This study constitutes the largest study to date on drug-associated hyperammonaemia and the first description of clinical features and severity for each potential liable drug. Most of the suspected drugs could be categorised into 4 main therapeutic classes: oncologic drugs, anti-epileptic drugs, immunosuppressants and psychiatric drugs. The drugs most frequently involved were valproic acid, fluorouracil, topiramate, oxaliplatin and asparaginase. In addition to these molecules known to be responsible for hyperammonaemia, our study reported 10 drugs not previously identified as responsible for hyperammonaemia (B1 extrinsic imputability). These include recently marketed molecules including anti-epileptics such as cannabidiol, immunosuppressants such as basiliximab, and anti-angiogenics agents such as tyrosine kinase inhibitors (sunitinib, sorafenib, regorafenib, lenvatinib) and monoclonal antibodies (bevacizumab, ramucirumab) [25]. Most reported cases were reported in the last 10 years. Several factors could explain this: first, the number of countries participating in VigiBase increases over time and so do the reports, second, many drugs have been released very recently in the field of oncology and finally, hyperammonaemia could be easily missed if the dosage is not performed.

It should be noted that hyperammonaemia secondary to immunosuppressants and oncological treatments were associated with the highest overall mortality rates. Hyperammonaemia secondary to anti-epileptic and psychiatric drugs had more favourable outcomes. This difference of outcome could be partly driven by more severe underlying comorbidities in patients with immunosuppressants and oncological treatments and the effect of the drugs themselves on mortality needs to be further characterised. Because of the absence of large series in the literature, this could not be compared to previous data, but these results seem to be in accord with the published case reports [26, 27]. It should be noticed that drugs known to be associated with hyperammonaemia such as asparaginase, 5-fluorouracil or capecitabine had lower mortality rates (6.9%, 4.3% and 6.7%, respectively) compared to more recently marketed molecules such as sunitinib, regorafenib or basiliximab (15.4%, 18.2% and 60%, respectively). This could suggest that the identification of the situation and its prompt management might improve the outcome and/or may indicate an awareness bias (i.e. serious cases and those involving new drugs are more likely to be reported). Tacrolimus is one of the drugs responsible for the most cases quantitatively of suspected drug-associated hyperammonaemia in VigiBase (6th out of 71) despite the absence of its mention in FDA and EMA’s drug labels. These latter cases are associated with a 57% mortality, but with data available in VigiBase, determining the exact cause of death in each individual case might be challenging, due to several limitations detailed below including exhaustivity of data information reported in records.

Clinical signs of hyperammonaemia are varied, but the most common presentation was coma or altered state of consciousness. Revealing symptoms could nevertheless be influenced by the underlying conditions of each patient and vary depending on the drug class involved. Indeed, neuropsychiatric symptoms were frequent for patients with psychiatric treatments and seizures often occurred in patients with anti-epileptic treatments. In patients taking immunosuppressants, hyperammonaemia was frequently accompanied by brain oedema or seizures. Time to onset of hyperammonaemia was not limited to the introduction period in the first few days following the liable drug start and could be delayed. Several mechanisms might explain this delayed reaction that has also been found in previous studies [16, 28 ,29]. First, it cannot be ruled out that hyperammonaemia is a dose-dependent adverse effect occurring at a "supratherapeutic" cumulative dose for some medications such as chemotherapies. Furthermore, a modification of the drug’s dose might also increase the risk of hyperammonaemia. Also, many intercurrent events (infection, acute renal failure, liver failure or the instauration of another drug with potential interactions, etc.) might cause hyperammonaemia in a patient taking one of the described drugs. For these reasons, when clinicians are faced with hyperammonaemia and after having excluded classical causes, a drug-associated origin should be suspected. The imputability of each drug should be assessed, based on an evaluation from the pharmacovigilance centre.

The main limitations of our analysis are those of retrospective pharmacovigilance studies from databases. Indeed, current post-marketing pharmacovigilance is strongly based on spontaneous notification and presents well‐known bias [21, 30]. Among these, the main bias concerns the lack of information and the under-notification. For instance, the number of cases with liver failure is strictly low and does not represent the prevalence of liver injuries in the population. This is related to the fact that hyperammonaemia is attributed—maybe wrongly—to the liver disease by hepatologists, who disregard the possibility of a drug-associated event. Due to its international nature, this database is particularly suitable for the identification of rare adverse events such as hyperammonaemia, but those results should be always confirmed by translational studies identifying the mechanisms at play [31, 32]. Moreover, underlying conditions such as cancer or auto-immune disease could also increase the risk of hyperammonaemia and constitute a potential bias. To address these issues and potential biases, we used a comprehensive approach using the informativity score and the extrinsic imputability score classically used in the French pharmacovigilance methodology of causality assessment [24]. Other methods, especially the Naranjo’s developed in Toronto would have been an alternative, but this method is designed for controlled trials rather than clinical routine practice. Furthermore, their uses are almost impossible here in view of the heterogeneity of the recordings and the available information (especially because of the absence of narratives). Finally, analyses on VigiBase do not systematically contain the exact values of lab results (such as ammonaemia or liver enzymes that can be absent or given as values relative to upper normal range), so interpretation of those parameters can only be qualitative. Furthermore, it is important to note that the imputability of each molecule is difficult to assess for drugs that are rarely prescribed in monotherapy, such as some chemotherapies or anti-epileptics. Furthermore, without data on numbers of exposed patients in VigiBase, this work cannot assess the incidence or risk of hyperammonaemia with these drugs. Finally, the description of the clinical presentation during hyperammonaemia should only be considered as exploratory because of the strong influence of the underlying conditions and the difficulty to attribute those symptoms to hyperammonaemia.