The effects of malnutrition on child mortality indeveloping countriesD.L. Pelletier,' E.A. Frongillo, Jr,1 D.G. Schroeder,2 & J.-P. Habicht1

Conventional methods of classifying causes of death suggest that about 70% of the deaths of children(aged 0-4 years) worldwide are due to diarrhoeal illness, acute respiratory infection, malaria, andimmunizable diseases. The role of malnutrition in child mortality is not revealed by these conventionalmethods, despite the long-standing recognition of the synergism between malnutrition and infectiousdiseases. This paper describes a recently-developed epidemiological method to estimate the percent-age of child deaths (aged 6-59 months) which could be attributed to the potentiating effects of malnutri-tion in infectious disease. The results from 53 developing countries with nationally representative dataon child weight-for-age indicate that 56% of child deaths were attributable to malnutrition's potentiatingeffects, and 83% of these were attributable to mild-to-moderate as opposed to severe malnutrition. Forindividual countries, malnutrition's total potentiating effects on mortality ranged from 13% to 66%, withat least three-quarters of this arising from mild-to-moderate malnutrition in each case.

These results show that malnutrition has a far more powerful impact on child mortality than is gen-erally appreciated, and suggest that strategies involving only the screening and treatment of the severe-ly malnourished will do little to address this impact. The methodology provided in this paper makes itpossible to estimate the effects of malnutrition on child mortality in any population for which prevalencedata exist.

According to conventional methods of classifyingcauses of death, an estimated 70% of the deaths ofchildren (aged 0-4 years) worldwide are due to diar-rhoeal illness, acute respiratory infection, malariaand immunizable diseases (1). These methods do notidentify malnutrition as a major cause of death indeveloping countries, despite its high prevalence anddespite the long-recognized synergism between mal-nutrition and infection in child mortality (2). Infec-tious diseases represent the immediate and moreobvious cause of death, while the role of malnutri-tion is only readily apparent when it is severeenough to cause clinical manifestations. Severe mal-nutrition is often classified under "nutritional defi-ciencies", and typically accounts for 1-5% of deathsin hospital-based mortality statistics from developingcountries (1). The object of this paper is (i) to esti-mate the percentage of child deaths attributable to

I Division of Nutritional Sciences, Cornell University, Ithaca,New York, USA. Requests for reprints should be sent to Dr D.L.Pelletier, Food and Nutrition Policy Program, Cornell University,3M28, Martha Van Rensselaer Hall, Ithaca, New York 14853,USA.2 Center for International Health, Emory University School ofPublic Health, Atlanta, GA, USA.

Reprint No. 5618

the potentiating effects of malnutrition on disease,using epidemiological (rather than clinical) methodswhich are capable of capturing the effects of mild-to-moderate as well as severe malnutrition, and (ii) todescribe a simple but reliable methodology to makethese estimates for specific countries, communitiesor population groups.

MethodsThe methodology described below builds uponearlier work which provided epidemiological con-firmation that the malnutrition-infection synergismobserved clinically and in biomedical studies doesindeed have a multiplicative impact on mortality atthe population level (3). These findings have sincebeen extended to permit estimation of malnutri-tion's quantitative impact on child mortality inspecific populations. The development and testingof this methodology have been described in detail(4) and are only summarized here. In addition, areview of the literature has been published (5) whichaddresses questions of confounding by the child'sage and socioeconomic factors, intercurrent morbid-ity, variation in length of follow-up, the relativeimportance of weight change versus attained weight,and other factors. This review supports the assump-

Bulletin of the World Health Organization, 1995, 73 (4): 443-448 © World Health Organization 1995 443

D.L. Pelletier et al.

tions made here, to the effect that the relationship be-tween low weight-for-age and mortality is not simplya function of confounding by these factors. Moreover,variation in study designs, methodologies andmeasurement error would be expected to producevariations in the slope of mortality on weight-for-ageacross studies (rather than the homogeneity inslopes described below). Thus, the existence of meth-odological differences across studies cannot explainthe consistency in the malnutrition-mortality relation-ship which is described below.

