Sirs, We noticed the analysis by Schmidt and Dalhoff of the effect of regular medication on the outcome of paracetamol poisoning.1 There are a number of methodological problems in the design of the study and difficulties with the presentation of the results that make the conclusions and interpretation less reliable than stated. There is, in fact, little evidence to support the observation that opioid analgesics have an effect on paracetamol poisoning.

  • (a) Any study of overdose patients that attempts to define risk factors for a particular outcome in any poisoning must clearly define the population that is being investigated. The population must be as homogeneous as possible or, at least, known differences in major factors must be adjusted for. The population sampled by Schmidt and Dalhoff includes both tertiary and primary referral cases of very differing severity.1 Seventy-three per cent of the patients (n = 536) were transferred from other hospitals with severe poisoning, and the mortality in this group is likely to be far greater than in the 201 patients who were admitted primarily from the local region. Unfortunately, the authors do not reveal the mortality in this severe group; however, a previous study by the same authors, which included only 671 patients (492 tertiary referrals and 179 local cases; does not include the 66 patients recruited in the first 6 months of 2001), showed that the mortality was 9.3% (46 of 492).2 The authors do state that the mortality in the 201 patients was zero,1 so that this group was clearly different and not as severely poisoned. The subsequent merging of these two groups for the analysis of the effect of regular medication on outcomes in paracetamol poisoning introduces significant bias, with over-sampling of the severely poisoned patients, and is thus not a valid methodology.

  • (b) During the construction of a logistic regression model to identify variables (risk factors) for the development of various markers of severe paracetamol poisoning, the authors do not clearly describe the model used and how they dealt with regular medications as a variable. It appears from Table 2 that they have performed a univariate analysis of each type of regular medication, and subsequently compared this to all cases without that regular medication, referring to the latter comparison group as ‘controls’. This means that, although they have controlled for known risk factors, they have not attempted to control for interaction between different regular medications. Thus, each regular medication ‘case’ is compared with a different set of ‘controls’ in Table 2. This methodology is not valid for investigating the extent to which regular medications alter paracetamol toxicity.

  • (c) The use of an International Normalized Ratio (INR) of > 3 as a measure of severity of hepatic dysfunction is confounded by the fact that 50% of patients with paracetamol poisoning, without hepatocellular injury, will have a raised INR.3 This increase appears to be due to the inhibition of activation of vitamin K-dependent coagulation factors (mainly factor VII), and is independent of hepatotoxicity.3 Schmidt et al., themselves, have recently shown that the INR increases after the initiation of N-acetylcysteine treatment (which all of the patients in this study received) without hepatocellular injury.4 Thus, an association between any regular medication and ‘hepatic dysfunction’ (as defined by INR > 3) in this study must be confounded by the effect of paracetamol or N-acetylcysteine on coagulation factors.3, 4

  • (d) The authors have over-emphasized a number of the results. Table 2 does not include any confidence intervals (CI) and there is only one result in the whole table that is significant after multivariate analysis. In the analysis of patients who were on beta-agonists and took a paracetamol overdose, the authors state that this was of borderline significance as an independent risk factor for hepatic dysfunction [odds ratio (OR) = 0.34; 95% CI, 0.15–1.07]. As the 95% CI crosses 1.0, this result is clearly not significant.

  • (e) The authors found that the regular use of opioid analgesics had a ‘significant’ effect on paracetamol poisoning, although they only showed this for one of the four outcomes, hepatic dysfunction (as defined by INR > 3, see above). This is concerning because this outcome may simply be a measure of increased INR from paracetamol itself or N-acetylcysteine, and not indicative of hepatic injury. The ORs for the other three outcomes are close to 1.0 and not significant, and the 95% CI for hepatic dysfunction is wide (1.13–25.77), suggesting low power in the study. This selective reporting on the only significant result is insufficient evidence to suggest that a heterogeneous group of opioid analgesics used as regular medication had an effect on paracetamol toxicity.

The clinical relevance of the study by Schmidt and Dalhoff in the management of paracetamol poisoning is also unclear.1 The population investigated included two groups with different severity of paracetamol poisoning. These results should not be applied to the current management of paracetamol poisoning because they cannot be generalized to most hospital emergency department or toxicology units, and their significance is suspect. Further studies in more uniform populations are required, and current evidence does not support the authors' conclusion regarding the effects of regular medications on paracetamol poisoning.


  1. Top of page