We found no evidence of an effect of a mass media campaign on child mortality. This finding comes against a background of rapidly decreasing mortality in both groups, which will have reduced our power to detect an effect on mortality (set at 80% to detect a 20% reduction in mortality). The decrease in mortality we recorded is broadly consistent with estimates for Burkina Faso as a whole from the UN Inter-Agency Group for Child Mortality Estimation (IGME). Recent improvements in child survival could reflect changes in national health policies, in particular two rounds of free national distribution of insecticide-treated bednets (2010 and 2013), and the addition of the pneumococcal and rotavirus vaccines to the expanded programme for immunisation in 2013. However, routine health facility data did provide evidence of increased utilisation of health services in intervention clusters relative to control clusters, especially with respect to care seeking for childhood illness. Self-reported behaviours might have been over-reported due to socially desirable bias, especially in the intervention group as a consequence of DMI’s campaign itself. Nevertheless, we observed some evidence of improved care seeking and treatment in the midline survey.9x9Institut National de la Statistique et de la Démographie (INSD). Enquête démographique et de santé et à indicateurs multiples du Burkina Faso 2010. INSD,
Calverton (MD); 2012 ()www.unicef.org/bfa/french/bf_eds_2010.pdf. ()
Google ScholarSee all References Although no overall difference was apparent at the endline survey, the survey data are consistent with increased care seeking among families living within up to 5 km of a facility, with no effect at greater distances.
With only a limited number of clusters available, a major limitation of our trial is that, despite randomisation, important differences between the intervention and control groups at baseline were not unlikely.14x14Hayes, RJ and Moulton, LH. Cluster randomised trials. Chapman and Hall,
Boca Raton (FL); 2009
Crossref | Google ScholarSee all References The use of pre-intervention mortality estimated at baseline survey was precluded by the intervention timeframe, and we therefore used a pair-matched randomisation procedure based on geography and estimated radio listenership. Nevertheless, intervention communities had a different ethnic and religious mix, tended to live further away from health facilities, and experienced higher mortality than the control communities. We generated a confounder score to account for imbalance between groups, but cannot exclude the possibility of residual confounding. Furthermore, contamination of one of the control areas occurred due to an increase in the strength of the transmission signal of the neighbouring radio partner, above that permitted by the national authorities. However, excluding women living in villages where contamination occurred had little effect on the results (data not shown).
The DMI campaign seems to have reached a high proportion of the primary target population as a high proportion of mothers interviewed in the intervention group reported recognising DMI’s spots and listening to the long format programmes. One in five women in the control clusters also reported recognising the spots or long format programme. Excluding the control cluster in which contamination occurred, only a few women mentioned one of DMI’s radio partners when asked on which radio station they listened to these broadcasts, which could suggest courtesy bias or confusion with other radio programmes.
In interpreting these results it should be considered that our survey data are likely to have much lower power than the facility data to detect a change in care seeking. While the survey data include 1000 or fewer sick children per group, the facility data record tens of thousands of consultations. However, both sources of data are prone to errors. Retrospective reporting of illness episodes and care seeking in surveys is known to have important limitations. We used a recall period of 2 weeks, as used in DHS, but it has been shown that recall of disease episodes tends to decline after a few days,20x20Feikin, DR, Audi, A, Olack, B et al. Evaluation of the optimal recall period for disease symptoms in home-based morbidity surveillance in rural and urban Kenya. Int J Epidemiol. 2010;
Crossref | PubMed | Scopus (74) | Google ScholarSee all References,21x21Alam, N, Henry, FJ, and Rahaman, MM. Reporting errors in one-week diarrhoea recall surveys: experience from a prospective study in rural Bangladesh. Int J Epidemiol. 1989;
Crossref | PubMed | Scopus (57) | Google ScholarSee all References,22x22Boerma, JT, Black, RE, Sommerfelt, AE, Rutstein, SO, and Bicego, GT. Accuracy and completeness of mothers’ recall of diarrhoea occurrence in pre-school children in demographic and health surveys. Int J Epidemiol. 1991;
Crossref | PubMed | Scopus (66) | Google ScholarSee all References,23x23Byass, P and Hanlon, PW. Daily morbidity records: recall and reliability. Int J Epidemiol. 1994;
