【正文】
標(biāo)準(zhǔn)差; ?正態(tài)分布數(shù)據(jù)進(jìn)行 Meta分析是安全的,偏態(tài)分布數(shù)據(jù)需特殊處理; ?若各研究使用統(tǒng)一的單位或量表,則對平均值差進(jìn)行 Meta分析; ?若各研究的單位或量表不同,則對標(biāo)準(zhǔn)平均值差進(jìn)行 Meta分析。 數(shù)據(jù)分析 —— 異質(zhì)性分析(一) ?異質(zhì)性:各研究結(jié)果之間的差異,包括: 多樣性 ( Diversity) : 各研究之間真實的差異 , 如因研究對象 , 干預(yù)措施 、 劑量等不同引起的差異; 偏倚 ( Bias) : 如因研究設(shè)計 、 質(zhì)量控制等引起的差異 。 數(shù)據(jù)分析 —— 異質(zhì)性分析(二) ?異質(zhì)性檢驗 統(tǒng)計檢驗 : χ 2檢驗 , Q值越大,異質(zhì)性越大 量化異質(zhì)性: I2值, I2值越大,異質(zhì)性越大 Q ?I2 = ?異質(zhì)性較大 數(shù)據(jù)分析 —— 異質(zhì)性分析(三) ?異質(zhì)性較大時 檢查提取的數(shù)據(jù) : 更換統(tǒng)計方法: 異質(zhì)性不明顯時:考慮固定效應(yīng)模型( fixed effect model) 異質(zhì)性明顯時:考慮不做 Meta分析,或隨機(jī)效應(yīng)模型( randomeffect model) 尋找異質(zhì)性原因: 亞組分析、 meta回歸 等 數(shù)據(jù)分析 —— 異質(zhì)性分析(四) ?固定效應(yīng)模型 假定所有研究的治療效果均相同; 各研究間的不同結(jié)果僅由隨機(jī)誤差引起。 ?隨機(jī)效應(yīng)模型 各研究之間的治療效果均不同; 各研究之間的治療效果服從一定的分布(一般為正態(tài)分布)。 SR7:報告的偏倚 Entry Judgement Support for judgement Random sequence generation (selection bias) Low risk. Quote: “patients were randomly allocated.” Comment: Probably done, since earlier reports from the same investigators clearly describe use of random sequences (Cartwright 1980). Allocation concealment (selection bias) High risk. Quote: “...using a table of random numbers.” Comment: Probably not done. Blinding of participants and personnel (performance bias) Low risk. Quote: “double blind, double dummy”。 “High and low dose tablets or capsules were indistinguishable in all aspects of their outward appearance. For each drug an identically matched placebo was available (the success of blinding was evaluated by examining the drugs before distribution).” Blinding of oute assessment (detection bias) (patientreported outes) Low risk. Quote: “double blind”. Comment: Probably done. Blinding of oute assessment (detection bias) (Mortality) Low risk. Obtained from medical records。 review authors do not believe this will introduce bias. Inplete oute data addressed (attrition bias) (Shortterm outes (26 weeks)) High risk. 4 weeks: 17/110 missing from intervention group (9 due to 39。lack of efficacy39。)。 7/113 missing from control group (2 due to 39。lack of efficacy39。). Inplete oute data addressed (attrition bias) (Longerterm outes (6 weeks)) High risk. 12 weeks: 31/110 missing from intervention group。 18/113 missing from control group. Reasons differ across groups. Selective reporting (reporting bias) High risk. Three rating scales for cognition listed in Methods, but only one (with statistically significant results) is reported. ?漏斗圖 ?Funnel plot SR8:陳述結(jié)果 ? 森林圖 ? 結(jié)果總結(jié)表 ? SR目的是為衛(wèi)生決策服務(wù),所以同時要用科普語言 SR9:解釋結(jié)果,得出結(jié)論 :如發(fā)表偏倚 效應(yīng)大小,變異 比較其它綜述 外延性:結(jié)果還適合于哪些患者