Clustering-based spatial analysis (CluSA) framework through graph neural network for chronic kidney disease prediction using histopathology images.

Machine learning applied to digital pathology has been increasingly used to assess kidney function and diagnose the underlying cause of chronic kidney disease (CKD). We developed a novel computational framework, clustering-based spatial analysis (CluSA), that leverages unsupervised learning to learn spatial relationships between local visual patterns in kidney tissue. This framework minimizes the need for time-consuming and impractical expert annotations. 107,471 histopathology images obtained from 172 biopsy cores were used in the clustering and in the deep learning model. To incorporate spatial information over the clustered image patterns on the biopsy sample, we spatially encoded clustered patterns with colors and performed spatial analysis through graph neural network. A random forest classifier with various groups of features were used to predict CKD. For predicting eGFR at the biopsy, we achieved a sensitivity of 0.97, specificity of 0.90, and accuracy of 0.95. AUC was 0.96. For predicting eGFR changes in one-year, we achieved a sensitivity of 0.83, specificity of 0.85, and accuracy of 0.84. AUC was 0.85. This study presents the first spatial analysis based on unsupervised machine learning algorithms. Without expert annotation, CluSA framework can not only accurately classify and predict the degree of kidney function at the biopsy and in one year, but also identify novel predictors of kidney function and renal prognosis.

Unsupervised machine learning for identifying important visual features through bag-of-words using histopathology data from chronic kidney disease.

Pathologists use visual classification to assess patient kidney biopsy samples when diagnosing the underlying cause of kidney disease. However, the assessment is qualitative, or semi-quantitative at best, and reproducibility is challenging. To discover previously unknown features which predict patient outcomes and overcome substantial interobserver variability, we developed an unsupervised bag-of-words model. Our study applied to the C-PROBE cohort of patients with chronic kidney disease (CKD). 107,471 histopathology images were obtained from 161 biopsy cores and identified important morphological features in biopsy tissue that are highly predictive of the presence of CKD both at the time of biopsy and in one year. To evaluate the performance of our model, we estimated the AUC and its 95% confidence interval. We show that this method is reliable and reproducible and can achieve 0.93 AUC at predicting glomerular filtration rate at the time of biopsy as well as predicting a loss of function at one year. Additionally, with this method, we ranked the identified morphological features according to their importance as diagnostic markers for chronic kidney disease. In this study, we have demonstrated the feasibility of using an unsupervised machine learning method without human input in order to predict the level of kidney function in CKD. The results from our study indicate that the visual dictionary, or visual image pattern, obtained from unsupervised machine learning can predict outcomes using machine-derived values that correspond to both known and unknown clinically relevant features.

Consent for Genetic Biobanking in a Diverse Multisite CKD Cohort.

INTRODUCTION: The goal of this study was to examine patterns in the likelihood of consent to genetic research among participants in a prospective kidney disease cohort and biobank, and to determine demographic, clinical, and socioeconomic factors linked to consent for ongoing and future genetic research. METHODS: The Clinical Phenotyping Resource and Biobank Core (C-PROBE) enrolled 1628 adult and pediatric patients with chronic kidney disease from 2009 to 2017 across 7 sites in the United States. Participants were asked at annual study visits for consent to provide DNA samples for future genetic studies. We compared characteristics of participants by initial consent outcome and consent status at their most recent study visit. RESULTS: Of the C-PROBE participants, 96% consented to genetic studies at their initial study visit. Although African Americans were slightly less likely to consent at baseline (93% vs. 97%, odds ratio = 0.3, P < 0.02), there were no significant racial or ethnic differences with longitudinal participation. Also, pediatric and adult genetic consent rates were equivalent. The major persistent differences in the likelihood of consent were based on enrollment site, which ranged from 85% to 100% (P < 0.0001). CONCLUSION: Overall, genetic consent rates for kidney research within the C-PROBE cohort were high. However, differences in consent rates over time and by recruitment site highlight the complexity of genetic consent for biobanking, and potential limitations for generalizability of observations.

An Outcomes-Based Definition of Proteinuria Remission in Focal Segmental Glomerulosclerosis.

