Quality improvement bundles to decrease hypothermia in very low/extremely low birth weight infants at birth: a systematic review and meta-analysis

Data sources and search strategy

A comprehensive literature search was conducted in electronic databases including PubMed, Embase, the Cochrane Library and Web of Science. The final search was completed on April 22nd, 2024, and no search limits or restrictions were applied initially to broaden the search scope. Appendix 1 describes the detailed search strategy across individual databases.

Inclusion and exclusion criteria

The criteria for including studies in this systematic review and meta-analysis are built on the PICOS framework: (1) Population (P), VLBW/ELBW infants, or preterm infants with a gestational age (GA) of 32 weeks or less. (2) Intervention (I), application of QI bundles aimed at preventing hypothermia at birth. (3) Comparison (C), VLBW/ELBW infants before the implementation of QI bundles as the comparison. (4) Outcome (O), the incidence of hypothermia from birth to transfer to the NICU. (5) Study design (S), the eligible study design is that of a Quality Improvement Project (QIP) that applies QI bundles. Exclusion criteria were as follows: (1) Participants being newborns with a gestational age over 32 weeks and a birth weight exceeding 1,500 g. (2) Studies not related to quality improvement, such as randomized or quasi-randomized trials of single or multiple interventions. (3) Temperature measurement is not between birth and admission to the NICU or the site of temperature measurement is not clear. (4) Studies only available as conference abstracts.

Study selection and data extraction

Two independent reviewers (GZ and JQ) screened titles and abstracts of identified studies for relevance. Full-text articles were retrieved for potentially relevant studies and assessed for eligibility based on predefined inclusion and exclusion criteria. Discrepancies between reviewers were resolved through discussion, with involvement of a third reviewer (BW) if consensus could not be reached. Two authors (GZ and JQ) independently extracted data using a standardized data collection form. The extracted information included study author, year of publication, location, study duration, sample size, participant demographics, site of temperature measurement, outcomes (incidence of mild hypothermia, moderate hypothermia, and hyperthermia) and components of the QI bundles. Disagreements were resolved through discussion or consultation with a third author (ZY) if necessary.

Quality evaluation and risk bias assessment

Two investigators (GZ and JQ) independently assessed the quality of the selected studies using the Quality Improvement Minimum Quality Criteria Set (QI-MQCS) (Hempel et al., 2015). This set encompasses 16 content categories: Organizational Motivation, Intervention Rationale, Intervention Description, Organizational Characteristics, Implementation, Study Design, Comparator, Data Source, Timing, Adherence/Fidelity, Health Outcomes, Organizational Readiness, Penetration/Reach, Sustainability, Spread, and Limitations. Each study was evaluated across these domains, earning 1 point for meeting the minimum criteria in each domain, and 0 points otherwise. The studies were subsequently classified based on their quality: scores above 10 signified high quality, scores between 7 and 10 indicated medium quality, and scores below 7 denoted low quality. Any discrepancies were resolved through group consensus.

Statistical analysis

When conducting the synthesis of effect sizes, data from VLBW infants and ELBW infants was categorized separately according to mild hypothermia, moderate hypothermia, and hyperthermia. Only studies included in the systematic review that provide binary variable data (or derived binary variables) for hypothermia or hyperthermia were included in the meta-analysis. If a study provides data for multiple groups, it will be separately included in the respective groups for data synthesis. We summarized the outcomes by reporting the odds ratio (OR) and 95% confidence interval (CI), and presented them using forest plots. Meta-analyses were performed with Review Manager 5.4.1 software. We assessed the heterogeneity of the studies using I² statistics. A random-effects model was employed to address substantial heterogeneity between studies (I² ≥ 50%), defaulting to a fixed-effects model when heterogeneity was absent (I² < 50%). Sources of heterogeneity were investigated, and sensitivity analyses were conducted to determine the impact of individual study results on the overall effect size. P < 0.05 was considered significant. When the number of studies included in the meta-analysis exceeds ten, a funnel plot generated by Review Manager 5.4.1 software will be used to assess publication bias.

Study selection

A total of 737 articles were retrieved from the electronic databases including PubMed (139), Embase (334), the Cochrane Library (65), and Web of Science (199). After removing duplicates, 457 articles remained and were screened for eligibility based on title and abstract. Out of the screened articles, 37 were selected for full-text screening to assess their suitability for inclusion. Among these, 19 articles were excluded due to reasons such as having the same or overlapping study cohort (three) (Frazer et al., 2018; Harriman et al., 2018; Wang et al., 2022), including newborns with a birth weight >1,500 g and GA > 32 weeks (nine) (Andrews et al., 2018; Choi et al., 2018; Harer et al., 2017; Howes & Keir, 2018; Patodia et al., 2021; Pratik et al., 2023; Ramjist et al., 2024; Shaw et al., 2018; Sprecher et al., 2021), unclear temperature measure time/site (two) (Glenn et al., 2021; Keir & Cavallaro, 2021), and only conference abstracts are available (five) (Bobby et al., 2014; Dale et al., 2021; Shaw et al., 2016; Thornton, Voos & McNellis, 2014; Woolley et al., 2019). Following the full-text screening, 18 articles met the criteria for qualitative synthesis. Among these, 12 articles were deemed suitable for quantitative synthesis through meta-analysis. Specifically, for the meta-analysis, for VLBW infants, four, eight, and six studies were included for mild hypothermia, moderate hypothermia, and hyperthermia, respectively; for ELBW infants, two, three, and three studies were included for mild hypothermia, moderate hypothermia, and hyperthermia, respectively. The detailed flowchart illustrating the screening and selection process is provided in Fig. 1, following the PRISMA guidelines.

