Association of Surgery and Anesthesia With Mental Disorder Diagnoses: What Would Sir Austin Bradford Hill Say?

In the current issue of the journal, Ing et al1 further contribute to the increasing research on whether exposure to anesthesia at an early age is detrimental to the developing brain. The potential for anesthetic agents to increase the risk of neurotoxicity has become both a paramount concern and a source of intense debate among perioperative caregivers over the last 2 decades.2–5 The fundamental question is whether children who undergo surgery and general anesthesia (GA) in the first few years of life are at risk of long-term or permanent neurobehavioral abnormalities. Given that millions of children undergo procedures requiring GA every year in the United States and around the world, this is an important clinical and epidemiologic question. One of the earliest studies on this topic demonstrated increased rates of neuronal death (neuroapoptosis) after the administration of N-Methyl-D-aspartate receptor blockers to neonatal rats.6 These early investigators subtly concluded that “these findings may have implications for pediatric anesthesia and that the mechanism of drug-induced neuro-apoptosis could contribute to a variety of neuropsychiatric disorders.”6 Since these early laboratory studies, an ever-increasing number of observational studies on human infants and preschool children has examined the association of multiple or single GA exposure and surgery with subsequent development of neurobehavioral problems in childhood.5,7–15 These human studies showed a variety of primary outcomes on factors ranging from learning disabilities, school performance evaluations, intelligence testing, and parental behavior assessments. Not surprisingly, the findings of these studies have been inconclusive and have generated more questions than answers.8 What complicates the question of surgery and anesthesia-related neurotoxicity is that there are no clear biomarkers or phenotypes for outcomes. Unlike other clinical problems, anesthesia-related neurotoxicity is an issue of “reverse association”; that is, laboratory studies in animals have preceded clinical observations. Animal studies are important in helping to delineate such issues, but one must remember that an animal model does not always equate to human conditions. One such example is studies on rodents that suggest that sevoflurane causes renal toxicity.16,17 Now, millions of anesthetics later, the animal model of renal toxicity does not appear to reflect the human experience. Given that studying anesthesia-related neurotoxicity is not amenable to classic randomized controlled trials (hard to find volunteers for surgery without anesthesia), most published works on this subject are by default, nonexperimental.2–5 What is not clear is which factors, if any, are associated with neurocognitive or behavioral impairment. Is there a threshold value for anesthetic toxicity? Does patient age at exposure, drug dosage, duration of anesthetic exposure, number of anesthetic exposures, or the area under the anesthetic concentration time curve of anesthetic exposure influence neurocognitive outcome? Human studies, whether retrospective6,7 or prospective,9 are conducted in inherently “noisy” environments. For example, standardizing the exposure variable (GA) is impossible. GA is a “state,” and no 2 GA concentrations produce the same anesthetic state. Also, is the patient inherently predisposed to the outcome, or did surgery/anesthesia simply unmask this inherent defect; that is, are surgery and anesthesia simply innocent bystanders? In this current issue of Anesthesia & Analgesia, Ing et al1 retrospectively studied a large cohort of children (N = 38,493) from the New York and Texas Medicaid databases who had undergone either pyloric stenosis, inguinal hernia, circumcision, or tonsillectomy and/or adenoidectomy before 5 years of age. The primary objective was to determine whether the age at GA exposure affected the risk of developing a mental disorder for up to 12 years after GA exposure. Mental disorder diagnosis was identified using International Classification of Diseases, 9th Revision coding and included the following: schizophrenia, bipolar disorder, depressive disorder, anxiety disorder, conduct disorders, autism, attention-deficit hyperactivity disorder (ADHD), intellectual disability developmental delay, delusional disorders, dissociative and somatosensory disorders, personality disorders, and adjustment disorders. The authors noted that exposed patients had an increased hazard ratio of 1.26 (95% confidence interval, 1.20–1.32) for developmental delay and 1.31 (95% confidence interval, 1.25–1.37) for ADHD, and that there did not appear to be an effect with respect to the age of the patient at the time of the anesthetic exposure. Unfortunately, the study design (retrospective database analyses) makes it difficult to draw any conclusions other than an association. Propensity matching involving 50 variables was used for comparative purposes to identify a group of unexposed patients. However, no mention was made as to whether children were raised in a single parent or 2-parent household, the number of household members, the level of maternal education, any history of parental substance abuse, or a family history of depression or mental illness. In addition, there was no way to ascertain who made the International Classification of Diseases, 9th Revision diagnosis of mental disorder, let alone the accuracy of the diagnosis. Although retrospective database studies make drawing any conclusions about causation difficult, the authors’ use of 2 different