Ethics statement

The study was performed in accordance with the Declaration of Helsinki [24]. Ethical approval was granted by the Lao Medical Ethics Committee (IFMT in Laos being the promoter of the study), and the Ministry of Health authorities in Madagascar. Parents/guardians were informed about the study in Malagasy language and given an information sheet describing the study. Children were included if their parents/guardians had given written informed consent. This was considered to be the most appropriate mode as the study was of a simple, observational design with minimal risk.

Study setting site

The study was conducted in the admissions unit of Befelatanana University Hospital, the main pediatric institution, in Antananarivo, the capital of Madagascar, between March and June 2010. In Madagascar (19 million inhabitants, 45% under 15 years) 68% of the population lives below the poverty line of USD 1.25 per day. Per capita income was USD 950 in 2011. The illiteracy rate is higher among women (32%) than men (22.2%). Malaria is mainly endemic in coastal regions, not in the highlands [25], [26]. The under-5 mortality rate is 106 per thousand while 15% of children below the age of five years suffer from acute malnutrition and 53% from stunting [27].

IMCI was introduced in Madagascar in 1995, and is currently used in the main hospitals and 90% of districts to triage severely ill children [28], [29].

Study site, patients and clinical procedures

The pediatric department of Befelatanana Hospital (75 beds) admits on average 3000 children per year [29]. The frequency of severe illness and general signs of danger is high in the admissions unit (56.7% and 38.2%, respectively) [29]. In-patient mortality rate dropped from 16% in 1998 (with 47.6% of deaths attributed to diarrhea) to 7% in 2009 [29]. Patients (or their parents) have to pay hospital fees and buy all necessary medical supplies (i.e. syringes and infusion sets) and prescribed medications from the pharmacy at the hospital before any treatment can begin. Incoming children are usually received at the admissions unit where a team of doctors and nurses are on duty 24 hours. The IMCI algorithm is used to triage the severity of children and determine who will be admitted to the ward, referred to peripheral units, or sent back home with a prescription [22], [29]. According to hospital guidelines, all children with general signs of danger, and all children brought in at night, are admitted to the pediatric ward, whatever their IMCI status.

Participants

All children (1 month–15 years) admitted to the pediatric service during the investigators' duty hours were eligible into the study. Emergency cases were included only if inclusion would not cause delays in their treatment. Children with known diabetes or hemophilia, or a history of neonatal hypoglycemia, were excluded as well as children who had been previously enrolled in the study.

Investigators were on duty for 48 hours, then given a 24-hour break.

Clinical evaluation

The medical history and examination findings were recorded in a standardized pre-tested clinical form. This form recorded demographic data, disease history, and previous treatment. Children were classified according to the algorithm of IMCI including the general danger signs: serious illness; signs of severe dehydration; signs of moderate/severe dehydration; pneumonia and nutritional status. Time and status at discharge were recorded from the hospitals forms.

Children diagnosed with hypoglycemia were treated with intravenous bolus administration of 20 mg/kg of 10% dextrose and the treatment for their baseline disease was started as early as possible. Children warranting hospitalization were sent to the pediatric ward for further treatment according to hospital guidelines.

In-hospital case fatalities were recorded prospectively for all patients admitted during the study period. Patients were discharged when the medical team considered them fully recovered. Children with aggravated status, who were discharged according to the wishes of their parents, and were not expected to survive, were recorded as additional case fatalities.

Children were weighed with 100 g precision and measured (length below 2 years, height above) with 1 mm precision. Nutritional Z-scores were calculated for children under the age of 5 years, using WHO software for anthropometrical Z scores. Malnutrition was defined as moderate or severe if one of the Z-scores was below −2 or −3 SD, respectively.

Level of consciousness was appreciated using the score routinely used in the ward: Blantyre Scale for children under 5 years, or the Glasgow Coma Score for those over.

Definitions

Severe illness was defined according to IMCI standards: presence of any general danger signs; inability to drink or drinking poorly; vomiting; convulsions; lethargy or unconsciousness; pneumonia or severe pneumonia; severe dehydration; some dehydration; persistent diarrhea; severe persistent diarrhea; very severe febrile disease; and severe malnutrition.

