Evaluation of sepsis in a critically ill surgical population

Machine-learning models for prediction of sepsis patients mortality

OBJECTIVES

Sepsis is an infection-caused syndrome, that leads to life-threatening organ damage. We aim to develop machine learning models with large-scale data to predict sepsis patients' mortality.

DESIGN

we extracted sepsis patients from two databases, Medical Information Mart for Intensive Care IV (MIMIC-IV) as a train set and Philips eICU Collaborative Research Database as a test set.

SETTING

ICUs in multicenter hospitals in the USA during 2012-2019.

PATIENTS OR PARTICIPANTS

A total of 21,680 sepsis-3 patients are included in the study, in which, 3771 patients were dead and 17,909 survived during hospitalization, respectively.

INTERVENTIONS

No interventions.

MAIN VARIABLES OF INTEREST

Basic information, examination items during hospitalization and some medication and treatment information are incorporated into analyzed. Seven different models were built with a Support vector machine, Decision Tree Classifier, Random Forest, Gradients Boosting, Multiple Layer Perception, Xgboost, light Gradients Boosting to predict dead or live during hospitalization.

RESULTS

Algorithms with an AUC value in the test set of the top three: light GBM, GBM, Xgboost. Considering the performance of the training set and the test set, the light GBM model performs best, and then the parameters of the model were adjusted, after that the AUC value was 0.99 in the train set, 0.96 in the test set, respectively.

CONCLUSIONS

Models built with light GBM algorithm from real-world sepsis patients from electronic health records accurately predict whether sepsis patients are dead and can be incorporated into clinical decision tools to enhance the prognosis of the patient and prevent adverse outcomes.

Multimedia abstract generation of intensive care data: the automation of clinical processes through AI methodologies

Medical errors from communication failures are enormous during the perioperative period of cardiac surgical patients. As caregivers change shifts or surgical patients change location within the hospital, key information is lost or misconstrued. After a baseline cognitive study of information need and caregiver workflow, we implemented an advanced clinical decision support tool of intelligent agents, medical logic modules, and text generators called the "Inference Engine" to summarize individual patient's raw medical data elements into procedural milestones, illness severity, and care therapies. The system generates two displays: 1) the continuum of care, multimedia abstract generation of intensive care data (MAGIC)-an expert system that would automatically generate a physician briefing of a cardiac patient's operative course in a multimodal format; and 2) the isolated point in time, "Inference Engine"-a system that provides a real-time, high-level, summarized depiction of a patient's clinical status. In our studies, system accuracy and efficacy was judged against clinician performance in the workplace. To test the automated physician briefing, "MAGIC," the patient's intraoperative course, was reviewed in the intensive care unit before patient arrival. It was then judged against the actual physician briefing and that given in a cohort of patients where the system was not used. To test the real-time representation of the patient's clinical status, system inferences were judged against clinician decisions. Changes in workflow and situational awareness were assessed by questionnaires and process evaluation. MAGIC provides 200% more information, twice the accuracy, and enhances situational awareness. This study demonstrates that the automation of clinical processes through AI methodologies yields positive results.

Generation and evaluation of intraoperative inferences for automated health care briefings on patient status after bypass surgery

OBJECTIVE

The authors present a system that scans electronic records from cardiac surgery and uses inference rules to identify and classify abnormal events (e.g., hypertension) that may occur during critical surgical points (e.g., start of bypass). This vital information is used as the content of automatically generated briefings designed by MAGIC, a multimedia system that they are developing to brief intensive care unit clinicians on patient status after cardiac surgery. By recognizing patterns in the patient record, inferences concisely summarize detailed patient data.

DESIGN

The authors present the development of inference rules that identify important information about patient status and describe their implementation and an experiment they carried out to validate their correctness. The data for a set of 24 patients were analyzed independently by the system and by 46 physicians.

MEASUREMENTS

The authors measured accuracy, specificity, and sensitivity by comparing system inferences against physician judgments, in cases where all three physicians agreed and against the majority opinion in all cases.

RESULTS

For laboratory inferences, evaluation shows that the system has an average accuracy of 98 percent (full agreement) and 96 percent (majority model). An analysis of interrater agreement, however, showed that physicians do not agree on abnormal hemodynamic events and could not serve as a gold standard for evaluating hemodynamic events. Analysis of discrepancies reveals possibilities for system improvement and causes of physician disagreement.

