職業病低報與實際解決方案

Abstract

職業病低報是許多國家長期存在的嚴重未解決問題。在眾多成因中，醫療可近性差是一個值得深思卻鮮少被討論的因素，這也是本研究首要探討的重點。首先，我們將評估患者移動到最近的主要職業醫學門診所需的長交通距離或交通時間，將如何阻礙職業病的通報，並估計在缺乏職業醫學門診的地區，例如新竹和苗栗，增設一家主要通報醫院可能帶來的影響。隨後，我們將追蹤臺大醫院新竹臺大分院正式成為主要通報醫院後的實際變化，以驗證我們的預估是否符合實際情況。 我們利用臺灣職業病通報系統（NODIS）的數據以及年度人力資源調查，基於各行政區勞工的行業和職業組成，去計算所有行政區（2008年至2018年）以及新竹和苗栗共 24/23 個目標行政區（2023年至2024年）的職業病發生率和預期職業病發生率；各行政區到到最近主要職業醫學門診的交通距離和交通時間，則透過 Google Maps 的 Distance Matrix API進行估算。我們採用準泊松迴歸模型來探究交通距離和交通時間對職業病發生率的影響，同時透過將預期職業病發生率作為Offset來校正行業和職業的影響。此外，我們也進行了次群組分析，以探究就業狀況、病假和通報年度所帶來的影響。 在2008年至2018年期間，我們的初步研究納入了3,420例確診職業病案例。經準泊松迴歸模型分析，在校正行業和職業類別後，交通距離和交通時間對職業病發生率具有顯著影響。當交通距離和交通時間分別增加10公里與10分鐘時，職業病發生率會分別下降 10.90%與10.73%。據此估計，每年會有約200例職業病或高達40%的職業病因此被低估。在次群組分析中，只有輕症的工人仍然受到交通距離和交通時間的顯著影響；在2023年至2024年期間，我們的後續研究納入了 82例確診職業病案例。同樣使用準泊松迴歸模型，在校正行業和職業類別後，交通距離和交通時間對職業病發生率仍有顯著影響。當交通距離和交通時間分別增加10公里與10分鐘時，職業病發生率會分別下降39.61%與44.71%。實際的職業病發生率（每百萬工作人年39.49與39.48例）比我們初步研究時的估計值（每百萬工作人年31.91與29.79例）高出約30%。 我們的初步研究顯示，醫療可近性差是導致職業病低通報的主因之一，尤其是輕症案例，且高達40%的職業病可能因此未被通報。利用這套方法，我們可以透過估計額外能通報多少職業病案例，來評估在醫療可近性差的地區增設通報醫院的成本效益。後續研究的發現則再次印證了醫療可近性在職業病低報中的重要作用，並指出其重要性甚至被嚴重低估。在缺乏職業醫學門診的地區，設立主要通報醫院後，職業病的發生率幾乎增加了兩倍。 除了主要通報醫院的可近性之外，職業病案例在這些醫院如何被轉診和診斷是另一個值得考量的因素，這也是本研究的第二個重點。診斷職業病需要基層醫療臨床醫師和職業醫學醫師共同進行臨床診斷並評估工作相關性。因此，兩者間精確且高效的轉診機制對於職業病的診斷至關重要。國際疾病分類代碼（ICD codes）曾被用作轉診標準，但對於多重病因的新興職業病，例如肌肉骨骼疾病（MSDs），ICD代碼的陽性預測值（PPV）低，使其成為不良的轉診標準。因此，我們研究的第二個目標是找到與工作相關性有關，而可能比現有 ICD 代碼有更高陽性預測值的轉介標準。 我們利用臺灣職業病通報系統（NODIS）在 2012 年至 2018 年間的數據，計算了不同人口學因素下，個案被認定與工作相關性大於50%的機率。後續我們進一步使用二項式迴歸模型來識別工作相關性的預測因子，並對每種肌肉骨骼疾病診斷進行次群組分析。 在2012年至2018年期間，我們的研究納入了4,651例職業性肌肉骨骼疾病通報個案，其中2,901例（62.37%）為工作相關性大於50%的個案。使用我們的二項式迴歸模型，包括工作年資、性別、病假、行業和職業類別等特徵均為肌肉骨骼疾病的工作相關性預測因子，且每種肌肉骨骼疾病都有一組獨特的預測因子，這反映了其職業病因與職業醫師評估其工作相關性的方式。 使用這種方法，我們不僅可以識別每種肌肉骨骼疾病的高風險特徵及其診斷勝算比（DOR），還可以將不同的特徵組合成一套轉介標準，並根據此套轉介標準，去計算轉介個案後續被認定工作相關性大於50%的機率，從而改善職業醫學與其他專科之間的轉診機制。 總結來說，根據我們的研究，在缺乏職業醫學門診的地區設立主要通報醫院，以及根據識別出的高風險特徵制定職業病的轉介標準，將是解決職業病通報不足的有效實用解決方案。有關當局可以利用這兩種策略來促進職業病的診斷，同時建議進行後續研究以驗證其實際效益。