The comerstone of the methodology is based onthe results of eight community-based, prospectivestudies of the relationship between anthropometryand child mortality in developing countries. Theseeight studies were chosen from a larger number,reviewed elsewhere (5), because they used compar-able methods for measuring and describing childnutritional status. They represent samples of children(aged 6-59 months) from the rural areas of Bangla-desh, India, Indonesia, Malawi, Papua New Guinea,and the United Republic of Tanzania; when properlyweighted by sample sizes they demonstrate remark-able consistency in the relative risk of mortality inrelation to the child's weight-for-age.a These studiessuggest that the risk of mortality increases at a com-pounded rate of 5.9% (±0.8% standard error) foreach percentage point decline in weight-for-agebelow the reference point of 90% weight-for-age.The equivalent rate in z-scores is 5.5% ± 0.8 for eachdecrease of 0.1 z-score units below -1. This trans-lates into relative risks of 8.4 for severe malnutrition(defined as <60% of reference or median weight-for-age), 4.6 for moderate malnutrition (60-69% ofmedian), and 2.5 for mild malnutrition (70-79% ofmedian). The standard errors in these estimates are2.1, 0.9 and 0.3, respectively. Formal statistical tests(regression of log mortality on weight-for-age,weighted by sample size and testing for interactionsbetween weight-for-age and study) reveal that thereis no significant variation in the response of mortal-ity to malnutrition across these eight studies, despitesignificant differences in disease ecology, prevalenceof stunting (low height-for-age) and wasting (lowweight-for-height), and age range of the studychildren. The above relative risk estimates can there-fore be applied to diverse populations to describe therisk of death as a function of child weight-for-age.Further discussion and testing of the assumptions

a Note that two of the eight studies, including that by Chen etal.(8), appear to have slopes that deviate from the others, butthese deviations are not statistically significant when variation insample sizes is taken into account through use of a weightedregression. This exception, along with the two exceptions fromZaire, are discussed elsewhere (4, 5).

inherent in applying these parameter estimates todiverse populations can be found in the originalmethodological paper (4).

Using the aforementioned relative risk estimates,the standard epidemiological statistic of population-attributable risk (PAR) is used to estimate the per-centage of child deaths attributable to malnutrition'spotentiating impact on infectious disease. The PARstatistic combines information on the strength of theassociation between low weight-for-age and mortal-ity and the prevalence of low weight-for-age, to esti-mate the percentage of total deaths statistically asso-ciated with low weight-for-age. Causal inferencesconcerning malnutrition's role cannot be based onthe PAR alone, but derive from other types of bio-logical, clinical and epidemiological evidence con-cerning these relationships.

The formula for PAR (6) is:

Deaths related to malnutritionPAR =

Total deaths

Prev x (RR-1)(1)

1 + [Prev x (RR-1)]

where Prev is the prevalence of malnutrition andRR is the ratio of mortality among the malnourishedto mortality among the non-malnourished (i.e., rela-tive risk).

When information is available on the distribu-tion of malnutrition across severe, moderate andmild grades, the above formula can be extended asfollows:

X(Previ x (RRi-l))PAR = (2)

1 + x, [Previ x (RR,-l)]Where Previ is the prevalence of malnutrition in

each of the three grades (severe, moderate and mild),RRi is the corresponding relative risk of mortalitywithin each grade, and , represents summationacross all grades of malnutrition. As implied by thisformula, the PAR attributable to all grades of malnu.trition can be estimated directly from the prevalenceof low weight-for-age, using the relative risk esti-mates of 8.4, 4.6 and 2.5 for severe, moderate andmild malnutrition, as described above. The testing ofalternative methodologies for estimating PAR (4)reveals greater precision when multiple grades ofmalnutrition are used (equation 2) rather than asingle grade (equation 1). Accordingly, the internation-al prevalence data used in this paper is based on thecompilation by UNICEF (7), which provides the

