全民健保開辦初期台灣中老年人自陳多重慢性病的時期與世代訊號：1993-2003年之年齡-時期-世代分析與2011年延伸

Period and birth-cohort patterns in self-reported multimorbidity in middle-aged and older Taiwanese adults during early National Health Insurance consolidation: an age-period-cohort analysis, 1993–2003, with an extension to 2011

目標：以年齡-時期-世代（age-period-cohort，APC）模型描述全民健保開辦初期台灣中老年人自陳多重慢性病的時期與出生世代訊號，區分「制度可見度」與「世代健康資本」，並界定哪些結論可辨識。方法：分析中老年身心社會生活狀況長期追蹤調查1993至2003年四波合併樣本（18，049人次／7，397人）。多重慢性病為八項自陳醫師診斷慢性病中至少二項，另以各病之日常生活不便題組建立嚴重度分級結果。以條件式APC邏輯迴歸搭配500次受訪者層次拔靴信賴區間（CI）估計，以正解線（solution line）分析檢驗部分識別、以界限分析處理追蹤流失，並另作2011年延伸分析。結果：65歲以上共同支持區間之年齡標準化盛行率由1993年33.8%升至2003年43.9%、2011年47.9%。時期成分由1996年低點（勝算比OR=0.78，95% CI [0.73，0.83]）升至2003年（1.36，[1.16，1.57]）；世代成分以1920-1929年為高平台，逐世代降至1950-1954年（0.46，[0.25，0.90]）。非線性成分顯示1996年低點與1925-1929年世代高點不依賴識別限制，線性趨勢幅度則隨假設變動。嚴重度分級結果之時期上升較小（2003年1.17）。結論：上升同時與可見度提升及世代健康資本改善相容；慢性病監測宜將兩種成分分開報告，並區分可辨識與不可辨識之部分。

Objectives: To describe period and birth-cohort patterns in self-reported multimorbidity in middle-aged and older Taiwanese adults during the early consolidation of Taiwan’s National Health Insurance, distinguish changes in institutional visibility from differences in health across birth cohorts, and clarify which findings can and cannot be identified from age-period-cohort analysis. Methods: Four pooled waves of the Taiwan Longitudinal Study on Aging (1993–2003) included 18,049 person-wave observations from 7,397 respondents. Multimorbidity was defined as two or more of eight self-reported physician-diagnosed conditions. Severity-graded outcomes also incorporated activity limitations associated with these conditions. Conditional age-period-cohort logistic regression was performed, and confidence intervals (CIs) were estimated using 500 person-level cluster bootstrap replicates. Partial identification was performed by fixing the linear age slope at different values along the solution line. Attrition was examined using inverse-probability weighting and extreme-value bounds. The 2011 wave (n = 3,727) was analyzed as an extension. Results: Age-standardized prevalence over the common age≥65-year support increased from 33.8% in 1993 to 43.9% in 2003 and 47.9% in 2011. The period component reached a nadir in 1996 (odds ratio [OR] = 0.78, 95% CI [0.73, 0.83]) and increased to 1.36 [1.16, 1.57] in 2003. The cohort component plateaued for the 1920–1929 birth cohorts (OR, approximately 1.44) and declined to 0.46 [0.25, 0.90] for 1950–1954 birth cohort. Constraint-free nonlinear estimates confirmed a nadir in 1996 (OR = 0.85 [0.82, 0.89]) and a peak for the 1925–1929 birth cohort (OR = 1.37 [1.19, 1.55]). However, the magnitude of the linear trends depended on the assumed age slope. The period-related increase was smaller for the severity-graded outcome (2003, OR = 1.17 vs 1.36). Conclusions: The observed multimorbidity trajectory is compatible with both period-related and birth-cohort interpretations. Chronic disease monitoring should report period and cohort components separately and distinguish quantities that are identified from those that depend on model assumptions.