Neuromuscular blockade in patients with ARDS: a rapid practice guideline.

The aim of this Intensive Care Medicine Rapid Practice Guideline (ICM-RPG) is to formulate an evidence-based guidance for the use of neuromuscular blocking agents (NMBA) in adults with acute respiratory distress syndrome (ARDS). The panel comprised 20 international clinical experts from 12 countries, and 2 patient representatives. We adhered to the methodology for trustworthy clinical practice guidelines and followed a strict conflict of interest policy. We convened panelists through teleconferences and web-based discussions. Guideline experts from the guidelines in intensive care, development, and evaluation Group provided methodological support. Two content experts provided input and shared their expertise with the panel but did not participate in drafting the final recommendations. We followed the Grading of Recommendations Assessment, Development, and Evaluation approach to assess the certainty of evidence and grade recommendations and suggestions. We used the evidence to decision framework to generate recommendations. The panel provided input on guideline implementation and monitoring, and suggested future research priorities. The overall certainty in the evidence was low. The ICM-RPG panel issued one recommendation and two suggestions regarding the use of NMBAs in adults with ARDS. Current evidence does not support the early routine use of an NMBA infusion in adults with ARDS of any severity. It favours avoiding a continuous infusion of NMBA for patients who are ventilated using a lighter sedation strategy. However, for patients who require deep sedation to facilitate lung protective ventilation or prone positioning, and require neuromuscular blockade, an infusion of an NMBA for 48 h is a reasonable option.

Point-of-care sensors for the management of sepsis.

Point-of-care sensors that enable the fast collection of information relevant to a patient's health state can facilitate improved health access, reduce healthcare costs and improve the quality of healthcare delivery. In the diagnosis of sepsis - defined as a life-threatening organ dysfunction caused by a dysregulated host response to infection, and the leading cause of in-patient death and of hospital readmission in the United States - predicting which infections will lead to life-threatening organ dysfunction and developing specific anti-sepsis treatments remain challenging because of the significant heterogeneity of the host response. Yet the use of point-of-care devices could reduce the time from the onset of a patient's infection to the administration of appropriate therapeutics. In this Perspective, we describe the current state of point-of-care sensors for the diagnosis and monitoring of sepsis, and outline opportunities in the use of these devices to dramatically improve patient care.

Surviving sepsis campaign: international guidelines for management of sepsis and septic shock: 2016

OBJECTIVE: To provide an update to "Surviving Sepsis Campaign Guidelines for Management of Sepsis and Septic Shock: 2012". DESIGN: A consensus committee of 55 international experts representing 25 international organizations was convened. Nominal groups were assembled at key international meetings (for those committee members attending the conference). A formal conflict-of-interest (COI) policy wasdeveloped at the onset of the process and enforced throughout. A stand-alone meeting was held for all panel members in December 2015. Teleconferences and electronic-based discussion among subgroupsand among the entire committee served as an integral part of the development. METHODS: The panel consisted of five sections: hemodynamics, infection, adjunctive therapies, metabolic, and ventilation. Population, intervention, comparison, and outcomes (PICO) questions were reviewed and updated as needed, and evidence profiles were generated. Each subgroup generated a list of questions, searched for best available evidence, and then followed the principles of the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system to assess the quality of evidence from high to very low, and to formulate recommendations as strong or weak, or best practice statement when applicable. RESULTS: The Surviving Sepsis Guideline panel provided 93 statements on early management and resuscitation of patients with sepsis or septic shock. Overall, 32 were strong recommendations, 39 were weak recommendations, and 18 were best-practice statements. No recommendation was provided for four questions. CONCLUSIONS: Substantial agreement exists among a large cohort of international experts regarding many strong recommendations for the best care of patients with sepsis. Although a significant number of aspects of care have relatively weak support, evidence-based recommendations regarding the acute management of sepsis and septic shock are the foundation of improved outcomes for these critically ill patients with high mortality.(AU)

A systematic review and meta-analysis of early goal-directed therapy for septic shock: the ARISE, ProCESS and ProMISe Investigators.