CONCLUSIONS

This evaluation shows that the laboratory inferences of the system have high accuracy. The lack of agreement among physicians highlights the need for an objective quality-assurance tool for hemodynamic inferences. The system provides such a tool by implementing inferencing procedures established in the literature.

Procalcitonin--a sepsis parameter in severe burn injuries

Procalcitonin (PCT) levels increase in patients with systemic infections; the highest levels have been found in sepsis. This study tested whether plasma procalcitonin level was related to sepsis, CRP, burn size, inhalation injury or mortality in severely burned patients over the entire clinical course. In 27 patients with 51 (20-91)% TBSA, PCT was measured three times weekly from admission over the entire course of stay in a single ICU. Daily scoring by the "Baltimore Sepsis Scale" was performed. The patients were assigned to three groups depending on the clinical course and outcome: A = no septic complications, B = septic complications-survivors, C = septic complications non-survivors. PCT levels were elevated slightly at admission (mean 2.1 ng/ml) except in three patients who suffered electrical burns (mean 15.7 ng/ml). PCT peak levels correlated well with the Scoring values (r = 0.84) while CRP did not (r = 0.64). Peak PCT levels were significantly higher (p < 0.005) in septic patients (B and C) who averaged 49.8+/-76.9 ng/ml, than in non-septic patients (A) who averaged peak levels of 2.3+/-3.7 ng/ml. The highest PCT levels were found immediately before death (86.8+/-97 ng/ml). Seven patients had an inhalation injury 3rd degree. In these patients at 24 h postburn, there was no relationship between PCT levels and inhalation injury but during the later days postburn there were significant differences in PCT levels in patients with versus without inhalation injury. All patients with inhalation injury 3rd degree developed septic complications. There was no positive correlation between the PCT-admission-levels and the TBSA, but there was a positive correlation between the TBSA and the mean peak PCT levels during the later days postburn (r = 0.73; p < 0.05). The cut-off value of 3 ng/ ml we found reliable to indicate severe bacterial or fungal infection. PCT values over 10 ng/ml increasing over the following days were found only in life-threatening situations due to systemic infections. The individual course of PCT in one patient is more important than absolute values. PCT presented in this study as a useful diagnostic parameter in severely burned patients.

Antithrombin III in patients with severe sepsis. A randomized, placebo-controlled, double-blind multicenter trial plus a meta-analysis on all randomized, placebo-controlled, double-blind trials with antithrombin III in severe sepsis

OBJECTIVES

To evaluate the safety and potential efficacy of antithrombin III (AT III) in reducing mortality in patients with severe sepsis.

DESIGN

Prospective, randomized, placebo-controlled, double-blind, phase II, multicenter, multinational clinical trial.

SETTING

Seven academic medical center intensive care units (ICU) in Belgium, Denmark, the Netherlands, Norway and Sweden.

PATIENTS

42 patients with severe sepsis who received standard supportive care and antimicrobial therapy, in addition to the administration of AT III or placebo.

INTERVENTIONS

Patients received either an intravenous loading dose of 3000 IU AT III followed by a maintenance dose of 1500 IU every 12 h for 5 days or equivalent amounts of placebo.

MEASUREMENTS AND RESULTS

All patients were evaluated for safety and for 30-day all-cause mortality.

CONCLUSIONS

The administration of AT III was safe and well-tolerated. It was followed by a 39 % reduction in 30-day all-cause mortality (NS). The reduction in mortality was accompanied by a considerably shorter stay in the ICU. Patients treated with AT III exhibited a better performance in overall severity of illness and organ failure scores (Acute Physiology and Chronic Health Evaluation II, multiple organ failure, organ system failure), which was noticeable soon after initiation of treatment. Patients treated with AT III demonstrated a better resolution of pre-existing organ failures and a lower incidence of new organ failures during the observation period. A meta-analysis comprising this and two other double-blind, placebo-controlled trials with AT III with a total of 122 patients suffering from severe sepsis confirms the positive trend. The results of the meta-analysis demonstrate a 22.9 % reduction in 30-day all-cause mortality in patients treated with AT III. Although still too small to be confirmative, the meta-analysis clearly points to the fact that a sufficiently powered phase III trial is warranted to prove whether AT III has a beneficial role in the treatment of severe sepsis.

