We thank Dr Jhirwal et al for their interest in our pragmatic trial. In this reply we shall address only those issues that have not already been covered in response to earlier comments.

The first concern regarded the omission of a placebo arm. A placebo group was initially considered but abandoned as clinicians felt this was unethical, difficult to justify and likely to pose practical difficulties in implementation. This was a pragmatic trial and the design was driven by what questions clinicians wanted answered and what interventions they (and the institution's ethics committee) would permit. Moreover, systematic reviews reveal no evidence that placebo interventions in general have clinically important effects, and the role of placebos in clinical trials, apart from helping to minimise bias, is questionable (Reference Hróabjartsson and GøstzscheHróbjartsson & Gøtzsche, 2004). Our pragmatic trial utilised adequate allocation concealment and masking of primary outcome assessors, two crucial features of trial design that significantly affect the internal validity of a randomised controlled trial (Reference Jüni, Altman and EggerJüni et al, 2001).

Those with toxic states, epilepsy or other organic conditions were invariably excluded from the study as treating clinicians were uncomfortable about their inclusion in a randomised trial with sedative agents. Investigation results are rarely available before the intervention is instituted for violent patients under normal conditions of clinical practice.

All those subjected to tranquillisation received standard levels of care that included monitoring of vital signs and intensive nursing support. Any adverse events with regard to autonomic instability were promptly reported. Only two patients on lorazepam reported any adverse events and this is described in the discussion (paragraph 4, page 65; a printing error in Table 2 ascribes this to the combination group instead of to lorazepam).

Table 2 records that equal numbers in both groups were given additional medication (always a single dose of 100 mg chlorpromazine) and this contradicts the speculation that differences in additional medication could have influenced improvement in favour of any particular group. The proportions that failed to sleep or were never tranquil reported in paragraph 1 on page 65 are correct. Table 2 reports the numbers who were tranquil/asleep and asleep at the times when outcomes were recorded. People who were tranquil or asleep at one assessment did not invariably remain so at other assessments, hence explaining the apparent discrepancy. We acknowledge the error in Table 5 where the proportion asleep at 120 min should be 69% and not 88% for TREC-India.

We hope the interest aroused by this paper will prompt greater use of trials of pragmatic design, free of industry sponsorship and aimed at answering clinical questions of relevance to real world clinical practice.