states’ databases makes for some interesting observations. It is notable that unexposed children in Texas had the same incidence of mental disorders (3.5–5.4 diagnoses per 100 person-years) as the exposed children in New York (3.0–5.3 diagnoses per 100 person-years). Analyzing a claims database for clinical outcomes is always going to be difficult given the inherent “noise” associated with these observations. Possible limitations noted with propensity matching coupled with possible misclassification errors when combining distinct databases further limit drawing any conclusions about causality. Another concern about the study was the inclusion of children who underwent adenotonsillectomy (tonsillectomy and adenoidectomy) in their analyses. Given that adenotonsillar hypertrophy and sleep-disordered breathing are often associated with the outcome variables reported in their study (ADHD and learning disability),18 this creates significant issues with confounding, which propensity score matching may not fix. Although association does not mean causation, nevertheless, investigators and the lay public often attempt to determine causation from nonexperimental studies. Given the limitations of determining causal factors in patients exposed to anesthesia, the epidemiologic approach involving the causality criteria by Hill19 can be utilized to examine the hypothesis that GA may cause neurobehavioral anomalies or mental disorder diagnoses. Thus, we ask the question: what would Sir Austin Bradford Hill do? In 1965, Hill19 proposed a set of 9 criteria to determine causality in observational or epidemiologic conditions.19 Establishing an argument for causation is a critical research activity because it affects the delivery of safe medical care. Given that anesthesia/surgery-related neurotoxicity can only be subjected to observational cohort studies (surgery without anesthesia is unethical), it represents the ideal scenario to subject to the criteria for causality by Hill.19 The criteria are as follows: (1) strength of association; (2) consistency; (3) specificity; (4) temporality; (5) biological gradient; (6) plausibility; (7) coherence; (8) experiment; and (9) analogy. 1. Strength of the observed association: This was listed as the first criterion in the article by Hill.19 The premise is that if condition A causes outcome B, then A and B can be demonstrably associated with each other. The association should be strong enough to be judged as clinically significant. However, Hill19 cautioned that we should not be too ready to discard slight associations. A small association does not mean causal association is absent, although larger the association, the more likely that it is causal. Additionally, this criterion ignores the fact that not every component cause will have a strong association with the outcome produced. Although the overall hazard ratio of mental disorder diagnosis in the study by Ing et al1 was only 1.26 (1.22–1.30), not a strong effect, other anesthesia and neurocognitive studies have found similarly small to moderate measures of association.3,4 Generally, an odds ratio (OR) of 1.5 or less is considered small, OR = 2 moderate, and OR ≥3 large effect size.20 2. Consistency of the observed association (reproducibility): Hill19 suggests that reproducible findings observed by different persons in different settings strengthen the likelihood of causation. Although the data by Ing et al1 will appear to score highly on this criterion, given that there are now many studies that have examined the association between GA and a wide variety of neurodevelopmental outcomes,2–8 certain caveats are necessary. The results from a review of these studies have been mixed largely due to inconsistency in the definition of the outcome variable.8 Diagnostic criteria for anesthesia-related neurobehavioral changes have varied widely and many of the studies have used a variety of diagnostic instruments, making comparison between studies extremely difficult. Another important caveat with the consistency criterion is that it is highly subject to publication bias because of the higher likelihood that positive findings get published. On the other hand, nonreproducible or inconsistent findings may be due to differences in research methodology. Furthermore, causal agents might require the presence of other factors. For example, does anesthesia/surgery-related neurotoxicity require hypoxia, hypercarbia, or other factors that present day limited knowledge has not identified? Finally, as Hill19 cautioned, shared flaws in study design would tend to replicate the same wrong conclusions. 3. Specificity: This is probably the weakest of all the criteria to establish by Hill,19 given that epidemiologic diseases are rarely caused by a single agent. There is no specific “causal agent” to be found in the report by Ing et al.1 Consider the outcome variable of ADHD. Was it the GA? The surgery? The prevailing environmental conditions in the states studied? Without knowing a threshold value for anesthetic toxicity, vulnerable age range of exposure, drug dosage, duration of anesthetic exposure, number of anesthetic exposures, or the area under the anesthetic concentration time curve make specificity a difficult component to achieve. However, the lack of specificity should in no way detract from the argument for causation.19 4. Temporality: The exposure must precede the disease or the outcome it is supposed to cause. The premise is that if A causes B, then A should occur before B, or as Hill19 reflected, “which is the cart, and which is the horse?” The best way to determine temporality in health research is to conduct prospective