Search strategy

Our protocol was registered on PROSPERO (PROSPERO 2017:CRD42017065650) on 11^th May 2017 (S1 File). We conducted a systematic literature search in Global Health, Embase and MEDLINE In-Process and Non-indexed citations databases between June and July 2017 with the last search on 19^th July. However, search alerts were turned on and new articles screened for eligibility until 28^th August, Search strategy was formulated using search terms and subject headings for 1) severe acute malnutrition, 2) child/infant, 3) therapeutic feeding intervention, 4) synonyms for the outcome parameter “follow-up”; using Medical Subject Headings and free text terms. No date limits were applied. Detailed information on the combinations of search terms used in our search strategy is shown in S2 File.

Study selection

Inclusion criteria were studies in which children aged 6–59 months in a Low- or Middle-Income Country experienced an episode of SAM, received an inpatient or outpatient therapeutic feeding intervention for the treatment of SAM, were subsequently discharged as cured, and who presented any clearly defined outcome between 6 and 24 months of follow-up after discharge. Definitions of SAM included in this review were <-3 WHZ or <70% weight-for-height (WFH) using the 1977 National Centre of Health Statistics (NCHS) reference median or the 2006 WHO Growth Standards, MUAC< 115mm, MUAC <110mm, bipedal-oedema or clinical diagnosis of marasmus, kwashiorkor or marasmic-kwashiorkor [23, 24]. Discharge criteria for cured was as per study definition.

Exclusion criteria were unpublished studies or grey literature, reviews, non-human studies, studies focusing on management of SAM in infants <6 months, studies in high-income countries and studies not published in English language or published in abstract form only.

Screening process

Titles and abstracts were screened against a screening form by the lead author only. The full text was then evaluated for complete suitability and included in review if it qualified. Data extraction forms were piloted then used on all studies to collect information on study details and outcomes. The standardized data extraction form is available in S3 File. Risk of bias was assessed at the study level using ‘Methods for the development of NICE public health guidance’ [25, 26].

Due to the heterogeneous nature of study designs, participants and outcomes, it was not meaningful to synthesize the results in a meta-analysis. The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guideline was followed, except for the items relating to meta-analysis (S1 Checklist).

Loss-to-follow-up

Loss-to-follow-up ranged from 0.0% to 45.1%. Aprameya et al. identified that distance from nutritional rehabilitation unit (NRU), travelling and food expenses, and loss of daily wages reduced compliance with follow-up visits [28]. Two studies found differences in baseline characteristics of children lost-to-follow-up. Losses and intermittent follow-up were more common in children who had been inpatients (p = 0.003) in Khanum et al. [33], and children lost-to-follow-up were more likely to be admitted with a smaller MUAC in Somasse et al. [32].

Mortality

Mortality was calculated solely for the period of 6–24 months after discharge from therapeutic feeding programmes, despite many studies reporting a combined total of in-programme and after-discharge mortality. The highest number of post-discharge deaths reported was 10.4% (49/471) in Kerac et al [27]. The lowest mortality reported was 0.6% (10/1659) by Burza et al. [29]; but when we consider only those who were traced successfully, this increases to 7.9% (10/1264), similar to the 8.5% (7/82) seen in Somasse et al. [32].

Post-discharge mortality of treatment programmes in Asia was generally lower compared to those operating in Africa. CMAM programmes reported lower mortality compared to inpatient programmes operating in the same continent. However, the number of studies is too small to decipher if this is significant.

