Association of postnatal severe acute malnutrition with pancreatic exocrine and endocrine function in children and adults: a systematic review

Severe acute malnutrition may lead both concurrently and subsequently to malabsorption and impaired glucose metabolism from pancreatic dysfunction. We conducted a systematic review to investigate the associations of current and prior postnatal wasting malnutrition with pancreatic endocrine and exocrine functions in humans. We searched PubMed, Google Scholar, Web of Science and reference lists of retrieved articles, limited to articles in English published before 1 February 2022. We included sixty-eight articles, mostly cross-sectional or cohort studies from twenty-nine countries including 592 530 participants, of which 325 998 were from a single study. Many were small clinical studies from decades ago and rated poor quality. Exocrine pancreas function, indicated by duodenal fluid or serum enzymes, or faecal elastase, was generally impaired in malnutrition. Insulin production was usually low in malnourished children and adults. Glucose disappearance during oral and intravenous glucose tolerance tests was variable. Upon treatment of malnutrition, most abnormalities improved but frequently not to control levels. Famine survivors studied decades later showed ongoing impaired glucose tolerance with some evidence of sex differences. The similar findings from anorexia nervosa, famine survivors and poverty- or infection-associated malnutrition in low- and middle-income countries (LMIC) lend credence to results being due to malnutrition itself. Research using large, well-documented cohorts and considering sexes separately, is needed to improve prevention and treatment of exocrine and endocrine pancreas abnormalities in LMIC with a high burden of malnutrition and diabetes.

Wasting malnutrition remains common both for children in low-and middle-income countries (LMIC) and for adults with severe infections, notably HIV or tuberculosis. As treatments for severe acute malnutrition improve (1) and drugs become increasingly available and effective for severe infections, more people survive but the long-term consequences of their malnutrition are not fully understood (2,3) . Acute nutritional deficits during the prenatal period affect the structure and function of organs such as the pancreas which have fundamental roles in metabolism (2,4) . While there is information from animal models and from human studies of prenatal malnutrition, usually indicated by a proxy of low birth weight (5) , the consequences of postnatal undernutrition on human pancreatic structure and function and later chronic disease development are not well documented.
Both pancreatic endocrine (i.e. production of hormones such as insulin or glucagon) and exocrine (i.e. production of enzymes to aid digestion and subsequent nutrient absorption) functions are critical for nutritional metabolism and chronic diseases including diabetes. A previous systematic review of effects of severe acute malnutrition on pancreatic exocrine function in children concluded that there was evidence of association but could not determine causality (6) . Diabetes mellitus (DM) is one of the most common non-communicable diseases worldwide and is rapidly increasing, particularly in LMIC (7) .While it is established that overweight and obesity in adult life increase the risk of type 2 DM (8) , the contribution of prior malnutrition to the aetiology of DM and its potential interaction with later overweight across the global context remains unclear.
In 1965, a WHO Expert Committee reported that 'the evidence that undernutrition protects adult populations from diabetes seems unassailable' (9) . In 1980, they reported that 'in some societies, malnutrition is probably a major determinant of diabetes' (10) . In 1985, malnutrition-related DM was included as a classification category of DM divided into two subtypes, protein deficiency pancreatic diabetes and fibrocalculous pancreatic diabetes, both commonly reported in tropical countries and usually associated with a history of undernutrition (11) . This classification has since been dropped and the literature is inconsistent in the terms and diagnostic criteria used for these atypical forms of diabetes. As well as the above classifications, other commonly used terms have included 'tropical diabetes', 'malnutrition-associated diabetes' and 'African diabetes'. A recent systematic review concluded that, based on currently limited data, two main phenotypes of atypical diabetes emerge, differing in usual age of onset and in the requirement for lifelong insulin but both occurring in younger ages than is typical for type 2 DM and in underweight individuals or normal weight/modestly overweight individuals; both phenotypes have some features similar to type 1 DM (12) . Previous reviews have assessed famine, or malnutrition in a particular age group, and either exocrine function or diabetes as an outcome of endocrine dysfunction but not detailed markers of glucose metabolism (6,13) . The present study includes detailed glucose metabolism markers and diabetes as well as exocrine pancreas functions in relation to the less studied, postnatal period of exposure to acute malnutrition, not limited to the postnatal period but including childhood and adulthood and from infection-associated malnutrition. Excluding this, most common type of malnutrition exposure could lead to underestimating the impact on populations which might underlie the large increase in diabetes in populations still experiencing a high burden of infectious diseases. Finally, the decision to include anorexia nervosa (AN) as another exposure was to allow the comparison with malnutrition in which diet restriction, rather than infection, has the main causal role.
This systematic review aims to describe the available evidence to determine if severe acute postnatal malnutrition causes persisting changes in pancreatic endocrine and exocrine function and later increased risk of DM.

