https://orcid.org/0000-0002-1586-9564
Figures
Abstract
During 2010–2012, extreme food insecurity and famine in Somalia were estimated to account for 256,000 deaths. Since 2014 Somalia has experienced recurrent below-average rainfall, with consecutive failed rains in late 2016 and 2017 leading to large-scale drought, displacement and epidemics. We wished to estimate mortality across Somalia from 2014 to 2018, and measure the excess death toll attributable to the 2017–2018 drought-triggered crisis. We used a statistical approach akin to small-area estimation, and relying solely on existing data. We identified and re-analysed 91 household surveys conducted at the district level and estimating the crude (CDR) and under 5 years death rate (U5DR) over retrospective periods of 3–4 months. We captured datasets of candidate predictors of mortality with availability by district and month. We also reconstructed population denominators by district-month combining alternative census estimates and displacement data. We combined these data inputs into predictive models to estimate CDR and U5DR and combined the predictions with population estimates to project death tolls. Excess mortality was estimated by constructing counterfactual no-crisis scenarios. Between 2013 and 2018, Somalia’s population increased from 12.0 to 13.5 million, and internally displaced people or returnees reached 20% of the population. We estimated an excess death toll of 44,700 in the most likely counterfactual scenario, and as high as 163,800 in a pessimistic scenario. By contrast to 2010–2012, excess deaths were widespread across Somalia, including central and northern regions. This analysis suggests that the 2017–2018 crisis had a lower, albeit still very substantial, mortality impact than its 2010–2012 predecessor. Despite modest elevations in death rate, crisis conditions were widespread and affected a population of millions. Humanitarian response to drought-related crises in Somalia needs to be strengthened, target the most vulnerable and emphasise very early interventions.
Citation: Warsame A, Frison S, Checchi F (2023) Drought, armed conflict and population mortality in Somalia, 2014–2018: A statistical analysis. PLOS Glob Public Health 3(4): e0001136. https://doi.org/10.1371/journal.pgph.0001136
Editor: Andreas K. Demetriades, Royal Infirmary of Edinburgh and University of Edinburgh, UNITED KINGDOM
Received: September 9, 2022; Accepted: March 22, 2023; Published: April 12, 2023
Copyright: © 2023 Warsame et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All data and analysis scripts, along with instructions for replicating the analysis, are available on https://github.com/francescochecchi/mortality_small_area_estimation.
Funding: This work was funded by the UK Department for International Development, now part of the UK Foreign Commonwealth and Development Office, FCDO (to AW, SF and FC) through the Research for Evidence Division (RED). However, the views expressed and information contained in it are not necessarily those of or endorsed by FCDO, which can accept no responsibility for such views or information or for any reliance placed on them. Additional funding for all authors (AW, SF, FC) came from UK Research and Innovation as part of the Global Challenges Research Fund, grant number ES/P010873/1. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Background
Somalia has experienced recurrent climate- and armed conflict-driven crises over the past 30 years. As a result, the population has limited resilience and is vulnerable to shocks brought on by such crises. In 2010–2012, largely as a result of consecutive failed rains and limited humanitarian access and assistance, a famine occurred across south-central Somalia. A study commissioned by the United Nations (and co-authored by some of us) estimated 256,000 deaths attributable to exceptional food insecurity during this period, of which approximately half in children under 5 years [1].
The two years following the famine saw modest improvements, coinciding with intensified humanitarian assistance and moderate rainfall. By end 2014, however, reduced humanitarian access due to renewed insecurity, as well as below-average rainfall, had returned pockets of Somalia to acute emergency conditions [2]. The years 2015 and 2016 saw a mixture of failed seasonal rains and exceptional flooding, complicated by the return of Somalis from Kenya refugee camps and war-stricken Yemen [3].
In 2016, the sustained underperformance of seasonal rains in Puntland and Somaliland triggered both governments to declare states of emergency. Most regions of Somalia experienced far below-average rainfall in September-December 2016, compromising key harvests and livestock viability [4]; as shown in Fig 1, in some regions drought persisted into 2017, and large movements of internally displaced persons (IDPs) were seen, topping 2.6 million by end 2018. Large-scale epidemics of cholera and measles, typical of food insecurity and displacement situations, also occurred during this period [5]. Abundant rainfall was however recorded from mid-2018. We aimed to update our 2010–2012 analysis by estimating mortality across Somalia during 2014–2018, while also quantifying the excess death toll attributable to drought during 2017–2018.
