grad-systematic-review
依 PRISMA 做可重現的系統性文獻回顧
中文翻譯
--- name: "grad-systematic-review" description: "依照 PRISMA 框架進行系統性文獻回顧，包含明確的搜尋策略、納入與排除標準、品質評估與透明的綜合。當使用者需要設計可重現的文獻搜尋、套用 PRISMA 流程記錄、制定納入與排除標準、評估研究品質，或問「系統性回顧怎麼做」「PRISMA 是什麼」「怎麼讓我的文獻回顧可重現」時使用。" metadata: category: "WP-32 質性方法" tags: ["systematic-review", "PRISMA", "search-strategy", "inclusion-exclusion", "quality-assessment", "meta-synthesis", "evidence-synthesis", "reproducibility"] --- # 系統性文獻回顧（PRISMA） ## 概述 系統性回顧使用明確、事先定義好的方法，來辨識、篩選、評讀並綜合針對某個特定問題的所有相關研究。與敘述性回顧不同，系統性回顧遵循可重現的研究計畫，把研究篩選與詮釋上的偏誤降到最低。PRISMA（Preferred Reporting Items for Systematic Reviews and Meta-Analyses）框架提供標準的報告結構，包含那張標誌性的流程圖，追蹤紀錄經過辨識、篩選、資格評估到納入的過程。 ## 何時使用 - 針對定義明確的研究問題，綜合所有證據 - 找出文獻中的缺口、矛盾或共識 - 在設計新的實證研究之前，先釐清目前已知的內容 - 用實證導向的綜合，為政策或實務指引提供依據 ## 何時不要使用 - 研究問題太廣，無法訂出清楚的納入標準 - 範圍型回顧（描繪整體概況）比系統性回顧（回答特定問題）更合適 - 時間限制使你無法執行所需的嚴謹計畫 - 這個主題已發表的研究很少（先考慮範圍型回顧） ## 假設 ``` 鐵律：系統性回顧必須「可重現」：每一個搜尋決定、納入標準、 品質評估都必須記錄下來，讓另一位研究者能複製整個流程。 如果你的回顧無法被複製，它就是敘述性回顧，「不是」系統性回顧。 ``` 關鍵假設： 1. 透明與可重現，是區分系統性回顧與敘述性回顧的關鍵 2. 事先註冊研究計畫，可減少研究篩選與分析中的偏誤 3. 至少要有兩位獨立的審查者來篩選與評估研究，以減少主觀偏誤 4. 對納入的研究做品質評估是必須的，不是所有證據都同等重要 ## 方法 ### 步驟 1：定義研究問題與計畫 用一個框架來擬定聚焦的問題（介入研究用 PICO、質性研究用 PEO、混合方法用 SPIDER）。註冊研究計畫（例如 PROSPERO）。定義資料庫、搜尋詞、日期範圍與語言限制。 | 框架 | 組成 | |-----------|-----------| | **PICO** | Population（族群）、Intervention（介入）、Comparison（比較）、Outcome（結果） | | **PEO** | Population（族群）、Exposure（暴露）、Outcome（結果） | | **SPIDER** | Sample（樣本）、Phenomenon of Interest（關注現象）、Design（設計）、Evaluation（評估）、Research type（研究類型） | ### 步驟 2：執行搜尋策略 至少搜尋 3 個資料庫（例如 Scopus、Web of Science、PubMed）。使用布林運算子（AND、OR、NOT），搭配控制詞彙與自由文字詞。記錄每一條搜尋式與日期。再以引文追蹤（往前與往後）、灰色文獻，以及手動搜尋重點期刊作為補充。 ### 步驟 3：篩選並選擇研究 分兩個階段套用納入/排除標準： 1. **標題與摘要篩選**：兩位審查者獨立進行，意見不一致時靠討論或第三位審查者解決 2. **全文篩選**：把標準套用到完整論文上，記錄排除的理由 用 PRISMA 流程圖記錄整個過程： - 辨識出的紀錄 → 移除重複 → 篩選 → 符合資格 → 納入 ### 步驟 4：擷取資料、評估品質並綜合 把資料擷取到標準化的表單中。用合適的工具評估品質（例如 RCT 用 Cochrane RoB、質性研究用 CASP、JBI 檢核表）。透過統合分析（量化）、主題綜合（質性）或敘述性綜合來整合。依照 PRISMA 2020 檢核表撰寫報告。 ## 輸出格式 ```markdown ## 系統性回顧：[研究問題] ### 研究計畫 - 問題框架：[PICO/PEO/SPIDER] - 註冊：[PROSPERO ID 或同等的註冊] - 搜尋的資料庫：[列表] - 日期範圍：[起-訖] ### 搜尋策略 | 資料庫 | 搜尋式 | 找到的紀錄 | |----------|--------------|---------------| | [名稱] | [布林查詢] | [N] | ### PRISMA 流程 - 辨識出：[N] 筆紀錄 - 移除重複：[N] - 篩選（標題/摘要）：[N] - 篩選時排除：[N] - 評估全文：[N] - 全文階段排除：[N]（理由：...） - 納入綜合：[N] ### 納入/排除標準 | 準則 | 納入 | 排除 | |-----------|---------|---------| | 族群 | [說明] | [說明] | | 研究類型 | [說明] | [說明] | | 語言 | [說明] | [說明] | | 日期 | [說明] | [說明] | ### 品質評估摘要 | 研究 | 使用的工具 | 整體評等 | 主要疑慮 | |-------|-----------|---------------|--------------| | [作者, 年份] | [RoB/CASP/JBI] | [高/中/低] | [具體問題] | ### 綜合 - [主要發現 1，並註明證據強度] - [主要發現 2，並註明證據強度] - 找出的缺口：[仍然未知的部分] ### 限制 - [搜尋的限制] - [評估的限制] ``` ## 常見陷阱 - 沒有 PRISMA 流程圖的系統性回顧是不完整的，流程圖不是選配 - 事先註冊研究計畫，可避免事後更改納入標準而引入偏誤 - 必須考慮灰色文獻（學位論文、研討會論文、報告），以減少發表偏誤 - 品質評估「不是」通過/不通過，它是用來決定在綜合時要給每項研究多少權重 - 「不要」把系統性回顧和統合分析混為一談，統合分析（統計合併）只是系統性回顧中一種可能的綜合方法 - 篩選與品質評估的評分者間信度應該要報告（例如 Cohen's kappa） ## 參考資料 - Page, M. J., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. *BMJ*, 372, n71. - Higgins, J. P. T., et al. (Eds.). (2023). *Cochrane Handbook for Systematic Reviews of Interventions* (Version 6.4). Cochrane. - Booth, A., Sutton, A., & Papaioannou, D. (2016). *Systematic Approaches to a Successful Literature Review* (2nd ed.). Sage.
英文原文
