Multimodal predictions of treatment outcome in major depression: A comparison of data-driven predictors with importance ratings by clinicians.

BACKGROUND: Reliable prediction models of treatment outcome in Major Depressive Disorder (MDD) are currently lacking in clinical practice. Data-driven outcome definitions, combining data from multiple modalities and incorporating clinician expertise might improve predictions. METHODS: We used unsupervised machine learning to identify treatment outcome classes in 1060 MDD inpatients. Subsequently, classification models were created on clinical and biological baseline information to predict treatment outcome classes and compared to the performance of two widely used classical outcome definitions. We also related the findings to results from an online survey that assessed which information clinicians use for outcome prognosis. RESULTS: Three and four outcome classes were identified by unsupervised learning. However, data-driven outcome classes did not result in more accurate prediction models. The best prediction model was targeting treatment response in its standard definition and reached accuracies of 63.9 % in the test sample, and 59.5 % and 56.9 % in the validation samples. Top predictors included sociodemographic and clinical characteristics, while biological parameters did not improve prediction accuracies. Treatment history, personality factors, prior course of the disorder, and patient attitude towards treatment were ranked as most important indicators by clinicians. LIMITATIONS: Missing data limited the power to identify biological predictors of treatment outcome from certain modalities. CONCLUSIONS: So far, the inclusion of available biological measures in addition to psychometric and clinical information did not improve predictive value of the models, which was overall low. Optimized biomarkers, stratified predictions and the inclusion of clinical expertise may improve future prediction models.

Automated High-Definition MRI Processing Routine Robustly Detects Longitudinal Morphometry Changes in Alzheimer's Disease Patients.

Longitudinal MRI studies are of increasing importance to document the time course of neurodegenerative diseases as well as neuroprotective effects of a drug candidate in clinical trials. However, manual longitudinal image assessments are time consuming and conventional assessment routines often deliver unsatisfying study outcomes. Here, we propose a profound analysis pipeline that consists of the following coordinated steps: (1) an automated and highly precise image processing stream including voxel and surface based morphometry using latest highly detailed brain atlases such as the HCP MMP 1.0 atlas with 360 cortical ROIs; (2) a profound statistical assessment using a multiplicative model of annual percent change (APC); and (3) a multiple testing correction adopted from genome-wide association studies that is optimally suited for longitudinal neuroimaging studies. We tested this analysis pipeline with 25 Alzheimer's disease patients against 25 age-matched cognitively normal subjects with a baseline and a 1-year follow-up conventional MRI scan from the ADNI-3 study. Even in this small cohort, we were able to report 22 significant measurements after multiple testing correction from SBM (including cortical volume, area and thickness) complementing only three statistically significant volume changes (left/right hippocampus and left amygdala) found by VBM. A 1-year decrease in brain morphometry coincided with an increasing clinical disability and cognitive decline in patients measured by MMSE, CDR GLOBAL, FAQ TOTAL and NPI TOTAL scores. This work shows that highly precise image assessments, APC computation and an adequate multiple testing correction can produce a significant study outcome even for small study sizes. With this, automated MRI processing is now available and reliable for routine use and clinical trials.

Clustering based on comorbidities in patients with chronic heart failure: an illustration of clinical diversity.

AIMS: It is increasingly recognized that the presence of comorbidities substantially contributes to the disease burden in patients with heart failure (HF). Several reports have suggested that clustering of comorbidities can lead to improved characterization of the disease phenotypes, which may influence management of the individual patient. Therefore, we aimed to cluster patients with HF based on medical comorbidities and their treatment and, subsequently, compare the clinical characteristics between these clusters. METHODS AND RESULTS: A total of 603 patients with HF entering an outpatient HF rehabilitation programme were included [median age 65 years (interquartile range 56-71), 57% ischaemic origin of cardiomyopathy, and left ventricular ejection fraction 35% (26-45)]. Exercise performance, daily life activities, disease-specific health status, coping styles, and personality traits were assessed. In addition, the presence of 12 clinically relevant comorbidities was recorded, based on targeted diagnostics combined with applicable pharmacotherapies. Self-organizing maps (SOMs; www.viscovery.net) were used to visualize clusters, generated by using a hybrid algorithm that applies the classical hierarchical cluster method of Ward on top of the SOM topology. Five clusters were identified: (1) a least comorbidities cluster; (2) a cachectic/implosive cluster; (3) a metabolic diabetes cluster; (4) a metabolic renal cluster; and (5) a psychologic cluster. Exercise performance, daily life activities, disease-specific health status, coping styles, personality traits, and number of comorbidities were significantly different between these clusters. CONCLUSIONS: Distinct combinations of comorbidities could be identified in patients with HF. Therapy may be tailored based on these clusters as next step towards precision medicine. The effect of such an approach needs to be prospectively tested.

Clustering of 27,525,663 Death Records from the United States Based on Health Conditions Associated with Death: An Example of big Health Data Exploration.

BACKGROUND: Insight into health conditions associated with death can inform healthcare policy. We aimed to cluster 27,525,663 deceased people based on the health conditions associated with death to study the associations between the health condition clusters, demographics, the recorded underlying cause and place of death. METHODS: Data from all deaths in the United States registered between 2006 and 2016 from the National Vital Statistics System of the National Center for Health Statistics were analyzed. A self-organizing map (SOM) was used to create an ordered representation of the mortality data. RESULTS: 16 clusters based on the health conditions associated with death were found showing significant differences in socio-demographics, place, and cause of death. Most people died at old age (73.1 (18.0) years) and had multiple health conditions. Chronic ischemic heart disease was the main cause of death. Most people died in the hospital or at home. CONCLUSIONS: The prevalence of multiple health conditions at death requires a shift from disease-oriented towards person-centred palliative care at the end of life, including timely advance care planning. Understanding differences in population-based patterns and clusters of end-of-life experiences is an important step toward developing a strategy for implementing population-based palliative care.