See Methods for data sources.
Methods
Study population, period and outcomes
Our analysis covered all of Somalia (including Somaliland and Puntland) and was stratified by district and month, spanning the period January 2014 to December 2018, with excess mortality estimated for 2017–2018, the drought-affected period of interest. We sought to estimate all-age and under 5y mortality: corresponding indicators are the crude death rate (CDR) and the under 5 years death rate (U5DR); the latter, unlike under 5y mortality ratios (which measure the probability of survival to age 5y), expresses the incidence of death among children.
Study design
Our method consists of six mostly sequential steps, summarised in Fig 2, and is broadly classifiable as a small-area estimation approach. It rests on constructing a statistical model to predict mortality based on a combination of previously collected data, and using the model, in conjunction with population denominators, to retrospectively estimate death rates and tolls across the crisis-affected population and period of interest. The model is also used to predict mortality in counterfactual scenarios of no crisis, and the difference between these counterfactual scenarios and the estimated mortality is taken as the excess, crisis-attributable mortality. Details on theory, data management, analytic steps and R software [6] implementation are presented in Checchi et al. [7] and S1 Text. All data and analysis scripts are uploaded to https://github.com/francescochecchi/mortality_small_area_estimation. Below, we summarise data sources, analysis steps and specify adaptations made for the Somalia context.
Data sources and management
All data were previously collected for routine humanitarian response and/or health service provision purposes, and were either in the public domain or shared in fully anonymised format.
Ground mortality data.
We sourced all available retrospective household surveys conducted during the analysis period by different humanitarian actors using the Standardised Monitoring of Relief and Transitions (SMART) methodology [8]: while these surveys are primarily conducted to estimate the prevalence of acute malnutrition, they often include a mortality questionnaire module, which elicits information from respondents on the demographic evolution (composition, births, deaths, in- and out-migration) of their households during a ‘recall’ period of 3–4 months [9,10]. Prior to 2016, SMART surveys often relied on an ‘aggregate’ questionnaire, which elicits only numbers of household members and demographic events; since 2016, only a more detailed ‘individual’ questionnaire has been used, whereby each household member present during the recall period is listed. Surveys use systematic random sampling or two-stage cluster sampling with probability of cluster selection proportional to size; we used surveys whose sampling universe was an entire district, or urban areas or IDP settlements within a district. We extracted meta-data on, cleaned and re-analysed all surveys with datasets available to us, estimating CDR, U5DR and other demographic indicators using generalised linear models at the household level, assuming a Poisson distribution offset by household person-time at risk, and adjusting standard errors for intra-cluster correlation. We also computed a quality score for each survey based on SMART criteria for anthropometry [7].
We had access to 201 surveys that estimated retrospective mortality in Somalia between 2013 and 2018 inclusive (Fig 3). Most (175, 87.1%) were conducted by the FSNAU, and the remainder by four other agencies. A raw dataset was available for all surveys, but a detailed report only for 28 (13.9%). We excluded 104 (51.7%) of the surveys from analysis. The main reason for exclusion (n = 87) was because surveys were designed to be representative of livelihood zones rather than districts: as population and predictor datasets were mostly stratified by district, analysing these surveys would have required additional assumptions (livelihood zones straddle multiple districts, and vice versa). Other exclusion reasons were unclear sampling frame (n = 11) and non-standard questionnaire and unclear recall period (n = 6). This left 97 eligible surveys, of which six done in 2013 were not used for modelling as predictor data coverage was insufficient during this year. Of the 91 surveys entered into statistical models, most (n = 63) used an individual questionnaire; the remainder (n = 28) used the older aggregate questionnaire. Survey availability by district-month is shown in Fig A in S1 Text.
Demographic and displacement data.
Somalia last conducted a census in 1975. The four alternative census estimates known to us, and falling within the analysis period, consisted of (i) the March 2014 United Nations Population Estimation Sample Survey (UNPESS), carried out by the United Nations Population Fund, based on a stratified sample of clusters within urban, rural, IDP and nomadic communities [11]; (ii) January 2015 estimates from AfriPop (now WorldPop: https://www.worldpop.org/), which used a validated statistical model to estimate population density by 100m2 pixel based on remote sensing [12] (iii) December 2018 data from the Polio Eradication Initiative, which in Somalia updates target vaccination population denominators through active enumeration (non-public); and (iv) January 2019 Expanded Programme on Immunization data, adjusted based on vaccination campaign performance data (non-public).