Study characteristics

The included 18 studies originated from diverse locations, with 11 from the USA (Bhatt et al., 2020; Billimoria et al., 2013; Castrodale & Rinehart, 2014; Croop et al., 2020; DeMauro et al., 2013; Dixon et al., 2021; Frazer et al., 2021; Godfrey et al., 2013; Lee, Ho & Rhine, 2008; Manani et al., 2013; Pinheiro et al., 2014), two from Australia (Kent & Williams, 2008; Singh et al., 2022), and one each from India (Sivanaridan, Sankar & Deorari, 2016), Singapore (Yip et al., 2017), Brazil (Caldas et al., 2018), the UK (Young et al., 2021) and China (Bi et al., 2022). The publication timeline of the selected studies spans from 2008 to 2022. Predominantly, the quality improvement initiatives (17/18) were conducted at a single center, and only one (Bi et al., 2022) was multicenter (5 NICUs). The sample sizes varied considerably, ranging from 59 to 1,199 (total = 6,552 infants). Table 1 displays the characteristics of the included studies.

Quality and risk of bias assessment

The assessment of the included studies was conducted utilizing the Quality Improvement Minimum Quality Criteria Set (QI-MQCS). The results showed study scores ranging from 9 to 15, with five categorized as medium quality and 13 as high quality, detailed in Appendix 2. Each study met the minimum quality requirements in seven of the 16 evaluated domains: organizational motivation, rationale behind the intervention, detailed description of the intervention, implementation strategies, description of the comparator, timing of the intervention, and the health outcomes observed. However, a predominant issue in most studies was the lack of detailed information on adherence to intervention protocols (adherence/fidelity: 14 out of 18 studies). Similarly, descriptions lacked details on the sustainability of interventions (sustainability: 12 out of 18 studies), potential for scalability or replication (spread: 12 out of 18 studies), and adequately outlining barriers and facilitators of organizational readiness for the interventions (organizational readiness: 11 out of 18 studies).

Bundle components

A total of 10 components were identified across the 18 studies, with the number of components per bundle varying. The number of components ranged from four to nine. The most common components included formation of a multidisciplinary expert team, development of evidence-based interventions, education of hospital staff (18/18); using plastic wrap or bag (18/18); and use of radiant warmer or incubator (18/18). These were followed by keeping the head in plastic wrap, using polyurethane-lined hats, or stocking-knit caps (16/18); increasing ambient temperature (14/18); regular reinforcements or ongoing process review and feedback loop (12/18); using warming mattresses (10/18); monitoring temperatures (6/18); maintaining plastic bag/wrap integrity during the resuscitation process (2/18); and delaying anthropometric data collection and non-essential procedures (2/18). Table 2 displays the 10 components included in the QI bundles.

Decrease in mild hypothermia (body temperature < 36.5 °C)

Among the 12 studies included in the meta-analysis, four provided dichotomous data on the incidence of mild hypothermia before and after QI in neonates with a birth weight of <1,500 g or GA < 32 weeks (Bi et al., 2022; Dixon et al., 2021; Singh et al., 2022; Yip et al., 2017). Westmead Hospital in Australia conducted a QI project over a 5-year period targeting VLBW infants and infants with a GA < 32 w (sample size: 568) (Singh et al., 2022). The project resulted in a significant reduction in the incidence of mild hypothermia, decreasing from 64.8% to 36.7%. KK Women’s and Children’s Hospital in Singapore also implemented a QI project for over 5 years with a sample size of 1,199 (including 168 before QI, 1,031 during and after QI) (Yip et al., 2017). The incidence of body temperature <36.5 °C decreased from 75.5% to 46.8%. A 36-month QI project conducted in Shandong, China, from January 2018 to December 2020, involved five NICUs and included a total sample size of 750 (Bi et al., 2022). The study observed a significant reduction in the incidence of mild hypothermia among VLBW infants, decreasing from 95.9% to 71.3%. Another QI project (sample size: 67) from Duke University School of Nursing (Dixon et al., 2021) reported that the incidence of mild hypothermia decreased from 33.3% (18/54) to 0% (0/13). The combined result showed that QI efforts substantially reduce the incidence of mild hypothermia in VLBW infants (OR 0.22, 95% CI [0.13–0.37]) (Fig. 2). Given the substantial heterogeneity between studies (I² = 71%), we identified the multicenter study (Bi et al., 2022) as a potential source of heterogeneity. Excluding that study led to homogeneity among the studies (I² = 0%), and the meta-analysis result remained stable (Appendix 3).

