Clinical cascades as a novel way to assess physical readiness of facilities for the care of small and sick neonates in Kenya and Uganda

Background Globally, there were 2.7 million neonatal deaths in 2015. Significant mortality reduction could be achieved by improving care in low- and middle-income countries (LMIC), where the majority of deaths occur. Determining the physical readiness of facilities to identify and manage complications is an essential component of strategies to reduce neonatal mortality. Methods We developed clinical cascades for 6 common neonatal conditions then utilized these to assess 23 health facilities in Kenya and Uganda at 2 time-points in 2016 and 2017. We calculated changes in resource availability over time by facility using McNemar’s test. We estimated mean readiness and loss of readiness for the 6 conditions and 3 stages of care (identification, treatment, monitoring-modifying treatment). We estimated overall mean readiness and readiness loss across all conditions and stages. Finally, we compared readiness of facilities with a newborn special care unit (NSCU) to those without using the two-sample test of proportions. Results The cascade model estimated mean readiness of 26.3–26.6% across the 3 stages for all conditions. Mean readiness ranged from 11.6% (respiratory distress-apnea) to 47.8% (essential newborn care) across both time-points. The model estimated overall mean readiness loss of 30.4–31.9%. There was mild to moderate variability in the timing of readiness loss, with the majority occurring in the identification stage. Overall mean readiness was higher among facilities with a NSCU (36.8%) compared to those without (20.0%). Conclusion The cascade model provides a novel approach to quantitatively assess physical readiness for neonatal care. Among 23 facilities in Kenya and Uganda, we identified a consistent pattern of 30–32% readiness loss across cascades and stages. This aggregate measure could be used to monitor and compare readiness at the facility-, health system-, or national-level. Estimates of readiness and loss of readiness may help guide strategies to improve care, prioritize resources, and promote neonatal survival in LMICs.