Under-reporting of occupational diseases (ODs) has been a serious un-solved issue in many countries. Among all the etiologies, poor medical accessibility is a considerable yet rarely discussed factor, which is the first focus of our study. We would first evaluate how ODs reporting is impeded by long travel distance/time (TD/TT) to nearest major occupational medicine clinics, and estimate impacts of establishing an additional major reporting hospital in areas lacking occupational medicine clinics, like Hsinchu and Miaoli. We would then follow-up the actual change after National Taiwan University Hospital Hsinchu branch became a major reporting hospital, to validate whether our estimation meets actual scenario or not. Using data from NODIS, Taiwan’s ODs surveillance system, and annual Manpower Survey, we calculate incidence/reporting rate of ODs (IROD) and expected IROD of all districts (2008 to 2018) and targeted 24/23 administrative districts in Hsinchu and Miaoli (2023 to 2024), based on industries and job titles. Each town’s TD/TT to nearest major occupation medicine clinics is estimated by Google Maps’ Distance Matrix API. Quasi-Poisson regression model is used to investigate effect of TD and TT on IROD, while industries and job titles are adjusted by offsetting expected IROD. Subgroup analysis is then carried out to check the effect of employment status, sickness absence, and reporting years. During 2008 to 2018, 3420 cases of definite ODs are included in our initial study. Using quasi-Poisson regression model, after adjusting industry types and job titles, TD and TT have significant effect on IROD. As TD/TT increase by 10 km/10 mins, IROD decreases by 10.90%/10.73%. It is estimated that around 200 OD cases per year or 40% ODs are therefore under-reported. In subgroup analysis, only mildly-sicked workers are still significantly affected by TD and TT. During 2023 to 2024, 82 cases of definite ODs are included in our follow-up study. Using quasi-Poisson regression model, after adjusting industry types and job titles, TD and TT have significant effect on IROD. As TD/TT increase by 10 km/10 mins, IROD decreases by 39.61%/44.71%. The actual incidence/reporting rates of occupational diseases (39.49/39.48 per million worker-years) are about 30% more than with estimation made by our previous study (31.91/29.79 per million worker-years). Our initial study shows how poor medical accessibility leads to under-reporting, especially for mildly-sicked cases, and up to 40% ODs could be under-reported. Using this method, we can evaluate cost-effectiveness of adding reporting hospital in areas with poor medical accessibility by estimating how many cases of occupational diseases could be reported additionally. Our follow-up study’s findings reiterate the import role of medical accessibilities in under-reporting of occupational diseases and point out that its importance is even largely underestimated. In areas lacking clinics of occupational medicine, the incidence/reporting rates of occupational diseases would almost triple after the establishments of major reporting hospital. In addition to the accessibility of major reporting hospital, how OD cases are referred and diagnosed in these hospitals is another considerable factor, which is the second focus of our study. Diagnosing occupational diseases (ODs) needs both primary care clinicians and occupational physicians to make clinical diagnoses and evaluate work-relatedness. Thus, a precise and efficient referral mechanism between them could be crucial for the diagnosis of ODs. ICD codes have once been used as referral criteria, yet novel ODs with multiple etiologies, like musculoskeletal diseases (MSDs), make ICD codes poor referral criteria with low positive predictive values. Thus, the second aim of our study is identifying criteria that are associated with work-relatedness and may have better positive predictive values to the existing ICD codes. Using data from Network of Occupational Diseases and Injuries Service (NODIS), Taiwan’s ODs surveillance system, during 2012 to 2018, we calculate the odds of cases being recognized as probable according to different demographic factors. A binomial regression model is further used to identify predictors of work-relatedness, and subgroup analysis is then carried out for each MSDs diagnosis. During 2012 to 2018, 4651 reported cases of occupational MSDs are included in our study, and 2901 (62.37%) cases are probable cases. Using our binomial regression model, characteristics including tenure, gender, sick leaves, industries and job titles are the predictors of work-relatedness of MSDs, and each MSD is associated with a unique set of predictors, which reflects its occupational etiologies and how occupational physicians evaluate work-relatedness. Using this method, we could not only identify high-risk characteristics and its diagnostic odd ratio (DOR) for each MSD, but also combine different characteristics into a set of referral criteria and calculate the odds of cases being recognized as probable, which could improve referral mechanisms between occupational medicine and other specialties. Based on our study, establishing major reporting hospitals in areas lacking occupational medicine clinics and developing referral criteria of ODs based on identified high-risk characteristics would be effective practical solutions to under-reporting of occupational diseases. Authorities concerned could use these two strategies to promote diagnosis of occupational diseases, and further follow-up studies are recommended to validate its actual effectiveness.