A facility-based unmatched case-control study design was employed to identify the determinants of neonatal sepsis. The rationale for the choice of the design relates to the rarity of neonatal sepsis within the facility’s neonatal catchment population, thus rendering the health centre a ready source of case patients. Although population-based controls would conceivably be more preferable, potential differences in health-seeking behavior between hospital and population-sourced controls suggested the need to recruit controls from the same facility as cases. The study conformed to the STROBE guidelines for reporting of a case-control study ²⁸ .

The study was conducted at the Kahawa Health Centre (KHC) which is a level three state-run facility in the northern part of Nairobi County. The estimated catchment population for this health centre is about 52,193 persons and includes the adjacent peri-urban localities. Most of these areas are predominantly informal settlements characterized by overcrowding, with 99% of inhabitants being young adults. Anecdotal reports connote high neonatal mortality rates in this area.

The study population comprised all neonates presenting to KHC for pediatric services during the span of August–October 2018. Case and control patients were selected from this population based on a predefined set of eligibility criteria. All primary visit neonates (incident cases) and infants whose guardians had consented to participation were included. Premature babies with gestational age less than 37 weeks, babies who had a lower than 2000 g birth weight and neonates with congenital anomalies were excluded from the study.

A case patient was a 0-28 day-old neonate, a resident of the study area, presenting to KHC during the study period with an elevated axillary temperature of ≥37.5°C and any one of the following symptoms of infection: purulent discharge (from ear/eye/umbilicus), respiratory distress/fast breathing (more than 60 breathes/minute), severe abdominal distension, poor feeding (persistent vomiting (last three feeds)/refusal to feed/inability to suck/weak suck), altered mentation (lethargic/unconsciousness/convulsions) or skin changes (deep jaundice/ periumbilical redness) ²⁹ . Considering that KHC registers around two to three neonatal sepsis cases per day, to attain the computed sample, all cases (who met the aforementioned eligibility criteria) presenting to the facility within the study period were prospectively recruited. Recruitment of cases occurred at pediatric outpatient consultation rooms.

Controls were neonates similarly defined as cases (though devoid of sepsis symptoms), presenting to the well-baby clinic during the same two-month time period. Controls were systematically randomly sampled from the well-baby clinic of the facility frequency-matched to the cases by time of presentation.

A definition for umbilical cord hygiene was adopted from WHO’s “five cleans” for postnatal care of the stump ²⁰ , based on indicators that comprised of method of folding the napkin, rooming-in, bathing, handwashing and substance application practices, as reported by caregivers. An aggregate score equal to or above the median would constitute good cord hygiene, while scores below the median would be considered as poor hygiene.

As specified by Kelsey et al. ³⁰ for case-control studies, the required sample size was derived:

n 1 = ( Z α + Z β ) 2 p ¯ q ¯ (r + 1) r (p 1 − p 2 ) 2

n 2 = r n 1

and

p 1 = p 2 OR 1 + p 2 ( OR − 1 )

p ¯ = p 1 + rp 2 r + 1

q ¯ = 1 − p ¯

Where: n ₁ = the number of cases; n ₂ = the number of controls; p ₁ = the proportion of cases with an unhygienic umbilical cord; p ₂= proportion of controls with an unhygienic umbilical cord specified at 37.6% based on a previous study ³¹ . Notably, Z _α/2 = 1.96 for the 2-tailed confidence level of 95%; Z _{1- β} = −0.84 for the desired statistical power of the study set at 80%; and r=2 as the specified ratio of controls to cases. The odds ratio (OR) for the umbilical cord hygiene-neonatal sepsis association was estimated at 2. With an anticipated 5% non-response rate, the required sample size was 312: 104 cases and 208 controls.

