Machine learning application for development of a data-driven predictive model able to investigate quality of life scores in a rare disease

Abstract

Background Alkaptonuria (AKU) is an ultra-rare autosomal recessive disease caused by a mutation in the homogentisate 1,2-dioxygenase (HGD) gene. One of the main obstacles in studying AKU, and other ultra-rare diseases, is the lack of a standardized methodology to assess disease severity or response to treatment. Quality of Life scores (QoL) are a reliable way to monitor patients’ clinical condition and health status. QoL scores allow to monitor the evolution of diseases and assess the suitability of treatments by taking into account patients’ symptoms, general health status and care satisfaction. However, more comprehensive tools to study a complex and multi-systemic disease like AKU are needed. In this study, a Machine Learning (ML) approach was implemented with the aim to perform a prediction of QoL scores based on clinical data deposited in the ApreciseKUre, an AKU- dedicated database. Method Data derived from 129 AKU patients have been firstly examined through a preliminary statistical analysis (Pearson correlation coefficient) to measure the linear correlation between 11 QoL scores. The variable importance in QoL scores prediction of 110 ApreciseKUre biomarkers has been then calculated using XGBoost, with K-nearest neighbours algorithm (k-NN) approach. Due to the limited number of data available, this model has been validated using surrogate data analysis. Results We identified a direct correlation of 6 (age, Serum Amyloid A, Chitotriosidase, Advanced Oxidation Protein Products, S-thiolated proteins and Body Mass Index) out of 110 biomarkers with the QoL health status, in particular with the KOOS (Knee injury and Osteoarthritis Outcome Score) symptoms (Relative Absolute Error (RAE) 0.25). The error distribution of surrogate-model (RAE 0.38) was unequivocally higher than the true-model one (RAE of 0.25), confirming the consistency of our dataset. Our data showed that inflammation, oxidative stress, amyloidosis and lifestyle of patients correlates with the QoL scores for physical status, while no correlation between the biomarkers and patients’ mental health was present (RAE 1.1). Conclusions This proof of principle study for rare diseases confirms the importance of database, allowing data management and analysis, which can be used to predict more effective treatments.

Towards a Precision Medicine Approach Based on Machine Learning for Tailoring Medical Treatment in Alkaptonuria

“…The classification was carried out using the RF algorithm and comparing its performance against LR and SVM in order to obtain the best result which were then validated. In accordance with our previous study, [ 14 ], the algorithm prediction performs best for KOOS daily living, KOOS sport and KOOS symptoms. In fact, despite the rather limited amount of data, about 70% of the records where correctly classified.…”

“…Each patient in the ApreciseKUre database is characterized by more than 100 features (for the complete list see Supplementary Materials S1 ), describing biochemical (i.e., SAA, CHIT1 and PTI), clinical, genotypic information and replies to questionnaires evaluating QoL scores. It has been performed patients assessment involving 11 QoL scores: (i) physical health score (PHS), (ii) mental health score (MHS); (iii) AKU Severity Score Index (AKUSSI) for joint pain (AJP) and (iv) AKUSSI spinal pain (ASP); (v) Knee injury and Osteoarthritis Outcome Score (KOOS) pain (KOOSp), (vi) KOOS symptoms (KOOSs), (vii) KOOS daily living (KOOSdl), (viii) KOOS sport (KOOSsp), (ix) KOOS QOL; (x) Health Assessment Questionnaire Disability Index (HAQ-DI) and (xi) global pain visual analog scale (hapVAS) (for more details about each score, see supplementary materials in [ 14 ]). Moreover, it includes information about drugs taken.…”

“…To overcome the limitations due to the scarcity of specimens and data available for AKU and the wide range of AKU symptoms, we have recently established a comprehensive digital ecosystem, ApreciseKUre, that integrates patient-derived information (QoL scores, lifestyle), clinician-derived information (urine, blood, plasma analysis), mutational analysis (genotypes, protein stability) and therapeutic treatments offering an exhaustive visualization of different informative layers, to support clinicians and researchers in a PM approach to AKU [ 10 , 11 , 12 , 13 , 14 , 15 ]. The ApreciseKUre database can be a good starting point for the creation of a new clinical management tool in AKU, which will lead to the development of a deeper knowledge network on the disease and will advance its treatment [ 10 , 11 , 12 ].…”

Quality of life of patients with alkaptonuria

“…В то же время при орфанных заболеваниях КЖ оценивается редко. Так, КЖ пациентов с АКУ изучалось лишь в четырех зарубежных исследованиях [12][13][14][15]. Выявлено существенное снижение КЖ, особенно физических функций по SF-36, значительные нарушения зафиксированы по опроснику КООS, характеризующему функциональное состояние коленного сустава.…”

HGDiscovery: an online tool providing functional and phenotypic information on novel variants of homogentisate 1,2- dioxigenase

“…An in-depth molecular characterization of the disease is needed in pharmacogenomics prediction for suitable medical treatment. To address the issue we developed ApreciseKUre platform, which includes data on potential biomarkers, patients' quality of life, biochemical outcomes and clinical information facilitating their integration and analysis in order to shed light on pathological characterization of every AKU patient in a typical Precision Medicine perspective [13][14][15][16] .…”