| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Corresponding author: Karen K. Giuliano, RN, PhD, 3000 Minuteman Rd, MS 500, Andover, MA 01810 (e-mail: Karen.Giuliano{at}philips.com).
 |
Abstract |
|---|
 Top Abstract The Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
• Methods The Project IMPACT data set was used to assess whether the physiological parameters of heart rate, mean arterial pressure, body temperature, and respiratory rate can be used to distinguish between critically ill adult patients with and without sepsis in the first 24 hours of admission to an intensive care unit.
• Results All predictor variables used in the analyses differed significantly between patients with sepsis and patients without sepsis. However, only 2 of the predictor variables, mean arterial pressure and high temperature, were independently associated with sepsis. In addition, the temperature mean for hypothermia was significantly lower in patients without sepsis. The odds ratio for having sepsis was 2.126 for patients with a temperature of 38°C or higher, 3.874 for patients with a mean arterial blood pressure of less than 70 mm Hg, and 4.63 times greater for patients who had both of these conditions.
• Conclusions The results support the use of some of the guidelines of the Surviving Sepsis Campaign. However, the lowest mean temperature was significantly less for patients without sepsis than for patients with sepsis, a finding that calls into question the clinical usefulness of using hypothermia as an early predictor of sepsis. Alone the group of variables used is not sufficient for discriminating between critically ill patients with and without sepsis.
|
Notice to CE enrollees: A closed-book, multiple-choice examination following this article tests your understanding of the following objectives:
To read this article and take the CE test online, visit www.ajcconline.org and click "CE Articles in This Issue."
|
Since their introduction into the literature, the term and concept of SIRS have been widely adopted by clinicians and investigators.1–4 The physiological criteria associated with SIRS and sepsis are listed in Table 1
. An international consensus panel that was convened in 2001 to review the 1992 definitions essentially reconfirmed the definitions.4
|
In 2004 an updated version of recommended clinical practice parameters was published.7 In this publication, the continuous physiological monitoring parameters identified for early sepsis screening continued to center on hypoperfusion and expanded the clinical definitions of hypoperfusion to include systolic blood pressure less than 90 mm Hg, MAP less than 65 mm Hg, a decrease in systolic blood pressure of more than 40 mm Hg, and a decrease in urine output.7
Other parameters for patients with sepsis that have been advocated and can be monitored on a continuous basis include body temperature, heart rate, respiratory rate, cardiac output, systemic vascular resistance, oxygen saturation, mixed venous oxygen saturation, right atrial venous oxygen saturation, urine output, and right ventricular filling pressure or volume.3–5,8
| Maintaining adequate mean arterial pressure is essential for managing septic shock.
|
|
The Surviving Sepsis Campaign |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
Because successful implementation of the bundles hinges largely on the early detection of sepsis, empiric data clearly are needed to support the usefulness of the clinical monitoring parameters that are being advocated as part of the SSC. The purpose of this study was to develop a predictive model for the early detection of sepsis that involves use of physiological parameters advocated for early screening by the SSC. This approach may provide further guidance to clinicians on how best to use the current sepsis criteria and clinical practice parameters to facilitate the earliest and most accurate diagnosis of sepsis in the clinical setting. The research question for the study was as follows:
On the basis of measurements obtained in the first 24 hours of admission to an intensive care unit (ICU), can a combination of the physiological parameters of heart rate, MAP, body temperature, and respiratory rate be used to distinguish between critically ill adult patients with and without sepsis?
|
Methods |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
Project IMPACT Data Set
The Project IMPACT data set, an international data set maintained in conjunction with the Society of Critical Care Medicine, was created for critical care practitioners in a multiyear effort by nearly 100 multidisciplinary critical care experts of the society. The main reason for its creation was to meet the data, information, and research needs of critically ill patients and their health-care providers. The physiological data from the Project IMPACT data set required to answer the research question included admission diagnosis, the Simplified Acute Physiology Score (SAPS) II, core body temperature, respiratory rate, and MAP. The highest and lowest values for each of these variables were collected for the first 24 hours of ICU admission.
| The Surviving Sepsis Campaign expects to reduce sepsis mortality significantly in 5 years.
|
Patients
Patients whose data are included in the Project IMPACT data set are adults (at least 18 years old). For this study, an effect size of 0.4 was used in the calculations for the power analysis. According to Cohen’s definitions of power, a value of 0.4 indicates a moderate to large effect.11 On the basis of a 2-tailed test with a power of 80%, an effect size of 0.4, and a .05 level of significance, the appropriate sample size per group for this study was 360.
