Can traditional birth attendants be trained to accurately identify septic infants, initiate antibiotics, and refer in a rural African setting?

INTRODUCTION

Neonatal mortality accounts for about 40% of all childhood mortality in low- and middle-income countries. Neonatal sepsis, birth asphyxia, and neonatal hypothermia are responsible for the largest number of preventable deaths.^1–3 These conditions are particularly difficult to manage in remote rural settings where home deliveries predominate. In such settings, traditional birth attendants (TBAs) provide 20%–40% of obstetrical care.⁴ Given their close access to the mother/infant pair, strengthening TBAs' capacity could be effective in reducing neonatal mortality.

The Lufwanyama Neonatal Survival Project (LUNESP) was a community-based effectiveness trial in Zambia that assessed the impact of a package of TBA-delivered interventions on all-cause mortality through postpartum day 28.^5,6 In LUNESP, TBAs were randomized either to continue their existing standard of care (controls) or to receive training and supplies enabling interventions targeting key preventable causes of neonatal mortality (intervention): birth asphyxia, hypothermia, and neonatal sepsis.⁷ The intervention had 2 components: (1) a simplified version of the neonatal resuscitation protocol (NRP), which targeted deaths from birth asphyxia and neonatal hypothermia; and (2) administration of oral antibiotics with facilitated referral (AFR) to a rural health center, which aimed to reduce deaths from neonatal sepsis.

The LUNESP interventions reduced all-cause day-28 mortality among live-born neonates by nearly half (relative risk [RR] = 0.45, 95% confidence interval [CI] = 0.33–0.9), with the largest reductions during the first 48 hours of life (7.8 deaths/1,000 live births vs. 19.9 deaths/1,000 live births, RR = 0.4, 95% CI = 0.19–0.83).⁶ However, mortality also trended lower during weeks 2 to 4 (RR = 0.47, 95% CI = 0.20–1.11). A late effect of NRP is unlikely to account for this but is consistent with a benefit from AFR. To explore this further, we conducted a secondary analysis focusing on the process by which the intervention TBAs identified, treated, and referred neonates to receive the AFR intervention. We addressed the following issues:

1. How often did the intervention group TBAs refer, and how often was referral coupled with the first dose of amoxicillin (or vice versa)?

2. What were the clinical indications cited for referrals, and how accurately did these predict a fatal outcome for the infant?

3. Is there evidence that TBAs were exercising reasonable clinical judgment in the implementation of the AFR intervention?

RESULTS

The intervention TBAs conducted a total of 2,007 deliveries, of which 38 were stillbirths (1.9%). Among live-born neonates, we had complete follow-up data on 1,889/1,969 (95.9%) (Figure). Table 1 summarizes baseline characteristics of the mothers and the 1,889 neonates who comprised the analysis set, stratified by the neonates' final vital status by day 28 of life (that is, alive vs. dead). Maternal baseline characteristics were similar between the 2 groups, except that more mothers of surviving neonates had been dewormed during pregnancy. Surviving neonates (n = 1,846) were significantly more likely to be female or to have been exclusively breastfed during the study period than those who died (n = 43). The gestational age of surviving neonates also trended lower, but the difference was not statistically significant.

• Download figure
• Open in new tab
• Download powerpoint

Figure.

Data on Live-Born Neonates, June 2006–November 2008

Table 2 summarizes characteristics of the TBAs. Most TBAs were married, identified farming as their primary occupation, and had worked as TBAs for an average of 6.3 years. Their educational backgrounds were quite limited: 83.3% either had no formal schooling or had failed to advance beyond primary school.

Of the 1,889 neonates, 208 (11.0%) were referred to an RHC, and 11 (0.6%) were referred twice. Thus, a total of 219 referrals occurred, 113 during week 1 (51.6% of referrals) and 106 during weeks 2–4 (48.4% of referrals). The majority of neonates (176/208 [84.6%]) completed their referral to the RHC. TBAs administered amoxicillin 188 times, 99 times during week 1 (52.7% of total administrations) and 89 times (47.3% of total administrations) during weeks 2–4. Among referred neonates, 171/208 (82.2%) received amoxicillin; among neonates given amoxicillin, 171/183 (93.4%) were referred (referral status unknown for 5 subjects).

Referred neonates had been ill for a median of 2 days prior to the TBAs' evaluation. TBAs cited a median of 3 criteria to justify their referrals, with a range of 0 (8 times) to 8 (2 times) reasons. A single reason for referral was cited 19 times; 2 reasons, 23 times; 3 reasons, 52 times; 4 reasons, 73 times; and 5 or more reasons, 41 times. The 6 most common reasons for referral, alone or in combination, were because: (1) the TBA thought the neonate “appeared ill” (82.8%), (2) the mother thought the neonate “appeared ill” (80.8%), (3) “felt hot” (41.2%), (4) “had a cough” (40.4%), (5) “was not making urine” (31.9%), or (6) was having “difficulty breathing” (26.4%).

