落實日常安全管理就能夠預防重大意外事故發生嗎？
Source ：International Journal of Risk Assessment and Management, Vol. 10, Nos. 1/2, 2008
Author ：Michael R. Elliott, Paul R. Kleindorfer, Joseph J. DuBois, Yanlin Wang, Isadore Rosenthal
用1996-2000年間，美國化學工廠為研究對象，用各工廠OSHA的reported occupational illnesses and injuries (OII)代表平常的安全管理績效，對應同一期間EPA（依據Clean Air Act當中的risk management programme (RMP)）的 database 中的accident histories（代表catastrophic loss）
而這個涵義撼動了古典的理論與典範，恐怕會讓一些被Heinrich洗腦的專家們會無法接受吧..
吃這行飯的可以點入繼續閱讀
結果顯示兩者關聯性不高。
Does Everyday Safety Prevent Catastrophic Loss？
作者s 含蓄的認為答案是：No
而這個涵義撼動了古典的理論與典範，恐怕會讓一些被Heinrich洗腦的專家們會無法接受吧..
吃這行飯的可以點入繼續閱讀
閱讀有水準的專業文章真是一種享受....
Abstract:
We link the risk management programme (RMP) database of accident histories collected by the US Environmental Protection Agency for the period 1996–2000 under section 112(r) of the Clean Air Act Amendments and OSHA reported occupational illnesses and injuries (OII) for the same period.
We explore various statistical associations between OIIs and RMP-reported accidents.
If we think of OIIs as reflecting everyday safety performance and RMP accidents as reflecting major accidents, then the analysis can be considered a test of whether good everyday safety performance is a foundation for preventing or mitigating relatively rare major accidents. We find only weak evidence that this is the case for the US
Keywords:
accident epidemiology意外流行學？; chemical accidents; occupational injuries; process safety management.
1 Introduction
Catastrophic chemical process accidents, such as those at Flixborough , UK in 1974 and Seveso , Italy
This ‘new’ paradigm was implicitly incorporated by OSHA (the US Occupational Safety and Health Administration) into its Process Safety Management standard (PSM) in 1992 and explicitly into the Seveso II Directive in the EU and the US Environmental Protection Agency (EPA) in its Risk Management Program regulation (RMP) in 1996.
The RMP regulation required all facilities storing, on-site, any of 77 toxic or 63 flammable substances above a threshold quantity (ranging from 250–20,000 lbs ) to develop a risk management programme (RMP) that included assessments of hazards, a summary of accidents at the facility during the past five years preceding the filing of the RMP, worst-case accident release scenarios and prevention and emergency response programmes (see Kleindorfer et al. (2003) for details on the RMP Rule). At the time these regulatory initiatives were launched, projections were made that these regulations would result in significant decreases in the incidence of process accidents. However, the process accident data available do not appear to support these expectations in either the US
由於化工廠相關事故引發社會關注與官方制定相關法規，當時針對些低可能性、高嚴重度（low probability-high consequence (LP-HC)）的事故，觀念上是認定這些公司在內部管理上一定有些瑕疵，才會導致事故發生。因此EPA在1996參照OHSA的製程安全管理標準PSM與歐洲Seveso II指引制定出”新的管理典範”RPM。
此一風險管理法規(RPM)規範，使用77種列管毒物或63種可燃性物質超過管制量的工廠，必須執行危害評估、事故回顧、最糟事故情境分析與預防/應變計畫assessments of hazards, a summary of accidents at the facility during the past five years preceding the filing of the RMP, worst-case accident release scenarios and prevention and emergency response programmes
預期應該可以有效降低與減少意外事故accident與虛驚事件incident的發生，然而美國與歐洲的記錄資料卻顯示：以上預期落空。
儘管如此業者卻多半認為有效的管理是預防意外事故發生的關鍵，而這些措施之所以無效，是因為它們不是實務有效的措施
most practitioners continue to believe that an ‘effective’ management system is the key to prevention. Such practitioners argue that the less than expected decrease in accident incidence exists because the newly adopted regulations have not resulted in the hoped for adoption of ‘effective’ process safety management systems by industry.
most practitioners continue to believe that an ‘effective’ management system is the key to prevention. Such practitioners argue that the less than expected decrease in accident incidence exists because the newly adopted regulations have not resulted in the hoped for adoption of ‘effective’ process safety management systems by industry.