Search strategy
An electronic literature search was performed on PubMed, Web of Science and Google Scholar to identify studies published in English from the earliest available date to 1 February 2022. Detailed search terms are shown in Supplementary Data 1. Studies were eligible for inclusion if they reported human pancreatic function in relation to exposure to postnatal malnutrition identified through clinical or anthropometric methods in hospitals, clinics or communities, famine or eating disorders. Studies were excluded if written in languages other than English, if the full text was unavailable, if study participants had no prior or current malnutrition exposure or only prenatal malnutrition exposure or if stunting (chronic malnutrition) without wasting (acute malnutrition) was the exposure. We included cross-sectional studies to investigate acute associations between malnutrition and pancreas function and trials, cohort studies or retrospective case-control studies to investigate longer term outcomes. Case series and case reports and studies with ≤ 10 participants were excluded from the review owing to the high potential for bias. Studies describing abnormal pancreatic function, for example, due to cystic fibrosis, leading to malnutrition were also excluded. Cancer studies were excluded since, although many cancers may result in malnutrition, the added metabolic complications of cancers and their treatment would make it difficult to determine the effects of malnutrition itself. However, we did include studies of malnutrition secondary to serious infections such as HIV or tuberculosis with the rationale that infections are virtually always part of severe malnutrition, including classical malnutrition in young children.
Duplicates were identified and removed, and the titles and abstracts were reviewed to determine possible eligibility by a single reviewer (FF). Additional studies were identified by manually searching the reference lists of included papers and previous reviews or meta-analyses. The full texts of the relevant articles were obtained and independently reviewed for final selection according to the eligibility criteria by at least two of the authors. Any differences in judgment were discussed with all authors to reach consensus.

Quality assessment
A quality assessment checklist was developed based upon the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) (14) and Consolidating Reporting of Clinical Trials (CONSORT) checklists (15) . The quality assessment checklist comprised sixteen items for cross-sectional studies, nineteen items for cohort studies, twenty items for case-control studies and twenty-two items for clinical trials where scores for each question item ranged from 0 to 2. Scores were assigned as follows: '0' for no information/unlikely or not reported/poor or inappropriate description; '1' for partially or possibly reported/ satisfactory and '2' meaningfully reported/good. Using these checklists, three authors (FF, SC, and NBS) independently evaluated the included articles. Scores and decisions were then discussed before assigning an overall quality rating of each study with each article's quality rated as 'high' if score ≥ 80 %; 'medium' if score 60-79 %; 'low' if score < 45-59 % or very low if score < 45 %.

Data extraction
Details of the included studies were extracted into an Excel file under the following headings: first author, year of publication, study design, country, quality category, age when participants were malnourished, number of participants, participant inclusion/exclusion criteria, diagnostic criteria or definition of malnutrition used, pancreatic outcomes assessed, a description of any interventions or treatment/nutritional rehabilitation received, the type, timing and frequency of pancreatic function outcome assessments conducted and the main or most salient results plus a column for further comments.

Articles included and excluded
After the removal of duplicates, 8108 articles were identified. After screening of abstracts, full texts of 259 papers were obtained (Fig. 1). An additional thirty articles meeting the study inclusion criteria were obtained from a manual screening of article reference lists. Of the 289 articles, 221 articles were excluded due to not meeting the inclusion criteria, leaving sixty-eight articles. The list of excluded studies with reasons is provided in Supplementary Data 2.