PURPOSE: To determine whether early goal-directed therapy (EGDT) reduces mortality compared with other resuscitation strategies for patients presenting to the emergency department (ED) with septic shock. METHODS: Using a search strategy of PubMed, EmBase and CENTRAL, we selected all relevant randomised clinical trials published from January 2000 to January 2015. We translated non-English papers and contacted authors as necessary. Our primary analysis generated a pooled odds ratio (OR) from a fixed-effect model. Sensitivity analyses explored the effect of including non-ED studies, adjusting for study quality, and conducting a random-effects model. Secondary outcomes included organ support and hospital and ICU length of stay. RESULTS: From 2395 initially eligible abstracts, five randomised clinical trials (n = 4735 patients) met all criteria and generally scored high for quality except for lack of blinding. There was no effect on the primary mortality outcome (EGDT: 23.2% [495/2134] versus control: 22.4% [582/2601]; pooled OR 1.01 [95% CI 0.88-1.16], P = 0.9, with heterogeneity [I(2) = 57%; P = 0.055]). The pooled estimate of 90-day mortality from the three recent multicentre studies (n = 4063) also showed no difference [pooled OR 0.99 (95% CI 0.86-1.15), P = 0.93] with no heterogeneity (I(2) = 0.0%; P = 0.97). EGDT increased vasopressor use (OR 1.25 [95% CI 1.10-1.41]; P < 0.001) and ICU admission [OR 2.19 (95% CI 1.82-2.65); P < 0.001]. Including six non-ED randomised trials increased heterogeneity (I(2) = 71%; P < 0.001) but did not change overall results [pooled OR 0.94 (95% CI 0.82 to 1.07); P = 0.33]. CONCLUSION: EGDT is not superior to usual care for ED patients with septic shock but is associated with increased utilisation of ICU resources.

A review of therapeutic attempts to recruit the microcirculation in patients with sepsis.

Microcirculatory dysfunction is a pivotal element of the pathogenesis of severe sepsis and septic shock. Technological development, including sidestream darkfield videomicroscopy, now allows for bedside assessment of the microcirculation. A number of clinical studies have established the importance of the microcirculation in sepsis. The objective of this review is to discuss human trials that have assessed interventions aimed at improving microcirculatory flow in patients with sepsis.

Surviving sepsis campaign: international guidelines for management of severe sepsis and septic shock, 2012

