PICO, PICOS and SPIDER: a comparison study of specificity and sensitivity in three search tools for qualitative systematic reviews

1Systematic reviews are a crucial method, underpinning evidence based practice and informing health care decisions. Traditionally systematic reviews are completed using an objective and primarily quantitative approach whereby a comprehensive search is conducted, attempting to identify all relevant articles which are then integrated and assimilated through statistical analysis. The comprehensiveness of the search process has been viewed as a key factor in preventing bias and providing a true representation of available research. Current research investigating the process of quantitative systematic reviews therefore focuses on methods for ensuring the most comprehensive and bias free searches possible. Because of the time and resources required to complete a systematic and comprehensive search, efforts have been made to investigate the sensitivity of searches, and thus lessen the amount of time spent reviewing irrelevant articles with no benefit.

2However, conducting comprehensive searches also forms the bedrock of qualitative or narrative reviews, now commonly referred to as qualitative evidence syntheses. Qualitative evidence syntheses are now acknowledged as a necessary and valuable type of information to answer health services research questions. However, difficulties in completing a sensitive yet comprehensive search of qualitative literature have been previously noted including: poor indexing and use of key words of qualitative studies, the common use of titles that lack the keywords describing the article, and unstructured abstracts.

3When devising a search strategy, a search tool is used as an organising framework to list terms by the main concepts in the search question, especially in teams where it is not possible to have an experienced information specialist as a member of the review team. The PICO tool focuses on the Population, Intervention, Comparison and Outcomes of a (usually quantitative) article. It is commonly used to identify components of clinical evidence for systematic reviews in evidence based medicine and is endorsed by the Cochrane Collaboration. Due to its target literature base several of these search terms such as “control group” and “intervention” are not relevant to qualitative research which traditionally does not utilise control groups or interventions, and therefore may not appropriately locate qualitative research. However, these terms may become more relevant in the future as more trials and interventions incorporate qualitative research.

4As the PICO tool does not currently accommodate terms relating to qualitative research or specific qualitative designs, it has often been modified in practice to “PICOS” where the “S” refers to the Study design, thus limiting the number of irrelevant articles.

5Cooke et al. also addressed this issue of relevance by developing a new search tool entitled “SPIDER” (sample, phenomenon of interest, design, evaluation, research type), designed specifically to identify relevant qualitative and mixed-method studies. The key features and differences of the SPIDER and PICO search tools are shown in Table 1. The addition of the “design” and “research type” categories to the SPIDER tool was intended to further increase the ability of this tool to identify qualitative articles, whilst removing irrelevant PICO categories such as the “comparison” group.

6Cooke et al. recommended that the SPIDER tool was tested further in qualitative literature searches. Although it has been used previously in a scoping review to investigate gaps in an evidence base on community participation in rural health care, SPIDER has not yet been tested and evaluated in a qualitative systematic narrative review context. The authors of this article recently completed a systematic review of the qualitative research investigating experiences of health care services for people with Multiple Sclerosis. On embarking on this review topic we faced many of the difficulties commonly discussed in identifying qualitative literature on a given topic, and identified SPIDER as a potential way of overcoming some of these difficulties. Therefore, the aim of this article was to test SPIDER by broadly replicating the work of Cooke et al., specifically by comparing the two approaches: 1) the traditional PICO method of searching electronic databases with 2) the newly devised SPIDER tool, developed for qualitative and mixed-method research. In addition we wished to build and expand on the work of Cooke et al. and so our third aim was to compare PICO and SPIDER to a modified PICO with qualitative study designs (PICOS, see Table 1 by investigating specificity and sensitivity across 3 major databases.

7Studies eligible for inclusion were those that qualitatively investigated patients’ experiences, views, attitudes to and perceptions of health care services for Multiple Sclerosis. No date restriction was imposed on searches as this was an original review. Qualitative research, for this purpose, was defined by the Cochrane qualitative methods group as using both a qualitative data collection method and qualitative analysis. Quantitative and mixed method studies were therefore excluded.