Definition of the malnutrition exposure
The definition of childhood severe malnutrition has varied over the publication dates of included studies from clinical definitions of kwashiorkor and marasmus, comparison with various different child growth standards, and more recently, the current WHO definition of severe acute malnutrition based on the 2005 growth standards or oedema (84) . While there may be minor differences between definitions, we believe these likely reflect broadly the same clinical conditions across time and have therefore considered the definitions of childhood clinical malnutrition together. The more important difference for effects on the pancreas appears to be whether or not the severe acute malnutrition involved oedema, that is, kwashiorkor. Adult malnutrition in the included papers has generally been caused by low BMI and/or a clinical diagnosis of AN. For the studies of long-term consequences of childhood famine exposure, there were no assessments of malnutrition at the time so exposure was defined by date and place of residence at the time.

Definitions of exocrine and endocrine pancreas outcomes
Tests used in the included studies of exocrine pancreas function fit into two main groups. Some earlier work measured enzymes in duodenal juice collected with a catheter, both basally and after stimulation with secretin or cholecystokinin, whereas recent studies were less invasive and collected only faeces or serum; this makes it hard to compare results. Studies using catheters and collecting duodenal fluid before and after simulation measured various enzymes, including amylase, lipase, phospholipase, trypsin and chymotrypsin, as well as bicarbonate and electrolytes (19,(22)(23)(24)(25)(26)28,48,49) . Recent studies assessed faecal elastase, which is low in pancreatic insufficiency (16,17,27,50) , or blood levels of enzymes such as trypsinogen, amylase or lipase which may be either high or low in pancreatic disorders (18,20,21) . One study measured pancreas head size in children using ultrasound (21) and another measured D-xylose and triglyceride absorption (50) .

Exocrine pancreas function during and after malnutrition in children and adults
Most papers addressing exocrine pancreas function were from hospital-based studies of acute malnutrition in young children, at admission, during treatment or at hospital discharge; two papers (24,28) also included some children longer after discharge (Table 1). Among the seven studies which measured duodenal enzymes, most both before and after pancreas stimulation (19,(22)(23)(24)(25)(26)28) , all but one (22) found decreases in several enzymes (amylase, lipase, trypsin, and chymotrypsin). Short-term nutritional therapy in hospital usually increased enzyme levels but not always to levels of well-nourished controls. Discrepancies among studies may be due to small sample sizes, differences in patient populations and the duration and type of nutritional or other therapy.
Three studies (16,17,27) , two from the same research group, used faecal elastase as the exocrine pancreas marker. The study in Indonesia (27) , which was concerned mainly with diarrhoea and did not clearly define malnutrition, found no differences in elastase between malnourished children at hospital admission and well-nourished children. The studies in Malawi (16,17) included no well-nourished children but investigated changes over time during therapy among malnourished children; children with oedematous malnutrition had lower faecal elastase than those with non-oedematous and generally did not recover within a month of nutritional therapy. In all these studies, there is a concern that diarrhoea, which is common among children hospitalised with malnutrition, could interfere with the use of faecal elastase as a marker of pancreas function.
Four studies, all with fairly large sample sizes, measured pancreatic enzymes in blood or serum. (17,18,20,21) . Circulating trypsinogen was higher in children with oedematous compared with non-oedematous malnutrition (17) , and in malnourished children than in healthy controls (20) . In one study, serum trypsinogen was not associated with malnutrition but was increased in children with gastroenteritis (18) . Serum amylase and lipase, as well as pancreas head size, were low in malnourished children at hospital admission compared with healthy controls and increased during treatment (21) .
There is limited information about exocrine pancreas function during or after malnutrition occurring in adulthood ( Table 2). There was no difference in faecal elastase among ten AN patients before and after nutritional recovery; this study also found normal D-xylose and triglyceride absorption (50) . Two papers from the same group in India studied possibly overlapping small numbers of malnourished adults and controls (48,49) . Before nutritional therapy, malnourished adults had steatorrhoea and low duodenal juice contents of trypsin and lipase basally and post-stimulation with cholecystokinin (pancreozymin) and secretin; amylase content was low only post-stimulation. After nutritional therapy, lipase and amylase differed little from values in healthy controls both before and after stimulation while trypsin remained low.

