Machine learning for hospital readmission prediction in pediatric population.

BACKGROUND AND OBJECTIVE: Pediatric readmissions are a burden on patients, families, and the healthcare system. In order to identify patients at higher readmission risk, more accurate techniques, as machine learning (ML), could be a good strategy to expand the knowledge in this area. The aim of this study was to develop predictive models capable of identifying children and adolescents at high risk of potentially avoidable 30-day readmission using ML. METHODS: Retrospective cohort study was carried out with 9,080 patients under 18 years old admitted to a tertiary university hospital. Demographic, clinical, and biochemical data were collected from electronic databases. We randomly divided the dataset into training (75 %) and testing (25 %), applied downsampling, repeated cross-validation with five folds and ten repetitions, and the hyperparameter was optimized of each technique using a grid search via racing with ANOVA models. We applied six ML classification algorithms to build the predictive models, including classification and regression tree (CART), random forest (RF), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), decision tree and logistic regression (LR). The area under the receiver operating curve (AUC), sensitivity, specificity, Youden's J-index and accuracy were used to evaluate the performance of each model. RESULTS: The avoidable 30-day hospital readmissions rate was 9.5 %. Some algorithms presented similar AUC, both in the dataset training and in the dataset testing, such as XGBoost, RF, GBM and CART. Considering the Youden's J-index, the algorithm that presented the best index was XGBoost with bagging imputation, with AUC of 0.814 (J-index of 0.484). Cancer diagnosis, age, red blood cells, leukocytes, red cell distribution width and sodium levels, elective admission, and multimorbidity were the most important characteristics to classify between readmission and non-readmission groups. CONCLUSION: Machine learning approaches, especially XGBoost, can predict potentially avoidable 30-day pediatric hospital readmission into tertiary assistance. If implemented in the computer hospital system, our model can help in the early and more accurate identification of patients at readmission risk, targeting health strategic interventions.

Nutritional Status and Growth Deficit in Children and Adolescents with Cancer at Different Moments of Treatment.

BACKGROUND: Nutrition deficits are common in children and adolescents undergoing cancer treatment and can contribute to a worse prognosis. There are scarce studies regarding this context considering different moments of treatment. The aim of this study was to evaluate the association between moment of treatment and nutritional status in children and adolescents with cancer. METHODS: A retrospective study was performed from January 2013 to December 2015, including data from all clinical records of patients under 18 years old with cancer. Clinical, nutritional support and anthropometric data were collected at four moments of treatment from cancer diagnosis: diagnosis (t0), 3 mo, (t1), 6 mo, (t2) and 1 year (t3). In addition, nutritional indicators were evaluated. Generalized Estimating Equation models were performed to analyze changes on anthropometric indices throughout four moments of treatment. RESULTS: The sample comprised 73 patients and frequency of nutritional deficits ranged from 13.0% to 18.6%. All nutritional indicators decreased at t1, showed a modest recovery at t2 and a stronger recovery at t3 (p < 0.001). Growth was also impacted during treatment, mainly on patients under 2 years in the first three months of treatment. CONCLUSIONS: Moment of treatment was associated with growth deficit and decreased percentiles in development indicators.

Serum Albumin is Independent Predictor of Hospital Mortality in Patients with Cancer / A Albumina Sérica é Preditor Independente de Mortalidade Hospitalar em Pacientes com Câncer / La Albúmina Sérica es un Predictor Independiente de Mortalidad Hospitalaria en Pacientes con Cáncer

