Modeling of covariates predicting renal allograft failure within six months after transplantation / Modelagem das covariáveis preditivas para falência do enxerto renal até seis meses do transplante

INTRODUCTION: Kidney transplantation (KT) is the renal replacement therapy (RRT) of choice for patients with chronic kidney disease (CKD). However, not every KT is successful and some patients persist on RRT. OBJECTIVE: To model a logistic regression with pre- and post-KT risk covariates capable of predicting secondary allograft dysfunction in need of RRT or reaching stage V of CKD until the first six months post-KT. METHODS: Cohort with KT recipients from Northeastern Brazil. Medical records of KT performed between 2011-2018 were analyzed. KT-recipients with insufficient data or who abandoned follow-up were excluded. The covariables analyzed were: demographic; infectious; pre- and post-KT comorbidities; panel reactive-antibodies; number of HLA mismatches; acute rejection episodes mediated by T-cell (ACR) or antibodies (AAR) six months after KT; and laboratory tests six months after KT. RESULTS: Covariates with higher risk for the analyzed outcomes six months after KT were: elderly KT recipients (OR:1.41; CI95%:1.01-1.99), time between onset of RRT and KT (ΔT-RRT&KT)>10years (OR:3.54; CI95%:1.27-9.87), diabetes mellitus (DM) pre-KT (OR:3.35; CI95%:1.51-7.46), pyelonephritis (OR:2.45; CI95%:1.24-4.84), polyomavirus nephropathy (OR:4.99; CI95%:1.87-13.3), AAS (OR:4.82; CI95%:1.35-17.2), 24h-proteinuria ≥300mg/24h (OR:5.05; CI95%:2.00-12.7) and serum calcium (Ca) <8.5mg/dL (OR:4.72; CI95%:2.00-11.1). The multivariate model presented an accuracy of 88.1% and the mean variance inflation factor is 1.81. CONCLUSION: Elderly-recipients, ΔT-RRT&KT>10 years, pre-KT DM, and post-KT aggressions until six months (pyelonephritis, polyomavirus nephropathy, ABMR, 24h-proteinuria≥300mg/24h, and Ca<8.5mg/dL) are associated with high predictive power for secondary allograft dysfunction in need of RRT or reaching CKD stage V until the first six months post-KT.

INTRODUÇÃO: Transplante renal (TR) é a terapia renal substitutiva (TRS) de escolha para pacientes com doença renal crônica (DRC). Entretanto, nem todo TR é bem-sucedido e alguns pacientes persistem em TRS. OBJETIVO: Modelar uma regressão logística com covariáveis de risco pré e pós-TR preditora da disfunção secundária do aloenxerto com necessidade de TRS ou alcance ao estágio V da DRC até os primeiros seis meses pós-TR. MÉTODOS: Coorte com receptores transplantados realizado em hospital no Nordeste brasileiro. Analisou-se registros médicos dos TR realizados entre 2011-2018. Receptores com dados insuficientes ou que abandonaram seguimento foram excluídos. Foram analisadas covariáveis: demográficas; infecciosas; comorbidades pré e pós-TR; painel de reatividade; incompatibilidades de HLA; episódios de rejeições agudas mediadas por células-T ou por anticorpos; exames laboratoriais seis meses pós-TR. RESULTADOS: Receptores idosos (OR:1,41; IC95%:1,01-1,99), tempo entre início da TRS e TR (∆T-TRS&TR)>10 anos (OR:3,54; IC95%:1,27-9,87), diabetes mellitus (DM) pré-TR (OR:3,35; IC95%:1,51-7,46), pielonefrite (OR:2,45; IC95%:1,24-4,84), nefropatia por poliomavírus (OR:4,99; IC95%:1,87-13,3), RAMA (OR:4,82; IC95%:1,35-17,2), proteinúria de 24h (Pt24h) ≥300mg/24h (OR:5,05; IC95%:2,00-12,7) e cálcio sérico (Ca)<8,5mg/dL (OR:4,72; IC95%:2,00-11,1) foram identificadas como covariáveis de maior risco para os desfechos analisados até seis meses pós-TR. O modelo multivariado apresentou acurácia de 88,1% e fator de inflação da variância médio de 1,81. CONCLUSÃO: Receptores idosos, ∆T-TRS&TR>10anos, DM pré-TR e agressões até seis meses pós-TR (pielonefrite, nefropatia por poliomavírus, RAMA, Pt24h≥300mg/24h e Ca<8,5mg/dL), apresentam alto poder preditivo para disfunção secundária do aloenxerto com necessidade de TRS ou alcance ao estágio V da DRC até os primeiros seis meses pós-TR.

