Introduction

Neonatal hypothermia, defined as an axillary temperature of less than 36.5°C, is a major contributor to preventable morbidity and mortality. The global burden of hypothermia is concentrated in low- and middle-income countries (LMICs), with prevalence estimates as high as 85% among hospitalized newborns [1]. Although causality is difficult to ascertain, it is a major comorbidity associated with leading causes of death, namely sepsis, prematurity, and asphyxia [2, 3]. Detecting, treating and preventing neonatal hypothermia aligns with the Sustainable Development Goals (SDG) to reduce neonatal (SDG Target 3.2) mortality [4].

A central challenge to addressing neonatal hypothermia is that it is most prevalent in settings with poor record-keeping systems; temperature information is not routinely documented, and the use of electronic medical records are far from universal [5, 6]. Studies that have evaluated neonatal hypothermia in LMICs have frequently relied on independent data collection instruments. These protocols often require more frequent assessments than routine vital sign measurements. Furthermore, evaluations are typically limited to admission temperatures, which is neither representative of nor pragmatic in most birth facilities globally [7–10]. An added complexity is that normothermia is a common performance metric, raising concerns about incentivizing the underreporting of abnormal temperature measurements.

The primary function of clinical decision support software (CDS) is to provide clinical recommendations based on patient-specific parameters and evidence-based guidelines [11]. Some CDS, particularly those that operate as standalone platforms, have begun to integrate data collection functionalities. CDS used voluntarily by clinicians has the potential to enhance our understanding of neonatal hypothermia by capturing temperature profiles in settings with limited data infrastructures. Conversely, where medical records do exist, CDS data can be leveraged to interrogate documentation accuracy and quality.

To date, the characteristics of data derived from a CDS in LMIC, particularly as it concerns neonatal hypothermia, have not been well-defined. In this prospective observational study, we analyze the 1,027 CDS entries submitted by a cohort of midwives and nurses at four rural hospitals in eastern Uganda using the NoviGuide 2.2 application, a novel CDS for newborn care [12]. We aimed to describe the profile of temperatures entered into the NoviGuide and evaluate clinical features associated with neonatal hypothermia.

Materials and methods

Results

Study site and newborn characteristics

Fifty-one midwives and nurses completed a total of 1,027 initial newborn assessments between September 2020 and November 2021 (Fig 1) (S2 Table). Most infants entered into the NoviGuide Initial Assessment pathway were born at term (88.6%, n = 910) and had a normal birth weight (82.6%, n = 848). The majority of assessments were performed at Tororo General Hospital (46.3%, n = 475). A total of 244 newborn assessments were started, but not completed, and were therefore excluded from the analysis. Differences in clinical characteristics between complete and incomplete entries are presented in S3 Table. Tororo General Hospital was the first site to adopt the NoviGuide platform (September 2020), followed by Mulanda (March 2021), Nagongera (April 2021), and Mukujju (May 2021). Once introduced, the NoviGuide was available throughout the study period at each study site without interruptions to usage.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study flow diagram for NoviGuide usage data.

https://doi.org/10.1371/journal.pgph.0000982.g001

Hypothermia

The median temperature was 36.8°C (Interquartile range 36.3–37°C) with a mode of 37.0°C (S1 Fig). Overall, neonatal hypothermia was recorded in 30.5% (n = 313) of CDS entries (Table 1). Among assessments with neonatal hypothermia, 79.5% and 20.5% had mild (36–36.5°C) and moderate (32–35.9°C) hypothermia, respectively. No CDS entries had severe hypothermia (<32°C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Clinical characteristics of study population, by temperature.

https://doi.org/10.1371/journal.pgph.0000982.t001

A higher proportion of preterm patients (<37 weeks gestational age) (24.6% vs. 5.1%) and a lower proportion of term (> = 37 weeks gestational age) patients (74.4% vs 94.9%) were observed among hypothermic patients relative to normothermic patients, respectively (p<0.001) (S2 Table). Similarly, more low birth weight patients (<2.5 kg) (32.3% vs. 11.2%) and fewer normal birth weight (> = 2.5 kg) patients (67.7% vs 88.8%) were observed with neonatal hypothermia than those with normothermia (p<0.001). Patient assessments with respiratory distress (41.9% vs. 16.2%, p<0.001), sepsis/concern for sepsis (77% vs 40%, p<0.001) and hypoglycemia/concern for hypoglycemia (36.4% vs 13.2%, p<0.001) were also noted in greater proportions among patients with neonatal hypothermia than normothermia. There were significant differences in neonatal hypothermia burden across study sites (p<0.001).

On multivariable logistic regression analysis, pre-term gestational age (Adjusted odds ratio [aOR]: 2.62, 95% CI: 1.38–4.98), sepsis/concern for sepsis (aOR: 2.73, 95% CI: 2.90–3.94), and hypoglycemia/concern for hypoglycemia (aOR: 1.78, 95% CI: 1.17–2.72) were among the clinical features associated with hypothermia (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Multivariable logistic regression of clinical characteristics associated with neonatal hypothermia^a.

https://doi.org/10.1371/journal.pgph.0000982.t002

Discussion

In this prospective, observational analysis of a novel CDS platform in eastern Uganda between September 2020 and November 2021, we found that nurses and midwives frequently entered hypothermic temperatures (30.5%) in a digital pathway for initial newborn assessment. Estimates of neonatal hypothermia extracted from CDS usage data differed across study sites. Additionally, clinical features associated with neonatal hypothermia reflect known risk factors, including prematurity, and symptomatic or laboratory proven hypoglycemia and acute infection. The Noviguide, and CDS more generally, may thus function as a valid and robust tool for data capture and analysis.