BACKGROUND AND OBJECTIVES: Proteinuria is used as an indicator of FSGS disease activity, but its use as a clinical trial end point is not universally accepted. The goal of this study was to refine proteinuria definitions associated with long-term kidney survival. DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS: Data on 466 patients with primary FSGS with proteinuria (urine protein-to-creatinine ratio >1 g/g) were analyzed from five independent cohorts. Proteinuria by months 1, 4, and 8 after study baseline was categorized by conventional definitions of complete (<0.3 g/g) and partial remission (<3.5 g/g and 50% reduction in proteinuria). Novel remission definitions were explored using receiver operating curves. Kaplan-Meier methods were used to estimate the associations of proteinuria with progression to ESRD or a 50% loss in kidney function. Propensity score-adjusted Cox proportional hazards models were used to adjust for baseline proteinuria, eGFR, and therapy. RESULTS: In the initial derivation cohort, conventional partial remission was not associated with kidney survival. A novel definition of partial remission (40% proteinuria reduction and proteinuria<1.5 g/g) on the basis of receiver operating curve analyses of 89 patients was identified (Sensitivity=0.70; Specificity=0.77). In the validation cohort analyses, complete remission was associated with better prognosis (6 out of 41 patients progressed to kidney failure; 6.6 per 100 patient-years) as was the novel partial remission (13 out of 71 progressed; 8.5 per 100 patient-years), compared with those with no response (51 out of 116 progressed; 20.1 per 100 patient-years). Conventional partial remission at month 8, but not month 4, was also associated with better response (19 out of 85 patients progressed; risk=10.4 per 100 patient-years). Propensity score-adjusted analyses showed the novel partial remission was associated with less progression at months 4 and 8 (month 4: hazard ratio, 0.50; P=0.01; month 8: hazard ratio, 0.30; P=0.002). CONCLUSIONS: Reaching either a complete or partial remission using a novel or conventional definition was associated with better long-term outcomes in patients with FSGS. PODCAST: This article contains a podcast at https://www.asn-online.org/media/podcast/CJASN/2018_02_20_CJASNPodcast_18_3_T.mp3.

Growth Differentiation Factor-15 and Risk of CKD Progression.

Growth differentiation factor-15 (GDF-15) is a member of the TGF-ß cytokine superfamily that is widely expressed and may be induced in response to tissue injury. Elevations in GDF-15 may identify a novel pathway involved in loss of kidney function among patients with CKD. Among participants in the Clinical Phenotyping and Resource Biobank (C-PROBE) study and the Seattle Kidney Study (SKS), we tested whether kidney tissue expression of GDF15 mRNA correlates with circulating levels of GDF-15 and whether elevations in circulating GDF-15 are associated with decline in kidney function. In matching samples of 24 patients with CKD from the C-PROBE study, circulating GDF-15 levels significantly correlated with intrarenal GDF15 transcript levels (r=0.54, P=0.01). Among the 224 C-PROBE and 297 SKS participants, 72 (32.1%) and 94 (32.0%) patients, respectively, reached a composite end point of 30% decline in eGFR or progression to ESRD over a median of 1.8 and 2.0 years of follow up, respectively. In multivariable models, after adjusting for potential confounders, every doubling of GDF-15 level associated with a 72% higher (95% confidence interval, 1.21 to 4.45; P=0.003) and 65% higher (95% confidence interval, 1.08 to 2.50; P=0.02) risk of progression of kidney disease in C-PROBE and SKS participants, respectively. These results show that circulating GDF-15 levels strongly correlated with intrarenal expression of GDF15 and significantly associated with increased risk of CKD progression in two independent cohorts. Circulating GDF-15 may be a marker for intrarenal GDF15-related signaling pathways associated with CKD and CKD progression.

Complete Remission in the Nephrotic Syndrome Study Network.

BACKGROUND AND OBJECTIVES: This analysis from the Nephrotic Syndrome Study Network (NEPTUNE) assessed the phenotypic and pathology characteristics of proteinuric patients undergoing kidney biopsy and defined the frequency and factors associated with complete proteinuria remission (CRever). DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS: We enrolled adults and children with proteinuria ≥0.5 g/d at the time of first clinically indicated renal biopsy at 21 sites in North America from April 2010 to June 2014 into a prospective cohort study. NEPTUNE central pathologists assigned participants to minimal-change disease (MCD), FSGS, membranous nephropathy, or other glomerulopathy cohorts. Outcome measures for this analysis were (1) CRever with urine protein-to-creatinine ratio (UPC) <0.3 g/g with preserved native kidney function and (2) ESRD. Continuous variables are reported as median and interquartile range (IQR; 25th, 75th percentile). Cox proportional hazards modeling was used to assess factors associated with CRever. RESULTS: We enrolled 441 patients: 116 (27%) had MCD, 142 (32%) had FSGS, 66 (15%) had membranous nephropathy, and 117 (27%) had other glomerulopathy. The baseline UPC was 4.1 g/g (IQR, 1.9, 7.7) and the eGFR was 81 ml/min per 1.73 m(2) (IQR, 50, 105). Median duration of observation was 19 months (IQR, 11, 30). CRever occurred in 46% of patients, and 4.6% progressed to ESRD. Multivariate analysis demonstrated that higher prebiopsy proteinuria (hazard ratio, 0.3; 95% confidence interval, 0.2 to 0.5) and pathology diagnosis (FSGS versus MCD; hazard ratio, 0.2; 95% confidence interval, 0.1 to 0.5) were inversely associated with CRever. The effect of immunosuppressive therapy on remission varied by pathology diagnosis. CONCLUSIONS: In NEPTUNE, the high frequency of other pathology in proteinuric patients affirms the value of the diagnostic kidney biopsy. Clinical factors, including level of proteinuria before biopsy, pathology diagnosis, and immunosuppression, are associated with complete remission.