--- name: "grad-systematic-review" description: "Conduct a systematic literature review following the PRISMA framework with explicit search strategy, inclusion and exclusion criteria, quality assessment, and transparent synthesis. Use this skill when the user needs to design a reproducible literature search, apply PRISMA flow documentation, develop inclusion and exclusion criteria, assess study quality, or when they ask 'how do I do a systematic review', 'what is PRISMA', or 'how do I make my literature review reproducible'." metadata: category: "WP-32 質性方法" tags: ["systematic-review", "PRISMA", "search-strategy", "inclusion-exclusion", "quality-assessment", "meta-synthesis", "evidence-synthesis", "reproducibility"] --- # Systematic Literature Review (PRISMA) ## Overview A systematic review uses explicit, pre-defined methods to identify, select, appraise, and synthesize all relevant research on a specific question. Unlike narrative reviews, systematic reviews follow a reproducible protocol that minimizes bias in study selection and interpretation. The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) framework provides the standard reporting structure, including the iconic flow diagram tracking records through identification, screening, eligibility, and inclusion. ## When to Use - Synthesizing the totality of evidence on a well-defined research question - Identifying gaps, contradictions, or consensus in a body of literature - Establishing what is known before designing new empirical research - Informing policy or practice guidelines with evidence-based synthesis ## When NOT to Use - When the research question is too broad to define clear inclusion criteria - When a scoping review (mapping the landscape) is more appropriate than a systematic review (answering a specific question) - When time constraints prevent the rigorous protocol required - When there is very little published research on the topic (consider a scoping review first) ## Assumptions ``` IRON LAW: A systematic review must be REPRODUCIBLE — every search decision, inclusion criterion, and quality assessment must be documented so another researcher can replicate the process. If your review cannot be replicated, it is a narrative review, NOT a systematic review. ``` Key assumptions: 1. Transparency and reproducibility distinguish systematic from narrative reviews 2. A pre-registered protocol reduces bias in study selection and analysis 3. At least two independent reviewers should screen and assess studies to reduce subjective bias 4. Quality assessment of included studies is mandatory — not all evidence is equal ## Methodology ### Step 1: Define the Research Question and Protocol Formulate a focused question using a framework (PICO for interventions, PEO for qualitative, SPIDER for mixed methods). Register the protocol (e.g., PROSPERO). Define databases, search terms, date ranges, and language restrictions. | Framework | Components | |-----------|-----------| | **PICO** | Population, Intervention, Comparison, Outcome | | **PEO** | Population, Exposure, Outcome | | **SPIDER** | Sample, Phenomenon of Interest, Design, Evaluation, Research type | ### Step 2: Execute the Search Strategy Search at least 3 databases (e.g., Scopus, Web of Science, PubMed). Use Boolean operators (AND, OR, NOT) with controlled