Quality improvement initiatives for hospitalised small and sick newborns in low- and middle-income countries: a systematic review

Studies were included if they met the following criteria:

The year cut-off was chosen in order to focus the review on recent practice in the context of changing healthcare systems. Studies that focused solely or primarily on practices in the delivery room that encompassed small/sick newborns only as a subset, or studies that focused on community interventions, were deemed not to meet the eligibility criteria of hospitalised small and sick newborns. Outcomes of self-assessed competency, or patient-assessed satisfaction, were also deemed not to meet the eligibility criteria of clinical outcomes, as this method of assessment was seen to be an insufficient proxy for an objectively measured clinical outcome. Finally, in order to simplify the review, studies focusing only on implementation of Kangaroo Mother Care (KMC), a specific method of care for preterm infants that focuses on encouraging skin-to-skin contact between mother and infant, were not included unless the study involved modification to KMC implementation, because KMC has been well covered in systematic reviews elsewhere [27, 28].

We searched the following electronic databases from January 2000 onwards: Medline, EMBASE, WHO Global Health Library, and Cochrane Library. We searched the trial registries: WHO ICTRP and ClinicalTrials.​gov for completed and ongoing studies. Searches were conducted in April 2017. The literature searches of peer-reviewed publications were supplemented by scanning the reference lists of relevant studies and systematic reviews. Interlibrary lending was used to access certain papers.

The following search strategy was designed to capture studies that were suitable for inclusion in the review. For example, the search strings used in EMBASE were:

The titles and abstracts were screened by two researchers independently for inclusion/exclusion, with disagreements resolved with arbitration from a third reviewer. The results shortlisted for inclusion and then underwent full-text screening, again undertaken by two researchers independently with arbitration from a third, to produce a final shortlist of articles to be included in the review. Data were extracted from each paper by one researcher and checked by a second, using a piloted worksheet, the details of which are supplied in Additional file 1. Summary results from fields that are not presented in the main manuscript are available on request.

To standardise study classifications, the following definitions were used, based on NICE definitions [29]:

QI approaches were classified according to the Kruk and Gage schema ‘Synthesizing improvement approaches’ [10]. This classifies approaches at the micro, meso, or macro level, meaning at the level of the patient-provider interaction, such as on the sick newborn care unit; at the district and clinic level; at an individual hospital; or at the above district level, such as across a health system. It then provides sub-classifications of approaches within each of these classes.

The relevant outcomes of the studies were extracted. Each outcome was then classified by a research team member. These classifications were based on the WHO-defined components of ‘quality of care’ [30]. This states that high-quality care should be Safe, Effective, Timely, Efficient, Equitable, and People-centred. Studies were sub-sorted by results, according to whether they reported that there was a significant increase in the metric during their study, a significant decrease, no significant change, or if statistical significance was not assessed or not reported. Significance was defined as either p < 0.05 or using the 95% confidence interval.

Barriers and promoters of quality improvement were classified as local level, meaning individual to the particular hospital or location where the intervention was based, and systems level, meaning a factor that would necessarily influence hospitals and locations beyond the local area. We utilised a wide scope for extraction of barriers and promoters, including both barriers and promoters that were specific objects of study and those that were informally reported such as through staff feedback reported in the discussion.

In order to assess bias in the included studies, we utilised the Cochrane Risk of Bias for Non-Randomised Studies of Interventions (ROBINS-I) and revised tool for Risk of Bias in randomised trials (RoB 2.0) to assess included quantitative studies [31, 32]. Any study with a Critical ROBINS-I Overall Bias or a High RoB 2.0 Overall Bias classification was omitted from the “Results” and “Discussion” sections of this review. Two independent reviewers conducted the quality appraisal, and adjudication was provided by a third reviewer if warranted.

In total, the searches returned 8110 results across the four databases, with 49 results identified from other sources. After 2228 duplicates were excluded, a total of 5931 results’ titles and abstracts were screened against inclusion criteria. Of these, 5677 results were excluded for not meeting inclusion criteria at this stage, as demonstrated by Fig. 1.

