Upgrade to Pro — share decks privately, control downloads, hide ads and more …

感染症の数理モデル4

 感染症の数理モデル4

Daisuke Yoneoka

April 19, 2024
Tweet

More Decks by Daisuke Yoneoka

Other Decks in Research

Transcript

  1. ⽬次 1. 感染症のコンパートメントモデル 2. 基本再⽣産数 3. 最終流⾏規模 4. R実装 5.

    ⼈⼝の異質性とSIR 6. 再⽣産⽅程式とエボラ vs インフル 7. R0 の推定⽅法(流⾏初期) 8. 内的増殖率の検定 本書の内容をカバーします。 具体的なコードなどは右の本 詳細なプログラムなどは https://github.com/objornstad/epimdr/tree/ master/rcode (結構間違ってる。。。) 2/48
  2. 最終流⾏規模 (Final epidemic size) もうちょっと現実的に⼈⼝あたりで考える SIRには2つの均衡点 (s, i, r) =

    (1, 0 , 0)と(s*, 0, r*) 8/20 t→∞としたときに感染者は0⼈になり、 s* (= s(∞)): 何%が感染を逃れたか r* (= r(∞)): 何%が感染したか <latexit sha1_base64="GZ7GT/t1BrHuSYH75Oq0B+oFT9Y=">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</latexit> ds dt = si di dt = si i dr dt = i <latexit sha1_base64="TWw88aSbJzklo45MPjvq1kp19nw=">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</latexit> s = S N , i = I N , r = R N <latexit sha1_base64="Aw425Lop8YfQ4vugYIF3fQA859A=">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</latexit> dS dt = SI dI dt = SI I dR dt = I <latexit sha1_base64="xH6Y+u0OyWqjHpz+UFtZmTKuzBw=">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</latexit> r⇤ = 1 exp( r⇤R0) <latexit sha1_base64="y4Drxc9cmIJphOIsqsikfGkQF1o=">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</latexit> R0 ⌘ = log(1 r⇤) r⇤ 新しい基本再⽣産数 r*にてつい て解く (解析的には解けないので)数値計算 <latexit sha1_base64="lxjKR111eVZZbnFb7RGt1wYQNmw=">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</latexit> r⇤ ⇡ 1 exp( R0) R0 が⼤きいとき はr*≒1なので R0 だけから、最終的な流⾏規模が⾒積もれる! 求め⽅:上の3番⽬を1番⽬に代⼊し て積分するだけ
  3. R0 の統計的推測 (1) 今までは、γやβは天下り的に与えてきた→データから推定したいよね 再⽣産⽅程式 時刻 tにおける新規感染者数 I(t) は、a⽇前の過去の新規感染者数を I(t−a)を⽤いて

