Angiopoietin-2 levels are associated with retinopathy and predict mortality in Malawian children with cerebral malaria: a retrospective case-control study*

Abstract

Objective: To investigate the relationship among the angiopoietin-Tie-2 system, retinopathy, and mortality in children with cerebral malaria.

Design: A case-control study of retinopathy-positive vs. retinopathy-negative children with clinically defined cerebral malaria.

Setting: Queen Elizabeth Central Hospital in Blantyre, Malawi.

Subjects: One hundred fifty-five children presenting with severe malaria and meeting a strict definition of clinical cerebral malaria (Blantyre Coma Score ≤ 2, Plasmodium falciparum parasitemia, no other identifiable cause for coma) were included in the study.

Interventions: None.

Measurements and main results: Clinical and laboratory parameters were recorded at admission and funduscopic examinations were performed. Admission levels of angiopoietin-1, angiopoietin-2, and a soluble version of their cognate receptor were measured by enzyme-linked immunosorbent assay. We show that angiopoietin-1 levels are decreased and angiopoietin-2 and soluble Tie-2 levels are increased in children with cerebral malaria who had retinopathy compared with those who did not. Angiopoietin-2 and soluble Tie-2 were independent predictors of retinopathy (adjusted odds ratio [95% CI], angiopoietin-2, 4.3 [1.3-14.6], p = .019; soluble Tie-2, 9.7 [2.1-45.8], p = .004). Angiopoietin-2 and soluble Tie-2 were positively correlated with the number of hemorrhages, the severity or retinal whitening, and the extent of capillary whitening observed on funduscopic examination (p < .05 after adjustment for multiple comparisons). Angiopoietin-2 and soluble Tie-2 levels were elevated in children with cerebral malaria who subsequently died and angiopoetin-2 was an independent predictor of death (adjusted odds ratio: 3.9 [1.2-12.7], p = .024). When combined with clinical parameters, angiopoetin-2 improved prediction of mortality using logistic regression models and classification trees.

Conclusions: These results provide insights into mechanisms of endothelial activation in cerebral malaria and indicate that the angiopoietin-Tie-2 axis is associated with retinopathy and mortality in pediatric cerebral malaria.

Figure 1. Endothelial biomarkers and lactate are associated with retinopathy in children with cerebral malaria

Bar graphs showing the median and scatter of endothelial biomarkers and venous lactate. (A) Angiopoietin (Ang)-1 (ng/mL), (B) Ang-2 (ng/mL), (C) sTie-2 (ng/mL), and (D) venous lactate (mmol/L) were measured in children with clinically defined cerebral malaria by ELISA with (n= 103) or without (n= 50) malarial retinopathy as defined by the presence of hemorrhage, retinal whitening or vessel changes. Analysis by Mann-Whitney (U statistic, p-value), †significant after Holms correction for 4 pair-wise comparisons: *p<0.05. **p<0.01, ***p<0.001. (A) Ang-1 (1969, p=0.018) †, (B) Ang-2 (1288, p<0.0001) †, (C) sTie-2 (1258, p<0.0001) †, and (D) venous lactate (1358, p=0.004) †.

Figure 2. Venn-diagram depicting the distribution of retinal changes in the CM patients with retinopathy

A Venn diagram depicting the distribution of retinal changes according to the number of children with hemorrhages (H), retinal whitening (W), or vessel abnormalities (V: orange vessels, white vessels, white capillaries). Data are presented as percentages followed by the number of children in each group. Detailed classification of retinal changes was available for 91 of the 103 retinopathy positive children. A total of 70.3% children had hemorrhages (n=64), 94.5% had retinal whitening (n=86) and 76.9% had vessel abnormalities (n=70).

Figure 3. Endothelial markers and venous lactate predict mortality

(A-D) Graphs showing the median and scatter of endothelial biomarkers and venous lactate, with the associated (E-F) receiver operating characteristic (ROC) curves and (I-L) decision plots of sensitivity and specificity generated from the ROC curves. Prognostic accuracy was assessed using receiver operating characteristic (ROC) curves and comparing the area under the ROC curve (AUC: 95% CI). Levels of Angiopoietin (Ang)-2 (B, F, J), sTie-2 (C, G, K) and venous lactate (D, H, L) were significantly elevated in children who died compared to those who survived (AUC (95% CI), p-value: Ang-2, 0.71 (0.63-0.78), p<0.0001; sTie-2, 0.64 (0.55-0.73), p=0.003; venous lactate, 0.67 (0.59-0.78), p=0.002). There was no difference in Ang-1 levels (A, E, I) between children who survived and children who died (AUC (95% CI), p-value: 0.56 (0.47-0.65), p=0.20). The AUC were compared between lactate, sTie-2, and Ang-2 and there were no significant differences (method of Delong et al.). The vertical line represents the cut-point for dichotomizing the biomarkers by CRT or established criteria (hyperlactatemia).

Figure 4. Classification and regression tree models for the prediction of mortality in children with cerebral malaria

We used classification and regression tree analysis (CRT) to generate an intuitive decision tree based on significant predictors of mortality according to the following conditions: misclassification cost, 10; cross-validation, 10-fold (if appropriate); prior probability of death, 23.8%; minimum number per child node, 25; and impurity measure, gini criterion. (A) Model generated from all significant independent variables (Angiopoietin (Ang)-2 (cut-point >3.85ng/mL): sensitivity- 93.2%, specificity- 36.5%, cross-validated misclassification rate (standard error)- 0.231 (0.032); Clinically informed models: (B) Blantyre Coma Score plus Ang-2 (sensitivity- 94.9%, specificity- 34.4%, misclassification rate (standard error)- 0.198 (0.038); (C) Respiratory distress plus Ang-2 (sensitivity- 93.2%, specificity- 27.1%, misclassification rate (standard error)- 0.228 (0.035).