Assessing the costs of healthcare technologies in clinical trials.

s were read and included where inclusion criteria were met. Articles were then skim read and included in the review where inclusion criteria were met. The aim was not to record the frequencies with which issues were raised but to be compre- hensive in identifying issues, so the final list of references represents all issues rather than all relevant articles discussing the issues. In-house search The second stage of the review was an in-house search of the relevant literature. The in-house database was organised by researchers in the field of economic evaluation in health care, so it pro- vided a useful starting point for the identification of articles for review, search terms and methodo- logical issues. The in-house database at the Health Economics Research Group, Brunel University, holds over 5000 references organised by using bibliographic software (Pro-cite version 2.21). The basic search terms used for the in-house searches were economic*, cost* or methodology* or trial* (where * indicates a truncated search term). Articles were retrieved when the inclusion criteria were met. Manual searches Manual searching of journals was also conducted. All journals were searched manually from 1990 until October 1997 unless they were first estab- lished after 1990. Where this was the case, the first year is presented in parentheses. The following journals were included: • British Journal of Medical Economics (to vol. 11 (1)) • Controlled Clinical Trials (to vol. 18 (5)) • Drug Information Journal (to vol. 31 (3)) • Health Economics (from 1992 to vol. 6 (5)) • Health Policy (to vol. 41 (2)) • International Journal of Technology Assessment in Health Care (to vol. 13 (3)) • Journal of Clinical Epidemiology (to vol. 50 (10)) • Journal of Epidemiology and Community Health (to vol. 51 (5)) • Journal of Health Economics (to vol. 16 (2)) • Medical Care (to vol. 35 (10)) • Medical Decision Making (to vol. 17 (3)) • PharmacoEconomics (from 1992 to vol. 12 (4)). Refinement of search strategy Following the in-house and manual searches, the articles retrieved were then skim read with a view to refining the search strategy and devising search terms. In developing the search strategy, use was made of truncated terms (*). Operators such as ‘and’, ‘or’ were also used and the operator ‘next to’ was used for expressions, such as economic evaluation, which became: economic ‘next to’ evaluation. The use of medical subject headings such as ‘costs and costs analysis’ and ‘economic’ were found to be too broad. The use of cost* in abstracts picked up too many articles concerned only with making minor comments about the cost implications of studies. The search terms were combined with their respective operators and further combined into sets: 001 cost* in title 002 cost-effective* in title or abstract 003 economic next to evaluation in title or abstract 004 methodology* in title or abstract 005 randomised controlled trial* in title or abstract 006 1 or 2 or 3 007 5 and 6 008 4 and 7 Electronic searches The search strategy detailed above was used for electronic searches, which were conducted simultaneously on MEDLINE, EMBASE and Healthstar (health administration database) via Dialog software. The simultaneous searching of the three databases permitted the easy identifi- cation of duplicate articles. Further searches were conducted on the Health Economic Evaluations Database (Office of Health Economics). The literature searches were limited to English language articles for the period 1986–1996 and excluded animal studies. Review of key articles and identification of key methodological issues The articles retrieved from the search were then reviewed in order to identify the methodo- logical issues. The issues were structured into four categories: study design, data collection, Health Technology Assessment 1999; Vol. 3: No. 6 53 data analysis and presentation of results. These categories also formed the outline of the report. Issue-specific searches The aim was to identify papers discussing a specific methodological issue or issues that are relevant to data collection and analysis for costing alongside clinical trials. This required a targeted search strategy; so, for each key issue, search terms were identified. Electronic searches were conducted for each issue on EMBASE, MEDLINE and Healthstar databases via Dialog. The issues identified were resource-use data collection and analysis issues (sample size, cost- effectiveness ratios, confidence intervals, sensitivity analysis, extrapolation and generalisation). The search strategy was as follows: 001 statistic* in title or abstract 002 stochastic in title or abstract 003 cost-effective* next to ratio* in title or abstract 004 extrapolat* in title or abstract 005 sensitivity next to analys* in title or abstract 006 generali* next to cost* in title or abstract 007 missing next to data in title or abstract 008 pool* next to data in title or abstract 009 cost-effective* in title or abstract 010 1 or 2 or 3 011 4 or 5 or 6 or 7 or 8 012 9 and 10 014 9 and 11 These search terms were then combined with those from the general search. Citation searches For each key issue, key articles were selected and used in the citation searches, which were conducted on EMBASE. The articles selected were those thought to be the earliest ones discussing the methodological issues. Identified articles were retrieved if they met the inclusion criteria and were not duplicates. Reference lists All reference lists from previously identified articles were reviewed in order to identify other relevant articles, subject to the inclusion criteria. Articles from experts In the consultation phase of the report, experts were asked whether any key articles were missing from those identified by the searches and, if so, to identify them. Review of articles The articles were photocopied and details entered on Pro-cite (a bibliographic software package) with keywords assigned for the source (in-house, handsearch, electronic search with database) and for the type of article (methodological, empirical, statistical, other review, data collection). The allocation of source keywords enabled easy cross- checking of in-house articles with those retrieved from the electronic searches. For each stage of the identification process, the number of articles identified and the number included were recorded. The results are shown in Table 2.