Objective: To provide an update to the "Surviving Sepsis Campaign Guidelines for Management of Severe Sepsis and Septic Shock," last published in 2008. Design: A consensus committee of 68 international experts representing 30 international organizations was convened. Nominal groups were assembled at key international meetings (for those committee members attending the conference). A formal conflict of interest policy was developed at the onset of the process and enforced throughout. The entire guidelines process was conducted independent of any industry funding. A stand-alone meeting was held for all subgroup heads, co- and vice-chairs, and selected individuals. Teleconferences and electronic-based discussion among subgroups and among the entire committee served as an integral part of the development. Methods: The authors were advised to follow the principles of the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system to guide assessment of quality of evidence from high (A) to very low (D) and to determine the strength of recommendations as strong (1) or weak (2). The potential drawbacks of making strong recommendations in the presence of low-quality evidence were emphasized. Recommendations were classified into three groups: (1) those directly targeting severe sepsis; (2) those targeting general care of the critically ill patient and considered high priority in severe sepsis; and (3) pediatric considerations. Results: Key recommendations and suggestions, listed by category, include: early quantitative resuscitation of the septic patient during the first 6 h after recognition (1C); blood cultures before antibiotic therapy (1C); imaging studies performed promptly to confirm a potential source of infection (UG); administration of broad-spectrum antimicrobials therapy within 1 h of the recognition of septic shock (1B) and severe sepsis without septic shock (1C) as the goal of therapy; reassessment of antimicrobial therapy daily for de-escalation, when appropriate (1B); infection source control with attention to the balance of risks and benefits of the chosen method within 12 h of diagnosis (1C); initial fluid resuscitation with crystalloid (1B) and consideration of the addition of albumin in patients who continue to require substantial amounts of crystalloid to maintain adequate mean arterial pressure (2C) and the avoidance of hetastarch formulations (1B); initial fluid challenge in patients with sepsis-induced tissue hypoperfusion and suspicion of hypovolemia to achieve a minimum of 30 mL/kg of crystalloids (more rapid administration and greater amounts of fluid may be needed in some patients (1C); fluid challenge technique continued as long as hemodynamic improvement is based on either dynamic or static variables (UG); norepinephrine as the first-choice vasopressor to maintain mean arterial pressure ≥65 mmHg (1B); epinephrine when an additional agent is needed to maintain adequate blood pressure (2B); vasopressin (0.03 U/min) can be added to norepinephrine to either raise mean arterial pressure to target or to decrease norepinephrine dose but should not be used as the initial vasopressor (UG); dopamine is not recommended except in highly selected circumstances (2C); dobutamine infusion administered or added to vasopressor in the presence of (a) myocardial dysfunction as suggested by elevated cardiac filling pressures and low cardiac output, or (b) ongoing signs of hypoperfusion despite achieving adequate intravascular volume and adequate mean arterial pressure (1C); avoiding use of intravenous hydrocortisone in adult septic shock patients if adequate fluid resuscitation and vasopressor therapy are able to restore hemodynamic stability (2C); hemoglobin target of 7­9 g/dL in the absence of tissue hypoperfusion, ischemic coronary artery disease, or acute hemorrhage (1B); low tidal volume (1A) and limitation of inspiratory plateau pressure (1B) for acute respiratory distress syndrome (ARDS); application of at least a minimal amount of positive end-expiratory pressure (PEEP) in ARDS (1B); higher rather than lower level of PEEP for patients with sepsis-induced moderate or severe ARDS (2C); recruitment maneuvers in sepsis patients with severe refractory hypoxemia due to ARDS (2C); prone positioning in sepsis-induced ARDS patients with a PaO 2/FiO 2 ratio of ≤100 mm Hg in facilities that have experience with such practices (2C); head-of-bed elevation in mechanically ventilated patients unless contraindicated (1B); a conservative fluid strategy for patients with established ARDS who do not have evidence of tissue hypoperfusion (1C); protocols for weaning and sedation (1A); minimizing use of either intermittent bolus sedation or continuous infusion sedation targeting specific titration endpoints (1B); avoidance of neuromuscular blockers if possible in the septic patient without ARDS (1C); a short course of neuromuscular blocker (no longer than 48 h) for patients with early ARDS and a PaO 2/FI O 2<150 mm Hg (2C); a protocolized approach to blood glucose management commencing insulin dosing when two consecutive blood glucose levels are >180 mg/dL, targeting an upper blood glucose ≤180 mg/dL (1A); equivalency of continuous veno-venous hemofiltration or intermittent hemodialysis (2B); prophylaxis for deep vein thrombosis (1B); use of stress ulcer prophylaxis to prevent upper gastrointestinal bleeding in patients with bleeding risk factors (1B); oral or enteral (if necessary) feedings, as tolerated, rather than either complete fasting or provision of only intravenous glucose within the first 48 h after a diagnosis of severe sepsis/septic shock (2C); and addressing goals of care, including treatment plans and end-of-life planning (as appropriate) (1B), as early as feasible, but within 72 h of intensive care unit admission (2C). Recommendations specific to pediatric severe sepsis include: therapy with face mask oxygen, high flow nasal cannula oxygen, or nasopharyngeal continuous PEEP in the presence of respiratory distress and hypoxemia (2C), use of physical examination therapeutic endpoints such as capillary refill (2C); for septic shock associated with hypovolemia, the use of crystalloids or albumin to deliver a bolus of 20 mL/kg of crystalloids (or albumin equivalent) over 5­10 min (2C); more common use of inotropes and vasodilators for low cardiac output septic shock associated with elevated systemic vascular resistance (2C); and use of hydrocortisone only in children with suspected or proven "absolute"' adrenal insufficiency (2C). Conclusions: Strong agreement existed among a large cohort of international experts regarding many level 1 recommendations for the best care of patients with severe sepsis. Although a significant number of aspects of care have relatively weak support, evidence-based recommendations regarding the acute management of sepsis and septic shock are the foundation of improved outcomes for this important group of critically ill patients.