8We define experience as “Patients’ reports of how care was organised and delivered to meet their needs p.301”. Patients’ reports could refer to either experience of health care services delivery and organisation overall or their experiences of care by specific health care personnel. We included studies that investigated adults (aged 18 years old and older) with a diagnosis of Multiple Sclerosis, who had experience of utilising health care services at any time point. There were no restrictions on subtype of Multiple Sclerosis, gender, ethnicity or frequency of use of health care. Health care in this sense referred to routine clinical care (either state funded or privately funded) not trial protocols or interventions. Excluded studies included studies that focussed on self-management and studies that investigated quality of life.

9Because of the focus on Multiple Sclerosis, studies were excluded if they used a mixed sample of various conditions (e.g. studies reported a mixed sample of people with neurological conditions) or if they used a sample of mixed respondents (i.e. people with Multiple Sclerosis and their carers) where results of patients with Multiple Sclerosis could not be clearly separated. If an article had a section or subtheme on health care services but this was not the main research area of the article, then that article was included; however only data from the relevant subtheme were extracted and included in the findings. Additional exclusion criteria were articles that only described carer or health care professional experiences not patient experiences. Conference abstracts, editorials and commentaries were not included.

10For this systematic search we developed a detailed search strategy in collaboration with a specialist librarian and information specialist. This search strategy was tailored to the three largest medical and nursing databases (Ovid MEDLINE, Ovid EMBASE, and EBSCO CINAHL Plus) as in Cooke et al.’s study and search terms used a mixture of medical subject headings and keywords. To investigate the benefit of the SPIDER,PICO and PICOS tools we used identical search terms but combined them in different ways as shown in Tables 2, 3 and 4 below.

11One reviewer judged titles and abstracts against the inclusion criteria. If a title and abstract met the inclusion criteria then full text copies of all articles were retrieved for further investigation. Two authors reviewed these full text articles independently for relevance to the search aim (i.e. patients/service users with multiple sclerosis, experiences of health care services and qualitative research). Any disagreements were resolved via discussion. Data from included studies were extracted by both reviewers independently to ensure accuracy and then stored on a Microsoft Excel spread sheet. No ethical approval was required for this study.

12All searches spanned from database inception until 12th October 2013. As in Cooke et al., we reviewed our findings based on two metrics; the number of hits generated and of these, the number relevant to the search aim (see Table 5).

13As found in Cooke et al., PICO created a much greater number of hits compared to SPIDER. A total of 23758 hits were generated using PICO, 448 hits were generated using PICOS and 239 hits were generated using SPIDER. Overall, the average reduction of hits (% across all three databases) was 98.58% for SPIDER vs. PICO, 97.94% for PICO vs. PICOS and 68.64% for PICOS vs. SPIDER. The time spent screening hits for relevant articles equated to weeks for the PICO hits and hours for the PICOS and SPIDER hits.

14Articles which met the inclusion criteria after full text review are displayed in Table 6. Examination of the titles and abstracts of the identified articles resulted in the obtainment of 18 full text articles relevant at full text, across all databases and search tools.

15For the PICO tool in CINAHL Plus, 5.78% of hits were deemed relevant after the title and abstract stage (78 articles/1350 articles), and 14/78 articles (17.95%) were confirmed to meet the inclusion criteria after full text review. For the PICO tool in MEDLINE, 0.42% of hits were deemed relevant after the title and abstract stage (34 articles/8158 articles) and 12/34 (35.29%) articles were confirmed to meet the inclusion criteria after full text review. For the PICO tool in EMBASE, 0.25% hits were deemed relevant after the title and abstract stage (35 articles/ 14250 articles) and 14/35(40%) articles were confirmed to meet the inclusion criteria after full text review.

