JUNE 2011 CRITICAL CARE JOURNAL CLUB

Reference as : Robbins RA, Singarajah CU. June 2011 Critical Club Journal Club. Southwest J Pulm Crit Care 2011;2:77-78. (Click here for PDF version )

This month’s Critical Care Journal club focused on two recent articles regarding nighttime coverage of the intensive care unit (ICU). Because of the commitment of Banner Good Samaritan to the electronic ICU (eICU), Dr. Raschke and his colleagues did not feel comfortable reviewing these articles. Therefore, Drs. Robbins and Singarajah are filling in for this month. Both practice at the Phoenix VA where nighttime coverage is neither by electronic nor on-site intensivists, but by second or third year medicine house officers with pulmonary/critical care fellow backup and medical attending physician backup.

Hospital mortality, length of stay, and preventable complications among critically ill patients before and after tele-ICU reengineering of critical care processes. JAMA 2011;305: 2175-83. (Click here for the abstract of the manuscript).

This article by Lilly et al. examined the effect of an adult eICU intervention on hospital mortality, length of stay, best practice adherence, and preventable complications from a compared to care performed by house officers at night during a preintervention period. The authors used a prospective stepped-wedge clinical practice study of 6290 adults admitted to any of 7 ICUs (3 medical, 3 surgical, and 1 mixed cardiovascular). The hospital mortality rate was 13.6% during the preintervention period compared with 11.8% during the eICU intervention period (p=0.005). The tele-ICU intervention period compared with the preintervention period was associated with higher rates of best clinical practice adherence for the prevention of deep vein thrombosis, prevention of stress ulcers, cardiovascular protection, prevention of ventilator-associated pneumonia and catheter-related bloodstream infection. Furthermore, use of the eICU shortened hospital length of stay (9.8 vs 13.3 days, p<0.001).

This article supports the use of eICU but has a number of caveats.  First, the comparison was a before and after comparison which often favor the intervention.  This may be due to other interventions which occurred concomitantly with the eICU implementation or the Hawthorne effect (observation alone tends to induce change). Second, the study was at a single institution and might not be applicable to other institutions. Third, the comparison was between eICU and night coverage with a house officer. It is not surprising that the outcomes might be better with a more experienced physician caring for the patient. The authors did not perform a comparison between an in-house physician intensivist and an eICU. Fourth, some details are missing. The educational level of the house officer is not stated. A first year house officer might not perform as well as sixth year senior pulmonary/critical care fellow. The number of patients being covered by the house officer or eICU physician is not stated. If the house officer was also covering floor patients or the eICU physician was covering only a few patients, this workload could make a difference. Clearly, large, multi-institutional, randomized, studies comparing the eICU with a beside intensivist are needed.

Economic implications of nighttime attending intensivist coverage in a medical intensive care unit. Crit Care Med 2011;39:1257–1262. (Click here for the abstract of the manuscript).

The article by Banerjee et al. assessed the cost implications of changing the intensive care unit staffing model from on-demand presence to mandatory 24 hour in-house critical care specialist presence. Like Lily’s study discussed above this was a pre-post comparison in a single academic center. Total cost estimates of hospitalization were calculated for each patient and adjusted mean total cost estimates were 61% lower after implementation of a 24 hour in-house critical care specialist for patients admitted during night hours (7 PM to 7 AM) who were in the highest Acute Physiology and Chronic Health Evaluation III quartile. The unadjusted intensive care unit length of stay fell in the post period relative to the pre period (3.5 vs. 4.8) with no change in non-intensive care unit length of stay on mortality.

Although virtually all the same criticisms of Lily’s study are also true in this study, both studies have the theme that care of the patient by the most experienced physician, whether in person or electronically, leads to improved outcomes. Future studies should address the various types of ICU coverage available – just house officers in internal medicine; house officers with critical care/pulmonary fellows; or either of the prior two with an in house attending. There are a variety of staffing models for attending coverage with critical care medicine attending physicians from internal medicine, surgery and anesthesiology available for coverage. Comparing in house coverage of a medical ICU by say surgery attending physicians as compared to medical attending physicians may be useful as well. One interpretation of the Bannerjee study is that it the change in model led to a more “efficient” way to achieve the same mortality. The eICU may be more of an administrative method to employ relatively limited resources (CCM attending physicians) more cost effectively, cost effectively that is, from the perspective of the payers and not necessarily for the physicians.

Several points remain unclear: does the physician need to be at bedside; or what staffing levels need to be present? For example, can one physician covering 200 patients electronically produce results as good as one physician covering 20 at the bedside? It is also unclear whether a physician monitoring all the patients is really necessary or can a physician extender such as a nurse practioner or physicians assistant monitor the patients electronically and have the physician become involved either electronically or in person when intervention is appropriate. Lastly, it should be pointed out that although the concept of electronic coverage has been applied to the intensive care unit because of potential staffing shortages, it may also be applied to other areas of the hospital. This includes not only physicians but also nursing and administration where it is unclear whether the presence of the nurse or administrator needs to be in the hospital or can be at a remote location.

