We performed a literature search using the PubMed/Medline and Web of Science databases, using the search terms “serotonin syndrome” and “serotonin toxicity”, “diagnostic criteria”, and “hyperthermia”.

To assess SS’s clinical relevance, we calculated the expected number of SS cases in several countries worldwide, for whom updated statistics on antidepressant use were available from the Organization of Economic Collaboration and Development (OECD). The OECD uses defined daily doses (DDD) of antidepressants (Code N06A according to the Anatomical Therapeutic Chemical (ATC) Classification System) per 1000 inhabitants per day¹ (¹Organisation for Economic Co-operation and Development (OECD). Pharmaceutical market. Pharmaceutical consumption. Antidepressants. https://stats.oecd.org/Index.aspx?ThemeTreeId=9 [accessed October 28]).DDDs provide an estimate of the proportion of a population treated daily with a particular drug or drug class. For instance, a DDD of 10/1000 inhabitants per day for a particular drug can be interpreted as 1% (10/1000) inhabitants receive this particular drug each day in that year. DDDs are most useful for drugs used chronically with a good agreement between the average prescribed daily dose (PDD) and DDD (WHO)² (²World Health Organization (WHO). Essential medicines and health products. https://www.who.int/medicines/regulation/medicines-safety/toolkit_indicators/en/ [accessed October 31]). The WHO states that DDDs only give “a rough estimate of consumption and not an exact picture of actual use”³ (³World Health Organization (WHO). Defined Daily Dose (DDD). https://www.who.int/toolkits/atc-ddd-toolkit/about-ddd [accessed December 8]). The WHO endorsed the ATC/DDD methodology for global use in 1996 as an international standard for drug utilization studies ³ (³World Health Organization (WHO). Defined Daily Dose (DDD). https://www.who.int/toolkits/atc-ddd-toolkit/about-ddd [accessed December 8]).

Our estimate for the incidence of SS was based on US insurance claims data collated between 2008 and 2013 from two national datasets, (i) the Veterans Health Administration (VHA) dataset and (ii) the Intercontinental Marketing Services PharMetrics Plus (IMS) dataset (Nguyen et al., 2017). To determine the minimum and the maximum number of SS cases to be expected, we used the maximum and minimum incidence figures for 2013. We multiplied these with the estimated number of patients per year taking antidepressants.

To establish the causes of severe SS, we used our database of SS cases obtained from a systematic review of cases published on PubMed/Medline or Web of Science between 1 January 2004 and 31 December 2018. We used 1 January 2004 as a starting point because all three diagnostic criteria systems for SS had become available by that time. We used the search terms “serotonin syndrome” or “serotonin toxicity” to identify cases. We counted as severe cases that had either resulted in (i) intensive care, (ii) intubation, (iii) coma, or (iv) death. We analyzed the following variables: (i) symptoms associated with severe SS, (ii) speed of onset of SS, and (iii) substances thought to have precipitated severe SS. For a potential association between monoamine oxidase inhibitors (MAOI) and severe SS, we also calculated odds ratios (OR) and 95% confidence intervals (CI). The systematic review method underlying this database has been published in detail elsewhere (Ott et al., 2019; Werneke et al., 2016; Wikström, 2019).

The prevalence of SS remains unknown. Our best estimates stem from a US insurance claims study. In this retrospective cohort study of 15 million patients exposed to at least one serotonergic agent, the incidence for SS lay between 0.9 and 2.3/1000 individuals exposed in 2013 (Nguyen et al., 2017). The absolute number of cases may vary widely across the world, depending on the population size of a respective country. For instance, each year, there may only be three to six cases in Latvia, a country with a small population and low antidepressant use. However, there may be hundreds or up to a thousand cases per year in countries with large populations and/or high antidepressant use, such as Turkey, Germany or Canada (Table 3). Therefore, although SS is rare in relative terms, the number of cases can be substantial in absolute terms. This makes SS a clinically relevant condition.

