From: Modeling adaptive response profiles in a vaccine clinical trial
2-segment piecewise (PW) linear | Yij=β0i+β1i.tj+β2i.(tj−tKNOT)++εij |
---|---|
 | β0i=γ00+ζ0i |
 | β1i=γ10+ζ1i |
 | β2i=γ20+ζ2i |
 | i: subject index, j: time index |
 | (tj−tKNOT)+ is a derived variable which becomes (tj−tKNOT) only when tj>tKNOT. Otherwise it is 0. |
 | γ00 and ζ0i are fixed and random intercepts |
 | γ10 and ζ1i are fixed and random first segment slopes |
 | γ20 and ζ2i are fixed and random incremental slopes |
3-segment PW linear | Yij=β0i+β1i.tj+β2i.(tj−tKNOT1)++β3i.(tj−tKNOT2)++εij |
 | β3i=γ30+ζ3i |
 | γ30 and ζ3i are also fixed and random incremental slopes |
Cubic | \(Y_{ij} = \beta _{0i} + \beta _{1i}. t_{j} + \beta _{2i}. t_{j}^{2} + \beta _{3i}. t_{j}^{3} + \epsilon _{ij}\) |
 | γ00 and ζ0i are fixed and random intercepts |
 | γ10 and ζ1i are fixed and random first order coefficients |
 | γ20 and ζ2i are fixed and random second order coefficients |
 | γ30 and ζ3i are fixed and random third order coefficients |
2-segment PW quadratic | \(Y_{ij} = \beta _{0i} + \beta _{1i}. (t_{j} - t_{KNOT})_{+} + \beta _{2i}. t_{j}^{2} + \beta _{3i}. (t_{j} - t_{KNOT})_{+}^{2} + \epsilon _{ij}\) |
 | γ00 and ζ0i are fixed and random intercepts |
 | γ10 and ζ1i are fixed and random first order coefficients |
 | γ20 and ζ2i are fixed and random second order coefficients |
 | γ30 and ζ3i are fixed and random incremental second order coefficients |
time as a categorical variable | Yij=γ00.D0+(γ10+ζ1i).D1+(γ20+ζ2i).D2+(γ30+ζ2i).D3+(γ40+ζ3i).D4+(γ50+ζ3i).D5+εij |
 | γ00,10,20,30,40,50 are fixed coefficients |
 | ζ1i,2i,3i are random coefficients shared between time points |
 | D0,1,2,3,4,5 are dummy indicator variables for time points |