Using bootstrap methods to assess model stability, malnutrition was the most robust predictor of mortality, with all models supporting the inclusion of either current (88% of models) or past history of malnutrition (remaining 12% of models). Other predictive factors were relatively robust, being supported by 50% to 70% of models. Although there was some support (inclusion in >25% of bootstrap models) for additional independent effects of prior hepatomegaly, splenomegaly, developmental delay, hospital admission for diarrhea, and the number of prior hospital admissions, the study was not powered to detect statistically significant independent effects of all these factors.
Growth Parameters Alone as Predictors of Survival (II)
Although age-adjusted weight, height, and BMI were univariable predictors of mortality, the only independent predictor was weight-for-age [Table 2(II), 100% bootstrap support]. Effects of primary carer and age (nonlinear) remained (Fig. 1).
Laboratory Markers Alone as Predictors of Survival (III)
Baseline CD4 percentage and hemoglobin were the only independent laboratory predictors of survival [Table 2(III), both 100% bootstrap support]; as expected, there was strong confounding between CD4 percentage and total lymphocyte count.
Clinical Events and Weight as Predictors of Survival (IV)
We then considered clinical events and weight together [Table 2(IV)]. The effects of primary carer, age, weight-for-age, and prior oral candidiasis were virtually unchanged, suggesting that these predictors are independent. Of interest, children with additional features of clinical malnutrition were at higher risk of death for any given weight-for-age. After adjusting for baseline weight-for-age, few other symptoms, signs, or reasons for hospitalization (including tuberculosis or single severe bacterial infection) were predictive in main or bootstrap models, suggesting that the effect of these diseases on mortality is captured mainly through their effect on weight-for-age.
Sixty-one of the 143 children with clinical malnutrition at or before baseline had been hospitalized for malnutrition. These children (who had presumably responded to treatment in hospital) were not at higher risk of mortality compared with children of the same weight-for-age who had not previously been diagnosed with other signs/symptoms of clinical malnutrition [HR = 1.09 (95% CI 0.64-1.83) P = 0.76]. Conversely, children with clinical malnutrition but who were never admitted to a hospital remained at higher risk of death compared with children with the same weight-for-age who had not had this diagnosis previously [HR = 1.76 (1.19-2.62) P = 0.005].
Clinical and Laboratory Markers as Predictors of Survival (V)
We then considered clinical events and laboratory markers together as predictors of survival [Table 2(V)]. The effects of primary carer, CD4 percentage, hemoglobin, prior malnutrition, and hospitalizations remained similar, suggesting that these predictors are independent and, specifically, that the effects of malnutrition and prior respiratory tract or recurrent severe bacterial infections are not directly or completely mediated through immunosuppression. Conversely, the effect of oral candidiasis completely disappeared after adjusting for CD4 percentage and hemoglobin (HR = 1.09, P = 0.59).
Overall Predictors of Survival (VI)
Considering all baseline factors together, mother as primary carer, low weight-for-age, low CD4 percentage, low hemoglobin, current malnutrition, and previous hospital admissions for 2 or more severe bacterial infections or other respiratory tract infections all independently predicted higher risk of death [Table 2(VI)]. Oral candidiasis only predicted survival when baseline CD4 percentage was not included, suggesting that it is a proxy for immunosuppression and thus mortality risk. None of the other variables univariably significant at P < 0.20 in Table 1 added important predictive information (P > 0.2). As indicated earlier (model IV), prior tuberculosis diagnosis or a single hospital admission for a severe bacterial infection did not independently predict higher risk of mortality after adjusting for current weight-for-age.
Similar effects were observed in all models with or without adjusting for age. However, increased mortality risk was more pronounced in younger children after adjusting for CD4 percentage (III, V, VI) (Fig. 2). Furthermore, mortality risk in younger children was more pronounced in models that did not adjust for clinical events (models II and III), with or without adjustment for CD4 percentage. Although this may be partially due to effects of HIV not mediated through CD4 percentage and/or long-term survival bias, this also suggests that such clinical events contribute significantly to mortality in younger children, regardless of the level of immunodeficiency or low weight.
Staging of HIV Infection According to Laboratory Parameters and Weight
The strong and independent effect of weight-for-age on survival is demonstrated by survival curves categorized by CD4 percentage, weight-for-age, and hemoglobin in Figure 2A. Children with CD4 percentage <15% generally have poorer survival than those with CD4 percentage ≥15%, as expected, and low weight-for-age and/or hemoglobin makes this worse (lowest 2 curves). However, survival among children with low weight-for-age but high CD4 percentage (≥15%) actually tends to be worse than among those with normal weight-for-age and hemoglobin but low CD4 percentage (<15%).