Acute care practices relevant to quality end-of-life care: a survey of Pennsylvania hospitals.

BACKGROUND: Improving end-of-life care in the hospital is a national priority. PURPOSE: To explore the prevalence and reasons for implementation of hospital-wide and intensive care unit (ICU) practices relevant to quality care in key end-of-life care domains and to discern major structural determinants of practice implementation. DESIGN: Cross-sectional mixed-mode survey of chief nursing officers of Pennsylvania acute care hospitals. RESULTS: The response rate was 74% (129 of 174). The prevalence of hospital and ICU practices ranged from 95% for a hospital-wide formal code policy to 6% for regularly scheduled family meetings with an attending physician in the ICU. Most practices had less than 50% implementation; most were implemented primarily for quality improvement or to keep up with the standard of care. In a multivariable model including hospital structural characteristics, only hospital size independently predicted the presence of one or more hospital initiatives (ethics consult service, OR 6.13, adjusted p = 0.02; private conference room in the ICU for family meetings, OR 4.54, adjusted p<0.001). CONCLUSIONS: There is low penetration of hospital practices relevant to quality end-of-life care in Pennsylvania acute care hospitals. Our results may serve to inform the development of future benchmark goals. It is critical to establish a strong evidence base for the practices most associated with improved end-of-life care outcomes and to develop quality measures for end-of-life care to complement existing hospital quality measures that primarily focus on life extension.

Sociodemographic differences in early access to liver transplantation services.

The question of whether health care inequities occur before patients with end-stage liver disease (ESLD) are waitlisted for transplantation has not previously been assessed. To determine the impact of gender, race and insurance on access to transplantation, we linked Pennsylvania sources of data regarding adult patients discharged from nongovernmental hospitals from 1994 to 2001. We followed the patients through 2003 and linked information to records from five centers responsible for 95% of liver transplants in Pennsylvania during this period. Using multinomial logistic regressions, we estimated probabilities that patients would undergo transplant evaluation, transplant waitlisting and transplantation itself. Of the 144,507 patients in the study, 4361 (3.0%) underwent transplant evaluation. Of those evaluated, 3071 (70.4%) were waitlisted. Of those waitlisted, 1537 (50.0%) received a transplant. Overall, 57,020 (39.5%) died during the study period. Patients were less likely to undergo evaluation, waitlisting and transplantation if they were women, black and lacked commercial insurance (p < 0.001 each). Differences were more pronounced for early stages (evaluation and listing) than for the transplantation stage (in which national oversight and review occur). For early management and treatment decisions of patients with ESLD to be better understood, more comprehensive data concerning referral and listing practices are needed.

Cost effectiveness of drotrecogin alfa (activated) for the treatment of severe sepsis in the United Kingdom.

Drotrecogin alfa (activated) is licensed in Europe for the treatment of severe sepsis in patients with multiple organ failure. We constructed a model to assess the cost effectiveness of drotrecogin alfa (activated) from the perspective of the UK National Health Service when used in adult intensive care units. Patient outcomes from a 28-day international clinical trial (PROWESS) and a subsequent follow-up study (EVBI) were supplemented with UK data. Cost effectiveness was assessed as incremental cost per life year and per quality adjusted life year saved compared to placebo alongside best usual care. Applying the 28-day mortality outcomes of the PROWESS study, the model produced a cost per life year saved of 4608 UK pounds and cost per quality adjusted life year saved of 6679 UK pounds. Equivalent results using actual hospital outcomes were 7625 UK pounds per life year and 11,051 UK pounds per quality adjusted life year. Drotrecogin alfa (activated) appears cost effective in treating severe sepsis in UK intensive care units.

