IJE Advance Access originally published online on March 11, 2005
International Journal of Epidemiology 2005 34(2):368-375; doi:10.1093/ije/dyh335
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Article |
Comorbidity in childhood in northern Ghana: magnitude, associated factors, and impact on mortality
1 Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK
2 Department of International Health, Johns Hopkins University School of Public Health, Wolfe Street, Baltimore, MD 21205–2179, USA
* Corresponding author. Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. E-mail: bridget.fenn{at}lshtm.ac.uk
 |
Abstract |
|---|
 Top Abstract IntroductionConceptual issuesData and methodsResultsDiscussionReferences |
|---|
Background It has been observed that in developing countries terminal illness in children under 5 yr of age is frequently characterized by comorbidity. This study seeks to quantify the co-occurrence of illness at the community level and investigates whether this co-occurrence increases the risk of mortality. We develop an appropriate measure of co-occurrence and investigate whether the comorbidity occurs by chance or whether it is due to shared risk factors.
Methods The data used for the analysis was taken from a study carried out from 1989 to 1991 in northern Ghana on children aged 2–59 months (n = 1879). Coding for diarrhoea, pneumonia, and measles was carried out using the classification system of the WHO/UNICEF strategy for the Integrated Management of Childhood Illness; malaria was confirmed by laboratory analysis. A bivariate probit analysis was conducted to quantify comorbidity. We used an additive regression model, implemented using the Generalized Estimating Equation approach, to examine the impact on mortality.
Results There is evidence of co-occurrence of diarrhoeal diseases and pneumonia, with greater comorbidity with increasing severity of disease. There is no evidence that the co-occurrence of diarrhoea with severe dehydration and severe pneumonia is characterized by a synergistic effect on mortality risk.
Conclusions Our study has shown that it is possible to have significant co-occurrence of illness at the community level. The bivariate probit procedure was easily adopted and considered appropriate for the analysis of comorbidity. The lack of suitable datasets for a more thorough analysis of comorbidity, and for the evaluation of synergistic effects on mortality, is a major limitation.
Keywords Comorbidity, childhood infectious diseases, northern Ghana, synergy
Accepted 11 August 2004
|
Introduction |
|---|
|
TopAbstractIntroductionConceptual issuesData and methodsResultsDiscussionReferences |
|---|
Terminal illness in children in developing countries is frequently characterized by the co-occurrence of more than one disease—a phenomenon referred to as comorbidity.1 If this phenomenon is common to a variety of settings, then many deaths might possibly be prevented with interventions against either one of a pair of co-occurring diseases. If comorbidity in childhood is prevalent, this could alter the ranking of different public health interventions with respect to the number of lives that could be saved.
Unfortunately, it is difficult to assess the true magnitude of comorbidity in childhood illness as it has received little attention in the research literature. While there is documented evidence of comorbidity in e.g. child psychopathology,2–4 the same does not appear to be true for childhood infectious diseases. A literature review, using Medline, Popline, and other search engines, and using the search terms (comorbid*) AND (child*) AND (diarrh* OR pneumonia OR ARI OR respiratory OR measles OR malaria), failed to reveal research specifically addressing comorbidity in childhood infectious illness or its public health consequences, though some research has been done on the co-occurrence of other diseases with malaria.5–7
Much of what we do know about the epidemiology of terminal illness in childhood in developing countries comes from retrospective studies of child deaths. However, many of the algorithms that are used to classify verbal post-mortem data result in a ‘one death, one cause’ profile, effectively disguising comorbidity. Furthermore, comorbidity in terminal illness can arise in more than one way, reflecting either the regular co-occurrence of particular illnesses in the community and/or synergistic effects, by which we mean an impact greater than the additive impact of the two diseases on an individual child's risk of death. It is important to distinguish between these two situations as the public health implications are different. Where co-occurrence in the community is common, prevention strategies for the two diseases should also be coupled; on the other hand, if two diseases act synergistically, caregivers in the home and in front-line health services need to be made aware of the particular dangers that are involved when the two diseases present simultaneously. Whatever may be the cause, it is clear that ignoring comorbidity in such populations could lead to an inappropriate prioritization of public health interventions for reducing under-five mortality.