Internal displacement data came from the United Nations High Commissioner for Refugees (UNHCR)-led Protection and Return Monitoring Network (PRMN) [13], which tracks displacement and returnee movements from one district to the next as reported by a network of humanitarian and civil society actors, with monthly data available from January 2016 to December 2018. Data also include the reported main reason for displacement. We also reviewed situation reports on the humanitarian ReliefWeb information platform (www.reliefweb.int) and on the UNHCR web site (www.unhcr.org) to approximate refugee flows in and out of districts.
Data on predictors of mortality.
Referring to a published causal framework [14] of excess mortality in crises, we searched for online open-access data sources and held discussions with likely data holders (United Nations agencies, humanitarian coordination mechanisms, Somali government agencies) to identify and source possible predictors of death rate for which datasets with reasonable completeness (≥ 70% complete for ≥ 70% district-months) were available. Candidate predictors for which we found data are detailed in Table 1. Where relevant, we divided predictor values by reconstructed population denominators (see below). Market price data, necessary to compute terms of trade indicators, featured substantial missingness, which we dealt with through manual imputation, by attributing to district-months with missing data a weighted mean of values from districts in the same region (relative weight of 1) and values from districts outside the region (relative weight 0.3); weights were chosen arbitrarily. After imputation, we also smoothed.the price time series using the smooth.spline R package, with spar (smoothing function) value set to 0.3, selected after extensive visual inspection.
Reconstruction of population denominators.
The UNPESS demographic survey estimated the prevalent proportion of IDPs within each district as of start 2014. As no incident displacement data were available before 2016, we simply assumed no further displacement during 2014–2015: this assumption may be reasonably accurate, given that throughout Somalia this period saw no major acute emergencies and relatively low levels of insecurity; humanitarian actors we liaised with also suggested that this was a plausible assumption. From 2016 onwards, we computed a monthly net flow of IDPs and refugees in and out of each district, based on UNHCR data. We applied these flows, along with an assumed 2.1% yearly growth rate (difference between UN-projected crude birth and death rates [21]) to the four alternative census estimates, back- or forward-calculating monthly district populations from the timepoint at which each estimated was centred. We consulted documentation available for each of the four demographic estimates to score their quality on a scale from 0 to 1, based on published criteria [22]. We used these scores as a weight to average the alternative population time series into a single estimate for our analysis. Lastly, we computed under 5y populations by applying the median proportion of children in this age group (25.0%) across the household surveys we reviewed.
Predictive models
We evaluated the accuracy of different candidate sets of predictors to predict mortality (CDR and U5DR) by fitting quasi-Poisson models of these predictors to mortality survey data (household-level counts of deaths, offset by household person-time at risk during the recall period); average values of each predictor over the survey’s recall period, and in the district surveyed, were used. After screening out predictors with poor fit, we did a brute-force search across all possible candidate models, selecting the final model out of the top 20% best-fitting alternatives. Model selection procedures are detailed in Checchi et al. [7]. Briefly, the validity (predictive accuracy) of each candidate model was assessed primarily through ten-fold cross-validation within the full dataset, with the mean Dawid–Sebastiani score across all folds as the metric of fit (a low score indicates good fit). As a secondary validation, we also fit models on a 80% random ‘training’ sample of surveys and observed predictive accuracy on the remaining 20% ‘holdout’ sample. Among shortlisted models, we manually selected one that had the highest possible predictive accuracy on both cross-validation and holdout data while also being as parsimonious as possible and featuring predictors (e.g. epidemic occurrence) that were plausibly sensitive to crisis conditions, and thus could be used to meaningfully construct counterfactual scenarios. We explored both categorical and continuous versions of variables, and lags up to 6 months where plausible. We introduced random effects (survey cluster and district) and plausible interactions, but retained neither as they worsened model fit. Lastly, we computed robust coefficient standard errors.
