Universal versus conditional day 3 follow-up for children with non-severe unclassified fever at the community level in Ethiopia: A cluster-randomised non-inferiority trial

Background: With declining malaria prevalence and improved use of malaria diagnostic tests, an increasing proportion of children seen by community health workers (CHWs) have unclassified fever. Current community management guidelines by WHO advise that children seen with non-severe unclassified fever (on day 1) should return to CHWs on day 3 for reassessment. We compared the safety of conditional follow-up reassessment only in cases where symptoms do not resolve with universal follow-up on day 3. Methods and findings: We undertook a 2-arm cluster-randomised controlled non-inferiority trial among children aged 2–59 months presenting with fever and without malaria, pneumonia, diarrhoea, or danger signs to 284 CHWs affiliated with 25 health centres (clusters) in Southern Nations, Nationalities, and Peoples’ Region, Ethiopia. The primary outcome was treatment failure (persistent fever, development of danger signs, hospital admission, death, malaria, pneumonia, or diarrhoea) at 1 week (day 8) of follow-up. Non-inferiority was defined as a 4% or smaller difference in the proportion of treatment failures with conditional follow-up compared to universal follow-up. Secondary outcomes included the percentage of children brought for reassessment, antimicrobial prescription, and severe adverse events (hospitalisations and deaths) after 4 weeks (day 29). From December 1, 2015, to November 30, 2016, we enrolled 4,595 children, of whom 3,946 (1,953 universal follow-up arm; 1,993 conditional follow-up arm) adhered to the CHW’s follow-up advice and also completed a day 8 study visit within ±1 days. Overall, 2.7% had treatment failure on day 8: 0.8% (16/1,993) in the conditional follow-up arm and 4.6% (90/1,953) in the universal follow-up arm (risk difference of treatment failure −3.81%, 95% CI −∞, 0.65%), meeting the prespecified criterion for non-inferiority. There were no deaths recorded by day 29. In the universal follow-up arm, 94.6% of caregivers reported returning for reassessment on day 3, in contrast to 7.5% in the conditional follow-up arm (risk ratio 22.0, 95% CI 17.9, 27.2, p < 0.001). Few children sought care from another provider after their initial visit to the CHW: 3.0% (59/1,993) in the conditional follow-up arm and 1.1% (22/1,953) in the universal follow-up arm, on average 3.2 and 3.4 days later, respectively, with no significant difference between arms (risk difference 1.79%, 95% CI −1.23%, 4.82%, p = 0.244). The mean travel time to another provider was 2.2 hours (95% CI 0.01, 5.3) in the conditional follow-up arm and 2.6 hours (95% CI 0.02, 4.5) in the universal follow-up arm (p = 0.82); the mean cost for seeking care after visiting the CHW was 26.5 birr (95% CI 7.8, 45.2) and 22.8 birr (95% CI 15.6, 30.0), respectively (p = 0.69). Though this study was an important step to evaluate the safety of conditional follow-up, the high adherence seen may have resulted from knowledge of the 1-week follow-up visit and may therefore not transfer to routine practice; hence, in an implementation setting it is crucial that CHWs are well trained in counselling skills to advise caregivers on when to come back for follow-up. Conclusions: Conditional follow-up of children with non-severe unclassified fever in a low malaria endemic setting in Ethiopia was non-inferior to universal follow-up through day 8. Allowing CHWs to advise caregivers to bring children back only in case of continued symptoms might be a more efficient use of resources in similar settings. This was a 2-arm community cluster-randomised non-inferiority trial (the TRAction study) carried out in 3 woredas (districts) in Southern Nations, Nationalities, and Peoples’ Region (SNNPR) in Ethiopia. The woredas were purposively selected based on (1) strength of iCCM programme (i.e., consistency in HEW supervision and supply), (2) HEW use rate among caregivers (more than 50 children assessed for fever each month over a 12-month period), and (3) regular concurrent community mobilisation and supportive supervision activities under other grants (to ensure that demand was kept high during the study period). Clusters, defined by health centre (the referral centre and practical training institution for HEWs, where their services are coordinated), were randomised into either the conditional or universal follow-up arm. All 25 health centres and 144 health posts with 284 HEWs in the 3 selected woredas were included in the study, and all children seeking care from the health posts in these clusters were potential recipients of the interventions, in addition to having access to routine care available from private and public health services. Caregivers of children who met the inclusion criteria (fever with a negative malaria rapid diagnostic test [mRDT], and in whom the HEW did not diagnose pneumonia or diarrhoea or identify other symptoms requiring referral on day 1) were counselled to either (1) return on day 3 (universal follow-up arm) or (2) return if symptoms persisted (conditional follow-up arm). Caregivers in both arms were advised to go to the health centre immediately if danger signs, such as convulsion, lethargy, not drinking/breastfeeding, or vomiting everything, developed. Of 25 included clusters, 13 were randomised to the universal follow-up arm and 12 to the conditional follow-up arm. The government of Ethiopia has deployed over 42,000 female CHWs, or HEWs [19,20], to provide preventive, promotive, and curative health services at the community level; since 2010, the full iCCM package has been included in the