Introduction: Cancer is one of the leading causes of morbidity and mortality worldwide. There are few studies showing adjusted models with other predictors of mortality by a conceptual model perspective. Objective: The objective of this study was to verify the prediction of albumin and Prognostic Nutritional Index (PNI) with in-hospital mortality in cancer patients. Method: Retrospective study was performed from 2014 to 2016 with 262 cancer patients (gastrointestinal tract, male genital organs, breast, metastasis, urinary tract, head and neck and others). Demographic data, blood counts, C-reactive protein, albumin, and haematological indexes (Prognosis nutritional index - PNI, Neutrophils to lymphocytes ratio - NLR, Monocytes lymphocytes ratio - MLR, Platelets to lymphocytes ratio - PLR and Platelets to albumin ratio - PAR), nutritional diagnoses and hospital outcomes (discharge or death) were collected. The cumulative probability of death was calculated by Kaplan-Meier curves, and survival analyses were performed using the Cox proportional hazards model. Results: The frequency of death among the study patients was 10.7% (28). Among the patients who died, 99.2% (26) presented some degree of malnutrition (p=0.004). In the multivariate analysis, serum albumin (<3 g/dL) was independently associated with in-hospital mortality (HR=3.43, 95% CI 1.11-10.63). On the other hand, the PNI was not associated with in-hospital mortality. Conclusion: Serum albumin levels during hospitalization were predictors of in-hospital mortality in the population evaluated. These results suggest that the serum levels of this protein can be used in clinical practice, adding prognostic information in patients with cancer

Introducción: El cáncer es una de las principales causas de morbilidad y mortalidad en todo el mundo. Hay pocos estudios que muestren modelos ajustados con otros predictores de mortalidad desde una perspectiva de modelo conceptual. Objetivo: El objetivo de este estudio fue verificar la predicción de la albúmina y el Índice Nutricional Pronóstico (IPN) con la mortalidad hospitalaria en pacientes con cáncer (tracto gastrointestinal, órganos genitales masculinos, mama, metástasis, tracto urinario, cabeza y cuello y otros). Método: Se realizó un estudio retrospectivo de 2014 a 2016 con 262 pacientes con cáncer. Se recogieron datos demográficos, hemogramas, proteína C reactiva, albúmina y índices hematológicos (Índice de Pronóstico Nutricional - IPN, proporción neutrófilos/linfocitos - NLR, proporción monocitos/linfocitos - MLR, proporción plaquetas/linfocitos - PLR y proporción plaquetas/albúmina - PAR), diagnósticos nutricionales y resultados hospitalarios (alta o muerte). La probabilidad acumulada de muerte se calculó mediante curvas de Kaplan-Meier y se realizaron análisis supervivencia utilizando el modelo de riesgos proporcionales de Cox. Resultados: La frecuencia de muerte entre los pacientes del estudio fue del 10,7% (28). Entre los pacientes fallecidos, el 99,2% (26) presentaba algún grado de desnutrición (p=0,004). En el análisis multivariado, la albúmina sérica (<3 g/dL) se asoció de forma independiente con la mortalidad hospitalaria (HR=3,43, IC 95% 1,11-10,63). Por otro lado, el IPN no se asoció con mortalidad intrahospitalaria. Conclusión: Los niveles de albúmina sérica durante la hospitalización fueron predictores de mortalidad intrahospitalaria en la población evaluada. Nuestros resultados sugieren que los niveles séricos de esta proteína se pueden utilizar en la práctica clínica, agregando información de pronóstico en pacientes con cáncer.

Introdução: O câncer é uma das principais causas de morbidade e mortalidade em todo o mundo. Existem poucos estudos mostrando modelos ajustados com outros preditores de mortalidade por uma perspectiva de modelo conceitual. Objetivo: Verificar a predição de albumina e do Índice Nutricional Prognóstico (IPN) com mortalidade intra-hospitalar em pacientes com câncer. Método: Estudo retrospectivo realizado de 2014 a 2016 com 262 pacientes com câncer (trato gastrointestinal, órgãos genitais masculinos, mama, metástases, trato urinário, cabeça e pescoço e outros). Foram coletados dados demográficos, hemograma, proteína C reativa, albumina e índices hematológicos (índice de prognóstico nutricional - IPN; relação neutrófilo por linfócitos - RNL; relação monócitos por linfócitos - RML; relação plaquetas por linfócitos - RPL; e relação plaquetas por albumina ­ RPA), diagnósticos nutricionais e desfechos hospitalares (alta ou óbito). A probabilidade cumulativa de morte foi calculada pelas curvas de Kaplan-Meier e as análises de sobrevivência realizadas usando o modelo de risco proporcional de Cox. Resultados: A frequência de óbito entre os pacientes do estudo foi de 10,7% (28). Entre os pacientes que morreram, 99,2% (26) apresentavam algum grau de desnutrição (p=0,004). Na análise multivariada, a albumina sérica (<3 g/dL) associou-se de forma independente à mortalidade hospitalar (HR=3,43, IC95% 1,11-10,63). Por outro lado, o IPN não foi associado com mortalidade intra-hospitalar. Conclusão: Os níveis de albumina sérica durante a internação foram preditores de mortalidade intra-hospitalar na população avaliada. Esses resultados sugerem que os níveis séricos dessa proteína podem ser utilizados na prática clínica, agregando informações prognósticas em pacientes com câncer