April 2011 Critical Care Journal Club

Reference as : Raschke RA. April 2011 Critical Club Journal Club. Southwest J Pulm Crit Care 2011;2:59-64. (Click here for PDF version )

This month’s Journal club focused on a couple of recent articles from the trauma literature.   Even though we don’t often have primary responsibility for the care of trauma patients, it’s important for the fellows to know about major advances in the field.  It’s also useful to consider how the trauma intensivists might deal differently than we would in areas of shared interest, such as hemorrhagic shock, and prevention of VAP.  Corey Detlefs MD – a senior member of our trauma team – was kind enough to join us to help lend perspective to our discussion.  Dr. Bajo, our senior intensivist, Dick Gerkin MD, our statistician, and Dr. Robbins all participated.    

The CRASH-2 Investigators.  The importance of early treatment with tranexamic acid in bleeding patients: an exploratory analysis of the CRASH-2 randomized controlled trial.  Lancet 2011; 377:1096-101.  (Click here for the abstract of the manuscript)

A difficult study for us to review.  Tranexamic acid is an inhibitor of fibrinolysis, and the original CRASH-2 study was a multicenter RCT with over 20,000 patients that showed a significant reduction in all cause mortality (RR 0.91 p=0.0035) when tranexamic acid was given to trauma patients who were bleeding or at risk for bleeding.  This study we reviewed is a re-analysis of the CRASH-2 data to determine if the observed benefit was due to a decrease in hemorrhagic death.  This makes sense since the original hypothesis was that tranexamic acid might prevent bleeding.  What confused us right from the start is that we apparently are not using tranexamic acid in our trauma patients – and we weren’t sure whether this was because the original study just hasn’t impacted bedside practice yet, or whether there was some reason to reject the conclusion of the original study based on critical appraisal. 

The study is quite impressive at face value.  It was conducted in 274 hospitals in 40 countries, and randomized 20,211 patients to tranexamic acid vs. placebo.  This did not seem like a data-dredging expedition - It makes sense to perform a re-analysis of death due to hemorrhage.  A subgroup analysis (that was planned a-priori) also made sense.  This examined the relationship between delay in treatment and benefit. 

The method of the original study was not fully reiterated in this re-analysis, but the general method and statistics seem appropriate.  The study showed a reduction in the risk of bleeding death (RR 0.85, 95% CI 0.76-0.96, p=0.0077).  The study showed strong evidence that this benefit depended on how rapidly the treatment was given p<0.0001, with the best benefit in patients who received it within 1 hour of injury (RR 0.68, 95% CI 0.57-0.82, p<0.0001).  These findings are not only highly statistically significant, but they are biologically plausible.  The main problem with the results is that the effect size is extremely small.  The absolute reduction in bleeding death observed in the treatment group is only 0.8%, so the number needed to treat is 125.  Thus, although statistically significant, the potential clinical benefit for an individual patient is quite small - more on this later, in our review of statistics for the month.

Roquilly A et al.  Hydrocortisone therapy for patients with multiple trauma.  JAMA 2011; 305:1201-9. (Click here for abstract of manuscript)

This was a randomized controlled trial that enrolled 150 trauma patients from 7 centers in France.  Patients were assigned to receive hydrocortisone infusions for a week, versus placebo infusions.  The main outcome measure was the incidence of hospital acquired pneumonia within 28 days.

Leaving out the unnecessary “modified” analysis, the study showed that steroids reduced hospital acquired pneumonia by about half (p=0.007), reduced ventilator length-of-stay (LOS) by 4 days (p=0.001), and reduced ICU LOS by 6 days (p=0.03).

But the study has some problems.  The authors provide some background for why they thought steroids might prevent pneumonia, but the biological plausibility seems a stretch.  Approximately 80% of potentially eligible patients failed to meet stringent inclusion/exclusion criteria, narrowing the generalizability of the study.  The diagnosis of hospital-based pneumonia is based on individual criteria that lack specificity – theoretically, even if the only action hydrocortisone produced was amelioration of fever, the apparent rate of pneumonia would decline, since fever is one of criteria for diagnosis of pneumonia.

But perhaps the strongest reason to doubt the conclusions of this study is historical.  Dozens of studies, over decades, have examined possible therapeutic effects of steroids in critically-ill patients.  Early enthusiasm is almost always followed by disappointment.  This study cannot be appraised without considering this history.  It would indeed be surprising if further studies unequivocally confirm these findings. 