Endocrine pancreas function
Studies in young children. This group comprises the bulk of the included papers but only one was scored as medium quality (39) and the rest were rated as low or very low quality. Fasting blood or plasma glucose (not always distinguished in the papers and referred to here as FBG), was generally unaffected by acute malnutrition (33)(34)(35)37,38,47) but decreased FBG was also reported (29,36,46) . HbA1c was higher in children with kwashiorkor or marasmus compared with controls and FBG and HbA1c were inversely associated (29) . Altered fluid balance, especially among children with oedematous malnutrition, and the more acute time frame represented by FBG than by HbA1c may have contributed to an apparently anomalous negative correlation between FBG and HbA1c.
Slow glucose disappearance rates during OGTT or IVGTT were commonly seen at hospital admission in malnourished children, especially those with kwashiorkor (32)(33)(34)37,41) but not always in those with marasmus (33) . One study (32) suggested that lack of insulin was not the reason for slow glucose disappearance since insulin infusion in four children with kwashiorkor did not increase the glucose disappearance rate. Another study (33) found that glucose disappearance rate in kwashiorkor patients was negatively correlated with blood fatty acids which suggest a role for insulin resistance. However, a small study which used stable isotopes to investigate glucose metabolism found no evidence for hepatic or peripheral insulin resistance but did note that glucose clearance was positively correlated with plasma albumin (39) . Nutritional therapy generally normalised glucose disappearance rate (33,37,41) although one study found this was still slow in children 6-12 years post-hospitalisation for kwashiorkor (42) .
Fasting plasma insulin in different studies of kwashiorkor or marasmus was variable, likely reflecting variable populations and small sample sizes (33,35,38) . Insulin release during an OGTT or IVGTT, measured either as plasma levels or in relation to glucose, was generally low for children admitted to hospital mainly with kwashiorkor but sometimes also with non-oedematous malnutrition (30,31,34,(39)(40)(41) , although one study found high peak insulin during an OGTT in malnourished children (35) . Insulin levels increased in the short term with nutritional therapy but were not always restored to normal even months after admission (30,33,34,37,38) , although the overall patterns of response normalised (31) . Interestingly, in a series of studies by the same research group with overlapping patient populations (30,31,40) , insulin/glucose AUC was more reduced and patterns of response more abnormal compared with controls during an OGTT than in an IVGTT, and a pattern of delayed insulin secretion in malnutrition was common; both these could indicate that part of the impaired insulin response was due to factors in the intestine. One study measured plasma glucagon and found low fasting levels in children admitted to hospital for malnutrition (38) .
Five studies investigated endocrine pancreas function using OGTT or IVGTT in participants several years after they were hospitalised for malnutrition in early childhood (30,(42)(43)(44)(45) . Ten years post-kwashiorkor, there was no difference compared with sibling controls in peak insulin or insulin/glucose AUC in an IVGTT (30) . There were two small studies from the same malnutrition unit in Uganda of children about 10 years after they recovered from kwashiorkor. One found a slower glucose disposal rate during an IVGTT in recovered malnourished children compared with controls (42) . The other study found lower fasting insulin in recovered kwashiorkor patients but normal insulin response in an OGTT and with a glucagon stimulus to elicit maximal insulin release (43) . More recent studies with larger sample sizes found larger differences in glucose metabolism of malnutrition survivors compared with controls. Jamaican adults who had experienced marasmus in early childhood had lower insulin secretion and poorer glucose tolerance in an OGTT compared with kwashiorkor survivors or not previously malnourished controls (44) . Among young adult Mexican male survivors of childhood malnutrition, plasma glucose concentration and AUC in an OGTT were higher than in not previously malnourished controls only after controlling for BMI, age and birth weight, whereas plasma insulin was higher both with and without controlling for these variables and the difference between cases and controls was greater in those with higher BMI (45) . A recent study in 1080 Chilean adults found higher glycaemia at age 22-28 years in those who were wasted or at risk of wasting at 12 months (WLZ score < -2 or < -1), including after adjustment for confounders including birth weight and gestational age. Those who were underweight (WAZ-score < -2) at 12 months had evidence of increased glycaemia in unadjusted but not adjusted analysis but increased insulin sensitivity when assessed using single point insulin sensitivity estimate but not homeostatic model assessment-insulin resistance (46) .

