評估結核病家庭災難性負擔測量之調查工具

Abstract

背景 結核病病人支出調查是監測世界衛生組織（World Health Organization, WHO）「終結結核」（End TB Strategy）目標進展的主要工具；該目標要求至2030年，沒有任何受結核病影響的家庭面臨災難性負擔（catastrophic total costs, CTC）。此類調查仰賴受訪者回溯性自陳其醫療利用情形，並依據不同長度的回憶期間加乘推估至完整治療期間。然而，這種方法尚未以登記申報資料作為標準進行驗證。臺灣全民健康保險（National Health Insurance, NHI）具有近乎全民涵蓋率及完整的電子申報紀錄，提供了彌補此研究缺口的獨特機會。 方法 本研究將2018年於臺灣執行、符合WHO方法學之全國性結核病病人支出調查資料（n = 516），與個人層級全民健保行政資料中之門診回診次數與住院天數進行串聯。研究共進行三項分析。第一，使用McNemar檢定比較調查資料與健保資料估計之災難性負擔比例。第二，辨識回報誤差絕對值最低的回憶期間；此分析將總回報誤差分解為三個部分，包括記憶誤差、由記憶誤差經加乘推估所造成的外推誤差，以及由醫療利用頻率變動所造成的外推誤差，並針對門診回診次數與住院天數進行分組描述性分析與局部加權散點圖平滑法（LOESS）分析，另於持續期門診病人中使用二次迴歸進一步估計最佳回憶期間。第三，使用邏輯斯迴歸評估低收入病人是否較可能低報醫療利用。 結果 調查資料估計之災難性負擔比例為26.6%，顯著高於健保資料估計之19.8%（p < 0.001）。整體而言，兩種資料來源之分類一致比例為88.2%；另有9.3%僅在調查資料中被判定為災難性負擔，2.5%僅在健保資料中被判定為災難性負擔，顯示不一致的主要方向為調查高估。此高估主要由住院天數高報所驅動，其影響最大程度反映於間接支出。調查與健保估計之差距在多重抗藥性結核病（multidrug-resistant TB, MDR-TB）病人中較為明顯（55.0% vs. 16.1%），而在一般結核病（drug-susceptible TB, DS-TB）病人中差距較小（23.0% vs. 20.2%）。回報誤差絕對值與回憶期間呈U型關係：在較短的回憶期間中，加乘推估誤差佔主導；而在較長的回憶期間中，記憶衰退造成的誤差則逐漸增加。二次迴歸（quadratic regression）結果顯示，最佳回憶期間約為106至158天，其他分析結果亦支持此一發現。低收入與醫療利用低報之間未達統計顯著關聯（OR = 1.161，95% CI: 0.679–1.984，p = 0.586）。 結論 本研究結果顯示，WHO建議之結核病病人支出調查工具在臺灣會系統性高估災難性負擔比例。未來仍需進一步研究，評估將訪談安排於3至5個月的回憶期間內，是否能有效降低此偏誤。至於本研究情境下是否存在社會經濟地位造成的報告準確性差異，仍需在較大樣本中進一步檢驗。這些發現凸顯在使用結核病病人支出調查結果作為政策依據之前，需先對調查工具進行驗證與改善。

Background Tuberculosis (TB) patient cost surveys are the primary tool for monitoring progress toward the WHO End TB target that no TB-affected household should face catastrophic total costs (CTC) by 2030. These surveys rely on retrospectively self-reported health care utilisation recalled across variable periods and extrapolated to full treatment duration, yet this approach has not been validated against an administrative standard. Taiwan's National Health Insurance (NHI), with near-universal coverage and comprehensive electronic claims records, offers a unique opportunity to address this gap. Methods This study linked a WHO-methodology-aligned national TB patient cost survey (n=516) conducted in Taiwan in 2018 to individual-level NHI administrative records of outpatient visits and inpatient bed-days. Three analyses were conducted: comparison of survey-based and NHI-based catastrophic total costs prevalence using McNemar's test; identification of the recall period associated with minimum absolute reporting error, using a novel three-component decomposition of total reporting error (memory error, extrapolation error by memory, and extrapolation error by frequency), descriptive categorical and LOESS analyses across both outpatient visits and inpatient bed-days, and quadratic regression among continuation-phase outpatient patients; and assessment of whether low-income patients differentially underreported health care utilisation using logistic regression. Results Survey-based estimation yielded a catastrophic total costs prevalence of 26.6%, significantly exceeding the NHI-based estimate of 19.8% (p0.001). Overall, 88.2% of households were concordantly classified by the two approaches, while 9.3% were classified as catastrophic by the survey-based estimate only and 2.5% were classified as catastrophic by the NHI-based estimate only, indicating that overestimation was the dominant direction of discordance. This overestimation was driven primarily by inpatient bed-day overreporting, with the largest effect operating through indirect cost components. The survey–NHI discrepancy was minimal among drug-susceptible TB patients (23.0% vs. 20.2%) but substantial among multidrug-resistant TB patients (55.0% vs. 16.1%). Absolute reporting error followed a U-shaped relationship with recall period: extrapolation error predominated at short windows while memory decay increased at longer ones. Quadratic regression identified an optimal recall window of 106–158 days, and other results were consistent with this finding. Low income was not significantly associated with underreporting (OR=1.161, 95% CI: 0.679–1.984, p=0.586). Conclusion These findings indicate that the WHO-recommended TB patient cost survey instrument systematically overestimates catastrophic total costs prevalence in Taiwan. Future research should examine whether scheduling interviews within a 3–5 month window can reduce this bias. Whether socioeconomic gradients in reporting accuracy exist in this setting requires investigation in larger samples. These findings highlight the need to validate and refine TB patient cost survey instruments before using them to guide policy.