This paper uses empirical data from northern Ghana to quantify the extent of comorbidity in a community-based cohort of children aged 2–59 months. We develop an appropriate measure of co-occurrence; test different combinations of illnesses for the presence of comorbidity; attempt to determine to what extent this comorbidity might be due to the concentration of illness at particular times of the year, in particular age groups, or among the undernourished; and seek to establish whether there is any evidence of a synergistic interaction between diseases in terminal illness.
|
Conceptual issues |
|---|
|
TopAbstractIntroductionConceptual issuesData and methodsResultsDiscussionReferences |
|---|
Comorbidity has been defined as: ‘any distinct additional clinical entity that has coexisted or that may occur during the clinical course of a patient who has the index disease under study’.8 Caron and Rutter3 add a qualifier that, for analytical purposes, is extremely useful, defining comorbidity as: ‘the occurrence of more than one disorder that exceeds that expected by chance alone’. We used this definition as the basis for identifying a suitable metric.
In our view, an ‘ideal’ measure of comorbidity would take the value 1 when there was an exact coincidence of the two diseases, 0 when there was only as much comorbidity as would occur by chance alone given the complete independence of the two diseases, and –1 when there was a complete separation of the two diseases. The value of rho estimated from a bivariate probit regression model9 has these properties. The bivariate probit model determines the inter-relatedness of two dependent variables, under the assumption that their error terms (which may include unobserved or unobservable determinants) are distributed as bivariate normal. The extent to which the error terms covary is denoted by rho, the ‘correlation parameter’.10 The two disease outcomes are thus viewed as realizations of two correlated probit distributions, and the degree of correlation is estimated empirically. No other model allows for the simultaneous estimation of two multivariable equations with dichotomous outcomes, with the correlation between the two error terms estimated as one of the model parameters.
In analyses based on this model, when rho 
0 the two disease outcomes are correlated; alternatively stated, the probability of one outcome is dependent on the probability of the other. Increasing values of rho between 0 and 1 indicate increasing correlation. Values between 0 and –1 indicate a negative correlation or separation. When rho = 0 then the outcomes are independent. The bivariate probit analysis allows for statistical significance testing of the null hypothesis that rho = 0.
Where comorbidity does exist, it may have arisen in a number of different ways. These have been extensively discussed by Caron and Rutter,3 and the list below illustrates each mechanism with an example from childhood infectious diseases. They include:
- Shared risk factors: ‘One possible reason for overlap between two disorders is that they share the same risk factors [because] many causal factors are not diagnosis-specific.’ Thus, for example, diarrhoea and pneumonia may both share age as a risk factor because, for both illnesses, peak incidence rates occur in infancy but beyond the neonatal period.11–12
- Overlap between multiple risk factors: Where the risk factors themselves are associated then this may result in comorbidity. This may be the case where a child has the risk factors for both diseases and the risk factors are clearly associated with one another, e.g. age (a risk factor for pneumonia) and (non-) breastfeeding (a risk factor for diarrhoea).
- Comorbid pattern constitutes a meaningful syndrome: In this case, comorbidity alters the clinical significance of the diagnosis, e.g. complicated measles is a distinct syndrome from uncomplicated measles.
- One disorder creates an increased risk for the other: For example, malaria may increase the risk of pneumonia by suppressing host resistance to bacterial or viral pathogens, or pneumonia might activate a latent malarial infection.13
- Overlapping diagnostic criteria: Where the same symptoms are used in the definition of two, or several, different diseases. For example, ‘fever’ in pneumonia and malaria may result in an artefactual effect of comorbidity due to the colinearity of the diagnostic criteria.13 There is also a danger of an artefactual association between severity and extent of comorbidity, as severe diseases with ‘many symptoms are likely to have a greater chance of fulfilling the criteria for more than one disease’;3 e.g. ‘danger signs’, such as vomiting, drowsiness, and food refusal, in severe pneumonia and severe malaria.
- One disorder is part of the other: For example, diarrhoea may be considered both a disease and a symptom, and may be classified incorrectly, as is possible in the diagnosis of malaria.
|
Data and methods |
|---|
|
TopAbstractIntroductionConceptual issuesData and methodsResultsDiscussionReferences |
|---|
This analysis uses data from a study carried out from 1989 to 1991 in northern Ghana that collected data on both mortality and morbidity on children aged 2–59 months, the Ghana Vitamin A Supplementation Trials (VAST).14
The Ghana VAST study was originally set up to assess the impact of vitamin A supplementation on child mortality and morbidity. The Child Health Study was one of two substudies conducted simultaneously. In this study, 1455 children aged 6–59 months, who were monitored every week for 1 yr, were randomized to receive vitamin A or placebo. A total of 1962 children aged 0–59 months were enrolled into the morbidity surveillance system.
