Criteria abstracted from The Users' Guide to Medical Literature, from the Health Information Research Unit and Clinical Epidemiology and Biostatistics, McMaster University

Highlighted lines and questions below provide links to the pertinent description of criteria in The EBM User's Guide, now available at the Canadian Centres for Health Evidence

Article Reviewed:

Please visit the new Evidence Based Journal Club Reviews

A 3 level prognostic classification in septic shock based on cortisol levels and cortisol response to corticotropin.

Annane D, Sébille V, Troché G, Raphael JC, Gajdos P, Bellisant E.

JAMA 2000;283:1038-1045. [abstract] [full-text for a limited time]

Reviewed by Xiomara Garcia, MD, Mona McPherson, MD, and Michele Mariscalco, MD. Pediatric Critical Care Section, Department of Pediatrics, Baylor College of Medicine, Houston, Texas

Review posted September 17, 2000

I. What is being studied?

Study objective:

To evaluate the prognostic value of cortisol levels and, a short corticotropin stimulation test, in patients with septic shock.

Study design

A prospective observational cohort design.

All consecutive patients hospitalized in the adult intensive care units of 2 teaching hospitals in France were enrolled if they met the criteria for septic shock. The criteria for septic shock were those accepted by the American College of Chest Physicians/Society of Critical Care Medicine. Patients were not eligible if they had a previous condition that may have disrupted their hypothalamic pituitary axis. The institutional review board approved the protocol and informed consent was obtained from the patient's next of kin.

Outcomes assessed

The main outcome was 28 day mortality as a function of variables collected at the onset of septic shock, including the cortisol levels before the corticotropin test and the cortisol response to corticotropin. A short corticotropin stimulation test was performed by intravenously injecting 0.25 mg of tetracosactrin; blood samples were taken immediately before the test (T0) and 30 (T30) and 60 (T60) minutes afterward. Change was defined as the difference between T0 and the highest value at either T30 or T60.

II. Are the results of the study valid?

Note: These questions follow from Randolph AG et al. Understanding articles describing clinical prediction tools. Crit Care Med 1998;26:1603-1612. [abstract]

1. Was a representative group of patients completely followed up?

Yes. The follow-up was sufficiently long, 28 day mortality is generally accepted as a primary outcome of mortality studies in the ICU. Of the 189 patients enrolled, all were followed and accounted for during the study period. We assume that the enrolled patients are representative of all patients who were eligible during the study period, though we can not be certain of this. The authors do not state how many patients were eligible, how many were excluded, or how many refused to participate.

2. Were all potential predictors included?

Most important predictors seem to be included, although, the authors did not explain why these variables were selected. The potential predictors were as follows: a) age, b) sex, c) surgical or medical admission, d) underlying disease, e) McCabe score, f) severity of illness as assessed by the number organ system failures (OSF), g) SAPS II (simplified acute physiology score II), h) vital signs (temperature, mean arterial pressure, heart rate, urinary output), and i) physician interventions (antibiotics, fluids, inotropes, steroids). Other predictors included laboratory results such as: cultures, site of infection, ABG, serum chemistries, arterial lactate, hematology labs, and cortisol levels before and after a short corticotropin test.

Other variables could have been included such as levels of inflammatory mediators. But in general these predictors seem reasonable.

3. Did the investigators test the independent contribution of each predictor variable?

Yes. The predictors significantly different between the survivors and non-survivors in the univariant analysis included McCabe scores, OSF, ASPS II, and other clinical and laboratory factors as well as cortisol levels. However when entered into a logistic regression model, only 5 factors remained independently associated with death: McCabe score, OSF, lactate, PaO2/FiO2 ratio, cortisol level at T0, and maximum variation in cortisol level after a corticotropin test.

4. Were outcome variables clearly and objectively defined?

Yes, all variables were objective parameters and were clearly defined.

