Case Study: 統計解析に関する査読者のコメントと解説 (4)

4. 統計学的手法の説明

・・・ Kaplan-Meier method and the log-rank test were used to compare differences in recurrence-free and overall survivals between the two groups. ・・・
The statistical test should be given in accordance to the variables to which they were applied. For example, Instead it should be stated which of the outcomes were tested with log rank test, and which with Proportional Hazards Model.
All analyses were done by intention-to-treat. Kaplan-Meier curves were generated for time to first recurrence, recurrence-free survival, disease-specific survival, and overall survival. We compared curves for the two groups with the log-rank test. All time estimates were done with the date of hepatectomy as the baseline. All patients were followed up for at least 3 years as of December, 1998, or until death.

We used Cox’s proportional-hazards model to estimate risk reduction in each of six recurrence types for competing risks (ie, when assessing risk of early recurrence, patients who did not have recurrence by the end of 2 years were not included; when assessing risks of four types of recurrence, the other three types of recurrence were treated as censored at the date of diagnosis).
推 定や検定を行うための統計手法は、解析の対象となる変数や評価尺度と対応づけて記述します。ここでは、生存率を推定するためのKaplan-Meier法 が推定手法として挙げられていますが、first recurrence, recurrence-free survival, disease-specific survival and overall survivalの評価尺度を使ったことが記述されています。群を分けるための変数の記載はありませんが、特定の変数のみを使用する場合には変数名も記述 すべきでしょう。ちなみに、初回原稿のsurvivalsはsurvivalが正しいと査読者からMinor commentで指摘を受けています。

Coxの比例ハザードモデルによる生存率解析では、通常、評価尺度は生死(死亡は全死因の場合と特定の死因による場合の2種類ありますが)で生存と死亡の 2つのカテゴリーしかありません。評価尺度が本論文のように3カテゴリー以上(ここでは6種類の再発)ある場合には競合リスクモデルによる分析が必要とな ります。競合リスクモデルの解析方法を丁寧に説明した部分がこの文です。
<< 前へ  | 目次 | 次へ >>