雙變項慢性病自然病史之探討:以高血壓及糖尿病之事件為例

Bivariate Events of Natural Course of Chronic Diseases: An Illustration with Hypertension and Diabetes Mellitus

賴和賢(Ho-Hsien Lai)

指導教授 ： 陳秀熙

國立臺灣大學/公共衛生學院/預防醫學研究所/碩士(2006年)

https://doi.org/10.6342/NTU.2006.02044

摘要

背景
　　第二型糖尿病(Type 2 D M, 簡稱糖尿病)及高血壓兩者是常見的慢性病，尤其於成年人常見糖尿病及高血壓同時共存。雖有許多研究進行糖尿病與高血壓共病性現象的探討，但大部分研究都以糖尿病或高血壓為主體進行研究，很少研究報導整個族群的糖尿病及高血壓的盛行率，甚至發生率的估算等等。但就臨床疾病管理觀點而言，不論高血壓或糖尿病之盛行率量化對於心血管疾病、大小血管併發症及死亡等等都相當重要。另一方面，對於糖尿病或高血壓發生率估算，甚至病程進展至兩者共病性時之估算，對於族群疾病預防亦相當重要；如果進一步根據人口學相關變項、生化及其相關危險因子等進行細部的盛行率及發生率估計，則對於疾病的預防醫學都相當有助益，例如影響各種病程狀態的重要因素，如能了解其嚴重性便可有效的進行預防或介入等，以達疾病預防之目的。
　　此外，對於這些常見慢性疾病的共病性問題，目前仍有許多潛在議題需進一步釐清。首先，我們可以從越來越多以橫斷面盛行個案的研究中，發現共病性問題持續被探討，然而有許多潛在可能發展共病性的個案，無法在橫斷式的調查中被發現。這個現象如同存活分析中的右設限問題，有可能低估共病率。另外，疾病發生先後之時間序位問題是相當困難回答的一項議題，因為慢性病精確的發病時間是不能得知的。再者，我們亦不清楚有關生活習慣因子如何影響疾病的產生。

研究目的

本研究目的乃利用社區健康篩檢世代進行
一、估計高血壓，糖尿病，及兩者共病症之盛行率，發生率，及五年或十年累積發生危險性。
二、探討不同生化危險因子間有關之高血壓，糖尿病，及兩者共病症之盛行率，發生率，及五年或十年累積發生危險性差異。
三、評估高血壓及糖尿病發生先後順序之可能。
四、發展數學模式進行可潛在發展共病症之比例，高血壓及糖尿病發生先後順序及高血壓和糖尿病發生率及發展成共病症之轉移機率。
五、評估上述可潛在發展共病症之比例，高血壓及糖尿病發生先後順序及高血壓和糖尿病發生率及發展成共病症之轉移機率在不同生化危險因子之差異。

材料與方法
研究對象為1996年至2000年間參與馬祖社區篩檢的個案，利用每年篩檢實行問卷，身體理學檢查，及血清學檢查。接著分別利用描述性統計，轉移模型，Cox Proportional Hazards Regression Model，此外，本研究發展一雙元事件(bivariate event)潛在變項(latent variable)模式，估計兩種事件的發生率，及單一事件發生後會再進展為共存狀態的轉移速率，以高血壓及糖尿病為例，該模式納入兩個潛在變項，一為個案在一生中可能發生兩種疾病的潛力，另一為具發生兩種疾病潛力的個案會先發生高血壓的比例，由於實際觀察到的資料會因右設限而無法直接估計上述兩個參數，因此我們以兩種潛在變項將所有個案分為四種可能：(1)兩疾病均會發生且高血壓發生在前，(2) 兩疾病均會發生且糖尿病發生在前，(3)僅可能發生高血壓，(4)僅可能發生糖尿病，再利用三個三階段馬可夫模式描述上述(1)，(2)及(3)+(4)，並據此建構總體概似函數且估計其最大概似估計值，以達成本研究原先所設定之研究目的。

