Pulmonary Tuberculosis Diagnosis Analysis through Dynamic Regression Models

Abstract

Introduction and objectives: Pulmonary tuberculosis (PTB) is the most common form of Tuberculosis and highly associated with airborne transmission. Time series analysis is a methodology highly valuable in the assessment of health trends and control program performance. The present study applies Dynamic Regression models in order to potentially contribute to a better characterization of the PTB monthly incidence rates in Portugal mainland and the impact of lagged predictor variables. Materials and Methods: Monthly PTB incidence rates (per 100,000 population) for Portugal mainland were analyzed, from 2000 to 2009. Dynamic regression models were employed to study PTB time series behavior and its association with up to six months lagged predictor variables. The predictor variables in the dynamic model were calculated as monthly proportions of TB cases that were alcohol consumers, smokers, drug users, inmates, homeless, HIV positive, community home, a new PTB case, immigrants from a high risk country and belonging to a certain age group. Results: A dynamic regression model with SARIMA(2,1,0)(1,0,0)12 structure for the errors revealed to be the best-fit model to monthly PTB incidence rates. Forecasts were performed for the year 2010. The predicted values were very close to the observed PTB monthly incidence rates for the year 2010. Overall, estimated future projections indicate a persistence of the decreasing trend in year 2010, with a peak in April and trough in December. We have seen that the model with predictor variables is better (in terms of fitting and forecasting) than the SARIMA model without predictors. Discussion and Conclusion: Understanding what characterizes patients who suffer from pulmonary tuberculosis is of great importance when establishing screening strategies to better control tuberculosis disease. Dynamic regression models with ARIMA errors prove to be useful in identifying high-risk groups associated with PTB incidence rates that should be targeted by control programs and in performing accurate predictions. Forecast analysis shown that the past pattern modeled herein will continue into the future. However, such assumption may not be valid in the near future, given that Portugal entered an economic crisis in 2008 which can affect the PTB incidence rates.