Incidence and definition of sepsis and associated organ dysfunction.

AIMS: To discuss the incidence, outcome and predisposing factors to systemic inflammatory response syndrome (SIRS), sepsis, and multiple organ failure. METHODS: A qualitative review of the literature. RESULTS: Case definitions of sepsis and severe sepsis, though clarified recently, are still arbitrary. It seems, however, that SIRS is not useful in identifying severe sepsis while organ failure has become a cornerstone for this definition. Incidence of severe sepsis appears to be approximately 10% of all ICU admissions, totaling nearly one million cases annually in the U.S. alone, and rising. Mortality associated with these events is still high, especially among ICU patients. Recent studies have been demonstrating an association between a variety of genetic polymorphisms and progression to and dying from sepsis. CONCLUSION: Recently there has been an increasing amount of information enabling characterization of the epidemiology of sepsis, which may help to direct appropriate care in the coming years.

Genetic variation and risk of sepsis.

Sepsis is the leading cause of death in non-coronary intensive care unit patients. Sepsis is caused by the immune response to infection and is manifest by pain, fever and edema as the result of the activation of coagulation and inflammatory responses. In severe cases, sepsis leads to organ dysfunction and failure. Sepsis affects more than 750,000 people each year in the US alone, with a mortality rate of over 35 percent making it one of the leading causes of death in developed countries. In addition many patients that die of other diseases have their hospital courses complicated by sepsis. Most patients with infection do not develop severe sepsis and septic shock and yet those that do have a significantly increased risk of death. Genetic and environmental variables may influence why one patient with infection gets sicker than the next. For example, people may be programmed to respond to infection in different ways; some with aggressive immune responses that may be able to wipe out infection before it manifests itself in physical symptoms, while others may have less aggressive immune systems that allow them to get sick more often. The discovery of various common genetic polymorphisms in genes that control the inflammatory response (e.g. tumor necrosis factor) has lent credence to this hypothesis. Yet discovery of the actual relationship between risks of infection / severe sepsis and individual genotypes will require larger, more rigorously designed studies.

Epidemiology of neonatal respiratory failure in the United States: projections from California and New York.

We wanted to determine the incidence, cost, outcome, and patterns of care for neonates requiring mechanical ventilation (MV) in the United States. Using 1994 state hospital discharge data from California and New York, we conducted an observational study of all neonatal hospitalizations (n = 16,405) with MV, comparing outcomes at centers of different technological capability, and generating national projections using census and natality reports. The MV rate was 18 per 1,000 live births. Although the incidence was much higher in lower birth weight (BW) babies, one-third had normal BW. The incidence was higher in boys (20 versus 15.6 per 1,000) and in blacks (29 per 1,000). Hospital mortality was 11.1%, higher in minority groups, and associated with low BW, congenital anomalies, and major hemorrhage. Mean hospital length of stay and costs were 31.1 d and $51,700. Half of all deaths occurred at lower level centers. There are 80,000 cases per year in the United States with 8,500 deaths and total hospital costs of $4.4 billion. We conclude neonatal respiratory failure is common, expensive, and frequently fatal. There are a surprisingly large number of normal BW cases and there are large racial differences.

Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care.

