Differences in Complexity of Glycemic Profile in ICU
Differences in Complexity of Glycemic Profile in ICU
Objective: To investigate glycemic dynamics and its relation with mortality in critically ill patients. We searched for differences in complexity of the glycemic profile between survivors and nonsurvivors in patients admitted to a multidisciplinary intensive care unit.
Design: Prospective, observational study, convenience sample.
Settings: Multidisciplinary intensive care unit of a teaching hospital in Madrid, Spain.
Patients: A convenience sample of 42 patients, aged 29 to 86 yrs, admitted to an intensive care unit with an Acute Physiology and Chronic Health Evaluation II score of ≥14 and with an anticipated intensive care unit stay of >72 hrs.
Interventions: A continuous glucose monitoring system was used to measure subcutaneous interstitial fluid glucose levels every 5 mins for 48 hrs during the first days of intensive care unit stay. A 24-hr period (n = 288 measurements) was used as time series for complexity analysis of the glycemic profile.
Measurements: Complexity of the glycemic profile was evaluated by means of detrended fluctuation analysis. Other conventional measurements of variability (range, sd, and Mean Amplitude of Glycemic Excursions) were also calculated.
Main results: Ten patients died during their intensive care unit stay. Glycemic profile was significantly more complex (lower detrended fluctuation analysis) in survivors (mean detrended fluctuation analysis, 1.49; 95% confidence interval, 1.44–1.53) than in nonsurvivors (1.60; 95% confidence interval, 1.52–1.68). This difference persisted after accounting for the presence of diabetes. In a logistic regression model, the odds ratio for death was 2.18 for every 0.1 change in detrended fluctuation analysis.
Age, gender, Simplified Acute Physiologic Score 3 or Acute Physiologic and Chronic Health Evaluation II scores failed to explain differences in survivorship. Conventional variability measurements did not differ between survivors and nonsurvivors.
Conclusions: Complexity of the glycemic profile of critically ill patients varies significantly between survivors and nonsurvivors. Loss of complexity in glycemia time series, evaluated by detrended fluctuation analysis, is associated with higher mortality.
In recent years, there has been a growing interest in hyperglycemia associated with critical illness. Hyperglycemia in critically ill patients is a consequence of several factors, including increased cortisol, catecholamines, glucagon, growth hormone, gluconeogenesis, and glycogenolysis. In addition, insulin resistance has been demonstrated in >80% of critically ill patients.
There is a wealth of observational evidence from different, mainly surgical, patient populations demonstrating that hyperglycemia is associated with poor clinical outcomes in critically ill patients. An important limitation in such observational evidence is that it cannot prove that hyperglycemia causes poor clinical outcomes; hyperglycemia may merely be a marker of severe illness.
There is evidence from some randomized trials that correction and prevention of hyperglycemia improve morbidity and may also decrease mortality in some critically ill patients. This suggests that there is a causal relationship between hyperglycemia and poor outcomes. Nevertheless, the optimal target blood glucose is controversial, and a widely accepted insulin regimen has not been established. Furthermore, other intervention studies have yielded contradictory results.
In recent studies, variability in blood glucose levels has emerged as a new predictor of mortality in intensive care unit (ICU) patients. In the retrospective studies by Egi et al and Krinsley, the glycemic variability was expressed as the sd of each patient's blood glucose levels extracted from electronically stored biochemical databases. The authors proposed that glycemic variability should be added as a metric to analyze ongoing and future clinical trials on intensive insulin therapy.
We hypothesized that new techniques derived from nonlinear dynamics and fractal geometry could offer a more in-depth view of the glucoregulatory process than the classic glucose variability measures (sd or Mean Amplitude of Glycemic Excursions), therefore allowing for the detection of slighter changes arguably correlated with the patient's physiologic status.
