Quality improvement collaboratives and the wisdom of crowds: spread explained by perceived success at group level

In 2002, Øvretveit et al. published their ‘lessons from research’ on quality improvement collaboratives (QICs) [1]. They established the need for more research on different types of QICs and their effectiveness. QICs bring ‘…together groups of practitioners from different healthcare organisations to work in a structured way to improve one aspect of the quality of their service. It involves them in a series of meetings to learn about best practice in the area chosen, about quality methods and change ideas, and to share their experiences of making changes in their own local setting.’ (p. 345) [1].

In the last decade, the effectiveness of QICs in stimulating improvement has been evaluated quite extensively. A systematic review concluded that the evidence underlying QICs ‘…is positive but limited and the effects cannot be predicted with great certainty’ and recommends further research into success factors [2]. Recent studies examined determinants for success or failure [3]–[7]. Yet, researchers focussed less on the function that Øvretveit et al. labelled ‘an intentional spread strategy’ [1]. This function is the topic of this article, which is based on data from an independent evaluation of a change programme for hospitals in the Netherlands.

Between 2004 and 2008, 24 hospitals joined a ‘multilevel quality collaborative’, a two-year programme based on a variety of QICs and a leadership programme for executives. In the first year, multidisciplinary teams comprised of health care professionals and managers from each hospital participated in the QICs and implemented improvement projects, pursuing different targets. In the second year, projects were disseminated over new units and patient groups within the hospitals (see Table 1). Whilst implementing the projects, hospitals were expected to enhance their internal quality improvement infrastructure [8],[9].

Before presenting our research question, we provide a brief overview of relevant earlier findings from the programme evaluation.

The black lines in Figure 1 represent the relationships between several domains that we assessed in other studies.

By the end of 2004, a first group of project teams from eight hospitals started with the programme, a second group of eight hospitals started one year later. Study data was collected from two sources: project team leaders (responsible for the implementation of individual projects) and programme coordinators (responsible for the coordinated implementation of the programme in their hospital) (Figure 2). In January 2006 and 2007, 148 project team leaders were asked to rate the overall success level of their project on a scale from 1 to 10. In the last quarter of 2006 and 2007, a questionnaire was sent to the programme coordinator appointed in each hospital for the duration of the programme. The coordinators were requested to fill out a table with the number of units or patient groups where projects were implemented in the second year (the same table as the one described by Dückers et al.; see the Additional file 1) [9]. Based on these data, an average perceived project success and an average second-year dissemination rate were calculated for each of the 12 QICs (six project types, two hospital groups). Associations and differences between groups were tested in SPSS 20.

The study protocol was approved by the Medical Ethics Review Committee of the VU University Medical Centre (registered with the US Office of Human Research Protections as IRB00002991). The study does not fall within the scope of the Medical Research Involving Human Subjects Act.

A total of 84 project team leaders (57%) rated the success of their project and 15 programme coordinators (94%) filled out the dissemination table. Table 2 contains general descriptive information on the data (i.e., mean, median, inter-quartile range, minimum and maximum values) and the average QIC scores. Each average perceived success is accompanied by the number of project team leaders who rated their project’s success level (between parentheses). The highest average perceived success was found among medication safety, process redesign and waiting list projects. The operation theatre QICs reveal the lowest average success and dissemination rare. The dissemination differs significantly between the twelve QICs (Kruskal-Wallis test; p < 0.001) and between both hospital groups (Mann–Whitney U test; p < 0.05). The non-parametric correlation between perceived success and dissemination at QIC level is 0.73 (Spearman’s rho; p < 0.01).

The programme provided a unique opportunity to test whether project successes at the QIC level, based on opinions of individual project team leaders, are indicative for dissemination. The study supports the potential of QICs as multi-organizational intentional spread strategy. QICs are a platform to reach professionals and managers from different organizations. The ability of QICs to generate team success stories appears to have an influence on the intensity of the spread. However, in the end, successful dissemination depends on the capacity of the teams’ supportive home organizations to establish an optimal receptive context [12],[13]. Organizational readiness suggests that people at all levels have sufficient motivation, capability, and opportunity to implement or facilitate necessary changes [14]. This is why different authors recommend a multilevel approach for large-scale implementation and spread [6],[15]. In case of the multilevel quality collaborative, this proved beneficial. What this study adds to our earlier work is an explanation for the impact of QICs on dissemination processes in organizations. Programme developers, managers, and evaluators are invited to look at what individuals have to say, but also to take the ‘wisdom of crowds’ into account [16]. Group data can be informative. The findings open other research avenues to explore, for example, on how policymakers and QIC planners can gather and use information on perceived success when conducting QICs.

Several limitations must be mentioned. Mittman emphasized how expectation biases and belief perseverance produce systematic overweighting of evidence and observations by respondents in QIC evaluations [17]. This is a typical risk [7]. The small sample, moreover, makes it necessary to be cautious with interpreting the results and should be seen as an encouragement to replicate the study in larger, independent samples. It is possible that actors within the second group of hospitals were exposed to or influenced in their dissemination decision-making by developments in and results from the first eight hospitals. Other issues we could not take into account are differences in the relative complexity of projects; the extent to which teams chose suitable interventions for their situation and if the same interventions were actually applied on or applicable to the new units and patient groups (which is in fact a broader operationalization of spread than the merely quantitative one in this article); the history hospitals had with the projects before joining the programme; and the possibility that not all projects were completely finished after year one. It is equally noteworthy that the dissemination potential is likely to depend on hospital size and variety in suitable patient groups, specialities, and locations. Consequently, the dissemination magnitude will likely differ between hospitals.