研究結果
　　在盛行率上，根據馬祖在1996年至2000年間所進行的社區篩檢所得到的資料，我們得知在此期間內該地三十歲以上民眾的高血壓，糖尿病盛行率分別為35%，以及6%。發生率上，單純的高血壓發生率整體為14%(13~16%);單純的糖尿病發生率整體為1.6%;就共病症的發生而論，相較於完全沒有病的人發生成為共病症的發生率，單獨具有糖尿病的人發生成為共病症的發生率為其88倍；相較於完全沒有病的人發生成為共病症的發生率，單獨具有高血壓的人發生成為共病症的發生率為其10倍。
　　在轉移模式的分析上，在控制了性別，年齡，BMI，及尿酸等變項之後，先前狀態中具有糖尿病及高血壓者具有分別高達92倍及6倍的風險性發展成為共病症。利用Cox 回歸模式分析時也可以找到相似的發現。當以共病症的發生作為事件發生時，先前狀態中具有高血壓及糖尿病者的粗危險對比值分別為2.5及13.2而經控制了性別，肥胖症，抽菸，及飲酒習慣後，校正之危險對比值對於先前狀態中具有糖尿病及高血壓者而言分別為9倍及2倍。
在隨機模式的估計上，可知在馬祖篩檢世代三十歲以上的個案中約有14%的個案具有發展為共病性的潛力(95%信賴區間(CI): 6.80%-20.75%)，這些人當中約有77.89%的個案會先發生高血壓，另外22.11%的個案則會先發生糖尿病再進展為與高血壓並存的共病性。相對於上述的14%，其他86%的個案則最多只可能發生一種疾病(高血壓或糠尿病)，或者兩種疾病都不發生。新發高血壓的發生率約為每100人年14.71例(95% CI: 13.03-16.39)，但糖尿病發生之後，再有高血壓的發生率則高出許多(0.9897, 95% CI: 0.4203-1.5590)。同樣的情況可見於糖尿病，新發糖尿病的發生率約為每100人年發生0.99例(95% CI: 0.42-1.56)，但在高血壓發生之後的糖尿病發生率則竄升至每`100人年30.92例(95% CI: 0.4203-1.5590)。
　　在檢視危險因子對於轉移速率影響的模式中，我們發現其具有共病性潛力比例及這些人中會先發生高血壓的比例(p)的估計值與上述結果差異不大，在調整不同危險因子的情況下，π的估計值介於11%及17%之間，p的估計值則介於72%至82%之間。對於轉移速率的影響，我們發現男性相較於女性的新發高血壓發生率約為兩倍(95% CI: 1.56-2.46) (p<0.05)，性別對於其他三個轉移速率則未達統計上顯著水準。相同的現象可見於高尿酸血症，抽菸及喝酒。肥胖個案(BMI≧25)則是具有較高的新發高血壓發生率(RR=2.02, 95% CI: 1.60-2.54)及高血壓後糖尿病發生率(RR=3.55, 95% CI: 1.55-8.14)。三酸甘油酯較高的個案(Triglyceride≧200)則是在高血壓(RR=2.66, 95% CI: 1.87-3.79)及糖尿病(RR=3.46, 95% CI: 1.00-11.99)的新發發生率有較高的危險性。高膽固醇(≧200mg/dL)對於轉移速率的影響則均未呈統計上顯著差異。
　　在檢視危險因子對於轉移速率影響的模式中同時再加入危險因子對於具共病潛在可能比例(π)的影響，結果發現各危險因子對於轉移速率的影響並不大，唯一的例外是原本肥胖對於高血壓後糖尿病的發生率之顯著影響變成不具統計上顯著水準。僅有三酸甘油酯異常對於π具統計上顯著水準的危險性(OR=2.12, 95% CI: 1.82-2.47)。

結論
整體而言，本論文分別估計了糖尿病、高血壓及共病性的盛行率、發生率及累積危險性，並藉由相關的生化指標找出糖尿病、高血壓及共病性的高危險群。研究結果同時也顯示高血壓的發生似乎早於糖尿病的發生。
　　而此結果也研究利用所發展出來的統計模式成功的同時量化高血壓及糖尿病兩種事件的疾病進展，並同時估計在族群中會發生共病性的潛在比例及其時序性而獲得支持。

