運用資料採礦分析技術探索晚發型阿茲海默失智症相關暴露危險及保護因子

Abstract

重要性 失智症為全球最失能與高經濟負擔的健康狀況之一，而阿茲海默病是失智症主要類型。之前的研究雖陸續發表與晚發型阿茲海默( Late-Onset Alzheimer’s Disease, LOAD)失智症相關之危險或保護的共病症，但罕見同時探討所有被診斷的相關共病症。 目的 本研究探討與LOAD失智症相關之危險或保護的共病症，並建構相關共病症和LOAD失智症之間的關係路徑。最後分析有耳聾患者其之後發生失智症的風險。 研究設計 進行全國人口為對象之配對病例對照研究及配對世代研究，並以第一次被診斷為LOAD失智症及第一次診斷耳聾為研究基準日。 資料來源 使用全民健康保險研究資料庫及重大傷病資料庫。 對象 台灣於2007-2013年重大傷病資料中之新診斷LOAD病患及2000-2011年新診斷健保資料耳聾的病患。病例對照研究以性別、年齡進行配對以選取控制組病患；世代研究以性別、年齡、投保薪資及居住地區進行配對。配對比例均為1:1。

暴露因子 篩選在診斷為LOAD失智症日期之前所有被診斷過去共病症，前述共病症認定以國際疾病分類第九版前三碼為準。另篩選新診斷耳聾症。 主要結果測量 LOAD失智症、失智症。 統計方法 (1)卡方或T檢定用於比較LOAD失智症新診斷病例與對照組之間的人口學差異。相關共病症透過多變數條件邏輯斯迴歸及逐步篩選法來進行篩選，顯著相關共病症認定為經多變數校正後之P值小於0.05。P值係執行雙尾檢定所得。 (2)路徑分析用於構建相關共病症和LOAD失智症之間的路徑，分別建構診斷1年之前及診斷4年之前的正、負及最後模型之相關路徑，並分析估計在滿足整體適合度指數(goodness of fit model) 達理想建議值的範圍，最後修正模型(Final model)中各共病症相關路徑間的路徑係數。 (3)使用存活分析計算有無耳聾世代之累積發生率，Cox比例風險模型計算有耳聾診斷的世代於調整干擾因子後發生失智症之風險(Hazard Ratio，HR)。顯著相關認定為經多變數及多重比較校正(False Discovery Rate, FDR)後之P值小於0.05。另以傾向分數配對法進行敏感性分析。 結果 (1)以第一次被診斷為LOAD失智症為研究基準日回溯篩選，在其診斷1年以前，總計有42項過去共病症與LOAD失智症呈顯著相關，其中25項為正相關，17項為負相關。在LOAD失智症診斷4年以前，總計有26項過去共病症與LOAD失智症呈顯著相關，其中19項為正相關，7項為負相關。 (2)在滿足整體適合度指數達理想建議值的範圍，於診斷1年之前及診斷4年之前的最後修正模型中，處於模式路徑的上層位置有4項過去共病症(ICD codes 300焦慮; ICD codes 307特殊症狀或徵候群心理疾患; ICD codes 564功能性消化道疾病; ICD codes 386眩暈徵候群及其他前庭系統之疾患)，且此4個共病症在1年之前及診斷4年之前的2項模型內，比其它相關共病症對LOAD的總影響力也是較大的。此外，3項共病症(ICD codes 564、386及389耳聾)為模式中其他多個共病症的常見中間變項。 (3)存活分析顯示，在觀察期間，耳聾患者累積失智症發生率比非耳聾患者顯著高(p< 0.001, log-rank test)，耳聾患者比非耳聾者的失智症發生率高 (耳聾：19.38 , 95% CI, 18.25-20.57; 非耳聾13.98, 95% CI,13.01-15.00，單位: 1000 person-year)。在控制潛在的干擾變項後，Cox迴歸模型分析顯示，耳聾與患失智症有顯著正相關(HR = 1.14, 95% CI, 1.04 - 1.25; p= 0.007, FDR p= 0.014)。年齡增加和高門診頻率與患失智症均呈顯著正相關(HR = 2.53, 95% CI, 2.40 - 2.67, FDR p< 0.001; HR = 1.05, 95% CI, 1.03 - 1.08; FDR p< 0.001)，投保薪資2萬及以上與失智顯著呈負相關(HR = 0.71, 95% CI, 0.62 - 0.81; p< 0.001, FDR p< 0.001)。 此外，心血管疾病、糖尿病、焦慮、憂鬱、酒精相關疾病及頭部外傷與發生失智症呈顯著正相關。以三個年齡組( 45-64 歲、65-74 歲、≥75 歲)的分層分析結果顯示，45-64歲組患失智的風險顯著高於其他年齡組(HR = 2.14, 95% CI, 1.51 - 3.03, FDR p<0.001)。根據敏感性分析結果，耳聾仍為發生失智症的危險因子(HR = 1.2, 95% CI, 1.09 - 1.31; FDR p< 0.001)，45-64歲組及65-74 歲組具患失智風險(HR = 1.44, 95% CI, 1.16 - 1.8, FDR p= 0.004; HR = 1.24, 95% CI, 1.07 - 1.45, FDR p=0.021)。 結論 本研究篩選出42項與LOAD相關的過去共病症，推測部分共病症為LOAD失智症之前驅期病症如焦慮、特殊症狀或徵候群心理疾患、情感性精神病、耳聾及大腦退化等(ICD codes 300、307、296、389及331)，對LOAD發生的總影響力也是較大。於神經精神疾病、功能性消化道疾病、眩暈徵候群及前庭系統之疾患及耳聾等疾病的相關路徑(ICD codes 300、307、296、386及389)支持“腸 - 腦軸”或“腦 - 腸軸”雙向通信系統作用在LOAD發生扮演重要角色。在世代研究分析顯示，耳聾是患失智症的顯著風險因子，特別是在45-64歲的族群。因此，聽力保護、篩檢及早期治療應視考慮列入預防失智症的策略之一。Importance Dementia is one of the most debilitating and burdensome health conditions worldwide. Alzheimer’s disease is a major type of dementia. Although studies have found some risk or protective factors regarding comorbidities associated with late-onset Alzheimer’s disease (LOAD), all comorbidities have rarely been screened simultaneously in the literature. Purpose The present study attempts to determine prior comorbidities associated with LOAD and construct pathways for these comorbidities. In addition, the incidence risk of all types of dementia in patients with hearing-loss