October 7, 2026 interva.net Editorial

After the Clinic Door: Sierra Leone's Cohort Puts a Number on Post-Discharge Deaths

A Lancet Global Health cohort followed 20,560 infants in three Sierra Leone districts and classified community deaths with the WHO verbal autopsy tool and InSilicoVA — while the facility side used clinical records. The split between the two routes is as informative as the causes themselves.

verbal autopsychild mortalitycause of deathsurveillance

An estimated 4.9 million children under five died in 2024, according to WHO’s current estimates — roughly one death every six seconds. Where each of those deaths happens often decides whether it enters any record at all. A child who dies in a hospital ward acquires a case file. A child who dies at home, weeks after being well enough to be sent home, may acquire nothing.

That asymmetry is what makes a cohort published in The Lancet Global Health worth reading for anyone who works with cause-of-death data. Kenneth Owusu-Kyei and colleagues, working with the ICARIA Safety Team, enrolled 20,560 infants aged six to ten weeks at 14 primary health facilities in three districts of Sierra Leone, recruiting them at their first dose of pentavalent vaccine between March 2021 and April 2024. Median follow-up ran to 16.6 months, ending on 31 December 2024, by which point about 80% of children had completed the full 18-month window.

A study built to notice deaths outside the ward

The measurement machinery is the point. Hospital admissions, deaths and immunisation status were assembled from facility surveillance, household visits, telephone follow-ups and community informants — a purpose-built net rather than the routine registration system most countries rely on. It caught 2,688 admissions (excluding readmissions and in-hospital deaths), an incidence of 8.67 per 1,000 child-months at risk.

Then the deaths were sorted twice over, by two different instruments. Children who died in a facility were classified clinically, using ICD-10. Children who died in the community — 425 of them — were classified with the WHO 2016 verbal autopsy instrument and the InSilicoVA algorithm, which turns a structured interview with a grieving family into a probabilistic distribution across causes. InSilicoVA belongs to the same methodological family as the InterVA model, the probabilistic cause-assignment tool developed at Umeå University, though this study did not use InterVA itself.

Two lists of causes, two kinds of resolution

The comparison is not a clean experiment, but the two distributions sit side by side in the paper. Among admissions, malaria dominated at 54.3%, followed by lower respiratory tract infections including pneumonia at 16.7%, diarrhoea at 9.6%, undernutrition at 7.4% and sepsis at 4.8%.

Of the 241 in-hospital deaths, malaria accounted for 122 (50.6%), lower respiratory infections for 41 (17.0%), sepsis for 33 (13.7%), undernutrition for 17 (7.1%) and diarrhoea for nine (3.7%).

The 425 community deaths coded by verbal autopsy produced a flatter picture: malaria 103 (24.2%), lower respiratory infections 76 (17.9%), “other infectious diseases” 73 (17.2%) and diarrhoea 54 (12.7%).

Read carefully, that contrast carries two different warnings. The children who reach hospital are already a filtered group — the sickest travel, and the ones who die before arrival or at home never enter the admission statistics at all. And where verbal autopsy coding is doing the classification, a large share of deaths lands in a residual “other infectious diseases” category; algorithmic cause assignment has a coarser grain than a clinician with a case file, and the study offers no reference standard against which those community assignments were checked. The two lists are not evidence that the causes of death genuinely differ between ward and village.

The stretch after discharge

The cohort’s most uncomfortable number concerns the period when a child is no longer anyone’s inpatient. Post-discharge mortality ran at 33.8 deaths per 1,000 child-months at risk (95% CI 27.1–42.2), a figure the authors report in the same units they use for admissions incidence. The two are not a like-for-like ratio — one counts deaths, the other admissions — but the size of the post-discharge figure is the study’s own way of saying that a large share of risk sits outside the ward.

That finding is not a surprise to the field. The CHAIN cohort, published in the same journal in 2022, followed 3,101 children aged two to 23 months admitted with acute illness across nine hospitals in six African and south Asian countries, and found that of 350 deaths, 168 (48.0%) occurred within 180 days of hospital discharge. More than half of those post-discharge deaths — 53.6% — happened at home. Almost any cause-of-death system built on facility records will miss them.

What to do about them is far less settled. A systematic review in The American Journal of Tropical Medicine and Hygiene this year screened 4,893 publications and found 17 studies of interventions to prevent post-discharge mortality in sub-Saharan Africa, covering 12,938 participants of whom 10.6% died after discharge. Nutrition supplementation, kangaroo mother care, antibiotics and micronutrients all appear in the literature; only two interventions showed a statistically significant reduction in post-discharge death — vitamin A for children with pneumonia and linkage to services for children with sickle cell disease — and both were graded low-quality evidence. The review’s own conclusion is that no single intervention type worked consistently.

Weight, immunisation, and the limits of association

Two associations in the Sierra Leone cohort are worth stating precisely, and no more strongly than the design allows. Being underweight was associated with higher in-hospital mortality (adjusted odds ratio 1.60, 95% CI 1.21–2.12). Full immunisation was associated with lower mortality overall (adjusted hazard ratio 0.65, 95% CI 0.55–0.76), lower in-hospital mortality (adjusted odds ratio 0.71, 95% CI 0.54–0.95) and lower post-discharge mortality (adjusted hazard ratio 0.56, 95% CI 0.33–0.92).

These are observational associations in a cohort nested within a randomised trial, not causal estimates. Immunisation status is also a marker of how closely a family is engaged with health services, which is exactly the kind of thing that confounds this sort of comparison.

What the study can claim is narrower and still significant: in a cohort with unusually good follow-up, mortality remained high and predominantly caused by preventable infections, and a measurable slice of it occurred after discharge. Its limits are the usual ones for a study of this type — three districts and 14 primary facilities rather than a national sample, roughly 80% follow-up completeness at 18 months, and community causes resting on model output rather than post-mortem verification. The CHAMPS network has taken the opposite route, pairing verbal autopsy with minimally invasive tissue sampling to name pathogens in individual child deaths; that evidence is far more precise and far harder to scale.

Verbal autopsy remains what it has always been: the only route to a cause of death for the children who die where no clinician is watching. The Sierra Leone paper adds a reminder of the accounting problem underneath — the deaths most likely to be unregistered are also the ones most likely to have been preceded by a clinic visit that ended with the child going home.

interva.net is an independent digest on verbal autopsy, the InterVA model and cause-of-death data. It is not affiliated with, endorsed by, or speaking for the InterVA project, Umeå University or WHO.

Sources

← Back to the news digest