Clinical Staging of HIV Infection
Finally, we applied the most recently defined 4-stage WHO clinical classification11 to these data. We considered current clinical malnutrition, weight-for-age less than −3, 2 or more prior hospitalizations for nonrespiratory severe bacterial infections, or Burkitt lymphoma as WHO stage 4 disease; prior oral candidiasis, tuberculosis, 2 or more hospitalizations for pneumonia or respiratory infections, hospitalization(s) for diarrhea, weight-for-age between −2 and −3, hemoglobin <8 g/dL, neutrophils <1.0 × 109/L, or platelets <50 × 109/L as WHO stage 3; or none of the above (stages 1 and 2). This classified 253 (49%), 228 (44%), and 33 (6%) children as stages 4, 3, and 1 or 2, respectively. Of these, 121 (48%), 43 (19%), and 1 (3%) died, corresponding to mortality rates per 100 child years at risk (95% CI) of 44.5 (37.2-53.2), 14.7 (10.9-19.8), and 2.3 (0.3-16.7), respectively. In addition, mortality risk was significantly higher among children classified with WHO stage 4 disease on the basis of signs of malnutrition or other clinical WHO stage 4 events with or without low weight-for-age (n = 146) compared with those with low weight-for-age alone (n = 107, P = 0.0003) (Fig. 2B).
Postmortem and cross-sectional clinical studies have emphasized the greater overlap between clinical manifestations in HIV-infected and -uninfected children in Africa than in well-resourced countries, particularly with regard to recurrent bacterial infections and malnutrition.1,4,23-25 A recent meta-analysis of birth cohorts of African HIV-infected infants born to HIV-infected mothers reported that around 50% died by age 2 years, around 7 times the mortality of uninfected children.26 Maternal death was an important risk factor in these young children, but other clinical or laboratory prognostic factors were not available. Small studies in Rwanda4 and Malawi27 reported similarly high mortality rates. All these African studies were birth cohorts with relatively short follow-up. In contrast, the median survival of HIV-infected children before effective ART in Europe and the United States is 8 to 10 years.28-30
In contrast, our prevalent study, with median age 4.4 years at entry, represents the way survivors after infancy with no previous HIV diagnosis present to medical services in many African countries. There are few longitudinal studies considering simple clinical and laboratory prognostic indicators for survival in such untreated children in resource-limited settings.31 In our study, average follow-up was nearly 2 years: low loss to follow-up (5%) is an additional benefit. Children with Pneumocystis jiroveci pneumonia or other active infections at baseline were not enrolled: very few infants were enrolled and were not included in this analysis. Those children identified during inpatient admission (7%) were recruited only after discharge from hospital. In total, one quarter had never been an inpatient, and most had received their HIV diagnosis after being referred from outpatient departments. Clinical information and laboratory measurements were obtained at baseline and supplemented by details of validated previous hospital admissions. As UTH was the main hospital for sick children living in Lusaka during the time of this study, these data are likely to be fairly complete. Although many diagnoses of previous or current illnesses were presumptive, this reflects the situation for most resource-limited settings where many HIV-infected children present late. As UTH is fairly typical of an urban poor setting, we believe our findings are generalizable to the prognosis of children in other resource-limited countries presenting to medical services with clinical symptoms/signs.
We found that malnutrition, previous bacterial infections, and oral candidiasis were the most common clinical diagnoses predicting mortality. The prognostic value of oral candidiasis disappeared after adjusting for CD4 percentage, reinforcing the point made by others that this is a good clinical proxy for low CD4 percentage.32 We found that weight-for-age was a better predictor than height-for-age or weight-for-height; weight has the considerable advantage of being easier to measure than height, especially in younger children. Of interest, weight and growth failure has been correlated with virologic response to ART in HIV-infected children in Europe and the United States.33,34
The relationships between weight-for-age and malnutrition are complex: we found prior clinical malnutrition with low current weight-for-age had a worse prognosis than low weight-for-age alone, and malnutrition predicted mortality independent of CD4 percentage. The latter is not surprising; however, the relationship is likely to be further complicated by the role of malnutrition itself in decreasing CD4 count. Recently, van Kooten Niekerk et al also reported a major effect of malnutrition on mortality in HIV-infected children in South Africa and stressed the importance of addressing access to food when assessing children for ART.35 Finally, an interesting finding in our study was that children with previous malnutrition per se were not at higher risk if they had been successfully treated during a previous hospital admission. This emphasizes the need to include response to treatment of malnutrition during evaluation of children for ART.