OBJECTIVE: To determine the incidence, cost, and outcome of severe sepsis in the United States. DESIGN: Observational cohort study. SETTING: All nonfederal hospitals (n = 847) in seven U.S. states. PATIENTS: All patients (n = 192,980) meeting criteria for severe sepsis based on the International Classification of Diseases, Ninth Revision, Clinical Modification. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We linked all 1995 state hospital discharge records (n = 6,621,559) from seven large states with population and hospital data from the U.S. Census, the Centers for Disease Control, the Health Care Financing Administration, and the American Hospital Association. We defined severe sepsis as documented infection and acute organ dysfunction using criteria based on the International Classification of Diseases, Ninth Revision, Clinical Modification. We validated these criteria against prospective clinical and physiologic criteria in a subset of five hospitals. We generated national age- and gender-adjusted estimates of incidence, cost, and outcome. We identified 192,980 cases, yielding national estimates of 751,000 cases (3.0 cases per 1,000 population and 2.26 cases per 100 hospital discharges), of whom 383,000 (51.1%) received intensive care and an additional 130,000 (17.3%) were ventilated in an intermediate care unit or cared for in a coronary care unit. Incidence increased >100-fold with age (0.2/1,000 in children to 26.2/1,000 in those >85 yrs old). Mortality was 28.6%, or 215,000 deaths nationally, and also increased with age, from 10% in children to 38.4% in those >85 yrs old. Women had lower age-specific incidence and mortality, but the difference in mortality was explained by differences in underlying disease and the site of infection. The average costs per case were $22,100, with annual total costs of $16.7 billion nationally. Costs were higher in infants, nonsurvivors, intensive care unit patients, surgical patients, and patients with more organ failure. The incidence was projected to increase by 1.5% per annum. CONCLUSIONS: Severe sepsis is a common, expensive, and frequently fatal condition, with as many deaths annually as those from acute myocardial infarction. It is especially common in the elderly and is likely to increase substantially as the U.S. population ages.

Epidemiology of sepsis: an update.

OBJECTIVE: We review the case definition, occurrence, and outcome of sepsis. We discuss whether the epidemiology of sepsis has changed over time and discuss issues important to our understanding of sepsis. DATA SOURCES: Literature review. DATA SUMMARY: Our understanding of the epidemiology of sepsis is hampered by the lack of a reliable case definition. Inconsistent application of sepsis definition criteria contributes to confusion and variability in the literature. Variability in the time course of sepsis also introduces difficulty. The Centers for Disease Control estimated an incidence of 73.6 per 100,000 population in 1979, rising to 175.9 per 100,000 in 1989. However, this study was of septicemia, not severe sepsis. There are several hospital-based studies of the occurrence of severe sepsis, defined using the American College of Chest Physicians/Society of Critical Care Medicine consensus criteria. These studies reported variable hospital and intensive care unit (ICU) occurrence rates, ranging from 2% to 11% of all hospital or ICU admissions. Most of these data are from academic, tertiary care centers, which limits generalizability. More population-based studies are required to better delineate the incidence and risk factors of sepsis in the general population. Hospital mortality from sepsis has ranged from 25% to 80% over the last few decades. Although mortality may be lower in recent years, sepsis is clearly still a very serious condition. Achieving a better understanding of whether the mortality rate for sepsis is falling, however, is confounded by the lack of a uniform definition. Risk factors for adverse outcome include the degree of physiologic derangement, organ dysfunction, underlying illness, site of infection, and microbiological etiology. We do not know, however, the factors that predict response to new therapies. This dilemma has led researchers to explore whether markers of the inflammatory cascade might be more specific for sepsis, more accurate for risk prediction, or more useful for predicting response to therapy. However, there as yet is no equivalent of the CPK-MB for acute myocardial infarction. Whether we will find such a marker as we develop a greater understanding of the genetic control of the inflammatory cascade is uncertain but promising. One might assume intuitively that the epidemiology of sepsis is changing. For example, the number of patients being treated in ICUs has increased over time, the technologies used in the ICU have changed, and the choice and the use of antibiotics have changed. Predisposing factors, such as chemotherapeutic regimens, have also changed, and there have been marked changes in antibiotic resistance. Furthermore, there have been wide changes in the microbiological etiologies of diseases such as pneumonia and acute exacerbations of chronic bronchitis. However, lacking good case definitions and true incidence studies, we can only make inferences about whether the epidemiology of sepsis is truly changing. CONCLUSION: Many studies have documented many aspects of the epidemiology of sepsis. However, the composite picture they provide, although rich in many aspects, remains incomplete and emphasizes the heterogeneity of the condition. Unfortunately, few population-based prospective cohort studies exist that allow us to accurately delineate the risk factors for sepsis, its course, and its outcome. To place new information, such as the role of genetic predisposition, in the correct context, it is essential that such studies be conducted.

