Combined unsupervised-supervised machine learning for phenotyping complex diseases with its application to obstructive sleep apnea

Unveiling diagnostic and therapeutic strategies for cervical cancer: biomarker discovery through proteomics approaches and exploring the role of cervical cancer stem cells

Cervical cancer (CC) is a major global health problem and leading cause of cancer deaths among women worldwide. Early detection through screening programs has reduced mortality; however, screening compliance remains low. Identifying non-invasive biomarkers through proteomics for diagnosis and monitoring response to treatment could improve patient outcomes. Here we review recent proteomics studies which have uncovered biomarkers and potential drug targets for CC. Additionally, we explore into the role of cervical cancer stem cells and their potential implications in driving CC progression and therapy resistance. Although challenges remain, proteomics has the potential to revolutionize the field of cervical cancer research and improve patient outcomes.

Predicting the Risk of Sleep Disorders Using a Machine Learning-Based Simple Questionnaire: Development and Validation Study

BACKGROUND

Sleep disorders, such as obstructive sleep apnea (OSA), comorbid insomnia and sleep apnea (COMISA), and insomnia are common and can have serious health consequences. However, accurately diagnosing these conditions can be challenging as a result of the underrecognition of these diseases, the time-intensive nature of sleep monitoring necessary for a proper diagnosis, and patients' hesitancy to undergo demanding and costly overnight polysomnography tests.

OBJECTIVE

We aim to develop a machine learning algorithm that can accurately predict the risk of OSA, COMISA, and insomnia with a simple set of questions, without the need for a polysomnography test.

METHODS

We applied extreme gradient boosting to the data from 2 medical centers (n=4257 from Samsung Medical Center and n=365 from Ewha Womans University Medical Center Seoul Hospital). Features were selected based on feature importance calculated by the Shapley additive explanations (SHAP) method. We applied extreme gradient boosting using selected features to develop a simple questionnaire predicting sleep disorders (SLEEPS). The accuracy of the algorithm was evaluated using the area under the receiver operating characteristics curve.

RESULTS

In total, 9 features were selected to construct SLEEPS. SLEEPS showed high accuracy, with an area under the receiver operating characteristics curve of greater than 0.897 for all 3 sleep disorders, and consistent performance across both sets of data. We found that the distinction between COMISA and OSA was critical for accurate prediction. A publicly accessible website was created based on the algorithm that provides predictions for the risk of the 3 sleep disorders and shows how the risk changes with changes in weight or age.

CONCLUSIONS

SLEEPS has the potential to improve the diagnosis and treatment of sleep disorders by providing more accessibility and convenience. The creation of a publicly accessible website based on the algorithm provides a user-friendly tool for assessing the risk of OSA, COMISA, and insomnia.

Association of Polysomnographic Sleep Parameters With Neuroimaging Biomarkers of Cerebrovascular Disease in Older Adults With Sleep Apnea

BACKGROUND AND OBJECTIVES

Our objective was to determine whether polysomnographic (PSG) sleep parameters are associated with neuroimaging biomarkers of cerebrovascular disease (CVD) related to white matter (WM) integrity in older adults with obstructive sleep apnea (OSA).

METHODS

From the population-based Mayo Clinic Study of Aging, we identified participants without dementia who underwent at least 1 brain MRI and PSG. We quantified 2 CVD biomarkers: WM hyperintensities (WMHs) from fluid-attenuated inversion recovery (FLAIR)-MRI, and fractional anisotropy of the genu of the corpus callosum (genu FA) from diffusion MRI. For this cross-sectional analysis, we fit linear models to assess associations between PSG parameters (NREM stage 1 percentage, NREM stage 3 percentage [slow-wave sleep], mean oxyhemoglobin saturation, and log of apnea-hypopnea index [AHI]) and CVD biomarkers (log of WMH and log of genu FA), respectively, while adjusting for age (at MRI), sex, APOE ε4 status, composite cardiovascular and metabolic conditions (CMC) score, REM stage percentage, sleep duration, and interval between MRI and PSG.

RESULTS

We included 140 participants with FLAIR-MRI (of which 103 had additional diffusion MRI). The mean ± SD age was 72.7 ± 9.6 years at MRI with nearly 60% being men. The absolute median (interquartile range [IQR]) interval between MRI and PSG was 1.74 (0.9-3.2) years. 90.7% were cognitively unimpaired (CU) during both assessments. For every 10-point decrease in N3%, there was a 0.058 (95% CI 0.006-0.111, p = 0.030) increase in the log of WMH and 0.006 decrease (95% CI -0.012 to -0.0002, p = 0.042) in the log of genu FA. After matching for age, sex, and N3%, participants with severe OSA had higher WMH (median [IQR] 0.007 [0.005-0.015] vs 0.006 [0.003-0.009], p = 0.042) and lower genu FA (median [IQR] 0.57 [0.55-0.63] vs 0.63 [0.58-0.65], p = 0.007), when compared with those with mild/moderate OSA.