Other than the primary exposure variable, the other predictor variables were maternal and neonatal factors. These were gathered using a semi-structured questionnaire, available as Extended data ³² . Maternal factors consisted of socio-demographic factors (age of mother, level of education, marital status, parity and religion), the antenatal history of the mother (number of antenatal care (ANC) visits, history of receiving health education, tetanus toxoid immunization, prenatal maternal bacterial infection, birth attendance, place of delivery and type of delivery) and post-natal history factors (history of illness or pregnancy related complications such as postpartum depression, nutritional status or other comorbidities). Neonatal risk factors included low APGAR scores of <7 at 5 minutes (whose signs included scores of appearance, pulse, grimace, activity, and respiration), neonate’s age, sex and invasive procedures (use of medically invasive instruments/resuscitation at birth). Table 1 displays assessment of these variables. Figure 1 provides a conceptual framework of the relationship between the aforementioned predictors and the outcome.

The research commenced after receiving written clearance from the Kenyatta National Hospital (KNH)-University of Nairobi (UoN) Ethics and Research Committee (P438/06/2018) and the Nairobi County Health Services (Ref. No. CMO/NRB/OPR/VOL.1/2018/91). Additionally, written informed consent was obtained from the mother/index care-giver for their neonate’s participation in the study.

Prior to commencing the data collection, two research assistants were trained on screening patients, complete neonatal medical examination and standardized interview techniques to reduce interviewer bias. Additionally, caregivers could have resorted to unhygienic cord practices (such as harmful applications) as a way to treat an already septic neonate. Hence, the possibility of reverse causality was reduced by focusing on incident cases. Attempts to minimize recall bias were made by referencing the mother-child booklet to ascertain information regarding some antenatal and perinatal information such as the number of antenatal visits, the neonate’s date of birth, neonatal APGAR score and resuscitation history.

The questionnaires were checked for completeness and qualitative data coded. The data were double-entered by two independent data entry clerks into EpiData version 3.1 spreadsheet. The principal researcher cross-checked the computerized data base against the questionnaires that had been administered. The dataset was exported to Stata software, version 13 (Stata Corporation, College Station, Texas, USA) for cleaning and analysis. For continuous variables’ descriptive statistics, data were summarized by means, medians and ranges. For categorical variables, data were summarized using frequency tables, proportions and percentages.

Scoring of the umbilical cord hygiene variable’s five components was standardized such that those responses that were desirable as per WHO essential newborn care guidelines received a higher value (≥1). A value of zero was awarded to responses inconsistent with these guidelines ^{41– 43} . A total score was then reached by summing up the individual component’s scores. Notably, a cord with an aggregate score below the median score of 7 was designated as improper cord hygiene, whereas one with a score equal to or above the median was deemed to have proper cord hygiene.

A logistic regression model was used to assess the crude association between umbilical cord hygiene and neonatal sepsis. For a sensible interpretation of APGAR score’s effect on neonatal sepsis, it was grouped into two categories ²⁷ : ≥7 or <7. To evaluate the potential confounding effect of neonatal and maternal factors on the umbilical cord hygiene-sepsis relationship, each of the predictors was screened for unconditional associations with neonatal sepsis at a 5% significance level. Qualifying variables were further screened for a significant association with umbilical cord hygiene at similar level of significance.

Variables that met these criteria were considered as potential confounders to the cord hygiene-neonatal sepsis relationship and therefore were included in a multivariable model to adjust for their confounding effect on this relationship. Here, a backward step-wise approach was applied to eliminate variables if there was not more than a 30% change in the regression coefficient for umbilical cord hygiene upon their exclusion ⁴⁴ . To evaluate the impact of umbilical cord hygiene in the neonatal population (the proportion of neonatal sepsis that could be prevented by adhering to proper umbilical cord care), a PAF was computed as described by Dohoo et al. ⁴⁴

AF p = p d ( aOR − 1 aOR )

Where: AF _p is the Population Attributable Fraction; pd is the proportion of total cases in the population arising from improper cord hygiene; aOR is the adjusted odds ratio for cord hygiene derived from the multivariable model.

A total of 312 participants (104 cases, 208 controls) were recruited into the study but those who consented to participation were 309. Of the 208 potential controls, three declined consent. Additionally, three others did not meet the eligibility criteria and were excluded, leaving 202 eligible controls that participated. A flow diagram illustrating the recruitment and enrollment process is shown in Figure 2.