Children were categorized into four groups according their blood glucose levels. The intervals of glucose blood levels were previously reported in the literature:

• Hypoglycemia: <2.2 mmol/l (<40 mg/ dl) [30].
• Low glycemia: 2.2–4.4 mmol/l (40–79 mg/ dl) [15].
• Normoglycemia: 4.4–8.3 mmol/l (80–149 mg/ dl).
• Hyperglycemia: over 8.3 mmol/l (≥150 mg/ dl) [17], [19].

Abnormal blood glucose, or dysglycemia, was defined as any of the 3 categories which differ from normoglycemia.

Laboratory measurements

As soon as it was possible after arrival, and after informed consent, 0. 6 µL of blood was collected by investigators through a finger prick to measure the blood glucose concentration (Accu-Chek® Performa glucometers from ROCHE Laboratories, with a sensitivity limit of 1 mmol/l according to the manufacturer). After every twenty-fifth measurement, a quality control by Accu-Chek® was performed. Blood glucose concentrations were recorded in mmol/l (conversion to mg/dl by multiplying by a factor of 18).

Outcome

The main outcome was the prevalence of dysglycemia. Secondary outcomes were: IMCI status; deaths; proportion of deaths within 24 hours; and time before discharge.

Statistical analysis

Data was entered in EpiData freeware. All records were cross-checked with the original data sheets. Analyses was carried out with STATA, Version 8 (Stata Corporation, College Station, TX, USA) and eventually re-checked with free software Open Epi software (http://www.openepi.com/Downloads/Downloads.htm).

Chi squared or Fisher's exact tests were used to compare categorical variables as appropriate. Data distribution was graphically evaluated using the kemel density estimate and eventually tested with the Skewness and Kurtosis test and the Shapiro-Wilk test for normality. Kruskal-Wallis test was used for non-normally distributed variables.

Relative risks (RR) were calculated with Taylor series 95% confidence intervals. We considered p <0.05 as statistically significant.

Children were classified into four glycemia categories (hypoglycemia, low glycemia, normoglycemia and hyperglycemia) and into 2 IMCI categories (severe, non-severe). All dysglycemic groups were compared to normoglycemic children. Associations between admissions glycemia and case fatality (died in hospital or taken home critically ill, not expected to survive) were investigated using univariate analysis. Mortality (died in hospital or taken home critically ill, not expected to survive) was evaluated in univariate analysis according to: child's gender; age (less or more than 3 years); duration of hospitalization; IMCI severity of illness; associated severe illness; dysglycemic status; residential area; mother's age; education; occupation; and daily expense on food (used as an estimated proxy for poverty). Variables with a degree of significance below 0.2 for mortality were then fitted into a multivariate logistic regression model with backward step-by-step analyses using odd ratios (Stata 8) to evaluate the factors associated with child in-patient mortality.

We have attempted to report the study according to the STROBE guidelines [31].

Hypoglycemia prevalence, detection and treatment

The overall prevalence of hypoglycemia in this non-malaria setting of Madagascar was lower than in some other studies regarding children with malaria [8], [11]–[13], [14], [32] but was similar to that previously observed by us in a high malaria-endemic area in Mali (3%) [3].

Though it is difficult to compare health facilities between different countries, endemic malaria enhanced the risk of hypoglycemia and may explain, at least partially, the differences observed [10], [13]. During malaria, parasitized erythrocytes consume 30 to 75 times the quantity of glucose that non-infected cells require and endogenous production of glucose is unbalanced by its over utilization [11].

Hyperinsulinemia after quinine treatment is a well-established iatrogenic cause linked to increased mortality [12], [33], [34], [35].

In addition, hypoglycemia in children is independently associated with poor outcome in the tropics. It was reported during malaria episodes [14], infections [11], pneumonia [5], diarrhea [11], [36], malnutrition [2], and intoxication [1], [37]. It is aggravated by fasting and vomiting [9].

Time between the last meal and unconsciousness was associated with hypoglycemia. Prolonged duration of illness and the delay of referral before admission increased the fasting period. Younger children and neonates are particularly susceptible to hypoglycemia, both in tropical and western countries [38]. Compared to adults, children have a limited tolerance to fasting because glycogen storage capability is less and therefore they are able to maintain a normal plasma glucose level for a fasting period of 12 hours only [39]. Anorexia induced by the disease, and traditional habits shorten the tolerance to fasting.