Two studies provided insights into mild hypothermia occurrences among ELBW infants or infants with a GA < 28 w (Godfrey et al., 2013; Singh et al., 2022). The first study involved 145 cases (Singh et al., 2022). It revealed that the incidence of mild hypothermia dropped from 59.1% to 38.9%. Conversely, the second study, conducted at the University of Pittsburgh School of Nursing and Magee-Women’s Hospital of the University of Pittsburgh Medical Center in the United States (Godfrey et al., 2013), included 81 cases and found no statistically significant change in the incidence of mild hypothermia. A subsequent meta-analysis combining the effect sizes from both studies determined an overall OR = 0.46, 95% CI [0.26–0.81], and I² = 0%, indicating a high degree of homogeneity among the findings (Fig. 3).

Decrease in moderate hypothermia (body temperature < 36.0 °C)

Among the 12 studies included in the meta-analysis, eight reported on the incidence of moderate hypothermia in VLBW infants or infants with a GA < 32 w (Bi et al., 2022; Caldas et al., 2018; Frazer et al., 2021; Lee, Ho & Rhine, 2008; Manani et al., 2013; Pinheiro et al., 2014; Singh et al., 2022; Sivanaridan, Sankar & Deorari, 2016). The combined sample size was 3,442, comprising 2,123 participants in the QI intervention group and 1,319 in the control group. Initially, the incidence of moderate hypothermia ranged from 25.5% to 76.1%. Following the introduction of the QI bundle, this rate decreased significantly to between 5.4% and 31.7%. A notable reduction in moderate hypothermia was observed across all studies, with the meta-analysis indicating a combined OR = 0.18, 95% CI [0.15–0.22], and a heterogeneity (I²) of 5% (Fig. 4).

Three studies highlighted the incidence of moderate hypothermia in ELBW infants or infants with a GA < 28 w (Billimoria et al., 2013; Lee, Ho & Rhine, 2008; Singh et al., 2022). These studies included a total of 461 subjects, split between 209 in the QI group and 252 in the control group. The incidence of moderate hypothermia before and after the QI intervention was reported as 76.6% to 56.5%, 35.2% to 14.2%, and 91.6% to 37.5%, respectively. The aggregated OR = 0.21, 95% CI [0.08–0.58], and I² = 78% (Fig. 5). After excluding an early study (Lee, Ho & Rhine, 2008), the heterogeneity disappeared (I² = 0%) (Appendix 4). Differences in standards of neonatal care across various temporal phases may contribute to heterogeneity.

Increase in hyperthermia (body temperature > 37.5 °C)

Six studies assessed the incidence of hyperthermia pre- and post-QI in VLBW infants and preterm infants with GA < 32 w (Bi et al., 2022; Caldas et al., 2018; Frazer et al., 2021; Manani et al., 2013; Singh et al., 2022; Yip et al., 2017). The total sample included 3,615 subjects, with 1,165 cases evaluated before and 2,450 after the interventions. Initially, hyperthermia incidence ranged from 0% to 2.4%, increasing to between 0.4% and 7.8% post-intervention. Among these six studies, two studies (Caldas et al., 2018; Frazer et al., 2021) highlighted a statistically significant rise in hyperthermia incidence post-intervention, yielding a combined OR = 2.79, 95% CI [1.53–5.09] (I² = 20%). Meta-analytic findings suggest that the implementation of a QI bundle significantly increased the incidence of hyperthermia in this demographic (Fig. 6).

Three studies investigated hyperthermia rates pre- and post-QI in ELBW infants and preterm infants with GA < 28 weeks (Billimoria et al., 2013; Godfrey et al., 2013; Singh et al., 2022), involving a total of 438 participants (250 pre-QI, 188 post-QI). According to Billimoria’s study (Billimoria et al., 2013), no hyperthermia cases were reported before or after QI. Conversely, the remaining two studies (Godfrey et al., 2013; Singh et al., 2022) reported hyperthermia incidences of 2.8% and 1.4%, and 2.1% and 5.7%, respectively, before and after QI. Meta-analysis indicated no significant difference in hyperthermia rates post-QI, with a combined OR = 1.10, 95% CI [0.22–5.43], I² = 4% (Fig. 7).

Publication bias

Due to the number of studies included for mild hypothermia, moderate hypothermia, and hyperthermia in both VLBW and ELBW groups not exceeding ten—with the maximum being only eight—a funnel plot was not generated to assess publication bias.