Interpretable prediction of mortality in liver transplant recipients based on machine learning.

BACKGROUND: Accurate prediction of the mortality of post-liver transplantation is an important but challenging task. It relates to optimizing organ allocation and estimating the risk of possible dysfunction. Existing risk scoring models, such as the Balance of Risk (BAR) score and the Survival Outcomes Following Liver Transplantation (SOFT) score, do not predict the mortality of post-liver transplantation with sufficient accuracy. In this study, we evaluate the performance of machine learning models and establish an explainable machine learning model for predicting mortality in liver transplant recipients. METHOD: The optimal feature set for the prediction of the mortality was selected by a wrapper method based on binary particle swarm optimization (BPSO). With the selected optimal feature set, seven machine learning models were applied to predict mortality over different time windows. The best-performing model was used to predict mortality through a comprehensive comparison and evaluation. An interpretable approach based on machine learning and SHapley Additive exPlanations (SHAP) is used to explicitly explain the model's decision and make new discoveries. RESULTS: With regard to predictive power, our results demonstrated that the feature set selected by BPSO outperformed both the feature set in the existing risk score model (BAR score, SOFT score) and the feature set processed by principal component analysis (PCA). The best-performing model, extreme gradient boosting (XGBoost), was found to improve the Area Under a Curve (AUC) values for mortality prediction by 6.7%, 11.6%, and 17.4% at 3 months, 3 years, and 10 years, respectively, compared to the SOFT score. The main predictors of mortality and their impact were discussed for different age groups and different follow-up periods. CONCLUSIONS: Our analysis demonstrates that XGBoost can be an ideal method to assess the mortality risk in liver transplantation. In combination with the SHAP approach, the proposed framework provides a more intuitive and comprehensive interpretation of the predictive model, thereby allowing the clinician to better understand the decision-making process of the model and the impact of factors associated with mortality risk in liver transplantation.

Weak Association between Skin Autofluorescence Levels and Prediabetes with an ILERVAS Cross-Sectional Study.

A large body of evidence demonstrates a relationship between hyperglycemia and increased concentrations of advanced glycation end-products (AGEs). However, there is little information about subcutaneous AGE accumulation in subjects with prediabetes, and whether or not this measurement could assist in the diagnosis of prediabetes is unclear. A cross-sectional study was conducted in 4181 middle-aged subjects without diabetes. Prediabetes (n = 1444) was defined as a glycosylated hemoglobin (HbA1c) level between 39 and 47 mmol/mol (5.7 to 6.4%), and skin autofluorescence (SAF) measurement was performed to assess AGEs. A multivariable logistic regression model and receiver operating characteristic curve were used. The cohort consisted of 50.1% women with an age of 57 [52;62] years, a BMI of 28.3 [25.4;31.6] kg/m2, and a prevalence of prediabetes of 34.5%. Participants with prediabetes showed higher SAF than control participants (2.0 [1.7;2.2] vs. 1.9 [1.7;2.2], p < 0.001). However, HbA1c was not significantly correlated with SAF levels (r = 0.026, p = 0.090). In addition, the SAF level was not independently associated with prediabetes (OR = 1.12 (0.96 to 1.30)). Finally, there was no good cutoff point for SAF to identify patients with prediabetes (AUC = 0.52 (0.50 to 0.54), sensitivity = 0.61, and 1-specificity = 0.56). Given all of this evidence, we can conclude that although there is an increase in SAF levels in participants with prediabetes, the applicability and clinical relevance of the results is low in this population.

Prediabetes Is Associated with Increased Prevalence of Sleep-Disordered Breathing.

Type 2 diabetes leads to severe nocturnal hypoxemia, with an increase in apnea events and daytime sleepiness. Hence, we assessed sleep breathing parameters in the prediabetes stage. A cross-sectional study conducted on 966 middle-aged subjects without known pulmonary disease (311 patients with prediabetes and 655 controls with normal glucose metabolism) was conducted. Prediabetes was defined by glycated hemoglobin (HbA1c), and a nonattended overnight home sleep study was performed. Participants with prediabetes (n = 311) displayed a higher apnea−hypopnea index (AHI: 12.7 (6.1;24.3) vs. 9.5 (4.2;19.6) events/h, p < 0.001) and hypopnea index (HI: 8.4 (4.0;14.9) vs. 6.0 (2.7;12.6) events/h, p < 0.001) than controls, without differences in the apnea index. Altogether, the prevalence of obstructive sleep apnea was higher in subjects with prediabetes than in controls (78.1 vs. 69.9%, p = 0.007). Additionally, subjects with prediabetes presented impaired measurements of the median and minimum nocturnal oxygen saturation, the percentage of time spent with oxygen saturations below 90%, and the 4% oxygen desaturation index in comparison with individuals without prediabetes (p < 0.001 for all). After adjusting for age, sex, and the presence of obesity, HbA1c correlated with the HI in the entire population (r = 0.141, p < 0.001), and the presence of prediabetes was independently associated with the AHI (B = 2.20 (0.10 to 4.31), p = 0.040) as well as the HI (B = 1.87 (0.61 to 3.14), p = 0.004) in the multiple linear regression model. We conclude that prediabetes is an independent risk factor for an increased AHI after adjusting for age, sex, and obesity. The enhanced AHI is mainly associated with increments in the hypopnea events.