Endocrine pancreas function after malnutrition in later life.
Of fifteen papers on endocrine pancreas function after malnutrition experienced in later childhood or adulthood, twelve are about AN patients, only one of which included males (59) , two are about Indian adults and one is about African adults ( Table 2). The quality scores for the majority in this group were classified as low or very low, with the exception of a study in Sweden (59) which utilised a national register with long-term follow-up and was rated as high quality, plus four studies rated as medium quality: one from the USA (58) , a case-control study in malnourished Indian adults (62) , a cohort study in African malnourished adults (65) and a study in Japanese women with AN before and 5 months after treatment (55) . When patients were admitted to hospital with AN, there was a fairly consistent finding of abnormal, that is, low or delayed, insulin production during OGTT, IVGTT, arginine or glucagon infusion or after a meal (51,52,(54)(55)(56)61,64) . Glucose metabolism during these tests was more variable: there was often poor glucose tolerance, as might be expected from the low insulin production (51,52,64) , but others found normal responses (55) . Plasma glucagon was generally not different between AN patients and controls (51,54) although one study found abnormal patterns of glucagon changes during an OGTT (52) , and another found low glucagon after insulin-induced hypoglycaemia but not after arginine infusion (55) . One study found that 24-h urinary excretion of C-peptide did not differ between AN patients and controls (53) . Several studies examined AN patients after weight recovery either just before discharge from care or several years later. Not all studies included controls so it is difficult to determine whether normal pancreas endocrine function was achieved following weight regain. Insulin production and glucose tolerance often improved compared with admission results by the time, usually after several months, AN patients had gained sufficient weight to be discharged (61,64) , but did not always reach normal levels (52,56) . A study of AN patients who had recovered weight, but which provided no information on time since diagnosis, found continued impairments in insulin production but heightened insulin sensitivity resulting in similar glucose responses following a test meal (57) . Insulin production remained low and glucose tolerance impaired 8-10 years after AN diagnosis in those who remained low weight but not in those who had achieved normal weight (58) . The incidence of diabetes diagnoses on a Swedish national register was lower among former AN patients than among the general population but not different from sibling controls; the low incidence among those with prior AN may have been related to their very low incidence of overweight as adults but the study was not designed to control for this (59) .
In a study of malnourished Indian adults, in which malnutrition duration and severity were not well defined and there were no well-nourished controls, insulin increase was slow but prolonged in an IVGTT, glucose disposal rate was low and both glucose and insulin responses to arginine infusion were blunted; all responses improved after 2-4 months of nutritional therapy (63) . A case-control study of tropical pancreatitis in Indian adults found that 13·5 % of the patients had diabetes, based on fasting and postprandial blood glucose levels, and that weight loss appeared a consequence, not a cause, of the impaired pancreas function (62) . A cohort study in African adults reported that malnutrition associated mainly with HIV or tuberculosis infection 7-12 years previously was later associated with lower insulin levels in an OGTT in men but not in women (65) .
Studies in adults exposed to famine in childhood. The famine follow-up studies (Table 3) represent the largest amount of information on long-term outcomes of childhood malnutrition, with by far the largest sample sizes, with generally robust statistics, and when participants were in middle age when diabetes is more common than at younger ages. The famine studies were the majority of studies rated good or medium in the quality assessments. The drawback of the famine studies is that the diagnosis of prior malnutrition is based on date and place of birth so it cannot generally account for local differences in famine exposure or individual or family response to famine, although one study asked participants what they recalled of their famine experience (74,81) . Another concern is that, particularly in the Chinese famine of 1959-1961 which was prolonged and had high mortality, there is likely to be a survivor bias, and this may have had a sex difference, that is, boys may have had higher mortality than girls (72,78,(80)(81)(82)(83) .
Eleven of the eighteen included studies were from China, were done by different research teams and together included data from six representative cohorts (one with two publications from the same group with different purposes (75,76) and another one with two publications by different research groups (72,77) ). Methods were similar in that famine exposure was determined by birth location and date with respect to the 1959-1961 famine. Most studies also examined fetal famine exposure which is not the concern here. There were differences among studies regarding the sex and postnatal age for which famine exposure carried the greatest risk of hyperglycaemia (assessed by fasting blood glucose and/or HbA1c) or diabetes: all ages from early to late childhood (75,76,80,82,83) , all ages but only in women (72,78) , early childhood in women only (79) , infancy only (77) or late childhood only (68,81) . Although both men and women were at increased risk of high fasting plasma glucose or HbA1c, if exposed prenatally, β-cell function, indicated by homeostatic model assessment-β, seemed to be the major problem in men, whereas it was insulin resistance, indicated by homeostatic model assessment-insulin resistance, in women (75) . Women exposed to famine at any stage of childhood had an increased prevalence of hyperglycaemia but not of diabetes, whereas men in the study had no difference in hyperglycaemia but lower prevalence of diabetes (72) . In another study that examined sex differences, this was only observed for risk of composite metabolic syndrome, with similar risks among men and women for hyperglycaemia (82) . There appears to be an interaction between famine exposure and diet or BMI at the time of glucose assessment: being overweight or currently eating a Western style rather than traditional Chinese diet increased the risk of hyperglycaemia or diabetes after childhood famine exposure (68,77,83) . One cohort study investigated the incidence of clinically diagnosed diabetes over about 7 years in middle age (69) . The incidence was increased among those exposed to famine in utero but not those exposed in childhood and was aggravated by adult abdominal obesity; however, about three times as many cases of prevalent diabetes were excluded • OGTT (0 and 120 min glucose) and standard criteria of: • Impaired fasting glucose (IFG) 0 min glucose 5·6 -6·9 mmol/l • Impaired glucose tolerance (IGT): 120 min glucose 7·8-11·0 mmol/l • DM: 120 min glucose ≥ 11·1 mmol/l • More underweight in fetal exposed • More overweight in postnatal exposed • No overall differences in glucose outcomes but interaction with current BMI and 120 min blood glucose: underweight fetal exposed higher than other groups • No differences in glucose outcomes between postnatal exposed and unexposed • Missing many statistical details and use of multiple post-hoc subgroup analyses • Likely survivor bias in those famine-exposed FPG > 6·1 mmol/l and/or 2-h glucose > 7·8 mmol/l • All exposed groups ↑ FPG compared with non-exposed • Late childhood exposed group ↑ hyperglycaemia, and DM v. unexposed, similar in severe and less severe famine. In contrast, fetal exposed group affects were limited to severe famine exposuresignificant interaction tests, • Prevalence of hyperglycaemia increased if exposed cohort consumed an affluent/ western diet/higher • Large sample size and consideration of effect modification by current diet • Area of current residence used as proxy as area of birth • Likely survivor bias because famine-exposed may have died in childhood • Childhood exposed group ↑ DM rate v. unexposed, similar in fetal exposed group • Age and Sex-adjusted relative risk 1·20 times ↑ in childhood famine exposed group v. nonexposed group, but non-significant when further adjusted, and in stratified analyses by non-ideal cardiovascular health metrics, or by sex • 1372 cases of incident DM • After adjustment for age and other confounders, ↑ DM incidence in fetal exposed compared with non-exposed and early childhood exposed combined, no differences by sex. • No evidence of effects in childhood exposed group • Adult obesity and abdominal obesity had additive effects with early MN on diabetes incidence, especially in women, significant interaction   • Adjusted DM risk ↑in childhood exposed compared with nonexposed. Effect appears to be due to effect in women with ↑risk in both childhood and adolescent exposed groups and • Sample representative of E China, stratified for urban/rural and high/low econ status • Indicators of insulin production and glucose  • ↑ Dysglycaemia prevalence in fetally and postnatally exposed groups than in control group • ↑Metabolic syndrome in both famine groups from the study as were found to have incident diabetes so it is difficult to determine the overall effect of famine exposure. The Dutch famine of 1944-1945 has been greatly studied for long-term effects of in utero exposure but less so for postnatal exposure. Famine exposure during adolescence, and to a borderline extent earlier in childhood, increased risk of a later diabetes diagnosis in women but not in men (70) . A study which included only women (74) found an increased risk of later diabetes by self-report or clinical diagnosis if exposed to famine at any time during childhood and the risk was increased if women reported their famine exposure as severe v. moderate.
Famines in Bangladesh, Hong Kong (on Chinese population), Nigeria, Russia and Austria are represented by one paper each. Early childhood famine exposure in Bangladesh did not affect fasting glucose or glucose response in an OGTT (66) . In Nigeria, childhood famine exposure had no effect on random blood glucose; unfortunately, the study's recruitment method precluded use of other tests of glucose metabolism (67) . In Hong Kong, self-reported childhood famine exposure was not associated with DM risk (81) . Childhood famine in Russia was not associated with glucose, insulin, proinsulin or C-peptide in adulthood (71) . Analysis of a large Austrian national database which included people of a wide age range covering fetal or childhood exposure to several 20 th century famines found clear evidence of increased risk of fetal famine exposure but not of childhood exposure (73) . It is unclear why these studies from other countries differ from the general findings of ongoing impaired endocrine pancreas function seen in the Chinese and Dutch famine studies, but differences in famine experience and mortality and in later environment and diet could be important.