------------------------------------------------------------------------------------

We had the opportunity at this point to pick Dr. Detlef’s brain regarding resuscitation in the trauma ICU, and I’m going to depart from discussion of specific articles for a moment to recall some of his comments.  Most came forth during our discussion of the CRASH-2 study.  One of the fellows pointed out that hemorrhagic shock in the trauma ICU is not all that different than hemorrhagic shock in the medical unit.  With that in mind, Dr. Detlef’s comments bear consideration.

One of the current concepts in trauma resuscitation came from observational data gathered by the military in Iraq and Afghanistan in regards to administration of whole blood.  Whole blood is generally not available in most stateside institutions, but the trauma intensivists have increasingly favored the concept of “whole blood equivalent” when transfusing blood products to patients with life-threatening hemorrhage.  This entails giving a 1:1:1 ratio of packed red blood cells, fresh frozen plasma, and single donor platelets. 

“Permissive hypotension” is another interesting concept in the trauma literature related to hemorrhagic shock.  This strategy involved intentional under-resuscitation – only giving the minimal amount of IV fluids and pressors necessary to maintain life until the source of bleeding is surgically controlled.  This is based on the concepts that early normotension may increase locally-uncontrolled bleeding, and that massive crystalloid administration might have deleterious effects.

These concepts are not yet supported by strong experimental data.  They should be considered as interesting hypotheses that might lead to future research.  Thanks Dr. Detlefs. 

-------------------------------------------------------------------------------------

Back to the articles:  Brief mentions:

Tierney LM et al.  Case 10-2011: A woman with fever, confusion, liver failure, anemia and thrombocytopenia.  New Engl J Med 2011;364:1259-70. (Click here for abstract of the CPC). This CPC was included to point out that secondary hemophagocytic lymphohistiocytosis (HLH) can present as SIRS / severe sepsis / septic shock in adults.  The diagnosis should be suspected when such a patient fails to improve with antibiotics, and multisystem organ failure with bi- or pancytopenia develops.  Bone marrow aspiration is often key to the diagnosis.  Secondary HLH is likely an under-appreciated clinical syndrome.  We have recognized several cases in the past two years. 

Dallas J et al.  Ventilator- in a mixed surgical and medical ICU population.  Chest  2011;139:513-8. (Click here for abstract of manuscript) I thought the concept of ventilator-associated tracheobronchitis (VAT) was interesting.  Essentially, the diagnostic criteria are the same as for VAP, except without lung infiltrates.  From a bedside perspective, this distinction seems impractical.  Few of our ICU patients have clear chest x-rays.  Fleeting infiltrates of unclear significance are commonly seen, and may be attributable to non-infectious processes in many cases.  The author’s conclusion that VAT should be treated with antibiotics may turn out to be proven someday, but is entirely unsupported by their data. 

Levitan RM, et al.  The complexities of tracheal intubation with direct laryngoscopy and alternative intubation devices.  Ann Emerg Med 2011;57:240-47. (Click here for abstract of manuscript) This article is worth reviewing if you are considering purchasing a laryngoscope with advanced optics, like a Glidescope® or C-mac®.   It reviews theoretical and clinical pros and cons of each of the devices.  The gist of the literature in this regard seems to be that these devices should be made available in institutions where physicians might be called upon to manage difficult airways, and that intensivists should pursue experience in their use. 

----------------------------------------------------------------------

Biostatistics comments:

Number needed to treat (NNT)

From: The CRASH-2 Investigators.  The importance of early treatment with tranexamic acid in bleeding patients: an exploratory analysis of the CRASH-2 randomized controlled trial.  Lancet  2011;377:1096-101. 

The NNT is a statistic used to express effect size.  It expresses the number of patients that would have to undergo a therapy in order to prevent a single bad outcome.   It is especially useful in very large clinical trials (like the CRASH-2 study; N = 20,211). Very large clinical trials tend to have enormous power.  Power is the ability to statistically demonstrate a difference when it really exists. This sounds like a good thing, and it usually is.  However, the power of a study is largely dependent on the effect size and sample size.  Thus, when a study has lots of power because of a huge sample size, it may yield a significant p value, even when the effect size is very small. This is why pharmaceutical companies invest millions of dollars to achieve a huge sample size (as in CRASH-2), when they suspect the benefit of their drug is very small.  In such cases, the statistical and clinical significance should be considered. 

NNT is mathematically defined as 1/(absolute effect size).  In CRASH-2, the absolute effect size is the difference in bleeding rates between the treatment and control groups (5.7% - 4.9% = 0.8% = 0.008).    1/0.008 = 125.  Thus the NNT is 125.  You have to give tranexamic acid to 125 patients in order to save one bleeding death.    