Quality-adjusted survival in the first year after the acute respiratory distress syndrome.

There is little information on long-term outcome after acute respiratory distress syndrome (ARDS). We measured quality-adjusted survival in the first year after ARDS in a prospective cohort (n = 200). All patients met traditional criteria for ARDS. Patients with sepsis and acute nonpulmonary organ dysfunction at presentation were excluded. The cohort was healthy before onset of ARDS as evidenced by high functional status (mean Karnofsky Performance Status index: 82.2/100 where >/= 80 = able to perform normal activities independently) and minimal comorbid illness (mean Charlson-Deyo comorbidity score: 0.32/17 where 0 = absence of chronic illness). We determined quality-adjusted life-years (QALYs) using the Quality of Well-being (QWB) scale (0 to 1 scale where 1 = optimal well-being), measured at 6 and 12 mo. Survival was 69.5 +/- 5.0% at 1 month, fell to 55.7 +/- 3.7% at 6 mo, and did not change at 12 mo, yielding a survival of 59 life-years in the first year per 100 patients with ARDS. QWB was low at 6 and 12 mo (0.59 +/- 0.015 and 0.60 +/- 0.015), yielding a quality-adjusted survival of 36 QALYs per 100 patients (sensitivity range: 21 to 46 QALYs). We conclude that ARDS developing in previously healthy patients is associated with poor quality-adjusted survival. These data are important for cost-effectiveness analyses and long-term care.

Predicting hospital mortality for patients in the intensive care unit: a comparison of artificial neural networks with logistic regression models.

OBJECTIVE: Logistic regression (LR), commonly used for hospital mortality prediction, has limitations. Artificial neural networks (ANNs) have been proposed as an alternative. We compared the performance of these approaches by using stepwise reductions in sample size. DESIGN: Prospective cohort study. SETTING: Seven intensive care units (ICU) at one tertiary care center. PATIENTS: Patients were 1,647 ICU admissions for whom first-day Acute Physiology and Chronic Health Evaluation III variables were collected. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We constructed LR and ANN models on a random set of 1,200 admissions (development set) and used the remaining 447 as the validation set. We repeated model construction on progressively smaller development sets (800, 400, and 200 admissions) and retested on the original validation set (n = 447). For each development set, we constructed models from two LR and two ANN architectures, organizing the independent variables differently. With the 1,200-admission development set, all models had good fit and discrimination on the validation set, where fit was assessed by the Hosmer-Lemeshow C statistic (range, 10.6-15.3; p > or = .05) and standardized mortality ratio (SMR) (range, 0.93 [95% confidence interval, 0.79-1.15] to 1.09 [95% confidence interval, 0.89-1.38]), and discrimination was assessed by the area under the receiver operating characteristic curve (range, 0.80-0.84). As development set sample size decreased, model performance on the validation set deteriorated rapidly, although the ANNs retained marginally better fit at 800 (best C statistic was 26.3 [p = .0009] and 13.1 [p = .11] for the LR and ANN models). Below 800, fit was poor with both approaches, with high C statistics (ranging from 22.8 [p <.004] to 633 [p <.0001]) and highly biased SMRs (seven of the eight models below 800 had SMRs of <0.85, with an upper confidence interval of <1). Discrimination ranged from 0.74 to 0.84 below 800. CONCLUSIONS: When sample size is adequate, LR and ANN models have similar performance. However, development sets of < or = 800 were generally inadequate. This is concerning, given typical sample sizes used for individual ICU mortality prediction.