Study selection
We identified 701 articles published from January 2010 to December 2019 in the databases searched (Fig 1) and 11 studies were added from the IHI website. A total of 319 duplicate articles were excluded, leaving 382. Of these, 245 titles were classified for analysis of the abstract. After reading the abstracts, 156 articles were excluded because they did not present at least one criterion of eligibility, leaving 89 articles for complete reading of the publication. Among the reasons for the exclusion of 55 articles were: (11) did not present pre- and post-data; (11) were not collaborative to improve quality; (10) were only a method/process description; (7) were studies on lessons learnt; (6) were descriptions of a project/network; (6) were non-accessible articles; (2) were not about health services; (1) was a collaborative evaluation; (1) was a study protocol.
Study characteristics
There were 34 studies [15-48] included for data extraction and qualitative analysis, of which only one was a randomized controlled trial, one was quasi-experimental with control, one was an observational study, and one was a multiple case study. All other studies were of the quasi-experimental type without a control group with time series analysis, in accordance with the BTS model of the IHI. Most studies were carried out in the USA (22), followed by countries in Africa (5), the United Kingdom (3), The Netherlands (2), Canada (1), and Indonesia (1), all in the English language. Two studies involved 2 or more countries.[43,46]
The great diversity of problems addressed in health structures with different levels of complexity and participation of professionals, sometimes from management, sometimes from patient care, characterised the set of complex interventions in this review. The characteristics of the studies are detailed in Table A - see Additional file 1 for the original data used to perform this analysis.
The number of institutions included in the collaboratives ranged from 1 to 744, with an average duration of the intervention of 20 months (ranging from 6 to 60 months). The number of health professionals participating in the collaboratives was mentioned in 16 studies and amounted to between 35 and 2000 professionals.
International patient safety goals were addressed in 10 studies, five focused on improved communication between teams and/or with the patient/family,[19,20, 26,37,39] three addressed falls reduction,[35,36,47] and two considered health care-related infection prevention.[30,38] Improvement of patient safety culture was the subject of three studies.[23,25,36] Improving care in different clinical conditions was the subject of 14 reports such as: neonatal death,[42] care with tracheostomies,[46] diagnosis,[41] child health,[18,27,45] HIV transmission or treatment,[16,22,43] mental health,[24] women's health,[31,34,44] and stroke.[48] Prevention of adverse events was the subject of four studies.[28,32,33,36]
Collaboratives were effective in improving the intended outcomes by 76.5% (26/34),[16,17,20-27, 29-34,36-38,40-42,44-47] reached partial results in 14.7 % (5 /34),[19, 28, 35,43,48] and did not obtain the desired improvement in 8.8% (3/34).[15,18,39]
Improvement sustainability was reported in 10 of the 26 studies with positive results.[22-24,29,30,32,34,36,38,42]
NTS such as leadership, teamwork, and communication are briefly cited in the discussions of 58.8% (20/34) of the studies as contributing factors for improvement. However, only five studies evaluated improvement in NTS among the participating professionals/teams and none mentioned or assessed their contribution to the dissemination and sustainability of improvement (Table 1- See Additional file 1). [19,28,37,42,46]
In the study by Stevens et al. (2010), the participants evaluated their improvement with the following parameters: delivery system design, decision support, clinical information systems, patient self-management, integration, healthcare system organisation, and community linkages. There was an improvement in the processes with a slight improvement in the clinical results.[19]
In the study by Zukoff et al. (2014) evaluation questionnaires were applied to the teams, not to the professionals individually, pre- and post-intervention, to measure the effort of the improvement intervention and the functioning of the teams, not identifying differences in their perception in terms of work improvement in a team.[28]
Nieuwsma et al. (2017) observed that chaplains and mental health professionals improved communication that favoured patient care.[37]
For Werdenberg et al. (2018) the NTS were important to the success of the proposed improvement objectives.[42]
The study by Bedwell et al. (2018) highlighted that nurses' knowledge focused on improving tracheostomy care increase in the following points: general knowledge, identifying emergencies, performance, situation, goals, coping.[46]
Risk of bias within and across studies
The selection of studies restricted to those with pre- and post-intervention data, may have generated a selection bias, insofar as it may have excluded qualitative or secondary studies that assessed the participation of people in collaboratives, which would likely bring elements related to NTS, the object of this study. Another risk of bias within the studies is the selection of participants, which, although voluntary, is directed towards team leaders, who are presumed to be professionals more open to change and who have a greater potential for engagement.
A common feature of most studies is the use of statistical control charts to measure the improvement achieved in the intervention. In accordance with the BTS model, the studies are not designed to evaluate the effectiveness of a new process, but to apply in practice what one wants to adapt, based on evidence already available in the literature. This was identified in most studies in this review.
The evaluation of compliance with Squire 2.0 items as a quality requirement for the study, showed that only seven (20.6%) of 34 studies met the six items under analysis.[21,22,25,29,34,39,42] Three studies did not present contextual elements in the methodology[18,20,35] and three others did not provide any information about the participating professionals.[19,26,38] A lack of recording in the results about unintended consequences, such as unexpected benefits, problems, failures, or costs associated with the intervention or details about lost data, was observed in 38% of the studies. Thus, 44.1% of the articles were classified as high quality with 5 to 6 Squire items;[15,21-23,25,26,28,29,34,38,39,41,42,45,48] 44.1% were classified as average quality with 3 to 4 Squire items[16-20,24,27, 30-33,37,40,43,44] and 11.8% were classified as low quality with 1 to 2 Squire items.[35,36,46,47]