Robert A. Rashke MD, Critical Care Journal Club Editor

March 2011 Critical Care Journal Club

Reference as: Raschke RA. March 2011 Critical Care Journal Club. Southwest J Pulm Crit Care 2011;2:46-53. (Click here for PDF version)

I’ve made the executive decision this month to only fully review the three best papers from Journal club.  These three were clearly a notch above the others, and provide plenty of food for thought.  I include a brief discussion of “Bland-Altman” plots and the significance of the R-squared statistic in linear regression.  These statistical issues were raised by the less-than-exemplary papers on Flo-trac cardiac output monitoring, and fluid resuscitation.  I want to begin to review at least one or two biostatistic/clinical epidemiology issues at each Journal club.  Bland-Altman plots seem a rather – well – “bland” topic.  But actually, once Dr. Richard Gerkin explained them, they made perfect sense (see below).

The three best articles: 

Swindell JS, et al.  Shaping patient’s decisions.  Chest.  2011;139:424.  (Click here for abstract).

 A very good discussion on the concept of autonomy, especially when the accompanying, somewhat contradictory editorial is also considered.  Swindell’s main points are that the ethical concept of autonomy is difficult to optimally apply in the ICU, and the physician’s advice should be considered an ethically-valid addition in decision-making.True autonomy requires more from a patient than simply making a decision.  The decision should be intentional - designed to carry out a specific plan, not simply based on intuition, impulse or habit.  Autonomy should be founded on an understanding of the consequences of the plan, and its alternatives, and it should be free from undue influences (external and internal).  These conditions are difficult, or even impossible for a patient to achieve in the ICU. 

Studies show that patients generally want the physician to give them advice, and that this desire increases as they become sicker.  It is also the physician’s responsibility to consider other, possibly conflicting, ethical concerns.  These might include issues of beneficence and social justice.   

This paper was not meant to refute the primacy of patient autonomy, but to bring out it’s limitations in the ICU setting.   Intensivists should do their best to support sound decisions based on autonomy.   But an appropriate degree of paternalism is likely to benefit many patients and their families.    Personally, I feel it is the responsibility of the intensivist to offer the patient and family firm advice when they are struggling with decisions in the midst of critical illness.  When properly accomplished, this can improve the soundness of the decision, and help take some of the emotional burden of the consequences off the patients/family’s shoulders. 

Waterer et al.  Management of community-acquired pneumonia in adults. Am J Respir Crit Care Med.  2011;183:157. (Click here for abstract).

 A good update on CAP.  Several key points:

Up to 50% of ICU admissions for CAP were initially admitted to the floor, and these patients suffer a higher mortality than those admitted directly to the unit.  Therefore clinical scoring tools to predict mortality risk are theoretically beneficial if they could help clinicians identify high risk patients that don’t obviously require ICU care.  There are a number of different scoring systems being tested.  The IDSA/ATS minor criteria (outlined below) seem to perform fairly well, but no scoring system to date has prospectively led to improved patient outcomes.   

“IDSA/ATS minor criteria” for severity of CAP.   ( 3 or more criteria predict severe illness warranting ICU admission) Respiratory rate ≥30 breaths/min PaO2/FiO2 ratio ≤ 250 Multilobar infiltrates Confusion/disorientation Uremia (BUN ≥ 20 mg/dL) Leukopenia (WBC count < 4000 cells/mm3 Thrombocytopenia (platelet count < 100,000 cells/mm3) Hypothermia (core temperature ≤ 36°C) Hypotension requiring aggressive fluid resuscitation

Research on biomarkers has progressed.  The role of procalcitonin is further clarified.  It cannot be relied upon to distinguish viral from bacterial pneumonia as once believed.  Currently no biomarker has been shown to improve the predictive value of clinical scoring tools described above. 

The authors present a series of references on the duration of antibiotic therapy for mild-moderate CAP that suggest that 5 days or less of antibiotics is adequate treatment in most cases.  One study even showed that a single dose of antibiotics cured 70% of mild – moderate cases of CAP. 

Evidence is accumulating however that two antibiotics are better than one in the treatment of severe CAP.  Mortality is clearly associated with empirical choice of antibiotics that cover the offending organism – two antibiotics are more likely than one to achieve this coverage.  In this regard, an increasing body of evidence suggests that macrolides are superior to fluoroquinolones and other agents as the second antibiotic.  This may not be directly related to the sensitivity of infecting organisms to macrolides.  Macrolides have other favorable qualities that might influence survival, including anti-inflammatory properties, and the ability to directly inhibit certain bacterial virulence factors.

The authors reviewed research on clinical care processes being implemented in an attempt to improve outcomes.  In some instances, efforts to improve outcomes by altering intermediate process variables are based on flawed logic.  For instance, efforts to speed up the time it takes to administer the first dose of antibiotics have shown no benefit, and resulted in serious negative effects including over diagnosis of CAP, overuse of antibiotics, and increased antibiotic toxicity including C. difficile colitis.  The authors point out that there may be valid reasons to delay antibiotic therapy in patients with multiple comorbidities, as physicians struggle to sort out a proper diagnosis.  Comorbidities, such as CHF and COPD, that may confuse the diagnosis of CAP, are independently associated with mortality, and therefore confound our understanding of the relationship between antibiotic timing and survival.  True clinical outcomes, such as severity-adjusted mortality, should remain the primary standard by which good care is measured.