Discussion
There is considerably more research on how postnatal malnutrition affects endocrine than exocrine pancreas function. This likely reflects the high prevalence of diabetes globally and its serious health consequences but is in part because there are fewer non-invasive tests of exocrine than endocrine pancreas function available. Some earlier work on exocrine pancreas function used catheters to collect duodenal juice, but such tests are unlikely now to be considered ethically justified for research and the earlier work was of generally poor quality due to low participant numbers, inadequate statistics and consideration of confounding and high risk of selection bias.
Overall, while there are differences among studies of exocrine pancreas function, it seems that secretion of many pancreatic enzymes is reduced in acute childhood malnutrition. The several papers reporting steatorrhoea suggest that this reduced enzyme secretion may have important functional consequences which, through impaired nutrient absorption, could have contributed to the malnutrition in the first place and would very likely exacerbate it. In most cases, nutritional therapy improved enzyme secretion although not always to control levels, possibly because of varying durations and quality of the therapy. One trial which investigated adding enzyme therapy to nutrition (16) found no additional benefits; this was a recent study using current WHO nutritional therapy guidelines which likely provide better nutritional support than was available in earlier studies. Our results are consistent with a previous review (6) which found an association between malnutrition and decreased exocrine pancreas function but could not determine causality.
Regarding endocrine pancreas function, there seem to be prolonged impairments in insulin production among people severely malnourished in childhood or adulthood but these are most profound in people who remain malnourished (46,47) . Adults in LMIC recruited when malnourished may have been so through much of their lives (48,49) and long-term impairments in AN patients are greatest in those who remain malnourished (55,57) . It would be interesting to investigate insulin in people in LMIC of previously good adult nutritional status who first became malnourished in adulthood. However, adult-onset malnutrition often follows serious infections, for example, with HIV or tuberculosis, or cancers so these factors confound the situation. AN remains the most common cause of severe malnutrition resulting mainly from low dietary intake. Since there appear similarities between observations in AN patients, people exposed to famine in childhood (74,78) and adults in LMIC, this suggests that it is malnutrition itself, rather than only the accompanying infections, environmental enteropathy and other aspects of living in poverty, that influence pancreatic insulin production (55,64,65,70,74) . Furthermore, since AN normally occurs in people who were previously adequately nourished in high-income countries, the results from AN patients suggest that the direction of causality is from malnutrition to impaired pancreas function, although the opposite direction of causality, with pancreas disease causing malnutrition, may also contribute (62) .
The mechanisms whereby malnutrition may result in longterm effects on the pancreas are unclear and the studies included in this review provide little information, in part because the more recent and high-quality studies have been mainly large ones investigating the epidemiology of glucose metabolism in famine survivors. Some studies in India have investigated pancreas calcification as the mechanism of the impaired function; such reports contributed to the earlier WHO classification of fibrocalculus pancreatic diabetes (11) but most such studies in the present review were excluded because the main exposure was not prior nutritional status. Environmental enteropathy, which is common in low-income countries and often associated with malnutrition (85) , may have contributed to impaired glucose tolerance in studies of acute malnutrition. Evidence for this comes from studies showing delayed insulin responses or larger effects on the insulin response to OGTT than to IVGTT which bypasses the gut. Possible mechanisms of intestinal epithelial involvement include delayed or reduced glucose absorption (86) during an OGTT and reduced insulin production because of low incretin production by enterocytes (87) .
Most abnormalities in endocrine pancreas function of severely malnourished people seem to improve in the short term with the treatment of the malnutrition; recovery of exocrine pancreas functions after malnutrition has been less studied. There is limited information about long-term pancreas function outside the famine studies. Several of those studies suggest early insults may interact with later diet and illness (68,77) .This is in keeping with the capacity load model of chronic disease (88) in which damage to a physiological capacity, for example, pancreas functions, earlier in life is most likely to result in health problems if in later life there is a greater load on the system, for example, due to overweight or consumption of a diet high in sugar. Differing prevalence of adult obesity in men and women, in addition to potential sex differences in survival from malnutrition, may contribute to the variable sex differences seen in some of the famine follow-up studies.