III. What are the results?

1. What is(are) the prediction tool(s)?

A prediction model wasn't provided by this study per se. Clinical prediction rules and models are developed by applying statistical techniques to find combinations of predictors that categorize a heterogeneous group of patients into subgroups of risk. To develop a prediction model, one needs usually two separate groups of patients. One to develop the model, the other to validate the model. (1)

In this study, the authors utilize many of the tools for risk stratification, but ultimately do not develop a prediction model. The authors identified variables associated with mortality in septic shock, but these variables were not arranged into a model that allows the prediction of risk of mortality. Using both multivariate logistic regression and Cox proportional hazards regression analysis they identified independent predictors that if present increased (or decreased) the odds of death. Predictors that increased the odds of death were: McCabe score > 0, Organ system failure score > 2, Lactate > 2.8, baseline cortisol > 26. Predictors that decreased the odds of death were: PaO2/FiO2 > 160 (odds ratio 0.49, CI 95% 0.33-0.74) and maximum variation in cortisol after corticotropin > 8 (decrease the odds ratio to 0.41, CI 95% 0.27-0.63).

A secondary finding in this study was the incidence of adrenal insufficiency as defined by a diminished response to corticotropin (change in cortisol level less than 9 mcg/dl). Using this definition, 54% of patients (95% C.I.: 47-61%) had "occult" adrenal insufficiency. This definition was based upon empiric data derived from septic ICU patients in that patients with this degree of hyporesponsiveness after corticotropin had impaired responsiveness to norepinephrine and a high mortality rate. The authors imply the diagnosis of adrenal insufficiency in the septic patient can NOT be determined using normal values.

2. How well does the model categorize patients into different levels of risk?

The authors attempted to use the baseline cortisol levels and the maximum change in cortisol after the corticotropin stimulation test as predictors of death. The authors developed receiver operating characteristic (ROC) curves to describe the discriminating ability of both these variables (i.e., tools) to predict death. The area under the ROC curves for both tools was less than 0.7 indicating that neither tool had an acceptable discriminating ability to predict death.

However, a multivariate regression model was developed using the threshold values for baseline cortisol and maximal change in cortisol after the stimulation test (see following paragraph) along with other variables identified in the univariate analysis (McCabe score, OSF score, lactate, and PaO2/FiO2). This model was both well-calibrated (Homer-Lemeshow goodness-of-fit test ) and discriminated well between patients who lived and those who died (ROC curve 0.884). Despite this, the authors chose not to fully develop the model into subgroups of risk. To do so, they would have had to include all these risk factors in their tool. They would also have had to validate their tool using an independent group of patients. Instead they used the information from the cortisol levels for prognostication.

Questions 2a and 3 are from the User's Guide on Prognosis Articles...

2a. How large is the likelihood of the outcome event in a specified period of time.

Though the cortisol levels could not be used alone to identify risk groups, they could be useful as prognosticators of outcome. The authors developed three prognostic categories for 28 day mortality based upon baseline cortisol levels and changes in cortisol level after the corticotropin test.

Prognostic categories and 28 day survival likelihood ratio (LR+)

Prognostic Categories	28 days Survival	Survival LR	CI 95%
Good	74%	3.2	2.53 —4.0
Intermediate	37%	0.66	0.46 — 0.94
Poor	18%	0.28	0.27 - 2.0

*LR: likelihood ratio, CI: confidence interval

Prognostic Categories based upon cortisol responses:

• Good: cortisol level < 34 mg/dl and max change > 9 m/dl.
• Intermediate: cortisol level of 34 m /dl and max change < 9 m/dl or > 34 m/dl and max change > 9 m/dl.
• Poor: cortisol level of > 34m /dl and max change < 9 m/dl.

As can be seen from the table the likelihood of survival is quite high in the "good" category. Survival likelihood decreases dramatically in the intermediate group (likelihood ration < 1.0)

Prognostic categories and 28 day mortality likelihood ratio (LR-)

Prognostic Categories	Mortality LR	CI 95%	28 days Mortality
Good	0.57	0.34 — 0.8	26%
Intermediate	1.5	1.15 — 2.2	63%
Poor	1.2	0.95 — 1.53	82%

 

The LR calculations were done by the reviewers, with the following formulas:
LR +: a/(a+c)/b/(b+d) or sensitivity/(1- specificity)
LR -: c/(a+c)/d/(b+d) or (1- sensitivity)/specificity

The LR for a test result compares the likelihood of that result in patients with the disease (in this case outcome) to the likelihood of that result in patients without the disease (or outcome).

