貝氏卜瓦松自我迴歸時間序列模型於創傷後壓力疾患應用

Bayesian Poisson Autoregressive Time Series Model for Posttraumatic Sress Disorder

王若青(Jwuo-Chin Wang)

指導教授 ： 陳秀熙

國立臺灣大學/公共衛生學院/流行病學與預防醫學研究所/碩士(2012年)

https://doi.org/10.6342/NTU.2012.10918

摘要

無資料

關鍵字

貝氏 ； 時間序列模型 ； 創傷後壓力疾患

並列摘要

Background Posttraumatic stress disorder (PTSD) attributed to non-disaster events has increasingly gained attention but only few study of PTSD was based on community-based samples. PTSD in the underlying community is poorly understood regarding the incidence, prevalence, time trend, seasonal variation, and past history of outcome. It is also unclear whether other putative factors are responsible for incident PTSD cases after controlling for time series factors.

Objectives Annual and age-and gender-specific incidence and prevalence of PTSD were estimated and compared. Bayesian time series model was used to model seasonal variation, time trend and autoregressive order together with demographic factors and other putative factors.

Method We used a five-year longitudinal follow-up retrospective cohort spawned from a community-based integrated screening program to identify incident PTSD cases. The autoregressive order was determined and Bayesian Poisson Autoregressive Time Series Model was applied to modeling time trend, seasonal variation, and autoregressive order (past history of PTSD) by using Bayesian Monte Carlo Markov Chain simulation method. We used two different approaches, including dummy variable and trigonometric function, to model the seasonal variation of PTSD. Statistical significance of time series component was assessed by a series of nested models using DIC criteria and mean squared predictive error. Risk factors accounting for incidence of PTSD were analyzed with univariate and multivariate models.

Results The annual incidence of PTSD was 0.08%, which is 10 times of the prevalence rate at baseline year (1999). Considering second order autocorrelation in the time series model, seasonal variation and time trend were statistically significant. For demographic characteristics, age, marital status, and education level were highly associated with incidence of PTSD. Significant risk factors included cardiovascular heart disease (adjusted relative risk (RR)=1.39，95% confidence interval (CI): 1.03-1.85), major depression (adjusted RR=2.67，95% CI: 1.70-4.14) and dysthymia (adjusted RR=7.52, 95% CI: 5.49-10.17).

Conclusions Prevalence and incidence rate of PTSD was estimated on the basis of a community-based cohort data. Bayesian time-series model was applied to ascertaining significant time trend, seasonal variation, and autoregressive order of PTSD. This model is also useful for identifying individual attributes related to PTSD, such as age, marital status, education, cardiovascular heart disease, major depression and dysthymia.

並列關鍵字

Bayesian ； Time Series Model ； PTSD

參考文獻

1. Lai, T.J. and C.M. Chang, Full and partial PTSD among earthquake survivors in rural Taiwan. J Psychiatr Res, 2004. 38(3): p. 313-322.

2. Solt, V., C.-J. Chen, and A. Roy, Seasonal Pattern of Posttraumatic Stress Disorder Admissions. Comprehensive Psychiatry, 1996. 37(1): p. 40-42.

3. Yehuda, R., Post-traumatic stress disorder. N Engl J Med, 2002. 346(2): p. 108-114.

4. Kramer, T.L. and B.L. Green, Posttraumatic stress disorder as an early response to sexual assault. Journal of Interpersonal Violence, 1991(6): p. 160-173.

5. APA, Diagnostic and Statistical Manual of Mental Disorders(4th edn) (DSM-IV)1994.