444 WHO Bulletin OMS. Vol 73 1995

Effects of malnutrition on child mortality

prevalence of severe (<-3 z-scores) and moderate(-2 to -3 z-scores) underweight in 53 developingcountries. This publication also provides an overallweighted average prevalence, based on these samecountries. All of the prevalence estimates in thatsource are derived from nationally representativesample surveys of children aged 0-59 months. Theseprevalence estimates are slightly lower than the truevalues for children aged 6-59 months, upon whomthe relative risk estimates are based and to whomthese results pertain. The z-score-based prevalencesfrom the UNICEF (7) were converted to the percent-age of median prevalences required in equation 2,using the procedure described in Pelletier et al. (4).The results of the PAR calculations are presented interms of the total PAR (arising from all three gradesof malnutrition combined) and the percentage oftotal PAR, which is accounted for by mild-to-moder-ate manutrition (MMM) (60-79% of median weight-for-age).

ResultsFig. 1 shows the estimates of total PAR and the per-centage attributable to MMM for the 53 countriesand the weighted average of all 53 countries (exactvalues shown in Table 1). Countries are orderedfrom lowest to highest, based on the value for totalPAR. The percentage of child deaths attributable tothe potentiating effects of malnutrition ranges from alow of 13 in Paraguay to a high of 67 in India, with aweighted average of 56%. The median for these 53countries, which is not heavily influenced by India,is 38%. The percentage of total PAR that is attribu-table to mild-to-moderate malnutrition ranges from alow of 73-74% in Bangladesh and India to a high of100% in those countries with very low malnutritionprevalences, with a weighted average of 83%. Thus,even in those countries with the highest prevalenceof severe malnutrition, at least three-quarters of allthe malnutrition-related deaths are attributable tomild-to-moderate malnutrition rather than severe.

Fig. 2 shows that there is a close curvilinearrelationship between total PAR in these 53 countriesand the total prevalence of low weight-for-age (i.e.,weight-for-age <80% of median). This relationship iswell described by the following quadratic equation:

Total PAR = (Prev8O x 1.42) + (Prev8O2 x -0.0075)+ 0.87 (3)

where Prev8O is the percentage of children lessthan 80% weight-for-age. This equation has a highR-squared (99.1%) and high precision of prediction(SE of the estimate = 1.37). In addition, with a stan-dard error of 0.76, the intercept (0.87) essentially

Fig. 1. The total population-attributable risk (PAR) forchild deaths due to the potentiating effects of malnutri-tion on infectious diseases but also related to malnutri-tion, which is severe or moderate-to-mild (MMM).

ParaguayDominica

Seycheise Related to:Barbados

_ PAR: MMMJordan

Tnnidad & Tobago _ PAR: SevereJamaica -------Uruguay

NicaraguaDomincan Repub_c

Tunisia _ _Peru

Zimbabwe

AnbguaIraq

CWte d'lvdreN.E. Brazi1

Egypt _----Bolia _

Lesotho 11--- _ _

Ecu adorChina

HondurasGuyanam

Cape Verde

Uganda ----_- _ThadlndZanbia

TogoS. Leone

D7_._.Ghana _Djibouti _ _ _ 1NamibiaRwanda - _-_-_--

Philippirnes _ -adagascarMali

Guatemaa

Sri Lanka 1 i_1_ _ _IBurunc_Nigeria

HaitiTanzaniaIndonesiaPaldstanViet Nam

BangladeshIndia

Waighted average

Percentage of child deaths

passes through zero. This is as expected becausetotal PAR should equal zero when the prevalence ofmalnutrition (Prev8O) is zero. Therefore, this equa-tion represents a simple approach for estimating total

WHO Bulletin OMS. Vol 73 1995 445

D.L. Pelletier et al.