Poor outcome from peritonitis is caused by disease acuity and organ failure, not recurrent peritoneal infection

OBJECTIVE

The purpose of the study is to determine whether organ failure develops in patients despite control of peritoneal infection and whether the process is, in part, neutrophil (polymorphonuclear leukocyte [PMN]) mediated.

SUMMARY BACKGROUND DATA

Peritonitis generally responds to prompt surgical intervention and systemic antibiotics; however, some patients continue a septic course and progress to organ failure and death.

METHODS

One hundred five consecutive patients with peritonitis between 1988 and 1996 who required operation and a postoperative hospital stay greater than 10 days were studied. Mice were injected with a monoclonal anti-PMN antibody 24 hours before cecal ligation and puncture (CLP) to deplete PMNs.

RESULTS

Thirty-eight patients died, and all but 1 had identified organ failure. Seventy-seven patients had either pulmonary failure alone (25 patients) or as a component of multisystem organ failure (52 patients). All but one of these patients showed resolution of their intraperitoneal infection as evident by clinical course, abdominal computed tomographic scan, second-look laparotomy, or autopsy. Recurrent intra-abdominal infection developed in 15 patients, but only 1 had organ failure, and 2 died. At 18 hours after CLP, lung injury, PMN content, interleukin-1 mRNA expression, and liver injury were significantly reduced by anti-PMN treatment, whereas serum endotoxin levels actually increased.

CONCLUSIONS

Disease acuity and organ failure, and not recurrent peritoneal infection, are the major causes of adverse outcome in patients with peritonitis. The authors' experimental data indicate that such organ injury is, in part, PMN mediated but not endotoxin mediated. Attraction of PMNs toward the site of primary infection, and thereby away from remote organs, is a logical future therapeutic approach in such patients who are critically ill with peritonitis.

Prediction of outcome from intensive care after gastroenterologic emergency

Prognostic factors determining the outcome from intensive care were studied in 952 patients admitted to 25 Finnish ICUs after gastroenterologic emergency. Logistic regression analysis was used to create predictive models based on the APACHE II-system. The models were constructed by using data from a random two-thirds of the study population and validated in the remaining independent one-third together with the original APACHE II-index. The Acute Physiology Score, age, and a pre-existing liver disease were the three most important determinants of outcome. The inclusion of the TISS score describing the intensity of treatment into a model did not enhance the accuracy of the prediction. Our models were better calibrated than the original APACHE II-equation when tested by the goodness-of-fit -statistics. These statistical models may help the clinicians to predict the outcome for an individual patient by providing them information about the relative impacts of predictive factors or about the probability of death. These probabilities should be interpreted cautiously, taking into account the limitations of statistical methods. This is especially important when assessing the highrisk patients. Their number in our study was too low for accurate outcome prediction.

Ruptured aortic aneurysm: early postoperative prediction of mortality using an organ system failure score

Ruptured abdominal aortic aneurysm (AAA) is a personal and public health catastrophe because of the high and unimproving mortality rate, increasingly long intensive care unit (ICU) stay and rising hospital costs. Criteria are needed to identify patients with a poor prognosis so that treatment and resources may be directed to those with a better outlook. This retrospective study reviews perioperative variables, outcome and length of ICU stay for 99 consecutive patients with operated ruptured infrarenal AAA between 1985 and 1992 who reached the ICU alive. Early death (within 48 h) occurred in 20 per cent using 24 ICU days and late death (after 48 h) occurred in 29 per cent using 420 ICU days; neither type of death could be predicted before operation. Multiple organ failure, the cause of 93 per cent of late deaths, was assessed for each patient 48 h after operation using an organ system failure score. There was a strong positive correlation between organ system failure score and mortality rate (P < 0.00001); all 20 patients with more than two failing organ systems died. If used in decision making the score would have saved 43 per cent of the ICU days associated with late mortality. Withdrawing postoperative treatment at an early stage in accordance with an organ system failure score may be a defensible and cost-effective option.