IMNCI guidelines (with the addition of treatment of pneumonia) and has been scaled up in most regions of the country. There are typically 2 HEWs assigned to a health post in a sub-district with a population of 3,000–5,000; they are supported by the Health Development Army, female volunteers who enhance community engagement and encourage use of maternal and newborn health services [21]. While IMNCI recommends conditional follow-up, HEWs and their supervisors report a range of other practices for children with unclassified fever, including universal follow-up advice, immediate referral to health centres, and treatment with antimalarial tablets [22]. Children aged 2–59 months who presented to the HEWs in the study area with fever (≥37.5°C) or a history of fever, a negative mRDT, no pneumonia or diarrhoea according to iCCM criteria, and no danger signs were eligible to participate in the study. Written informed consent was obtained from each caregiver before enrolment in the study. Cluster randomisation was at the health centre area level; the 25 study health centres had an average of 5 health posts and 7.5 HEWs each. Allocation to the universal versus conditional arm was done via restricted randomisation whereby clusters were balanced on health area estimates of (1) population size, (2) prior 6-month likelihood of mRDT-negative febrile children (number of children mRDT negative/under-5 population), and (3) geographic distance from health post to zonal referral hospital, was performed to minimise the differences between conditional and universal follow-up arms [23]. Sorting of clusters and random selection of schemes were carried out by the study statistician (MP) in Stata 13 (StataCorp, College Station, TX, US). HEWs collected data at enrolment (day 1) using an Open Data Kit (ODK) Collect (version 1.9.1) data collection form on mobile phones, including date of enrolment, a child identifier code, and clinical indicators such as fever (axillary temperature ≥ 37.5°C or, if a functional thermometer was unavailable, hot to touch reported by HEW or caregiver-reported fever in past 2 days), cough, respiratory rate, diarrhoea, and danger signs. The enrolment data were synchronised with a server accessed by a data manager for scheduling of follow-up visits. Six independent evaluators (IEs), with a bachelor’s degree in a health-related discipline, clinical experience using IMNCI, and a minimum of 2 years’ research experience, and who were able to communicate in Amharic and English, were trained for 2 days in study procedures and in follow-up of enrolled children. Each district was assigned 2 IEs, who were blinded to the cluster allocation of the children they were reassessing. In the conditional follow-up arm, HEWs counselled caregivers on how to detect danger signs and to seek care immediately at the health centre if danger signs developed, how to reduce fever using paracetamol, and the need to return at any point to the HEW for reassessment if symptoms remained the same, or worsened. In the universal follow-up arm, caregivers were counselled on all of the above, as well as the need to return on day 3 to the HEW for a follow-up assessment, even if the child had recovered. Caregivers in both arms were informed that a follow-up home visit by an IE would take place. Clinical outcomes were assessed by an IE during a home visit after 1 week (on day 8); if the child had not fully recovered, the child was assessed again by the IE after 2 weeks (day 15) and, if still not recovered, at after 4 weeks (day 29). Caregivers of all children were followed up by a phone call to assess vital status (alive/dead) after 4 weeks. Management of illness at any follow-up visit (i.e., return to HEW on any day, including scheduled assessments) followed national IMNCI guidelines. IEs initially used ODK to collect reassessment data on enrolled children; halfway through the study the data collection software was changed to CommCare (version 2.38.1, Dimagi, Cambridge, MA, US), which allowed for automatic linking of follow-up forms, as well as scheduling of subsequent visits, once the children were registered in the 1-week follow-up form. The replacement system used an automatically generated child identifier code, which reduced the effort of having to manually link the forms, as well as supporting the IEs in tracking the follow-up visits that were due. The data collected during the household follow-up visits included the child identifier code, clinical data, additional antimicrobial treatment, hospitalisation, care-seeking history, and costs, as well as caregiver and household characteristics. For children who were brought back on day 3 for reassessment in the universal follow-up arm and for any spontaneous revisit in both arms, a full reassessment was done by the HEW. If the child still had unclassified fever and a negative mRDT on reassessment, the child was referred to the nearest health centre, as recommended in the national IMNCI guidelines. A rigorous monitoring system implemented by the study team was part of the continuous quality assurance. The data manager reviewed forms submitted to the server daily, and checked for duplicates, completeness, and accuracy before storing them in the project database. Discrepancies, overdue follow-up visits, and other issues were resolved by phone calls to the IEs and during weekly supervision meetings with field research staff. Biweekly field supervision visits to all HEWs were carried out, and district HEW supervisors were trained to monitor HEW trial activities during routine weekly group supervisions. While the protocol stated that a minimum of 10% of all enrolled cases and 50% of children with treatment failure should