Our updated systematic review of published SS cases expanded our database from 299 to 412 cases collated from 350 articles. Of all 412 cases, 42% (173) had resulted in severe SS. Of all 173 severe cases, 13% (22) resulted in death. Our updated database’s re-analysis showed that of all 412 SS cases, the Hunter criteria would have missed 37%. The Sternbach criteria would have missed 10%, and the Radomski criteria 24%. Of all 173 severe cases, the Hunter criteria would have missed 36%, and the Sternbach criteria would have missed 8% and the Radomski criteria 11%. In terms of missing cases SS overall and severe SS cases, the Hunter criteria performed worse than the Sternbach and Radomski criteria.

Our updated systematic case review, including only the 234 cases for whom a temperature was explicitly mentioned, 36% did not have a temperature $>$ 38 ${}^{\circ }$ C (100.4 ${}^{\circ }$ F). Of the 128 severe cases with a temperature explicitly mentioned, 27% did not have a temperature $>$ 38 ${}^{\circ }$ C (100.4 ${}^{\circ }$ F). Hence, SS can occur in the absence of an elevated temperature, even in severe course cases.

Information on the onset speed was available for 314 of the 412 cases in the total sample and 122 of the 173 severe cases. Of the 314 cases in the total sample, 115 concerned surgical procedures, trauma care, overdoses or substance misuse and 199 concerning medical or psychiatric treatments. In the 115 cases relating to surgical procedures, trauma care, overdoses or substance misuse, 51% of all SS episodes presented within six hours and 17% after 24 hours. In the 199 cases relating to medical or psychiatric cases, 12% of all SS episodes presented within six hours and 65% after 24 hours. Of the 122 severe SS cases, 75 concerning surgical procedures, trauma care, overdoses or substance misuse and 47 concerning medical or psychiatric treatments. In the 75 severe cases relating to surgical procedures, trauma care, overdoses or substance misuse, 60% of all SS episodes presented within six hours and 9% after 24 hours. In the 47 severe cases relating to medical or psychiatric cases, 11% of all SS episodes presented within six hours and 62% after 24 hours. Our findings suggest that the speed of onset seems to depend on pharmacokinetic and pharmacodynamic factors. Therefore, fast onset cannot be regarded as a reliable clinical sign of SS.

We explored this further using our updated systematic case review. We identified the top ten causes of severe SS as associated with any of the three diagnostic criteria systems and determined MAOI involvement for each cause. Of the 173 severe SS cases diagnosed by any of the three diagnostic criteria systems, only 41% (71) involved MAOI (Table 4). The odds ratio for MAOI involvement was 3.68 (CI 2.32-5.83; P 0.001). As expected, most cases related to non-selective or selective MAOI-A inhibitors. Thus, our data confirm that MAOI is indeed significantly more frequently implicated in severe SS. But the absence of MAOI treatment does not exclude a diagnosis of SS.

To understand why the Hunter criteria achieved gold standard status for the diagnosis of SS, it is helpful to recall how all three diagnostic criteria systems came about. The Sternbach criteria were developed in 1991, based on 38 psychiatric cases (Sternbach, 1991). In 2000, Radomski et al. (2000) suggested a new classification building on Sternbach criteria. They expanded the underlying evidence to 62 cases, adding 24 further published cases. They also suggested that symptoms should be divided into major and minor symptoms. Radomski et al. (2000) added rigidity (hypertonicity) as a major criterion and insomnia, dilated pupils, tachycardia, tachy- or dyspnea, diarrhea, and hyper- hypotension as minor criteria. Uncoordination was another minor symptom in Radomski criteria, corresponding to the criterium of incoordination in the Sternbach criteria. Radomski et al. (2000) recognized that SS varied in severity and might manifest itself as mild, full-blown and toxic states. A third classification system, the Hunter criteria, was proposed in 2003. The Hunter criteria were derived from 473 cases of single serotonin-reuptake inhibitor (SSRI) overdoses. The derived criteria were then tested (validated) on 2222 cases of overdoses with any serotonergic drug. Prior to the validation exercise, rigidity was added as an additional symptom. The Hunter criteria firmly shifted SS’s focus to neuromuscular symptoms, particularly to clonus in various forms (Dunkley et al., 2003).

