Pediatric respiratory disease is a major cause of morbidity and mortality in the developing world with 70% of the global deaths occurring in Africa and Southeast Asia [1]–[4]. In Kenya, 16% of annual deaths in children <5 years of age are attributed pneumonia [5].

Lack of early and accurate detection of high-risk patients contributes to preventable complications and deaths associated with respiratory illnesses [6]. The limited diagnostic and treatment options available in most hospitals in developing countries have led to the development of syndrome-based guidelines for care. In 1999, the Kenyan Government introduced the Integrated Management of Childhood Illness (IMCI) to standardize and improve treatment for major causes of death in children <5 years [7]. Although implementation of IMCI remains suboptimal in Africa [8], IMCI includes guidelines to help health care workers identify “danger signs” for children at risk of severe illness who should be hospitalized [9].

Whereas IMCI seeks to identify young infants and children in need of hospitalization in developing country settings, no guidelines exist to help health care workers identify children who are at higher risk of dying from respiratory infections within the hospital setting. Recently, Reed and colleagues developed a Respiratory Index of Severity in Children (RISC) scoring system to quantify the severity of pediatric pneumonia for HIV positive and HIV negative children aged <24 months in South Africa [10]. This system has not been validated in other lower-resource countries in tropical sub-Saharan Africa. We used a similar methodology to create a modified Respiratory Index of Severity in Children (mRISC) scoring system for children under age five who were hospitalized with a respiratory illness in a government referral hospital in western Kenya.

During August 2009 to July 2012, 3974 children aged <5 years were hospitalized at SDH with SARI, representing 56% of all hospitalizations among children under age 5 at SDH during this time period. Ninety percent (3581/3974) of children had complete data on the variables of interest and were included in the analysis (Figure 1). When the 393 excluded patients were compared with those who were included in the analysis, no significant differences were seen in terms of demographic characteristics or mortality. Death was the outcome for 218 (6%) patients. The median length of stay for all SARI admissions was 3 days [inter quartile range (IQR) = 2–5 days] but was significantly shorter for those who died [median = 1 day; p<0.001]. Most (75%) of the children aged <5 years and admitted with SARI were aged <2 years of age. Although HIV testing was offered to all admitted patients, 1322 (37%) were not tested. Of the 2259 children who had ever been tested for HIV, 198 (9%) were HIV-infected, and among them 130 (66%) had either been enrolled in HIV care (provided cotrimoxazole) or started on ARVs. A total of 2028 (56%) and 388 (11%) of the admissions had IMCI-defined pneumonia and severe/very severe pneumonia respectively. Of the children who died in the hospital, 67 (31%) had IMCI-defined severe/very severe pneumonia upon admission (Table 1).

We found that the mRISC scoring system accurately predicted the likelihood of fatal outcomes of children hospitalized for respiratory illness in Western Kenya using measurements taken at admission. In many Kenyan hospitals, where tools like radiography and laboratory testing are not commonly available, the mRISC score may prove a practical and objective tool for health care workers (Table 4) by helping to better guide decisions surrounding level of admission ward (i.e., high dependency units or intensive care unit vs. regular or low risk wards) and the prioritization of the use of limited resources such as high flow oxygen. The mRISC could also serve as a complementary tool to the IMCI guidelines, which are intended to help clinicians determine whether or not children need to be hospitalized, but which have less discriminating power for use in triage of hospitalized patients. Indeed, this is supported by the finding that while 17% of children classified as having severe/very severe pneumonia according to IMCI guidelines died in the hospital, 40% and 80% of children admitted with mRISC scores of ≥3 and ≥6, respectively, died in the hospital.

Sub-optimal implementation of IMCI standards, coupled with fairly routine admission of children with uncomplicated malaria or recent Mycobacterium tuberculosis infection (TB), can together serve to lower the threshold for hospital admission and complicate patient triage and management within the hospital setting. This increases the chances of misclassifying subsets of children at low risk of severe outcomes. Such hospitalizations would take up resources that should have been focused on severe cases and therefore further underscores the need for a complementary tool such as the mRISC to help guide in-hospital triage. A recent study conducted in Kenya showed that clinicians were not universally following IMCI guidelines; 80% of the health workers had not been trained; and community members faced financial barriers in accessing services [8]. In Pakistan, even in situations where admissions with severe pneumonia as defined by IMCI were made, some of these children could have successfully been managed from home [23].

Our finding of the negative association between laboratory-confirmed malaria and risk of mortality among children hospitalized with respiratory illness suggests that these children may have had a lower threshold for admission and less severe disease and therefore less likely to die. This negative association is consistent with findings from a recent study conducted within the same area [24]. While this may highlight the value of rapid diagnosis of malaria, the same finding may not hold true in an area where malaria diagnosis is poorly sensitive or specific [25]–[27].

Although other studies have demonstrated increased mortality among HIV-infected patients with respiratory illness [28]–[30], we did not find this in our study. This could in part be explained by the fact that most of the patients who were HIV positive had either been started on cotrimoxazole or were on ARVs. Alternatively, this may be because we detected relatively few HIV-infected children, and this may have reduced our power to detect a difference in mortality. Also in contrast to other studies, low-oxygen saturation was not associated with increased mortality in the multivariable model. While unexpected, this makes the mRISC score more feasible to use in most health facilities in Kenya, where working pulse oximeters are not available. The observations that chest wall in-drawing, inability to drink or breastfeed and admission diagnosis of dehydration were only associated with mortality among children aged <2 years could in part be explained by the fact that these measures are most commonly made during the evaluation of very young children.

The study had several limitations. First, this study utilized data collected from only one site and as such further evaluation and external validation at other sites in Kenya should be undertaken. Secondly, it also possible that some of the children who were discharged alive may have had other serious complications or even died shortly afterwards, and are therefore misclassified in this analysis. Third, we did not include radiographic measurements in our analysis; it is conceivable that chest radiographic findings would have added discrimination if included in the scoring system, although this was not the case in South Africa [10]. Finally, as much as we adjusted for the type of drug treatment received and participation in clinical studies, other differences in care received could have be an unmeasured source of confounding.

In conclusion, this study shows that this adaptation of the RISC score initially developed in South Africa, may also have practical utility to rapidly identify children most at risk for fatal outcomes due to respiratory disease in rural Western Kenya. As a complementary tool for use alongside the IMCI guidelines, the mRISC could be useful in the triage of children with respiratory illness in the admission process and help improve their clinical management, and perhaps also serve as a standard measure of severity for use in epidemiological studies.