This study was reviewed and approved by the University of California San Francisco Human Research Protection Program Committee on Human Research, IRB #10-02198, Reference #004193. All respondents provided written informed consent. Written informed consent was obtained from the parents/guardians of respondents younger than 18 years of age.

The non-governmental organization Muso and the Malian Ministry of Health (MoH) launched their health system strengthening intervention in the periurban area of Yirimadjo in Mali in September 2008. Mali is ranked amongst the world’s poorest countries and has the world’s eighth highest under-five child mortality rate, projected to be 128/1000 live births in 2012 [27]–[28]. Since the Bamako Initiative of 1987 [29], the public sector health system has utilized a user fee for service health care financing model. Approximately 3.5 square miles on the outskirts of Bamako, Yirimadjo has an estimated population of 56,000, and has experienced rapid population growth due to high birth rates and in-migration. At baseline, the public sector health provided primary care through a primary care center that consisted of a consultation room, medicine dispensary, observation room, and two rooms for labor and delivery. Relays, local volunteers, had been trained through the MoH to share key messages with other community members about maternal-child health. At baseline, there were no Community Health Workers providing community-based management of childhood illness.

In partnership with local, regional, and national structures of the Malian MoH, Muso undertook a three-component intervention to overcome health system barriers to timely access to preventative and curative care (Table 1). The model is diagrammed in Figure S1.

The first component of the intervention involved (i) selecting, training, and employing Community Health Workers to conduct active door-to-door case finding, identifying 16 danger signs and symptoms of childhood illness for children 0–59 months, diagnosing malaria in the home using HRP-2 rapid antigen diagnostic testing, treating malaria in the home with artemisinin-based combination therapy (ACT), referring or accompanying patients with other illnesses to a MoH community health center, conducting follow-up visits for treated patients at 24 and 48 hours, and connecting pregnant women with prenatal care and birthing services (ii) removing user fees to provide free care for all patients who could not afford to pay, as determined by patient self-reporting to their CHW that they were unable to pay (iii) constructing and renovating clinical infrastructure at the MoH community health center and (iv) training of health care providers at the MoH community health center. User fee removal, CHWs, and health center construction were all fully deployed by September 2008.

A team of 20 CHWs and 3 CHW assistants provided outreach and care to the 11,000 households in the area of the intervention. Residents of the communities they serve, this new cadre of community-based health professionals were nominated by their communities and selected by a committee that included representatives of the community, the MoH health center, and Muso. The CHWs were jointly trained and supervised by Muso and the MoH health center clinical team. The area of the intervention, estimated population 56,000, was divided into 20 zones, each served by a CHW. Thus each CHW provided outreach and care to a zone with an average population of 2800, of which 560 were children aged 0–59 months. CHW diagnosis and treatment algorithms were built from WHO recommendations for the Integrated Management of Childhood Illness. These CHW protocols (Figure S2 and Figure S3) are included as supporting information.

The second component of the intervention involved the creation of a rapid referral network to identify sick children and refer them early for assessment. The rapid referral network included 13 local religious leaders, 14 education centers, and 238 community organizers, as well as the households they engaged through outreach work. Members of this network trained in the importance of early diagnosis and treatment of sick children. They identified patients, particularly children, who were sick and referred them to a CHW for assessment. They also taught their neighbors and family members about the importance of early diagnosis and treatment for children with fever.

The third component addressed the socioeconomic determinants of health disparities through microenterprise support, education, and community organizing. A microenterprise program provided training, savings structures, and low-interest and no-interest loans to women to start or expand revenue generating activities. The microenterprise program was designed to increase women’s income. In doing so, the program aimed to increase women’s decision-making power relative to health decisions for themselves and their children, as well as to increase their capacity to spend on health related commodities, such as soap and nutritious foods. Testing these specific hypotheses was beyond the scope of the current study and will be the subject of future research. The non-formal education curriculum, developed and implemented by the organization Tostan, is a human rights-based curriculum offered to adults and adolescents, which trains community members in community-led development, improving living conditions that influence community health. The three-year curriculum includes modules on human rights, problem solving, project management, democracy, health, hygiene, literacy, numeracy and income generation skills. Health skills modules include family planning, prenatal and perinatal health, vaccinations, malaria prevention and care, diarrheal disease prevention and home-based management, and nutrition. Classes met eight hours per week during the three year curriculum. Muso and Tostan together trained 238 Community Organizers to identify and implement community-designed health and development projects.

We conducted independent, randomized cross-sectional household surveys at baseline, 12, 24, and 36 months in the area of intervention starting in June 2008. We included households with a woman aged 16 and older. We sampled 400 households in June 2008 (baseline), June 2009 (12 months), June 2010 (24 months), and 1170 households in June 2011 (36 months). We powered the sample to estimate the percentage of under-five children with a fever in the two weeks prior who received an effective antimalarial within 24 hours of symptom onset, in a population of 56,000, with a 95% confidence level, a 5% margin of error, and a conservative 50% response distribution.

