried out by the woman and/or an ordinal number count of an ongoing pregnancy.
  4. 4. The method of any of the preceding claims, wherein the first condition is fulfilled if the woman is having her first pregnancy and/or wherein the second condition is fulfilled if the woman is having her second or subsequent pregnancy.
  5. 5. The method of any of the preceding claims, further comprising: providing an assessment of the health risk parameter using a third machine-learning model (3) if the pregnancy-related parameter fulfils the second condition as well as a third condition, in particular a third condition relating to a prior history of the woman with respect to the health condition whose risk is assessed using the health risk parameter.
  6. 6. The method of claim 5, further comprising: providing an assessment of the health risk parameter using the second machine-learning model (2) if the pregnancy-related parameter fulfils the second condition, but not the third condition, in particular with the third condition relating to a prior history of the woman with respect to the health condition whose risk is assessed using the health risk parameter.
  7. 7. The method of any of the preceding claims, performed during the first trimester of an ongoing pregnancy.
  8. 8. A computer-implemented machine-learning model structure for predictive assessment of a health risk parameter, such as a risk for diabetes, in particular gestational diabetes mellitus, and/or preeclampsia, for a woman based on at least one pregnancy-related parameter, comprising a first machinelearning model (1) and a second first machine-learning model (2), wherein the first machine-learning model (1) may provide an assessment of the health risk parameter if the pregnancy-related parameter fulfils a first condition; and the second machine-learning model (2) may provide an assessment of the health risk parameter if the pregnancy-related parameter fulfils a second condition, wherein the first machine-learning model (1) and the second machine-learning model (2) are deployed on separate cohorts, and wherein the first machine-learning model (1) and the second machine-learning model (2) have been trained on separate cohorts.
  9. 9. A method of training the machine-learning model structure of claim 8, comprising training one or more machine-learning models using nulliparous and multiparous cohort data; and deploying the one or more machine-learning models on separate cohorts as first machinelearning model and the second machine-learning model; particularly: wherein training the first machine-learning model (1) is based on the nulliparous cohort data whereas training the second machine-learning model (2) is based on the multiparous cohort data.
  10. 10. A data processing device such as a computer or a mobile phone, in particular an electronic personal and/or professional health device, or a controller for such a data processing device, comprising means for carrying out the method of any one of the preceding claims 1-7 or 9.
  11. 11. A computer program, or a computer-readable medium storing a computer program, the computer program comprising instructions which, when the program is executed on a computer, cause the computer to carry out the method of any one of claims 1-7 or 9. A data structure comprising the machine-learning model structure of claim 8.

Description

PREDICTIVE ASSESSMENT OF HEALTH RISK PARAMETERS REGARDING GESTATIONAL DIABETES MELLITUS AND PREECLAMPSIA FIELD OF THE INVENTION The present invention generally concerns the field of health risk prediction, in particular in the context of coming and/or ongoing pregnancies. Exemplary embodiments of the present invention have applications with respect to gestational diabetes mellitus (GDM) or preeclampsia. High accuracy in prediction can be accomplished using only limited data and limited computational resources. BACKGROUND OF THE INVENTION Digitalization is ubiquitous in today’s world and fundamentally changes the way we live and communicate. It may bring many unprecedented advantages to society. To that end, digitalization not only enables a more efficient realization of traditional processes, but also allows very different and new approaches to problems that were not available before. This is equally true for the health sector. Gestational diabetes mellitus (GDM) is a type of diabetes that can develop during pregnancy. Around 1 in 7 pregnant women will develop GDM during pregnancy. GDM develops if excessive glucose is accumulated in maternal blood. During pregnancy certain pregnancy hormones secreted by the placenta, such as oestrogen, cortisol and lactogen inhibit insulin functioning. Insulin therefore becomes less effective in transferring glucose from the blood stream to the expecting mother’s tissues. This process is essential to ensure an adequate nutrient supply to the baby. However, the process also causes insulin resistance in the mother’s body, which also increases as the pregnancy advances. If the mother’s body does not secrete enough insulin to balance this resistance, her blood sugar increases and GDM occurs. Women with gestational diabetes do not have diabetes before their pregnancy and their GDM usually resolves after giving birth. Gestational diabetes can occur at any stage of a pregnancy but is often diagnosed during the second or third trimester. Women who develop GDM might need to take medication to regulate their blood glucose. If the level of blood glucose is not controlled, it might contribute to developing other pregnancy related complications such as pre-eclampsia (high blood pressure and protein in the urine) and in some cases might require a caesarean section. Furthermore, they might become prone to developing diabetes later in life. For babies, potential consequences of uncontrolled gestational diabetes of the mother include growing larger than they normally would or having hypoglycaemia (low blood sugar) after birth. Generally, starting pregnancy with a (significantly) high weight (i.e., BMI), having diabetes in the family, or having developed GDM in a previous pregnancy might increase the chance of GDM for the current pregnancy. Scientific research has shown that early lifestyle modifications during pregnancy can reduce the chance of developing GDM. This effect is enhanced if lifestyle adjustments are made early on, typically before week 15 of pregnancy, and maintained during pregnancy. There is thus a need to accurately predict the risk of GDM at an early stage during pregnancy. The accuracy of models for early risk prediction of GDM may depend on the risk factors that are included in training the model. It is known from the prior art to use factors in GDM risk prediction models such as: Age, weight or BMI, Ethnicity (race), family history of diabetes, parity, gravidity and history of GDM, as well as certain blood-based biomarkers such as hsCRP, SHBG, adipokines as well as inflammatory markers like TNF-alpha. The present GDM early risk prediction models are, however, providing limited risk prediction accuracy with area under curve (AUC) in the range between 0.70 to 0.80. This has hence disadvantages resulting from the limited accuracy of the prediction, leading to false decisions in treatment, preventions and adaptions of lifestyle. Higher accuracy and precision would hence be highly desirable. It is hence an object of the present invention to provide methods for improving the risk prediction accuracy of risk prediction models, such as particularly GDM early risk prediction models. Often, data for model training is difficult to obtain and quantitatively limited. It is hence another object of the invention to provide for high-precision results based on limited data and limited computational resources. US Patent Application Number US 2022/181030 Al describes a method for predicting the risk of pre-eclampsia in pregnant obese women based on metabolite biomarkers. SUMMARY OF THE INVENTION A solution to the problem has now been devised by the subject-matter of the independent claims. Accordingly, a computer-implemented method for predictive assessment of a health risk parameter is provided as defined in independent claim 1. The health risk parameter may, for example, be a risk for diabetes, in particular gestational diabetes mellitus, and/or pre-eclampsia. The health risk parameter