The data were collected using extensive questionnaires designed to capture numerous possible symptoms, defined on the basis of both ‘emic’ and ‘etic’ disease classifications, and resulting in a dataset with multiple outcomes. The morbidity survey was carried out by weekly surveillance of all children under 5 years of age within the study area. Information was collected weekly from mothers in their homes and by simple physical examination by trained non-medical field workers.
Coding of diseases was carried out using the classification system of the WHO/UNICEF strategy for the Integrated Management of Childhood Illness (IMCI).15 For the purpose of this analysis, four diseases from the IMCI guidelines were included (Table 1) based on specificity of diagnosis and whether or not they were direct causes of death: diarrhoeal diseases, pneumonia, malaria, and measles. However, classification of malaria from the IMCI guidelines was not considered specific enough for this study and so only slide-confirmed malaria was included in the analysis.
|
Four complementary datasets were used for the comorbidity analysis. These were: (i) weekly surveillance data, 79 974 records: repeated data on 1962 infants and children, infants under 2 months of age (n = 1435) and older infants and children (n = 78 539). Information was collected on morbidity symptoms. Of these children 1839 were actually present in their homes for at least 1 week during the surveillance period and 53 died before the end of the surveillance period. (ii) Monthly surveillance data, 20 245 records: infants under 2 months of age (n = 220) and older infants and children (n = 20 025). Information was collected on weight, breastfeeding, and number of people sleeping in the same room as the child (as a proxy for crowding). (iii) Malaria data (a random cross-sectional subsample of the study children collected at 12 points in time, over 2 days every 2 months), 421 records: infants under 2 months of age (n = 13) and older infants and children (n = 408). Information was collected on parasite type and density (on thick blood films), axillary temperature, and reported febrile illness. (4) Vitamin A supplementation treatment group.
These four datasets were merged to provide the final dataset for analysis. The analysis was carried out only on the 2–59 month group as the 0- up to 2-month group was considered too small to allow for meaningful analysis. The final dataset consisted of 1867 children (78 541 records) 2–59 months: 50.4% girls, 49.6% boys. The age distribution of the population and the number of visits per child and prevalence of disease by age group are shown in Table 2.
|
Statistical analysis
Bivariate probit analysis (adjusted for within-child clustering to take into account the dependence among observations due to repeated measures on the same children) was carried out using Stata 8.0 (StataCorp, College Station, TX). Risk factors identified for comorbidity were: anthropometric status, age, season, breastfeeding, vitamin A supplementation allocation status, and crowding. The analysis was carried out with and without adjustments for these risk factors, separately and in combination.
Adjustments for age were made taking into account the age distribution of each disease. It was assumed that the associations between the disease outcomes and the children's ages (in months) would not necessarily be linear; flexible curve fitting using fractional polynomial regression analysis16 was therefore used to identify a ‘best fit’ model. For seasonality, it was not considered possible to categorize the data into distinct seasons, and a dummy variable was therefore constructed to represent each calendar month.
Anthropometric status was modelled as a continuous variable: weight for age (W/A). W/A is a general indicator of nutritional status and is measured in Z-scores, which equate to standard deviations from the median of a reference population (US data collected by the National Centre for Health Statistics).17 Breastfeeding was modelled as a categorical variable; children were breastfed exclusively, breastfed with solids or semi-solids at least once a day, or not breastfed; the last category consisting of those children who were either not breastfed or were breastfed no more than once a week. A crowding variable was constructed as a continuous variable and vitamin A supplementation status was modelled as a binary variable.
The bivariate probit model was first run for all the different illness combinations without taking into consideration any of the risk factors, generating unadjusted, or crude rhos. More complex models were then estimated adjusting separately for age, and season, and then for both variables together. For each model, the change in rho was determined by comparing adjusted and unadjusted values. The model was then rerun adjusting for all of the other risk factors, with age and season adjusted for a priori, and this was compared with the age- and season-adjusted model.
To examine directly whether illnesses occurring in combination were more likely to result in a fatal outcome than illnesses occurring singly, we calculated week-to-week mortality risks, and identified those morbid conditions associated with a significantly raised probability of death in the following week. We then tested whether the incremental mortality risk associated with one life-threatening condition was modified by the presence of a second condition. We used an additive rather than a multiplicative regression model because our null hypothesis was that each illness would increase a child's risk of death over the next week by a fixed amount. That is to say that the risk of death associated with co-occurrence of A and B would be equal to the risk of death associated with A plus the risk of death associated with B. Any greater-than-additive increase in risk would constitute evidence of synergy. This model was implemented using the Generalized Estimating Equation approach of Liang and Zeger.18 The error distribution was specified as binomial, in accordance with the dichotomous nature of the outcome variable. It was assumed that there would be clustering within children because of unobserved between-child heterogeneity in mortality risk, but because of uncertainty about the exact form of the covariance matrix, the models were estimated using robust standard errors.