The etiology of hypoglycemia is not completely understood and is likely to be multifactorial. During acute infectious diseases, hypoglycemia may result in reduced calorie intake and stress-induced cytokines. Glucose metabolism is similar in children and adults in terms of glycogenolysis and glycolysis. However, carbohydrate stocks are limited in children and glucose storage is depleted by starvation. Parasite utilization or cytokine-induced impairment of gluconeogenesis has been implicated in the mechanism of hypoglycemia. During critical illnesses, the physiological responses induced by hypoglycemia may be frequently impaired. Prolonged hypoglycemia may induce biological toxicities, such as increased systemic inflammatory response, cerebral vasodilation, and neuroglycopenia, resulting in neuronal excitotoxicity and cellular damage.

Estimating hypoglycemia at the bedside is particularly difficult due to the lack of specificity of the clinical signs such as the usual vegetative symptoms observed in healthy children [13]. Critically ill or comatose children do not develop specific symptoms despite hypoglycemia levels lower than 1.2 mmol/l. Moreover, hypoglycemia also remains undetected biologically because of limited laboratory resources. Rapid bedside glucose tests are seldom available but should be considered a crucial first-aid element, in the management of a sick child [40]. Despite concerns at the underestimation in cases of moderate to severe anemia and overestimation of hypoglycemia for glucose levels below 3 mmol/l, [30] the usefulness of bedside tests is essential in areas with limited resources and in emergency units. The glucometer was found to be easy to use by health workers, and sensitive enough for blood sugar determination in emergency cases for confirming the diagnosis [41].

Early correction of hypoglycemia is an important therapeutic measure as it is associated with increased mortality though more research is needed to show evidence that systematically preventing it could reduce child mortality in the tropics [12], [17].

WHO guidelines now recommend intravenous treatment of ‘symptomatic hypoglycemia’ using 10% glucose, with continued monitoring and feeding or continued glucose infusion thereafter [22]. However, IV therapy is only feasible where trained personnel and essential supplies are available. Longer delay in initiating resuscitative measures is a contributory factor to early death and should be addressed in tropical pediatric health facilities. In our study, more than 50% of the deaths occurred during the first 24 hours after admission and 14 others died before being able to enter the pediatric ward. This high mortality could be explained by the frequent non-availability of emergency medications in the pediatric wards, or by excessive delay in providing care at the referral hospital [29]. In many hospitals, as here, parents have to buy all necessary medical equipment and prescribed medications from a pharmacy, losing precious time in the race for the survival of their children. Attempts to keep emergency medications in the facility are regularly hindered by financial constraints.

In resource-constrained settings where administering intravenous dextrose is delayed or impossible, there is increasing evidence that sub-lingual administration of sugar could be an immediate and effective ‘first-aid’ measure [3], [15]–[16].

Low glycemia prevalence and prognosis

Low glycemia was the most common blood glucose dysregulation (20%) and was associated with a poorer IMCI status than in children with normal blood glucose. Contrary to the findings in our previous study in Mali, low glycemia was not associated with increased mortality suggesting the additional role of malaria as a factor of mortality there [17]. These children were older than those with hypoglycemia, suggesting a better tolerance to the illness. Furthermore, the children who died in this low glycemia group were as young as those in the hypoglycemia group. The frequency of vomiting and the duration of the fasting period were more important than in the normoglycemic group, contributing to the decrease of blood glucose concentration. This suggests recommending the continued feeding of severely ill children. This data also suggests that low glycemia should be assessed further in young children as a pejorative factor associated with their deaths.

Hyperglycemia prevalence, prognosis, significance and treatment

There are no definite criteria for diagnosing hyperglycemia at admissions among children without diabetes mellitus. However a strong association was demonstrated between the risk of dying and the likelihood of having a persistent glucose blood level higher than 150 mg/dl in critically ill children [42].

In our study, hyperglycemia, a common dysglycemia (11%) carried an increased risk of death (RR: 2.2, 95% CI: 1–4.7). It was diversely evaluated in previous studies. In Mali, hyperglycemia was associated with fewer deaths and considered a potentially protective mechanism in children with severe malaria [17]. However, hyperglycemia was more often associated as a pejorative aggravating factor in another report [13] and in emergency units [43].