Acute renal failure in the critically ill: management by continuous veno-venous hemodiafiltration

The consequences of newer techniques of continuous renal replacement therapy in critically ill patients are not yet fully known. The clinical and biochemical impact of continuous veno-venous hemodiafiltration (CVVHD) was, therefore, prospectively studied in 60 critically ill patients with acute renal failure. Prospective clinical, biochemical, and hematological data were collected from patients receiving CVVHD. Over the initial 24 hours of therapy, CVVHD resulted in a decrease in mean plasma urea from 34.5 mmol/L (95% confidence interval [CI], 29.4 to 39.6) to 25 mmol/L (95% CI, 21.8 to 28.2). With continued CVVHD, the mean plasma urea reached a plateau level of 17.6 mmol/L (95% CI, 15.8 to 19.4) at 72 hours. This degree of azotemia control was achieved with ease and essentially without complications during 8,360 hours of therapy despite the presence of multi-organ failure and the aggressive administration of protein nitrogen (0.25 to 0.35 g/kg/day). No abnormalities of serum electrolytes developed during treatment. Survival to intensive care discharge was 46.6% and to hospital discharge 41.6%, despite a mean Acute Physiology and Chronic Health Evaluation (APACHE) II score at presentation of 27.7. Continuous veno-venous hemodiafiltration offers superior azotemia control and a safe approach to renal replacement therapy in critically ill patients. Its use is associated with a comparatively favorable outcome. CVVHD may be regarded as the treatment of choice in such patients.

Acute continuous hemodiafiltration: a prospective study of 110 patients and a review of the literature

One hundred ten critically ill patients with acute renal failure receiving acute continuous hemodiafiltration (ACHD) in our intensive care unit were studied prospectively. Acute continuous hemodiafiltration consisted either of continuous arteriovenous hemodiafiltration or of continuous veno-venous hemodiafiltration, and was used for 17,817 hours (mean duration of patient treatment, 161.9 hours), resulting in a fall from a mean pre-ACHD urea of 35.7 mmol/L to a plateau value of 16.8 mmol/L at 72 hours of treatment. The mean urea clearance achieved was 24.9 mL/min. Eighty of these patients (72.7%) were receiving artificial ventilation at the time of ACHD and 45 (40.9%) had more than four failing organs. The mean APACHE II score was 27.7. Despite the degree of illness severity, 42 patients (32.2%) survived to discharge from hospital. The use of ACHD was associated with hemodynamic stability, rapid normalization of electrolytes, and the ability to freely administer drugs, blood, and/or blood products. It also allowed for maintenance of an aggressive, nitrogen-rich, nutritional regimen. Support of these critically ill patients with acute renal failure using ACHD was achieved safely and without the employment of additional dialysis-trained nursing staff. Our own experience and a review of the available literature strongly suggest that the advantages associated with the use of ACHD therapies are clinically significant and support the view that ACHD is a modality of renal replacement most suited to critically ill patients with acute renal failure.

Characterization of intensive care unit patients using a model based on the presence or absence of organ dysfunctions and/or infection: the ODIN model

OBJECTIVE

To evaluate the sensitivity, specificity and overall accuracy of a model based on the presence or absence of organ dysfunctions and/or infection (ODIN) to predict the outcome for intensive care unit patients.

DESIGN

Prospective study.

SETTING

General intensive care unit in a university teaching hospital.

PATIENTS

1070 consecutive, unselected patients.

INTERVENTIONS

There were no interventions.

MEASUREMENTS AND MAIN RESULTS

We recorded within the first 24 h of admission the presence or absence of dysfunction in 6 organ systems: respiratory, cardiovascular, renal, hematologic, hepatic and neurologic, and/or infection (ODIN) in all patients admitted to our ICU, thus establishing a profile of organ dysfunctions in each patient. Using univariate analysis, a strong correlation was found between the number of ODIN and the death rate (2.6, 9.7, 16.7, 32.3, 64.9, 75.9, 94.4 and 100% for 0, 1, 2, 3, 4, 5, 6 and 7 ODIN, respectively; (p < 0.001). In addition, the highest mortality rates were associated with hepatic (60.8%), hematologic (58.1%) and renal (54.8%) dysfunctions, and the lowest with respiratory dysfunction (36.5%) and infection (38.3%). For taking into account both the number and the type of organ dysfunction, a logistic regression model was then used to calculate individual probabilities of death that depended upon the statistical weight assigned to each ODIN (in the following order of descending severity: cardiovascular, renal, respiratory, neurologic, hematologic, hepatic dysfunctions and infection). The ability of this severity-of-disease classification system to stratify a wide variety of patients prognostically (sensitivity 51.4%, specificity 93.4%, overall accuracy 82.1%) was not different from that of currently used scoring systems.