DISCUSSION

We found that reduced slow-wave sleep and severe OSA were associated with higher burden of WM abnormalities in predominantly CU older adults, which may contribute to greater risk of cognitive impairment, dementia, and stroke. Our study supports the association between sleep depth/fragmentation and intermittent hypoxia and CVD biomarkers. Longitudinal studies are required to assess causation.

Clinical applications of artificial intelligence in sleep medicine: a sleep clinician's perspective

BACKGROUND

The past few years have seen a rapid emergence of artificial intelligence (AI)-enabled technology in the field of sleep medicine. AI refers to the capability of computer systems to perform tasks conventionally considered to require human intelligence, such as speech recognition, decision-making, and visual recognition of patterns and objects. The practice of sleep tracking and measuring physiological signals in sleep is widely practiced. Therefore, sleep monitoring in both the laboratory and ambulatory environments results in the accrual of massive amounts of data that uniquely positions the field of sleep medicine to gain from AI.

METHOD

The purpose of this article is to provide a concise overview of relevant terminology, definitions, and use cases of AI in sleep medicine. This was supplemented by a thorough review of relevant published literature.

RESULTS

Artificial intelligence has several applications in sleep medicine including sleep and respiratory event scoring in the sleep laboratory, diagnosing and managing sleep disorders, and population health. While still in its nascent stage, there are several challenges which preclude AI's generalizability and wide-reaching clinical applications. Overcoming these challenges will help integrate AI seamlessly within sleep medicine and augment clinical practice.

CONCLUSION

Artificial intelligence is a powerful tool in healthcare that may improve patient care, enhance diagnostic abilities, and augment the management of sleep disorders. However, there is a need to regulate and standardize existing machine learning algorithms prior to its inclusion in the sleep clinic.

Using Cluster Analysis to Overcome the Limits of Traditional Phenotype-Genotype Correlations: The Example of RYR1-Related Myopathies

Thanks to advances in gene sequencing, RYR1-related myopathy (RYR1-RM) is now known to manifest itself in vastly heterogeneous forms, whose clinical interpretation is, therefore, highly challenging. We set out to develop a novel unsupervised cluster analysis method in a large patient population. The objective was to analyze the main RYR1-related characteristics to identify distinctive features of RYR1-RM and, thus, offer more precise genotype-phenotype correlations in a group of potentially life-threatening disorders. We studied 600 patients presenting with a suspicion of inherited myopathy, who were investigated using next-generation sequencing. Among them, 73 index cases harbored variants in RYR1. In an attempt to group genetic variants and fully exploit information derived from genetic, morphological, and clinical datasets, we performed unsupervised cluster analysis in 64 probands carrying monoallelic variants. Most of the 73 patients with positive molecular diagnoses were clinically asymptomatic or pauci-symptomatic. Multimodal integration of clinical and histological data, performed using a non-metric multi-dimensional scaling analysis with k-means clustering, grouped the 64 patients into 4 clusters with distinctive patterns of clinical and morphological findings. In addressing the need for more specific genotype-phenotype correlations, we found clustering to overcome the limits of the "single-dimension" paradigm traditionally used to describe genotype-phenotype relationships.

Multidimensional assessment and cluster analysis for OSA phenotyping

STUDY OBJECTIVES

Obstructive sleep apnea (OSA) is a heterogeneous disease with varying phenotype. A cluster analysis based on multidimensional disease characteristics, including symptom, anthropometry, polysomnography (PSG), and craniofacial morphology, in combination with auto-continuous positive airway pressure (CPAP) titration response and comorbidity profiles was performed within a well-characterized cohort of patients with OSA, with the aim to refine current phenotypic expressions of OSA with clinical implications.

METHODS

Two hundred and ninety-one subjects with a new diagnosis of moderate to severe OSA, referred for auto-CPAP titration to the sleep center were included for analysis. In-laboratory PSG and craniofacial computed tomography (CT) scanning was performed, followed by an auto-CPAP titration. The symptom of excessive daytime sleepiness (EDS) was assessed by Epworth sleepiness scale (ESS).

RESULTS

Three patient phenotypes, corresponding to the "normal weight, non-sleepy and moderate OSA", the "obese, non-sleepy and severe OSA" and "obese, sleepy, very severe OSA with craniofacial limitation" were identified. Among the PSG parameters, only N3% and mean pulse oxygen saturation (SPO₂) were found to be associated with ESS, and they only explain small fraction of the variation (R²=0.136). Neck circumference and craniofacial limitation were associated the more severe phenotype, which had higher prevalence of hypertension, metabolic syndrome, greater diurnal blood gas abnormalities and worse PAP titration response.

CONCLUSIONS

Three OSA phenotypes were identified according to multiple aspect of clinical features in patients with moderate to severe OSA, which differed in prevalence of hypertension, metabolic syndrome, diurnal blood gas parameters and CPAP titration response. Self-reported EDS was not related with the severity of sleep breathing disturbance, and craniofacial limitation was associated the more severe phenotype. These findings highlight the necessity of integrate multiple disease characters into phenotyping to achieve better understanding of the clinical pictures of OSA.