Relapse

There were multiple discrepancies between the studies’ definition of relapse, and the timing of its measurement. Some studies reported relapse as a point prevalence, based on a single data collection point, whilst others reported incidence based on multiple data collection points. In Khanum et al., 0.6% (3/437) re-met the admission criteria at 1-year post-discharge. In Khare et al. 6.3% (2/32) had re-experienced SAM during the 15 months, but none of the children were experiencing it at the time of follow-up. In Aprameya et al. children were defined as recovered (defined as WHZ ≥1), or not recovered (WHZ ≥-3 but <-1) at 6 months. 24.1% (22/91) were classified as not recovered, and there were more children in this group who had SAM due to an underlying systemic illness than from poor nutrition (p = 0.023) [28]. 41.7% (38/91) of the children were defined as having recovered. At a single measurement 15 months after discharge in Bahwere et al. 26/1783 (1.5%) were malnourished (WFH <80% reference median and/or bilateral pitting oedema). Although not included in their definition of relapse, one additional child had a MUAC <110mm, and five had MUAC <125mm. There were stark differences in relapse rates according to HIV serostatus; 14.3% (4/28) in HIV+ compared to 2.0% (22/1094) in HIV-, in those children with known outcomes. 2.7% (45/1659) of all children discharged as cured met the criteria for SAM admission (MUAC<115mm and/or bipedal oedema) at a single measurement either 3, 6, 9, 12, or 18 months (average 9 months) post-discharge in Burza et al. [29]. As was done with mortality, when it is calculated for only children with known outcomes this increases to 3.6% (45/1264). When also including the case definition for MAM, 27.5% (347/1264) relapsed to SAM or MAM at an average follow-up time of 9 months.

Morbidity

In Aprameya et al. 42.9% (39/91) and 15% (14/91) of children post-discharge had recurrent respiratory or diarrhoea illnesses during follow-up (28). 43.8% (14/32) in Khare et al. had some form of infection, 100% (32/32) were anaemic and 78.1% (25/32) had helminthiasis [31]. 402/437 (92%) in Khanum et al. experienced at least one episode of diarrhoea over the course of the 12-month follow-up, but the mean was seven episodes (range = 0–30) [33]. 292/437 (67%) attended an outpatient health centre with diarrhoea (105/437 (24%) at least three times), 319/437 (73%) with fever (114/437 (26%) at least three times) and 254/437 (58%) with pneumonia (44/437 (10%) at least three times). 420/437 (96%) reported a cough with fever; although this was less frequent in the domiciliary care group (p = 0.03). However, the domiciliary care group had an older age at admission (29 months vs. 25 and 26 months for inpatient and day-care groups, respectively) which could be a factor in this.

Anthropometry

Kerac et al. found that from discharge to 12-months children underwent large catch-up in WHZ [27]. A mean improvement in WHZ of 1.92 (95% CI: 1.76, 2.08) brought them to a mean value of 0.04 WHZ at 12 months post-discharge; comparable to sibling controls and the global norm. WAZ also improved by 1.66 (1.50, 1.82) giving a mean value of -1.77. However, this was below that of sibling controls by -0.55 WAZ (-0.71, -0.38; p<0.01). This could be explained by HAZ, as there was only a small mean improvement of 0.37 (0.21, 0.53, p<0.01) to reach a final value of -2.97 (1.3 SD), below that of the sibling controls by -1.13 (-1.34, -0.93; p<0.01). Burza et al. noted similarly small levels of improvement in HAZ: a 0.7 (0.5, 0.9, p<0.001) increase from discharge to 12-months [29].

Khanum et al. also observed increases in mean WHZ and WAZ from discharge to 12-months of -1.6 to -0.9 and -3.70 to -3.08, respectively [33]. Mean HAZ did not improve over follow-up and remained at -4.14; classified as severe stunting. No associations were found between time with diarrhoea, fever or cough and levels of weight or height gain.

In Aprameya et al. there was an increase in mean weight from 9.2kg (2.0 SD) at discharge to 10.3kg (2.2) at 12-months for children without underlying illness (p<0.001) [28]. The same increase was observed for children with underlying illnesses from 8.1kg (2.4) to 8.9kg (2.6) (p<0.001), albeit lower at each time point. The difference in weight gain between groups was also significant (p<0.001).

