This was a retrospective cross-sectional survey of the association between characteristics of the regulatory environment and the times to site activation, ethics approval and governance approval for multi-centre clinical trials. The project received ethics approval from the UNSW Sydney Human Research Ethics Committee on the 29th of March 2021. The STROBE checklist was used to standardize this report [10]. The guiding question for the study was “what are the factors associated with site initiation times and how large are their effects?”.

This study used a convenience sample of trial initiations obtained from the database (GrantPlan) of a collaborating Clinical Research Organisation, George Clinical. A primary search of the company records identified a large number of potentially eligible site initiations, though it was recognised from the outset that incompleteness of information in the GrantPlan database, and limited resource available to ascertain missing data from other sources, would limit the final number of site initiations included in the analysis. Data required for eligibility were disease area, trial phase, and country as well as the dates of initial site contact, ethics submission and approval, governance submission and approval, as well as the date of site activation or first patient recruited.

George Clinical team members extracted standard data from the clinical trial management database (GrantPlan), the contracts database and other data repositories held by the company between July 2021 and September 2022. The data were provided to a researcher external to George Clinical who de-identified the information by replacing the project name, principal investigator name and site name with unique identification numbers. The standard set of variables were extracted into an Excel database for each trial [11]. Two researchers also examined the websites, contract templates and governance agreement templates of each trial site to collect standard information about the ethics and governance approval processes applicable to each trial. This information was either publicly available on the trial site’s website or available upon request from their research office. This was done in duplicate and independently for the first half of the sites but since there was a very high alignment of findings was done by just one researcher for the remainder of the sites.

The factors that were considered as potential determinants of site activation time fell into three categories—trial characteristics, site characteristics and characteristics of the ethics and governance processes. The latter variously operated at national, sub-national or site levels. The selection of exposures for evaluation was based on two systematic reviews that assessed factors previously reported to influence time to trial start-up [9, 12]. Site initiations were also grouped into pre-2020 and post-2020 groups to explore the impact of COVID-19.

The key trial characteristics considered were the disease area (nephrology; oncology; neurology; endocrinology; paediatric) and the trial phase (1, 2, 3 or 4). The site characteristics recorded were the ownership status (government or private) and the country (Australia, Hong Kong SAR, South Korea, New Zealand, and Taiwan). Researchers were unable to find the ownership status of many trial sites for the Asian countries. The exposures relating to the ethics review process were the use of scope guidelines by the ethics committee; a requirement for mutual acceptance of other ethics committee findings; and an ethics process that triaged applications as low, medium, or high risk. The exposures relating to the governance review were the use of scope guidelines by the governance body; and a governance approval process that triaged applications as low, medium, or high risk.

An outcome was the time taken to achieve site activation defined as the date that ‘site activation’ was recorded in the clinical trial management system. The date of initial site contact made by the clinical research organisation was used as the date at which the trial start-up process commenced. For 19 trials without a documented ‘site activation’ date, the date of site activation was imputed as the date the first patient was recruited. Additional outcomes were the time from submitting to obtaining ethics approval and the time from submitting to attaining governance approval. The unit of measurement for time was days.

The primary analysis was based on data from 150 site activations including 19 for which missing ‘site activation’ dates were imputed as the date of first patient recruitment. Since the distributions of the times across the included trials were noted to be substantively right-skewed, summary data for site activation, ethics approval and governance approval were reported as medians and ranges. Overall median times and ranges were first summarised for each outcome with all data available and then summarised separately for each subgroup for factors of interest. Differences in the median times across each subset for factors of interest were tested first using univariable regressions and non-parametric tests (Kruskal–Wallis rank sum tests or Wilcoxon rank sum tests) on log-transformed outcome data. We then repeated the analyses using multiple regression methods including all the variables from the univariable analysis, to assess the joint effects of the exposures. Records with missing or unknown values were excluded from relevant analyses with, for example, analysis of the effects of government versus private ownership restricted to Australia because this information could not be obtained for overseas sites. For the analyses of factors influencing time to site activation, a subsidiary analysis was done including only the 131 trials for which a date of site activation was recorded, without the 19 trials with imputed data (Supplementary Table 1). The data was also divided depending upon the year at which trial start-up was commenced (Jan 2014–Dec 2019 versus Jan 2020–Dec 2022) to test for effects of the COVID-19 pandemic. P-values < 0.05 were considered statistically significant. All analyses were done using R version 4.1.0 and RStudio 2022.07.2 Build 576 [13, 14]. Packages “gtsummary” and “ggplot2” were used for summarizing the data [15, 16].