Descriptive statistics for the demographic variables are indicated in Table 2. Notably, males comprised 55.8% ( n=58) of cases and 47.0% ( n=95) of controls. The mean neonatal age was 19.7 days; the mean age of cases and controls being 16.5 days (range: 5-28 days) and 21.3 days (range: 3-28 days), respectively. Regarding marital status, 69.2% ( n=72) of cases’ mothers were married compared to 81.7% ( n=165) of controls’. Only 13.5% ( n=14) of the cases’ caregivers had received up to tertiary level of education compared to 22.3% ( n=45) of the controls’.

A description of the participants’ cord care practices is displayed in Table 3. In this population, majority of mothers reported use of chlorhexidine/surgical spirit (64%, n=197). Among cases, slightly over a third (35.6%, n=37) had surgical spirit/chlorhexidine applied as compared to about four-fifths (79.2%, n=160) of the controls. Of concern, saliva/ash was applied among 10.6% ( n=11) of cases compared to 2.5% ( n=5) of the controls. In this study setting, about two-thirds (65.7%, n=201) of mothers fastened their babies’ diapers below the umbilical stump. Roughly 30% (29.8%, n=31) of the case respondents revealed that they folded the neonate’s napkin below the cord in comparison to 84% (84.2%, n=170) of the control participants.

Almost four-fifths (78.1%) of caregivers declared that they washed their hands when changing the napkins/diapers. This consisted of 54.8% (n=57) of the cases’ and 90.1% (n=182) of the controls’ mothers. Regarding the cleansing substance employed by those who reported handwashing, only 44.4% (n=136) used both water and soap. In particular, whereas 61.4% (n=124) of controls’ mothers stated they used water and soap before cord handling, only 11.6% (n=12) of cases’ mothers did the same. The practice of rooming-in was followed by approximately all (99.4%, n=304) of the participants. Amongst these were 98.1% (n=102) of the cases and all the controls. Sponge-bathing was the bathing practice recorded by most (64.4%, n=197) of the participants in the present study. However, only 28.9% (n=30) of the cases were sponge-bathed.

The proportion of mothers/care-givers who had improper hygiene practices was 35.3% (n=108), with unhygienic cord status being disproportionately high in case (72.1%, n=75) than in control mothers (16.3%, n=33), Table 3.

The crude association between cord hygiene and neonatal sepsis is captured in Table 4. Notably, the odds of neonatal sepsis in infants who had improper hygiene was approximately 13 times higher (OR=13.24; 95% CI: [7.5; 23.4]) compared to those with proper hygiene.

Of the variables screened, neonatal factors registering a significant association with neonatal sepsis were: low APGAR score ( P=0.001), invasive procedures ( P=0.007) and neonate’s age ( P<0.001). With respect to maternal factors, marital status ( P=0.05), initiation of breastfeeding ( P=0.006), type of feed ( P<0.001) and pregnancy-related events ( P=0.005) were found to be significantly associated with neonatal sepsis ( Table 4). To qualify as potential confounders, these significant factors were further evaluated for an association with the primary exposure as presented in Table 5. Following the assessment, the variables: low APGAR score, invasive procedure, neonate’s age, marital status, type of feed and pregnancy-related events were significantly associated with cord hygiene and thus were offered to the multivariable model to adjust for their potential confounding effect.

From the multivariable analysis, none of the six factors assessed confounded (resulted in a >30% change in the coefficient for umbilical cord hygiene) the primary association between umbilical cord hygiene and neonatal sepsis ( Table 6), and as such, the OR of 13.34 was used to compute the PAF. The estimated PAF was 66.7% (95% CI: 62.5; 69.0).

Harvard Dataverse: Replication Data for: Association between umbilical cord hygiene and neonatal sepsis among neonates presenting to a primary care facility, Kenya: A case-control study. https://doi.org/10.7910/DVN/FSXPR8 ³² .

This project contains the following underlying data:

Harvard Dataverse: Replication Data for: Association between umbilical cord hygiene and neonatal sepsis among neonates presenting to a primary care facility, Kenya: A case-control study. https://doi.org/10.7910/DVN/FSXPR8 ³² .

This project contains the following extended data:

Data are available under the terms of the Creative Commons Zero “No rights reserved” data waiver (CC0 1.0 Public domain dedication).