Crossref | PubMed | Scopus (26) | Google ScholarSee all References,24x24Ramakrishnan, R, Venkatarao, T, Koya, PK, and Kamaraj, P. Influence of recall period on estimates of diarrhoea morbidity in infants in rural Tamil Nadu. Indian J Public Health. 1999;
PubMed | Google ScholarSee all References as well as reporting of clinic visits.20x20Feikin, DR, Audi, A, Olack, B et al. Evaluation of the optimal recall period for disease symptoms in home-based morbidity surveillance in rural and urban Kenya. Int J Epidemiol. 2010;
Crossref | PubMed | Scopus (74) | Google ScholarSee all References Thus, our population-based surveys almost certainly missed some episodes of recent illness. However, the routine facility records might also be subject to recording errors and come without precise and up-to-date denominator data. The population of Burkina Faso is estimated to be increasing by about 3% per year11x11Sarrassat, S, Meda, N, Ouedraogo, M et al. Behavior change after 20 months of a radio campaign addressing key lifesaving family behaviors for child survival: midline results from a cluster randomized trial in rural Burkina Faso. Glob Health Sci Pract. 2015;
Crossref | PubMed | Scopus (6) | Google ScholarSee all References and it is therefore likely that the under-5 child population served by the facilities for which we have data was increasing over time. Interpretation of the observed differences between intervention and control groups as being attributable to the intervention requires the assumption that any increases in the underlying populations served by the facilities were of similar magnitude in both groups (or smaller in the intervention group). However, we have no reason to believe that population growth differed between groups.
The facility data suggest a large increase in under-5 consultations in the intervention group in the first year of the intervention. The estimated impacts in subsequent years are smaller. While this apparent decline could be a chance finding, it might reflect attenuation in the effect of the intervention. In Burkina Faso, in-depth interviews with health workers and patients have revealed low satisfaction with the quality of care in public facilities.25x25Gemignani, R and Wodon, Q. How households choose between health providers? Results from qualitative fieldwork in Burkina Faso. World Bank,
; 2012https://mpra.ub.uni-muenchen.de/45375/. ()
Google ScholarSee all References,26x26Melberg, A, Diallo, AH, Tylleskar, T, and Moland, KM. “We saw she was in danger, but couldn’t do anything”: Missed opportunities and health worker disempowerment during birth care in rural Burkina Faso. BMC Pregnancy Childbirth. 2016;
Crossref | PubMed | Scopus (0) | Google ScholarSee all References The low use of and dissatisfaction with community-based insurance in northwest Burkina Faso has been attributed, in part, to the suboptimal quality of care provided, including poor health worker attitudes and behaviours.27x27Robyn, PJ, Fink, G, Sié, A, and Sauerborn, R. Health insurance and health-seeking behaviour: evidence from a randomized community-based insurance rollout in rural Burkina Faso. Soc Sci Med. 2010;
Crossref | Scopus (12) | Google ScholarSee all References In the same area, Mugisha and colleagues found that, while many factors influence initiation of the demand for services, only perceived quality of care predicted “retention” in modern health-care services.28x28Mugisha, F, Bocar, K, Dong, H, Chepng’eno, G, and Sauerborn, R. The two faces of enhancing utilization of health care services: determinants of patient initiation and retention in rural Burkina Faso. Bull World Health Organ. 2004;
PubMed | Google ScholarSee all References They concluded that increasing patient initiation and patient retention require different interventions and that the latter should focus on improving the perceived quality of care. A possible, admittedly speculative, explanation for our findings is that women were initially encouraged by the campaign to take their children to a facility, but that poor perceived quality of care may have discouraged some from returning for subsequent illnesses.
Our findings showed no effect of the campaign on self-reported habitual behaviours, such as child feeding practices, handwashing, and child stool disposal practices. The campaign’s broadcasts were heavily weighted to care seeking rather than home-based behaviours, and as we have discussed previously, it might be harder to achieve sustained changes in habitual behaviours that need to be performed daily with little obvious immediate benefit, than for behaviours that are only performed occasionally and for which some immediate benefit may be perceived.9x9Institut National de la Statistique et de la Démographie (INSD). Enquête démographique et de santé et à indicateurs multiples du Burkina Faso 2010. INSD,
Calverton (MD); 2012 ()www.unicef.org/bfa/french/bf_eds_2010.pdf. ()
Google ScholarSee all References The confidence intervals for the effect of the intervention on habitual behaviours are wide and do not preclude modest but important changes in these behaviours.