have a quality control reassessment by a research assistant, the actual percentage was significantly higher. Six months into the trial, all HEWs and their district supervisors had a refresher training in study procedures. In addition, the regional ethical clearance committee members did a field supervision visit during implementation of the project in all 3 districts selected for the study (9 health posts; 3 from each district), and provided feedback recommendations to the study team. The final dataset was analysed in Stata 13. The primary outcome was treatment failure rate on day 8, defined as the proportion of children whose illness was not resolved (child had any of the following: reported fever, danger sign(s), hospital admission, death, malaria, pneumonia, or diarrhoea). Three progressively stricter, and more objective, definitions of treatment failure were added in post hoc analysis to be consistent with a concurrent sister study in the Democratic Republic of the Congo (DRC) [24]: (1) reported fever ≥3 days, danger sign, hospital admission, death, malaria, pneumonia, or diarrhoea; (2) measured axillary temperature ≥ 37.5°C, danger sign, hospital admission, death, malaria, pneumonia, or diarrhoea; and (3) danger sign, hospital admission, death, malaria, pneumonia, or diarrhoea. Secondary outcomes were percentage of caregivers who brought the child to the HEW for the follow-up visit on day 3 in the universal follow-up arm; percentage of caregivers who spontaneously re-presented to HEW for persistence or worsening of symptoms in the conditional follow-up arm, and the timing of these visits; percentage of children receiving antimicrobial treatment in each arm; and severe adverse events in each arm. Severe adverse events were defined as hospitalisation or death. A data monitoring committee (DMC) convened twice during the study to review enrolment rates, demographic and clinical characteristics of enrolled children, and follow-up rates at 1, 2, and 4 weeks, in order to monitor the overall conduct of the study. The DMC was advisory to a study steering committee (SC), which comprised the implementing study team from Malaria Consortium and lead study investigators, who jointly had responsibility for the design, conduct, and analysis of the trial. The SC was responsible for reviewing the DMC recommendations, to decide whether to continue or terminate the study, and to determine whether amendments to the protocol or changes in study conduct were required. We hypothesized that treatment failure at day 8 would not be more common with conditional than universal follow-up. We assumed that about 5% of children in the universal follow-up arm and 6% in the conditional follow-up arm would have treatment failure at day 8 (based on rates of 3% and 8% in previous studies [9,25]). Sample size for a non-inferiority trial was calculated in PASS 15 (NCSS, Kaysville, UT, US). Assuming that the proportion of failure was 5% in the universal follow-up arm and 6% in the conditional follow-up arm, and allowing for non-inferiority if the proportion of failure was as high as 9% in the conditional follow-up arm, a sample size of 2,142 per arm was needed to ensure that the upper limit of the 1-sided 95% confidence interval would exclude a difference in treatment failure of more than 4% with power of 80%. Using a design effect of 3 to account for clustering at the health post and health centre levels, the total sample size required was 4,284 children, with 2,142 per arm; this was inflated to 4,900 to account for potential losses to follow-up. The primary analysis was conducted on the per-protocol population (only including children for whom the primary outcome was collected on day 8 ± 1 and whose caregiver reported receiving follow-up advice from the HEW that was aligned with the study arm). In addition, an intention-to-treat analysis was done, whereby all children with a primary outcome defined were included. We also calculated cluster-specific failure rates on the per-protocol population and with the same model specifications as for the primary outcome. The primary outcome was compared between arms using generalised linear models with a binomial distribution and identity link using a robust variance estimator, treating cluster as a random effect. We applied a conventional statistical non-inferiority test using a CI approach using the exact binomial CI for the difference in overall treatment failure between study arms. Here, we claimed non-inferiority if the upper bound of the 95% CI lay on the negative side of the 4% margin, using a 1-sided test done at the 2.5% significance level. The main analysis was done using the per-protocol population, as is appropriate for non-inferiority and equivalence studies, together with sensitivity analysis in the per-protocol and intention-to-treat populations [26]. All p-values for categorical data were calculated using the Pearson’s chi-squared test, whereas the adjusted Wald test was used for continuous data, accounting for clustering using the svy command in Stata 13. The trial protocol was approved by the SNNPR State Health Bureau on September 23, 2015 (ref P026-19/4511). In addition, approval was obtained from the district authorities and local leaders in the study-area woredas. US Centers for Disease Control and Prevention investigators participated under a non-engaged determination from their Office for Human Research Protections. The study protocol has been published [23] and was registered (ClinicalTrials.gov; identifier {"type":"clinical-trial","attrs":{"text":"NCT02926625","term_id":"NCT02926625"}}NCT02926625) after the first participant was randomised due to a miscommunication between study investigators.