Mortality

Routine SAM programme monitoring would have missed the post-discharge mortality that we observed. It is important to consider that the mortality reported here was calculated solely for the period of 6–24 months post-discharge (depending on the length of study follow-up) following nutritional cure, and does not include in-programme mortality. Mortality even among this group of “healthy survivors” is thus striking and is likely to reflect the fact that some children who present with SAM have pre-existing vulnerabilities, whereby SAM is manifested as a symptom and/or sign of those vulnerabilities, rather than due to poor dietary quality and/or quantity alone. Therapeutic feeding is thus sometime analogous to treating infection-related fever: an important part of the “package” of care but not necessarily addressing the root issue. Aside from infection (which is already treated in existing therapeutic feeding programmes) many other possible causes of poor growth are recognised which can be difficult to identify and address even in resource-rich settings, for example biological causes such as malabsorption, metabolic disorders and disability, and social causes such as neglect (either deliberate or, more often, due to complex family circumstances such as illness and poverty) [34, 35].

The highest post-treatment mortality reported by Kerac et al. of 10.4% is above even the in-programme SPHERE standards of <10% mortality [27]. This reflects the initial high-risk cohort: children were all originally admitted to inpatient care in 2006–2007, when CMAM services were unavailable locally and hence cases tended towards the ‘complicated’ end of the SAM spectrum. Malawi also has a high background child mortality as well as a high prevalence of HIV (43.5% (445/1024) of the original cohort of admissions were HIV seropositive) [36]. Another study in Malawi which followed-up children for four months after discharge found that HIV seropositive children were nearly three times more likely to die during nutritional rehabilitation; a 35.4% mortality risk compared to 10.4% (p<0.001) [37].

Relapse

The current lack of a standard definition for post-discharge relapse was evident in these studies. The inconsistency in the definition and measurement of relapse means it is largely not possible to compare differences in, and burden of relapse across the different contexts. Although relapse reported in studies included in this review range from 0% to 6%, many other studies with shorter follow-up periods cite much higher figures. A study in Ethiopia found that 14 weeks after admission into an OTP, 34.6% had relapsed to SAM and a further 37.5% had MAM [17]. Relapse to SAM was 11.1% measured at monthly follow-up visits over the course of 6 months in the Democratic Republic of Congo, and this increased to 44.2% when relapse to MAM is included [38].

Morbidity

High levels of morbidity, in particular diarrhoeal and respiratory illnesses, were reported in all three studies that measured morbidity [28, 31, 33]. A study in Ethiopia that followed-up children six months post-discharge found that there was approximately a 1.7 times higher incidence of fever, diarrhoea and cough in children discharged as cured compared to controls [39]. The synergistic effect of malnutrition and infection on risk of mortality is well-documented, and this review furthermore highlights the potential need for additional care for recovered malnourished children in order to prevent infection, as vulnerability to infection is still present despite recovery in WHZ [40].

Anthropometry

Both included studies that measured WHZ at 6–24 months showed that children were no longer wasted [27, 33]. This has been observed by multiple other studies, including a Bangladesh follow-up that found WHZ increased from -2.5 (0.6 SD) at discharge to -1.9 (1.0) at 6 months [21, 41, 42]. A study from India that followed-up children between one and three, and five and seven years after an episode of SAM found that their mean WHZ was higher than their siblings who had not experienced malnutrition [21]. Whilst it is reassuring that complete or near-complete catch-up in WHZ is possible, there are potentially important life-course implications for these children as suggested by the accumulation of evidence on the Developmental Origins of Health and Disease (DOHAD) highlighting associations between rapid weight gain in infancy and later risk of non-communicable diseases (NCDs) [43].