關鍵字

糖尿病 ； 高血壓 ； 共病症 ； 自然病史 ； 雙變項事件

並列摘要

Background
Diabetes Mellitus (DM) and hypertension are two common chronic diseases and may coexist in an adult. Either treating hypertension or DM as outcome, both directions in previous studies have consistent findings. In spite of these findings, it is very rare to report prevalence rate and estimate incidence rate in the same study based on data from an underlying population rather than from hypertensive or diabetic subjects. Quantifying the prevalence rate has a significant implication for clinical management of both diseases in order to reduce macro-vascular, macro-vascular complications, and deaths. Estimating incidence from general population may give a clue to onset of hypertension and DM and progression to co-morbidity. Moreover, cumulative risk for co-morbidity or incidence by different demographic features or biochemical variables may be of great help for identifying high-risk group for developing newly diagnosed hypertension or DM.

There are several potential issues that still remain elusive in investigating co-morbidities of these common chronic diseases. First, reporting the preponderance of co-morbid diseases is frequently based on prevalent cases with a cross-sectional survey. This may underestimate co-morbidity rate because those who are potential of developing co-morbidity may not be observed yet at the time of survey, which is so-called right-censored problem in survival analysis. Second, the temporal sequence between serial events, which is a thorny issue because the exact onset time of chronic disease is unknown, has been neglected and hardly addressed. Third, how life-style factors affect occurrence of serial events is also unclear.

Aim
By using the dataset of community-screening program in Matzu we carry out following tasks and try to solve up all the potential issues.
1. Estimate the prevalence, incidence, 5-year or 10-year cumulative risk for hypertension, diabetes mellitus, and comorbidity.
2. Estimate the 5-year or 10-year cumulative risks based on incidence by different biological measures.
3. Evaluate the temporal sequence between serial events of hypertension and diabetes mellitus.
4. Develop a mathematical method to evaluate the potential proportion of being comorbidity case, the temporal sequence between serial events of hypertension and diabetes mellitus, the incidence of hypertension and diabetes mellitus, and the transition probability of developing comorbidity.
5. Estimate the potential proportion of being comorbidity case, the temporal sequence between serial events of hypertension and diabetes mellitus, the incidence of hypertension and diabetes mellitus, and the transition probability of developing comorbidity by different biological measures.

Material and Methods
Study population is consisted of those who participate community-screening program in Matzu between 1996 and 2000. In the community-screening program, questionnaire, physical examination, and serological examination were performed annually. After data collection is completed, we apply basic descriptive statistics, Markov transition model, and Cox proportional hazards regression model to perform analysis. Besides, we also develop a bivariate event model to estimate two latent variables, the incidence of each single event, and the transition rate form being single event state to the state of two events. Take hypertension and diabetes mellitus for example, latent variables include the proportion of an individual with potential of developing hypertension (HTN) and diabetes mellitus (DM) and the proportion of getting hypertension first and diabetes mellitus latter among all the cases of comorbidity. For the sake of right censoring, it is unable to estimate these two latent variables directly. In order to achieve the aim we set, based on these latent variables we divided all the cases into four conditions, getting hypertension first and comorbidity latter, getting diabetes mellitus first and comorbidity latter, getting hypertension only, and getting diabetes mellitus only. Then, we constructed three three-state stationary Markov models to describe the transitional probability of every condition that mentioned above. Total likelihood function was derived, then the maximum likelihood estimates were obtained by using Newton-Raphson optimisation. 95% confidence intervals were calculated using variance estimated from the inverse Hessian matrix.

Result
The overall prevalence rate was 35% and 6% for hypertension and DM, respectively. The overall incidence rate of hypertension was 14% (13%-16%). The overall incidence rate of diabetes mellitus was 1.6%. DM and hypertension had 88 and 10 times, respectively, higher risk for developing co-morbidity compared with subjects free of DM and hypertension.

In the results of transition model, we found that after controlling for age, gender, BMI, and uric acid, DM and hypertension in previous state had 92-fold and 6-fold risk for developing co-morbidity. Similar findings are found by using Cox proportional hazards regression model. The crude hazard ratios for DM and hypertension in previous state were 2.5 and 13. After controlling for gender, obesity, smoking, and alcohol drinking, the adjusted hazard ratios for developing co-morbidity were 2-fold and 9-fold for the past history of hypertension and DM.

