Introduction: Sepsis is a life-threatening condition in which dysregulated host response to infection leads to multiorgan failure. Early detection of deterioration in sepsis is key to improving overall patient outcomes. Studies investigating complex disorders using a network physiology approach have shown a distinct pattern of organ system connectivity between different patient populations with positive and negative disease outcomes. Physiological network mapping is not currently used in intensive care units. This study used a parenclitic network approach to compute organ connectivity for individual patients with sepsis using records of routine laboratory test results.

Aims/objectives: The goal of this study was to investigate whether analysing organ connectivity in individual patients with sepsis using a parenclitic network could predict the deterioration and mortality of sepsis patients.

Method: Electronic patient records from 162 patients with sepsis were obtained from the MIMIC-III database (Ethics IRB protocol nos. 2001P001699). Patients were studied retrospectively, and clinical data on 48-hour deterioration and 30-day survival were obtained. Fifteen physiological variables (serum phosphate, arterial pH, urea, haemoglobin, lactate, white blood cell count, serum sodium, international normalized ratio, platelets, total bilirubin, blood glucose, serum creatinine, alanine transaminase, bicarbonate, and serum potassium) representing different organ systems were extracted from the laboratory data. Correlation analysis of physiological variables was performed to study the pattern of interaction between patient groups with different outcomes (i.e., survivors versus non-survivors). The parenclitic network investigated organ connectivity in individual patients by examining how individual patient data deviated from the characteristics of organ relationships established in a reference population (survivors). Parenclitic deviations were computed for everyone, and Cox regression was used to investigate if parenclitic deviations could predict 48-hour deterioration and 24-hour survival in sepsis patients.

Results: Correlation analysis identified 7 and 9 pairs of unique correlations in physiological variables that were significantly different between survivors and non-survivors for 30-day survival and between deteriorated and non-deteriorated patients for 48-hour deterioration, respectively. Parenclitic deviations for the pH-bicarbonate axis (hazard ratio = 2.081, p < 0.001) and pH-lactate axis (hazard ratio = 2.773, p = 0.024) were able to significantly predict 30-day mortality in sepsis patients independent of the measures of organ dysfunction severity, SOFA, and ventilation status. None of the parenclitic deviations were able to predict 48-hour deterioration.

Conclusions: Investigating organ connectivity in individual patients using parenclitic network analysis significantly predicted 30-day mortality in sepsis population. Parenclitic deviation may potentially offer useful insight into pathophysiology of sepsis and give useful insight into different physiological response towards sepsis between survivors and non-survivors.