Introduction
Globally, there were 2.7 million neonatal deaths in 2015 [1]. The leading causes were preterm birth, defined as birth before 37 completed weeks of gestation (16%), intrapartum-related events (11%), and sepsis or meningitis (7%) [1]. Deaths in the neonatal period are responsible for 45% of all deaths in children under age 5, with preterm birth being the leading cause [1]. Major mortality reduction could be achieved by improving care in low-and middle-income countries (LMIC) [2][3][4][5]. Within the neonatal period, 36% of deaths occur on the day of birth and 73% occur in early neonatal period, defined as the first 7 days [2]. Between 1980 and 2015, early neonatal mortality has decreased more slowly than all other age-categories of under-5 mortality [6]. Thus, the immediate postnatal and early neonatal periods represent critical windows of opportunity to improve neonatal survival globally.
Facility-based care of small and sick neonates, including neonatal resuscitation, kangaroo mother care (KMC), intravenous (IV) fluids, feeding support, oxygen, antibiotics, and phototherapy, could avert an estimated 580,000 neonatal deaths annually [7]. Estimates suggest that provision of available interventions in facilities could decrease prematurity, intrapartum, and infection-related causes of neonatal mortality by 58%, 79%, and 84%, respectively [7]. Conversely, some of these interventions also carry a risk of harm when administered by inadequately trained staff or without proper equipment. For example, provision of oxygen therapy in the absence of pulse oximetry monitoring increases the risk of retinopathy and subsequent visual impairment in preterm neonates [8,9]. Although an estimated 72% of deliveries globally occurred in facilities from 2012 to 2017 [10], lack of access to delivery and postnatal care remains a challenge in LMICs [11][12][13]. Further, essential interventions are not successfully implemented in many LMICs due to an array of underlying constraints, including shortages of skilled providers, inadequate funding, poor distribution of newborn care services, and weak referral systems [14]. To better understand such barriers and ultimately address the functionality of a health system as a whole, facility-level capacity limitations must first be identified.
However, analyzing facility capacity has been an ongoing focus of public health for decades and competing theories exist on how to best approach such an analysis. To develop and further standardize an approach, the United Nations Children's Fund (UNICEF) and the World Health Organization (WHO) published guidelines describing "signal functions" related to facility readiness for provision of Essential Obstetric Care (EOC) [15]. These signal functions represent a selection of key interventions used to treat obstetric complications, which classify and monitor the level of care (basic or comprehensive) being provided by a facility, rather than listing all EOC services that should be provided [15]. In 2009, EOC was replaced by Emergency Obstetric and Newborn Care (EmONC), and the list of signal functions was expanded to include neonatal resuscitation (with a bag and mask) at facilities providing basic or comprehensive levels of care [16]. Availability, use, and quality of EmONC have been suggested as pragmatic indicators to monitor and evaluate the progress of health systems towards reducing maternal and neonatal mortality [16][17][18][19][20][21], yet there have been few attempts to develop additional signal functions for neonates. One study evaluated indicators for the quality of pediatric hospital care in LMICs and found broad support among experts for several newborn indictors, including availability of tetracycline, vitamin K, parenteral antibiotics, and drugs for the prevention of mother-to-child transmission (PMTCT) of HIV [22]. In 2012, new signal functions were proposed for routine and emergency newborn care, including KMC, IV fluids, cup feeding, oxygen, antibiotics, and PMTCT [18]; however, these have not been widely adopted globally. A recent study delineated over 600 structural characteristics, including infrastructure, equipment, drugs, providers and guidelines, for facility readiness to deliver care for small and sick newborns, and work is currently underway to finalise recommendations for newborn signal functions [23].
Notably, the signal function approach is subject to limitations and recommendations for alternative approaches have led researchers to the rethink the analysis of capacity. Potter and Brough emphasized that the key components of a functioning health system (i.e., facility/staff, skills, tools) are hierarchical in nature, and depend greatly upon the availability and functionality of each other [24,25]. Applied to a clinical context, a medical intervention can only be effectively administered if the necessary infrastructure, staff, and tools are first in place. Through precise identification of resource shortages, capacity needs can be tied to specific gaps found within each step of the hierarchy [24]. The HIV treatment cascade was developed as a tool to identify gaps in care delivery and to prioritize resources with the goal of improving public health [26][27][28]. At each step of the cascade (e.g., diagnosis, linkage to care), patients may be lost to follow-up and, as a result, fail to access or benefit from available health interventions. The cascade approach has subsequently been applied to other areas of public health, including PMTCT [29], hepatitis C [30], diabetes [31], hypertension [32] and, most recently, emergency obstetric care [33]. The latter introduced the clinical cascade model, which highlights the fact that multiple resources are required sequentially or simultaneously in order to provide realtime patient care [33]. For example, a provider can effectively treat a sick neonate requiring immediate care only when all resources needed to identify and treat the underlying condition are simultaneously present in the facility. In the obstetric cascade study, Cranmer and colleagues assessed 44 primary care facilities in Kakamega County, Kenya and found that 39-100% had the resources required for identification, 7-57% had resources for treatment, and 0-2% had resources to monitor or modify treatment across five common maternal emergencies [33].
Informed by Potter and Brough's capacity pyramid and based on the obstetric emergency cascade [24,33], we aimed i) to develop clinical cascades to evaluate facility readiness to care for small, sick neonates and ii) to utilize these to assess 23 health facilities in Kenya and Uganda.

Study setting
In Kenya and Uganda, annual neonatal mortality rates have slowly decreased over the last decade, but remain high at 21 and 20 deaths per 1,000 live births, respectively (2017) [34]. Estimated preterm birth rates are 12% for Kenya and 14% for Uganda [35]. We conducted facility assessments at 23 health facilities-17 in Migori County, Kenya and 6 in Busoga Region,

Statistical analysis
We described facility characteristics and neonatal care variables with standard descriptive statistics, including mean, standard deviation (SD), median, interquartile range (IQR), frequency, and proportion. Point estimates for resource availability across all facilities at each time-point were summarized as counts and proportions. Changes in resource availability over time by facility were calculated using McNemar's test. Since the dataset had fewer than 100 observations or data were not normally distributed, non-parametric statistics with two-sided tests of significance were used for all analyses. Loss of readiness was calculated by subtracting readiness at a given stage from readiness in the preceding stage. In the identification stage, readiness loss was calculated by subtracting readiness from 100%. Means were used to estimate overall readiness and loss of readiness because these measures were based on few observations, thus medians would not accurately capture the range of observations. Variability was summarized using the absolute range and SD since resource availability varied greatly. Readiness loss between stages was quantified with percentages. Since resource requirements differ based upon the expected level of care provision [8,18,37], we estimated readiness for each clinical cascade and stage of care by facility level as well as by country. In the sub-analysis of health clinics, we excluded items that are not expected to be available at this level of facility [37]. Further, we compared the readiness of facilities with a functional newborn special care unit (NSCU) to those without using the two-sample test of proportions. All statistical analyses were carried out using Stata 15.1 (StataCorp LP, College Station, Texas, United States of America).

