@techreport{oai:ipsj.ixsq.nii.ac.jp:00225715,
 author = {北村, 憲太 and Irvan, Mhd and 山口, 利恵},
 issue = {7},
 month = {May},
 note = {医療においては，論文に記載されるなどで一般に手に入るデータに patient characteristics と呼ばれる統計データがある．このデータ形式を用いてクロス集計表という医学的エビデンスの生成に使われるデータの推定を行う手法として，Cross-tabulation Table Estimation from Multiple Patient characteristics（CTEMP）が提案されている．CTEMP では，複数の patient characteristics を用いてクロス集計表の推定を行うが，複数の patient characteristics を集める場合，匿名性侵害のリスクが生じる．そこで本報告では，発行する複数の patient characteristics から匿名性が低い patient characteristics をマハラノビス距離を用いて削除する Mahalanobis' distance CTEMP（MCTEMP）を提案する．MCTEMP の匿名性は，patient characteristics の匿名性として提案されている Patient Family Detect on Overall Category（PFDOC）entropy を用いて評価し，有用性は推定精度で実験的に評価した．実験では，マハラノビス距離を小さくすることで匿名性が任意に確保されたが，推定精度は CTEM と比べて明らかな低下を確認できなかった．, In the field of medicine, there is a type of statistical data called ”patient characteristics,” which is generally available through research papers. Cross-tabulation Table Estimation from Multiple Patient characteristics (CTEMP) has been proposed as a method for generating medical evidence using this data format. CTEMP estimates the data using multiple patient characteristics, but collecting multiple patient characteristics poses a risk of violating anonymity. Therefore, in this paper, we propose a Mahalanobis' distance CTEMP (MCTEMP) method that removes patient characteristics with low anonymity from multiple patient characteristics using Mahalanobis distance. The anonymity of the proposed method is evaluated using PFDOC entropy proposed as the anonymity of patient characteristics, and usefulness is evaluated experimentally in estimation accuracy. In the experiment, anonymity was arbitrarily ensured by reducing the Mahalanobis distance, but the estimation accuracy did not show a clear decrease compared to CTEM.},
 title = {マハラノビス距離を利用した複数のpatient characteristicsからのクロス集計表推定法},
 year = {2023}
}