was investigated. Study design We conducted a total population-based matched case–control study and a population-based matched cohort study. The index date was set as the first diagnosis date of LOAD and hearing loss. Data source Data were collected from Taiwan’s National Health Insurance Research Database and Catastrophic Illness Database. Study participants The study participants included patients newly diagnosed with dementia (International Classification of Diseases, ninth revision, ICD-9-CM code: 290) from 2007 to 2013 and those prescribed any acetylcholinesterase inhibitors (AChEIs; Anatomical Therapeutic Chemical code: N06D) as well as those newly diagnosed with hearing loss (ICD-9-CM codes 389 and A241) from 2000 to 2011. These patients were exactly matched according to sex and age, with a matching ratio of 1:1 for case control study and a matching ratio of 1:1 by sex, age, residence, insurance premium for the cohort study. Exposures Prior comorbidities were screened out for 1, 2, 3, 4, 5, 6, 7, 8, 9 years before the first diagnosis of LOAD. Comorbidities were defined according to the first three digits of ICD-9-CM codes. Furthermore, a diagnosis of hearing loss was defined according to ICD-9-CM codes 389, A241. Main Outcome Measure LOAD and dementia. Statistics analysis (1) Chi-squared or t-tests were performed to examine differences in the demographics and characteristics of patients newly diagnosed with LOAD and those of controls. A conditional logistic regression model was used to identify associated comorbidities. To evaluate the association between these comorbidities and the risk of LOAD by performing stepwise multivariate logistic regression, each significant independent factor was adjusted P< 0.05; all tests were two-sided. (2)Path analysis was performed to construct the pathways between associated comorbidities and LOAD for 1 and 4 years before the first diagnosis of LOAD. The path coefficients of each pathway were estimated in goodness-of-fit of model. (3) A Kaplan–Meier analysis was performed to calculate the cumulative incidence of dementia. Cox’s hazards model was used to compute the hazard ratios (HRs) of dementia progression after potential confounding factors were adjustedfor in the hearing-loss cohort study. Each significant independent factor was considered as having a false discovery rate (FDR) adjusted p value of< 0.05. To validate the robustness of the main study findings, we performed a sensitivity analysis by matching propensity scores. Results (1)Of the total 42 diagnostic prior comorbidities associated with LOAD in 1 year before the first diagnosis date of LOAD, 25 comorbidities had positive effects and 17 comorbidities had negative effects. Moreover, of the 26 comorbidities associated with LOAD in the 4 years before the first diagnosis date, 19 comorbidities had positive effects and 7 comorbidities had negative effects. (2) In the final model for 1 and 4 years before the first diagnosis date of LOAD, anxiety, psychopathology-specific symptoms, functional digestive