Finally, the authors review the long term consequences of CAP.  Studies show that survivors of CAP suffer a doubling of subsequent mortality for as long as two years post discharge.  This might be due to an increase in cardiovascular deaths.  It’s possible that the inflammatory response to CAP leads to long-lasting changes in coronary plaque that increase the chances of plaque rupture.  Further studies in this area are needed.      

Volpicelli G.  Sonographic diagnosis of pneumothorax.  Intensive Care Medicine.  2011;37:224. (Click here for abstract).

Another excellent paper with a manageable number of take-home points.

Studies suggest that the sensitivity of the ultrasound for detecting pneumothorax is actually better than an anterior-posterior chest x-ray, and rivals CT scanning. 

Specificity is not as good, unless the “lung point” sign is found.  This has a specificity of 100%.

Clinicians need to read this paper and the accompanying illustrations to better understand how to perform a good ultrasound examination for pneumothorax.  I won’t try to bypass that necessity, but will mention the four cardinal elements of the exam. 

1) Sliding lung.  Horizontal movement of lung underlying pleura with respiration.  The presence of sliding lung rules strongly against pneumothorax at the anatomical area being examined.

2) B lines.  Artifactual vertical bright lines that radiate distally from the pleural surface -  B-lines rule strongly against pneumothorax at the anatomical site under exam.

3) Lung pulse.  Vertical movement of lung seen when respiration is interrupted.  As with the first three findings, it’s presence rules pneumothorax out.  All three of the above tests are highly sensitive.

4) Lung point.  Determination of lung point involves ultrasound exam for sliding lung at multiple points across the chest.  The lung point is the anatomical site at which sliding lung is found, when it was absent at less dependent sites.  The lung point correlates with the least dependent portion of the lung that is not collapsed and still in direct contact with the parietal pleura.  Unlike the above findings, lung point is highly specific for pneumothorax (strongly rules it in).

The clinical utility of bedside ultrasound is clearly dependent on the experience of the operator.  I recommend that our fellows do an ultrasound examination of the lungs, anytime they have the ultrasound in a patient’s room for any reason.  It is particularly important to practice ultrasound exam of the lungs in all patients with known pneumothorax, if time allows.  Becoming comfortable with these findings in controlled situations will enhance the utility of ultrasound in emergencies.   

As an aside, I want to plug again for all intensivist groups to be equipped with a good portable ultrasound machine, with cardiac and vascular probes.  Our group has accumulated an increasingly positive experience with the use of our ultrasound machine to place central venous and arterial lines, to guide thoracentesis and paracentesis, and to determine the cause of PEA arrest in codes.  This article provides a further use – ultrasound should improve the sensitivity of early detection of spontaneous, ventilator-associated or post-procedural pneumothorax.   

Biostatistics comments:

Bland-Altman plots

From: Backer DD et al.  Arterial pressure-based cardiac output monitoring: a multicenter validation of the third-generation software in septic patients.  Intensive Care Med 2011;37:233. (Click here for full text).

The  Bland – Altman plot allows comparison of two simultaneous measurements of the same parameter by two methods.  Ideally, one method should be the gold standard.

In the plot below, iCO refers injectate cardiac output, determined by thermodilution with a Swan Ganz catheter – the gold standard.  CCO refers to continuous cardiac output measured simultaneously by the Swan (without injectate).  The idea here is to see how accurately and precisely the CCO measures the cardiac output, compared to the gold standard iCO.  Note on the plot, the X-axis is the mean of each of the paired measurements, and the Y-axis is the difference.  The result is a graphical display that illustrates accuracy and precision of the CCO measurement.  

Overall (mean) accuracy is shown by the solid horizontal line, which illustrates the mean difference.  In this case, CCO overestimates iCO cardiac output by 9.5%.  This parameter is sometimes called “bias” – in this case the CCO has a 9.5% bias to overestimate iCO.

Precision is illustrated by the spread of the points and their relationship to the two dotted horizontal lines, which represent the 95% confidence intervals of the differences.  This facet of the plot shows that in individual data points, the precision of the CCO is pretty poor – commonly off by as much as 30% too high or 20% too low.  This doesn’t seem too reassuring in light of our recent switch from iCO to CCO Swan Ganz catheters!     

Interpretation of the R² statistic in linear regression

from: Boyd JH, et al.  Fluid resuscitation in septic shock: A positive fluid balance and elevated central venous pressure are associated with increased mortality.  Crit Care Med 2011;39:259. (Click here for abstact).