Nonlineal dynamics (the study of the behavior of nonlineal deterministic systems) is increasingly been used in physiologic studies. Nonlineal systems display an extremely complex output that, although being rigorously deterministic, is unpredictable and, at first glance, seems to be random (thus, the term "pseudorandom"). These systems have some striking similarities with certain physiologic mechanisms: They exhibit a pseudorandom behavior, they tend to develop spontaneous rhythms, and most notably, they have a strong tendency to remain in a narrow range of values ("strange attractors"), displaying a behavior that could easily be called homeostatic.
Complexity analysis of time series has been widely used in the study of variability of biological phenomena, such as cardiac interbeat interval, cardiac arrhythmia, intracranial hypertension, sepsis and organ failure, temperature and electroencephalogram activity. Ogata et al and Churruca et al have analyzed diabetes-related alterations of glucose control by means of complexity analysis of the glycemic profile. They have reported diminished complexity of glycemic profile in diabetic patients vs. healthy volunteers. Several authors have proposed that critical illness and multisystem organ dysfunction are characterized by the phenomenon of decomplexification. Healthy state exhibits some degree of (pseudo)random variability in physiologic variables, such as heart rate or temperature. Loss of such irregularity (and consequently of complexity) is one of the hallmarks of critical illness.
It should be noted that complexity and variability, although seemingly related, are quite different and often contradictory concepts. A key difference between variability and complexity metrics is that, although variability is based on conventional statistics (range, sd) and, thus, takes each measurement as an independent value, in complexity analysis, each measure is related to its neighbors. This arguably allows complexity analysis to detect minor systemic dysfunctions, not perceived by variability studies. In general, a healthy regulatory system displays a complex output, with frequent and quick corrections of even small deviations. On the other hand, a failing regulatory system will be sluggish and allow for greater deviations before reacting. Therefore, as a rule of thumb, healthy systems have a high complexity and low variability, while failing systems display lower complexity and higher variability.
The aim of this pilot study was to investigate the complexity of the glycemic profile in critically ill patients. Namely, we searched for differences in complexity of glycemic profile between the patients who survive and the patients who die in the ICU. We hypothesized that the glycemic profile would be less complex in nonsurvivors.
Abstract and Introduction
Abstract
Objective: To investigate glycemic dynamics and its relation with mortality in critically ill patients. We searched for differences in complexity of the glycemic profile between survivors and nonsurvivors in patients admitted to a multidisciplinary intensive care unit.
Design: Prospective, observational study, convenience sample.
Settings: Multidisciplinary intensive care unit of a teaching hospital in Madrid, Spain.
Patients: A convenience sample of 42 patients, aged 29 to 86 yrs, admitted to an intensive care unit with an Acute Physiology and Chronic Health Evaluation II score of ≥14 and with an anticipated intensive care unit stay of >72 hrs.
Interventions: A continuous glucose monitoring system was used to measure subcutaneous interstitial fluid glucose levels every 5 mins for 48 hrs during the first days of intensive care unit stay. A 24-hr period (n = 288 measurements) was used as time series for complexity analysis of the glycemic profile.
Measurements: Complexity of the glycemic profile was evaluated by means of detrended fluctuation analysis. Other conventional measurements of variability (range, sd, and Mean Amplitude of Glycemic Excursions) were also calculated.
Main results: Ten patients died during their intensive care unit stay. Glycemic profile was significantly more complex (lower detrended fluctuation analysis) in survivors (mean detrended fluctuation analysis, 1.49; 95% confidence interval, 1.44–1.53) than in nonsurvivors (1.60; 95% confidence interval, 1.52–1.68). This difference persisted after accounting for the presence of diabetes. In a logistic regression model, the odds ratio for death was 2.18 for every 0.1 change in detrended fluctuation analysis.
Age, gender, Simplified Acute Physiologic Score 3 or Acute Physiologic and Chronic Health Evaluation II scores failed to explain differences in survivorship. Conventional variability measurements did not differ between survivors and nonsurvivors.
Conclusions: Complexity of the glycemic profile of critically ill patients varies significantly between survivors and nonsurvivors. Loss of complexity in glycemia time series, evaluated by detrended fluctuation analysis, is associated with higher mortality.
