The application of Poisson hidden Markov model to forecasting new cases of congenital hypothyroidism in Khuzestan province

Abstract

Background & Aim: Congenital hypothyroidism (CH) is one of the most common endocrine diseases and is a major cause of preventable mental retardation. Early diagnosis of CH can help prevent future diseases. Although time series techniques are often utilized to forecast future status, they are inadequate to deal with count data with overdispersion. The aim of this study was to apply Poisson hidden Markov model to forecast new monthly cases of CH disease. Methods & Materials: This study was based on the monthly frequency of new CH cases in Khuzestan province of Iran, from 2008 to 2014. We applied stationary Poisson hidden Markov with different states to determine the number of states for the model. According to the model, with the specified state, new CH cases were forecast for the next 24 months. Results: The Poisson hidden Markov with two states based on Akaike information criterion was chosen for the data. The results of forecasting showed that the new CH cases for the next 2 years comforted in state two with the frequency of new cases at 6-18. The forecast mode and median for all months were 12 and 13, respectively. Each state is explained by each component of dependent mixture model. Conclusion: Our estimates indicated that state of frequency of CH case is invariant during the forecast time. Forecast means for the next 2 years were from 13 to 14 new CH cases. Furthermore, forecasting intervals were observed between 7 and 25 new cases. These estimates are valid when the general fertility rate and crude birth rate were been fixed.

Keywords

• Congenital hypothyroidism
• Poisson hidden Markov
• forecast
• Khuzestan