16For the PICOS tool in CINAHL Plus, 38.36% of articles were relevant after the title and abstract stage (56 articles/146 articles) and 12/56 (21.43%) were confirmed to meet the inclusion criteria after full text review. For the PICOS tool in MEDLINE 14.16% of articles were relevant after the title and abstract stage (16 articles/ 113 articles) and 6/16 (37.5%) were confirmed to meet the inclusion criteria after full text review. For the PICOS tool in EMBASE 7.94% of articles were deemed relevant after the title and abstract stage (15 articles/189 articles) and 7/15 (46.67%) were confirmed to meet the inclusion criteria after full text review.

17For the SPIDER tool in CINAHL Plus 38.36% of articles were relevant after the title and abstract stage (56 articles/146 articles) and 12/56 (21.43%) were confirmed to meet the inclusion criteria after full text review. For the SPIDER tool in MEDLINE, 36.81% hits were deemed relevant at the title stage (14 articles/38 articles) and 5/14 articles (35.71%) were confirmed to meet the inclusion criteria after full text review. For the SPIDER tool in EMBASE, 16.36% were relevant at the title stage (9 articles/55 articles) and 3/9 (33.33%) were confirmed to meet the inclusion criteria after full text review.

18The SPIDER tool identified 13 relevant articles out of 239 articles across all three databases (5.43%) compared to PICOS which identified 13 articles out of 448 articles (2.90%) and PICO which identified 18 articles out of 23758 articles (0.076%). Of the 18 relevant articles identified by the PICO tool, 66.66% came from both MEDLINE and CINAHL Plus (12 articles each), and 72.22% came from EMBASE (13 articles). Of the 13 relevant articles identified by the PICOS tool 46.15% came from MEDLINE (6 articles), 53.84% came from EMBASE (7 articles) and 92.31% came from CINAHL Plus (12 articles). Of the 13 relevant articles identified by SPIDER, 38.46% came from MEDLINE (5 articles) and 23.07% came from EMBASE (3 articles) and 92.30% came from CINAHL Plus (12 articles) Table 7.

19Different articles were found across different tools and databases (as shown in Table 6). All three databases were checked for all articles. One article was available in CINAHL Plus but not identified by any of the tools. Two papers were identified in all databases through all search tools. Five papers were identified in MEDLINE through all search tools, three identified in EMBASE through all search tools and 12 identified in CINAHL through all search tools. Five papers were identified solely in CINAHL Plus, with one of these papers only identified using the PICO search method. One paper was identified by all search tools in EMBASE but not identified by any in MEDLINE. No new studies were identified using the SPIDER or PICOS tools alone in any database.

20In this article we addressed the aim of replicating a comparison between the SPIDER, PICOS and PICO search tools. As previously described in Cooke et al., the SPIDER tool produced a greatly reduced number of initial hits to sift through, however in this study it missed five studies that were identified through the PICO method. This may be partly be explained by the nature of the research question prompting the search. As this study included subthemes of studies whose focus differed from the initial research question (i.e. only a smaller section of the paper related to health care) then it’s possible that these studies were picked up by a broader search but not the highly specific SPIDER search. Other authors researching the process of qualitative literature reviews have previously commented that there appears to be a decision to be made about the benefits of comprehensiveness of findings versus the accuracy of the studies identified. Given the common nature of using sub-sections of papers for systematic reviews then our findings suggest that comprehensiveness needs to be the key for this type of search.

21The PICOS tool was more specific than the PICO tool, but did not identify any additional relevant hits to the SPIDER tool, suggesting it is of approximately equal sensitivity. PICOS identified the same number of papers as the SPIDER tool and both demonstrated a substantially lower number of hits generated than a regular PICO search. The SPIDER tool showed the greatest specificity due the small number of hits generated. This may mean that review teams with very limited resources or time, and who are not aiming for a totally comprehensive search (i.e. in the case of scoping studies), would benefit from using the SPIDER tool. This might be applicable particularly to studies such as qualitative syntheses, where the research aim is theoretical saturation, not a comprehensive search. In addition, articles written to influence policy often require swift publication, providing another area in which either SPIDER or PICOS might improve current practice.

