Algorithm 1 Multi-Variable Hybrid Attentive Model

Input:      Multivariable physiological signals Output:      The predicted result of intravenous intervention: 0 or 1 1:  s = getSeg(S); // split S into M equal length segments 2:  $t{r}_{\left(kt\right)}$ = getDiff(s); // calculate the difference between $s_k$ of all channels 3:  P = conv(s); // convert s into features 4:  O = sum(P); // output of the CNN layer 5:  $\alpha$ = getFluAtt(O); // calculate the fluctuant level attention weights 6:  H = biLSTM($\alpha$O); // convert $\alpha$O into recurrent features 7:  Z = sum(H); // output of the LSTM layer 8:  X = getFull(Z); // convert Z into X through the full connected layer 9:  $\beta$ = getTreAtt(X); // calculate the trend level attention weights 10:${\widehat{y}}_i$ = getPre($\beta$X, $t{r}_{\left(kt\right)}$); // obtain the prediction result