Ethics
We obtained ethical approval from the Institutional Review Boards of Makerere University, the Uganda National Council for Science and Technology, the Kenya Medical Research Institute, and the University of California San Francisco. The facility assessment data did not require individual informed consent.
Across all 23 facilities, there were a median of 0 pediatricians, 1 general doctor, 0 clinical officers, and 8 nurse-midwives working in neonatal care (Table 1). Pediatricians were only available at 2 (9%) facilities, both of which were regional-or district-level hospitals in Uganda. All 6 Ugandan facilities had �1 general doctor working in neonatal care. Among the 17 Kenyan facilities, 6 had �1 general doctor, 6 had �1 clinical officer, and 5 had only nurse midwives working in neonatal care.
In 2017, 14 of 23 facilities (61%) had the resources necessary to identify a non-vigorous neonate requiring resuscitation, including water and soap (or hand disinfectant), stethoscope, and disposable gloves ( Table 2; Fig 3). Of those, 12 (52%) had a resuscitation area with heat lamp, Any clinical officer, n (%) 6 (26)   10 (44%) had a ventilation bag, and 8 (35%) had an appropriately-sized mask and suction device. Only 1 facility (4%) also had guidelines for referral of sick newborns, which should be present in all facilities offering newborn care as this is essential to help providers identify neonates who require a higher level of care (Table 2; Fig 3). In 2016, 12 facilities (52%) had water and soap (or disinfectant), stethoscopes, and thermometers for identifying neonatal infections (Table 3; Fig 4). Only 3 (13%) additionally had supplies to dose and administer parenteral antibiotics (IV cannula sets, IV bags or tubing, and weighing scale). Although 87% of facilities had ampicillin (or penicillin) and 78% had gentamicin (S2 Table), far fewer had the resources to identify infections and accurately administer first-line antibiotics (13% stage 2 readiness; Table 3; Figs 2 and 4).

Readiness loss by cascade
Along the cascades, there were notable differences in readiness loss from identification (stage 1) through monitoring-modifying therapy (stage 3). At both time-points, it varied least for neonatal resuscitation (range: 13-35) and most for respiratory distress-apnea (range: 65-74; Table 4; Figs 1-3; S1 Fig). There was mild to moderate variability in when readiness was lost along the cascade. In 2016, the majority of readiness was lost during the identification stage for all cascades except essential newborn care, which lost most readiness in the treatment stage (Table 4; Fig 4; S1 and S2 Figs). In 2017, the majority of readiness was again lost in the identification stage for respiratory distress-apnea, jaundice, and neonatal resuscitation (Table 4; Fig  3). In contrast, the infection-convulsions, essential newborn care, and poor feeding-hypothermia cascades lost most readiness in the treatment stage (Table 4; S3 and S4 Figs). Incubator or radiant warmer j 3 (13) 2 (9)

Monitor-Modify
Guidelines: neonatal jaundice k Postnatal gestational age assessment tool k 1 (4) 2 (9) Newborn weighing scale l 1 (4) 2 (9) Guidelines: referral of sick newborns 1 (4) 2 (9) a For each successive item in a clinical cascade, readiness requires the simultaneous presence of all preceding items in that cascade.
b Caffeine citrate (preferred) or aminophylline is required to help prevent and treat apnea in preterm infants.
c Term (size 1) masks are required for normal-weight infants and preterm (size 0) masks are required for infants weighing <2500g [36]. In this study, the presence of term or preterm size masks was assessed in facilities. d Oxygen therapy guidelines are needed to help providers modify oxygen therapy based on oxygen saturation and clinical signs. e CPAP is required to provide respiratory support to infants with severe respiratory distress (in secondary/referral-level facilities). f A weighing scale is required for accurate dosing of antibiotics and other medications.
g Ceftriaxone or cefotaxime is required as a second-line therapy for meningitis and other severe infections not responding to initial antibiotics within 2-3 days.
Ceftriaxone is also used as a first-line therapy with tetracycline eye ointment (in Essential Newborn Care cascade) for ophthalmia neonatorum. h Phenobarbital is required to treat infants who are having convulsions (not assessed in this study

Comparison of readiness in facilities with and without newborn special care units
Across both time-points, overall mean readiness was higher among facilities with a functional NSCU (37%) compared to those without (20%). For both groups, readiness was again lowest for respiratory distress-apnea (27% and 3%, respectively) and highest for essential newborn care (54% and 39%, respectively; Table 5). Among facilities with a NSCU, there was Clinical cascades to assess facility readiness for neonatal care significantly increased identification readiness for respiratory distress-apnea (2016: 63% vs. 7%; 2017: 38% vs. 7%) and jaundice (2016: 75% vs. 33%), and treatment readiness for essential  Table 3 Table 5). At the monitoring-modifying stage, a significant difference was only identified for essential newborn care (2016: 38% vs. 7%; Table 5).