而學術上要判定這些措施是否有效，則尚需鑑別出哪些才是所謂的有效措施
Testing the validity of this belief requires the ability to define and identify the essential elements of ‘effective’ facility process safety management plans.
因此只能間接的量測與觀察這些措施對於減少小事故(諸如員工的工傷病狀況)，如果這些措施可以有效減小輕微事故的發生，理論上也可以減小重大事故的發生。
Among other issues, it will be important to separate out the effects that a given process safety management system has on everyday safety events from the effects, if any, that such a system might have in preventing or mitigating the consequences of larger events, including catastrophic failures.
本研究想了解工廠日常安全水準(報告給OHSA的員工職業傷病)與重大災害（報告給EPA 的major accident）兩者間的關聯性，進而間接評估與推斷這些管理措施是否”有效”
The main point of the present paper is to examine whether there is any relationship between the performance of chemical facilities on everyday safety and major accidents in the US
關於安全文化氣候與管理系統減少和預測職業傷病已經有相當多的研究，關鍵包含：(Carder and Ragan, 2003; Donald, 1998; Petersen, 2005). Key elements required include:
- management commitment to safety管理階層的安全承諾
- workforce educated and knowledgeable with respect to worker safety勞工工作安全知識的教育
- effectiveness of the supervisory process有效的內部稽核與督導
- employee involvement and commitment.員工參與與承諾
Carder and Ragan的實證研究survey 12 facilities correlated with three year average OII results with a very high level of statistical significance（越多管理問題的廠區，其職業傷病率越高）
然而找出這些低可能性、高嚴重度（low probability-high consequence (LP-HC)）的事故背後所隱藏的深層內部管理問題很困難
Unfortunately, acquiring comparable knowledge on the factors underlying major accidents is very difficult because of the low incidence of LP-HC process accidents.
因此本研究僅能間接量測職業傷病率與重大事故間的關連性，研究動機與觀念是：低嚴重度高發生機率的職業傷病(OII)可視為發生低可能性高發生機率巨災(catastrophic)的預兆The motivating idea here is that the more frequent and lower consequence OIIs could serve as a precursor or prior indicator of the likelihood of low-incidence, high-consequence accidents such as those reported under the RMP rule.進而利用職業傷病率的大小來預測可能發生重大巨災的工廠
備註：這個idea與管理意涵可以補充到個人的paper上，但個人的實證與經驗也否認此一看法（Can’t use Gas/Liquid Leak& fire alarm as proxy or indicator to predict departmental work injury rate. Because operational characteristic dominate safety performance and feature in department level.）
然而有另一派的學者認為：由於管理系統運作與相關經費支出，容易投注於可見常發生的職業傷病與小意外，因此容易忽略對於嚴重度高鋒機率低重大災害的預防（也就是說：OII與Catastrophic間的關連性不高），
Dalzell and Ditchburn (2003) note, for example, that there is a potential “disconnect between the management of occupational safety, health and . . .environmental risk management and the management of major accident hazard risks.”
OIIs measure routine injuries, while LP-HC incidents are rare and contribute little to the OII measures in the routine course of events. Hopkins
例如BP的相關事故
As the Baker Panel Report on the Texas City
2007; p.14):
“BP has emphasised personal safety in recent years and has achieved significant improvement in personal safety performance but BP did not emphasise process safety. BP mistakenly interpreted improving personal injury rates as an indication of acceptable process safety performance at its US
BP強調個人安全，也以工傷率為各廠區安全指標，然而卻忽略了製程安全面
基於以上兩派的爭議，作者建立以下研究假說：
Hypothesis 1能夠有效的降低職業傷病(OII)的管理系統 足夠作為指標 進而預測該廠區是否會發生RPM定義的重大事故
Management system effectiveness in reducing OII incidence is a sufficient indicator of facility management system effectiveness in reducing RMP reportable accidents.
Hypothesis 2能夠有效的降低職業傷病(OII)的管理系統 不夠以作為指標 進而預測該廠區是否會發生RPM定義的重大事故
Management system effectiveness in reducing OII incidence is a necessary but not sufficient indicator of facility management system effectiveness in reducing RMP reportable accidents.
