Background: Statistical models for analysis of correlated count data are important for answering epidemiological questions that involve taking individual count measurements repeatedly over time through the use of longitudinal studies. Conventional regression models for this type of data are inadequate, leading to improper conclusions and inference. An important application of longitudinal studies in Public Health is the evaluation and monitoring of patients with infectious diseases, such as HIV/AIDS, to determine their health status, to verify the treatment effects, and to make prognosis concerning the evolution of the disease, including interdependencies of clinical manifestations. The purpose of this article is to characterize different statistical strategies for analysis of longitudinal count data, emphasizing how to choose the most suitable model for the data and how to interpret the results.
Methods:We illustrate their applicability by evaluating the effect of associated factors on lymphocyte CD4+T cell count in HIV seropositive patients in Salvador/Bahia – Brazil. We describe Poisson and Negative Binomial models using multilevel (ML) approach and generalized estimations equations (GEE) for analysis of longitudinal count data.
Results: It is worth noting that the interpretation of the results from ML and GEE differs and they should not be compared directly.
Conclusion: We believe that the statistical methodology for analysis of longitudinal studies with correlated count data can be useful to address several important questions in public health, particularly by helping to monitor patients and checking the effectiveness of treatments.