Overall good recovery of WHZ needs to be seen in context of less impressive changes in other anthropometric indicators. All three studies included in this review that measured HAZ found there were, at best, limited improvements, which is consistent with the wider literature [27, 29, 33, 42, 44–46]. The lack of catch-up in HAZ is a significant concern and of critical importance as severely stunted children are 5.5 times more likely to die, and furthermore are 12.3 times more likely to die when any degree of stunting and wasting occur concurrently [47, 48]. Long term adverse health associated with stunting has been well described and includes an increased risk of suboptimal cognitive and physical development and risk of adult NCDs [49]. A recent Technical Briefing Paper on the relationship between wasting and stunting has helped the international community begin to recognise the gap in both knowledge and programming between these two forms of malnutrition, but this remains a problem to be addressed [50].

Limitations

The use of different admission criteria (WHO versus NCHS; WHZ versus MUAC), which do not overlap perfectly, identifying similar but not the same children, has likely resulted in the subsequent study populations having variable initial risks of mortality and morbidity [51–53]. This issue was compounded by the use of different discharge criteria which resulted in distinct groups of children classified as recovered, making it hard to differentiate between post-discharge mortality and morbidity in recovered children and those who never achieved nutritional rehabilitation (had they been in a different programme).

Relatively high losses to follow-up also limit what can be learnt from our review. Of children not found at 6–24 months post-treatment, some will have died, some will be alive but unwell or still malnourished, and some will be alive and well [54]. What percentage are in each category is likely to vary programme to programme, but what is almost certain is that known mortality and morbidity are underestimates of the true value.

There is a significant risk of bias across all the studies in the review. Firstly, survivor bias will have likely affected all the studies and is critical to their interpretation. Only the healthiest children of those originally admitted for SAM would have made it to nutritional cure in the first place. It is therefore particularly striking that some of these are still at risk of mortality and morbidity over six months post-nutritional cure.

Kerac et al. was the only study which employed a control group to assess relative risk of observed outcomes: siblings living in the same household [27]. However, although 90% of siblings had never experienced SAM, they may have exhibited a degree of undernourishment themselves, which could reduce the effect size. In addition, HIV status was not known or reported on in all studies; despite being an important risk factor for poor outcomes in African studies [55]. In the absence of control groups in all studies except for Kerac et al., we cannot estimate the relative risk of our outcomes and the attributable risk of these related to a previous SAM episode. Despite being unable to say whether there is an excess of e.g. mortality and stunting among children after an episode of SAM, we feel the observed high rates give some cause for concern.

We must also be aware of other variables that influence the extent to which the study results are generalisable. For example, the infectious disease burden, health facility coverage, and food security status of a local population are factors that may influence the outcomes considered in this review, and therefore necessitate context-specific interpretation. Such factors need to be measured and analysed in future studies to better assess the associations between being treated for SAM and the observed outcomes in these cohort studies.

The final constraint when it comes to critical appraisal of the studies in this review is the lack of targets or standard literature to assess whether and to what extent observed mortality and morbidity represents an excess over that in the general population. All settings in which SAM is a public health problem are by definition high-risk. Without more background information on reference population mortality and morbidity, it is hard to quantify the excess risks in our post-SAM children. More context-specific information would be helpful to assess the additional level of support required for SAM survivors.

There are several methodological limitations of the review which warrant mention. Restricting to English language, published papers in the search strategy means not all relevant research may have been identified, and omission of important studies remains a possibility. Selection or reviewer bias could also have occurred given studies were not screened or abstracted in duplicate.

Programmatic recommendations

Our results highlight the urgent requirement of a uniform definition of relapse. Once defined a simple method to enable programs to assess relapse should be devised, with consideration to the limited capacity of most programmes to collect data. Providing mothers with MUAC tapes could facilitate collection of such data. This would be in line with the current field of thinking, as in 2017 The Council of Research and Technical Advice on Acute Malnutrition (CORTASAM) recommended that MUAC should be used as a primary admission criterion for SAM programmes [56]. Mobile phones within households or communities could be a useful tool for reporting relapse and mortality.

