GEE logistic regression models for quality of care over time
| Measure | OR | 95% CI | P value | Model QIC* |
| First cohort VHWs, overall correct care† | 14 587 | |||
| Months since iCCM services initiation | 0.994 | 0.988 to 0.999 | 0.032 | |
| First cohort VHWs, overall correct care, with spline knot at 3 years after iCCM care initiation† | 14 524 | |||
| Months since iCCM services initiation, months 14–36 | 1.022 | 1.005 to 1.038 | 0.009 | |
| Months since iCCM services initiation, months 36–70 | 0.978 | 0.970 to 0.986 | <0.001 | |
| First cohort VHWs, overall correct care excluding referral of patients with negative RDT† | 11 747 | |||
| Months since iCCM services initiation | 0.999 | 0.993 to 1.004 | 0.632 | |
| First cohort VHWs, overall correct care excluding referral of patients with negative RDT, with spline knot at 3 years after iCCM care initiation† | 11 718 | |||
| Months since iCCM services initiation - Months 14–36 | 1.023 | 1.005 to 1.043 | 0.015 | |
| Months since iCCM services initiation - Months 36–70 | 0.984 | 0.976 to 0.993 | <0.001 | |
| Second cohort VHWs, overall correct care | 6057 | |||
| Months since iCCM services initiation | 1.007 | 0.989 to 1.025 | 0.475 | |
| Second cohort VHWs, overall correct care, with spline knot at 6 months after iCCM care initiation | 6061 | |||
| Months since iCCM services initiation, months 1–6 | 1.068 | 0.971 to 1.175 | 0.175 | |
| Months since iCCM services initiation, months 7–30 | 0.998 | 0.973 to 1.022 | 0.845 | |
| Second cohort VHWs, overall correct care, with spline knot at 18 months after iCCM care initiation | 6047 | |||
| Months since iCCM services initiation, months 1–18 | 1.027 | 1.004 to 1.050 | 0.023 | |
| Months since iCCM services initiation, months 19–30 | 0.963 | 0.926 to 1.000 | 0.053 | |
| Second cohort VHWs, overall correct care excluding referral of patients with negative RDT | 5272 | |||
| Months since iCCM services initiation | 1.026 | 1.007 to 1.045 | 0.006 | |
*Quasi-likelihood under the independence model criterion. This is a modification of the Akaike information criterion so that it can be applied to GEE regression models to assess goodness of fit of different models. A lower QIC term reflects a better-fitting regression model.
†One VHW joined the first cohort of VHWs later and is thus excluded from this analysis.
GEE, generalised estimating equation; iCCM, integrated community case management; QIC, quasi-likelihood under the independence criterion; RDT, rapid diagnostic test; VHW, village health worker.