This graph from the paper shows the relationship between CVP and fluid resuscitation in septic shock.  The authors state that this relationship is statistically significant  - and indeed, the p value is < 0.05.  However, to a non-statistician, it looks like they fired a shotgun at a piece of paper to generate the plot.  How can this relationship be clinically significant?

In linear regression, the R² value is often as important as the p value in interpretation of significance.  R²  – in this case: (0.20)² = 0.04 – gives the percent variation of the dependent variable that is explained by changes in the independent variable(s).  Thus, this study shows that only 4% of the variability in CVP is accounted for by how much fluid resuscitation the patient received.  Even though the relationship is statically significant, it is clearly of no clinical value to try to use such a weak predictor as CVP to guide fluid management.  The authors of this paper were incorrect in their conclusions on this point. 

Robert A. Rashke MD, Critical Care Journal Club Editor

February 2011 Critical Care Journal Club

Reference as: Raschke RA. February 2011 Critical Care Journal Club. Southwest J Pulm Crit Care 2011;2:29-33. (Click here for PDF version)

Lee et al.  Analysis of overall level of evidence behind Infectious Diseases Society of America practice guidelines.  Arch Intern Med 2011;171:18-22.  (Click here for abstract) and . . .

Kett et al.  Implementation of guidelines for management of possible multidrug-resistant pneumonia in intensive care: an observational, multicentre cohort study.  The Lancet.com/infection Published online. Jan 20,2011.  (Click here for abstact)

Several very thought-provoking articles offered by Dr. Robbins for our consideration (click here for accomanying editorial).  Lee’s paper used an elegant graphic display to illustrate that 55% of over 4000 recommendations made by the IDSA are based on level 3 evidence – essentially just expert opinion.  Only 14% were supported by an RCT (level 1).  The authors point out that guidelines can only summarize the best evidence we have – sometimes expert panels are tasked with making recommendations regarding questions for which randomized controlled trials may be logistically or ethically difficult to perform.  However, this study shows that simply following guidelines cannot always be equated with providing evidence-based practice.

In a related study, Kett and colleagues suggested that adherence with IDSA guidelines for the treatment of hospital-acquired (and related pneumonias) may actually increase mortality!  

Kett’s study was a multicenter retrospective cohort with 303 patients suspected at risk for multi-drug resistant pneumonia. They used a propensity model to attempt to control for variables that might bias or confound their results.  A Kaplan-Meier analysis showed that survival in the group of patients treated in adherence with guidelines was significantly worse than in patients whose treatment was noncompliant  -  65% vs. 79% (p<0.004).  This difference persisted after adjustment for severity of illness.  Non-adherence was largely due to failure to double-cover pseudomonas, and to empirically cover MRSA.   Yet, once cultures and sensitivities were analysed, patients in the non-adherent group were actually slightly more likely to have received active empirical antibiotics. 

The authors contention that toxic side effects of potentially unnecessary antibiotics might explain the (huge) mortality difference doesn’t seem plausible.  Antibiotic complications would have to have killed 1 out of every 7 patients in the adherence group to explain these results.  It’s more likely that the retrospective study design and statistical analysis did not generate adequate internal validity to answer this complicated question.

The author’s conclusion is sound.  An RCT is needed.  Although we doubt that adhering to IDSA guidelines actually kills patients, these two studies taken together suggest that physician’s should not necessarily be villainized for failure to comply with guidelines that are sometimes based on little more than expert opinion.  

Biostatistics:    Propensity scoring is a statistical technique used to adjust for bias and confounding in cohort studies.  It is akin to multiple linear and multiple logistic regression in that it allows the researcher to statistically adjust for differences between groups of non-randomized patients that might impact the outcome of interest.  For instance, in Kett’s study, suppose that physicians were more likely to adhere to IDSA guidelines in sicker patients.  Since severity of illness is likely related to mortality, this might make it falsely appear that adherence increases mortality.   However, inclusion of an Apache severity score in a propensity model allows the researcher to essentially match subjects and controls for severity of illness.  All variables that might effect the propensity for the physicians to adhere to the guidelines can be included.  However, unrecognized factors cannot be adjusted for.  This is one reason a randomized controlled trial is a stronger general study design than a cohort study regardless of what statistical analysis is employed.

Levy et al.  Vascular hyporesponsiveness to vasopressors in septic shock: from bench to bedside.  Intensive Care Med 2010;36:2019-29. (Click here for article).  and...

Boerma et al.  The role of vasoactive agents in the resuscitation of microvascular perfusion and tissue oxygenation in critically ill patients.  Intensive Care Med 2010;36:2004-18. (Click here for article).

These reviews took some effort to get through, but were full of good information.  Although much is known about the mechanisms of vascular hyporesponsiveness in septic shock, it hasn’t translated yet into the development of any new effective therapies.  Specifically, pharmacological  inhibition of nitric oxide, prostaglandin synthesis, or K_ATP channels, and substitutive doses of vasopressin have all failed to improve clinical outcomes. 