Table 1: Percentage of child deaths attributable to thepotentiating effects of malnutrition (total PAR) and theproportion of this that Is related to mild-to-moderatemalnutrition (MMM) in 53 countries or areas

Total Related toCountry or area PAR (%) MMM (%)

Paraguay 13 100Dominica 14 100Seychelles 15 100Barbados 15 100Jordan 17 100Trinidad & Tobago 17 100Jamaica 18 100Uruguay 19 100Nicaragua 23 100Dominican Republic 23 100Tunisia 23 100Peru 23 100Zimbabwe 24 100Colombia 25 100Antigua 25 92Iraq 25 100C6te d'lvoire 26 100N.E. Brazil 26 99Egypt 27 100Bolivia 27 100Lesotho 29 100Morocco 31 95Ecuador 32 94China 35 98Honduras 36 94

Country or area

GuyanaCape VerdeSenegalUgandaThailandZambiaTogoSierra LeoneGhanaDjiboutiNamibiaRwandaPhilippinesMadagascarMaliGuatemalaMyanmarSri LankaBurundiNigeriaHaitiTanzaniaIndonesiaPakistanViet NamNepalBangladeshIndiaWeighted average

PAR in any population for which the prevalence oflow weight-for-age has been estimated.

Fig. 3 shows a similar strong relationshipbetween the percentage of total PAR attributable tomild-to-moderate malnutrition and the prevalence ofsevere malnutrition (weight-for-age <60%). Theequation describing this relationship is:

Percent MMM = (Prev6O x -9.02) + Prev602 x

0.8058) + 99.2 (4)

where Prev6O is the percentage of children lessthan 60% of median weight-for-age. The R-squaredvalue for this equation is 95.0% and the SEE of 1.84reveals high predictive precision. In addition, theintercept of 99.2 conforms to the expectation thatwhen severe malnutrition (Prev6O) is zero, then thePercent MMM is essentially 100%. Therefore, thisequation can be used in conjunction with equation 3to estimate the percentage of total PAR which isattributable to mild-to-moderate malnutrition.

DiscussionThese results add greatly to our understanding of therelationship between child malnutrition and mortali-ty. First, they reveal that malnutrition, by virtue of itssynergistic relationship with infectious disease, has a

powerful impact on child mortality (total PAR =

56%) and one that is much larger than suggested bythe category of "nutritional deficiencies" in mostroutine reporting systems. Second, they reveal thatthe vast majority of malnutrition-related deaths(83%) are attributable to mild-to-moderate, ratherthan severe, malnutrition. This is contrary to thewidespread perception that mortality is elevated onlyamong the severely malnourished, a perception thatarose in part because one of the earliest studies ofthis relationship found no elevated risk above 65%weight-for-age (8). This result has generally not beenconfirmed in community-based studies since thattime, including several conducted in the same areas

of Bangladesh (5). One exception is a recent report

WHO Bulletin OMS. Vol 73 1995

TotalPAR (%)

3738393940404142424344444648484849505252535354555665666756

Related toMMM (%)

9485899194918883908090969385828383868380799392797880737483

446

Effects of malnutrition on child mortality

Fig. 2. Data from 53 developing countries on the totalpopulation-attributable risk (PAR) for child deaths inrelation to the prevalence of malnutrition (<80% medianweight-for-age).

70

60

50 _ , *

40.< ..

30

20

10

0 20 40 60 80Malnutrition prevalence (%)

from Zaire (9), which the authors hypothesize isattributable to the exceptional importance of malariaas a cause of morbidity and mortality in that setting.

Most studies, to date, have quantified mortalityrisks only as relative risks and other parameterswhich do not take account of the proportions of chil-dren in different grades of malnutrition. The presentfinding, that 83% of all malnutrition-related deathsare attributable to the potentiating effects of mild-to-moderate malnutrition, clearly suggests that interven-tion strategies that rely heavily or exclusively onscreening and treating only the severely malnour-ished will fail to address the major source of themalnutrition-infection synergism. Equations 3 and 4make it possible to estimate the effects of malnutri-tion (and mild-to-moderate malnutrition) on childmortality in any population for which prevalence

Fig. 3. Data from 53 developing countries on the popu-lation-attributable risk (PAR) associated with mild-to-moderate malnutrition (MMM) in relation to severe mal-nutrition (<60% median weight-for-age).

