“[Is] delirium causative of cognitive impairment, or is it a symptom of a ‘vulnerable’ brain that was already in decline?”

Although the earliest descriptions of delirium date back to 500 BC, it was the English doctor Philip Barrough who first reported in 1583 a “loss of memory and reasoning power” after the resolution of delirium.  Modern population-based studies have supported this observation by demonstrating an association between delirium and subsequent accelerated cognitive decline.  This begs a fundamental question: is delirium causative of cognitive impairment, or is it a symptom of a “vulnerable” brain that was already in decline? Proving causation in this area is difficult but nonetheless extremely important to the study of delirium prevention. One tool to help us prove causation is the demonstration of neuropathology in delirium: if people incur greater brain damage during an episode of delirium, this would provide a pathophysiological connection between delirium and greater subsequent cognitive decline. In this issue of Anesthesiology, Khalifa et al.  assess the relationship between a marker of neuronal damage (neurofilament light) and delirium in a cohort of 220 cardiac surgery patients. They found that preoperative levels of serum neurofilament light correlated with postoperative delirium but changes from baseline to postoperative levels did not. Herein, we describe how these findings build on previous work and outline the unanswered questions.

There are two findings that have been broadly consistent across this field, which are reinforced by Khalifa et al.  The first is that, on average, serum neurofilament light increases during the perioperative period. This may be reflective of the (very small) harmful effect of surgery on cognition that has been observed in large cohorts  or of peripheral nerve damage during surgery. Importantly, we have shown that cerebrospinal fluid levels of neurofilament light rise postoperatively somewhat mitigating concerns about the role of peripheral nervous system contributions. The second is that baseline (preoperative) neurofilament light is associated with postoperative delirium. This is an expected finding: people with higher neuronal damage at baseline are at higher risk of postoperative delirium. However, to understand the toxicity of delirium, what is most interesting is the change in neurofilament light from baseline. This would offset the known correlation between baseline neurofilament light levels and delirium and focus attention on relative neuronal insults during the perioperative period. In 2020, both Casey et al. and Fong et al. published about an association between delirium severity and early postoperative change in neurofilament light. This gave credence to the idea that people with delirium incurred relatively greater neuronal damage than those without delirium. Furthermore, we showed proportional associations with delirium severity and inflammation, as well as associated covert stroke on magnetic resonance imaging.  However, Khalifa et al.  do not observe evidence of an association of delirium and a rise in neurofilament light, something also recently reported by Brown et al.  in the British Journal of Anaesthesia. This has many possible explanations. First, Khalifa et al.  use a binary delirium outcome (present or absent), which sacrifices power to study dose–response relationships compared to delirium severity. Second, Khalifa et al. used the Confusion Assessment Method for the intensive care unit (ICU) to assess delirium for all patients in the ICU, which is suboptimal compared to the Confusion Assessment Method for patients who are not intubated. Third, Khalifa et al. studied cardiac surgery patients, whereas Fong et al. and Casey et al. studied noncardiac surgery patients. These observations could point to the idea that neurofilament light is related to delirium in noncardiac but not cardiac patients. If this is truly the case, the underlying reasons for it are unknown.

Although these articles explore neurofilament light in delirium, none of them assess the relationship between postoperative neurofilament light and long-term cognition. Brown et al.in a cohort overlapping that cited above, observed a correlation between postoperative day 1 change in neurofilament light and 1 yr cognition. It is difficult to reconcile this finding with the reported lack of association of neurofilament light and delirium in the same cohort, given that postoperative delirium is considered predictor of cognition at 1 yr postoperatively. It is certainly possible that the neuronal damage incurred after surgery is (in part) pathophysiologically independent of delirium: only a subset of patients with covert stroke get delirium. Stated another way, neuronal injury is only one of many pathophysiological drivers of delirium. While the pathophysiology of delirium remains largely opaque, it will be difficult to isolate this subset of patients who have a “neuronal damage” delirium phenotype that can be specifically targeted in clinical trials without a rapid bedside test for neuronal injury and simultaneous ability to exclude other causes.

Other explanations for the discordance of the cardiac and noncardiac surgical findings include that noncardiac surgery may be a more heterogenous surgical population.  Heterogeneity in the precipitating insult allows us to identify factors driving that insult more easily (particularly when it is unethical to randomize patients to different insults). Notably, operative time, blood transfusion, and type of surgery (but not cardiopulmonary bypass time) did not differ between the delirious and nondelirious groups in the study by Khalifa et al.  However, according to Casey et al.operative time and blood loss did predict neurofilament light rises, as well as delirium. Perhaps larger cohorts of cardiac surgical patients may detect associations of neurofilament light and delirium. Further exploratory analyses of these data, with appropriate validation cohorts, may yield important advances in our understanding of the dynamics of perioperative neuronal injury and its inter-relationship with delirium.

Underlying all of the unanswered questions in this field is the caveat that there is significant heterogeneity in the methods used to assess the relationship between neurofilament light and cognition. Investigators may choose to use peak or postoperative day 1, 2, or 3 neurofilament light; peak or postoperative day 1, 2, or 3 delirium severity; delirium incidence; plasma or cerebrospinal fluid samples; fixed or mixed effect models; and so on. The cognitive tests used vary widely and are sometimes combined in a Z score and sometimes not. The covariables that are adjusted for are never identical between research groups. With rates of delirium on the rise in aging populations in many countries and no currently proven pharmacologic option to prevent delirium, the clock is ticking; as a field, we will need more collaboration, transparent reporting, open dialogue, and consensus to move forward quickly enough.