    33 <latexit sha1_base64="KRYtOBHRA80qCdkLhBLyz4qB/bc=">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</latexit> I(t) = Z 1 0 A(a)I(t a)da <latexit sha1_base64="vFZ72HbnVeU8w6LkYvqK9aOFGaY=">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</latexit> R0 = Z 1 0 A(a)da ただし、 <latexit sha1_base64="QvFYNraTbOF3WY09zCRxI8hPGLg=">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</latexit> A(a) R 1 0 A(a)da は世代時間のpdf この解を と⼀旦してみる (rは内的増殖率という) <latexit sha1_base64="g1gRI6Y400KhcDNHvGyOUuELeYQ=">AAAB/XicbVBNS8NAEN34WetXqkcvi0VoQUoiRb0IRS96q2A/oA1ls920SzebsDtRSyn+Eo/qRbz6Szz4b9y2OWjrg4HHezPMzPNjwTU4zre1tLyyurae2chubm3v7Nq5vbqOEkVZjUYiUk2faCa4ZDXgIFgzVoyEvmANf3A18Rv3TGkeyTsYxswLSU/ygFMCRurYuZsCFPEFHrTZY1xQGIodO++UnCnwInFTkkcpqh37q92NaBIyCVQQrVuuE4M3Igo4FWycbSeaxYQOSI+1DJUkZNobTU8f4yOjdHEQKVMS8FT9PTEiodbD0D/2Q9McEujreXsi/ue1EgjOvRGXcQJM0tmuIBEYIjyJAne5YhTE0BBCFTfnYtonilAwgWVNDu7814ukflJyT0vl23K+cpkmkkEH6BAVkIvOUAVdoyqqIYoe0DN6RW/Wk/VivVsfs9YlK53ZR39gff4AmWaTYQ==</latexit> I(t) = k exp(rt) <latexit sha1_base64="YqBVCmEG4Xltxg7eyG3luZYdW48=">AAACG3icbVBNSwMxEM36bf2qevQSLEILWnelqBfBj4vHCrYKbS2zabYNzWaXZFYsxZ/h0V/iUb2IVw/+G9N2D9r6YJjHezMk8/xYCoOu++1MTc/Mzs0vLGaWlldW17LrG1UTJZrxCotkpG99MFwKxSsoUPLbWHMIfclv/O7FwL+559qISF1jL+aNENpKBIIBWqmZ3e/W+UOc1xQL9ITWhcKmS+9sD7BHz/JQSP087kGh0IJmNucW3SHoJPFSkiMpys3sV70VsSTkCpkEY2qeG2OjDxoFk/wxU08Mj4F1oc1rlioIuWn0h4c90h2rtGgQaVsK6VD9vdGH0Jhe6O/6oR0OATtm3B6I/3m1BIPjRl+oOEGu2OitIJEUIzoIiraE5gxlzxJgWtjvUtYBDQxtnBmbgzd+9SSpHhS9w2LpqpQ7PU8TWSBbZJvkiUeOyCm5JGVSIYw8kRfyRt6dZ+fV+XA+R6NTTrqzSf7A+foBwBeeuw==</latexit> k exp(rt) = Z 1 0 A(a)k exp(r(t a))da でないといけない。両辺 で割れば (オイラー=トロカの特性⽅程式) <latexit sha1_base64="g1gRI6Y400KhcDNHvGyOUuELeYQ=">AAAB/XicbVBNS8NAEN34WetXqkcvi0VoQUoiRb0IRS96q2A/oA1ls920SzebsDtRSyn+Eo/qRbz6Szz4b9y2OWjrg4HHezPMzPNjwTU4zre1tLyyurae2chubm3v7Nq5vbqOEkVZjUYiUk2faCa4ZDXgIFgzVoyEvmANf3A18Rv3TGkeyTsYxswLSU/ygFMCRurYuZsCFPEFHrTZY1xQGIodO++UnCnwInFTkkcpqh37q92NaBIyCVQQrVuuE4M3Igo4FWycbSeaxYQOSI+1DJUkZNobTU8f4yOjdHEQKVMS8FT9PTEiodbD0D/2Q9McEujreXsi/ue1EgjOvRGXcQJM0tmuIBEYIjyJAne5YhTE0BBCFTfnYtonilAwgWVNDu7814ukflJyT0vl23K+cpkmkkEH6BAVkIvOUAVdoyqqIYoe0DN6RW/Wk/VivVsfs9YlK53ZR39gff4AmWaTYQ==</latexit> I(t) = k exp(rt) <latexit sha1_base64="J/e5A2LBdnws9TuyTGuk9tcHYis=">AAACDnicbVDLSgNBEJz1GeMr6tHLkCAkoGFXRL0IUS8eI5gHJDH0Tmbj4OzsMtMrhpC7R7/Eo3oRr/6AB//GyeOg0YKmi6puZrr8WAqDrvvlzMzOzS8sppbSyyura+uZjc2qiRLNeIVFMtJ1HwyXQvEKCpS8HmsOoS95zb89H/q1O66NiNQV9mLeCqGrRCAYoJXamaxHT2hTKGy79Nr2AHv0NA+FJr+P83saCh1oZ3Ju0R2B/iXehOTIBOV25rPZiVgScoVMgjENz42x1QeNgkk+SDcTw2Ngt9DlDUsVhNy0+qNbBnTHKh0aRNqWQjpSf270ITSmF/q7fmiHQ8AbM20Pxf+8RoLBcasvVJwgV2z8VpBIihEdZkM7QnOGsmcJMC3sdym7AQ0MbYJpm4M3ffVfUt0veofFg8uDXOlskkiKbJMsyROPHJESuSBlUiGMPJAn8kJenUfn2Xlz3sejM85kZ4v8gvPxDWZHmc8=</latexit> 1 = Z 1 0 A(a) exp( ra)da (上のただしの部分を使うと)よって <latexit sha1_base64="DDEx2FgrbA2KW0FzqNF85JdxCjA=">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</latexit> R0 = 1 Z 1 0 A(a) R 1 0 A(b)db exp( ra)da
  4. R0 の統計的推測 (2) 世代時間のpdfをg(t)と書くと したがって、rとg()の形が決まれば、R0 は推定可能 • 実際の現場的には、g()は過去⽂献から持ってくることが多い • rは右図のようなepi