We also attempted to analyse associations with slide-confirmed malaria, but because of the overall small number of deaths and the fact that malaria slides were taken at predetermined dates on a subset of all the children in the study, there turned out to be no deaths at all in this subsample of children in the weeks immediately following slide preparation.
|
Results |
|---|
|
TopAbstractIntroductionConceptual issuesData and methodsResultsDiscussionReferences |
|---|
The unadjusted comorbidity rhos ranged from 0.67 for severe pneumonia + diarrhoea with severe dehydration to –0.81 for dysentery + malaria. The median value of rho was 0.24. Adjusting for all other shared risk factors, the comorbidity rhos ranged from 0.64 for severe pneumonia + diarrhoea with severe dehydration to –0.79 for severe persistent diarrhoea + malaria, with a median value of rho of 0.25. Figures 1 and 2 show the unadjusted and adjusted distributions of rhos for different illnesses combinations, and indicate similar left-skewed distributions with many of the rhos clustered between 0 and 0.5. The extreme negative values included only combinations with malaria. No values for measles + malaria are reported due to sparse data problems.
|
|
For the unadjusted results, 23/28 (82%) combinations had significant rhos (P < 0.05), and for the adjusted results 17/24 (72%) combinations were significantly different from zero. Adjusting for all other risk factors resulted in four of the comorbid pairs with malaria being dropped due to sparse data. Combinations with malaria were rarely significant, although this may have been due to the much smaller sample size (total number of weeks of observations = 404; malaria positive = 46).
The effect of age or season (independently) on comorbidity was minimal. Adjusting for age resulted in changes in rho ranging from –0.03 to 0.06 (median 0.01). Adjusting for season resulted in changes in rho ranging from 20.04 to 0.08 (median 0.01). As expected, if the age or seasonal patterns of two diseases were similar, then the adjustment resulted in a decrease in rho, whereas dissimilarities in the age or seasonal distributions of the two diseases resulted in an increase in rho.
Adjusting for all other shared risk factors (with age and season adjusted a priori) showed a number of greater, but selective, changes in rho (change in rho ranging from –0.14 to 0.45; median 0.03) than adjusting for age and season alone, although these changes were still very small. Some patterns emerged: increasing comorbidity with dysentery; dysentery and severe pneumonia (rho increased from 0.25 to 0.39) and dysentery and pneumonia (rho increased from 0.19 to 0.29); and decreasing comorbidity with measles; measles and diarrhoea with severe dehydration (rho decreased from 0.44 to 0.22) and measles and severe pneumonia (rho decreased from 0.35 to 0.10). An increase in comorbidity implied that the associations between the various shared risk factors and the two co-occurring diseases were different, whereas decreasing comorbidity reflected similar associations between the various shared risk factors and each of the two co-occurring diseases.
Table 3 lists all possible pairs of disease co-occurrence ordered (not adjusted), with those with the greatest comorbidity rhos at the top of the table and those showing separation at the bottom of the table. It can be seen that increasing severity is associated with higher levels of co-occurrence, e.g. the combination diarrhoea with severe dehydration + severe pneumonia has the highest comorbidity coefficient (unadjusted rho = 0.67, adjusted rho = 0.64). Negative coefficients, indicating statistical separation of the two diseases, occur only in combinations with malaria. However, as mentioned previously, the malaria data should be interpreted with caution because of the small sample size.
|
Table 4 shows the results of the analysis of mortality-risk interaction between diarrhoea with severe dehydration and severe pneumonia. Both diarrhoea with severe dehydration and severe pneumonia are independently associated with death in the following week (P < 0.005 in each case). However, the interaction term is statistically indistinguishable from zero, indicating that there is no evidence that the two diseases have a synergistic effect on mortality risk. The 95% confidence interval for this interaction effect is very wide, ranging from –23.3 per thousand to 26.2 per thousand. No other possible interactions were investigated because other disease outcomes were either not significantly associated with mortality in the following week or were so rare that the test for interaction would have had no statistical power.
|
|
Discussion |
|---|
|
TopAbstractIntroductionConceptual issuesData and methodsResultsDiscussionReferences |
|---|
Our study has demonstrated significant co-occurrence of illness at the community level. Certain disease combinations, particularly those involving pneumonia and diarrhoea, were more likely to be observed than others, and increasing severity of disease was associated with greater comorbidity. The comorbidity that we have detected is not explained by the fact that different diseases have the same age and/or seasonal patterns, or by similar distributions of other shared risk factors.
