I do not like to think I gamble, but no physician can predict the future. I worry about failing to treat conditions I should and treating conditions inappropriately. As an intensivist, my patients often have so little margin that treating one condition worsens another. In strategizing for optimal outcomes, the adverse effects of what I do and choosing which condition to prioritize are real concerns.
Historically, we clinicians have taken a complex cognitive approach to this problem by combining recognition (of a disorder), prediction (of the clinical trajectory), and selection of an intervention (to fix the problem). Recognition, prediction, and intervention are interconnected, and a change in 1 alters the other 2. By allowing the clinician to observe the response to therapy, interventions improve recognition and prognosis. By appropriately assessing the pretest probability, prediction informs recognition and helps us choose an intervention. By correctly identifying disease type and severity, recognition leads to prediction and intervention. Over time, these 3 elements are refined into the entity we call clinical judgment because accumulated experience generates rich mental models of diseases and their trajectories.
One theme persists when I encounter uncertainty: the effect of unseen tradeoffs in my decision-making. Calculating the risk and benefit is particularly challenging when one or the other is not measureable. Some perioperative data already exist to help the clinician judge the consequences of their decisions. Unplanned reintubation after surgery for example carries an odds ratio of 9.2 for 30-day mortality,1 and returning to the operating room after cardiac surgery carries an odds ratio of approximately 3.2 Although crude, such data allow the clinician a starting point for a risk–benefit analysis. In this issue of Anesthesia & Analgesia, Kim et al.3 shed some light on risk–benefit analyses in the surgical intensive care unit (ICU).
By using the large National Surgical Quality Improvement Program database, Kim et al. examined not the outcomes of sepsis, acute kidney injury (AKI), or acute respiratory failure (ARF), but examined the outcomes of combinations of these conditions. After multivariate analysis to account for the effects of time and a variety of covariates, they investigated the effect of all 7 possible combinations of the 3 conditions on 30-day mortality. ARF was the strongest single predictor of mortality (hazard ratio, 27.4), and the combination of AKI and ARF had a particularly strong synergistic effect on mortality (hazard ratio, 107). Adding respiratory failure to AKI increased mortality risk by a factor of 3 to 6, depending on assumptions of the analysis. Kim et al. also noted that adding a third condition to any other 2 conditions did not exert a further synergistic effect on 30-day mortality.
The findings of Kim et al. provide a potential window to clarifying clinical decision-making in the ICU. If caring for a patient with renal failure, for example, I might restrict IV fluids to minimize the risk of respiratory failure and the consequent more than additive risk of mortality from having renal and respiratory failure together. Conversely, if a patient remains on the ventilator and his or her kidney function is at risk of failing, perhaps aggressive circulatory support, even if it worsens oxygenation, is warranted to avoid that particularly deadly combination of respiratory and renal failures. These decisions require several leaps of faith but suggest a strategy for choosing the least risky outcome. Just as a baseball manager might “play the numbers” by, for example, pinch-hitting a left hander against a right-handed pitcher, a clinician might also someday consult a series of tables attributing risk to combinations of diagnoses and manage with a goal of achieving the “least worst” combination.
The above is, of course, conjecture, and no matter how significant the P value, mechanism cannot be proven from correlations in large retrospective database analyses. Without prospective trials, we cannot know whether supra-additive 30-day mortality rates with the combination of AKI and ARF have a physiologic basis or, instead, represent a futility trigger for withdrawal of support. As a result, skeptics may find many flaws in the organ failure synergy argument. Kim et al. only investigated 3 primary dysfunctions: sepsis, respiratory failure, and renal failure. Each is an umbrella diagnosis covering a heterogeneous group of disorders. With adjustment for relevant variables, the synergistic effects erode. In fact, the synergistic mortality of renal failure with respiratory failure decreases to below that of renal failure with sepsis when the model includes other complications. In addition, some respiratory failure may be a consequence of sepsis or renal failure. Insertion of a femoral dialysis catheter, for example, may inhibit mobilization, which may then limit the pulmonary toilet activities. Surrogates for the factors leading to the mortality outcome are likely embedded in the definitions of sepsis, respiratory failure, and AKI. A strategy of combining multiple combinations of prognostic markers will produce many papers but may not yield meaningful clinical insights.
However, Kim et al. offer some daylight to intensivists forced to choose among different organ failure conditions. We see only 3 organ systems, but these 3 are an important part of care in the ICU. Moreover, the effect sizes are large. Perhaps the combination of respiratory failure and renal failure identifies a subset of patients who have lost the ability to preserve acid-base status and lack the respiratory reserve to compensate for the ensuing metabolic acidosis. If so, how do we then explain the observation that many AKI cases came after respiratory failure and that this sequence correlated with higher mortality? Questions abound, but the concept of synergistic organ system morbidities may have a physiologic basis and is worth consideration.
The existence of diagnostic combinations with supra-additive mortality, if substantiated, may lead clinicians to make a more informed risk and benefit calculation in the face of uncertainty. Other investigators suggest that combinations of factors may help clinicians discriminate among different types of patients, for whom uncertain decisions become clear in the context of their new “phenotype.”4
Could it be the association reflects a real physiologic linkage and that organs “talk” to each other? Evidence supports the idea of uncoupled organ crosstalk in critical illness.5 These data are consistent with the concept. Perhaps clinicians should think about new axes of organ interaction when considering their patients’ conditions.
All clinicians wrestle with uncertainty. Each develops their own approach to weighing potential tradeoffs and making a decision when the outcomes of each path are unclear. New data regarding how failing organs interact with each other may help intensivists better understand their clinical decisions. If a supra-additive effect of respiratory and renal failure persists in future investigations, it would help us determine whether a patient was getting better or worse and how to steer clear of bad combinations of complications. I do not know exactly what to do with the data from this study when I am considering the least bad intervention for my patients, but I want to find out.
Name: Mark E. Nunnally, MD, FCCM.
Contribution: This author wrote the manuscript.
Attestation: Mark E. Nunnally approved the final manuscript.
This manuscript was handled by: Avery Tung, MD.
1. Ramachandran SK, Nafiu OO, Ghaferi A, Tremper KK, Shanks A, Kheterpal S. Independent predictors and outcomes of unanticipated early postoperative tracheal intubation after nonemergent, noncardiac surgery. Anesthesiology. 2011;115:44–53
2. Haneya A, Diez C, Kolat P, Suesskind-Schwendi Mv, Ried M, Schmid C, Hirt SW. Re-exploration for bleeding or tamponade after cardiac surgery: impact of timing and indication on outcome. Thorac Cardiovasc Surg. 2015;63:51–7
3. Kim M, Brady JE, Li G. Interaction effects of acute kidney injury, acute respiratory failure, and sepsis on 30-day postoperative mortality in patients undergoing high-risk intraabdominal general surgical procedures. Anesth Analg. 2015;121:1536–46
4. Knox DB, Lanspa MJ, Kuttler KG, Brewer SC, Brown SM. Phenotypic clusters within sepsis-associated multiple organ dysfunction syndrome. Intensive Care Med. 2015;41:814–22
5. Godin PJ, Buchman TG. Uncoupling of biological oscillators: a complementary hypothesis concerning the pathogenesis of multiple organ dysfunction syndrome. Crit Care Med. 1996;24:1107–16