Table 3 Inverse Probability of Treatment Weighting (IPTW) diagnostics based on summaries of stabilized weights \(SW_i\) by different specifications of multinomial logistic regressions for the denominator \(\Pr \left( A_i \big | \varvec{L}_i\right)\)
IPTW Specification | Stabilized weights: \(\varvec{SW}_{\varvec{i}}\) | ||
|---|---|---|---|
Method | Descriptiona | Mean (s.d.) | Min/Max |
IPTW 1 | Categorical/binary confounders and binary effect modifier as main effect only; each continuous MOTox score linearly related to the log-odds. | 0.988 (0.663) | 0.311/5.154 |
IPTW 2 | Same as in IPTW 1 + interaction terms for toxicity confounders linearly related to the log-odds. | 0.989 (0.691) | 0.334/5.102 |
IPTW 3 | Same as in IPTW 1 + interaction terms between toxicities and trial linearly related to the log-odds. | 0.977 (0.685) | 0.305/5.245 |
IPTW 4 | Categorical confounders and effect modifier as main effect only; B-spline basis matrix for cubic polynomial splines with three internal knots were used to model the relationship between each continuous MOTox score and the log-odds. | 0.969 (0.813) | 0.306/7.027 |
IPTW 5 | Same as in IPTW 1 + interaction terms between toxicities and histological response linearly related to the log-odds | 0.963 (0.694) | 0.279/8.386 |