McKnight et al assessed the AEs of lithium on the thyroid, as reported by 77 studies (including 4 RCTs) with varying methodological approaches. The risk of clinical, subclinical and laboratory (as measured by thyroid-stimulating hormone levels) hypothyroidism was increased by five times. As McKnight et al. point out, it was difficult to include some papers published more than 30 years ago because of difficulty in comparing the laboratory resultss. The meta-analysis also found that serum parathormone and calcium levels increased by about 10%. Overall, a quarter of patients had abnormalities of the thyroid and/or the parathyroid gland. Of practical importance is the finding that stimulation of parathyroid function is likely to be more common than assumed to date. Regular monitoring of serum calcium levels should, therefore, be mandatory, and if found to be consistently raised, they should be properly managed .
In the meta-analysis, long-term lithium treatment was found to produce modest but significant weight gain, with an odds ratio of 1.89, and this was distinctly lower than the weight gain induced by olanzapine. Weight gain can be a problem with lithium treatment, but it is less pronounced than with the most frequently prescribed 'atypical' neuroleptics . The latter are prescribed increasingly for patients with BD [12, 13], and serious concernshave already been expressed about this therapeutic strategy and its unfortunate metabolic effects [14–17]. Furthermore, unlike atypical neuroleptics, lithium has not been found to induce diabetes mellitus.
McKnight et al. mention that hair loss was described mostly in case reports; in two RCTs, hair loss occurred in 3% to 8% of patients on lithium, compared with 0% to 6% in the placebo group. Surprisingly, McKnight et.al. could not find sufficient evidence for the existence of lithium-induced skin abnormalities; however, a long series of case reports (available online in the appendix of their paper) tends to contradict their statistical finding.
We feel, and McKnight et al. readily admit, that their meta-analysis has limitations, which include both the nature of the available data and the non-consideration of potential confounding variables, and therefore, is likely to lead to a biased estimation of important AEs. The claim by McKnight et al that they undertook 'a clinically informative systematic toxicity profile of lithium' is only partly fulfilled because the group failed to assess and discuss many of the AEs of lithium that are particularly important to patients and may influence compliance, such as gastrointestinal problems, tremor, cognitive impairments, ataxia, or speech disorders. To give one example: in the German spontaneous reporting system of adverse drug reactions, tremor accounted for nearly 9% of the 654 reports (received until 10 October 2008, referring to lithium mono as well as combined treatment), changed gait and ataxia for more than 10%, and confusional states for 9.5% (Drug Commission of the German Medical Association, personal communication).
Below, we discuss several specific issues that need to be considered critically.
Practicing clinicians are primarily interested in the potential AEs experienced by properly treated patients, that is, patients who are correctly selected for long-term lithium treatment, are maintained on a minimum effective dosage, and are regularly monitored and managed. In this context, lithium treatment generates AEs in a relatively small proportion of patients. However, the reports included in the meta-analysis unavoidably included numerous patients with bipolar spectrum disorders for whom lithium treatment was not the optimal treatment choice , and who were treated by physicians with insufficient experience in lithium treatment. Hence, an important aspect of future work would be to compare the rates of side-effects in specialized programs as opposed to general psychiatric practice.
Sufficient information about dosage and serum levels of lithium could not be taken into account although the authors made efforts to exclude patients with pre-existing lithium intoxication. Many side-effects, including endocrine and metabolic effects are dose-dependent, and therefore can often be controlled by proper dose titration. Details on patient compliance and the time intervals between lithium plasma levels and AEs would have been important background information, and their absence raises questions about the validity of the data and the statistical processing. This criticism was also raised by Malhi and Berk  in the same issue of the journal. As clinicians, we often come across referred patients whose dosage has not been adjusted for many years, and they frequently experience side-effects because their lithium levels are frequently much higher than 1.0 mmol/l.
The potential confounding factors of age and sex on lithium treatment could not be considered adequately in the meta-analysis.
Another important piece of information for the correct interpretation of side-effects is the patient's clinical response to long-term lithium treatment. Non-responders develop side-effects distinctly more frequently than responders, even at comparable serum lithium levels .
There is also the problem of potentially harmful co-medication. McKnight et al. did not elaborate on whether co-prescription of potentially nephrotoxic drugs (for example, antibiotics, cyclooxygenase-2 inhibitors) or compounds that can change the pharmacokinetics of lithium was controlled for, as this can be a strong potential confounding factor particularly with regard to renal toxicity.
Finally, it remains unclear whether AEs such as impaired thyroid function occurred during lithium medication given alone or were seen during treatment with thyroxin supplementation.