    curveに対して あたりをfitting • rの推定は別に最⼩⼆乗法とかでもOK 34 <latexit sha1_base64="DDEx2FgrbA2KW0FzqNF85JdxCjA=">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</latexit> R0 = 1 Z 1 0 A(a) R 1 0 A(b)db exp( ra)da <latexit sha1_base64="XFoBPaOXbyk7gBRw5r4TPBJUrdQ=">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</latexit> R0 = 1 R 1 0 g(s) exp( rs)ds = 1 M( r) M()は統計家ならみんな ⼤好きgのモーメント⺟関数 <latexit sha1_base64="g1gRI6Y400KhcDNHvGyOUuELeYQ=">AAAB/XicbVBNS8NAEN34WetXqkcvi0VoQUoiRb0IRS96q2A/oA1ls920SzebsDtRSyn+Eo/qRbz6Szz4b9y2OWjrg4HHezPMzPNjwTU4zre1tLyyurae2chubm3v7Nq5vbqOEkVZjUYiUk2faCa4ZDXgIFgzVoyEvmANf3A18Rv3TGkeyTsYxswLSU/ygFMCRurYuZsCFPEFHrTZY1xQGIodO++UnCnwInFTkkcpqh37q92NaBIyCVQQrVuuE4M3Igo4FWycbSeaxYQOSI+1DJUkZNobTU8f4yOjdHEQKVMS8FT9PTEiodbD0D/2Q9McEujreXsi/ue1EgjOvRGXcQJM0tmuIBEYIjyJAne5YhTE0BBCFTfnYtonilAwgWVNDu7814ukflJyT0vl23K+cpkmkkEH6BAVkIvOUAVdoyqqIYoe0DN6RW/Wk/VivVsfs9YlK53ZR39gff4AmWaTYQ==</latexit> I(t) = k exp(rt)
  5. 感染性の⽐較のための逐次検定法 感染性を⽐較したい (新型インフル vs 季節性インフル) 世代時間が同じくらいとすると、結局は内的増殖率 r 同⼠の⽐較 【問題設定】 ある時間tまでの累積感染者数(C新型

    (t)、C季節 (t))に有意差あるか? But, 時間tが変わるごとに結論が出るまで検定していきたい。 Wald (1945)の逐次確率⽐検定(Sequential probability ratio test, SPRT): • 検定の最適性を満たす (from ネイマン・ピアソンの定理) • Uniformlu most powerful test (UMP): (ざっくり⾔うと)あるType 1 errorの値のもとで、検出⼒が最⼤ になる検定 • もうちょっといいのにMaxSPRTというのもある CDCなどでサーベイランスのアウトブレイク検知などで使⽤ 35 逐次検定〜
  6. (ざっくり)SPRT、逐次確率⽐検定 H0 (新型の⽅が内的増殖率が⼤きい): r新型 >r季節 H1 (季節性インフルの⽅が内的増殖率が⼤きい): r新型 <r季節 ゴール:

    H0とH1のどっちなのか“なる早”で決定したい: i.e., for the smallerst value of t Step 1: 2つの定数を「あなたが」決める γ1 > γ0 Step 2: Λt ≥γ1 になった時点でH1を採択 Λt ≤γ0 になった時点でH0を採択 γ0 ≤Λt ≤γ1 なら決定は保留する→次のtでまた検定する 当たり前だが⼤きいγ1 と⼩さいγ0 を設定すると、決定が保留され続ける代わりに、正確な決定可能 36 <latexit sha1_base64="hJbwm0ZHdgahdRNlldUS840TXCE=">AAAConicfZBdb9MwFIbd8DXKxzq45MaiIA0UVckEbDcTE3CB+BCboFulpkSOc9pZc+zMPkFUUX4UvwZxB/+Ek7ZIbEMcydKj97zH9nmzUiuPUfSjE1y6fOXqtbXr3Rs3b91e723cOfS2chKG0mrrRpnwoJWBISrUMCodiCLTcJSdvGz7R1/AeWXNJ5yXMCnEzKipkgJJSntvk3dkzkWKPGlvM3B6ypPS2Tyt1W7cfCY9nzoh6zKNN0epetQQRUsKOfJdHodbYZJb9GmvHw2iRfGLEK+gz1a1n250ntOgrAowKLXwfhxHJU5q4VBJDU03qTyUQp6IGYwJjSjAT+rF1g1/SErOp9bRMcgX6t8TtSi8nxcZOQuBx/58rxXDrPhXe1zhdGdSK1NWCEYu35pWmqPlbYo8Vw4k6jmBkE7Rd7k8FhQTUtZnHmrvRmu1p21eAW3p4D1JH0pwAq17XCfCzQplGtp6loQt/c8ovv4xEnW7FHl8PuCLcLg1iJ8Nnh486e+9WIW/xu6x+2yTxWyb7bHXbJ8NmWTf2Hf2k/0KHgRvgoPg49IadFYzd9mZCpLfN/XQPg==</latexit> ⇤t := t Y i=1 p1(Xi) p0(Xi) , t = 1, 2, . . .
  7. Reid (1956)の⽅法 Notations - : 時点tの累積患者数の確率変数(1が新型インフル、2が季節性) - は内的増殖率ri 以外は既知の分布 【ゴール】H1:

    r1 < r2 or H2: r1 > r2 Step 1: 上の仮説を改定する。Hʼ1: x1 , x2 の同時分布が Hʼ2: x1 , x2 の同時分布が Step 2: 単純出⽣過程(微⼩時間において1⼈の感染者は⾼々1⼈の感染者を⽣む)を仮定すると、 決定関数は Step 3: αをtype 1 error rateとして 37 <latexit sha1_base64="Nj1TEJwYVA9X7aFpUNFaIHvCTWM=">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</latexit> X1(t), X2(t) <latexit sha1_base64="L/hH7FT+TWtUJbbttQXJu121m+8=">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</latexit> pri (xi, t) = P(Xri (t) = xi), i = 1, 2 <latexit sha1_base64="1WQOQ0pEDO0R2MNAChuOSUgf2Lw=">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</latexit> d(t)  log{↵2/(1 ↵1)} <latexit sha1_base64="m1K5jhgZ/bhf19q6J/dFBHTxu9g=">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</latexit> d(t) log{(1 ↵2)/↵1 } H2を採択 H1を採択 それ以外なら、検定を続ける <latexit sha1_base64="cLs2iVEUNYf3cskjE5j8cKeGbIQ=">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</latexit> pr1 (x1, t)pr2 (x2, t) <latexit sha1_base64="JHzZSVZ8g1xkA6skk823qmFkyJE=">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</latexit> pr2 (x1, t)pr1 (x2, t) <latexit sha1_base64="q7CkWI80aiNYArX32JfPwg1xQxs=">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</latexit> d(t) = [x2(t) x1(t)] log 1 exp( r1t) 1 exp( r2t) <latexit sha1_base64="bEBy+0ErgV3JLFbMr9e+2xAa30I=">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</latexit> p 1 (x, t) = e it(1 e t)x 1, t 0, x 0, X(0) = 1 単純出⽣過程