The relationship between immediate preoperative serum 25-hydroxy-vitamin D3 levels and cardiac function, dysglycemia, length of stay, and 30-d readmissions in cardiac surgery patients.

OBJECTIVES: Vitamin D has pleiotropic effects on cardiac, renal, and endocrine diseases like diabetes mellitus and deficiency has been correlated with increased Intensive Care Unit (ICU) morbidity and mortality. We studied the relationship between preoperative Vitamin D levels and several short-term endpoints including cardiovascular events, glucose levels, ICU, and hospital length of stay. METHODS: Standard demographic data were obtained. Blood samples were drawn for 25-hydroxy-vitamin D3 (Vit D) levels at baseline (just before induction of anesthesia) and on postoperative day (POD #1). The number of inotropes used on POD # 0, 1, and 2 was recorded as well as the Cardiac Index (CI). Baseline glucose, Blood Urea Nitrogen and Creatinine (Cr) levels were obtained and repeated on POD # 1 & 2. Other variables studied are number of days of ICU and hospital stay. RESULTS: Of the 64 patients included in the cohort, 3 were excluded because of inadequate data. 69% had Vit D levels <20 ng/mL and 31% had levels ≥20 ng/mL. More than 90% of the cohort had a significant decrease in POD # 1 Vit D levels (P < 0.001). Age, sex, race, and body mass index did not predict the preoperative Vit D levels; however, the timing of surgery was associated with preoperative Vit D levels, lowest in subjects who had surgery performed during winter. Preoperative Vit D levels had no effect on postoperative glycemic control, cardiac index, or composite outcome-arrhythmias, respiratory failure, or prolonged inotropic support. On regression analysis, preoperative Vit D levels did show a significant effect on ICU and hospital length of stay in this cohort. CONCLUSIONS: The low levels in this study truly represent the Vit D status as they were obtained before any intervention, including surgery or fluid administration. Vit D levels decreased rapidly after surgery and hence future studies on Vit D may need to focus on premorbid levels obtained at the time of initial presentation and not those obtained after resuscitation or ICU admission. In contrast to epidemiologic reports, we found no association between low Vit D levels and postoperative cardiovascular events. However, low Vit D levels did affect the ICU and hospital length of stay in patients who were undergoing cardiac surgery. This is an important finding especially when many institutions and regulatory agencies are investigating novel therapies and processes to reduce the length of hospitalization. More studies are required to investigate the effect on hospital length of stay of early preadmission or preoperative Vit D supplementation before elective surgery.

Hypertension awareness, treatment, and control in adults with CKD: results from the Chronic Renal Insufficiency Cohort (CRIC) Study.

BACKGROUND: A low rate of blood pressure control has been reported in patients with chronic kidney disease (CKD). These data were derived from population-based samples with a low rate of CKD awareness. STUDY DESIGN: Cross-sectional. SETTING & PARTICIPANTS: Data from the baseline visit of the Chronic Renal Insufficiency Cohort (CRIC) Study (n = 3,612) were analyzed. Participants with an estimated glomerular filtration rate of 20-70 mL/min/1.73 m(2) were identified from physician offices and review of laboratory databases. OUTCOMES: Prevalence and awareness of hypertension, treatment patterns, control rates, and factors associated with hypertension control. MEASUREMENTS: Following a standardized protocol, blood pressure was measured 3 times by trained staff, and hypertension was defined as systolic blood pressure > or =140 mm Hg and/or diastolic blood pressure > or =90 mm Hg and/or self-reported antihypertensive medication use. Patients' awareness and treatment of hypertension were defined using self-report, and 2 levels of hypertension control were evaluated: systolic/diastolic blood pressure <140/90 and <130/80 mm Hg. RESULTS: The prevalence of hypertension was 85.7%, and 98.9% of CRIC participants were aware of this diagnosis and 98.3% were treated with medications, whereas 67.1% and 46.1% had hypertension controlled to <140/90 and <130/80 mm Hg, respectively. Of CRIC participants with hypertension, 15%, 25%, 26%, and 32% were using 1, 2, 3, and > or =4 antihypertensive medications, respectively. After multivariable adjustment, older patients, blacks, and those with higher urinary albumin excretion were less likely, whereas participants using angiotensin-converting enzyme inhibitors and angiotensin receptor blockers were more likely to have controlled their hypertension to <140/90 and <130/80 mm Hg. LIMITATIONS: Data were derived from a single study visit. CONCLUSIONS: Despite almost universal hypertension awareness and treatment in this cohort of patients with CKD, rates of hypertension control were suboptimal.

