R Garcia-Carretero1, M Olid-Velilla2, D Perez-Torrella3, N Torres-Pacho2, M-T Darnaude-Ortiz4, A-D Bustamate-Zuloeta4, J-A Tenorio5. 1. Department of Internal Medicine, Mostoles University Hospital, Rey Juan Carlos University, Madrid, Spain. rgcarretero@salud.madrid.org. 2. Department of Internal Medicine, Mostoles University Hospital, Madrid, Spain. 3. Department of Laboratory of Clinical Analysis, Mostoles University Hospital, Madrid, Spain. 4. Department of Clinical Genetics, Mostoles University Hospital, Madrid, Spain. 5. Institute of Medical and Molecular Genetics (INGEMM), La Paz University Hospital, Madrid, Spain.
Abstract
Approximately half of individuals with hypophosphatasemia (low levels of serum alkaline phosphatase) have hypophosphatasia, a rare genetic disease in which patients may have stress fractures, bone and joint pain, or premature tooth loss. We developed a predictive model based on specific biomarkers of this disease to better diagnose this condition. INTRODUCTION: Hypophosphatasemia is a condition in which low levels of alkaline phosphatase (ALP) are detected in the serum. Some individuals presenting with this condition may have a rare genetic disease called hypophosphatasia (HPP), which involves mineralization of the bone and teeth. Lack of awareness of HPP and its nonspecific symptoms make this genetic disease difficult to diagnose. We developed a predictive model based on biomarkers of HPP such as ALP and pyridoxal 5'-phosphate (PLP), because clinical manifestations sometimes are not recognized as symptoms of HPP. METHODS: We assessed 325,000 ALP results between 2010 and 2015 to identify individuals suspected of having HPP. We performed univariate and multivariate analyses to characterize the relationship between hypophosphatasemia and HPP. Using several machine learning algorithms, we developed several models based on biomarkers and compared their performance to determine the best model. RESULTS: The final cohort included 45 patients who underwent a genetic test. Half (23 patients) showed a mutation of the ALPL gene that encodes the tissue-nonspecific ALP enzyme. ALP (odds ratio [OR] 0.61, 95% confidence interval [CI] 0.3-0.8, p = 0.01) and PLP (OR 1.06, 95% CI 1.01-1.15, p = 0.04) were the only variables significantly associated with the presence of HPP. Support vector machines and logistic regression were the machine learning algorithms that provided the best predictive models in terms of classification (area under the curve 0.936 and 0.844, respectively). CONCLUSIONS: Given the high probability of a misdiagnosis, its nonspecific symptoms, and a lack of awareness of serum ALP levels, it is difficult to make a clinical diagnosis of HPP. Predictive models based on biomarkers are necessary to achieve a proper diagnosis. Our proposed machine learning approaches achieved reasonable performance compared to traditional statistical methods used in biomedicine, increasing the likelihood of properly diagnosing such a rare disease as HPP.
Approximately half of individuals with hypophosphatasemia (low levels of serum alkaline phosphatase) have hypophosphatasia, a rare genetic disease in which patients may have stress fractures, bone and joint pain, or premature tooth loss. We developed a predictive model based on specific biomarkers of this disease to better diagnose this condition. INTRODUCTION: Hypophosphatasemia is a condition in which low levels of alkaline phosphatase (ALP) are detected in the serum. Some individuals presenting with this condition may have a rare genetic disease called hypophosphatasia (HPP), which involves mineralization of the bone and teeth. Lack of awareness of HPP and its nonspecific symptoms make this genetic disease difficult to diagnose. We developed a predictive model based on biomarkers of HPP such as ALP and pyridoxal 5'-phosphate (PLP), because clinical manifestations sometimes are not recognized as symptoms of HPP. METHODS: We assessed 325,000 ALP results between 2010 and 2015 to identify individuals suspected of having HPP. We performed univariate and multivariate analyses to characterize the relationship between hypophosphatasemia and HPP. Using several machine learning algorithms, we developed several models based on biomarkers and compared their performance to determine the best model. RESULTS: The final cohort included 45 patients who underwent a genetic test. Half (23 patients) showed a mutation of the ALPL gene that encodes the tissue-nonspecific ALP enzyme. ALP (odds ratio [OR] 0.61, 95% confidence interval [CI] 0.3-0.8, p = 0.01) and PLP (OR 1.06, 95% CI 1.01-1.15, p = 0.04) were the only variables significantly associated with the presence of HPP. Support vector machines and logistic regression were the machine learning algorithms that provided the best predictive models in terms of classification (area under the curve 0.936 and 0.844, respectively). CONCLUSIONS: Given the high probability of a misdiagnosis, its nonspecific symptoms, and a lack of awareness of serum ALP levels, it is difficult to make a clinical diagnosis of HPP. Predictive models based on biomarkers are necessary to achieve a proper diagnosis. Our proposed machine learning approaches achieved reasonable performance compared to traditional statistical methods used in biomedicine, increasing the likelihood of properly diagnosing such a rare disease as HPP.
Entities:
Keywords:
Alkaline phosphatase; Hypophosphatasemia; Hypophosphatasia; Machine learning; Predictive model
Authors: C Tornero; V Navarro-Compán; A Buño; K E Heath; M Díaz-Almirón; A Balsa; J A Tenorio; J Quer; P Aguado Journal: Orphanet J Rare Dis Date: 2022-03-03 Impact factor: 4.123