Energy Consumption Forecasting for Smart Meters Using Extreme Learning Machine Ensemble.

The employment of smart meters for energy consumption monitoring is essential for planning and management of power generation systems. In this context, forecasting energy consumption is a valuable asset for decision making, since it can improve the predictability of forthcoming demand to energy providers. In this work, we propose a data-driven ensemble that combines five single well-known models in the forecasting literature: a statistical linear autoregressive model and four artificial neural networks: (radial basis function, multilayer perceptron, extreme learning machines, and echo state networks). The proposed ensemble employs extreme learning machines as the combination model due to its simplicity, learning speed, and greater ability of generalization in comparison to other artificial neural networks. The experiments were conducted on real consumption data collected from a smart meter in a one-step-ahead forecasting scenario. The results using five different performance metrics demonstrate that our solution outperforms other statistical, machine learning, and ensembles models proposed in the literature.

Logistic Regression Model in a Machine Learning Application to Predict Elderly Kidney Transplant Recipients with Worse Renal Function One Year after Kidney Transplant: Elderly KTbot.

BACKGROUND: Renal replacement therapy (RRT) is a public health problem worldwide. Kidney transplantation (KT) is the best treatment for elderly patients' longevity and quality of life. OBJECTIVES: The primary endpoint was to compare elderly versus younger KT recipients by analyzing the risk covariables involved in worsening renal function, proteinuria, graft loss, and death one year after KT. The secondary endpoint was to create a robot based on logistic regression capable of predicting the likelihood that elderly recipients will develop worse renal function one year after KT. METHOD: Unicentric retrospective analysis of a cohort was performed with individuals aged ≥60 and <60 years old. We analysed medical records of KT recipients from January to December 2017, with a follow-up time of one year after KT. We used multivariable logistic regression to estimate odds ratios for elderly vs younger recipients, controlled for demographic, clinical, laboratory, data pre- and post-KT, and death. RESULTS: 18 elderly and 100 younger KT recipients were included. Pretransplant immune variables were similar between two groups. No significant differences (P > 0.05) between groups were observed after KT on laboratory data means and for the prevalences of diabetes mellitus, hypertension, acute rejection, cytomegalovirus, polyomavirus, and urinary infections. One year after KT, the creatinine clearance was higher (P = 0.006) in youngers (70.9 ± 25.2 mL/min/1.73 m2) versus elderlies (53.3 ± 21.1 mL/min/1.73 m2). There was no difference in death outcome comparison. Multivariable analysis among covariables predisposing chronic kidney disease epidemiology collaboration (CKD-EPI) equation <60 mL/min/1.73 m2 presented a statistical significance for age ≥60 years (P = 0.01) and reduction in serum haemoglobin (P = 0.03). The model presented goodness-fit in the evaluation of artificial intelligence metrics (precision: 90%; sensitivity: 71%; and F 1 score: 0.79). CONCLUSION: Renal function in elderly KT recipients was lower than in younger KT recipients. However, patients aged ≥60 years maintained enough renal function to remain off dialysis. Moreover, a learning machine application built a robot (Elderly KTbot) to predict in the elderly populations the likelihood of worse renal function one year after KT.