While we detected evidence that the intervention was associated with an increase in care-seeking in facilities we did not detect any evidence of a reduction in mortality. There are several possible explanations for this apparent inconsistency. First, our mortality data do not exclude the possibility of an impact on mortality with the lower bounds of the 95% confidence interval for the mortality risk ratio compatible with an important reduction in mortality. The impact of the campaign on child mortality has been modelled using the Lives Saved Tool and showed an estimated 8% reduction in the first year, and 5% reduction in the second and third years (unpublished data). In addition, mortality at baseline differed between the two groups despite randomisation. Although we adjusted for pre-intervention mortality risk and a confounder score, which performed reasonably well in explaining the baseline mortality imbalance, we cannot exclude the possibility of residual confounding which might have masked an intervention effect. Second, while the numbers of consultations with diagnoses of malaria, pneumonia and diarrhoea, three of the leading causes of child death in Burkina Faso, all increased (unpublished data), we have no data on the severity of the episodes for which children were taken to facilities. If most of the increase in consultations was due to children with mild self-limiting illness, then limited impact on mortality might be expected. In some parts of Burkina Faso a preference for traditional care has been reported for some severe manifestations of illness, such as cerebral malaria.29x29Beiersmann, C, Sanou, A, Wladarsch, E, De Allegri, M, Kouyaté, B, and Müller, O. Malaria in rural Burkina Fao: local illness concepts, patterns of traditional treatment and influence on health-seeking behaviour. Malar J. 2007;
Crossref | PubMed | Scopus (65) | Google ScholarSee all References,30x30Some, DT and Zerbo, R. Atypical etiology of malaria: local perceptions and practices for treatment and prevention in the department of Gaoua, Burkina Faso. Med Trop. 2007;
Google ScholarSee all References Third, if the quality of care received at the facility was low, this could limit any mortality reduction through increased care seeking. An evaluation of the quality of care at health facilities for children under-5, conducted in 2011 in two regions in the north of Burkina Faso, found that on average only six of ten tasks that should be performed as part of IMCI were performed.31x31Kouanda, S and Baguiya, A. Evaluation de la qualité des soins prodigués aux enfants de moins de cinq ans dans les formations sanitaires des régions du Nord et de Centre-Nord du Burkina Faso. Ministère de la Santé du Burkina Faso,
Google ScholarSee all References Only 28% of children were checked for three danger signs, and 40% of children judged to require referral by an Integrated Management of Childhood Illness expert were referred by the health worker. In addition, the 2010 DHS indicated that among children in rural areas who were taken to a public primary health facility, 54% of those with fever received an antimalarial, 35% of those with diarrhoea received oral rehydration solution, and 77% of those with cough and fast or difficult breathing received an antibiotic. Fourth, mortality data were collected by interviewing women about their pregnancy histories. Such data are subject to measurement errors. We did a number of checks on the data, similar to those routinely performed by DHS. Apart from heaping of deaths at age 12 months, which occurred to a similar degree in both groups and should not have affected the under-5 (post-neonatal) mortality estimates, these analyses did not identify any major concerns. The cluster-level estimates of mortality risk at baseline correlated well with subnational estimates from the 2010 DHS and the time trend in mortality is broadly consistent with that estimated by the UN Inter-agency Group for Child Mortality Estimation.
In summary, there is evidence that DMI’s campaign led to increased use of health facilities, especially by sick children. However, we noted no effect of the campaign on child mortality. The small number of clusters available for randomisation together with the substantial between-cluster heterogeneity at baseline, and rapidly decreasing mortality, limited the power of the study to detect modest changes in behaviour or mortality. Caution should be exercised in interpreting these results since, despite randomisation, there were important differences between intervention and control clusters at baseline. Nevertheless, this study provides some of the best evidence available that a mass media campaign alone can increase health facility utilisation for maternal and child health in a low-income, rural setting.