Estimation of excess death tolls
Working with the selected models for CDR and U5DR, we predicted death rates and (after multiplying by population denominators) tolls, by district-month. We also estimated what mortality would have been in the most likely, reasonable best- and reasonable worst-case counterfactual scenarios (Table 2): each scenario was constructed by varying both population denominators and model predictor values in accordance with assumptions on what would have happened in the absence of a crisis. We generated 10,000 bootstrap sets of actual and counterfactual predictions by sampling from model error distributions, and for each set computed excess mortality as the difference. We then computed point estimates (modes) and 95% percentile intervals from the distributions of bootstrap samples.
Sensitivity analyses
We wished to examine the sensitivity of estimates to two key possible error sources. Firstly, we explored potential systematic or unsystematic error in the population and displacement input data by replicating the analysis for different combinations of relative errors in district population and displacement estimates. Secondly, we considered possible under-reporting of under 5y deaths, as noted in a previous South Sudan analysis [23], and suggested by the surprisingly low ratio of U5DR to CDR (see below). For different assumed proportions of child deaths not reported, we computed the corresponding ‘unseen’ number of deaths in each survey; then, for each of 10,000 bootstrap replicates, we attributed these unseen deaths at random across the surveyed households and ran all subsequent steps of the analysis.
Ethics
The study was approved by the Ethics Committee of the London School of Hygiene & Tropical Medicine (ref. 15334) and the Research and the Ethics Review Committee of the Ministry of Health and Human Services, Somali Federal Republic (ref. MOH&HS/DGO/1944/Dec/2018). SMART survey participants (all adults) provided verbal informed consent on behalf of their households, in accordance with the SMART survey protocol [8]. Our study relied only on previously collected anonymised and unlinkable datasets, and did not collect any primary data.
Results
Crude survey estimates
Table 3 reports descriptive statistics for the 97 eligible mortality surveys. The highest death rates were estimated by surveys done in 2017. Eligible surveys featured a median ratio of U5DR to CDR of around 1.5, with 43% of under 5y deaths among infants; ‘injury’ caused some 5% of deaths on average. Net migration from households was mostly negative throughout the period, with the exception of 2017. Additional crude trends in demographic indicators over time and by region are shown in Fig C of S1 Text.
Evolution of population denominators
Between 2013 and 2018, we estimated that Somalia’s population increased from 12.0 to 13.5 million, but with a discrepancy among sources of up to 3.5 million (Fig 4). Some regions (Bay, Bakool, Lower Shabelle) saw substantial declines in population, while Banadir and Lower Juba experienced a marked increase in population, reflecting internal displacement during the period (Figs D, F in S1 Text). Overall, 2,319,000 people were reported to have become newly displaced or returned to their district of origin during 2016–2018, with reasons reported as drought (1,262,000, 54.4%), insecurity (698,000, 30.1%), flooding (291,000, 12.6%) and other (68,000, 2.9%). As a proportion of the population, this corresponds to an increase from about 8% to 20% in IDPs and returnees (Fig E in S1 Text), with marked regional differences (Fig F in S1 Text). We estimated that net refugee migration was 144,000 during the analysis period.
UNPESS = United Nations Population Estimation Sample Survey. WHO = World Health Organization.
Predictive model
Table 4 summarises final predictive models for both CDR and U5DR (for the latter, the same model specification as for CDR was selected; while other models with equivalent predictive accuracy were available, we considered it advantageous for interpretation that the CDR and U5DR estimates arise from the same statistical process). In both models, death rate increased with incidence of armed conflict, the rate of SAM admissions into nutritional therapy and with non-zero measles incidence, but decreased as the facility-based incidence of malaria increased (see Discussion). Both models featured comparable fit on the training dataset, on 10-fold cross-validation, and on the holdout dataset (Figs G, H in S1 Text).
Estimates of mortality
In the most likely counterfactual scenario, the excess death toll between Jan 2017 and Dec 2018 across Somalia was estimated at 44,700 people, out of some 454,500 total deaths (Table 5). A considerably lower excess was estimated for children under 5y, amounting to about 21% of the all-age excess death toll. However, under a pessimistic scenario the death toll rose to as many as 163,800 (and 61,400 children under 5y). The 2017 excess death toll was about double that in 2018.