Eleven percent of children in this study had prior hospitalizations for multiple severe bacterial infections (mainly pneumonia) or other respiratory tract infections (bronchitis, otitis media, tonsillitis). As with malnutrition, these predicted mortality independent of CD4 percentage. Whereas a single bacterial infection or diagnosis of tuberculosis (almost all presumptive) had some predictive value when analyzed univariably, these no longer predicted mortality after adjusting for weight-for-age. This is not surprising, as wasting may be both a feature of having these diseases and a risk factor for developing them. Previous studies have also reported that survival of children with HIV infection and tuberculosis is closely related to nutritional status.36
We observed a borderline protective effect on mortality of care not being provided by the mother at baseline. Previous studies have reported that maternal death increases the child's risk of death 3-fold.26 However, a recent study suggested that although higher mortality was observed around the time of maternal death, there was no evidence that risk of death was increased among children surviving their mothers by 12 months.37 Motherless children in CHAP may have had stable alternative uninfected carers at baseline who were therefore less likely to become sick and/or die of HIV during the course of the study. We plan further analyses to assess the role of changing carers on mortality in these children.
Low hemoglobin remained a predictor of mortality in this study even after adjusting for CD4 percentage. Total lymphocyte count was highly confounded with CD4, but seemed to be a less robust predictor than suggested by a recent meta-analysis of data from untreated children in Europe and North America38 and, in particular, did not add independent information. The joint effects of total lymphocyte count and hemoglobin need evaluation in a larger data set, planned through the 3cs4kids collaboration pooling natural history data from pediatric cohort studies in resource-limited settings (http://www.3cs4kids.org).
It was not possible to assign many WHO stage 4 diagnoses to the new WHO staging system11 because of difficulties in diagnosing most opportunistic infections in these children. A further limitation of this study was that symptoms and signs specific to all WHO stages were not sought prospectively, particularly for stages 1 and 2. However, despite this, we observed considerable discrimination in mortality risk by stage at baseline. The dominant contribution of low weight-for-age to stage 3 and 4 disease again emphasizes the importance of managing malnutrition before starting ART.
We thank the families and children enrolled in CHAP and other staff from the University Teaching Hospital and the School of Medicine, Lusaka, Zambia.
1. Chintu C, Mudenda V, Lucas S, et al, On behalf of the UNZA-UCLMS Project Paediatric Post-mortem Study Group. Lung diseases at necropsy in African children dying from respiratory illnesses: a descriptive necropsy study. Lancet
2. Graham SM, Mtitimila EI, Kamanga HS, et al. Clinical presentation and outcome of Pneumocystis carinii pneumonia in Malawian children. Lancet
3. Zar HJ, Dechaboon A, Hanslo D, et al. Pneumocystis carinii pneumonia in South African children infected with human immunodeficiency virus. Pediatr Infect Dis J
4. Spira R, Lepage P, Msellati P, et al. Natural history of human immunodeficiency virus type 1 infection in children: a five-year prospective study in Rwanda. Pediatrics
5. Taha TE, Kumwenda NI, Hoover DR, et al. Association of HIV-1 load and CD4 lymphocyte count with mortality among untreated African children over one year of age. AIDS
6. HIV Paediatric Prognostic Markers Collaborative Study Group. Short-term risk of disease progression in HIV-1 infected children receiving no antiretroviral therapy or zidovudine monotherapy: a meta-analysis. Lancet
7. Galli L, de Martino M, Tovo PA, et al. Predictive value of the HIV paediatric classification system for the long-term course of perinatally infected children. Int J Epidemiol
8. Barnhart HX, Caldwell MB, Thomas P, et al. Natural history of human immunodeficiency virus disease in perinatally infected children: an analysis from the Pediatric Spectrum of Disease Project. Pediatrics
12. Centers for Disease Control and Prevention. 1994 revised classification system for human immunodeficiency virus infection in children less than 13 years of age. MMWR
. 1994; 43 (No RR-12).
13. Chintu C, Bhat GJ, Walker AS, et al, On behalf of the CHAP Trial team. Co-trimoxazole as prophylaxis against opportunistic infections in HIV-infected Zambian children (CHAP): a double-blind randomised placebo-controlled trial. Lancet
14. McIntosh N, Helms P, Smyth R. Forfar and Arneil's Textbook of Pediatrics
, 6th ed. London: Churchill Livingstone, 2003:576-577.
