Contributing factors to hypophosphatemia development in critically Ill ventilated patients: a retrospective cohort study.

Hypophosphatemia (serum phosphate < 2.5 mg/dL) is a major concern when initiating nutritional support. We evaluated which factors contribute to hypophosphatemia development in critically ill patients, as well as the association between hypophosphatemia and mortality. A retrospective cohort study of patients who were ventilated for at least 2 days in a 16-bed mixed ICU. Data collected includes demographics, Acute Physiology & Chronic Health Evaluation 2 (APACHE2) admission score, Sequential Organ Failure Assessment score at 24 h (SOFA24), hourly energy delivery, plasma phosphate levels during the first 2 weeks of admission, ICU length of stay (LOS), length of ventilation (LOV), and mortality (ICU and 90 days). For the hypophosphatemia development model, we considered mortality as a competing risk. For mortality analysis, we used the Cox proportional hazards model considering hypophosphatemia development as a time-varying covariate. 462 patients were used in the analysis. 59.52% of the patients developed hypophosphatemia. Several factors were associated with a decreased risk of hypophosphatemia: age, BMI, pre-admission diabetes diagnosis, APACHE2, SOFA24, first kidney SOFA score, hospital admission time before ICU admission, and admission after liver transplantation. Admission due to trauma was associated with an increased risk of hypophosphatemia. Survival analysis with hypophosphatemia as a time-varying covariate showed a protective effect of hypophosphatemia from mortality (HR 0.447, 95% CI 0.281, 0.712). Age, APACHE2, and SOFA24 score were found to be significantly associated with ICU mortality. Fasting duration in the ICU before nutritional support initiation was not found to be significantly associated with hypophosphatemia. We examined several fasting intervals (12 h, 24 h, 36 h, 48 h, 60 h, 72 h). In each fast interval, we compared the prevalence of hypophosphatemia among patients who fasted the specified length of time, with those who did not fast for the same length of time. In each fasting interval, hypophosphatemia prevalence was lower in the fasting group compared to the non-fasting group. However, this difference was insignificant. BMI, APACHE2, and hospital LOS before ICU admission were inversely associated with hypophosphatemia development. Fasting for up to 72 h in the ICU before starting nutritional support did not affect hypophosphatemia occurrence. Hypophosphatemia was associated with lower mortality.

The future of artificial intelligence in clinical nutrition.

PURPOSE OF REVIEW: Artificial intelligence has reached the clinical nutrition field. To perform personalized medicine, numerous tools can be used. In this review, we describe how the physician can utilize the growing healthcare databases to develop deep learning and machine learning algorithms, thus helping to improve screening, assessment, prediction of clinical events and outcomes related to clinical nutrition. RECENT FINDINGS: Artificial intelligence can be applied to all the fields of clinical nutrition. Improving screening tools, identifying malnourished cancer patients or obesity using large databases has been achieved. In intensive care, machine learning has been able to predict enteral feeding intolerance, diarrhea, or refeeding hypophosphatemia. The outcome of patients with cancer can also be improved. Microbiota and metabolomics profiles are better integrated with the clinical condition using machine learning. However, ethical considerations and limitations of the use of artificial intelligence should be considered. SUMMARY: Artificial intelligence is here to support the decision-making process of health professionals. Knowing not only its limitations but also its power will allow precision medicine in clinical nutrition as well as in the rest of the medical practice.

Using Machine-Learning to Assess the Prognostic Value of Early Enteral Feeding Intolerance in Critically Ill Patients: A Retrospective Study.

BACKGROUND: The association between gastrointestinal intolerance during early enteral nutrition (EN) and adverse clinical outcomes in critically ill patients is controversial. We aimed to assess the prognostic value of enteral feeding intolerance (EFI) markers during early ICU stays and to predict early EN failure using a machine learning (ML) approach. METHODS: We performed a retrospective analysis of data from adult patients admitted to Beilinson Hospital ICU between January 2011 and December 2018 for more than 48 h and received EN. Clinical data, including demographics, severity scores, EFI markers, and medications, along with 72 h after admission, were analyzed by ML algorithms. Prediction performance was assessed by the area under the receiver operating characteristics (AUCROC) of a ten-fold cross-validation set. RESULTS: The datasets comprised 1584 patients. The means of the cross-validation AUCROCs for 90-day mortality and early EN failure were 0.73 (95% CI 0.71-0.75) and 0.71 (95% CI 0.67-0.74), respectively. Gastric residual volume above 250 mL on the second day was an important component of both prediction models. CONCLUSIONS: ML underlined the EFI markers that predict poor 90-day outcomes and early EN failure and supports early recognition of at-risk patients. Results have to be confirmed in further prospective and external validation studies.

Gastrointestinal failure, big data and intensive care.

