Statistical Modelling for Time Trends of Healthcare-Associated Infections

• Main Authors: Ruei-Fang Wang, 王瑞芳
• Other Authors: 陳秀熙
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/82199762261873312959

博士 === 國立臺灣大學 === 流行病學與預防醫學研究所 === 103 === Background Systematic evaluation of time-series factors (time trend, seasonal variations, and autocorrelation) and factors responsible for heterogeneity accounting for healthcare-associated infections (HAIs) plays an important role in the surveillance of HAIs, particularly for evaluation of the efficacy of interventions and the outbreak of pathogens probably due to drug-resistance. However, forecasting for the long-term dynamic evolution and evaluation of the efficacy of interventions is often confronted with a series of methodological issues if the conventional time-series model is applied, including non-Gaussian data, stationarity and invertiblity, hierarchical data structure, and heterogeneity beyond time-series factors. Aims By using a longitudinal follow-up time-series data on HAIs from a medical center, my thesis aimed to, from the practical aspect of HAIs control, (1) identify the risk factors responsible for HAIs incidence; (2) elucidate how time-series factors such as time trend, seasonal variation, and autoregressive order made contribution to incident HAIs using the conventional time series model and the extended Poisson time-series model for the overall HAIs, site-specific, pathogen-specific and department-specific HAIs; (3) to forecast the evolution of HAIs making allowance for time-series components as identified in (2) and heterogeneity contributed from other covariates such as age and gender with 95% confidence interval; (4) to evaluate the efficacy of interventions related to HAIs control in the site-specific, pathogen-specific, and department-specific reduction in HAIs. My thesis also aimed to, form the aspect of methodology, (5) to develop a novel generalized linear mixed ARIMA model to achieve the objective (3); (6) devise a time-series model-based design together with the proposed model in (5) to evaluate the efficacy of interventions associated with HAIs in the absence of randomized controlled trial as mentioned in (4) . Data Sources: A cohort of healthcare-associated infections was followed during the period of January 1, 1994 and December 31, 2013 in an urban tertiary medical center in northern Taipei with 921-bed and approximately 27,000 inpatient admission annually. Intervention programs indicators: Intervention of PDCA, Hygiene programs, Taiwan Centers for Disease Control (CDC) National Hand Hygiene Campaign and the urinary tract infection quality improvement program of Taiwan Joint Commission on Hospital Accreditation (TJCHA) called CDC/TJCHA, and Bundle care program. Study Design: The first part of study design was in the light of an incident follow-up cohort over time to identify the HAI episode. There are two study designs proposed for evaluation of efficacy of these intervention programs. The first is based on before and after comparison of counts of HAIs. The estimated HAIs counts that were computed on the basis of the posterior distribution with the same length of period conducted with the intervention program were compared with the observed HAIs after the intervention program. The second study design was based on a pseudo randomized controlled trial design. The HAIs counts in the observed were compared with the control group created by predicting rather than estimating the HAIs counts based on the predictive distribution formed by the posterior distribution estimated from the time series data before interventions (i.e. the year before 2005). Model Specification The analysis framework began with the conventional time-series model including decomposition method and Bayesian dynamic linear model and then step-by-step developed the proposed Bayesian linear mixed autoregressive moving average model, combining with for forecasting the long-term time trend of HAIs based on the empirical data presented here and also for evaluation of the efficacy of intervention programs. Results and Conclusions As far as factors affecting the occurrence of HAIs are concerned, the summary of results and conclusions consists of the following points: (1) The elderly males are more likely to be susceptible to HAIs than the young female by using demographic features. (2) The most frequent infection sites are UTI and bacteremia and there is much variation of HAIs across departments. (3) There was much preponderance in summer but less in winter seasons, a decreasing time trends with linear and non-linear (quadratic and cubic) pattern, the consideration of autoregressive orders depending on the site of infection and pathogens. Regarding the efficacy of intervention, the summarized findings and conclusions were as follows. (1) Around 26% and 39% reduction resulting from CDC/TJCHA and Bundle care program, respectively, after 2010 were estimated with adjustment for age, gender, time trend, seasonal variation, and third-order of autoregressive order. However, there was a 10% non-significant reduction for hygiene program and lacking of significant benefit for PCDA. (2) The 36% reduction resulting from time lag (6 months) of either CDC/TJCHA or Bundle care program after 2010 was estimated with adjustment for age, gender, time trend, seasonal variation, and autoregressive order. (3) The similar findings on (1) were found when random-effects considering the hierarchical structure of department, infection site, and pathogen were allowed. (4) The results of efficacy of CDC/TJCHA and Bundle care varied with site of infection. CDC/TJCHA was conducive to 36% reduction in HAIs for bacteremia and SSI, 16% for UTI, 81% for others but there was lacking of any benefit for pneumonia. Bundle care was conducive to 37% reduction in HAIs for bacteremia, 44% for SSI, 38% for UTI, 88% for others but only 3%for pneumonia. (5) The reduction in HAIs for CDC/TJCHA was the greatest in emergency department (almost 94%) and the least in pediatrics (7%). There was lacking any benefit for oncology. The reduction in HAIs for Bundle care was the greatest in infection department (almost 77%) and the least in surgical (34%). There was lacking any benefit for oncology and pediatric department. (6) The results of efficacy of CDC/TJCHA and Bundle care largely varied with pathogen. The reduction in HAIs with CDC/TJCHA was the greatest for anaerobic pathogen (65%), followed by Gram-positive (31%) and Gram-negative (30%), but smallest for Fungi pathogen (5%). There was lacking of any benefit for other pathogens. The reduction in HAIs with Bundle care was the greatest for others (91%), followed by anaerobic pathogen (82%), by Fungi (52%), Gram-positive (34%), and Gram-negative (31%). Regarding the novelty of methodology, there are two parts pertaining to the novelty of methodology presented in this thesis, the development of a Bayesian generalized linear mixed ARIMA model and the model-based design for evaluation of the efficacy of intervention dispensing with the randomized controlled trial. Specifically, this thesis developed a generalized linear mixed effect ARIMA model by combining the generalized linear mixed model widely used in longitudinal follow-up study and ARIMA model widely used in economic studies. It can be useful for monitoring the episodes of HAIs by projecting time-series-featuring HAIs with the relevant parameters estimated by Bayesian approach making allowance for both properties of heterogeneity and time series components. The thesis has devised a time-series model-based design to evaluate the efficacy of intervention associated with HAIs in the absence of randomized controlled trial. Such a time-series model-based design is very flexible in the evaluation of any kind of evaluation of intervention in association with HAIs without needing a randomized controlled trial design.