15. Wade AM, Ades AE. Age related reference ranges: significance test for models and confidence intervals for centiles. Stat Med
16. Cole TJ, Freeman JV, Preece MA. British 1990 growth reference centiles for weight, height, body mass index and head circumference fitted by maximum penalized likelihood. Stat Med
17. van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis. Stat Med
18. Royston P. Multiple imputation of missing values. Stata J
19. Royston P, Altman DG, Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med
20. Royston P, Altman DG. Regression using fractional polynomials of continuous covariates: parsimonious parametric modelling. Appl Stat
21. Sauerbrei W, Royston P. Building multivariable prognostic and diagnostic models: transformation of the predictors by using fractional polynomials. J R Stat Soc Ser A
22. Royston P, Sauerbrei W. Stability of multivariable fractional polynomial models with selection of variables and transformations: a bootstrap investigation. Stat Med
23. Chintu C, Luo C, Bhat G, et al. Impact of the human immunodeficiency virus type-1 on common pediatric illnesses in Zambia. J Trop Pediatr
24. Marum LH, Tindeybwa D, Gibb DM. Care of children with HIV infection and AIDS in Africa. AIDS
. 1997;(suppl 11):S125-S134.
25. Horwood C, Liebeschuetz S, Blaauw D, et al. Diagnosis of pediatric HIV infection in a primary health care setting with a clinical algorithm. Bull WHO
26. Newell ML, Coovadia H, Cortina-Borja M, et al. Mortality of infected and uninfected infants born to HIV-infected mothers in Africa: a pooled analysis. Lancet
27. Taha TE, Graham SM, Kumwenda NI, et al. Morbidity among human immunodeficiency virus-1-infected and -uninfected African children. Pediatrics
28. Blanche S, Newell ML, Mayaux MJ, et al. Morbidity and mortality in European children vertically infected by HIV-1. The French Pediatric HIV Infection Study Group and European Collaborative Study. J Acquir Immune Defic Syndr
29. Tovo PA, de Martino M, Gabiano C, et al. Prognostic factors and survival in children with perinatal HIV-1 infection. The Italian register for HIV infections in children. Lancet
30. Scott GB, Hutto C, Makuch RW, et al. Survival in children with perinatally acquired human immunodeficiency virus type 1 infection. N Engl J Med
31. Marston M, Zaba B, Salomon JA, et al. Estimating the net effect of HIV on child mortality in African populations affected by generalized HIV epidemics. J Acquir Immune Defic Syndr
32. Yeung S, Wilkinson D, Escott S, et al. Paediatric HIV infection in a rural South African district hospital. J Trop Pediatr
33. Paediatric European Network for Treatment of AIDS (PENTA). Comparison of dual nucleoside-analogue reverse-transcriptase inhibitor regimens with and without nelfinavir in children with HIV-1 who have not previously been treated: the PENTA 5 randomised trial. Lancet
34. Lindsey JC, Hughes MD, McKinney RE, et al. Treatment-mediated changes in human immunodeficiency virus (HIV) type 1 RNA and CD4 cell counts as predictors of weight growth failure, cognitive decline, and survival in HIV-infected children. J Infect Dis
35. van Kooten Niekerk NK, Knies MM, Howard J, et al. The first 5 years of the family clinic for HIV at Tygerberg Hospital: family demographics, survival of children and early impact of antiretroviral therapy. J Trop Pediatr
36. Hussey GD, Reijnhart RM, Sebens AM, et al. Survival of children in Cape Town known to be vertically infected with HIV-1. S Afr J Med
37. Zaba B, Whitworth J, Marston M, et al. HIV and mortality of mothers and children: evidence from cohort studies in Uganda, Tanzania, and Malawi. Epidemiology
38. HIV Paediatric Prognostic Markers Collaborative Study Group. Use of total lymphocyte count for informing when to start antiretroviral therapy in HIV-infected children: a meta-analysis of longitudinal data. Lancet
THE CHAP TRIAL TEAM
R. Chileshe, C. Kalengo, A. Musweu Muyawa, J. Kaluwaji, M.M. Mutengo, V. Bwalya, P. Chitambala provided counseling care and follow-up for the children and their families; M. Choongo, L. Namakube, N. Kaganson, and P. Kelleher from the data entry and management team; P. Kelleher, N. Kaganson, and S. Mutambo did data monitoring. L. Farelly, N. Kaganson from the Clinical Trials Unit; J. Mwansa, D. Mwenya, K. Mutela from Microbiology; G. Mulundu, F. Kasolo from Virology; M. Yumbe from Haematology; B. Mandanda, M. Mutengo from Parasitology; V. Mudenda from Pathology; and L. Banda, T. Chipoya, B. Chanda were support staff for the CHAP team.
S. Patel and C. Ling provided microbiology support from the Royal Free and University College Medical School, London. I. Chitsike (Zimbabwe) and C. Luo (Zambia) assisted with the early development of the CHAP trial.
Data and Safety Monitoring Committee: T. Peto (Chairman), M. Sharland, M. Quigley, and G. Biemba.
Keywords:© 2006 Lippincott Williams & Wilkins, Inc.
HIV; pediatric; natural history; Africa