Development and Validation of a Model to Predict Severe Hospital-Acquired Acute Kidney Injury in Non-Critically Ill Patients.

BACKGROUND: The current models developed to predict hospital-acquired AKI (HA-AKI) in non-critically ill fail to identify the patients at risk of severe HA-AKI stage 3. OBJECTIVE: To develop and externally validate a model to predict the individual probability of developing HA-AKI stage 3 through the integration of electronic health databases. METHODS: Study set: 165,893 non-critically ill hospitalized patients. Using stepwise logistic regression analyses, including demography, chronic comorbidities, and exposure to risk factors prior to AKI detection, we developed a multivariate model to predict HA-AKI stage 3. This model was then externally validated in 43,569 non-critical patients admitted to the validation center. RESULTS: The incidence of HA-AKI stage 3 in the study set was 0.6%. Among chronic comorbidities, the highest odds ratios were conferred by ischemic heart disease, ischemic cerebrovascular disease, chronic congestive heart failure, chronic obstructive pulmonary disease, chronic kidney disease and liver disease. Among acute complications, the highest odd ratios were associated with acute respiratory failure, major surgery and exposure to nephrotoxic drugs. The model showed an AUC of 0.906 (95% CI 0.904 to 0.908), a sensitivity of 89.1 (95% CI 87.0-91.0) and a specificity of 80.5 (95% CI 80.2-80.7) to predict HA-AKI stage 3, but tended to overestimate the risk at low-risk categories with an adequate goodness-of-fit for all risk categories (Chi2: 16.4, p: 0.034). In the validation set, incidence of HA-AKI stage 3 was 0.62%. The model showed an AUC of 0.861 (95% CI 0.859-0.863), a sensitivity of 83.0 (95% CI 80.5-85.3) and a specificity of 76.5 (95% CI 76.2-76.8) to predict HA-AKI stage 3 with an adequate goodness of fit for all risk categories (Chi2: 15.42, p: 0.052). CONCLUSIONS: Our study provides a model that can be used in clinical practice to obtain an accurate dynamic assessment of the individual risk of HA-AKI stage 3 along the hospital stay period in non-critically ill patients.

Idiopathic pulmonary fibrosis cluster analysis highlights diagnostic delay and cardiovascular comorbidity association with outcome.

INTRODUCTION: Idiopathic pulmonary fibrosis (IPF) prognosis is heterogeneous despite antifibrotic treatment. Cluster analysis has proven to be a useful tool in identifying interstitial lung disease phenotypes, which has yet to be performed in IPF. The aim of this study is to identify phenotypes of IPF with different prognoses and requirements. METHODS: Observational retrospective study including 136 IPF patients receiving antifibrotic treatment between 2012 and 2018. Six patients were excluded due to follow-up in other centres. Cluster analysis of 30 variables was performed using approximate singular value-based tensor decomposition method and comparative statistical analysis. RESULTS: The cluster analysis identified three different groups of patients according to disease behaviour and clinical features, including mortality, lung transplant and progression-free survival time after 3-year follow-up. Cluster 1 (n=60) was significantly associated (p=0.02) with higher mortality. Diagnostic delay was the most relevant characteristic of this cluster, as 48% of patients had ≥2 years from first respiratory symptoms to antifibrotic treatment initiation. Cluster 2 (n=22) had the longest progression-free survival time and was correlated to subclinical patients evaluated in the context of incidental findings or familial screening. Cluster 3 (n=48) showed the highest percentage of disease progression without cluster 1 mortality, with metabolic syndrome and cardiovascular comorbidities as the main characteristics. CONCLUSION: This cluster analysis of IPF patients suggests that diagnostic and treatment delay are the most significant factors associated with mortality, while IPF progression was more related to metabolic syndrome and cardiovascular comorbidities.

Pipeline design to identify key features and classify the chemotherapy response on lung cancer patients using large-scale genetic data.