LR's are independent of disease incidence, and the importance of the test can be estimated by the size of the LR. An LR higher than 10 or smaller than 1 imply a big change in outcome.

Once again from the tables the likelihood of mortality is much higher in the intermediate and poor prognosis group (LR > 1), and likelihood of mortality is quite low in the "good" category. (LR < 1)

3. How precise are the estimates of likelihood?

The precision for the estimates of likelihood related to mortality is good for the "good prognosis" group, the precision falls dramatically in the "intermediate" and "poor " groups (i.e. confidence intervals approach or cross 1.0).

IV. Will the results help me in caring for my patients?

1. Does the tool maintain its prediction power in a new sample of patients?

The authors did not develop a prediction tool, though they could have using cortisol levels and maximum changes after corticotropin test and the variables in their model. In order to validate the tool, the authors would need to use the tool in a new sample of patients to ensure that it maintains its prediction power.

2. Are your patients similar to those patients used in deriving and validating the tool(s)?

Although this study was done with adults, the causes of septic shock were similar to those seen with pediatric patients. The diagnostic tests, severity of illness scores related to disease, organ failure and laboratory abnormalities used in this study could be applied to a pediatric population.

To date only one pediatric study examined the incidence of adrenal insufficiency associated with septic shock (3). Hatherill et all examined baseline cortisol levels and the response to corticotropin in 33 pediatric patients. Hatherhill defined adrenal insufficiency as a maximum change in cortisol less than 8 mcg/dl, similar to the present study of Annane. Using this definition, 52% of pediatric patients had occult adrenal insufficiency, similar to that of adult patients as demonstrated in this present study . However the death rate in pediatric patients is lower (33% vs 58%). Furthermore there was NOT an increased incidence of death in the pediatric patients with occult adrenal insufficiency compared to the those with an adequate adrenal response (35% vs 31%, p = 0.5). If we re-analyze Annane's (2) adult group of septic patients who are adrenal insufficient compared to the sufficient group (using the same definitions as Hatherhill), mortality is significantly increased (72% vs 46%, p < 0.001). From Hatherhill's study therefore we can see that application of the prognostic tools developed in adults may well NOT be useful in pediatric patients with septic shock.

3. Does the tool improve your clinical decisions?

Again, this study did not provide a tool to predict risk of mortality. They did create a multivariate model based on all the variables they found to be significant. The model was well-calibrated and discriminated well, but the authors did not determine the combination of predictors that categorize a homogenous population. . With the likelihood ratios and the confidence intervals calculated by the reviewers, we can prognosticate who and who may not have a significantly higher chance of survival. It may be useful in adult patients, but its applicability to pediatrics is questionable. It is unclear that the results of this study will lead directly to selecting or avoiding therapy, though it could. However, it may be useful to design future studies in adult patients in particular those who may benefit from steroid therapy.

The results of this study should prompt investigators to re-examine the prognostic factors of acute adrenal insufficiency in pediatric septic shock. As demonstrated in Haverhill's and Ceneviva's (4) studies, mortality rate in pediatric septic shock is much lower than adult despite the high occurrence of adrenal insufficiency. (Haverhill) Is there a use for steroids in pediatric patients with septic shock? The question is unanswered at this time. Does adrenal insufficiency prognosticate for death in pediatric patients? At this point, it does not appear so.

References

1. Randolph AG et al. Understanding articles describing clinical prediction tools. Crit Care Med 1998;26:1603-1612. [abstract]
2. Annane D, Raphael JC, Gajdos P. Steroid replacement in sepsis: an unexplored side of a multifaceted drug class. Crit Care Med. 1996;24(5):899-900. [abstract]
3. Hatherill M, Tibby SM, Hilliard T, Turner C, Murdoch IA. Adrenal insufficiency in septic shock. Arch Dis Child. 1999;80(1):51-5. [abstract]
4. Ceneviva G, Paschall JA, Maffei F, Carcillo JA. Hemodynamic support in fluid-refractory pediatric septic shock. Pediatrics. 1998;102(2):e19. [abstract]

 

Comments

Back to the EB Journal Club Index