Eighty-Two per Thousand: The Child Deaths Burkina Faso's National Survey Could Not Reach
A phone-based mortality survey in Burkina Faso found under-five mortality rising from 47 to 82 deaths per 1,000 live births in conflict-affected regions — a reversal the 2021 DHS did not show, and one that the survey's own sampling frame helps explain.
The 2021 Demographic and Health Survey for Burkina Faso planned 41 survey clusters in the Sahel, the region where the insurgency has been most intense. It dropped 28 of them. Across the country as a whole, only five of thirteen regions were surveyed in full, and the survey report itself asks readers to treat the Eastern and Sahel results with caution.
What follows from that omission is the subject of a study published in June in Tropical Medicine & International Health and now in its October issue. A team led by Kévin Dianou and Bruno Masquelier, with colleagues at the Institut Supérieur des Sciences de la Population in Ouagadougou and at LSHTM, put a mortality questionnaire on the phone instead. In the regions affected by armed conflict, their data show under-five mortality rising from 47 deaths per 1,000 live births in 2018–2019 (95% CI 38–57) to 82 (67–98) in 2020–2022. The 2021 DHS, by contrast, recorded a decline in the same conflict-affected areas, from 54 (43–64) to 45 (36–55).
Two estimates of the same phenomenon, one nearly twice the other, both published in good faith. The difference between them is largely a question of which households could be reached at all.
What the survey could not walk to
The Burkina Faso Rapid Mortality Mobile Phone Survey ran quarterly cross-sectional waves from September 2021 to October 2022. It drew respondents two ways: random digit dialling, with a screening step run by a commercial provider to confirm that numbers were live, and re-contact of people who had answered the 2018–2019 household living conditions survey face to face. Between the two arms, 52,650 numbers were tried and 11,387 women of reproductive age were interviewed. Response rates were 59% in the re-contact arm and 45% in the random-digit arm; the re-contact group needed two calls per completed interview, the random group four.
Weighting carried much of the burden of making the phone sample look like the country. The unweighted respondents were older, wealthier, better educated and more urban than the census population, and raking weights were applied against census margins for age, residence, household assets, education and region. What weighting cannot do is bring back a province that no interviewer could enter. In the paper’s own accounting, only 1.8% of the children in the phone sample (204 of 11,222) lived in provinces entirely outside the DHS sampling frame — the Sahel provinces of Oudalan and Yagha and the Eastern provinces of Kompienga, Komondjari and Tapoa. Yet death rates were higher there, which is why the authors read the gap as a downward bias in the face-to-face survey rather than an upward one in theirs. In the Eastern region the phone estimates stayed above the DHS even for provinces the DHS did cover, which the authors attribute to insecurity that varied inside officially “secure” provinces: the field teams reached the accessible parts, the phone calls reached further.
The phone is not a neutral instrument either
Mobile ownership is the obvious objection, and the authors do not dodge it. Citing DHS figures, they note that about 90% of urban women of reproductive age own a phone against roughly 68% of rural women, and that ownership runs from 38% in the Sahel to 90% in the Centre region around Ouagadougou. Quotas and post-stratification narrow the resulting skew, in the expected direction, but in the absence of a gold standard nobody can demonstrate that it disappears.
A smaller study from Cameroon, published in Conflict and Health in 2020, is a useful check on the assumption that phone surveys simply over-count deaths. In Mora, in the Far North region, MSF and ministry researchers ran telephone interviews and home visits in parallel: the phones produced a crude mortality rate of 0.30 deaths per 10,000 person-days (95% CI 0.16–0.43), the home visits 0.16 (0.05–0.27). Mode effects do not point in a single direction, and they are not uniform across indicators.
A floor, not a ceiling
There is a second reason to read the Burkina numbers as conservative. To keep interviews to around twenty minutes, the survey used a truncated birth history rather than a full one, and the authors state plainly that the truncated instrument tends to produce downward-biased under-five mortality estimates. On their own account, then, 82 per 1,000 in the conflict-affected regions is closer to a floor than to a best estimate. The increase also sits disproportionately among children aged one to four: the relative gap between the phone and face-to-face estimates was 124.6% for that age group against 60.3% for infants, a pattern the authors connect to food insecurity, mobility and delayed care-seeking once children stop being carried everywhere.
The deaths without a cause
Everything above concerns how many children died. The questionnaire recorded deaths and their timing; it did not assign causes, and the paper reports no cause-of-death distribution for the 2020–2022 period in which the mortality reversal appears.
That absence is not unusual, and it is the division of labour the field has settled on: household instruments count deaths, while verbal autopsy — most often the WHO verbal autopsy instrument, in its 2016 or 2022 versions — supplies the causes, by interviewing a caregiver about the symptoms and circumstances preceding a death and having the record coded, whether by physician review or by an algorithm such as those the InterVA model, developed at Umeå University, provides. The measurement problem documented here does not stop at the count. A verbal autopsy sample drawn from a surveillance site, an HDSS or a cluster survey inherits the accessibility of the platform that collects it, and the provinces that fell outside a DHS sampling frame are unlikely to be the ones where cause-of-death data are thickest.
What the study does and does not show
This is a comparison of survey modes, not a validation exercise; there is no gold standard against which either estimate can be scored, and the authors say so more than once. Non-response cannot be broken down by region, because the region of a respondent who did not complete an interview is unknown. Conflict exposure and displacement were not measured directly, so the mechanism behind the rise is inferred rather than tested. And the study covers one country across a defined window, at a moment when the wider survey system is itself under strain: the authors point out that uncertainty over DHS funding after the dismantling of USAID raises the stakes for cheaper alternatives. Earlier this month we looked at a related question from Bangladesh, where a phone survey found deaths that facility- and cemetery-based surveillance could never have captured.
What the Burkina study does establish is narrower and more useful than a national rate. It shows that a sampling frame is a statement about who can be visited, and that a survey which quietly confines itself to the reachable parts of a country will report a stable mortality trend during a period when mortality was rising. That is not a failure of the survey’s arithmetic. It is a failure that leaves no trace in the table.
Sources
- Dianou K, Masquelier B, Helleringer S, et al. Under-five mortality in humanitarian emergencies: a comparison of face-to-face and mobile phone survey estimates from Burkina Faso. Tropical Medicine & International Health 2026;31(10):1370-1380 (open access via PubMed Central) pmc.ncbi.nlm.nih.gov
- The DHS Program — Burkina Faso: DHS 2021, Final Report (FR378) dhsprogram.com
- Gignoux EMH, Donfack Sontsa OT, Mudasiru A, et al. A telephone based assessment of the health situation in the far north region of Cameroon. Conflict and Health 2020;14:82 (open access via PubMed Central) pmc.ncbi.nlm.nih.gov
- World Health Organization — 2022 WHO verbal autopsy instrument who.int