In the results of stochastic model, we found that around 14% (95% confidence interval (CI): 6.80%-20.75%) subjects aged 30 years and above in the Matzu cohort have potential to developing cormobidity. Among them, 77.89% subjects would develop hypertension first. The other 22.11% would have diabetes mellitus first. It yielded 86% cases could either develop single disease (either hypertension or diabetes mellitus) or none of both diseases in lifetime. The incidence of fresh hypertension is 14.71 cases per 100 person-year (95% CI: 13.03-16.39). However, the incidence of hypertension after diabetes mellitus diagnosed was much higher (0.9897, 95% CI: 0.4203-1.5590). For fresh diabetes mellitus, the annual incidence was estimated as 0.99 cases per 100 person-year (95% CI: 0.42-1.56). After hypertension, the incidence of diabetes mellitus was much higher (0.3092, 95% CI: 0.4203-1.5590).

When looking into the effect of single covariate on transition rates, we found that the estimates of p and p were quite stable. Estimates of p were between 11% and 17% when considering different covariate. Estimates of p were around 72% to 82%. We found that male had 2-fold risk (95% CI: 1.56-2.46) of incidence of fresh hypertension (p<0.05). The effects were not statistically significant on the other three transition rates. This was similar to the effects of hyperuricemia, smoking, and alcohol drinking. For obese subjects (BMI³25), they had statistically significant higher risks (RR=2.02, 95% CI: 1.60-2.54) on fresh incidence of hypertension and on the risk of developing diabetes mellitus after hypertension already diagnosed (RR=3.55, 95% CI: 1.55-8.14). Compared to those with normal value of triglyceride, cases with hypertriglycemia (³200) had statistically higher risk of both fresh incidence, 2.66-fold on hypertension (95% CI: 1.87-3.79) and 3.46-fold on diabetes mellitus (95% CI: 1.00-11.99). In this model, the effects of elevated total cholesterol (³200) were not statistically significant on all four-transition rates.

The effects of covariates on transition rates with simultaneous consideration of the effects on the potential of comorbidity (p) were similar to the results in Table 2, except the one of obesity on the incidence of diabetes mellitus after hypertension diagnosed became insignificant. We found the only significant covariate on p was hypertriglycemia (OR=2.12, 95% CI: 1.82-2.47).

Discussion
In conclusion, prevalence, incidence and cumulative risk of hypertension, DM, and co-morbidity were estimated, of which high risk group for developing hypertension, DM and co-morbidity by relevant biochemical variables were identified. It also suggests that occurrence of hypertension seems prior to the development of DM.
Finally, a statistical model was developed for modeling bivariate events of disease natural history for hypertension and DM taking susceptibility to co-morbidity and temporal order of hypertension and DM.

並列關鍵字

Diabetes Mellitus ； Hypertension ； Comorbidity ； Natural History ； Bivariate Events

參考文獻

1. Barrettconnor, E., Criqui, M. H., Klauber, M. R. & Holdbrook, M. Diabetes and Hypertension in A Community of Older Adults. American Journal of Epidemiology 113, 276-284 (1981).

2. Ostrande, L. D., Francis, T., Hayner, N. S., Kjelsber, M. O. & Epstein, F. H. Relationship of Cardiovascular Disease to Hyperglycemia. Annals of Internal Medicine 62, 1188-& (1965).

3. Teuscher, A., Egger, M. & Herman, J. B. Diabetes and Hypertension - Blood-Pressure in Clinical Diabetic-Patients and A Control Population. Archives of Internal Medicine 149, 1942-1945 (1989).

4. Kannel, W. B., Wilson, P. W. & Zhang, T. J. The epidemiology of impaired glucose tolerance and hypertension. American Heart Journal. 121(4 Pt 2):1268-73, (1991).

5. Tai, T. Y., Chuang, L. M., Chen, C. J. & Lin, B. J. Link Between Hypertension and Diabetes-Mellitus Epidemiologic-Study of Chinese Adults in Taiwan. Diabetes Care 14, 1013-1020 (1991).