22The issue of time was also related to the number of relevant articles identified per database. Whilst EMBASE generated nearly twice as many hits as MEDLINE, only one additional paper was found. The PICO tool identified all articles, suggesting that where time is not a factor, it might be of more benefit to use this tool, as SPIDER demonstrated lower sensitivity, did not identify any new articles and identified fewer relevant articles than PICO.

23Our findings indicate that it is worthwhile testing a chosen search tool across various databases as they produce different results; i.e. CINHAL Plus identified papers not identified in MEDLINE or EMBASE databases. It is therefore important for future research to investigate the potential of the SPIDER vs. PICOS and PICO tools as a base for the recommended comprehensive searching process, by investigating the contribution of the SPIDER and PICOS tools at every stage from the initial search hits, to the final included relevant articles.

24As CINAHL is a database dedicated to nursing and allied health research, it was expected that it would produce a greater number of relevant articles than more medically focussed databases, as nursing and allied areas have traditionally been at the forefront of qualitative investigations into Multiple Sclerosis.

25SPIDER proved to be a tool designed to formulate search terms easily, as it naturally fits the crucial elements of the search question. However, even though some qualitative keywords are necessary to identify qualitative studies, including the words “qualitative research” AND the name of the type of research e.g. “grounded theory” might be too restrictive, particularly given the poor use of the qualitative index term, and might partially explain the fewer studies identified by SPIDER in comparison to PICO. Studies not identified by the SPIDER model in MEDLINE and EMBASE databases did not use keywords such as “qualitative”, but some described qualitative methods, such as “phenomenological-hermeneutic” or “interview(s)”.

26In all PICO searches for MEDLINE and EMBASE the word “qualitative” combined with the phrase “multiple sclerosis” identified many quantitative studies reporting brain scan assessments that were wholly unrelated to the search aim. This was because the word “qualitative” in this context referred to using a qualitative method to provide information about the quality of the scan and any potential flaws. This caused a problem with specificity, resulting in thousands of inappropriate hits as there was no way to exclude studies with the word “qualitative” unless all articles clearly utilised and indexed qualitative research methods in the title, abstract and keywords.

27Many studies were excluded at the full text stage on the basis that the samples were mixed: being comprised of either various neurological conditions or mixed groups of people i.e. patients and carers/patients and health care professionals and so forth. Without clearer titles and abstracts, and potentially an indexing phrase that indicates mixed samples, there is no way of avoiding this issue. Excluding the phrases “caregivers” or “health care professionals” would have excluded any studies that used these phrases (for example in the introduction or implication for future research sections) and therefore it is difficult to see how this could be prevented. A strength and limitation of our study is that whilst it details a real world example of evidence searching, it only addresses one topic. Further research should test these search tools against a wider variety of narrative review and meta-synthesis topics.

28SPIDER greatly reduced the initial number of articles identified on a given search due to increased specificity, however because of lower sensitivity omitted many relevant papers. The PICOS tool resulted in an overall more sensitive search, but still demonstrated poor specificity on this topic. Further investigations of the specificity and sensitivity of SPIDER and PICOS on varied topics will be of benefit to research teams with limited time and resources or articles necessary to impact on policy or change current practice. However, where comprehensiveness is a key factor we suggest that the PICO tool should be used preferentially. Part of the lower identification rate for SPIDER (in comparison to PICO) was poor labelling and use of qualitative keywords in indexing studies. As both individual research submissions and journal/database indexers improve, or standardise, the indexing of qualitative studies, it is likely that the relevance of the SPIDER tool will increase. The recommendation for current practice therefore is to use the PICO tool across a variety of databases. In this article we have shown that SPIDER is relevant for those researchers completing systematic narrative reviews of qualitative literature but not as effective as PICO. Future research should investigate the use of SPIDER and PICOS across varied databases.