Some interesting physiology was brought forth by the authors in regards to resuscitation.  Studies suggest that increasing arterial pressure does not necessarily improve microcirculatory perfusion in septic shock.  In fact, pharmacological vasoconstriction can actually worsen microperfusion despite an increase in blood pressure.  Additionally, vasopressors lower the capillary hematocrit (an interesting effect explained in Boerma’s paper), and worsen tissue oxygen diffusion, potentially diminishing cellular oxygen delivery.  Norepinephrine has never been shown to improve mortality in septic shock.  We’re not suggesting it shouldn’t be used, but it is disquieting to consider these arguments against a central tenet of hemodynamic support in the ICU.  Maybe in the future, we’ll find more elegant ways to measure and support microcirculatory perfusion besides simply cranking up the MAP with norepinephrine.  

Shin et al.  Extracorporeal cardiopulmonary resuscitation in patients  with in-hospital cardiac arrest: A comparison with conventional cardiopulmonary resuscitation.  Crit Care Medicine 2011;39:1-7. (Click here for abstract).

This was another retrospective cohort study.  Four-hundred six cardiac arrests lasting > 10mins were analyzed.  The decision whether or not to perform eCPR was at the discretion of the physician running the code, observing specific contraindications.  A portable self-priming ECMO unit was employed to rapidly intiate veno-arterial ECMO in 85.  Propensity scoring was used to select 60 matched pairs from the study population for analysis.  In each matched pair, one patient had undergone standard CPR, and one had undergone extracorporeal cardiopulmonary resuscitation (eCPR).  The patients were matched based on a large number of variables that were related to the propensity for the patients to have received eCPR.  These included age, comorbidities, cause of arrest, location of arrest, initial rhythm,  and study period.  The odds ratio for mortality before discharge or suffering significant neurological injury was significantly reduced in the patients receiving eCPR (OR: 0.17 95%CI: 0.04-0.68; p=0.012).  Mortality benefit was also significant after 6 months of followup.

Retrospective cohort studies should be considered “hypothesis-generating”.  This study provides strong rationale to support an RCT.   As an aside, our fledgling ECMO program has been busy this winter with the onset of influenza season.  We have yet to attempt eCPR, which obviously requires a well-coordinated system for rapid ECMO implementation.   

Brief mentions:

Jubran et al.  Post-traumatic stress disorder after weaning from prolonged mechanical ventilation.  Intensive Care Med 2010;36:2030-7. (Click here for abstract).

In this prospective cohort study, 72 patients who had undergone tracheostomy after at least 3 weeks of mechanical ventilation were followed up by a psychologist.  Five of 41 patients (12%) followed up 3 months after vent weaning were found to meet criteria for post-traumatic stress disorder.  A post traumatic stress syndrome questionnaire (PTSS-10) performed within a week of vent weaning predicted eventual diagnosis of PTSD with a sensitivity of 100% and specificity of 76%. 

Dr. Owen Reece pointed out that other studies have shown the rate of PTSD in patients who have experienced ICU care ranges from 10-50%.  Significant flaws in this study include the 43% drop-out rate, and the ethically-questionable failure to provide treatment to patients predicted to be at risk for PTSD by the PTSS-10 questionnaire done before discharge.   

Khouli H et al.  Performance of medical residents in sterile techniques during central vein catheterization.  Chest 2011;139:80-7. Click here for abstract).

This study employed an inexplicably bizarre and weak study design.  Residents were randomized to learn central line insertion via video training vs video plus simulation center training.  For reasons that are unclear, the line-related complication rate between simulation trained residents and controls were never presented.  Instead, the researchers used a convenience control group (surgical residents not involved in the randomization) to show a difference in the rate of catheter-related blood stream infection. 

If you are going to go through the trouble of performing a randomized controlled trial, you should analyze the main clinical outcome variable between the experimental and control groups.  Failure to present this analysis seems so strange that it raises the suspicion that the results may have been omitted purposefully.  We believe simulation center training is highly worthwhile, but this study does not add much support to that contention due to its weak study design.   

November 2010 Critical Care Journal Club

Reference as: Raschke RA. November 2010 critical care journal club. Southwest J Pulm Crit Care 2010;1:10-13. (Click here for PDF version)

Vincent J-L et al.  Critical Care: Advances and future perspectives.  Lancet. 2010;376:1354-61.  A very broad overview of the history of evidence-based Critical Care.  The authors observe that results of most reported therapeutic RCTs in Critical Care Medicine have been negative – and some even harmed patients.  Paradoxically, the main contribution of RCTs in Critical Care has been to show that overtreatment is harmful (e.g., high tidal volumes, excessive blood transfusions, oversedation, targeting supranormal oxygen delivery, etc.).  Additionally, the authors point out inconsistency in clinical implementation of therapies supported by RCTs, compiling a table of positive RCTs that have failed to become part of bedside Critical Care. The example they provide regarding drotrecogin alpha is very enlightening, especially for those that didn’t follow that story as it was unfolding.  This review is well worth reading.  It’s uncommon that an editorial is published with this breath and depth of perspective. 

