Background:  In resource-limited settings, high background infection rates and malnutrition increase mortality in HIV infected children. Pediatric treatment guidelines, based mainly on markers from industrialized country studies, must be validated since commitments have been made to improve access to ART worldwide.

Methods:  Longitudinal data on untreated children aged >12 months from 11 studies (10 African, 1 Brazilian) were combined. Prognostic values for laboratory markers (CD4 percentage, CD4 count, total lymphocyte count, hemoglobin), growth (weight-for-age and height-for-age) on mortality in the next 12 months were assessed using crude mortality rates and Poisson regression, adjusting for age, cotrimoxazole (CTX) use and study effects.

Results:  During 3769 patient-years at risk, 2510 children aged >12 months contributed 357 deaths with 81% follow-up occurring after start of CTX. At first measurement, median age was 4.0 (IQR 2.2 to 7.0) years, median CD4 was 15% and weight z-score –1.9. CD4 percentage and CD4 count were the strongest predictors of mortality, followed by weight for age and hemoglobin. After adjusting for these markers, total lymphocyte count and height-for-age had only weak effects. Children with both weight-for-age z-score <–3 and hemoglobin <8 had particularly high mortality (rate 55.2/100 patient-years compared to 6.9 for those with weight-for-age z-score –3 to –1 and hemoglobin 8 to <10, and 1.4 when Weight-for-age  z-score ≥1 and hemoglobin ≥10). This was true even for children with CD4 ³15%:  mortality risk was 17.3 vs 1.2/100 patient-years for weight-for-age z-score <–3 and ³–3, respectively, and 13.2 vs 2.4/100 patient-years for hemoglobin <8 and ≥8 g/dL. Where weight-for-age z-score ³–3 and hemoglobin ³8 g/dL (75% follow-up time), mortality rates increased from 1.3/100 patient-years with CD4 15 to 25%, to 6.6 and 12.6 at CD4 10 to 14% and <10%, respectively. In children ³5 years, where CD4 count may be the preferred marker, corresponding rates increased from 1.1/100 patient-years when CD4 ³400, to 4.5, 4.5, and 17.2 when CD4 300 to 399, 200 to 299, and <200, respectively. Although numbers are small, for both CD4 percentage and CD4 count, mortality was higher and thresholds less marked where weight-for-age z-score was <–3 or haemoglobin <8 g/dL.

Conclusions:  CD4 percentage and count strongly predict mortality as in European/U.S. studies, whereas total lymphocyte count  appears less predictive. However, for children with severe malnutrition or anemia, mortality was high, even at high CD4 percentage and count. Weight-for-age and hemoglobin need consideration in clinical management of children in resource-limited settings, as recommended by recent revisions to WHO ART treatment guidelines.