When aggregated by region (Table 6), the estimated death rates per 10,000 person-days ivaried from 0.33 in Woqooyi Galbeed to 0.60 in Gedo (CDR), and from 0.42 in Awdal to 1.02 in Hiraan (U5DR), with a markedly lower ratio of U5DR to CDR in the North-East and North-West. The highest excess death rates for all ages were observed in the North-East and in Hiraan, while among children under 5y South-Central Somalia had the highest excess mortality estimates (see also Figs I, J, K, L in S1 Text). Estimates by district are shown in Table A in S1 Text.
During 2014–2018, CDR across Somalia remained within a limited range, but an appreciable elevation (up to 0.09 per 10,000 person-days higher than the most likely counterfactual) was estimated from early 2017 to mid-2018, coinciding with the food security crisis period (Fig 5). The pattern was similar for U5DR, but with a less marked elevation in 2017–2018 (Fig 5).
Trends in the estimated crude death rate (top panel) and under 5 years death rate (bottom panel). Actual estimates, i.e. under observed conditions, are shown with a dotted solid orange line (point estimate) and orange shaded area (95%confidence interval). The green lines indicate counterfactual levels (most likely scenario: Solid green line; reasonable worst-case and best-case scenario: Dashed green lines).
On visual inspection, estimates of the excess death rate correlated broadly with the forecast severity of the crisis as of early 2017, as summarised by the proportion of people projected to be in phases 3 (crisis) or 4 (emergency) of the Integrated Phase Classification (IPC) system for each region of Somalia (Fig 6).
Discussion
To our knowledge this is the first comprehensive analysis of mortality patterns in Somalia since 2014 and in particular death tolls attributable to the food security crisis of 2017–2018. It follows on from a similar study [1] covering the period October 2010 to April 2012, when a very severe crisis, featuring large pockets of famine, swept through the country [24]. Despite some methodological and data source differences with the 2010–2012 study, the present analysis suggests that the 2017–2018 crisis had a lower mortality impact (about 45,000 excess deaths, compared to 258,000 in 2010–2012, when CDR across South-Central Somalia peaked at some three times the baseline).
Nevertheless, excess mortality estimates for 2017–2018 remain staggering, indicating considerable unmet needs in the response to this crisis. This large death toll, occurring despite an increase in CDR of ‘only’ about 15%, also reflects a much larger population at risk than in 2010, due to demographic growth and more geographically widespread drought conditions: in 2010–2012, mortality was concentrated in Bay, Lower and Middle Shabelle and Banadir regions, but in 2017–2018 Central and North-East Somalia (Puntland) also seemed heavily affected. During the period 2016–2018 mortality surveillance within IDP camps in the Afgooye corridor reported under 5 deaths rate as several times higher than the emergency threshold on several occasions [25]. The findings of this surveillance system may be an underestimation as deaths in transit were not fully captured.
By contrast to the afore-mentioned 2010–2012 study (1), we found no consistent association of terms of trade (purchasing power) with mortality. The 2010–2012 study suggested that mortality increased steeply below the 10,000 Kcal cereal per daily wage threshold. As shown in Fig M in S1 Text, this threshold was crossed in most regions both in 2008 and in 2010–2011, while in 2017 terms of trade varied little or remained above 20,000 Kcal cereal per daily wage. Despite this, the unusual levels of rainfall, displacement and nutritional therapy admissions in 2017–2018, along with country-wide measles and cholera epidemics (typical of food insecurity crises), all indicate that a substantial crisis did occur in Somalia during this period, with estimated mortality roughly correlating with IPC severity phase classifications.
We conducted a recent ecological analysis exploring drivers of displacement in Somalia between 2016–18. We found that largescale early displacement was strongly associated with a number of crisis related risk factors such as failed rains and subsequent food insecurity while conflict intensity was weakly associated [26]. We plan further analyses to explore reasons the causal relationships among drought, food insecurity and outcomes upstream of mortality, including prevalence of acute malnutrition and epidemic incidence, so as to better understand the dynamics of this latest crisis, and the role of food security versus other drivers. Notably, large-scale displacement in this recent drought has been linked to a ‘pull’ factor from urban centres offering humanitarian assistance, and the widespread sale of land by impoverished farmers in south-central Somalia [27].