2 Data and methods of analysis
Low-probability, high-consequence event data were obtained from the RMP*Info database of the EPA 1996 RMP regulation. With certain exceptions, all facilities storing on-site at least one of 77 toxic or one of 63 flammable substances above a threshold quantity are required to develop a risk management programme (RMP). The information contained in the RMP*Info database is extensive and includes details about on-site chemicals and processes, regulatory programme coverage, geographic location, number of full-time employees (FTE) and other descriptive information on the facility.
Rates and counts of occupational illnesses and injuries (OIIs) were obtained from the US
Annually, OSHA contacts and obtains these records a minimum of once every four years among facilities with 40 or more full-time employee equivalents (FTEs) who use one or more of 558 NAICS processes (NAICS = North American Industry Classification System, which provides a classification system that differentiates between agricultural, manufacturing, warehousing, wholesale trade and medical facilities).
在2000/12/04 ，總共有15219個廠區回報RMP資料，3201個廠區有40人以上的勞工須回報職災率，兩個資料庫的交集有922個廠區（＝樣本數？）
As of December 7th 2000, a total of 15,219 facilities reported on their covered facilities to the EPA’s RMP*Info database as required by the provisions the RMP regulation. Of these, 3201 facilities had 40 or more employees and used one or more of the 558 NAICS process codes that OSHA uses to determine whether or not to include a facility in the OII survey. Of these 3201 facilities, 922 could be linked to the OSHA OII survey list under the criterion that the RMP*Info and OSHA facility street addresses matched or that the facility name and city matched with a near match on the address. An additional 164 facilities with fewer than 40 employees or with processes that did not match the standard OII survey NAICS process codes were nonetheless matched in the OSHA OII surveys and are included in the analysis below, for a total of 1086 facilities.
2.2 Data analysis
We considered one outcome from the OSHA OII data:
the total number of occupationally-related illnesses and injuries reported to OSHA during 1996–2000.
Since facilities differed in size and might have reported to OSHA for more than one year, OIIs are usually normalised across facilities by measuring the average number per 100 FTE (full-time equivalent employees) per year. This is the approach we use here, as it controls for the number of exposed employees in a facility.
For the RMP data, we considered six outcomes:
death
property damage
evacuations
sheltering in place
environmental damage (reported fish/animal kills, defoliation or other environmental damage)
injury.
解釋變數OII的定義：100F TE（the facility in the given time period 1996–2000 that involved an injury either on-site or off-site.）
被解釋變數：
whether or not an incident took place at the facility in the given time period 1996–2000 that involved any of the following outcomes:
(a) major property damage(USD100000)
(b) evacuations
(c) sheltering in place
(d) environmental damage
(e) deaths.
直覺來說，危害性的製程比較可能導致嚴重的災害與事故發生，然而也可能因為考量製程的潛在危害與風險，而導致人員盡量不要在危害區域作業
Intuitively, more hazardous processes and chemicals might reasonably be expected by their very nature to lead to more frequent or more severe accidents. However, hazardousness might also lead to greater attempts to mitigate underlying risks, perhaps in ways that lead to the substitution of capital for labour in the control infrastructure of the affected facilities and processes.
因此考量製程的本質危害，研究將防護措施的投資以及防止勞工暴露於危害的努力納入考量
To account for the fact that more ‘intrinsically’ hazardous processes tend to involve capital-intensive infrastructure that might confound relationships between OII events and LP-HC events, we first developed a ‘total hazard’ measure as a proxy for facility hazardousness. We then included this measure in our statistical analysis as a control to account for the potential effects of both increased mitigation investments and decreased employee/labour exposure to the underlying process hazards.
各廠區的本質危害（‘total hazard’）大小：
用該廠區使用/存放的危害物質超過管制恕限值的倍數表示：a total hazard measure of 1 means 1 chemical is kept at up to twice threshold level, 2 means two chemicals kept at up to twice threshold level or one chemical at up to four times threshold level and so forth.