Strengths and limitations of the review
A strength of the review is that it included a large number of studies from many countries of varying income levels and with multiple study designs and participant characteristics. The overall similarity in results from very different studies, that is, clinical malnutrition in young children, AN in older children and adults and follow-up of famine studies, lends credence to the findings. At least two of the authors reviewed all the included studies. The review has limitations resulting from the heterogeneity of data from varying methodologies, settings and populations enrolled which also precluded being able to conduct any meaningful meta-analyses. Many of the included studies were of poor quality with small sample sizes, poorly defined populations and unclear statistics. We did not analyse the findings of only good quality studies separately since this would have excluded too many that provided data not available elsewhere. Many of the early studies were conducted to define the aetiology and biology of kwashiorkor v. marasmus so we were not addressing our interests specifically. Similarly, much research on AN was not specifically investigating pancreas function. The techniques used in older compared with newer studies differed and were not always validated, so results are hard to compare. Authors of some of the famine studies were interested in prenatal famine exposure, so they included postnatal exposure as a control for that. The absence of an original aim to investigate postnatal malnutrition and pancreas function could have meant that, even if the study contained data relevant to our search, the title and abstract might not have mentioned it so it would have been missed at the first level of the search; this may explain why a large proportion of the included studies were actually located from the reference lists of other articles found. In addition, we did not include studies where the population was selected based on diabetes because this could have resulted in bias in relation to our aims; for example, studies looking at malnutrition as one of many risk factors for diabetes in a population may have included it in the abstract only if the association was statistically significant, for example, Fekadu et al. (89)