This left 254 papers, which underwent full-text review. As shown in Fig. 1, the most common reason for exclusion at this stage was that studies took place in countries that were not LMICs (n = 79) according to the World Bank classifications [26]. The second most common reason was that studies were not describing QI interventions (n = 68). Twenty-eight studies were ultimately identified as suitable for inclusion in the narrative synthesis. Owing to the heterogeneity of QI methodologies, settings, and outcomes measured, meta-analysis was not undertaken.

The 28 studies consisted of 25 controlled before and after studies, two intervention studies (non-random), and one randomised controlled trial. Table 1 provides a summary of all the included studies.

The 28 included studies included data from 23 different countries. This included data from India (4 studies), China (3 studies), Bangladesh (3 studies), Ghana (3 studies), Brazil (3 studies), Iran (2 studies), El Salvador (2 studies), Mexico (2 studies), the Philippines (2 studies), Tunisia (2 studies), Turkey (2 studies), Uganda, Nepal, Kenya, Guatemala, Mozambique, Lebanon, Sierra Leone, Pakistan, Argentina, Colombia, Peru, and Morocco. The review includes two multi-country analyses, with one covering Argentina, Colombia, El Salvador, India, Mexico, Morocco, Peru, the Philippines, Tunisia, and Turkey and the other covering El Salvador, Mexico, the Philippines, and Tunisia [33, 34].

There were a total of 65,642 reported participants included in analyses across the studies, although one large study included 29,508 deliveries in its analysis and another included 15,249 [35, 36].

Table 2 presents the ROBINS-I and RoB 2.0 ratings awarded to each of the studies. One study scored Critical in Overall Risk of Bias, and thus was excluded from data synthesis, but is included in Tables 1 and 2 [37].

The core narrative themes extracted from the papers are presented under the headings, classified QI approaches, groups of outcomes measured in QI approaches, and barriers and promoters to implementing QI approaches. Henceforth, summary data comes from the 28 studies with Overall Risk of Bias of Serious or lower.

We categorised the overarching approaches for quality improvement used for sick newborn care using the Kruk and Gage ‘Synthesizing improvement approaches’ schema [10]. We found 11 studies with micro interventions [33, 38‐47], 23 studies with meso interventions [34‐36, 38‐42, 44‐46, 48‐59], and two studies with macro interventions [59, 60]. Nine of the studies had mixed-level interventions, with eight of these being meso and micro and one being macro and meso.

Table 3 provides information on approaches for quality improvement by subtype. In addition to the included subtypes in Table 3, there were additional categories according to the Kruk and Gage schema, for which no studies utilised those particular methods; these were, at the macro level, pay for performance, other financing and incentives, pre-service training, and external to health system and, at the meso level, mortality audits and social franchising [10].

The most frequent subtype of intervention was the meso approach ‘In-service training’, used by 20 studies. Such interventions were often delivered as part of a group of innovations—for example, Rosenthal et al. introduced a multifaceted infection control bundle incorporating education on hand hygiene and asepsis, and Agarwal et al. developed a package of interventions that included on-job training of nurses in common neonatal skills [34, 36]. The intervention of Clark et al. was based on a WHO Emergency Triage and Treatment training course, and Sethi et al. utilised a Comprehensive Post-Natal Counselling package, comprised of education of health care providers and family members [50, 54]. In Zhou et al., key staff members attended a training workshop run by the Canadian Neonatal Network for 2 days, and attendees in turn then trained the other Neonatal Intensive Care Unit (NICU) team members [46].

The second most frequent subtype of intervention was the meso approach, ‘Service organisation’, used by nine studies. For example, Rahman et al. described an approach that involved service reorganisation with triage and fast assessment and use of a record system [44]. Erdeve et al. evaluated the impact of individual rooms on patients and families in the NICU [52].

The most frequent micro approach was ‘Distribution of Referencing Materials to Providers’ by eight studies. Pinto et al. disseminated a new protocol proposed by the Brazilian National Health Surveillance Agency for antibiotic usage in low birth weight infants [43]. Salehi et al. also described the implementation of new guidelines, as did Mais et al. and Gathara et al. as part of their approaches [39, 41, 45].