廠區有OII資料的這個群組稱為‘apparently OII eligible’.，同時又具備RMP資料的群組稱為‘OII-matched’ facilities，底下結果部份將顯示兩個群組得特性相近，OII-matched可代表OII eligible
另外考量變數不是線性常態分布，此一研究使用非母數的計量方法進行變數間的關連性與回歸(這方法要問老師了，如果版內有統計與計量方法方面的達人，亦懇請不吝出手相救)
To consider the association between LP-HC outcomes and OII rates in a more detailed fashion, a logistic regression model is considered. Standard logistic regression models assume a linear relationship between the predictor and the log-odds of the event. Since this relationship may not be linear, a nonparametric generalised additive model (Hastie and Tibshirani, 1999) was fit to the present data. This allows a very general functional form to be fit to the log-odds outcome variable, based on predictor variables such as OII rates and total hazard measure. The result may be thought of as a very general polynomial fitting process that captures potential non-linearities linking the underlying predictive factors to the log-odds outcome variable.
另外底下這段是關於因果關聯推論的重要說明，算是解釋本質危害大小對於OII與RMP accident 關聯性的影響
A caveat for all statistical analyses is that finding a statistical association between two factors does not prove that one causes the other. For instance, one might view an association between factors A and B as being due to confounding by factor C. That is, A and B might have no association at a given level of C but, due to a common association between A and C and B and C, the unadjusted analysis shows an association between A and B, while the adjusted analysis which compares A and B at similar levels of C shows no association.
底下結果顯示：OII rate竟然和accident呈現負相關，作者認為或許有些關於製程危害特性的因素被忽略了，所以導致這個結果現象。因此用無母數的計量迴歸方程式估計與調整total hazard的影響
For example, in the analysis below, a negative association between accident outcomes and OII rates might not be due to something intrinsic to OII behaviour at the facility but, rather, to the underlying hazardousness (or lack thereof) of the processes that tend to be used in facilities where OII rates are high. Thus, this confounding might mask a positive association between accident risk and OII rates. To account for this particular potential confounding effect, the nonparametric logistic regression model is always adjusted for facility ‘total hazard’ using the total hazard measure described above.
3 Results
ORWRII to denote the total OSHA reportable work-related illnesses or injuries for that facility for the time period 1996–2000. Figure 1 provides a histogram of the OII rates among the 1086 facilities in the RMP*Info database that could be matched to OII surveys during 1996–2000; the mean (SD) OII rate was 3.42 (4.67) ORWRII/Year/100 FTE. Table 1 provides descriptive statistics for key measures taken from the RMP*Info database overall and stratified by OII status (not eligible, eligible but not matched and matched).
標準的 Poisson Distribution
不同群組的OII敘述統計資料
所有的工廠數15219，不須提報OII的工廠數11854，有OII但非RMP的廠區數2279，有OII又符合RMP的產區數：1086
有RMP的廠區，相關的損失（%injury, damage, evacuation…）較小=>符合假說1
這個研究的罩門露餡跑出來了：沒有辦法把行業別的本質作業危害特性量化&納入迴歸；但這是所有進行跨部門、不同產品、行業別安全績效衡量的先天障礙
研究上克服這點的唯一手法：用相同類型部門、相似產品與相同行業別來進行比較與迴歸，但缺點是：樣本數會更少…表3顯示：
有報告LPHC事故的群組（n=95）， OII rate反而較其他沒有報告的群組（n=985）低2%
有報告property damage, evacuation/sheltering, environmental damage or death的群組（n=53），其OII反而較其他沒有報告的群組（n=1027）低24%
Table 3 considers the bivariate association between the LP-HC events reported in
RMP*Info and the OIIs reported to OSHA. In general, facilities reporting LP-HC events had lower OII rates than facilities without such events: facilities reporting one or more injuries during the period had OII rates 2% lower and facilities reporting one or more incidents of major property damage, evacuation/sheltering, environmental damage or death had OII rates 24% lower than facilities without the given type of LP-HC event.
These differences in OII rates are marginally statistically significant in both cases.