We utilized a two-stage cluster sampling methodology with probability of selection proportionate to size to achieve a self-weighting sample of households. In the first-stage of sampling, we generated 40 (2008, 2009, 2010) or 65 (2011) random non-overlapping latitude-longitude coordinates with 100 m radial buffers within the intervention area. Using less than six month old satellite imagery, we counted the number of structures within each cluster as a measure of population density. In the second stage of sampling, interviewers selected households by arriving at the center of the cluster, spinning a pen to identify a random direction, and visiting every second household and choosing a pre-determined number of households proportional to the population density in that cluster. Within each household, one respondent, a female aged 16 or over, was randomly selected from all eligible respondents based on a KISH table. The survey was adapted from existing validated tools created by ICF International [30] and Population Services International (PSI) and included questions about household demographics, fever in children under five, and birth histories (10-year birth histories in 2008, and 6-year birth histories in 2009–2011). Interviewers were hired for the sole purpose of survey administration, and were not members of the communities they surveyed.

Data on patient encounters were collected through patient encounter forms completed by MoH Primary Health Center staff and CHWs. The number of patient encounters in which user fees were effectively removed was tracked through an electronic database maintained by a data entry professional based at the Primary Health Center. The number of sick patient visits per month was tracked by the Primary Health Center and provided through the Ministry of Health District Health Information Systems office. Program staff submitted forms to track other program output including education center enrollment and microloan repayment rates, which were then verified by program supervisors.

To track exposure to Community Health Worker outreach, we added questions to the 2011 household survey regarding whether and how recently each household had been visited by a CHW.

We conducted an adequacy evaluation based on a full coverage intervention to measure changes in early access and under-five mortality outcomes in the surveyed area [31].

We calculated under-five mortality rates from all births reported by respondents in the five years prior to the survey. We calculated under-five mortality rates and compared risk of death before age five across surveys with Cox proportional hazards regression using survey year as the explanatory variable. Children still alive and under age five at the time of survey were right censored. We accounted for clustering of observations by sampling area with the shared frailty option and compared the difference in under-five mortality rates across surveys with the 2008 survey as the reference. The proportional hazard criteria were met. We multiply imputed missing age of death, month of birth, and year of birth based on other covariates using the MICE system of chained equations, and analyzed the results in Stata (v.10) using the multiply imputed datasets to preserve the original sample variability including variability due to missing data.

We calculated the proportion of children in the household under age five (both biological and non-biological children of respondents) who were reported to be febrile in the prior two weeks and the proportion of febrile children under-five who received an effective antimalarial treatment within 24 hours of symptom onset. We estimated prevalence in SAS (v.9.2) accounting for clustering with the surveyfreq command, and compared estimates with chi-square statistics, reporting prevalence ratios and p-values with 2008 as the reference.

During the period of the intervention, CHWs performed on average 7,109 screening home visits per month actively searching for cases of sick children. CHWs identified sick children at an average of 356 sick child assessments per month, 12,120 sick child home visits between 2008 and 2011. They performed malaria rapid antigen diagnostic testing at 6,943 of those visits, an average of 204 malaria rapid diagnostic tests per month. CHWs identified one or more danger signs of severe malaria, pneumonia, diarrheal disease, or other severe childhood illness requiring immediate accompaniment for health center level care during 4,925 of those home visits. Of the sick children they cared for, they reached 35% within 24 hours of symptom onset, 52% within 48 hours of symptom onset, and 78% within 72 hours of symptom onset (Table 3).

A 2900 square foot clinical care center was constructed for the MoH Yirimadjo Community Health Center and opened September 2008. The expanded Yirimadjo Community Health Center provided 66,631 sick patient consultations during the period of the intervention (September 2008 to June 2011). User fees were removed for consultations, diagnostic studies, services, and medications at 33,800 of those health center visits, for those patients who were determined by CHWs to be unable to pay for care themselves. Figure 1 depicts patient visits at the Health Center and in the community from 2002 to 2011, before and after the September 2008 launch.

Community Organizers implemented community initiatives including a potable water advocacy campaign, after which the government installed more than 30 new clean water access points in the area. The microenterprise program provided training loans to 247 women, who achieved a greater than 99% repayment rate. 1245 adults and adolescents participated in the non-formal education program across 14 education centers.

Figure 2 shows the downward mean trend in under-five mortality over the four surveys, with under-five mortality starting at a pre-intervention rate of 155 deaths per 1000 live births, and ending at 17 deaths per 1000 live births in 2011. We present these results against the trends of annual Demographic and Health Survey (DHS) urban under-five mortality estimates and annual UN national estimates of under-five mortality derived from DHS and census data (Figure 2). In the first year of the intervention, hazard of death before age five were nearly quartered (Hazard Ratio (HR) = 0.27, p = 0.0020). In 2010 and 2011, the hazards of under-five mortality were approximately one tenth that of baseline (2010 HR = 0.07, 2011 HR = 0.10, p<0.0001 for both ratios) (Table 4).

The prevalence of febrile illness in the two weeks prior to the survey among children under-five was 38.2% in 2008, and 23.3% in 2011 (Prevalence Ratio (PR) = 0.61, p = 0.0009) (Table 5). Among febrile children, the proportion that received antimalarial treatment within 24 hours of symptom onset nearly doubled during the study period (PR = 1.89, p = 0.0195) (Table 6).