100

p..90 :

80>.80

70

60 T Z0 2 4 6 8Prevalence of < 60% weight-for-age

data exist.b The PAR methodology used here rests onthe empirical observation, based on eight prospectivestudies, that the quantitative impact of malnutritionon child mortality exhibits little variation across pop-ulations with different disease ecology. This obser-vation suggests that the malnutrition-infection syner-gism may have mortality consequences for manyforms of morbidity and is not restricted to the case ofdiarrhoeal illness and measles (10). The aforemen-tioned study from Zaire (9) suggests that malariamay be an exception, when it is the dominant causeof morbidity and mortality, but this finding requiresfurther epidemiological and clinical confirmation.

The quantitative estimates provided here areremarkably similar to those found by the Pan Ameri-can Health Organization (PAHO) in the Inter-Ameri-can Investigation of Mortality in Childhood, con-ducted in 15 countries (11). This study, using directclinical, autopsy and family interview methods toascertain the primary and associated causes of deathfor each individual, found that malnutrition wasimplicated in 56% of all deaths among children aged1-4 years; of these, malnutrition was identified asthe primary cause of death in 15% of cases (indicat-ing severe forms) and as an associated cause of deathin 85% of cases (indicating less severe forms). Thus,the estimates arising from the anthropometry-basedmethod in the present paper are of the same magni-tude as those found using direct methods for ascer-taining the cause of death. They are also similar toPAR estimates by Schroeder & Brown (12) whoused five of the eight studies used here, butemployed different assumptions, methods and datasources.

The malnutrition-infection synergism appears tobe a physiological vicious cycle and has been recog-nized as important for public health policy and prac-tice for over two decades (2). Similarly, the findingsby the Inter-American Investigation conceming itsquantitative effects on mortality at the populationlevel were known twenty years ago, and led PAHOto give high priority to reducing malnutrition as partof programmes to reduce mortality (11). Morerecently, it has been confirmed that malnutrition haspotentiating effects on mortality at the populationlevel (3), precisely as would be predicted from thevicious-cycle concept. An important contribution ofthe present findings is to provide quantitative esti-mates of malnutrition's effects for a broader set ofregions and countries of the world, in order to

b It should be noted that the accuracy of PAR estimates,derived from equations 3 and 4, depends on the accuracy withwhich the prevalence of low weight-for-age was estimated, inaddition to the errors inherent in the equations themselves (4).

WHO Bulletin OMS. Vol 73 1995 447

D.L. Pelletier et al.

encourage a re-examination of the priority given tonutrition in child survival strategies and in relatedresource allocation decisions. It is important to notethat significant reductions in child mortality havebeen and can be achieved through selective healthinterventions, even in the face of persisting malnutri-tion (13, 14). The present results do not contradictthat fact. However, they do imply that malnutritionstrongly exacerbates the burden of life-threateningdiseases in developing countries, increases the asso-ciated health care costs, and increases the mortalityrisk for individuals not adequately covered by thehealth care system. Our findings suggest that theseconcerns could be significantly reduced throughenhanced efforts to reduce malnutrition of all grades,and not the severe cases only. Thus, there are ben-efits in cost-effectiveness, in addition to the humani-tarian and ethical reasons, for ensuring that adequateattention is paid to nutrition in health and develop-ment policies.