The bivariate probit procedure was found to be an appropriate way of analysing comorbidity. The datasets used, although adequate for analyses of diarrhoeal disease and pneumonia (and possibly measles), were inadequate for analyses of parasite-confirmed malaria due to the small sample size of the malaria dataset. For the malaria subset, where the number of child-weeks of observation with a particular disease outcome was very limited [as in the cases of dysentery (n = 6), severe pneumonia (n = 2), pneumonia (n = 4), and diarrhoea with severe dehydration (n = 4)], the model became unstable, resulting in rhos at the extreme end of the negative range. This effect was even more marked when attempting to adjust for multiple covariates, using the malaria subset; for these estimations, the model did not converge at all. An alternative dataset, if available, would be necessary for analysing malaria comorbidity. However, studies already carried out,5–7 describing the lack of association, beyond chance, between malaria and pneumonia and between malaria and diarrhoea, suggest that more emphasis should be placed on examining associations between other comorbid pairs. Our risk-modelling approach seems promising in that, it shows that the IMCI diagnostic categories for diarrhoea with severe dehydration and severe pneumonia do predict mortality in this setting. Unfortunately, the study only had limited power to detect significant interactions between different conditions afflicting the same child simultaneously. To make definitive statements about the effects of such interactions, one would need to analyse a comparable study with many more deaths. We looked for, but were unable to locate, such a study.
A couple of points that should be kept in mind when interpreting these results concern the data source. First, since the VAST study was carried out over 10 years ago, the questionnaires for data collection were designed prior to the IMCI disease classification system; as a result not all the necessary information for complete disease classification was available. Second, the fact that we are using a dataset from a study that was not designed specifically for this analysis has implications for the levels of comorbidity identified. The observed comorbidity may be the true correlation between the two diseases or may, in fact, be a result of residual confounding (specification error) due to missing, but crucial, covariates from the model. Although we have tried to adjust for as many risk factors as possible, it has not been possible to adjust for all known risk factors of the diseases under study due to the unavailability of this data from the original study. One other point to consider in the interpretation of our results relates to multiple testing. Many significance tests were carried out, and we acknowledge the possibility of Type 1 error occurring. However, at this stage, we are more concerned with looking for regular patterns occurring between the correlations between different comorbid pairs. The occurrence of increasing correlation with increasing severity of disease is one such pattern identified.
This study provides a preliminary quantification of comorbidity using one set of data from a specific population. Although the bivariate probit analysis could be replicated in other settings to quantify comorbidity, generalizability of these specific results to other populations is limited since comorbidity is expected to vary depending on the population characteristics, e.g. the incidence and distribution of diseases and distribution of their risk factors. The evidence of comorbidity at the community level, in this setting, suggests the need for further research in this area, especially with regard to possible synergistic effects of comorbidity on child mortality. These questions need to be answered before we are able to make unequivocal recommendations on the importance of comorbidity for public health interventions in child survival. Unfortunately, the lack of suitable datasets is a major limiting factor in this research.
KEY MESSAGES
|
|
Acknowledgments |
|---|
This work was carried out under the auspices of the Child Health Epidemiology Reference Group of the Child and Adolescent Health and Development Department of the World Health Organization, with funding from the Bill and Melinda Gates Foundation. In addition to the members of this group, the authors would like to thank Jennifer Bryce, Jean Pierre Habicht, and Simon Cousens for their useful comments on earlier versions of this analysis.
|
References |
|---|
|
TopAbstractIntroductionConceptual issuesData and methodsResultsDiscussionReferences |
|---|
1 Black RE, Morris SS, Bryce J. Where and why are 10 million children dying every year? Lancet 2003; 361:2226–34.[CrossRef][Web of Science][Medline]
2 Achenbach T. ‘Comorbidity’ in child and adolescent psychiatry: categorical and quantitative perspectives. J Child Adolesc Psychopharmacol 1990/91;1:271–78.