Once the data were received, they were searched for a sample of all patients with an admitting ICU diagnosis of sepsis (International Classification of Diseases, Ninth Revision codes 995.91 and 995.92). This step resulted in a group of 380 patients. SAPS II matching and a stratified random sampling procedure were used to create a comparison group of patients without sepsis matched for acuity.
Data Cleaning and Preparation
Descriptive statistics were generated for the entire sample of both groups of patients. Medians, means, and SDs were computed for all continuous data; frequency counts and percentages were computed for all categorical data. The variables were first checked for missing data. Because no single variable had more than 5% missing data, no data substitution methods or procedures were necessary.
Next, the data were manually inspected with regard to the ranges and data distribution for each variable. This inspection revealed 2 patients in the sample who had both highest and lowest blood pressure and heart rates of zero and ICU stays of only 0.1 day. On the basis of these data, these patients most likely were dead on arrival at the ICU and therefore were not included in the sample. A decision was also made at that time to include only patients with a minimum ICU stay of 0.4 days (9.6 hours). This step was taken to ensure that enough physiological data would be generated by these patients to reasonably contribute to the final model. The final sample consisted of 364 patients in the no sepsis group and 363 patients in the sepsis group (total sample=727).
The data were then examined for skewness by using the Fisher measure of skewness, calculated by dividing the measure of skewness by the SE of the skew.12 With this formula all values ±1.96 are considered significantly skewed because 95% of the scores in a normal distribution fall between ±1.96 SD from the mean. In this data set, all variables with the exception of the lowest respiratory rate in the first 24 hours of ICU admission were significantly skewed.
|
Results |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
) and a t test was run on the single continuous variable (lowest respiratory rate in 24 hours) that was not significantly skewed (Table 3). A 
2 analysis was used to make group comparisons for the categorical variables (Table 4). Because multiple comparisons were being made, the Bonferroni correction test was applied to all continuous variables. As a result, for the Mann-Whitney U, a result of P = .004 (.05/13) or less was considered significant, and for the t test comparisons a result of P = .025 (.05/2) or less was considered significant. For descriptive purposes only, means and SDs were reported for all variables, even though these calculations were not used as part of the Mann-Whitney analysis.
|
|
|
. According to Munro,12 any correlation of 0.25 or less represents little if any relationship, and values of 0.26 to 0.49 represent a low correlation. As seen in Table 5, all of the correlation values fall into these 2 ranges. Therefore, most likely no correlation was high enough to have had an effect on the final logistic regression model.
|
| A low mean arterial pressure in the first 24 hours after admission to the intensive care unit increases sepsis risk by almost 4-fold; a high temperature, by 2-fold.
|
Logistic Regression
Predictors, or the independent variables for a logistic regression, can be at any level of measurement. In binary logistic regression, the dependent variable is categorical and dichotomous, and in this study the dichotomous outcome variable was sepsis or no sepsis. The first assumption of logistic regression is that the outcome groups are mutually exclusive. Because the SSC guidelines for monitoring in sepsis use cutoff values for all the physiological predictors, before the logistic regression was run all predictor variables were recoded into dichotomous variables by using the currently recommended cutoff values4,8 (Table 6).
|
The classification table (Table 7) based on this model indicates that the correct classification was much higher for sepsis (sensitivity 78.9%) than for no sepsis (specificity 45.1%). Sensitivity is an estimate of the probability that an instrument can correctly indicate those individuals who have a particular attribute; in this instance, patients with sepsis. Conversely, specificity is an estimate of the probability that an instrument will exclude those individuals who do not have a particular attribute; in this instance, patients without sepsis.13
|
) were significantly and independently associated with sepsis: MAP (P < .001) and high temperature (P < .001). Patients who in the first 24 hours of ICU admission had a MAP of 69 mm Hg or less and a temperature of 38°C or higher were more likely to have sepsis than were those who did not. Respiratory rate (P =. 21) and heart rate (P = .43) did not differ significantly between the 2 groups.