Discussion
The clinical cascade model offers a novel, stepwise approach to quantitatively estimate facility readiness for neonatal care in LMICs. Cascade-derived indicators, including overall readiness, readiness loss by cascade, readiness loss by stage, and aggregate readiness loss, can be used by health administrators, policy-makers, program managers, and researchers to assess and monitor the availability of drugs, supplies, and equipment for facility-based neonatal care in such contexts. By precisely identifying the timing and location of readiness loss, by stage or clinical condition, the cascades could help guide resource allocation decisions and facilitate provision of available, evidence-based interventions to reduce neonatal morbidity and mortality [7]. Further, aggregate readiness loss may be utilized to evaluate and compare readiness for neonatal care across health systems, countries, or geographic regions [33]. In contrast to health facility inventories and signal functions widely used to evaluate EmONC capacity [16,44,45], the cascades pragmatically assess and quantify a facility's capacity to identify and manage six common neonatal conditions. This is accomplished by modeling the hierarchical and interdependent relationship among the resources required to identify a condition, provide initial treatment, monitor clinical response, and modify treatment if indicated [24,25]. Although additional signal functions for emergency newborn care were proposed in 2012 [18], these have not been incorporated in recent WHO guidelines [38,39]. Further, a standardized approach to evaluate readiness for basic newborn care using existing indicators is lacking [16,33].
To help address this gap, the cascade model provides an intuitive set of overall, conditionspecific, and health system readiness indicators for basic and comprehensive levels of newborn care. Notably, this approach entails a negligible increase in data collection requirements compared to existing facility assessment inventories [44,45]. In addition, the neonatal cascades provide detailed information about when and where readiness loss occurs. Aggregate readiness loss can be used as a standardized indicator to quantify and compare readiness at the facility-, health system-, or national-level. This study identified a consistent pattern of 30-32% overall readiness loss across cascades and stages, which is comparable to that seen in the obstetric cascade study in Kenya [33]. The majority of readiness loss occurred in the identification stage; however, loss of treatment readiness increased in 2017, with this stage accounting for the majority of overall loss for essential newborn care, poor feeding-hypothermia, and infectionconvulsions. Comparatively, the obstetric cascade study found increased variability in timing of readiness loss. For example, most loss for hypertensive emergencies occurred in the identification stage, whereas most loss for hemorrhage occurred in the monitoring-modifying stage [33].  Clinical cascades to assess facility readiness for neonatal care Using the cascade model, we found that overall readiness for neonatal care was 26% among the 23 facilities at both time-points. In comparison, three studies using the EmONC signal function classification to assess a total of 431 facilities in sub-Saharan Africa and South Asia found that 0-9% and 0-23% were able to provide basic and comprehensive levels of care, respectively [19][20][21]. Readiness was consistently highest for essential newborn care. This may be related to the fact that resources required for comprehensive neonatal care are more expensive or difficult to maintain than those needed for essential newborn care. Additionally, previous studies evaluating facility or health system capacity have largely focused on indicators related to basic newborn care, including cleanliness, breastfeeding, cord care, tetracycline eye ointment, vitamin K, and resuscitation at birth [18][19][20][21][22][46][47][48]. Conversely, readiness was consistently lowest for respiratory distress-apnea, with only 17-26% of facilities having a pulse oximeter and other supplies required for identification. Two previous studies in Kenya similarly found that 14-18% of public referral hospitals had functional pulse oximeters for pediatric and neonatal care in 2012 [49,50]. Not surprisingly, overall readiness was higher in regional/district/county level and mission/PNFP facilities relative to sub-county and health clinic level facilities. In line with the WHO guidelines on managing possible serious bacterial infections in young infants when referral is not feasible [51], we replaced IV cannula sets with sterile syringes and needles (for IM injection) in the neonatal infection-convulsions cascade sub-analysis of health clinics. By country, overall readiness was higher in Uganda (32-39%) than in Kenya (22-23%). This is likely related to the fact that 100% of Ugandan facilities had a functional NSCU, whereas nearly 90% of Kenyan facilities did not.