Obstructive Sleep Apnea Syndrome Comorbidity Phenotypes in Primary Health Care Patients in Northern Greece

Background: Obstructive sleep apnea syndrome (OSAS) is a significant public health issue. In the general population, the prevalence varies from 10% to 50%. We aimed to phenotype comorbidities in OSAS patients referred to the primary health care (PHC) system. Methods: We enrolled 1496 patients referred to the PHC system for any respiratory- or sleep-related issue from November 2015 to September 2017. Some patients underwent polysomnography (PSG) evaluation in order to establish OSAS diagnosis. The final study population comprised 136 patients, and the Charlson comorbidity index was assessed. Categorical principal component analysis and TwoStep clustering was used to identify distinct clusters in the study population. Results: The analysis revealed three clusters: the first with moderate OSAS, obesity and a high ESS score without significant comorbidities; the second with severe OSAS, severe obesity with comorbidities and the highest ESS score; and the third with severe OSAS and obesity without comorbidities but with a high ESS score. The clusters differed in age (p < 0.005), apnea–hypopnea index, oxygen desaturation index, arousal index and respiratory and desaturation arousal index (p < 0.001). Conclusions: Predictive comorbidity models may aid the early diagnosis of patients at risk in the context of PHC and pave the way for personalized treatment.

Automatic Segmentation to Cluster Patterns of Breathing in Sleep Apnea

Annotation of polysomnography (PSG) recordings for diagnosis of obstructive sleep apnea (OSA) is a standard procedure but an expensive and time-consuming process for clinicians. To aid clinicians in this process we present a data driven unsupervised hierarchical clustering approach for detection and visual presentation of breathing patterns in PSG recordings. The aim was to develop a model independent of manual annotations to detect and visualize respiratory events related to OSA. 10 recordings from the Sleep Heart Health Study database were used, and the proposed algorithm was evaluated based on the manually annotated events for each recording. The algorithm reached an F1-score of 0.58 across the 10 recordings when detecting the presence of an event vs. no event and a 100% correct diagnosis prediction of OSA when predicting if apnea-hypopnea index (AHI) ≥ 15, which is a clinically meaningful cut-off. The F1-score may be due to imprecise placement of events, difficulty distinguishing between hypopneas and stable breathing, and variations in scoring. In conclusion the performance can be improved despite the strong agreement in diagnostics. The method is a proof of concept that a clustering method can detect and visualize breathing patterns related to OSA while maintaining a correct diagnosis.

Data-driven identification of temporal glucose patterns in a large cohort of nondiabetic patients with COVID-19 using time-series clustering

Objective

Hyperglycemia has emerged as an important clinical manifestation of coronavirus disease 2019 (COVID-19) in diabetic and nondiabetic patients. Whether these glycemic changes are specific to a subgroup of patients and persist following COVID-19 resolution remains to be elucidated. This work aimed to characterize longitudinal random blood glucose in a large cohort of nondiabetic patients diagnosed with COVID-19.

Materials and Methods

De-identified electronic medical records of 7502 patients diagnosed with COVID-19 without prior diagnosis of diabetes between January 1, 2020, and November 18, 2020, were accessed through the TriNetX Research Network. Glucose measurements, diagnostic codes, medication codes, laboratory values, vital signs, and demographics were extracted before, during, and after COVID-19 diagnosis. Unsupervised time-series clustering algorithms were trained to identify distinct clusters of glucose trajectories. Cluster associations were tested for demographic variables, COVID-19 severity, glucose-altering medications, glucose values, and new-onset diabetes diagnoses.

Results

Time-series clustering identified a low-complexity model with 3 clusters and a high-complexity model with 19 clusters as the best-performing models. In both models, cluster membership differed significantly by death status, COVID-19 severity, and glucose levels. Clusters membership in the 19 cluster model also differed significantly by age, sex, and new-onset diabetes mellitus.

Discussion and Conclusion

This work identified distinct longitudinal blood glucose changes associated with subclinical glucose dysfunction in the low-complexity model and increased new-onset diabetes incidence in the high-complexity model. Together, these findings highlight the utility of data-driven techniques to elucidate longitudinal glycemic dysfunction in patients with COVID-19 and provide clinical evidence for further evaluation of the role of COVID-19 in diabetes pathogenesis.

The Future of Sleep Measurements: A Review and Perspective

This article provides an overview of the current use, limitations, and future directions of the variety of subjective and objective sleep assessments available. This article argues for various ways and sources of collecting, combining, and using data to enlighten clinical practice and the sleep research of the future. It highlights the prospects of digital management platforms to store and present the data, and the importance of codesign when developing such platforms and other new instruments. It also discusses the abundance of opportunities that data science and machine learning open for the analysis of data.