Risk factors for high cerebral blood flow velocity and death in Kenyan children with Sickle Cell Anaemia: role of haemoglobin oxygen saturation and febrile illness

High cerebral blood flow velocity (CBFv) and low haemoglobin oxygen saturation (SpO2) predict neurological complications in sickle cell anaemia (SCA) but any association is unclear. In a cross-sectional study of 105 Kenyan children, mean CBFv was 120 ± 34·9 cm/s; 3 had conditional CBFv (170–199 cm/s) but none had abnormal CBFv (>200 cm/s). After adjustment for age and haematocrit, CBFv ≥150 cm/s was predicted by SpO2 ≤ 95% and history of fever. Four years later, 10 children were lost to follow-up, none had suffered neurological events and 11/95 (12%) had died, predicted by history of fever but not low SpO2. Natural history of SCA in Africa may be different from North America and Europe.

and conditional cerebral blood flow velocity (CBFv) >200 cm/s and 170-199 cm/s respectively, measured by transcranial Doppler (TCD) ultrasonography of the intracerebral arteries, respectively, predicted 40% and 7% risk of stroke over the subsequent three years (Adams et al, 1992;Kirkham et al, 2001) while CBFv > 150 cm/s may indicate intracranial vessel stenosis (Adams et al, 1990). Peripheral haemoglobin oxygen saturation (SpO 2 ) is also a predictor of overt stroke (Kirkham et al, 2001;Quinn & Sargent, 2008) and might also predict death, since it is associated with pulmonary hypertension (Minniti et al, 2009). There has been no previous description of CBFv or SpO 2 in African SCA patients, their predictors or their use for identifying those at high risk of adverse outcomes. In this report we describe the spectrum of CBFv in a Kenyan paediatric SCA cohort, exploring factors associated with high CBFv and relating these data to mortality.

Study design
The study started in 2002 in Kilifi District Hospital (KDH), following approval by the Kenya National Ethical committee. There is a dedicated outpatient clinic for all SCA patients, seen every 3-4 months after clinical diagnosis as there is no newborn screening. Proguanil is given as antimalarial prophylaxis but none of the patients are treated with penicillin prophylaxis or Hydroxycarbamide. Study patients were recruited when well. Following informed consent, history of clinical events, such as febrile illness and episodes of neurological compromise including seizures, was obtained, systematic general and neurological examination conducted and daytime SpO 2 (pulse oximetry; Nellcor, Pleasanton, CA, USA) measured. Laboratory investigations included blood count (MDII; Beckman Coulter, Fullerton, CA, USA) and sickle genotype confirmed by alkaline haemoglobin electrophoresis (Helena, Sunderland, Tyne & Wear, UK). All information was collected in standardized clinical research forms. Patients were followed up to November 2006 through clinic visits, with admissions, neurological events and deaths captured by active surveillance.
CBFv was measured following the STOP (Stroke prevention in sickle cell disease) protocol (Nichols et al, 2001) using the Companion II (Nicolet, Warwick, UK) machine. The highest CBFv (time-averaged maximal mean velocity in the distal internal carotid artery and middle cerebral artery on either side) was determined. CBFv was classified as low (<50 cm/s), normal (50-149 cm/s) or high ( ‡150 cm/s) based on previous criteria (Adams et al, 1990) and our Kenyan control data (Newton et al, 1996).
STATA 9 (StataCorp, College Station, TX, USA) was used for statistical analysis. Continuous variables were analysed using two-sided t-tests. Categorical variables were analysed by chi-squared test. Logistic regression was used to explore associations between CBFv and clinical and laboratory variables, presenting the results as odds ratio (OR) with 95% confidence interval (CI) . P-values <0AE05 were considered statistically significant. Multivariate analysis included all variables with univariate significance of <0AE1. Using backward elimination, the final model included all variables significant at the 0AE05 level. Binary logistic regression was also used to assess factors predicting death (as date of death was not known).
Patients with high CBFv were significantly more anaemic (haematocrit 22AE9 ± 4AE8% vs. 25AE5 ± 4AE3%, P = 0AE04) with lower mean SpO 2 (97 ± 2AE97% vs. 98AE5 ± 2AE39%, P = 0AE04) compared to the normal CBFv group (Table I). There was a trend for ‡3 reported febrile episodes in the past year to be associated with high CBFv (P = 0AE08; Table I). Multivariate analysis using a model that included age and haematocrit showed that SpO 2 £ 95% (P = 0AE02) and history of fever were associated with high CBFv, with the likelihood of having high CBFv increasing with the number of episodes of reported fever (OR 13AE8; P = 0AE01 for ‡3 reported febrile episodes; Table I).
By 2006, 10 were lost to follow-up, none of whom had a high baseline CBFv or SpO 2 £ 95%. For the 95 children with available data (Table II) none had suffered neurological complications but eleven (12%) had died. There was no significant difference in mortality between those with normal or low CBFv, compared to those with high CBFv (20 vs. 9%; P = 0AE38). In univariate analysis, history of febrile illness (P = 0AE02) at baseline and blood transfusion (P = 0AE04) during follow-up were associated with increased risk of death. Although not statistically significant, survivors had a higher haemoglobin and CBFv, but a paradoxically lower SpO 2 , at baseline. History of febrile illness and blood transfusion were not predictors of death in a multivariate model that also contained peripheral SpO 2 (Table II).