In order to ascertain a more complete evaluation of CMAM programmes, mortality and relapse 6 months post-discharge should be initiated as critical indicators of performance. In particular, post-treatment relapse rates and causal factors have been identified as a priority research area by the No Wasted Lives Coalition [57]. Conducting secondary analysis on existing high-quality programme data could aid the creation of a validated tool for use at discharge that systematically identifies children at highest risk of mortality and relapse. Routine referral to post-discharge nutrition-specific and nutrition-sensitive interventions such as the key Lancet evidence-based interventions, could be put in place for the most vulnerable children [58].

Research recommendations

The dearth of literature in this area must be addressed, therefore conducting prospective cohort studies in a broader range of geographic and endemic disease settings in children post-SAM should be made a priority, alongside intervention studies exploring how to improve post-SAM outcomes and identify high-risk children. New studies should aim to provide a more comprehensive description of the problem, with particular focus on mortality and stunting. Mortality, stunting and relapse to SAM should be compared with sibling or community controls, alongside anthropometry, to give a more complete picture of the vulnerability of post-SAM children, and allow us to assess the attributable risk of these adverse outcomes to a previous episode of SAM. Furthermore, improved understanding of the risk factors that lead to relapse and excess mortality is needed. This can be achieved via case-control studies comparing relapsed with non-relapsed treated SAM children.

Randomised control trials are needed to determine interventions that can prevent relapse. One possibility is unconditional cash transfers, which were found to decrease the proportion of relapse to SAM following treatment in a recent study [38]. Determining whether this approach works in different settings and prevents relapse in the long-term should be made a priority. Another approach that needs urgent consideration is whether to treat SAM children with antibiotics after discharge from treatment. A recently published study suggests that treatment with antibiotics can reduce mortality in children even in the absence of malnutrition in different settings [59]. Children treated for SAM should be included in intervention studies that aim to improve HAZ through a variety of methods, including during adolescence [60]. Qualitative data should be collected to investigate and understand the role of the home environment and child feeding practices in children who relapse or die.

The increased risk of prevalence of chronic disease in children treated for SAM warrants further investigation. There is clearly a need to balance the short-term benefits of rapid weight gain with the possible long-term adverse consequences. Randomised control trials are required to explore the trade-off between slow and rapid weight gain. If the evidence suggests rapid weight gain is problematic, different treatment protocols should be tested with the aim of minimising future adverse consequences. A RUTF with a lower sugar content should be trialled in the light of the growing body of literature suggesting that high sugar foods can predispose to obesity [61].

As many studies in this review were excluded because they admitted on WAZ (See S3 File), admission criteria for future studies should be based on WHO definition of SAM, and discharge criteria on the WHO Protocol in order to standardise results. If studies choose to admit on WHZ, WAZ and HAZ, disaggregated data should be presented according to type of undernutrition, alongside age and sex.

As this review suggests, meeting anthropometric criteria for recovery may not equate to complete recovery and return to baseline risk of mortality and morbidity. The mortality and morbidity reported in studies suggests that children affected by an episode of SAM take longer to recover immunologically and physiologically than their weight gain would suggest and remain susceptible to problems like severe infection even after successful hospital discharge. Whether this is true for all ex-SAM children of a specific subset is important to determine in future work. Persisting vulnerability even after apparent nutritional cure may be explained by SAM being a symptom of other underlying problems (such as disability, malabsorption, immune dysfunction) rather than a random exposure, and calls for increased understanding of the aetiology. In order to increase our future understanding of SAM, future studies should report on a wider range of outcomes alongside mortality and relapse rates, such as body composition, cognition, behaviour, educational attainment, metabolism, immunity, and micronutrient deficiencies. This may enable us to broaden or re-define our definition of recovery in the future to one that, for example, could include immune and body composition parameters. This will allow us to determine when a child has truly returned to healthy physiological state, and also allows identification and further input to children who remain vulnerable to mortality and morbidity, despite being deemed recovered based on anthropometric criteria.