A time series analysis of HIV linkage to care rates in Kampala District, Uganda, 2015 2023

Abstract

Routine monitoring of the HIV care cascade is critical for evaluating health service delivery, yet routine data are frequently summarized descriptively without formal time series modelling. This study conducted a time series analysis of monthly HIV linkage to care rates in Kampala District, Uganda, using secondary routine data from the District Health Information Software covering the period from January 2015 to December 2023. Following data cleaning which involved the exclusion of the first six months of 2015 due to incomplete reporting an analytic series of 102 monthly observations was evaluated. The primary objectives were to examine the trend pattern, assess seasonal variation, and fit an advanced predictive model to generate short term forecasts for the 2024 calendar year. Decomposition and feature extraction analyses revealed a weak underlying upward trend Ft = 0.142. Furthermore, the seasonal strength score Fs = 0.213 confirmed the absence of meaningful annual seasonality, demonstrating that linkage to care rates do not follow a repeating calendar cycle. To capture the complex, nonlinear dynamics and high variance of the series, a Neural Network Autoregression (NNAR) model specifically an NNAR (1, 1, 2)12 architecture trained as an ensemble of 20 networks was fitted to the data. The validated model projected a dynamic, nonlinear linkage to care rate for the 2024 calendar year, averaging in the mid 80s with monthly point forecasts fluctuating between a low of 82.5 percent in March and a peak of 89.1 percent in May. Wide 95 percent prediction intervals (spanning roughly 64 to 110 percent) successfully accounted for the extreme historical volatility and periodic reporting anomalies, such as episodes where recorded linkage exceeded 100 percent. The findings indicate that Kampala District's healthcare network has reached operational maturity, transitioning from expansion to sustainment. The study recommends shifting broad linkage campaigns toward targeted, localized interventions for hard to reach populations, improving cross month client tracking between testing points and ART clinics to resolve reporting lags, and maintaining consistent year round resource allocation.