As previously described in Mali, hyperglycemia was associated with more frequent seizures. About 15.7% of our patients with convulsions had a blood glucose concentration greater than 8.3 mmol/l at admission. In an Italian study, 12.9% of children with febrile convulsions had stress hyperglycemia [19].

In an American study in a community pediatric hospital, hyperglycemia was associated with a greater need for intensive care and ICU monitoring but not with increased in-hospital mortality [44]. In a similar study, the frequency of hyperglycemia was evaluated at 14.3% for children admitted to intensive care centers without any association between hyperglycemia and mortality [45].

On the contrary, our results in Madagascar are in agreement with the Kenya study which suggests that the difference in mortality in ICU settings and their centre at Kilifi is based on the lack of ICU support [17], [19], [46]. A higher mortality rate was observed among hyperglycemic (13.0%) than in normoglycemic (5.0%) children, suggesting a greater severity state. The role and proper treatment of hyperglycemia in the tropics should be addressed further.

While it is well-known that hyperglycemia is a possible component of the neuroendocrine response to stress, it is not clear how different stimuli may trigger it. Stress response commonly includes elevations in plasma concentrations of hyperglycemic hormones like glucocorticoids, catecholamines, glucagon and growth hormones resulting in alterations to the metabolism of glucose and other energy substrates. Insulin concentrations may be inappropriately low for serum glucose concentration, or insulin may have diminished receptor responsiveness in seriously stressed patients. Many alterations in carbohydrate metabolism have been proposed including increased gluconeogenesis, depressed glycogenesis, glucose intolerance and insulin resistance consecutive to decreased glucose uptake in skeletal muscle. It is not always clear when an altered metabolic or hormonal state is an appropriate response to stress, or represents a general failure of the body.

The cellular glucose overload induces excessive glycolysis and oxidative phosphorylation with increased production of reactive oxygen species. These highly reactive species lead to increased apoptosis, and consequently, cellular death and organ system failure.

Clinical conditions such as traumatic injuries, febrile seizures and elevated body temperature (>39°C) are often associated with hyperglycemia. It has been suggested that cytokines, commonly increased during febrile diseases like interleukin-1 or tumor necrosis factor, may inhibit insulin release and increase the secretion of cortisol, further explaining hyperglycemia. In association with fever and seizures, hyperglycemia was found to be three times more frequent than that observed in other febrile conditions.

Until recently, hyperglycemia was not routinely controlled in ICUs, except among patients with known diabetes mellitus. Recent investigations have demonstrated that glycemic management with insulin therapy reduces morbidity and mortality in those critically ill. The appreciable frequency (11%) of hyperglycemic children at admission, and the association with an increased risk of death raise concerns regarding the use of such presumptive treatment. Further studies are needed for a better understanding of hyperglycemic outcomes and the necessity to initiate insulin treatment in severely ill hyperglycemic children admitted to pediatric emergency wards in the tropics.

The very short period between admission and death is a common proxy of tropical pediatric wards where, unfortunately, children are taken too late and in conditions beyond treatment. The fact that 18 children died before being admitted, emphasize this problem. Part of the solution lies in early diagnosis, triage and treatment of severely ill children. This is dependent on improved diagnostic tools such as IMCI classification and rapid tests, improved quality and accessibility to primary care, and the efficacy of the referral system [47].

Limitations

This study presents some limitations. It was performed in field conditions within a busy pediatric ward of a university hospital. Also, children admitted outside the duty hours of the investigators could not be included. However, using periods of 48 consecutive hours probably decreased a potential selection bias, and results regarding the severity of patients are very similar to those presented the previous year. The 14 children who died on arrival were not assessed. This may have underestimated the rate of hypoglycemia, and partially explains the differences with other studies.

Comparisons with other studies are difficult due to the different thresholds used for hypoglycemia or hyperglycemia. Additionally, health facilities and access to health care differ widely within the areas.

Conclusions

Hypoglycemia and hyperglycemia are associated with a high risk of mortality for children admitted to general pediatric hospitals in non-malaria areas. The pejorative prognosis of low glycemia was observed only in young children. These findings underline the need for the use of rapid screening tests to initiate early treatment. Alternative treatments such as oral or sublingual administration of glucose should be developed in structures with limited resources.