為什麼表3的樣本數只剩1077？而不是表1 OII matched的1086？
以上因果關聯的詮釋與推論：
- 有出過事的公司才會重視工安，所以OII較低
- 回報給官方資料的偏誤
- 本文作者的解釋：產業作業特性差異，例如石化廠，OII低，但是危害性卻高；食品加工產業：OII高，但是危害特性低
圖2把所有OII matched的1086個廠區，用OII rate為X軸座標，Total Hazard為Y軸座標做圖，搭配使用spearman 相關性檢定，結果顯示兩者呈現非常顯著的負相關(p<0.001)
圖3用無母數回歸的方法，以樣本的OII資料來預測發生injury或財產損失等accident的或然率，然而兩者均未達統計上顯著
(p = 0.14 for injury and p = 0.091 for major property damage/evacuation/sheltering/environmental damage/death).
很炫的圖，樣本數一樣都是1086個廠區，兩個圖X對Y的線性與線型都不理想
圖4挑食品製造業（n=335）與化學品製造業（n=228）的工作傷害來做圖，結果一樣不理想(p = 0.13 for food processors and p = 0.46 for chemical manufacturers)
4 Discussion
結果討論
樣本廠區的職業傷病率(OII)和RMP報告當中的低可能性高嚴重度(LPHC)的重大事故之間，沒有存在顯著與正相關性；而且有發生死亡與財產損失等重大事故的樣本廠區，通常會較沒有發生重大事故的廠區，有更低的職業傷病率。
然而此一負相關可能是由於有較高職業傷病率的廠區通常其危害性較低，所以也比較不可能發生重大意外(例如食品處理業與石化業對照的例子)
然而即便把這項因素考量進來(不同行業個別比對)，職業傷病率和LPHC之間仍然沒有統計上的顯著關聯，或許只能說高職業傷病率的廠區，有著較高的或然率會發生LPHC的事故(但統計檢定還是不顯著p=0.14)
而此一現象可能因為OHSA沒有特別確認各廠的職業傷病率的申報或廠商職業傷病率的申報吃案而導致沒有顯著的結果。
然而此一研究同時比對了是否被RMP列管的兩個不同群組，以上廠商填報不實資料的誤謬應該是隨機發生，以上的分析不太可能受到太大的誤導。
5 Conclusions and future research
結論與未來研究方向
研究顯示：低職業傷病率的廠商，不見得發生LPHC事故的機率就低
而且以(高)職業傷病率來預測廠商是否會發生LPHC的事故，預測效力有限
前面假設的兩個假說隱含著：「安全管理與氣候文化是決定職業傷病率高低的關鍵因素&擁有良好安全管理與文化的廠區有較少的LPHC重大事故」的假設，此一實証結果顯示不成立。
後續研究應該加入更多衡量安全管理與文化的變數，例如是否通過系統認證等，或許更能反映出對OII與LPHC的影響
Beyond the valuable studies noted just above, there remain many important insights for industry and policy makers related to the continuing assessment of the RMP data. For example, recent results in Kleindorfer et al. (2007) show some significant decreases in injury rates associated with RMP accidents between the first wave of RMP filings studied in this paper (the 1999–2000) and the second wave of RMP filings (filed in 2004–2005). Studying whether these decreases in injury rates are associated with corresponding decreases in the same facilities in their OII rates and relating both RMP and OSHA trends to changes in technology and management systems in specific sectors could be an important source of greater understanding of the foundations of process safety and major accident prevention.
感想：
- 災害與事故都是因為預料不到，所以才叫”意外”？！管理系統與組織內的安全文化，只能預防”可預期”的風險與危害，所以無法產生功效；工安”意外”其實等同於是無法預測的”黑天鵝”事件！
- 這個研究有機會可以搞一個台灣版(evidence from Taiwan)，只要有經濟部工業局、勞委會、消防署與環保署的數據資料庫即可
- 就未來研究發展而言，除了作者講的納入是否有管理系統認證的這個因素(也可哪考慮把拿國家工安環保獎的廠商作為對照組，代表有效管理的群組) 外；還應該區分行業別，以消弭行業危害特性本質差異(paired的比較)；另外公司規模與人數多寡也是一個應該納入考量的因素。
這個題目還有以下更深入的命題：
如何偵測與判定一家公司的內部安全管理是否”有效”？！
不同行業的作業風險與文化，是否也塑造了這個行業的安全管理與文化？
有發生過重大事故教訓的公司才會重視安全？
沒有留言:
張貼留言