|
The previous interpretation of the odds ratios highlighted the odds of having sepsis according to each of the 2 individually significant predictors of temperature greater than 38°C and MAP less than 70 mm Hg. However, to calculate the odds of a patient having sepsis if both conditions (temperature > 38°C and MAP < 70 mm Hg) occurred simultaneously, additional calculations were necessary.12 According to these calculations, if a patient had both a temperature of 38°C or greater and a MAP less than 70 mm Hg, the odds of having sepsis were 4.63 times higher than if the patient had a temperature less than 38°C and a MAP of 70 mm Hg or greater.
| Based on blood pressure and temperature, almost 80% of septic patients can be correctly identified.
|
|
Discussion |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
The Cox and Snell R2 (0.102) and the Nagelkerke R2 (0.136) indicated that the model only explained between 10.2% and 13.6% of the variance. Whereas the significant differences between patients with sepsis and those without support the use of all of the variables in the final logistic regression model, the R2 results indicate that some other significant variables were probably excluded from the final model. This notion was well understood at the onset of the study.
The purpose of the study was not to include every relevant predictor of sepsis but to separate and assess the clinical predictive value of the physiological monitoring parameters advocated by the SSC. Additional physiological measurements (eg, particularly global and microperfusion measures) as well as laboratory values (eg, blood cultures, serum levels of lactate) most likely would be even more useful in the diagnosis of sepsis. However, in the real clinical setting, clinicians require some level of suspicion for sepsis before these types of tests are performed. Hence, the purpose of this study was to assess the value of the more common and routine critical care monitoring parameters in raising a clinician’s level of suspicion for sepsis, so that more definitive ways to test and diagnose can then be ordered earlier in the course of the pathophysiological process of sepsis.
The problem with this predictive model and these associated predictive measures is that although most patients with sepsis will have positive results in screens for sepsis (true-positives), a fairly high number of patients without sepsis who are equally ill also will have positive results (false-positives). Nonetheless, the clinical implications of this classification finding are more positive than negative. For clinicians working in the ICU, resource allocation is always an issue. If such allocation were not an issue, every ICU patient could have every conceivable test for sepsis. If each patient did have all these tests, most likely sepsis would be correctly detected in most patients who had it and incorrectly diagnosed in no or very few patients who did not have it. Because treating patients with sepsis is costly, time sensitive, and resource intensive, it is probably most prudent to miss the fewest number of patients for early screening, because with a few additional diagnostic tests the patients with false-positive results can be quickly screened out.
When the cost of some additional inexpensive diagnostic tests (such as serum levels of lactate) for patients is weighed against the cost of missing large numbers of patients who truly have sepsis and could benefit tremendously from the earliest possible treatment, this classification table has some clinical usefulness. With only 2 simple and inexpensive physiological measures, the current predictive model can be used to correctly detect sepsis in approximately 80% of patients who have it. Therefore, this predictor set of clinical monitoring parameters supports some of the current recommendations, may actually simplify those current recommendations, and thus has some real clinical value.
One interesting finding was the usefulness of hypothermia as a potential predictor of sepsis. Despite a lack of published empirical evidence, current practice protocols associate hypothermia with sepsis. A temperature less than 38°C is most often used as the cutoff value.4,5 With this criterion, it was expected that the lowest temperature in the first 24 hours of ICU admission would be lower in the group with sepsis than in the group without sepsis. However, the converse was found. Patients without sepsis (mean temperature 36.11°C, SD 0.88) had a significantly lower mean 24-hour temperature than did patients with sepsis (mean temperature 36.31°C, SD 0.96). Specifically, more patients without sepsis (n = 123) than patients with sepsis (n = 95) had a temperature less than 36°C. Therefore, in this sample of patients, hypothermia was not related to sepsis and for that reason was not used in the final logistic regression analysis.