Monitor-Modify
Endorsed in 2014, the Every Newborn Action Plan (ENAP) is a global multi-partner initiative to prevent stillbirths and reduce neonatal mortality, with national targets of 10 or fewer deaths per 1000 livebirths by 2035 [52]. To help improve the provision of facility-based care, ENAP has recommended defining indicators for intervention packages by level of care (basic, special, or intensive care), noting that many small and sick neonates can be appropriately managed in NSCUs [3,8,23]. A Delphi study suggested that special care, including KMC, feeding support, IV fluids, oxygen, and management of infections and jaundice, could prevent 70% of deaths in preterm neonates [7]. We identified increased overall readiness among facilities with a NSCU compared to those without. Further, facilities with a NSCU had significantly increased treatment readiness for essential newborn care, poor feeding-hypothermia, respiratory distress-apnea, and infection-convulsions, relative to facilities without a NSCU. Recognizing the need to improve care for small and sick neonates, the Government of India scaled-up the establishment of NSCUs in district hospitals across the country [53][54][55]. A study of eight NSCUs in eight Indian states, all established within the preceding five years, demonstrated that cause-specific mortality due to sepsis and low birthweight decreased significantly over a two-year period [55].
Notably, poor-quality care is now considered to be a greater barrier to mortality reduction in LMICs than insufficient access [56]. The WHO has developed a quality of care framework for pregnant women and newborns in facilities, which highlights the overarching need for both competent human resources and essential physical resources and additionally requires evidence-based practices for routine and emergency care; actionable information systems; functional referral systems; effective communication; respect and dignity; and emotional support [57]. In recent years, increased emphasis has been placed on promotion of respectful maternity care and elimination of abuse during childbirth [58][59][60]. One study found that women who experience discrimination or abuse during childbirth are less likely to seek facility-based delivery care in the future [61]; such experiences may also deter postnatal care-seeking [12]. To help improve clinical outcomes for small and sick newborns, a culture of capability should be promoted in places where fatalism on the part of healthcare providers is common [62,63]. Evidence from high-and middle-income countries has also demonstrated the value of family-centered developmental care for this vulnerable population [64][65][66]. A study in Colombia found that continuing education for care providers, provision of materials for positioning of neonates, and use of an informative video for parents were helpful in promoting related care practices [67].
This study has several limitations. The cascade model assesses the physical readiness of facilities to provide newborn care, but it does not assess human resource availability or healthcare providers' skills. Observational data from 18 LMICs, including Kenya and Uganda, showed that providers fulfilled 45% and 64% of recommended elements of sick child and delivery care, respectively [56], highlighting the fact that provider skill assessment is also imperative. These data are from 23 facilities in two regions within two countries of East Africa, which limits generalizability to all LMIC contexts. All facility assessments were routinely conducted as part of a broader maternal and newborn health research initiative. As a result, a few variables necessary for complete modeling of the essential newborn care, infection-convulsions, and jaundice cascades were not available. Significant and unanticipated reductions in the availability of certain durable goods (e.g., resuscitation area with warmer, glucometer) were identified; however, we did not obtain data about potential reasons why these items were no longer present (or functional). Certain tracer items for basic and special levels of newborn care are poorly defined, which may slightly limit comparisons of this study with previous studies using these indicators from the literature. For example, the EmONC tracer for basic neonatal resuscitation does not specify the ventilation bag or mask size [16], and the ENAP tracer for IV fluid does not specify the type of fluid [8]. Clearly defined tracer items are imperative to standardize readiness estimates for neonatal care and promote comparability across study results and settings.
In the future, research should evaluate the neonatal cascade model in a variety of cultural, regional, and national contexts. In addition, studies comparing this novel model of neonatal care readiness with previous models may be indicated. Notably, the third stage (monitoringmodifying treatment) of each cascade includes one or more guidelines related to newborn care practices, e.g., referral of sick newborns; however, few previous studies assessing facility readiness for EmONC have utilized clinical guidelines as tracer items [33,47,68]. To assess the quality of facility-based neonatal care and compare readiness estimates using different models, inclusion of tracers for key clinical guidelines is essential. Finally, research could evaluate the ability of the cascades, employed as one component of a broader model, to predict neonatal mortality and morbidities related to the 6 conditions and explore the association between aggregate readiness loss and neonatal mortality across countries or regions.

Conclusion
In conclusion, the clinical cascade model provides a novel, stepwise approach to quantitatively assess facility readiness for neonatal care. We identified a consistent pattern of 30-32% readiness loss across cascades and stages at both time-points at 23 facilities in Kenya and Uganda. This aggregate measure could be used to monitor and compare readiness at the facility-, health system-, or national-level. Cascade-derived estimates of readiness and capacity loss may help guide strategies to improve care, prioritize resources, and promote neonatal survival in LMICs.
Supporting information S1