Study limitations
The validity of the predictive model is central to the estimates’ robustness. Cross-validation and the accuracy of prediction on holdout samples support external validity, with little evidence of systematic bias. Internal validity is suggested by the plausible associations in the final models, namely evidence (albeit with weak significance) of higher mortality as a function of armed conflict intensity, admissions for severe acute malnutrition and occurrence of measles. Less interpretably, mortality was associated with increasing malaria incidence in health facilities: this may be a proxy for levels of drought (lower rainfall could have led to decreased mosquito breeding and thus malaria transmission: otherwise put, malaria transmission may have correlated with better-than-average food security) or better access to health services if malaria caseload was a proxy for outpatient service utilisation.
A possible source of bias (albeit with unclear directionality) is inaccuracy in the source demographic estimates. We attempted to mitigate this by averaging the four available estimates based on their assumed robustness. However, the observed discrepancy among estimates suggests considerable uncertainty, particularly when considering estimates at the regional and district scale. Moreover, our redistribution of population among districts based on internal displacement movements rests on the accuracy of PRMN reports, which are not based on statistically representative estimation methods, but rather ground informants. Though the PRMN project captures both departures and returns, it is plausible that the latter flows would be less systematically reported.
We estimated a relatively low excess mortality among children under 5y, compared to among all ages: by comparison, about half of excess deaths in 2010–2012 were in this age group, and meta-analyses of SMART surveys have generally found the U5DR to CDR ratio to be around two, rather than 1.5 in this study [28,29]. A similar analysis in South Sudan [23] suggested that deaths among infants had been substantially underestimated by SMART surveys, possibly due to stigma associated with child deaths or faulty administration of the mortality questionnaire. It is possible that surveys included in this analysis were also subject to some underestimation in neonatal and infant events, as suggested by a lower crude birth rate (36 versus 43 per 1000 person-years) and proportion of infants among under 5y deaths (43% versus 60%) in comparison to estimates by demographic models for Somalia [21].
To explore sensitivity of our estimates to key potential error sources, we (i) assumed different levels of bias in source demographic estimates and reported displacement movements, and (ii) varying proportions of under-reporting among deaths under 5y (see Checchi et al. [7] for detail). Generally, excess mortality estimates appeared relatively insensitive to bias in displacement figures, but decreased substantially as bias in demographic estimates (either over- or underestimation) increased (Fig N in S1 Text). As under-estimation in U5DR rose from 0% to 50%, a doubling in the excess death toll under 5y was projected, but all-age death tolls increased only moderately (Fig O in S1 Text). Our overall assessment of systematic bias in our estimates is presented in Table 7.
Conclusions
This study finds evidence of elevated mortality during a drought-triggered crisis in Somalia over 2017–2018, despite a lesser effect on food security than in previous similar events, and a more proactive and far-reaching humanitarian response [30]. Our findings indicate that even at moderate levels of population stress (for example, IPC phase 3), excess mortality accrues. Therefore, even if the sole aim of humanitarian actors is to support survival, a response needs to be implemented to scale very early in the curve of deterioration in food security and other upstream crisis indicators [31]. While our analysis does not have granularity below district level, we speculate that mortality risk is likely to be clustered within particularly vulnerable communities and households: more specific targeting of limited resources to support these communities and individuals would probably increase the cost-effectiveness and efficiency of humanitarian responses, particularly in food insecurity crises.
Somalia and other fragile regions of the world appear to face an increased threat of drought due to climate change, compounded by global inflation in food and commodity prices. At the time of writing, Somalia and parts of the Horn of Africa face a new, potentially severe food insecurity crisis [32]. Understanding the mortality impact of these events can support rational resource allocation and benchmark the adequacy of humanitarian responses. As such, mortality analyses should arguably become a systematic component of monitoring and evaluation in drought-triggered crises. It is also critical that the narrative of Somalia’s crisis not lose sight of human security as a key driver of mortality, as shown in our model. Armed conflict and the securitisation of much of Somalia exacerbate the severity of drought and underlie the chronic vulnerability of Somali people to unfavourable climate: the restoration of peace and governance to all of Somalia holds perhaps the greatest potential for attenuating the impact of droughts and other natural disasters on its population.
Acknowledgments
We are grateful to Anna Carnegie for project management support and the Federal Ministry of Health and Social Services of Somalia for facilitating data acquisition. We are also grateful to Amy Gimma and Christopher Jarvis for technical advice. Lastly, we thank all agencies that provided data, and the field data collectors from the Food Security and Nutrition Analysis Unit and other organisations who collected primary mortality household data.