Two studies utilised macro approaches, which were regulation and governance, and task shifting. Ramaswamy et al. used regulation and governance in their development of the Ridge-Kybele Model for Obstetric and Neonatal Care, an integrated approach for systems change which prioritises capacity building in order to properly embed change practices [59]. Yawson et al. utilised task shifting, by using a tool to identify service gaps which led to national and regional operating plans being developed and implemented to improve neonatal care [60].

A total of 23 broad outcomes were used to assess the efficacy of the QI interventions, shown in Table 4. The majority of these, 13, were outcomes that were classified as aiming at delivering safe care, defined as “delivering health care which minimises risks and harm to service users, including avoiding preventable injuries and reducing medical errors” [30]. Five were aimed at delivering effective care, two efficient care, two people-centred care, and one timely care. No studies were identified as including outcomes addressing the delivery of equitable care.

Mortality rate was the most frequently measured outcome, assessed by 16 studies. Studies used a variety of mortality metrics, including 28-day mortality, mortality within the study period, and specific cause mortality. Eight studies found that introduction of the QI intervention was associated with a significant decrease in mortality, four found no significant change, and five others reported results but not the statistical significance of those results. Amongst the more successful interventions, Bhutta et al. found that survival in their NICU increased from 65 to 84%, p < 0.05, after a policy change to create a stepdown unit and involve mothers earlier in the care of their at risk infants [49]. Pinto et al. also found a significant reduction in mortality, from 20.9 to 4.4%, p = 0.009, after the dissemination of a new antibiotic protocol to their NICU, which was supervised by two neonatologists in charge of the clinical routine [43]. Leng et al. also found mortality rates decreased, from 12 to 7%, p = 0.03, amongst newborns transferred from eight hospitals to their Level III NICU, after the introduction of a package of interventions including the STABLE programme and staff education [40]. Crouse et al. found that overall mortality for the critically ill patients decreased with their new emergency triage process, from 12.4% pre-intervention to 6.0% post-intervention, but this was not statistically significant (p = 0.15) [38].

Length of admission was the second most assessed outcome, by ten studies. Studies reported mixed outcomes, with one reporting a significant increase, four a significant decrease, two no change and three where significance was not assessed. Soni et al. found that length of stay in the NICU increased with identified Kangaroo Mother Care champions, at a median of 9 days, compared to a median of seven without the champions [61]. Conversely, Bastani et al. found that, in their RCT, mean length of stay in the NICU for their family-centred care group was 6.96 days compared to 12.96 in the control group, p < 0.001 [48]. Bhutta et al. found that length of stay more than halved after their stepdown unit was introduced, from a mean of 34 days pre-intervention to 16 post-intervention [49].

Sepsis rates were the joint third most frequently measured outcome, assessed by eight studies. Three studies reported a significant decrease in sepsis rates, and five reported no significant change. Amongst the studies that reported significant decreases were Agarwal et al., who found that there was a severe reduction in deaths in their neonatal unit due to sepsis after the introduction of their multi-faceted intervention package, from 37.9% pre-intervention to 15.5% post, p < 0.01 [36]. Their package included greater involvement of mothers in caregiving, as with Bhutta et al., alongside enforced aseptic routines, greater use of protocols, education, and other features [36, 49]. However, in Gilbert et al., the introduction of an educational package, POINTS of Care, did not change sepsis rates across five neonatal units nor did the multi-faceted intervention incorporating education, feedback, and other elements in Leng et al. [40, 53].

Patient infection rates, across a variety of different specific infections, were assessed by nine studies. Seven studies reported a significant decrease in infection rates, one a significant increase and one reported no significant change. Amongst the studies that reported significant decreases was Rosenthal et al., who found that a multi-faceted ventilator-associated pneumonia (VAP) bundle intervention was associated with a reduction in VAP rates from 17.8/1000 ventilation days pre-intervention to 12.0/1000 ventilation days post-intervention, across 15 NICUs in ten countries [33]. Rosenthal et al. found that after an infection control bundle, central line-associated bloodstream infection (CLABSI) rates reduced across four NICUs in four countries, resulting in a relative risk of 0.45 post-intervention (95% CI 0.33–0.63) [34]. Both interventions utilised infection control teams and surveillance. Mais et al. also reported on CLABSI rates, which declined from 15/1000 central line days before the introduction of a bundle of interventions to 6.4/1000 afterwards (p < 0.05) [41]. For Zhou et al., CLABSI rates also fell significantly from 16.7/1000 before the introduction of a nursing training programme to 5.2/1000 afterwards [46].