BACKGROUND: During the last decade, the interest to apply machine learning algorithms to genomic data has increased in many bioinformatics applications. Analyzing this type of data entails difficulties for managing high-dimensional data, class imbalance for knowledge extraction, identifying important features and classifying individuals. In this study, we propose a general framework to tackle these challenges with different machine learning algorithms and techniques. We apply the configuration of this framework on lung cancer patients, identifying genetic signatures for classifying response to drug treatment response. We intersect these relevant SNPs with the GWAS Catalog of the National Human Genome Research Institute and explore the Regulomedb, GTEx databases for functional analysis purposes. RESULTS: The machine learning based solution proposed in this study is a scalable and flexible alternative to the classical uni-variate regression approach to analyze large-scale data. From 36 experiments executed using the machine learning framework design, we obtain good classification performance from the top 5 models with the highest cross-validation score and the smallest standard deviation. One thousand two hundred twenty four SNPs corresponding to the key features from the top 20 models (cross validation F1 mean >= 0.65) were compared with the GWAS Catalog finding no intersection with genome-wide significant reported hits. From these, new genetic signatures in MAE, CEP104, PRKCZ and ADRB2 show relevant biological regulatory functionality related to lung physiology. CONCLUSIONS: We have defined a machine learning framework using data with an unbalanced large data-set of SNP-arrays and imputed genotyping data from a pharmacogenomics study in lung cancer patients subjected to first-line platinum-based treatment. This approach found genome signals with no genome-wide significance in the uni-variate regression approach (GWAS Catalog) that are valuable for classifying patients, only few of them with related biological function. The effect results of these variants can be explained by the recently proposed omnigenic model hypothesis, which states that complex traits can be influenced mostly by genes outside not only by the "core genes", mainly found by the genome-wide significant SNPs, but also by the rest of genes outside of the "core pathways" with apparent unrelated biological functionality.

Characterization of the CPAP-treated patient population in Catalonia.

There are different phenotypes of obstructive sleep apnoea (OSA), many of which have not been characterised. Identification of these different phenotypes is important in defining prognosis and guiding the therapeutic strategy. The aim of this study was to characterise the entire population of continuous positive airway pressure (CPAP)-treated patients in Catalonia and identify specific patient profiles using cluster analysis. A total of 72,217 CPAP-treated patients who contacted the Catalan Health System (CatSalut) during the years 2012 and 2013 were included. Six clusters were identified, classified as "Neoplastic patients" (Cluster 1, 10.4%), "Metabolic syndrome patients" (Cluster 2, 27.7%), "Asthmatic patients" (Cluster 3, 5.8%), "Musculoskeletal and joint disorder patients" (Cluster 4, 10.3%), "Patients with few comorbidities" (Cluster 5, 35.6%) and "Oldest and cardiac disease patients" (Cluster 6, 10.2%). Healthcare facility use and mortality were highest in patients from Cluster 1 and 6. Conversely, patients in Clusters 2 and 4 had low morbidity, mortality and healthcare resource use. Our findings highlight the heterogeneity of CPAP-treated patients, and suggest that OSA is associated with a different prognosis in the clusters identified. These results suggest the need for a comprehensive and individualised approach to CPAP treatment of OSA.

Rehabilitation Profiles of Older Adult Stroke Survivors Admitted to Intermediate Care Units: A Multi-Centre Study.

BACKGROUND: Stroke is a major cause of disability in older adults, but the evidence around post-acute treatment is limited and heterogeneous. We aimed to identify profiles of older adult stroke survivors admitted to intermediate care geriatric rehabilitation units. METHODS: We performed a cohort study, enrolling stroke survivors aged 65 years or older, admitted to 9 intermediate care units in Catalonia-Spain. To identify potential profiles, we included age, caregiver presence, comorbidity, pre-stroke and post-stroke disability, cognitive impairment and stroke severity in a cluster analysis. We also proposed a practical decision tree for patient's classification in clinical practice. We analyzed differences between profiles in functional improvement (Barthel index), relative functional gain (Montebello index), length of hospital stay (LOS), rehabilitation efficiency (functional improvement by LOS), and new institutionalization using multivariable regression models (for continuous and dichotomous outcomes). RESULTS: Among 384 patients (79.1±7.9 years, 50.8% women), we identified 3 complexity profiles: a) Lower Complexity with Caregiver (LCC), b) Moderate Complexity without Caregiver (MCN), and c) Higher Complexity with Caregiver (HCC). The decision tree showed high agreement with cluster analysis (96.6%). Using either linear (continuous outcomes) or logistic regression, both LCC and MCN, compared to HCC, showed statistically significant higher chances of functional improvement (OR = 4.68, 95%CI = 2.54-8.63 and OR = 3.0, 95%CI = 1.52-5.87, respectively, for Barthel index improvement ≥20), relative functional gain (OR = 4.41, 95%CI = 1.81-10.75 and OR = 3.45, 95%CI = 1.31-9.04, respectively, for top Vs lower tertiles), and rehabilitation efficiency (OR = 7.88, 95%CI = 3.65-17.03 and OR = 3.87, 95%CI = 1.69-8.89, respectively, for top Vs lower tertiles). In relation to LOS, MCN cluster had lower chance of shorter LOS than LCC (OR = 0.41, 95%CI = 0.23-0.75) and HCC (OR = 0.37, 95%CI = 0.19-0.73), for LOS lower Vs higher tertiles. CONCLUSION: Our data suggest that post-stroke rehabilitation profiles could be identified using routine assessment tools and showed differential recovery. If confirmed, these findings might help to develop tailored interventions to optimize recovery of older stroke patients.