CONCLUSIONS

These findings suggest that determination of the number and the type of organ dysfunctions and infection offers a clear and reliable method for characterizing ICU patients. Before a widespread use, this model requires to be validated in other institutions.

A comparison of conventional dialytic therapy and acute continuous hemodiafiltration in the management of acute renal failure in the critically ill

OBJECTIVE

To compare and contrast the clinical outcomes in critically ill patients with acute renal failure managed with either acute continuous hemodiafiltration or conventional dialytic therapies.

DESIGN

Retrospective review of the medical records of 167 consecutive cases of acute renal failure treated at a single center (July 1982-July 1991). Scoring for illness severity (APACHE II, number of failing organs) and assessment of outcome in terms of biochemical control of azotemia, ARF therapy-related morbidity, and overall morbidity and mortality.

SETTING

Tertiary institution.

PATIENTS

167 consecutive critically ill patients with multiorgan failure and acute renal failure.

MEASUREMENTS AND MAIN RESULTS

84 patients received conventional dialytic therapy (CDT) (1982-1988) and 83 acute continuous hemodiafiltration (ACHD) (1988-1991). The etiology of ARF and illness severity indices were similar in both groups (organ failure scores: CDT 3.9 vs. ACHD 4.1; NS). All patients were critically ill, with more severely ill patients within the ACHD groups (mean APACHE II score: CDT 25.8 vs. ACHD 28.1; p < .01). There were no significant differences in pretreatment serum creatinine, glucose, bicarbonate and phosphate, white cell and platelet counts, incidence of disseminated intravascular coagulation, prevalence of sepsis, or evidence of pulmonary and/or peripheral edema. Overall survival was 29.8% for the CDT groups and 41% for the ACHD group (NS). When patients were stratified by severity of illness, survival in those with 2 to 4 failing organs was significantly greater in the ACHD group (CDT 31.1% vs. ACHD 53.8%; p < .025). Similarly, overall survival in patients with intermediate APACHE II scores (24 to 29) was significantly better in those treated with ACHD (CDT 12.5% vs. ACHD 46.4%; p < .025). During the course of ARF, in comparison to CDT, ACHD was associated with greater overall reductions in serum creatinine, and in phosphate and plasma urea, and an increased net nutritional intake.

CONCLUSIONS

ACHD provided biochemical and outcome indicator advantages over conventional dialytic therapy. In patients with 2 to 4 failing organs or an intermediate APACHE II score (24 to 29) a significant survival advantage was demonstrated for ACHD over CDT. Although this study is a retrospective analysis, with all the inherent limitations of such studies, it suggests that ACHD is the treatment of choice for ARF in the critically ill, with maximum benefits seen in those with 2 to 4 failing organs and/or intermediate APACHE II scores.

Cost of intensive therapy. A description of methodology and initial results

A preliminary study was performed to calculate the cost of intensive therapy on an individual patient basis. The fixed (equipment, supporting services and land opportunity), semi-fixed (staff) and marginal (treatment) costs of 20 critically ill patients were calculated individually. The results show that there is wide variation in intensive therapy costs. The average daily cost for a spontaneously breathing patient was 399 pounds (95% confidence intervals 388 pounds-460 pounds) while that for a ventilated patient was 726 pounds (656 pounds-795 pounds). The mean total cost per patient was 1980 pounds, but the cost per survivor increased by 16% (347 pounds) because of four deaths on the intensive care unit. High total costs are associated with increased severity of illness and higher marginal (treatment) costs are associated with increased semi-fixed (staff) costs. The cost of intensive therapy was three to five times that for general ward care.

