The proportional odds assumption will be examined when sufficient data have been collected. Should it not appear to hold we will perform logistic regression analyses yielding separate odds ratios for (i) all or some requested information versus none, and (ii) all versus some or none. The former will be the primary outcome.

Stopping rule
An advantage of the Bayesian framework[26] is that a trial may be continuously monitored without fear of obtaining spuriously significant results. First Contact will be analysed at regular intervals and the results will be made available to all participants (though not before results from six different reviewers are available). Participants will therefore be able to use the results to change the certainty or uncertainty they have about how to contact investigators. The trial will continue for a minimum of two years. Thereafter it will be stopped when potential participants are sufficiently certain of the superiority of one intervention, or the equivalence of the two, to refrain from randomising. As a last resort, the trial will be stopped when the primary analysis indicates that (i) there is 99% probability that the odds ratio is greater than 1; or (ii) there is 99% probability that the odds ratio is less than 1; or (iii) there is 99% probability that the odds ratio lies between 0.9 and 1.1.

Sample size issues
Though the primary analysis will take a Bayesian approach, indications of appropriate sample sizes[27] are given in the table below. These are based on the supposition that 50% of investigators will respond and 20% will provide complete information, though the sample sizes are robust to some variation in these[27]. The calculations indicate the following for a trial with 80% power and a 5% significance level: to detect an increase in the non-response rate from 50% to 60%, 663 investigators would need to be approached; to detect an increase from 50% to 64% would require 224 investigators.

		Control response (proportion)			Experimental response (proportion)
sig. level	power	none	some	full	none	some	full	Odds ratio	Sample size
5%	80%	0.5	0.3	0.2	0.40	0.33	0.27	1.5	663
5%	80%	0.5	0.3	0.2	0.36	0.33	0.31	1.75	345
5%	80%	0.5	0.3	0.2	0.33	0.33	0.33	2	224
5%	90%	0.5	0.3	0.2	0.40	0.33	0.27	1.5	887
5%	90%	0.5	0.3	0.2	0.36	0.33	0.31	1.75	462
5%	90%	0.5	0.3	0.2	0.33	0.33	0.33	2	298

The initial recruitment target for First Contact is approximately 450 investigators though, as described above, the Bayesian approach to the analysis renders sample size issues less important. Assuming a minimum average of six investigators per reviewer this would require the participation of 75 reviewers. The first issue of the Cochrane Database of Systematic Reviews in 2000 contained 672 protocols, 98 of which were new to that issue. Assuming a consistent rate of reviews being undertaken, a recruitment of 10% of Cochrane reviewers would see the target reached within two years. These estimates are considered conservative, given the expected recruitment of keen reviewers with many more than six investigators to contact.

Secondary outcomes
The first of the secondary outcomes (time to response) will be analysed using standard log-rank "survival" analyses techniques for time to first response (of any type) and time to first complete response. The second secondary outcome (contact made, irrespective of data retrieval) will be analysed using an overall odds ratio. The cost and time outcomes will be analysed using t-tests. Postage and telephone calls will be transformed to equivalent costs in the United Kingdom.

A further secondary analysis will be performed if the trial stops in favour of the experimental intervention. Logistic regression will be used to determine whether response rates differ according to (i) time since publication and (ii) request for information or data (the distinction being that the investigator would normally be able to provide the former without needing to refer to files). The dichotomised outcome of some or all information retrieved versus no information retrieved will be used for this analysis.

Exploratory analyses may be undertaken if the trial stops in favour of the experimental intervention in order to generate hypotheses concerning which aspects of the experimental intervention are responsible for the improved response. Pre-notification and follow-up may be investigated by examining the time at which responses were received. Investigation of other factors is dependent on there being variability in how the experimental intervention is implemented.

Bayesian analyses of the primary outcome will be undertaken using a community subjective prior distribution elicited from members of the First Contact Steering Group, participants in First Contact and experienced systematic reviewers. All elicitation will take place before any results are available.