Table 3 summarizes the accuracy for the 16 referral criteria for identifying neonates who subsequently died. Individually, none displayed both high sensitivity and high specificity. Moreover, several signs were nearly ubiquitous in their presence and some were tautological. For example, 95% of referred neonates included the criterion that the TBA thought the baby “appeared ill,” but logically a TBA would be unlikely to refer otherwise. Moreover, this criterion was never cited alone but was always accompanied by other, more specific criteria. Criteria that were cited frequently were “felt hot/cold,” “had cough,” “refused to feed,” had “difficulty breathing,” and “not making urine.” Conversely, “rapid breathing,” “bulging fontanel,” and “infected umbilicus” were rarely cited, and no neonates were referred for “diarrhea” or “chest wall in-drawing.”

Overall, the most useful sign for both predicting and excluding a fatal outcome was the criterion “difficulty breathing,” with sensitivity of 65% and specificity of 78%, yielding positive and negative LRs of 3.0 and 0.4, respectively. Infrequent signs that predicted a fatal outcome were: “baby sleepy/unarousable” (LR+ 9.1), “rapid breathing” (LR+ 5.9), “bulging fontanel” (LR+ 5.9), “refuses to feed” (LR+ 5.4), “convulsions/fits/seizures” (LR+ 5.4), “infected umbilicus” (LR+ 4.5), and “floppy/poor muscle tone” (LR+ 3.4). Conversely, the only criterion strongly predictive of infant survival was if the TBA did not feel that the infant looked sick, where the LR- was 0.3. The opinion of the mother was neither sensitive nor specific (LR+/− both 1.0).

Among live-born neonates, 43 (2.3%) died during the first 28 days of life. Among referred neonates, 9.6% (20/208) ultimately died. By contrast, among neonates who were not referred, only 1.2% died (22/1,658), with 1 neonate whose referral status was missing. Since a neonate who dies shortly after birth has little opportunity to be referred, we assessed referral status relative to the timing of deaths. Of the non-referred deaths, 10/22 (45%) occurred on the first day of life, while only 4/20 (20%) of the referred deaths occurred on the first day of life.

Neonates sick enough to warrant referral (regardless of whether given amoxicillin or not) were nearly 8 times more likely to die than babies who were not referred (RR = 7.93, 95% CI = 4.4–14.3). Similarly, neonates deemed sick enough to receive amoxicillin (regardless of referral status) were nearly 5 times likely to die as those not given amoxicillin (RR = 4.7, 95% CI = 2.5–8.7).

The TBAs recorded a subjective impression of illness severity for 185/208 referred neonates (88.9%), as summarized in Table 4. Neonates judged as “extremely sick” by the referring TBA were far more likely to die than those deemed “not sick” or only “moderately sick” (RR = 8.61, 95% CI = 4.0–18.5).

DISCUSSION

In LUNESP, about 11% of neonates cared for by the intervention TBAs were targeted for AFR, a rate that is consistent with the incidence of serious bacterial infections during the first 28 days of life.¹⁰ Notably, neonates who were referred were approximately 8 times more likely to have a fatal outcome than those who did not, and those who received amoxicillin were nearly 5 times more likely to die. It makes little sense that referral in itself causes death and implausible that a safe and well-tolerated antibiotic like amoxicillin would increase mortality. Therefore, a more logical interpretation is that referral and/or use of amoxicillin were both markers of very ill children at high risk of death. This interpretation is supported by the observation that neonates deemed “extremely sick” by the referring TBA were about 9 times more likely to die than neonates judged less severely ill. Taken together, we conclude that the TBAs demonstrated sound clinical judgment and reacted according to their training, which strongly supports the feasibility of using TBAs in this role.

With that said, in order for an intervention like AFR to be effective, several conditions must be met. First, the CHW must have sufficient clinical judgment to accurately identify those children who are in crisis. Second, they must be empowered, trained, and equipped to act upon that determination. Third, the actions that ensue as a consequence must be sufficient to interrupt the disease process that would ultimately lead to death.

Our analysis suggests that the first 2 conditions were met, but evidence for the third is more ambiguous. In our main effects analysis, during weeks 2–4 of life, the period during which AFR would presumably have been most effective (and the effects of NRP minimal), mortality among neonates cared for by intervention TBAs was reduced by about 50%, but this did not reach statistical significance.⁶ Our study had been powered to detect an overall effect on mortality due to the combination of NRP and AFR, so the infrequency of sepsis could have left us underpowered to isolate the effect of AFR. This possibility could be evaluated in a larger clinical trial. However, another possibility is that the receiving RHCs were insufficiently prepared to manage neonatal sepsis, which would be an essential pre-condition for the third condition listed above. If so, 2 potential solutions are suggested.

