泌尿系統醫療照護相關感染之時間序列分析

Abstract

前言 醫療照護相關感染長時期的資料型態依時間為軸排序可呈現出趨勢、季節、循環等成分。尤其是醫療照護相關感染最常見的泌尿系統部位，描述這樣資料型態的趨勢及季節型態是相當重要的，此外，此資料型態亦可能存在不同時間點觀察值之間的自我相關性，增加了描述上的挑戰性。 目的 為了描述醫療照護相關感染的資料型態，應用一系列系統時間序列來描述之。呈現菌種發生次數及發生率對於趨勢、季節、循環演變，找尋菌種發生次數及發生率以時間為軸的自我相關現象，進而對未來主要菌種發生次數及發生率做出預測。 材料與方法 自1994年至2009年6月期間，以在台灣台北新光吳火獅紀念醫院感染控制中心由醫師判斷收集之醫療照護相關感染病患族群做分析。此研究收納其中泌尿系統部位之醫療照護相關感染病患進行研究。收集資料包含有:(1)病患背景資料;(2)患病資料:住院出院日期、病房;(3)感染病原菌種資料:菌種、採檢部位、菌種抗藥性敏感試驗;(4)侵入性治療處置:放導尿管、使用中心靜脈營養、置放中央靜脈導管、手術；(5)全院之住院人數計數及人日數。將一個月做為最小單位之發生感染次數及感染發生率依菌種做區分，分別使用分解法、Durbin-Watson統計量、以及Box-Jenkins模式分析，並使用模式化之方法進一步做未來一年之感染預測。 結果 自1994年至2008年期間，此醫學中心共收集有6519人次的泌尿道系統醫療照護相關感染。就菌種分類而言感染累計前四名，依序是酵母菌、大腸桿菌、綠膿桿菌，腸球菌。以革蘭氏陰性菌占多數，為61.7%(4024/6519)。時間序列分析顯示分解法在酵母菌、大腸桿菌、綠膿桿菌及腸球菌發生上顯示出具有季節差異、趨勢的變化。其中綠膿桿菌感染發生隨時間呈現下降的趨勢，而其他主要三種菌種則呈現上升趨勢。Box-Jenkins模式在酵母菌及大腸桿菌之感染發生次數及發生率均具有級數一之移動平均模式，綠膿桿菌時間序列為隨機模式，腸球菌在感染發生率上為級數二即級數四之自我回歸模式。 結論 本研究發現根據感染之菌種不同，時間序列分析法能呈現出不同的趨勢、季節、循環模式、及自我相關性模式，依據感染事件及發生率過去的變化型態是建立個別菌種之感染預測模式，以及作出未來的預測相當有用的方式。

Background The time-series longitudinal incidence data on healthcare-associated infection (HAI) can reveal its trend, seasonal, and cyclical components. As far as the urinary tract system, the most common type of healthcare-associated infection, is concerned modeling trend and seasonal components is therefore of great interest. However, the time-series data are often challenged by autocorrelations between successive time measurements. Objectives This study aimed to utilize a combination of time series models to analyze the long-term healthcare-associated infection time series data, to elucidate the trend, seasonal, cyclical patterns, making allowance for autocorrelations, and to forecast future events and incidences of each major urinary system HAI pathogens. Material and Methods Those who were diagnosed as healthcare-associated infection during the period of January, 1994 and June, 2009 in Shin-Kong Wu Ho-Su memorial hospital in Taipei, Taiwan were included for analysis. This study focused on the major pathogens of healthcare-associated urinary tract infections, such as yeast, Escherichia coli, Pseudomonas aeruginosa, and Enterococcus species. Information used for the study includes (1) patients background; (2) infection data: admission and discharged date, ward of getting infection; (3) pathogen classification: species, sampling sites, antibiotics sensitivity tests; (4) invasive procedures: urinary catheter usage, parenteral nutrition via central line, central venous catheter usage, and surgery; (5) whole hospital admission patients number and person-day of the admitted patients. The infection events per month and the incidences time series of different pathogens were analyzed using decomposition methods, Durbin-Watson statistics, and Box-Jenkins model. We also projected the outcome of the coming year by using these models.

Results There were 6519 infection events during the inclusion period of 15.5 years in the Shin-Kong Wu Ho-Su medical center. The first four ranks of urinary tract healthcare-associated infections in order were yeast, Escherichia coli, Pseudomonas aeruginosa, and Enterococcus species. The Gram-negative bacteria were the most common pathogens, accounting for 61.7%(4024/6519) of all urinary system HAIs。 Time series analysis of decomposition methods revealed the remarkable trend and seasonal patterns in these four major pathogens. The time trend of infection incidence and events of Pseudomonas aeruginosa has been decreasing, whereas those of the other three major pathogens have been increasing with time. Box-Jenkins models of both yeast and Escherichia coli had first-order auto-regression, and those of Enterococcus species had second-order and fourth-order auto-regression (time lag, 2 and 4 months). The time series of Pseudomonas aeruginosa are shown in stochastic process.

Conclusion We observed that the trend, seasonal, cyclic patterns, and autocorrelations were different with respect to different infectious pathogens associated with healthcare-associated urinary tract infections by using time series analysis. The historical patterns of the events and the incidences patterns are very useful for model construction and forecasting of such time-series data.