The variability of symptoms between the three diagnostic systems has raised which system is the best. One major concern of the Sternbach criteria has been their potentially low specificity. For instance, a possible symptom constellation could consist of mental status changes, agitation and fever. This constellation is commonly found in many medical conditions. Yet, it is important to remember that the Sternbach criteria require the presence of a serotonergic agent, absence of a neuroleptic agent and the ruling out of other possible etiologies.

According to their developers, the Hunter criteria have better sensitivity and specificity than the Sternbach criteria (Dunkley et al., 2003). Therefore, the Hunter criteria have widely been recommended as the gold standard of SS (Boyer and Shannon, 2005; Buckley et al., 2014), but their sensitivity and specificity may have been substantially overstated. The derivation dataset, from which the Hunter criteria were generated, was also part of the validation data set, on which the Hunter criteria then were tested. 21% of all cases in the validation dataset had also been used to derive Hunter criteria. Given the magnitude of potential bias, the claim that Hunter criteria were more sensitive (84% vs. 75%) and more specific (97% vs. 96%) (Dunkley et al., 2003) than Sternbach criteria cannot be upheld (Werneke et al., 2016). Only re-analysis of the original validation set with the derivation cases’ exclusion could unambiguously settle this question. The developers of the Hunter criteria themselves point out that their criteria would require validation in other settings (Dunkley et al., 2003). This, to our knowledge, has not occurred. Another concern is the exclusive reliance of Hunter criteria on overdoses for derivation and validation. In overdoses, symptoms are likely to be much more clear-cut and severe. Other cases with more discrete symptoms may be missed (Werneke et al., 2016). The Hunter criteria requirement of the presence of clonus in its various forms or a combination of tremor and hyperreflexia for SS diagnosis may also be too narrow. For instance, patients who have developed substantial rigidity may have become too constrained in their ability to move (Boyer and Shannon, 2005). Clonus and hyperreflexia may not be visible in patients in whom peripheral polyneuropathy masks upper motor neuron signs (Prakash et al., 2014).

There is substantial variability of symptoms sampled in the three classification systems. Of all symptoms, five symptoms appear in all three criteria sets. Seven are shared by two of the three criteria sets. Six symptoms are only found in one of either diagnostic criteria set. In our previous systematic review of 299 SS cases, we showed that agreement beyond chance lacked between the Hunter and Sternbach criteria and between the Hunter and Radomski criteria (Werneke et al., 2016).

Some symptoms, such as clonus, myoclonus and hyperthermia, require particular attention since their definitions can be ambiguous.

It has been generally accepted that-SS develops fast (Boyer and Shannon, 2005). In our previous work, we have already put this claim in question (Werneke et al., 2016). Our results show that the speed of onset of SS depends on drug administration and dose. SS may develop fast in the context of overdoses or in surgical or intensive care settings, in which serotonergic agents may be administered quickly. SS may develop slowly in medical and psychiatric settings, in which serotonergic drugs are often slowly introduced or cross-tapered.