The authors are solely responsible for the analyses presented here, and acknowledgment of data sources does not imply that the agencies providing data endorse the results of the analysis. The opinions expressed in this article are those of the authors only and do not necessarily represent the decisions, policies, or views of the London School of Hygiene & Tropical Medicine.
References
- 1. Food and Agriculture Organisation, Famine Early Warning Systems Network. Mortality among populations of southern and central Somalia affected by severe food insecurity and famine during 2010–2012. Rome and Washington, DC: FAO and FEWS NET; 2013 May.
- 2. Food Security and Nutrition Analysis Unit—Somalia, Famine Early Warning Systems Network. Technical release: Over one million people in Somalia face acute food insecurity as food crisis worsens. Nairobi: FSNAU; 2014. Available: http://www.fsnau.org/downloads/FSNAU-FEWSNET-Technical-Release-August-2014.pdf.
- 3. Food Security and Nutrition Analysis Unit—Somalia. Somalia Food Security & Nutrition Analysis Post Deyr 2015/16 Technical Series Report No VII. 64. Nairobi: FSNAU; 2016. Available: http://www.fsnau.org/downloads/FSNAU-Post-Deyr-2015-16-Food-Security-and-Nutrition-Analysis-Technical-Report.pdf.
- 4. United Nations Office for the Coordination of Humanitarian Affairs. Somalia: Humanitarian Response Plan Revision—May 2017. Mogadishu: UN OCHA; 2017. Available: https://www.humanitarianresponse.info/en/operations/somalia/document/somalia-humanitarian-response-plan-revision-may-2017.
- 5. Seal A, Checchi F, Balfour N, Nur AH, Jelle M. A weak health response is increasing the risk of excess mortality as food crisis worsens in Somalia. Confl Health. 2017/07/07 ed. 2017;11: 12. pmid:28680460
- 6. R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2021. Available: https://www.R-project.org.
- 7. Checchi F, Testa A, Gimma A, Koum-Besson E, Warsame A. A method for small-area estimation of population mortality in settings affected by crises. Popul Health Metr. 2022;20: 4. pmid:35016675
- 8.
Standardised Monitoring and Assessment of Relief and Transitions (SMART). Measuring Mortality, Nutritional Status, and Food Security in Crisis Situations: SMART Methodology. SMART Manual Version 2. 2017. Available: https://smartmethodology.org/survey-planning-tools/smart-methodology/.
- 9. Cairns KL, Woodruff BA, Myatt M, Bartlett L, Goldberg H, Roberts L. Cross-sectional survey methods to assess retrospectively mortality in humanitarian emergencies. Disasters. 2009;33: 503–21. pmid:19500327
- 10. Checchi F. Estimation of population mortality in crisis-affected populations: Guidance for humanitarian coordination mechanisms. Geneva: World Health Organization; 2018. Available: https://www.who.int/health-cluster/resources/publications/LSHTM-Mortality-Estimation-Options-oct2018.pdf.
- 11. United Nations Population Fund. Composition and Demographic Characteristics of the Somali People. Nairobi: UNFPA; 2016. Available: http://www.dns.org.so/docs/Analytical_Report_Volume_2.pdf.
- 12. Linard C, Alegana VA, Noor AM, Snow RW, Tatem AJ. A high resolution spatial population database of Somalia for disease risk mapping. Int J Health Geogr. 2010;9: 45. pmid:20840751
- 13. United Nations High Commissioner for Refugees. UNHCR Somalia Protection & Return Monitoring Network (PRMN): Notes on methodology. Nairobi: UNHCR; 2017. Available: https://reliefweb.int/sites/reliefweb.int/files/resources/PRMN-MethodologyNotesv1.0.pdf.
- 14. Checchi F, Warsame A, Treacy-Wong V, Polonsky J, van Ommeren M, Prudhon C. Public health information in crisis-affected populations: a review of methods and their use for advocacy and action. Lancet. 2017/06/13 ed. 2017;390: 2297–2313. pmid:28602558
- 15. Desert Research Institute and University of Idaho. Climate Engine. 2020. Available: http://climateengine.org.
