We developed and validated a clinical prediction score, the RISC score, which uses a simple set of clinical factors to discriminate between young children with varying risks of death from LRTI. Variables retained in the score represent known risk factors for severe outcomes of respiratory illness in children, including: hypoxemia, chest indrawing, refusal to feed, malnutrition, age, and stage of HIV disease 
. We demonstrate that a simple combination of these factors in a scoring model provides good discrimination and calibration for predicting mortality in young children with LRTI, without the need for radiographic or laboratory measurements.
This severity index has several potential uses in both research and clinical settings. In the research setting this tool may provide a useful means of quantifying the severity of LRTI in epidemiologic studies and clinical trials. In clinical settings, the ability to quantify the risk of mortality may aid clinicians in the management of young children that present with LRTI and compliment current IMCI guidelines. Recent studies have indicated that some children with severe pneumonia by IMCI criteria may be successfully managed at home 
, and that in regions with barriers to hospital referral, limiting referrals to those most in need actually improved the overall number of children who received appropriate treatment and may have reduced mortality 
. The use of the RISC score may help refine decisions about case management in resource-limited countries or areas with geographic and socioeconomic barriers to hospital referral by facilitating decisions about the most appropriate site of treatment (i.e., home vs. hospital) or the need for additional supportive care (i.e., supplemental oxygen or intensive care).
Hypoxemia was an important risk factor among both HIV-infected and non-infected children. While pulse oximeters may not be currently available in all settings, our finding supports several recent studies that have shown the importance of pulse oximetry in guiding the use of oxygen therapy and other supportive care for reducing the mortality in children with respiratory infection 
. In contrast, wheezing was associated with reduced mortality in both models. There have been concerns about the potential for misclassifying pneumonia in children with wheeze, which can also result in rapid breathing and thus an IMCI classification of pneumonia. Studies of children with IMCI-defined pneumonia and wheeze have found deterioration was more likely in children with additional danger signs such as chest indrawing or malnutrition 
. In the absence of additional danger signs, children with wheeze would have a low RISC score, which supports suggestions that in many instances these young children might be successfully managed without hospitalization 
Predictive models are expected to perform better in the population in which they were developed than in other populations 
. We used bootstrapping techniques to estimate the possible optimism associated with the apparent performance of the RISC model in this population. However, while the RISC score includes recognized predictors of severe LRTI, differences in the distribution of risk factors and/or health seeking behaviors in different populations may impact the performance of this model at predicting the risk of death, and will need to be evaluated. Efforts are ongoing to validate these scoring models in other populations of young children with respiratory illness to further define the predicted mortality across the RISC score in a variety of settings. Further study should also focus on evaluating the usefulness and applicability of this tool in different settings 
For children with HIV infection, this study was conducted prior to antiretroviral medication use in South Africa. Antiretroviral medication programs can reduce the incidence and severity of HIV-associated pneumonia in children 
, and the impact on this score is unclear. This highlights the need for further evaluation of this model in other populations to determine how the RISC score may best be applied or refined in other contexts.
In order to develop a score that would have high utility in resource-limited settings, which have the greatest burden of childhood mortality from LRTI, we considered the effect of radiographic and laboratory measurements only after developing a strong multivariable model. In this population, these variables did not provide any additional discrimination after other risk factors in the multivariable model and thus was not included in the final RISC score, but may warrant future consideration.
Finally, these data come from a cohort of children enrolled in a pneumococcal vaccine trial. Although there were no differences in mortality between children hospitalized with LRTI in either the vaccine or comparison group, half of the children did receive pneumococcal conjugate vaccine, and all received Hib vaccine. Additionally, although study investigators passively monitored hospitalizations with LRTI and did not directly participate in their clinical care, mortality observed in a vaccine trial may underestimate the risk of mortality that would exist in populations with less access to care. This underscores the need for further study of RISC in additional populations, to understand how it might be adapted in different settings.
The RISC score incorporates a simple set of variables that discriminate the probability of death in children hospitalized with LRTI. The ability to estimate a child's risk of mortality with limited clinical information may provide the ability to improve outcomes for children in resource-limited settings where the burden of pediatric pneumonia is highest. With further validation in additional populations, the RISC score may be a tool to more effectively manage LRTI, a major cause of death in children worldwide, and to better understand the impact of vaccination or other public health interventions on the severity of childhood respiratory illness.