vocabulary and free-text terms. Document every search string and date. Supplement with citation chaining (forward and backward), grey literature, and hand-searching key journals. ### Step 3: Screen and Select Studies Apply inclusion/exclusion criteria in two phases: 1. **Title and abstract screening** — two reviewers independently; resolve disagreements by discussion or third reviewer 2. **Full-text screening** — apply criteria to full papers; record reasons for exclusion Document the process in a PRISMA flow diagram: - Records identified → duplicates removed → screened → eligible → included ### Step 4: Extract Data, Assess Quality, and Synthesize Extract data into a standardized form. Assess quality using appropriate tools (e.g., Cochrane RoB for RCTs, CASP for qualitative, JBI checklists). Synthesize via meta-analysis (quantitative), thematic synthesis (qualitative), or narrative synthesis. Report per PRISMA 2020 checklist. ## Output Format ```markdown ## Systematic Review: [Research Question] ### Protocol - Question framework: [PICO/PEO/SPIDER] - Registration: [PROSPERO ID or equivalent] - Databases searched: [list] - Date range: [start-end] ### Search Strategy | Database | Search String | Records Found | |----------|--------------|---------------| | [name] | [Boolean query] | [N] | ### PRISMA Flow - Identified: [N] records - Duplicates removed: [N] - Screened (title/abstract): [N] - Excluded at screening: [N] - Full-text assessed: [N] - Excluded at full-text: [N] (reasons: ...) - Included in synthesis: [N] ### Inclusion/Exclusion Criteria | Criterion | Include | Exclude | |-----------|---------|---------| | Population | [specification] | [specification] | | Study type | [specification] | [specification] | | Language | [specification] | [specification] | | Date | [specification] | [specification] | ### Quality Assessment Summary | Study | Tool Used | Overall Rating | Key Concerns | |-------|-----------|---------------|--------------| | [author, year] | [RoB/CASP/JBI] | [high/moderate/low] | [specific issues] | ### Synthesis - [Key finding 1 with evidence strength] - [Key finding 2 with evidence strength] - Gaps identified: [what remains unknown] ### Limitations - [Search limitations] - [Assessment limitations] ``` ## Gotchas - A systematic review without a PRISMA flow diagram is incomplete — the flow diagram is not optional - Pre-registering the protocol prevents post-hoc changes to inclusion criteria that introduce bias - Grey literature (theses, conference papers, reports) must be considered to reduce publication bias - Quality assessment is NOT pass/fail — it informs how much weight to give each study in the synthesis - Do NOT conflate systematic review with meta-analysis; meta-analysis (statistical pooling) is one possible synthesis method within a systematic review - Inter-rater reliability for screening and quality assessment should be reported (e.g., Cohen's kappa) ## References - Page, M. J., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. *BMJ*, 372, n71. - Higgins, J. P. T., et al. (Eds.). (2023). *Cochrane Handbook for Systematic Reviews of Interventions* (Version 6.4). Cochrane. - Booth, A., Sutton, A., & Papaioannou, D. (2016). *Systematic Approaches to a Successful Literature Review* (2nd ed.). Sage.