Rfsum6Les effets de la malnutrition sur lamortalit6 infantile dans les pays end6veloppementLes methodes habituelles de classification descauses de deces conduisent a penser qu'environ70% des deces d'enfants de 0 a 4 ans dans lemonde sont dus a la diarrhee, aux infections respi-ratoires, au paludisme et aux maladies evitablespar la vaccination. Ces methodes ne mettent pasen evidence le r6le de la malnutrition, bien que l'onsache depuis longtemps qu'il existe une synergieentre la malnutrition et les maladies infectieuses.Le present article decrit une m6thode 6pid6miolo-gique r6cemment mise au point pour estimer lepourcentage de deces qui pourraient etre attribu6saux effets potentialisateurs de la malnutrition surles maladies infectieuses chez les enfants ages de6 a 59 mois. Les donn6es recueillies dans 53 paysen d6veloppement ou il existe des statistiquesnationales sur le poids des enfants en fonction del'age montrent que 56% des deces d'enfants sontattribuables aux effets potentialisateurs de la mal-nutrition et que dans 83% des cas il s'agit d'unemalnutrition l6gere a moderee, et non d'une malnu-trition grave. Selon les pays, les effets potentialisa-teurs de la malnutrition varient entre 13% et 66%,une forme de malnutrition legere a moderee etanten cause dans au moins trois cas sur quatre.

Ces r6sultats montrent que la malnutrition aune incidence beaucoup plus importante sur lamortalite infantile qu'on ne le croit g6neralement.

lls donnent aussi a penser que des strat6giesvisant seulement a depister et a traiter les cas lesplus graves ne contribueront guere a r6soudre leprobleme. La m6thodologie pr6sentee dans cetarticle permet d'estimer les effets de la malnutri-tion sur la mortalite infantile dans toute populationpour laquelle il existe des donnees sur la preva-lence de ce phenomene.

References1. World Bank. World development report, 1993.

Investing in health. New York, Oxford UniversityPress, 1993.

2. Scrimshaw NS, Taylor CE, Gordon JE. Interactionof nutrition and infection. Geneva, World HealthOrganization, 1968 (Monograph Series 57).

3. Pelletier DL, Frongillo EA, Habicht JP. Epidemi-ologic evidence for a potentiating effect of malnutri-tion on child mortality. American journal of publichealth, 1993, 83: 1130-1133.

4. Pelletier DL et al. A methodology for estimating thecontribution of malnutrition to child mortality indeveloping countries. Journal of nutrition, 1994, 124(1OS): 2106-2122.

5. Pelletier DL. Relationship between child anthro-pometry and mortality in developing countries: impli-cations for policy, programs and future research.Journal of nutrition, 1994, 124(1 OS): 2047-2081

6. Kleinbaum DG, Kupper LL, Morgenstern H. eds.Epidemiologic research: principles and quantitativemethods. New York, Van Nostrand Reinhold Com-pany, 1982.

7. UNICEF. Child malnutrition: progress toward the WorldSummit for Children goal. New York, UNICEF, 1993.

8. Chen LC, Chowdhury A, Huffman SL. Anthropo-metric assessment of protein-energy malnutritionand the subsequent risk of mortality among pre-school-aged children. American journal of clinicalnutrition, 1980, 33: 1836-1845.

9. Van den Broeck J, Eeckels R, Vuylsteke J. Influ-ence of nutritional status on child mortality in ruralZaire. Lancet, 1993, 341: 1491-1495.

10. Morley D. Paediatric priorities in the developingworld. London, Butterworth, 1973.

11. Puffer RC, Serrano CV. Patterns of mortality inchildhood. Washington, Pan American HealthOrganization, 1973, (Scientific Publication No. 262).

12. Schroeder DG, Brown KH. Nutritional status as apredictor of child survival: summarizing the associa-tion and quantifying its global impact. Bulletin of theWorld Health Organization, 1994, 72: 569-579.

13. Rutstein SO. Levels, trends and differentials ininfant and child mortality in less developed coun-tries. In: Hill K, ed. Child health priorities for the1990s. Baltimore, Johns Hopkins University Schoolof Hygiene and Public Health, Institute for Interna-tional Programs, 1992.

14. Walsh JA, Warren KS. Selective primary healthcare: an interim strategy for disease control indeveloping countries. New England journal of medi-cine, 1979, 301: 967-974.

448 WHO Bulletin OMS. Vol 73 1995