Summary of results
Utilising all five methods of data collection, suspected infection was identified in 55.6% of the study participants, a high figure especially when considering some of these infants were ill on multiple occasions. None of the methods of data collection successfully identified all cases of suspected neonatal infection. The IMCI screening forms successfully identified infection in 30.3% of the total study participants and 54.5% of the total number of infants with suspected infection. Evidence for infection from clinical notes was found for only 34.5% of the infants with suspected infection, and microbiological evidence was obtained for only 9.1% of the total suspected with infection. Similar to reports in previous studies, Staphylococcus spp. and Escherichia coli were the most common causative pathogens identified [2, 12]. Maternal report of infection captured the greatest amount of data, as mothers reported illness for 81.8% of the total presumed infected infants. Finally, evidence for antibiotic use was only found in 40% of the total infected infants.
Whilst it is desirable to have the sensitivity and specificity for each method of data collection, this had to be calculated (and must also be interpreted) with great caution. The analysis was calculated on all 99 infants (including those who did not test positive) and assumes, probably erroneously, that the absence of a positive result for any test method equated to a negative test result. Furthermore, the evidence available in the clinical notes had to be utilised as the gold standard and that itself is suspect.
Comparison to other studies
The main difference between this study and others was the prospective rather than retrospective element. A recent cross-sectional study in Eastern Uganda involved conducting face to face interviews with 2237 women who had delivered in the past 12 months, to determine the neonatal mortality rate and influential factors . The retrospective nature of the study meant it was therefore subject to recall bias, and the sole reliance of maternal assessment of ‘danger signs at birth’ (e.g. ‘yellow skin’, ‘difficulty breathing’) was arguably subjective and potentially unreliable. A similar community-based prospective cohort study was carried out in Mbale, Eastern Uganda, to determine perinatal mortality and risk factors of perinatal death for 7 days postnatally in 835 women . There were 34 perinatal deaths (18 excluding stillbirths), none of which were officially registered or had death certificates (again illustrating the difficulty of monitoring infection and NMR using the accepted estimated statistics for the country). Again, the women were interviewed within 4 weeks of delivery to recall their antenatal care attendance, socio-demographic characteristics, and infant mortality. Other studies which considered community infection in neonates generally required the women to attend hospital or antenatal clinics for data collection, due to the difficulty of regularly monitoring infants in the community home.
Limitations of study
The prospective element of this study meant data had to be collected in real time, and the obstacles to data collection faced in this trial highlight why there is a paucity of published research on rural Ugandan mothers and children. Numerous challenges prevented widespread use of the screening forms. The VHWs were not always notified when infants fell ill and may not have been available to visit the mother at home. This likely resulted in cases whereby infection was missed. Mothers also often bypassed the study screening process, attending hospital or health centres directly with their children without notifying the research team, presumably due to concerns about their baby’s health and to avoid delays in seeking care. Others screened positive on the IMCI form, but there was no record of their subsequent attendance at the study health centres as should have happened per protocol. Mothers may have taken their infants to local pharmacies or private clinics to prevent the informal charges that are sometimes demanded by health centre staff. Furthermore, mothers were sometimes absent from home during day 1 or 90 screening, and others failed to notify researchers that their child had been born for some time. All these presented enormous difficulties in collecting accurate, reliable data on infant infection.
Although clinical notes were obtained where possible from hospitals and health centres following referral, delays in researcher notification meant that clinical notes were often no longer traceable. Finite members of staff at the health centres and hospitals were notified of the study, and therefore, the safeguarding of participant notes was limited by staff awareness. This was an important learning point from the pilot study and may have been exacerbated by parental refusal to attend hospital following referral which has been documented in previous community-based studies [20, 31].
Unlike the IMCI screening forms, and clinical or microbiological data which made diagnoses in real time, maternal report of infection captured the greatest number of infants with suspected infection, but was subject to recall bias. Likewise, data obtained on antibiotic exposure was occasionally gathered from free-text comments on the screening forms and therefore also subject to recall bias.
Furthermore, the IMCI form is a validated tool for severe illness, whilst the other methods used in this study are more specific for infections. This makes comparison of infection rates challenging. Mothers were asked: ‘Has your baby been unwell since the last visit?’ ‘How many times has baby been unwell?’ and ‘Has your baby ever been admitted to the hospital since the last visit?’. Besides recall bias, the closed questions neither establish the severity of the suspected infection nor the time period. Thus, it would be possible for the quoted rates to include both minor infections and relapses rather than individual episodes of severe illness. Only 18 of the 45 infants (40%) identified via maternal report were clinically diagnosed. This can be partially explained by lack of care-seeking at health facilities and the difficulty in collecting clinical documentation from busy health centres and might also indicate limitations in maternal recognition and report of severe infection.
Collecting hard evidence for infant infection was challenging. Whilst clinical notes reported infections ranging from neonatal septicaemia to skin pustules and diarrhoea, microbiological evidence for infection was very limited. Although funding for tests was made available for the study, established local practice is not to rely on microbiological testing but rather to provide empirical treatment. Furthermore, in Mbale, antibiotics can be obtained with ease over the counter from local sources without a qualified diagnosis being made. Antibiotic use could therefore not be confidently utilised as an indicator of infection (and especially severe infection) in participants, even though the information was easily obtained retrospectively from mothers.