High red blood cell distribution width is associated with a risk of short-term mortality in hospitalized surgical, but not clinical patients.

BACKGROUND & AIMS: The distribution width of red blood cells (RDW) is a known factor risk for mortality. However, the association between high RDW and short-term mortality in surgical patients is poorly understood. The aim of this study was to evaluate the association of high RDW with all-cause in-hospital mortality, in surgical and non-surgical patients. METHODS: A retrospective study was performed with patients aged 18 years or older, hospitalized in Clinical Medical and Surgery wards, using adjustments based on a conceptual model. Cox regression was used to determine the independent predictors of in-hospital mortality. The RDW cutoff value was 13.6%. RESULTS: Of the 2923 patients, 46.1% were over 60 years old, 58.7% were male and 4.7% died. The area under the ROC curve was 0.677 (CI 95%: 0.619-0.712). RDW was associated with an adjusted risk for all-cause in-hospital mortality in surgical (HR 1.17 - CI 95%: 1.03-1.32), but not in clinical patients. For every 1% increase in RDW, the risk of all-cause hospital death in surgical patients increased by 17%. RDW ≥13.6% was associated with an adjusted risk of all cause in-hospital mortality in surgical (HR 2.65 - 95%CI: 1.22-5.73), but not in clinical patients. CONCLUSIONS: High RDW was associated with a risk of in-hospital mortality independent of age, sex, hemoglobin level, multimorbidity, nutritional status and immunological condition. We therefore recommend the use of RDW as a possible marker of mortality risk in clinical practice in surgical patients.

Failure to achieve proteic goals in non-critical patients increases risk for death: Old discussion, ongoing problem.

OBJECTIVES: Providing adequate nutritional support for hospitalized patients continues to be a challenge. The aim of this study was to evaluate the association of energy and protein provision with in-hospital mortality in non-critically ill patients. METHODS: A retrospective study (2014-2016) was performed with all patients >18 y of age who were admitted to medical and surgical clinic wards and given exclusive enteral therapy. The mean of energy and protein was estimated per day and per kilogram of body mass from the enteral prescription data, over the entire period of hospitalization. A prescription mean was considered hypocaloric or hypoproteic at <20 kcal · kg · d-1 or <0.8 g · kg · d-1, respectively. RESULTS: Of the 240 patients, 58.3% were >60 y of age and 60% were men. The frequencies of in-hospital mortality (19.2%) and malnutrition (78.8%) were high. The means of protein (0.75 g· kg· d-1) and energy (17.60 kcal · kg · d-1) were below the general recommendations and 37.8% did not reach a mean of 20 kcal · kg · d-1 during the entire hospitalization period. Hypocaloric (hazard ratio [HR], 5.78; 95% confidence interval [CI], 1.59-21.04) and hypoproteic nutrition (HR, 3.69; 95% CI, 1.25-10.93) were predictors of all-cause in-hospital mortality in adjusted multivariate models. However, when we adjusted the hypoprotein nutrition by energy (non-protein calories to nitrogen ratio), hypoprotein nutrition seems to maintain the increased risk for death (HR, 3.15; 95% CI, 1.04-9.53). CONCLUSIONS: Hypoproteic nutrition appears to be more significant than hypocaloric nutrition in predicting all-cause in-hospital mortality. Protocols should be implemented to ensure that target caloric and protein levels are reached as quickly as possible to optimize patient survival.