The young woman had a multidrug overdose including metoprolol succinate, extended release. She presented in shock. The authors state "Gastrointestinal decontamination was not performed because her ingestion was suspected to have occurred several hours before admission" [emphasis mine]. I have already in an early post on this blog, discussed the inadequacy of the existing data and experimental models on the timing of charcoal administration, which interested readers can read about here.
Studies of the timing of Activated Charcoal (AC) administration use normal volunteers taking therapeutic doses of drugs (for ethical reasons). We have inadequate data on the speed of absorption of drugs in people who are shocked or who take very large doses of drugs, or extended release drugs. To dichotomize the efficacy of AC by a bright line of 30 or 60 or even 120 minutes is a gross oversimplification of reality that belies overconfidence in the existing data and experimental models. This patient took a large dose of a sustained release medication and at some point became shocked presumably causing splanchnic vasoconstriction. She also took Tramadol and other medications which may slow gut motility via anticholinergic or other effects. And there may be other unknowns - other medications she took that slowed absorption of the sustained release metoprolol that we don't even know about.
Superimposed on assumptions about the absorption of the drug, a further assumption was made: that her ingestion was several hours prior to admission. What is the evidence for this? How do we know she wasn't drunk successively taking bottles of pills until she became unconscious? How do we know that the metoprolol was not the last thing that she took, just prior to being found? What is meant by "several", where did the "several hours" estimate come from, and what is the confidence interval surrounding this estimate? Suppose "several" means 3 hours - does the 95% confidence interval surrounding it include 1-2 hours, that bright line cutoff for efficacy of AC advocated by, for example, the authors of the uptodate.com chapter Gastrointestinal decontamination of the poisoned patient? Overconfidence in our imperfect estimates leads us to fail to consider the range of reasonable estimates, and makes it impossible to utilize a decision framework based, even loosely, on expected utility theory and the threshold approach to clinical decision making.
These two caveats encourage us to assign a degree of uncertainty to our estimation of the efficacy of AC in this case, and to refuse to dichotomize AC as either effective or ineffective at the time of presentation. In light of that probabilistic uncertainty, we next must consider the consequences of giving versus not giving AC. The patient had a protected airway, so the risk of aspiration of AC was reduced to some background level. There is a risk of ileus or even bowel obstruction with AC. Overall, as described in the UTDOL.com link above, the risks of giving AC in this case are low. The risk of not giving AC is that the patient will have a prolonged toxidrome from delayed absorption of metoprolol and prolonged cardiac toxicity. That is what happened in this case, and the patient received glucagon, GIK (as we used to call it; glucose-insulin-potassium) or HIE (hyperinsulinemic euglycemia) therapy, and later ECMO for prolonged respiratory support for ARDS complicating her toxidrome. (Interestingly, to my knowledge, none of these therapies have more robust support of efficacy than AC.) Whether or not this prolonged course could have been shortened with AC or MDAC (multi-dose AC) cannot be determined. However, anecdotally I will say that the toxidromes appear to improve more rapidly when it is given.
Ideally, decision makers will consider the utility of two courses of action (such as giving or not giving AC) and choose the path that is expected to lead to the best outcome. By giving AC, the net benefit would have been:
Probability(shortened course)*Value(shortened course)
- Probability(adverse effects)*Value(adverse effects)
= Net Utility
Of course, we don't know the exact probabilities or their values, but we can do a back of the envelope or sensitivity analysis to discover whether giving or not giving has stochastic dominance.
There is some probability of a shortened course with AC and its value is going to be high especially if it obviates ECMO or prolonged mechanical ventilation. We established above from the UTDOL.com article that the probability of adverse effects is low, say 5% (based on 20 years experience giving AC, I can assure you that the upper bound of a reasonable confidence interval for the risk of serious complications is not higher than this). Aspiration is a dangerous prospect with AC, but she's intubated and there are other ways of mitigating the risk (HOB upright, giving aliquots rather than a large bolus, assuring proper ETT maintenance). We can treat ileus and prevent bowel obstruction. Even if you treat the negative value of adverse effects as equivalent but opposite in sign to the value of a shortened course, then to justify not giving charcoal, you have to think that the probability of the shortened course is less than the 5% risk of adverse effects (which may cause a prolonged course!). Perhaps I have convinced you that the probability is higher for an extended release medication in a shocked patient, perhaps not. But we do see that there was a prolonged course, and it doesn't require too great a leap of faith to infer that the probability was greater than 5% that it would have been shortened by AC. In fact I estimate that there is a 50% chance that the course would have been shortened with AC.
It is difficult to apply expected utility theory to clinical decision making, as the above example highlights, because we don't have reliable estimates of probabilities or assignments of relative values of outcomes. But the exercise does at least make explicit the relevant variables and force us to consider the uncertainty about our estimates. It is my firm belief that deconstructing clinical decisions and making them explicit is the key to avoiding cognitive errors. Doing so also allows you to evaluate the decisions in retrospect to analyze where estimates appear to have been in error and to determine if any desiderata were ignored.
In sum, we should be careful about:
- overconfidence in data and expert recommendations that are based on inadequate data or flawed or assumption-prone experimental models with poor external validity
- dichotomizing decisions such as "qualifies for" or "doesn't qualify for" a therapy based on a bright line cutoff
- Failing to recognize or ignoring confidence intervals around uncertain estimates
- Failure to consider future possible states of reality (such as the possibility that the patient will have ongoing absorption, a prolonged toxidrome, and wind up on ECMO)
- Failure to carefully consider the consequences of both doing and not doing some therapeutic action