Hypothermia defined as a temperature less than 36°C is currently part of the SSC screening criteria for the early detection of sepsis. However, because the criterion of hypothermia is based primarily on theoretical evidence and expert consensus, these findings suggest that hypothermia may not be valid as an accurate and early predictor of sepsis. Most likely hypothermia in sepsis is a late sign and represents the body’s inability to continue to mount an appropriate thermal response to infection. If this notion is correct, hypothermia does not belong in the SSC list of screening criteria intended to detect the onset of sepsis.
| Hypothermia may not be related to sepsis, though current practice protocols often include it.
|
|
Limitations |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
Another potential problem with the use of large extant data sets is possible underreporting of patients with a primary diagnosis of sepsis. Because the starting data set included information on approximately 120 000 patients at the time the data for this study were obtained, the number of patients with an admitting diagnosis of sepsis that could have been used in the final analysis was expected to be much higher than the 363 achieved (<0.003%). This expectation was based on the projection by Angus et al14 of 750 000 new cases annually in the United States.
Another recent document15 based on Medicare data and the newly established patient safety indicator software of the Agency for Healthcare Research and Quality cited sepsis as the fourth most common principal diagnosis in hospitalized patients. In the Agency for Healthcare Research and Quality sample, the incidence of sepsis for at-risk hospitalized patients was 1.2 per 1000 cases. In contrast, the overall percentage of critical care patients with sepsis as a principal diagnosis reported in the data set of the Healthcare Cost and Utilization Project for the years 1993 to 2002 was only 0.0002%; for sepsis included as any one of the diagnoses, it was only slightly higher at 0.0008%.16
These data suggest that (perhaps because of the International Classification of Diseases, Ninth Revision coding structure as well as the relatively new codes used to identify this sample of patients) the incidence and prevalence of sepsis may be underreported in some large national databases or may be buried in some other diagnostic categories such as pneumonia and urinary tract or other types of infectious processes. Because the Project IMPACT data set includes information on a large number of postoperative patients (who would generally not be admitted to an ICU with sepsis), this characteristic also could have had an effect on the expected number of patients with sepsis in the sample used for this study.
Another consideration is the study design. The rationale for choosing patients with an admitting diagnosis of sepsis was that this procedure would be the best way to ensure selection of patients in the early stage of sepsis. This method also would be likely to eliminate patients who did not have sepsis, because only measurements from the first 24 hours of ICU admission were used. However, because this study was a secondary analysis, no method was available to accurately measure the stage of sepsis. Patients whose admission diagnosis was misclassified or in whom sepsis developed within the first 24 hours of ICU admission could have been inadvertently included in the no sepsis group of patients. However, since the group of patients without sepsis was drawn by stratified random sampling from approximately 100 000 patients, most likely large numbers of patients with sepsis were not inadvertently classified as not having sepsis.
|
Implications for Clinical Practice |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
|
Recommendations for Future Research |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
A logical next step would be to use continuously collected monitoring data and data mining techniques to assess more comprehensively the value of the currently recommended clinical monitoring parameters in the detection of critically ill patients with sepsis and to better identify the role of blood pressure in sepsis. Data mining, defined as the semiautomatic exploration and analysis of large quantities of data to discover meaningful patterns and rules, is a newer approach to data analysis that can be used to derive meaningful information from massive amounts of data.17 Because current patient monitoring systems, coupled with bedside clinical information systems, can be used to collect and store massive amounts of clinical data, data mining of these systems would seem an obvious choice for assessing the effectiveness of current clinical practices in the care of critically ill patients with sepsis.
|
ACKNOWLEDGMENTS |
|---|
FINANCIAL DISCLOSURE
Funding was provided, in part, by an American Association of Critical-Care Nurses clinical practice grant.
To purchase reprints, contact The InnoVision Group, 101 Columbia, Aliso Viejo, CA 92656. Phone, (800) 809-2273 or (949) 362-2050 (ext 532); fax, (949) 362-2049; e-mail, reprints{at}aacn.org.
|
REFERENCES |
|---|
|
TopAbstractThe Surviving Sepsis CampaignMethodsResultsDiscussionLimitationsImplications for Clinical...Recommendations for Future...References |
|---|
This article has been cited by other articles:
|
|
 |
|
  R. Sincic and K. K. Giuliano Reader Affirms Association Between Hypothermia and Sepsis Am. J. Crit. Care., July 1, 2007; 16(4): 332 - 333. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