Many factors have the potential to either promote or inhibit the successful implementation of interventions for quality improvement [62, 63], and several of these were covered by the studies. These factors are noted in Table 5 and are identified as either barriers or promoters operating at a local or systems level. In total, 11 barriers were identified, with six at the local level and five at the systems level, and 13 promoters were identified, with nine at the local level and four at the systems level. No study assessed the statistical significance of any barrier or promoter; identified barriers and promoters were largely inductive by the investigators.

Nine promoters were identified at the local level by seven studies. Three studies highlighted the importance of motivation of key individuals. Darmstadt et al. highlighted the will and effort of key individuals as being important, especially the nursing supervisor; Srofenyoh et al. and Namazzi et al. noted the importance of local champions [35, 42, 51].

Four promoters were identified at the systems level by four studies. Yawson et al. stated that good quality national data collection is essential for designing QI interventions [60]. Namazzi et al. highlighted structured community relationships, Cavicchiolo et al. said NGO collaboration could be helpful, and Srofenyoh et al. valued formal support from the Ghanian Health Service [35, 42, 58].

Six barriers to quality improvement initiatives were identified at a local level by eight studies. The barrier posed by overburdened staff was the joint most frequently mentioned barrier, by four studies. This includes Gilbert et al. who mentioned staff being overstretched as an impediment, and Namazzi et al. who discussed the fact that facility staff had competing demands on their time [42, 53]. Four studies also mentioned the problem of insufficient equipment, such as Crouse et al. who stated that paper supplies running low and no computerised patient records hampered record keeping [38].

Five barriers were identified at a systems level by four studies. Namazzi et al. highlighted both the problems of increasing demand for services at all hospitals and lack of finances for necessary medicines [42]. Gilbert et al. said government policies enforcing redistribution of staff from study NICUs to underserved areas also created barriers [53].

Programme planners should consider that the majority of interventions are at the meso level, and many studies (20 studies) involved in-service training as part of their interventions. Educational interventions may be frequently represented because exposure of the intervention to the relevant staff can be appropriately monitored, implemented, and tailored to local needs, whereas macro-level interventions like regulation and governance or task shifting require greater continuous coordination. For planners designing their first QI projects, in-service training may be an advisable first step. However, structural adjustments remain necessary in order to facilitate QI at all levels of care. It will be especially relevant to formulate and implement policies to retain skilled nursing staff alongside providing financing to achieve national standards of minimum infrastructure and equipment, and to increase staffing for optimal nurse-patient ratios. Increasing adoption of perinatal death reviews in LMICs will also provide an opportunity to review and address common gaps in a country through macro-level interventions.

The most frequently assessed outcomes were mortality rates, sepsis and infection rates. Eight of the 12 studies that investigated statistically significant differences in mortality rates observed a statistically significant decrease in mortality. Particularly high reductions occurred in Bhutta et al. and Pinto et al., with mortality falling by 19% in the former after the creation of a stepdown unit and mortality falling by 16.5% in the latter after the dissemination of a new antibiotic protocol [43, 49]. These studies demonstrate the potential for QI to produce swift and significant benefits for this vulnerable patient group. However, it is important to note that the majority of the included studies were non-randomised before-after studies, and a review by Schouten et al. found that observational studies tend to demonstrate larger effects than more rigorous designs [65]. There are many feeders into mortality rates and a number of the studies had important confounders. With regards to other outcomes, in general, studies did not report greater involvement of family members as part of their interventions, but those that did reported significant positive results for mortality, sepsis, and readmission rates, suggesting this area could be explored further [36, 48, 49]. It will be important to focus future QI efforts on sepsis due to rising rates of antibiotic resistant infections and sepsis in this group [66]. Some successful studies in this area did incorporate holistic interventions involving family members [36, 49].