disorder, vertiginous syndromes, and disorders of the vestibular system (ICD codes 300, 307, 564, and 386) located at the upper position of final model had stronger positive effects on LOAD incidence than other associated comorbidities. Vertiginous syndromes and disorders of the vestibular system, hearing loss, functional digestive disorder, and disorders of the urethra and urinary tract (ICD codes 564, 386, 389, and 599) were frequent mediators in the model. (3) In the Kaplan–Meier analysis, cumulative dementia incidence rates were significantly higher in the hearing-loss group than those in the non-hearing-loss group (P< 0.001, log-rank test). The dementia incidence rate in the hearing-loss group was higher than that in the non-hearing-loss group (19.38, 95% CI, 18.25–20.57; 13.98, 95% CI, 13.01–15.00, 1000 person-years) during the follow-up time. After the fully adjusted multivariate cox regression model was applied for risk analysis, it was found that the patients with hearing loss had a significant risk of developing dementia (HR = 1.14, 95%CI, 1.04–1.25; FDR P = 0.014). Increased age and highly frequent outpatient visits were significantly and positively associated with a risk of dementia, but insurance premium amount (NT$20,000 and above per moth) was significantly and negatively associated with dementia. In addition, cerebrovascular disease, diabetes, anxiety, depression, alcohol-related illnesses, and head injury were significantly and positively associated with a risk of dementia (FDR P< 0.01). A stratified analysis of three age groups (45–64 years, 65–74 years, and ≥75 years) showed that the 45–64-year group was at a significantly higher risk of developing dementia than the other age groups (HR = 2.14, 95% CI, 1.51–3.03, FDR P < 0.001). After a sensitivity analysis was performed, it was revealed that patients with hearing loss still had a significant risk of developing dementia (HR = 1.2, 95% CI, 1.09–1.31; FDR P < 0.001), and patients in the 45–64-year and 65–74-year age groups were at a significantly higher risk of developing dementia than those in the ≥75-year age group (HR = 1.44, 95% CI, 1.16–1.8, FDR P = 0.004; HR = 1.24, 95% CI, 1.07–1.45, FDR P = 0.021). Conclusion A total of 42 prior comorbidities related to LOAD, such as anxiety, psychopathology-specific symptoms, hearing loss, episodic mood disorders, and cerebral degenerations (ICD codes 300, 307, 296, 389, and 331), before diagnosis were related to prodromal LOAD and yielded a high total positive effect on LOAD. The corrected pathways among neuropsychiatric, digestive, vestibular, and hearing system diseases (ICD codes 300, 307, 564, 386, and 389, respectively) and LOAD may support the gut–brain axis and brain–gut axis causal effect in LOAD pathogenesis. In our cohort study, the results showed that hearing loss is a significant risk predictor for development of dementia, especially among patients in the 45–64 years age group. Hearing protection, screening, and treatment can be employed as the strategies for preventing dementia.