PURPOSE OF REVIEW: Enteral feeding is the main route of administration of medical nutritional therapy in the critically ill. However, its failure is associated with increased complications. Machine learning and artificial intelligence have been used in intensive care to predict complications. The aim of this review is to explore the ability of machine learning to support decision making to ensure successful nutritional therapy. RECENT FINDINGS: Numerous conditions such as sepsis, acute kidney injury or indication for mechanical ventilation can be predicted using machine learning. Recently, machine learning has been applied to explore how gastrointestinal symptoms in addition to demographic parameters and severity scores, can accurately predict outcomes and successful administration of medical nutritional therapy. SUMMARY: With the rise of precision and personalized medicine for support of medical decisions, machine learning is gaining popularity in the field of intensive care, first not only to predict acute renal failure or indication for intubation but also to define the best parameters for recognizing gastrointestinal intolerance and to recognize patients intolerant to enteral feeding. Large data availability and improvement in data science will make machine learning an important tool to improve medical nutritional therapy.

Correlations between First 72 h Hypophosphatemia, Energy Deficit, Length of Ventilation, and Mortality-A Retrospective Cohort Study.

INTRODUCTION: Hypophosphatemia may prolong ventilation and induce weaning failure. Some studies have associated hypophosphatemia with increased mortality. Starting or restarting nutrition in a critically ill patient may be associated with refeeding syndrome and hypophosphatemia. The correlation between nutrition, mechanical ventilation, and hypophosphatemia has not yet been fully elucidated. METHODS: A retrospective cohort study of 825 admissions during two consecutive years was conducted. Using the electronic medical chart, demographic and clinical data were obtained. Hypophosphatemia was defined as a phosphate level below 2.5 mg/dL (0.81 mmol/L) in the first 72 h of ICU admission. Comparisons between baseline characteristics and outcomes and multivariate analysis were performed. RESULTS: A total of 324 (39.27%) patients had hypophosphatemia during the first 72 h of ICU admission. Patients with hypophosphatemia tended to be younger, with lower APACHE-II, SOFA24, and ΔSOFA scores. They had a longer length of stay and length of ventilation, more prevalent prolonged ventilation, and decreased mortality. Their energy deficit was lower. There was no effect of hypophosphatemia severity on these results. In multivariate analysis, hypophosphatemia was not found to be statistically significant either with respect to mortality or survivor's length of ventilation, but lower average daily energy deficit and SOFA24 were found to be statistically significant with respect to survivor's length of ventilation. CONCLUSION: Hypophosphatemia had no effect on mortality or length of ventilation. Lower average daily energy deficit is associated with a longer survivor's length of ventilation.

Towards personalized nutritional treatment for malnutrition using machine learning-based screening tools.

Early identification of patients at risk of malnutrition or who are malnourished is crucial in order to start a timely and adequate nutritional therapy. Yet, despite the presence of many nutrition screening tools for use in the hospital setting, there is no consensus regarding the best tool as well as inadequate adherence to screening practices which impairs the achievement of effective nutritional therapy. In recent years, artificial intelligence and machine learning methods have been widely used, across multiple medical domains, to aid clinical decision making and to improve quality and efficiency of care. Therefore, Yin and colleagues propose a machine learning based individualized decision support system aimed to identify and grade malnutrition in cancer patients by applying unsupervised and supervised machine learning methods on nationwide cohort. This approach, demonstrate the ability of machine learning methods to create tools to recognize malnutrition. The machine learning based screening serves as a first layer in a nutritional therapy workflow and provides improved support for decision making of health professionals to fit individualized nutritional therapy in at-risk patients.

Improvement in the Prediction of Coronary Heart Disease Risk by Using Artificial Neural Networks.

BACKGROUND AND OBJECTIVES: Cardiovascular diseases, such as coronary heart disease (CHD), are the main cause of mortality and morbidity worldwide. Although CHD cannot be entirely predicted by classic risk factors, it is preventable. Therefore, predicting CHD risk is crucial to clinical cardiology research, and the development of innovative methods for predicting CHD risk is of great practical interest. The Framingham risk score (FRS) is one of the most frequently implemented risk models. However, recent advances in the field of analytics may enhance the prediction of CHD risk beyond the FRS. Here, we propose a model based on an artificial neural network (ANN) for predicting CHD risk with respect to the Framingham Heart Study (FHS) dataset. The performance of this model was compared to that of the FRS. METHODS: A sample of 3066 subjects from the FHS offspring cohort was subjected to an ANN. A multilayer perceptron ANN architecture was used and the lift, gains, receiver operating characteristic (ROC), and precision-recall predicted by the ANN were compared with those of the FRS. RESULTS: The lift and gain curves of the ANN model outperformed those of the FRS model in terms of top percentiles. The ROC curve showed that, for higher risk scores, the ANN model had higher sensitivity and higher specificity than those of the FRS model, although its area under the curve (AUC) was lower. For the precision-recall measures, the ANN generated significantly better results than the FRS with a higher AUC. CONCLUSIONS: The findings suggest that the ANN model is a promising approach for predicting CHD risk and a good screening procedure to identify high-risk subjects.