- 16. Huntington JL, Hegewisch KC, Daudert B, Morton CG, Abatzoglou JT, McEvoy DJ, et al. Climate Engine: Cloud Computing and Visualization of Climate and Remote Sensing Data for Advanced Natural Resource Monitoring and Process Understanding. Bull Am Meteorol Soc. 2017;98: 2397–2410.
- 17. Food Security and Nutrition Analysis Unit—Somalia. Early Warning—Early Action Dashboard. Nairobi; 2020. Available: https://dashboard.fsnau.org/.
- 18. Raleigh C, Linke A, Hegre H, Karlsen J. Introducing ACLED-Armed Conflict Location and Event Data. J Peace Res. 2010;47: 1–10.
- 19. Humanitarian Outcomes. Aid Worker Security Database. 2020. Available: https://aidworkersecurity.org/.
- 20. Food Security and Nutrition Analysis Unit—Somalia. Somalia: Livelihood Zones. Nairobi: FSNAU; 2015. Available: https://www.fsnau.org/downloads/FSNAU-FEWSNET-Updated-Somalia-Livelihood-Zones-Map-April-2015.pdf.
- 21. Population Division of the Department of Economic and Social Affairs of the United Nations Secretariat. World Population Prospects: The 2019 Revision. New York: UN; 2020. Available: https://population.un.org/wpp/DataQuery/.
- 22. Abdelmagid N, Checchi F. Estimation of population denominators for the humanitarian health sector: Guidance for humanitarian coordination mechanisms. London: London School of Hygiene and Tropical Medicine; 2018. Available: https://www.who.int/entity/health-cluster/resources/publications/LSHTM_Population_Guidance_GHC_15-Nov-2018.pdf.
- 23. Checchi F, Testa A, Warsame A, Quach L, Burns R. Estimates of crisis-attributable mortality in South Sudan, December 2013- April 2018: A statistical analysis. London: London School of Hygiene and Tropical Medicine,; 2018. Available: https://crises.lshtm.ac.uk/2018/09/25/south-sudan/.
- 24. Maxwell D, Fitzpatrick M. The 2011 Somalia famine: Context, causes, and complications. Glob Food Secur. 2012;1: 5–12.
- 25. Seal AJ, Jelle M, Grijalva-Eternod CS, Mohamed H, Ali R, Fottrell E. Use of verbal autopsy for establishing causes of child mortality in camps for internally displaced people in Mogadishu, Somalia: a population-based, prospective, cohort study. Lancet Glob Health. 2021;9: e1286–e1295. pmid:34416214
- 26. Yuen A, Warsame A, Checchi F. Exploring the temporal patterns and crisis-related risk factors for population displacement in Somalia (2016–2018). J Migr Health. 2022;5: 100095. pmid:35434680
- 27. Jaspars S, Adan GM, Majid N. Food and power in Somalia: business as usual? A scoping study on the political economy of food following shifts in food assistance and in governance. London: London School of Economics and Political Science Conflict Research Programme; 2020. Available: http://eprints.lse.ac.uk/103138/4/Food_and_Power_in_Somalia_business_as_usual_FINAL21Jan2020.pdf.
- 28. Guha-Sapir D, Panhuis WG. Conflict-related mortality: an analysis of 37 datasets. Disasters. 2004/12/01 ed. 2004;28: 418–28. pmid:15569382
- 29. Guha-Sapir D, van Panhuis WG, Degomme O, Teran V. Civil conflicts in four african countries: a five-year review of trends in nutrition and mortality. Epidemiol Rev. 2005;27: 67–77. Available: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Citation&list_uids=15958428. pmid:15958428
- 30. DuBois M, Harvey P, Taylor G. Rapid Real-Time Review: DFID Somalia Drought Response. London: Humanitarian Outcomes; 2018. Available: https://assets.publishing.service.gov.uk/media/5a8566e4e5274a2e87dc3e43/DFID_Somalia_2017_IRF_Real_Time_Review_FINAL.pdf.
- 31. Maxwell D, Khalif A, Hailey P, Checchi F. Viewpoint: Determining famine: Multi-dimensional analysis for the twenty-first century. FOOD POLICY. 2020;92.
- 32. Famine Early Warning Systems Network. Key Message Update: Food aid urgently needed to mitigate loss of life and avert Risk of Famine (IPC Phase 5). May 2022 [cited 26 Aug 2022]. Available: https://fews.net/east-africa/somalia.