All three diagnostic criteria systems require the presence of a serotonergic agent without giving a specific definition (Dunkley et al., 2003; Radomski et al., 2000; Sternbach, 1991). Pragmatically, a serotonergic agent may be defined as a substance capable of leading to excess serotonin activity (Boyer and Shannon, 2005). Sternbach (1991) and Radomski et al. (2000) suggested that SS might be due to activation of the central 5HT ${}_{1A}$ receptor based on animal experiments findings. Other evidence from animal experiments has pointed towards 5HT ${}_{2A}$ agonism (Isbister and Whyte, 2002; Nisijima et al., 2000). This is further supported because cyproheptadine, an antihistamine with strong 5HT ${}_2$ antagonistic effects (Kapur et al., 1997), is recommended for the treatment of SS (Boyer and Shannon, 2005). This implies that the withdrawal of substances with 5HT ${}_2$ antagonistic properties could also cause SS (Harmouche, 2018; Stevenson et al., 2013). Ultimately though, no single receptor seems to be responsible for the development of SS (Boyer and Shannon, 2005). Serotonin toxicity may be caused by an increase in serotonin’s intrasynaptic concentration in the central nervous system (CNS) (Dunkley et al., 2003).

Clinicians may associate serotonergic drugs with mainly psychiatric treatments. However, many other substances also have serotonergic properties, including somatic medicines, recreational drugs and complementary remedies (Table 2) (Boyer and Shannon, 2005; Buckley et al., 2014; Truedson et al., 2020). For instance, some opiates, such as pethidine (meperidine) and tramadol, can inhibit serotonin reuptake. Not only some antidepressants but also several somatic medicines can act as MAOI, either type A or type B. Examples of MAOI-A include methylene blue and linezolid. Methylene blue is used to treat methemoglobinemia and as a contrast dye used in specific surgical procedures. Linezolid is used as an antibiotic against methicillin-resistant Staphylococcus aureus (MRSA). Examples of MAOI-B include selegiline (L-deprenyl), rasagiline, and safinamide to treat Parkinson’s disease.

In our results, we have challenged the assumption that severe SS “usually” (Buckley et al., 2014; Isbister and Buckley, 2005) or “almost exclusively” (Foong et al., 2018; Isbister and Buckley, 2005) involves MAOI. MAOI are classified according to three pharmacological criteria, (i) affinity to subtype of the enzyme (isoenzyme), (ii) selectivity, and (iii) reversibility (Table 5). There are two forms of MAO, MAO-A and MAO-B. Both MAO-A and MAO-B metabolize dopamine and tyramine. However, MAO-A is relatively selective for the metabolism of norepinephrine and serotonin (Lotufo-Neto, 1999) with an about ten-fold higher affinity for serotonin than MAO-B (Finberg, 2014). Reversibility refers to the ease with which the MAOI can be displaced from the enzyme. Irreversible MAOI cannot be displaced from the enzyme. They essentially stay put for the lifetime of the molecule of about 14 to 28 days and deactivate the enzyme during this time (Lotufo-Neto, 1999). Reversible MAOI may clear the enzyme within a day. For instance, for moclobemide, a reversible MAOI-A, MAO-A inhibition rises to 80% within two hours of administration and lasts between eight and ten hours. After 24 hours, the MAO-A activity is completely restored. Judged on the pharmacological properties alone, irreversible MAOI with a high affinity for MAO-A, irrespective of selectivity, will carry the highest risk of SS, followed by reversible MAOI-A. Reversible MAOI-B should carry the lowest risk of SS.

Despite non-selective and selective MAOI-A having strong serotonergic properties, there is no logical reason why other serotonergic drugs could not cause SS. Neither were the three diagnostic criteria exclusively derived from MAOI. In the original 38 Sternbach criteria cases, 85% involved MAOI. However, the Hunter criteria were derived from single SSRI overdoses with any cases due to other serotonergic drugs being expressly excluded. The Hunter criteria added rigidity prior to the validation to not miss severe cases with rigidity in whom clonus or hyperreflexia was not demonstrable (Dunkley et al., 2003). These may most likely have included some SS cases caused by MAOI. Generally, SS may occur when levels of serotonin rise. Levels can rise with dose increases, the addition of serotonergic drugs or withdrawal of serotonin antagonists. Hyperserotonergic states can even arise with impairment of metabolism (e.g., via cytochrome P 450 system) or elimination (Ott et al., 2019).