Heart rate time series characteristics for early detection of infections in critically ill patients.

It is difficult to make a distinction between inflammation and infection. Therefore, new strategies are required to allow accurate detection of infection. Here, we hypothesize that we can distinguish infected from non-infected ICU patients based on dynamic features of serum cytokine concentrations and heart rate time series. Serum cytokine profiles and heart rate time series of 39 patients were available for this study. The serum concentration of ten cytokines were measured using blood sampled every 10 min between 2100 and 0600 hours. Heart rate was recorded every minute. Ten metrics were used to extract features from these time series to obtain an accurate classification of infected patients. The predictive power of the metrics derived from the heart rate time series was investigated using decision tree analysis. Finally, logistic regression methods were used to examine whether classification performance improved with inclusion of features derived from the cytokine time series. The AUC of a decision tree based on two heart rate features was 0.88. The model had good calibration with 0.09 Hosmer-Lemeshow p value. There was no significant additional value of adding static cytokine levels or cytokine time series information to the generated decision tree model. The results suggest that heart rate is a better marker for infection than information captured by cytokine time series when the exact stage of infection is not known. The predictive value of (expensive) biomarkers should always be weighed against the routinely monitored data, and such biomarkers have to demonstrate added value.

Circulating phthalates during critical illness in children are associated with long-term attention deficit: a study of a development and a validation cohort.

PURPOSE: Environmental phthalate exposure has been associated with attention deficit disorders in children. We hypothesized that in children treated in the pediatric intensive care unit (PICU), circulating phthalates leaching from indwelling medical devices contribute to their long-term attention deficit. METHODS: Circulating plasma concentrations of di(2-ethylhexyl)phthalate (DEHP) metabolites were quantified in 100 healthy children and 449 children who had been treated in PICU and were neurocognitively tested 4 years later. In a development patient cohort (N = 228), a multivariable bootstrap study identified stable thresholds of exposure to circulating DEHP metabolites above which there was an independent association with worse neurocognitive outcome. Subsequently, in a second patient cohort (N = 221), the observed independent associations were validated. RESULTS: Plasma concentrations of DEHP metabolites, which were virtually undetectable [0.029 (0.027-0.031) µmol/l] in healthy children, were 4.41 (3.76-5.06) µmol/l in critically ill children upon PICU admission (P < 0.001). Plasma DEHP metabolite concentrations decreased rapidly but remained 18 times higher until PICU discharge (P < 0.001). After adjusting for baseline risk factors and duration of PICU stay, and further for PICU complications and treatments, exceeding the potentially harmful threshold for exposure to circulating DEHP metabolites was independently associated with the attention deficit (all P ≤ 0.008) and impaired motor coordination (all P ≤ 0.02). The association with the attention deficit was confirmed in the validation cohort (all P ≤ 0.01). This phthalate exposure effect explained half of the attention deficit in post-PICU patients. CONCLUSIONS: Iatrogenic exposure to DEHP metabolites during intensive care was independently and robustly associated with the important attention deficit observed in children 4 years after critical illness. Clinicaltrials.gov identifier: NCT00214916.

Anterior pituitary morphology and hormone production during sustained critical illness in a rabbit model.

Prolonged critical illness is hallmarked by striking alterations in the somatrope, thyrotrope, and lactotrope axes, the severity of which is associated with the risk of morbidity and mortality. The exact role of the pituitary gland in these alterations is unknown. We studied the impact of sustained critical illness on pituitary morphology and hormone production in a standardized rabbit model of prolonged (7 days) burn injury-induced critical illness. In healthy and prolonged critically ill rabbits, we determined pituitary weight, size, morphology and orientation of the somatrope, lactotrope and thyrotrope cells and the pituitary expression of GH, PRL, and TSH at gene and protein level. The weight of the pituitary gland was unaltered by 7 days of critical illness. Also, spatial orientation and morphology of the GH, PRL, and TSH producing cells remained normal. In prolonged critically ill rabbits GH mRNA levels were higher and PRL mRNA levels were lower than in healthy controls, whereas TSH mRNA was not affected. The sizes of GH, PRL, or TSH producing cells and the pituitary content of GH, PRL, and TSH proteins were unaltered. In conclusion, in this rabbit model of prolonged critical illness, the morphology of the pituitary gland and the pituitary GH, PRL, and TSH content was normal. The alterations in pituitary hormone mRNA levels with sustained critical illness are compatible with altered hypothalamic and peripheral regulation of pituitary hormone release as previously suggested indirectly by responses to exogenous releasing factors.

Prediction of clinical conditions after coronary bypass surgery using dynamic data analysis.

This work studies the impact of using dynamic information as features in a machine learning algorithm for the prediction task of classifying critically ill patients in two classes according to the time they need to reach a stable state after coronary bypass surgery: less or more than 9 h. On the basis of five physiological variables (heart rate, systolic arterial blood pressure, systolic pulmonary pressure, blood temperature and oxygen saturation), different dynamic features were extracted, namely the means and standard deviations at different moments in time, coefficients of multivariate autoregressive models and cepstral coefficients. These sets of features served subsequently as inputs for a Gaussian process and the prediction results were compared with the case where only admission data was used for the classification. The dynamic features, especially the cepstral coefficients (aROC: 0.749, Brier score: 0.206), resulted in higher performances when compared to static admission data (aROC: 0.547, Brier score: 0.247). The differences in performance are shown to be significant. In all cases, the Gaussian process classifier outperformed to logistic regression.