There is significant heterogeneity among previous reports of neonatal hypothermia; rates have been found to vary widely in African countries and across diverse geographies, climates, and facilities [9, 15–18]. The lack of comprehensive temperature data globally obscures the epidemiologic picture of neonatal hypothermia and limits the advancement of evidence-based thermoregulatory care in LMIC. Our findings suggest that the NoviGuide has the potential to bridge key data gaps where traditional data collection instruments may not exist [19]. This study builds upon a previous report, which found that the NoviGuide is adaptable, well-received, and significantly improves provider confidence in newborn care [12].

The downstream applications of CDS-derived data extend to both individual facilities and across health systems. Internal review of CDS data could inform indicators for quality improvement initiatives. NoviGuide implementation more broadly could enhance cross-site comparisons, enabling high-risk areas to be more easily targeted. The range in temperature distributions in our study locations, for example, illustrates the importance of high data resolution. More than half of infants were found to be hypothermic at Tororo, compared to less than 5% at Mulanda and Mukujju. The mechanisms underlying site-to-site variation likely encompass both clinic-level factors–such as acuity level, catchment area, and case volumes–and individual-level factors–such as provider usage patterns and experience. As a larger district hospital, Tororo may see a comparatively greater number of home births and medically complex cases, leading to a higher prevalence of hypothermia. Further research is needed to understand how to leverage these insights to support data-driven policy and resource allocation efforts.

Importantly, data captured through CDS usage should be interpreted within unique technical and operational considerations [20]. There are functional differences between CDS software and medical record data. Whereas medical records document clinical care, CDS software serves primarily as an aid in point-of-care decision-making and when used voluntarily, cannot necessarily provide meaningful prevalence or outcome estimates. In our study, CDS usage was dictated by users’ perceived benefit of the technology rather than an administrative mandate. This is reflected in the patient and temperature profiles. Sepsis/concern for sepsis and respiratory distress were more common than what would be expected for a general newborn population [21, 22]. Accordingly, there may be a preferential bias of the NoviGuide towards acutely ill newborns, leading to an overestimation of hypothermia in our cohort.

We also observed a significant number of 37°C temperatures, the default entry in the NoviGuide. Providers may have lacked functioning thermometers or elected not to enter a real temperature. Several quality control steps have been incorporated into the NoviGuide to account for these and other potential biases and discrepancies. Internal logic checks ensure data entries are physiologically consistent and feasible [23]. Users are alerted at the point of care to address incongruencies, such as unlikely weight-for-age combinations. Duplicate entries, numerical outliers, and missing data can be further handled with retrospective aggregate data analysis.

The NoviGuide is specifically designed to improve adherence to clinical guidelines in the settings least able to produce high quality medical records. As a result, research based on CDS usage data is somewhat constrained beyond descriptive studies. We were unable to ascertain the relationship between neonatal hypothermia and mortality, for example, without linkage to individual patient outcomes. This challenge is not easily solved by interoperability alone [24]. Embedding CDS platforms into an electronic medical record, while technically feasible, will not provide the sought-after clarity if documentation is incomplete or of poor quality. Indeed, temperature data is often missing from the medical record in LMICs [25]. Alternatively, introducing novel data collection instruments to measure clinical impact alongside the NoviGuide may distort CDS data if the technology creates a separate clerical burden [26]. As information systems become more accessible globally, integration between CDS, electronic medical records, and other digital platforms should be prioritized to maximize data collection and analysis at a local and public health scale.

Conclusion

Ugandan nurses and midwives using the NoviGuide frequently documented neonatal hypothermia and usage data closely mirrored known clinical risk factors. These results should be contextualized within intrinsic limitations of data captured through CDS. Neonatal hypothermia is a global problem, but is most severe in settings that also lack high quality medical records and specialized neonatal nurses. By prompting clinicians to take temperatures, alerting clinicians to the presence of hypothermia, guiding corrective action and then gathering data about temperature recordings, CDS may be a useful adjunct to efforts to improve thermoregulation and, by extension, newborn survival.

Supporting information

Acknowledgments

We would like to thank the medical superintendents namely; Dr. Ochar Thomas (Tororo General Hospital), Dr. Jagile Onyango (Mukujju Health Center), Dr. Onen Jacob (Nagongera Health Center) and Dr. Were Micheal (Mulanda Health Center) for their generous support. We also acknowledge the midwives and the nurses for their participation in the parent study and specifically Sr. Daisy Wandera, Sr. Atim Grace and Sr. Musungu Beatrice, the NoviGuide Trainers for leading the initial onsite NoviGuide training sessions and for providing mentorship and site support at the participating hospitals. Lastly, Amee D. Azad, MD MS provided significant support with the statistical analysis.