Introduction
In recent years, there has been a growing interest in hyperglycemia associated with critical illness. Hyperglycemia in critically ill patients is a consequence of several factors, including increased cortisol, catecholamines, glucagon, growth hormone, gluconeogenesis, and glycogenolysis. In addition, insulin resistance has been demonstrated in >80% of critically ill patients.
There is a wealth of observational evidence from different, mainly surgical, patient populations demonstrating that hyperglycemia is associated with poor clinical outcomes in critically ill patients. An important limitation in such observational evidence is that it cannot prove that hyperglycemia causes poor clinical outcomes; hyperglycemia may merely be a marker of severe illness.
There is evidence from some randomized trials that correction and prevention of hyperglycemia improve morbidity and may also decrease mortality in some critically ill patients. This suggests that there is a causal relationship between hyperglycemia and poor outcomes. Nevertheless, the optimal target blood glucose is controversial, and a widely accepted insulin regimen has not been established. Furthermore, other intervention studies have yielded contradictory results.
In recent studies, variability in blood glucose levels has emerged as a new predictor of mortality in intensive care unit (ICU) patients. In the retrospective studies by Egi et al and Krinsley, the glycemic variability was expressed as the sd of each patient's blood glucose levels extracted from electronically stored biochemical databases. The authors proposed that glycemic variability should be added as a metric to analyze ongoing and future clinical trials on intensive insulin therapy.
We hypothesized that new techniques derived from nonlinear dynamics and fractal geometry could offer a more in-depth view of the glucoregulatory process than the classic glucose variability measures (sd or Mean Amplitude of Glycemic Excursions), therefore allowing for the detection of slighter changes arguably correlated with the patient's physiologic status.
Nonlineal dynamics (the study of the behavior of nonlineal deterministic systems) is increasingly been used in physiologic studies. Nonlineal systems display an extremely complex output that, although being rigorously deterministic, is unpredictable and, at first glance, seems to be random (thus, the term "pseudorandom"). These systems have some striking similarities with certain physiologic mechanisms: They exhibit a pseudorandom behavior, they tend to develop spontaneous rhythms, and most notably, they have a strong tendency to remain in a narrow range of values ("strange attractors"), displaying a behavior that could easily be called homeostatic.
Complexity analysis of time series has been widely used in the study of variability of biological phenomena, such as cardiac interbeat interval, cardiac arrhythmia, intracranial hypertension, sepsis and organ failure, temperature and electroencephalogram activity. Ogata et al and Churruca et al have analyzed diabetes-related alterations of glucose control by means of complexity analysis of the glycemic profile. They have reported diminished complexity of glycemic profile in diabetic patients vs. healthy volunteers. Several authors have proposed that critical illness and multisystem organ dysfunction are characterized by the phenomenon of decomplexification. Healthy state exhibits some degree of (pseudo)random variability in physiologic variables, such as heart rate or temperature. Loss of such irregularity (and consequently of complexity) is one of the hallmarks of critical illness.
It should be noted that complexity and variability, although seemingly related, are quite different and often contradictory concepts. A key difference between variability and complexity metrics is that, although variability is based on conventional statistics (range, sd) and, thus, takes each measurement as an independent value, in complexity analysis, each measure is related to its neighbors. This arguably allows complexity analysis to detect minor systemic dysfunctions, not perceived by variability studies. In general, a healthy regulatory system displays a complex output, with frequent and quick corrections of even small deviations. On the other hand, a failing regulatory system will be sluggish and allow for greater deviations before reacting. Therefore, as a rule of thumb, healthy systems have a high complexity and low variability, while failing systems display lower complexity and higher variability.
The aim of this pilot study was to investigate the complexity of the glycemic profile in critically ill patients. Namely, we searched for differences in complexity of glycemic profile between the patients who survive and the patients who die in the ICU. We hypothesized that the glycemic profile would be less complex in nonsurvivors.
Source...