3 Caron C, Rutter M. Comorbidity in child psychopathology: concepts, issues and research strategies. J Child Psychol Psychiatry 1991; 32:1063–80.[Web of Science][Medline]
4 Nottelmann E, Jensen P. Comorbidity of disorders in children and adolescents: Developmental perspectives. In: Ollendick T, Prinz R (eds). Advances in Clinical Child Psychology Vol. 17, 109–55. New York: Plenum Press, 1995.
5 Sodemann M, Jakobsen MS, Møllbak K et al. Malaria parasitemia and childhood diarrhea in a peri-urban area of Guinea-Bissau. Am J Trop Med Hyg 1999; 61:336–38.[Abstract]
6 Greenwood BM, Byass P, Greenwood AM et al. Lack of an association between acute gastroenteritis, acute respiratory infections and malaria in young Gambian children. Trans R Soc Trop Med Hyg 1989; 83:595–98.[CrossRef][Web of Science][Medline]
7 Sodeinde O, Gbadegesin RA, Ademowo OG et al. Lack of association between falciparum malaria parasitemia and acute diarrhea in Nigerian children. Am J Trop Med Hyg 1997; 57:702–05.
8 Feinstein AR. The pre-therapeutic classification of comorbidity in chronic disease. J Chronic Dis 1970; 23:455–68.[CrossRef]
9 Greene WH. Econometric Analysis, 3rd edn. New York: Macmillan, 1997.
10 Zorn C. Advanced Regression-Bivariate Probit. Available at: http://www.emory.edu/POLS/zorn/Classes/POLS585AU02/notes4-2.pdf (Accessed September 2003).
11 Williams BG, Gouws E, Boschi-Pinto C et al. Estimates of world-wide distribution of child deaths from acute respiratory infections. Lancet Inf Dis 2002; 2:25–32.[CrossRef][Web of Science][Medline]
12 Kosek M, Bern C, Guerrant RL. The global burden of diarrhoeal disease, as estimated from studies published between 1992 and 2000. Bull World Health Organ 2003; 81:197–204.[Web of Science][Medline]
13 World Health Organisation. The Overlap in the Clinical Presentation and Treatment of Malaria and Pneumonia in Children: Report of a Meeting. Geneva: World Health Organisation, 1992.
14 Ghana VAST Study Team. Vitamin A supplementation in northern Ghana: effects on clinic attendances, hospital admissions, and child mortality. Lancet 1993; 342:7–12.[CrossRef][Web of Science][Medline]
15 WHO/UNICEF strategy for the Integrated Management of Childhood Illness (IMCI). Available at: http://www.who.int/child-adolescent-health/integr.htm (Accessed April 2003).
16 Royston P, Altman DG. Approximating statistical functions by using fractional polynomial regression. Statistician 1997; 46:411–22.
17 Hamil PVV, Drizid TA, Johnson CL, Reed RB, Roche AF, Moore WM. NCHS Growth Curves for Children, Birth-18 years. Vital and health statistics series 11, no. 165. DHEW publication no. (PHS):78–1650. Hyattsville, MD: USHEW-PHS, 1977.
18 Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika 1986; 73:13–22.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  W.-P. Schmidt, S. Cairncross, M. L Barreto, T. Clasen, and B. Genser Recent diarrhoeal illness and risk of lower respiratory infections in children under the age of 5 years Int. J. Epidemiol., June 1, 2009; 38(3): 766 - 772. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. L F. Walker and R. E Black Commentary: What is the role of co-morbidity in child mortality? Int. J. Epidemiol., June 1, 2009; 38(3): 772 - 774. [Full Text] [PDF]
|
|
|
|
 |
|
 D C Fuhr and M J De Silva Physical long-term health problems and mental comorbidity: evidence from Vietnam Arch. Dis. Child., August 1, 2008; 93(8): 686 - 689. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Gwer, C. R.J.C. Newton, and J. A. Berkley Over-Diagnosis and Co-Morbidity of Severe Malaria in African Children: A Guide for Clinicians Am J Trop Med Hyg, December 1, 2007; 77(6_Suppl): 6 - 13. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H P S Sachdev Commentary: Utilizing information on causes of neonatal deaths in less-developed countries Int. J. Epidemiol., June 1, 2006; 35(3): 718 - 719. [Full Text] [PDF]
|
|
|
|
|
|
K. Mulholland Commentary: Comorbidity as a factor in child health and child survival in developing countries Int. J. Epidemiol., April 1, 2005; 34(2): 375 - 377. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||