Discussion
This is the first description of CBFv in African SCA children. The lower mean value for CBFv (120 vs. 129 cm/s) and absence of abnormal CBFv, in contrast to the 3-10% prevalence in the USA (Adams et al, 1992), are surprising, given the reportedly higher incidence of stroke in Africa. This may be due to rapid progression of disease, with death swiftly following the development of high CBFv in young children, although we found no evidence for any link between CBFv and mortality. In addition to the possibility of an alternative mechanism, such as venous sinus thrombosis, neurological complications in African children with SCA may be secondary to progressively worsening stenosis, with Short Report ª 2009 The Authors decrease in CBFv and sometimes moyamoya collaterals (Hogan et al, 2005). In our study, low CBFv compatible with, but not diagnostic of, moyamoya (Takase et al, 1997) was documented in seven SCA patients, one of whom had had convulsions; none of these patients was known to have died or have neurological complications 4 years later.
In univariate analysis, patients with high CBFv had lower haematocrit and oxygen saturation and were more likely to have had ‡3 febrile episodes; however on multivariate analysis, only the latter two factors were independently associated with high CBFv. Recent evidence suggests that low oxygen saturation predicts neurological events (Kirkham et al, 2001;Quinn & Sargent, 2008); this might in part be explained by an association between low SpO 2 and high CBFv but any link to stroke risk cannot be ascertained from our data in view of the low prevalence of conditional or abnormal CBFv and the lack of neurological events. 1AE90 (0AE86-4AE20) 0AE11 [Mean (SD)] Geometric mean, standard deviation in parenthesis. Prevalence of patients in the respective groups. Odds ratio [95% Confidence Intervals (95% CI)] and P-value to show the association between the respective factor and death. *Fourteen children were excluded from the multivariate analysis of death due to missing data. Peripheral oxygen saturation was used as a continuous variable as none of the children who died had SpO2 <95%. Although not statistically significant, those who died had a higher SpO 2 at baseline. In studies in the USA, low SpO 2 is associated with pulmonary hypertension, a predictor of death in adults, but these relatively young Kenyan children may not have yet developed this chronic complication (Minniti et al, 2009). Variation in exposure and adaptation to anaemic and hypoxic hypoxia may explain some of the phenotypic variability in SCA (Kirkham & Datta, 2006).
We report an association between parental report of ‡3 febrile illnesses in the past year and high CBFv, independent of age, haematocrit and SpO 2. Mortality was also predicted by previous febrile illness. However, cause of death was not recorded so we cannot infer any direct link with infection or exclude neurological causes. The majority of SCD patients do not currently receive penicillin prophylaxis in most African countries, as there are conflicting data regarding the importance of infection, particularly streptococcal septicaemia, as a cause of morbidity and mortality (Makani et al, 2007). Assuming that febrile illness is commonly associated with infection, it is important to determine the causal agent(s) so that preventative strategies can be evaluated in this setting.
Given that this was a cross-sectional study conducted in 2002, with historical data about CNS events and febrile episodes and limited information available from clinical review and surveillance up to 2006, it is difficult to accurately determine the cause-effect relationship of the various factors and their change over time. Larger longitudinal populationbased birth cohorts, with repeated laboratory measurements, such as CBFv and SpO 2 , and details of acute illness and clinical events before death, are required to identify risk factors and document the natural history of SCA in Africa.
This study shows a novel association between low SpO 2 and increased CBFv, which should be explored in non-African settings. The absence of documented abnormally high CBFv suggests that the course and mechanism of cerebral vasculopathy and neurological events in Africa are different from those in North America and Europe. More detailed neuroimaging studies with magnetic resonance angiography are required to elucidate this further.