trials where subjects are examined for the presence of the outcome before the exposure variable. Subjects are then monitored for the outcome after a sufficient lag period. In the case of anesthesia-related neurotoxicity, GA administration must predate the disorder it is purported to cause. This is ostensibly the easiest of the criteria by Hill19 to fulfill but, in reality, the most difficult. GA is a clear categorical variable. A patient either received GA or they did not. Unfortunately, this clear variable can be difficult to establish from retrospective database queries. In their report, Ing et al1 utilized surgical procedure as a proxy for GA, arguing that it is impossible to perform the selected procedures without anesthesia. While this may be true for tonsillectomy and adenoidectomy and pyloromyotomy, some of the circumcision and herniorrhaphy cases could have been performed without GA, which would introduce recall or attribution bias to the outcome of the study. Furthermore, since mental disorders were not prospectively evaluated before the anesthetic exposure, it is impossible to determine whether the patients had preprocedural behavioral disturbances from a simple retrospective database study. This limits the assessment of temporality. 5. Biological gradient: Hill19 suggested that evidence of a dose–response relationship (or biological gradient) was indicative of a causal relationship.19 Greater exposure should generally lead to a greater incidence of the effect. While this criterion may be required for an outcome due to exposure to pathogens or toxins, it may not be appropriate or easy to establish in anesthesia-related neurotoxicity. Though studies have shown patients with repeated exposures have fared worse, it is unclear whether the repeat exposure or the patient’s underlying disease process that is responsible. As previously noted, it becomes essential to discern a threshold value for anesthetic toxicity, vulnerable age, drug dosage, duration of anesthetic exposure, number of anesthetic exposures, or the area under the anesthetic concentration time curve of anesthetic exposure. Furthermore, what is the best measure of the putative brain insult resulting from GA? Is it language impairment, behavioral abnormalities, or something else? Most outcomes such as language delay and attention-deficit disorder are indirect measures of brain injury because they may be influenced by noncerebral factors, such as parental education, income, home stressors such as parental abuse, relocation, job losses, etc. None of these issues appeared to be addressed in propensity matching. 6. Plausibility: Does it make biologic sense that the exposure is causing the outcome? Hill19 was quick to point out that this is a feature we cannot demand to establish causality. The principal reason is that biologic plausibility is often determined by the prevailing knowledge and belief system because “truth” changes with our belief system. For example, it was once “true” that blood-letting was the standard of care for the prevention and treatment of illnesses. Also, it was generally accepted until the late 1980s that neonates did not feel pain and major surgeries were performed on babies without anesthesia or analgesia.21 These lines of thinking, popular in those days, make no sense in contemporary medicine. Therefore, biological plausibility depends on the prevailing knowledge of the day. Consequently, we must be willing to accept, regardless of what we believe about anesthesia-related neurobehavioral disorders, that an argument that makes no sense to us does not mean it is not true. The remaining 3 criteria by Hill19 (coherence, experiment, and analogy) can be succinctly summarized. Coherence between known information, clinical and laboratory, increases the likelihood of causality. Unfortunately, there is no consensus on the role of GA with neurobehavioral or mental disorder diagnoses. There are presently insufficient data to support the experimental evidence criterion because randomizing children to surgery without anesthesia is unethical. Finally, regarding analogous evidence, if neurobehavioral changes are observed in other situations comparable to GA, this could provide analogous evidence. For example, children sedated in the intensive care unit may provide some analogous evidence. However, eliminating confounders in this setting would be almost impossible. SUMMARY Although each of the criteria by Hill19 is not a sine qua non for causality, the criteria do provide a framework against which anesthetic exposure and outcome can be examined. The question underlying the hypothesis that GA causes neurobehavioral or mental disorder is far from being answered. To this end, further research on this subject is needed. However, future studies should address the methodologic flaws of existing studies and reexamine the value of retrospective database studies, with their inherent biases and presumed matching criteria. Finally, as with many other studies in this area, “what we don’t know, we don’t know” may be more of a contributor to the findings than what we think we know. Studies that can identify a biomarker or develop a phenotype will become important elements in determining causation, as well as defining threshold values for toxicity with respect to anesthetic agents, doses, and duration and age of exposure. DISCLOSURES Name: Olubukola O. Nafiu, MD, FRCA, MS. Contribution: This author helped with idea concept, literature review, and manuscript preparation. Name: Peter J. Davis, MD. Contribution: This author helped with manuscript preparation and editing. This manuscript was handled by: James A. DiNardo, MD, FAAP.