Management of Pediatric Septic Shock. Progress through Applied Insight

Septic shock is rapidly progressive and often fatal; the impact can be appalling and seems especially tragic when the victims are children. Although immense effort has been expended in the science of sepsis, as well as in well-intentioned attempts to improve practice by consensus, it is difficult to identify any new intervention in the last 2 decades that has proved beneficial to patients. In this issue of the Journal, Wong and colleagues (pp. 309–315) might help shift this impasse for children with septic shock (1).

The background is important. This group previously demonstrated that gene expression using genome-wide arrays from children with septic shock naturally aggregated into three subclasses (so-called unsupervised analysis), and that these subclasses predicted outcome (2). In short, blood from children with septic shock was analyzed to determine which genes (among the whole genome) were activated and which were depressed. Although the expression data of all the individual genes was (not surprisingly) overwhelming, the patterns could be portrayed as simple pictures, which in turn corresponded to level of risk.

Although that was a genuine advance, the current study goes one better: the investigators report on the 100 genes previously found to be the best subclass predictors (1). Using only these 100 genes, there was reliable separation of gene expression into two subclasses, one of which, subclass A, was associated with lower overall gene expression and worse outcome. All these findings were prospectively validated using different patients, and multivariable analysis helped ensure the bad outcome predicted by subclass A was independent of age, comorbidity, and severity of illness. Finally, many of the relevant genes are involved in glucocorticoid signaling, and the use of steroids in patients in subclass A (low gene expression) was associated with mortality.

Is this approach feasible? To be effective in septic shock, an intervention will have to be rapidly available and affordable. This 100-gene platform can be analyzed in about 10 hours, and likely faster in the future, in contrast to whole-genome arrays that take days. Also, at an estimated $100 per patient, this mini-array costs about the same as a day’s therapy with newer antibiotics.

There are important concerns. All studies based on gene arrays are prone to overinterpretation, principally because of the extensive noise from overwhelming numbers of genes (3), and establishing the biologic roles for all statistically relevant genes is always a challenge. In addition, there is more to an illness than gene activation, and the roles of coexisting disease, as well as specific pathogens, need to be understood. In addition, it is worrisome that this study (1) identified two gene subclasses (A and B), whereas the original report (2) identified three (A, B, and C); although the missing subclass was not clinically different than subclass B, it is unclear why it was not identified. Finally, all such studies, as with any solid science, must ultimately stand the test of reproducibility by different investigators studying different patients.

How can this approach, which is not a therapy, be of use to patients? The answer may lie in better diagnosis. At this time, “sepsis” approximates to a pathophysiologic state, and it remains a clinical conclusion. The gene array approach has provided an assessment of risk, and perhaps another metric for severity-of-disease profiles. In the current study, group A is the higher-risk group, possibly reflecting the greater use of steroids. Dissecting out the molecular basis for this increased risk will be important to enable individualized focus and care.

For sepsis, as in all of medicine, diagnosis is fundamental to practice. In sepsis, diagnosis is a problem, as it is often based on a consensus definition (4) that has high sensitivity and low specificity (5); therefore, although good for screening, it is bad for diagnosis. Indeed, a founder of our specialty, Max Harry Weil, reinforced this concern, reminding us that such definitions “were, and are useful to the extent that they call attention to an immediate life-threatening clinical sequence, for which a more precise diagnosis and targeted therapy are imperative” (6). Modern (and traditional) personalized care dictates that the better the clinician understands the individual patient’s illness, the better will be the care and the more accurate the prognosis.

A major problem in sepsis research is that while we rely heavily on randomized clinical trials, almost all of them conclude that the tested interventions are not beneficial. A key reason for the negative randomized controlled trials may be reliance on consensus definitions for diagnosis, about which Dr Weil raised concerns. For example, we would not perform a randomized controlled trial of an antiarrhythmic for any dysrhythmia; instead, we would determine for which dysrhythmia (e.g., atrial fibrillation, nodal tachycardia) the intervention was likely to be effective and study only patients with that diagnosis. Indeed, 20 years ago, a landmark study illustrated precisely this error: anti-HA1A antibody appeared to improve survival in gram-negative sepsis (albeit with a worse outcome in gram-positive sepsis) (7). The optimism was short-lived, with two unfortunate lessons for research: gram-negative sepsis was impossible to diagnose without laboratory testing (8), and the necessary test of study reproducibility failed (9).

Nevertheless, sepsis studies continue to rely on consensus definitions (or variants thereof; e.g., severe sepsis, septic shock) and have not yet shifted to biologic characterization; in addition to pathogen detection, incorporating biologic characterization would be expected to outperform the consensus-only approach. If such stratification were valid, randomized controlled trials would be restricted to patients most likely to benefit (and exclude those unlikely to benefit). For example, investigators planning to study immunomodulation in sepsis who need to identify the highest-risk patients, or those investigating use of corticosteroids, might benefit from the current molecular insights (1). More generally, if sepsis trials successfully incorporate genomic medicine, this might help the field catch up to where cancer therapeutics has been for some years (10).

The current data (1) might not apply to adults because children are less susceptible to septic shock, usually have fewer coexisting conditions, and have a lower attributable mortality. Although mortality is less, two recent reports indicate that contemporary characterization of children with sepsis may be insufficient (11, 12). In one study, protocolized fluid boluses such as those commonly recommended in the developed world increased mortality compared with usual care by local practitioners (11). In the other study, protocolized fluid regimens appeared to increase mortality in children with concomitant respiratory failure (12). The reasons for these unfortunate outcomes are not understood, but it is precisely that knowledge gap (and its consequences) that should direct us away from consensus-type definitions of fatal illness and point us instead toward placing our patients on the potentially firmer footing of biologic insight.

In summary, the big lesson from the paper by Wong and colleagues is that biologic insights into how (and who) septic shock can kill might soon be applied to the advantage of individual patients (1). In an age of consensus, guidelines, and big data, it is easy to forget that the biology of the individual patient matters; important work such as this reminds us that in pediatric septic shock, it almost certainly does.