Conclusion
Much of the world is currently facing a double burden of underand over-nutrition in which there is an increasing prevalence of overweight, diabetes and other chronic diseases but an ongoing high prevalence of malnutrition, both in young children and in older individuals with severe infections. There is a need for a better understanding of how these conditions interact in order to improve prevention and treatment of chronic conditions. This review suggests that malnutrition at any postnatal age can have both acute and long-term adverse effects on pancreas function so that diabetes treatments should consider insulin production as well as insulin resistance. Currently, the common first-line pharmacological treatment for diabetes in many settings, including low-income ones where detailed metabolic investigations are often not possible, is metformin which acts primarily on insulin resistance; however, it may not be the best treatment in populations where low insulin production is a major concern (90) . The similarity of findings from very different populations, including children living in poor environments, adults with malnutrition secondary to severe infections, AN patients and famine survivors, suggests that it is malnutrition itself which can result in impaired pancreas functions. If infection-mediated malnutrition has life-long impacts on diabetes risk, this provides added impetus to prevent and treat this malnutrition beyond achieving favourable outcomes of the original infection, for example, tuberculosis or HIV. More well-designed research with clearly defined populations, adequate sample sizes, consideration of the sexes separately and using robust current techniques to determine the contribution of low insulin production or increased insulin resistance, is needed in order to understand both the epidemiology and mechanisms of interactions between malnutrition and pancreas functions.

Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114522001404