Lee, et al.  Probiotic prophylaxis of VAP.  Am J Respir Crit Care medicine. 2010;182:1058-64.  (Also - please see editorial on pages 993-4).  This was a well-performed randomized controlled trial (RCT) that included 146 patients on mechanical ventilation, and showed that patients treated with the probiotic bacteria - Lactobacillus rhamnosus – instilled into their oropharynx and down their nasogastric tubes – had a significant reduction in the rate of VAP (19% vs. 40% p<0.007) compared to placebo.  The theory behind therapeutic use of probiotic bacteria is that they compete with, and mitigate overgrowth of the upper respiratory tract by pathogenic bacteria.  This study had several minor design flaws (e.g., misuse of intention-to-treat), and shared a problem that all such studies have – difficulty in firmly establishing the diagnosis of VAP.  However, these shortcomings did not seem critical enough to overturn the author’s conclusions.  We feel that in general (and particularly in the area of VAP), RCTs that originally show positive results are often disproven by subsequent studies.  Therefore, we wouldn’t jump on this bandwagon yet with both feet.  But the theory behind probiotics seems compelling and we await further studies with interest.  

Guo Y-L, et al.  Accuracy of BAL galactomannin in diagnosing invasive aspergillosis.  Chest. 2010;138:817-24.  This paper was of great interest because we often have difficulty diagnosing invasive pulmonary aspergillosis, especially in the bone marrow transplantation unit.  Unfortunately, this paper was disappointing.  The authors made a critical error in their study design.  The “gold standard” they used to calculate the operating characteristics did not allow patients to be firmly categorized as to whether or not they actually had invasive aspergillosis.  This approach included patients with “possible” invasive aspergillosis in the category of “no disease”.  This is nonsensical.  The concept of sensitivity is to describe how good a test is at ruling a disease out.  Therefore the sensitivity of a test can’t be calculated unless your gold standard firmly rules the disease out.  This basic error propagated through much of the advanced statistics the authors provide (for instance the ROC curves, which are partially derived from sensitivity calculations).  We used to call this technique “putting lipstick on a pig”.  No useful conclusions can be drawn from this study.

Benson AB, Austin GL et al, Transfusion-related acute lung injury in ICU patients admitted with gastrointestinal bleeding.  Intensive Care Med. 2010 Oct;36(10):1710-7. Epub 2010 Jul 24.  This paper was a retrospective cohort study that found the incidence of transfusion-related acute lung injury (TRALI) was 15% among 150 ICU patients who received transfusions for GI bleeds.  Fresh frozen plasma (FFP), rather than packed red blood cells, were most likely to be temporally-related to TRALI onset. In patients with end stage liver disease (ESLD), each unit of FFP increased the risk of TRALI by 11%.  Although we are unlikely to drastically change our approach to transfusion therapy based on this small retrospective study, we agree with the author’s conclusion that we should carefully consider the risk of each unit of FFP we infuse – this may be another case in Critical Care of “less is better”.  Strangely, these authors found TRALI to occur almost exclusively in patients with ESLD (21 of 22 cases).  This is not consistent with other trials, and doesn’t have clear biological plausibility, since TRALI is thought to be due to the presence of anti-granulocyte antibodies, or granulocyte-activating cytokines in the donor plasma. 

Summers, DM et al.  Analysis of factors that affect outcome after transplantation of kidneys donated after cardiac death in the U.K.: a cohort study.  Lancet. 2010;376:1303-11.  This paper shows that organs harvested in patients who undergo controlled cardiac death are equivalent to organs harvested in patients who are declared brain dead (with essentially nil warm ischemic time).  Briefly, controlled cardiac death may be considered in a patient on life support who is an organ donor, in whom an independent decision has been made to withdraw life support (based on the patient’s wishes or best interests).  Support can then be withdrawn in the OR, with the harvest team hovering to immediately remove organs at the moment of death.  The intensivist will usually be asked to determine this moment – typically after 5-minutes of asystole.  This is a fascinating topic that focuses on a question that is surprisingly tough to answer – how can we define the exact moment of death?  A significant number of physicians and ethicists feel that controlled cardiac death is unethical.  A very good editorial in the NEJM discusses this:  Truog RD, Miller FG.  The dead donor rule and organ transplantation.  N Engl J Med. 2008 Aug 14;359(7):674-5.

Robert A. Raschke, M.D., Critical Care Journal Club Editor