First, one could focus resources on strengthening the “back end” of the referral process, that is, the receiving RHCs. In this approach, one would continue to limit the role of TBAs to giving a first dose of antibiotics and facilitating referral, essentially using the TBAs to extend the reach of the RHCs into the community while focusing resources to strengthen the capacity of the RHCs to manage newborns with serious bacterial infections.

Alternatively, one could invest in strengthening the “front-end” by increasing the capacity of the TBAs themselves, following the model of Bang and Bang in India.^11–13 There, village health workers were responsible not just for identifying sepsis but also for administering the full antibiotic treatment course in the community using a combination of oral and injectable antibiotics. This approach is ambitious since it requires that CHWs be trained to use injectable antibiotics dosed according to infant body weight and that systems be established for managing contaminated sharps, storing antibiotics appropriately, and for reclaiming expired drugs. However, this strategy minimizes the delay between identification of a sick child and the start of definitive antibiotic therapy, and it largely eliminates the problem of non-adherence to referrals.

One distinction: the CHWs from studies in South Asia had significantly more training than the LUNESP TBAs, both in terms of their background education and in the intervention training specifically.^14–16 Even so, a hybrid of these 2 models might yet be feasible: TBAs could be responsible for identifying neonates with possible sepsis, initiating therapy, and referring, but the single-dose amoxicillin could be replaced with broader spectrum antibiotics with longer half-lives, or combinations of drugs such as the oral cotrimoxazole and injectable gentamicin used by Bang and Khanal.^12,16 Decisions about which strategy is most appropriate will depend on the capacity of the CHWs, the local epidemiology of neonatal sepsis, the capacity of RHCs, and especially the average transit times to the RHCs. Logically, the strategy of embedding more capacity in TBAs or other CHWs becomes more attractive as distances to the RHCs increase.

Several other findings in this analysis were of interest. First, one of the most common reasons cited for initiating AFR was that the TBA thought the neonate “appeared ill.” This criterion had been included to empower the TBA to initiate AFR based on general clinical impressions, absent more specific definable signs or symptoms.⁸ Notably, among referred neonates who did not meet this criterion, mortality was about one-third lower. This lends further credibility to the clinical judgment of the TBAs and is consistent with other researchers' experience with more highly skilled CHWs in Bangladesh, India, and Nepal.^14–16 By contrast, the criterion “mother thought infant looked sick” was neither sensitive nor specific.

Second, the most clinically useful criterion was “difficulty breathing.” This criterion predicted mortality when present (LR+ 3.0), and it was protective when absent (LR- 0.4). Conversely, “rapid breathing” (a sign of “severe pneumonia”) was rarely cited, and “chest wall in-drawing” (a criterion for “very severe pneumonia”), was never cited. Both signs require some skill to identify, and, in the case of rapid breathing, a timer to count respiratory rates, so their absence as reasons for referral by TBAs should be interpreted cautiously. The clinical sign that most strongly predicted a fatal outcome was “difficulty arousing the child,” with a 10-fold increased likelihood of death. It is relevant that several of the referral criteria that performed best in the hands of TBAs were included in the young infant integrated management of childhood illness algorithm and validated in the Young Infant Clinical Signs Study.^17,18

Third, the AFR training appeared successful at coupling the actions of referring and administering amoxicillin. In more than 80% of instances, the one behavior predicted the other. While one can lament the fact that roughly one-fifth of referrals were not completed, this rate is comparatively high relative to other similar community intervention strategies of this kind.^19,20 Nonetheless, it is clearly a vulnerable point in the AFR strategy. While our data do not indicate the reason for failure to complete a referral, previous research has suggested that mothers may decline referral for a variety of reasons, including distance to facility, financial barriers, and reluctance to comply in the absence of a spouse's authorization.^{15,19,21–24}

A general limitation of this analysis was that despite having very complete data from the TBAs about the reasons for referral, data were not collected from the RHCs, which were outside of the LUNESP study. Future studies should assess both sides of the referral system, and also assess the average elapsed time between a TBA's first contact with a sick infant and the first dose of amoxicillin, as well as average transit times to the nearest RHC.

Despite these limitations, our findings provide cause for optimism. Can TBAs be trained to identify sick neonates, refer them, and initiate treatment at the community level? The answer is “yes.” Whether this translates into reduced mortality is a testable question. Moreover, these results should be highly generalizable. The LUNESP TBAs were relatively advanced in that they had all received formal training in obstetrical care prior to this study, but they were not highly educated (83% never advanced beyond primary school), and most were illiterate, requiring that our trainings be done without text materials. As such, our study demonstrates that lack of formal education does not mean that individuals are unintelligent or un-trainable. Quite the opposite: despite their lack of formal schooling, the Lufwanyama TBAs proved to be eager and successful students, capable of mastering complex concepts and acting upon them appropriately. Our experiences justify further investigations to assess the effectiveness of community-based identification and presumptive treatment of sepsis in the hands of a relevant cadre of CHWs who, by dint of their obstetrical work, have close access to newborns during their period of greatest vulnerability.