The free-text comments documented on the screening forms indicated liberal use of freely available antibiotics within the rural setting without a prescription. Antibiotic exposure prior to admission meant limited samples could be obtained from infants, resulting in few cases of illness with microbiological evidence. Although 31% of culture samples obtained were positive (comparable to the 35 and 37% obtained in the Mbarara Regional Referral Hospital and Mulago Hospital studies respectively) [2, 12], only 10 infants with suspected illness had samples taken for culture. Furthermore, the process of transporting samples was not without error: one baby had samples taken for complete blood count (CBC) with blood culture and sensitivity which was delivered to the hospital where there were no personnel to receive it. The sample subsequently haemolysed and was rejected due to a public holiday resulting in the sample remaining at room temperature for > 48 h.
A further reason for the limited number of samples was the lack of skilled staff. Hospitals and health centres in Uganda are often under-staffed, and therefore, necessary tests and investigations cannot be carried out as required, or advised by local protocol. Arguably, this is one of the greatest challenges of carrying out research of this type in this environment. With appropriate funding, it would be possible to set up a study in a high-quality European standard paediatric ward with all the associated resources, but this would ultimately alter the study setting and prevent this being a pragmatic study.
Strengths of study
The study had a high follow-up rate, with all participants being tracked to day 90 postnatally, except the women who withdrew their consent or relocated and those with perinatal deaths. The prospective format gives a variety of strengths: the potential for clear temporal chronology to link the outcome to the exposure, limited risk of recall bias due to rapid assessment with the IMCI screening forms, and clinical or microbiological diagnosis in real time. The variety of methods utilised for data collection meant that a spectrum of illness was captured. In cases whereby multiple methods successfully identified singular episodes of infection, the combined data was complementary, creating a sound clinical picture, which could be utilised to differentiate severe from minor infection.
In future studies, alterations must be made to the methods of data collection and the details gathered. The choice of method will depend on a balance of cost, feasibility, and acceptability. Unless all methods measure severe infection only, differences in infection risk will be incomparable; limiting the study outcome to major infections would permit comparative analysis. The IMCI screening form should only identify severe infection, as should the clinical notes of those presenting directly to health care providers. In contrast, the maternal report of infection captured a huge number of infants with potentially minor or suspected infections. This was compounded by the difficulty of VHWs accessing the community homes making it difficult to verify maternal reports. In order to reduce recall bias and ensure the infections being identified are severe, the maternal questionnaire about recent infections should include greater detail on the symptoms, signs, date, and length of illness. Regular, weekly communication with participants via a toll-free number would improve communication between the researchers and mothers to determine if the child has been ill in the past week (whilst also reducing the risk of recall bias). A researcher could attend the home to assess the ill child or visit the health centre/hospital the mother attended during the week to collect clinical notes if the baby was admitted, to ensure as many infants with suspected infection are captured as possible. Particular attention should be paid around the expected date of delivery and final follow-up, with frequent communication to guarantee adequate data collection.
To improve the collection of both clinical and microbiological data, improved communication between the health centres and laboratory alongside more reliable services will be required for future studies. Although multiple samples were obtained from infected infants, only singular samples yielded results, confirming the importance of obtaining multiple samples where possible and relevant. This thorough method optimised the chance of achieving measurable pathogenic growth and should continue to be utilised in future studies. As many members of clinical staff as possible should be notified of the research taking place at the health centres and local hospital, so they have an awareness of the mothers participating and can notify researchers, keeping notes and samples relating to the participants safe. Where screening forms result in a referral to a health centre or hospital, researchers should attend with the infant in order to collect the clinically relevant notes and make the health practitioners aware that the infant is participating in the study.
The challenge of antibiotic use prior to hospital admission and self-prescribed antibiotic use within the community is difficult to address. During the 90-day interview, numerous mothers described occasions on which their child had been ill, and they had self-medicated them and/or not notified the VHWs. Future studies could monitor antibiotic use reported both verbally and with evidence of prescription or drug packaging to more accurately capture the cohort of infants with suspected infection. Administration of antibiotics is a useful indicator of how ill the mother deems their child to be, particularly as a limitation of maternal report of infection is the risk of identifying minor illness rather than severe infections. The benefit of capturing data through maternally reported antibiotic use includes identifying suspected infection regardless of where the care was sought and would be helpful to monitor as a public health outcome given the financial cost of antibiotic use and development of antibiotic resistance, which is a growing international concern [20, 32]. Drawbacks include a potential overestimation of the frequency of true infections and difficulty in ascertaining which drugs were administered as often drugs are provided over the counter without a prescription. This might be addressed by ensuring prescriptions and packaging are retained for inspection by the research team. Alternatively, to reduce the risk of overestimation, evidence for antibiotic use could be restricted to that provided by qualified health care professionals. The downside would be that it might also result in cases being missed. There is currently a paucity of data on the use of antibiotics within Uganda’s rural communities , and monitoring this in future would be valuable [34, 35].