Several studies benefitted from focusing their efforts on single pathologies (ventilator-associated pneumonia, central line-associated blood stream infections) for quality improvement, which allowed for collation of findings across multiple centres [33, 34]. Other interventions utilised training courses included the POINTS of Care training (a six module training programme covering topics including pain control and nutrition interventions) [53, 67] and the STABLE programme (a continuing education course that focuses on stabilising sick newborns) [40]. Such approaches allow easy monitoring of attendance and hence exposure for future studies that may choose to utilise control groups.

For programme planners, our identification of barriers and promoters to successful QI interventions will be helpful to accelerate efforts for meeting the objectives of Every Newborn Action Plan, Every Woman Every Child, and other quality of care agendas [68, 69]. Planners should consider resolving barriers such as overburdened staff, which may be more severe during times of higher seasonal demand, and consider identifying ‘staff champions’ to promote their projects which were reported as promoters in several studies [35, 42, 51]. A previous review of barriers and enablers of KMC identified similar influential factors, such as the barrier posed by a high workload [70].

Finally, it is important to reflect that quality improvement approaches are heavily dependent on data linked to action and evidence to drive positive change. Moxon et al. have advocated for the urgent need to improve health management information systems and monitoring of hospital care for newborns for better measurement of quality of care and to identify and address quality gaps [62]. Thus, the allocation of resources on improving quality of sick newborn care needs to factor in requirements for improving local, country-led sustainable information systems, as well as systematic use of perinatal death audits [71]. Mortality audits were not utilised by any of the included studies, despite a recent focus via Maternal and Perinatal Deaths Surveillance and Response projects [72]. The bottlenecks of health financing, health workforce, data and community engagement need more varied approaches to implementation and research, especially at Macro level, and advanced information systems would be valuable in order to optimise QI if used for action at the correct level of the health system.

Regarding future research into QI initiatives in LMICs, we recommend QI implementation takes place in tandem with strong data collection and monitoring. The majority of research takes place in high-income countries, whose settings may not be fully applicable to LMICs [12, 25]. We also note the impact of several interventions was hard to assess because studies did not analyse statistical significance. These interventions would benefit from further, larger-scale studies, or more rigorous evaluation. The sharing of outcome measures across studies would also allow for improved future quantitative synthesis.

We found that research often focused on single centres, also noted by Dettrick et al. [23]. More work should be done to evaluate quality improvement at multiple centres in different settings, such as comparisons between rural and urban hospitals, and to investigate the impact of scaling up existing projects [12, 73]. This may need specific capacity building of public health programmes and clinical staff on research in LMICs and provision of grants to undertake context-specific projects on a range of interventions. Barriers and promoters were not assessed for statistical significance by any study, and multi-centre studies could enable such analysis and the provision of richer qualitative data. Generally, we would recommend that QI projects are rigorously evaluated and the experience documented or reported in peer-reviewed literature wherever feasible.

The heterogeneity of our study population, the interventions for quality improvement and multiple outcome measures were key challenges. Quality improvement approaches is an umbrella term for a variety of interventions, and though we cast a wider net, some studies may have been missed from our selection criteria, especially if they did not use the exact term. Many healthcare interventions in LMICs are community- rather than hospital-based, and care during pregnancy was not considered, so many interventions that may be useful in mitigating overall newborn morbidity and mortality were not eligible [74]. In general, many quality improvement projects do not progress to published literature, particularly small-scale projects, so publication bias is likely to be present. Unpublished ongoing initiatives ongoing in collaboration with WHO, UNICEF, USAID and others could not be included.

This study did not capture solely qualitative literature, and qualitative synthesis would be useful going forward, particularly for identifying barriers and promoters. There was generally little information on the methodology for identifying barriers and promoters in the included studies, which may mean that findings are not fully representative. The study method did not allow for collection of information relating to parental experience, which is a critical dimension of quality of care improvement. This study is also limited by only including studies published from 2000 onwards and published in English. We did not conduct overall outcome-specific assessments of quality of evidence with a tool such as GRADE. Finally, although we aimed to analyse with reference to wealth, rural/urban, and type of facility as measures of healthcare equity, data on financing of health settings were available for only a minority of the included papers, and many hospital settings were anonymised. As a result, these analyses were not undertaken.