Forecasting Pneumonia Mortality in Children Under-5 at Kiambu County Referral Hospital 2019-2021

Abstract

Pneumonia has been a menace since the 20th century. By 2000, approximately 156,000,000 pneumonia cases occurred each year, with 151-million reported in developing countries. The cases in Sub-Saharan Africa were 35-million (Rudan 2008). Approximately 1,900,000 under-5s die yearly where half of this occurred in Africa (William 2002). By 2010, the pneumonia cases globally were approximately 120-million with 880,000 deaths occurring yearly, but the countries that contributed to the highest deaths were 15, with Kenya being among them (Biloglav Z 2008). Kenya had 30,000 under-5 deaths in 2008 caused by pneumonia alone. It was the second killer of under-5 children by 2012 (Black RE 2010). 2015, is when the cases had decreased to 102-million globally, and the deaths had reduced to 700,000 (McAllister DA 2019). It was still the second cause of death in 2017 among children under-5 in Kenya. Kenya’s pneumonia under-five mortality in 2018 was 15%. That is, the deadly disease killed almost 9000 children in 2018. This translates to more than one child dying per hour (UNICEF 2019). To model the past values of pneumonia mortality and forecast the future mortality of children under-5 at Kiambu County Referral Hospital using an ARIMA model. The main aim was to use an ARIMA model with this data; thus, the hypotheses formulated were to help identify the appropriate model. The hypothesis for the data at the Kiambu County Referral Hospital was 1. Pneumonia mortality ratio doesn’t have a constant mean and consistent variance, not stationary. versus Pneumonia mortality ratio has a constant mean and consistent variance, and is stationary. 2. Pneumonia mortality ratio has no trend. versus the pneumonia mortality ratio has a trend. 3. Pneumonia mortality ratio doesn’t follow an ARIMA (p,d,q) model for forecast. versus Pneumonia mortality ratio follows an ARIMA (p,d,q) model for forecast. ARIMA (2,1,1) model with AIC 206 was used to achieve the objective. For the ARIMA model to be realized, the Dickey-Fuller test was used to test for a unit root. This test had a p-value 0.4469 > 0.05, implying data was not stationary. Differencing was done once; therefore, d=1. ACF and PACF were used to form the order of p and d in p,d, and q. A simple moving average of n=8 revealed the trend was downward in 2019, upward and downward in 2020, and has been upward since 2021. Out of 217 deaths, boys were 89 (41%), and girls were 128 (59%). The age group where most deaths occurred was 0-12 months, with 81.6%, while the other 13-36 months and 37 to ≤ 60 had 15.2% and 3.2%, respectively. The forecast for May-July was done, predicted, and the actual values were within the confidence interval. ARIMA (2,1,1) had MASE < 1; thus, the model gave an acceptable forecast. Pneumonia mortality was found to be increasing from the ARIMA (2,1,1) forecast, even if there was a pandemic going on from May to December 2022.