C-reactive protein as a diagnostic test of sepsis in the critically ill

Changes in the plasma concentration of C-reactive protein were assessed as a diagnostic test for sepsis in critically ill patients. Forty-nine episodes of secondary sepsis were identified in 31 patients. In 43 out of the 49 episodes there was a 25% or greater change in the concentration of C-reactive protein on the day that sepsis was diagnosed but in six episodes of sepsis the change was less than 25%. A 25% rise in the plasma concentration of C-reactive protein in the absence of other non-infective causes of a raised C-reactive protein, such as inflammation, tissue injury or surgery, is highly suggestive of infection, but failure of the C-reactive protein to rise does not eliminate a diagnosis of sepsis.

Outcome prediction of acute renal failure in medical intensive care

Data acquired prospectively from 134 patients with acute renal failure requiring dialysis in a medical intensive care unit (ICU) were analysed in order to derive indicators predicting ICU-survival. Mortality in the ICU was 56.7%. Linear discriminant analysis correctly predicted outcome in 79.9% at the start of dialysis, and 84.7% at 48 h after the first dialysis. The most important predictive variables were mechanical ventilation and low blood pressure. On the other hand, the total correct classification rates achieved by a standardised system for scoring ICU-patients (APACHE II) did not exceed 58.2%. It is concluded that outcome prediction by APACHE II and even by the discriminant functions is too inaccurate to become the basis for clinical decisions either concerning the initiation or the continuation of dialysis treatment in ARF.

Are elderly people less responsive to intensive care?

Older patients may be excluded from intensive care units because of the perception that they will benefit less than younger patients. To determine if advanced age is associated with increased mortality independent of severity of illness, we compared older and middle-aged patients admitted to a medical intensive care unit. We reviewed the charts of 130 patients age 75 years or older and 135 patients age 55 to 65 admitted over a 30-month period. We controlled for severity of illness using the Acute Physiology Assessment and Chronic Health Evaluation (APACHE II) system without including points for age (APACHE IIM). The groups were similar with regard to gender, whether or not they had a private attending physician, mean APACHE IIM score, and diagnoses, except that older patients had more chronic obstructive pulmonary disease. Hospital stay was slightly longer in the older group (37 vs. 39 days, rank sum, P less than .02). Hospital mortality was significantly greater in the older group (39% vs. 51%, Chi-square P less than .05) with a crude relative risk of 1.32 (95% confidence interval [CI]: 1.01, 1.73). However, the relation of age group to mortality differed for patients with different diagnoses. When we used logistic regression to adjust for APACHE IIM, whether the patient had a private attending physician, primary admitting diagnosis, or presence of cancer, older patients did not have a significantly greater risk of dying (adjusted relative risk, 1.05; 95% CI: 0.97, 1.12). When pulmonary artery catheterization was added to the model, it independently predicted mortality adjusted relative risk, 1.47; 95% CI: 1.05, 2.06.(ABSTRACT TRUNCATED AT 250 WORDS)

Proposed definitions for diagnosis, severity scoring, stratification, and outcome for trials on intraabdominal infection. Joint Working Party of SIS North America and Europe

Analysis of the experience with scientific studies on patients with secondary intraabdominal infection has revealed that problems of interpretation and comparability between studies exist as they relate to variable diagnostic criteria, unmeasured severity of disease, and unclear outcome measures. A consistent system of definitions has been developed to address these deficiencies. Intraabdominal infection is defined as clinical peritonitis requiring both operative and microbiological confirmation for proof of infection. The APACHE II system is proposed for grading the severity of the infection and for stratification of patient risk of mortality. Mortality and time until death, on one hand, and recovery and time until recovery, on the other, are proposed as the main outcome measures, both being independently and positively defined. It is anticipated that this system of minimum rules will produce studies that can be compared, hence, accelerating knowledge and understanding about intraabdominal infection and its best treatment.

Cell therapy in shock: a reasonable prospect or a lost cause?

Almost any insult to the body can set in motion the sequence of physiologic and biochemical changes leading to shock. In the absence of effective treatment widespread disruption of the body's biological mechanisms inevitably follows. This review will describe some of cellular and subcellular disturbances associated with the shock syndrome, the persistence of which lead to (multiple) organ failure. Current views as to the mechanisms of these disturbances will be outlined with particular reference to hypovolemic and septic shock. Finally, an indication will be given of the relevant therapeutic implications, now and in the future.