Aortic PWV in chronic kidney disease: a CRIC ancillary study.

BACKGROUND: Aortic pulse wave velocity (PWV) is a measure of arterial stiffness and has proved useful in predicting cardiovascular morbidity and mortality in several populations of patients, including the healthy elderly, hypertensives and those with end-stage renal disease receiving hemodialysis. Little data exist characterizing aortic stiffness in patients with chronic kidney disease (CKD) who are not receiving dialysis, and in particular the effect of reduced kidney function on aortic PWV. METHODS: We performed measurements of aortic PWV in a cross-sectional cohort of participants enrolled in the Chronic Renal Insufficiency Cohort (CRIC) study to determine factors which predict increased aortic PWV in CKD. RESULTS: PWV measurements were obtained in 2,564 participants. The tertiles of aortic PWV (adjusted for waist circumference) were <7.7 m/s, 7.7-10.2 m/s, and >10.2 m/s with an overall mean (+/- s.d.) value of 9.48 +/- 3.03 m/s (95% confidence interval = 9.35-9.61 m/s). Multivariable regression identified significant independent positive associations of age, blood glucose concentrations, race, waist circumference, mean arterial blood pressure, gender, and presence of diabetes with aortic PWV and a significant negative association with the level of kidney function. CONCLUSIONS: The large size of this unique cohort, and the targeted enrollment of CKD participants provides an ideal situation to study the role of reduced kidney function as a determinant of arterial stiffness. Arterial stiffness may be a significant component of the enhanced cardiovascular risk associated with kidney failure.

Aortic pulse pressure is associated with carotid IMT in chronic kidney disease: report from Chronic Renal Insufficiency Cohort.

BACKGROUND: Patients with chronic kidney disease (CKD) have a disproportionate risk of cardiovascular disease. This study was designed to assess the association between two noninvasive measures of cardiovascular risk, pulse wave analysis (PWA), and carotid intima-media thickness (IMT), in a cohort of CKD patients enrolled in the Chronic Renal Insufficiency Cohort (CRIC) study. METHODS: Three hundred and sixty-seven subjects with CKD enrolled in the CRIC study at the University of Pennsylvania site (mean age 59.9 years, blood pressure 129/74 mm Hg, estimated glomerular filtration rate 48 ml/min/1.73 m2, IMT 0.8 mm) had both carotid IMT and PWA measurements. Carotid ultrasound was also used to determine the presence of plaque. PWA was used to determine augmentation index (AI), amplification ratio (AMPR), aortic pulse pressure (C_PP), and central aortic systolic pressure (C_SP). RESULTS: IMT was significantly associated with all PWA-derived measures. However, on multivariable linear regression analysis, only AMPR (regression coefficient -0.072, P = 0.006), C_PP (regression coefficient 0.0025, P < 0.001), and C_SP (regression coefficient 0.0017, P < 0.001) remained significantly associated with IMT. The prevalence of carotid plaque in the cohort was 59%. Of the PWA-derived measures, only C_PP was significantly associated with the presence of carotid plaque (P < 0.001). CONCLUSIONS: PWA-derived measures are associated with carotid IMT and plaque in the CKD. Of these measures, C_PP was most associated with carotid IMT and plaque.

The impact of evidence on physicians' inpatient treatment decisions.

OBJECTIVE: Previous studies have shown that most medical inpatients receive treatment supported by strong evidence (evidence-based treatment), but they have not assessed whether and how physicians actually use evidence when making their treatment decisions. We investigated whether physicians would change inpatient treatment if presented with the results of a literature search. DESIGN: Before-after study. SETTING: Large public teaching hospital. PARTICIPANTS: Random sample of 146 inpatients cared for by 33 internal medicine attending physicians. INTERVENTIONS: After physicians committed to a specific diagnosis and treatment plan, investigators performed standardized literature searches and provided the search results to the attending physicians. MEASUREMENTS AND MAIN RESULTS: The primary study outcome was the number of patients whose attending physicians would change treatment due to the literature searches. These changes were evaluated by blinded peer review. A secondary outcome was the proportion of patients who received evidence-based treatment before and after the literature searches. Attending physicians changed treatment for 23 (18%) of 130 eligible patients (95% confidence interval, 12% to 24%) as a result of the literature searches. Overall, 86% of patients (112 of 130) received evidence-based treatments before the searches and 87% (113 of 130) after the searches. Changes were not related to whether patients were receiving evidence-based treatment before the search (P =.6). Panels of peer reviewers judged the quality of patient care as improved or maintained for 18 (78%) of the 23 patients with treatment changes. CONCLUSIONS: Searching the literature could improve the treatment of many medical inpatients, including those already receiving evidence-based treatment.