Table 1.
Hypothesis number | Outcome variables | Primary predictor variables | Hypothesis description | Data sources | Analytical models |
---|---|---|---|---|---|
Hypothesis 1.1 | Retention in care (no missed clinic visits vs. at least 1 missed clinic visit) | Condition (MI vs. SOC) | At each follow-up timepoint, the MI condition clinics will have a higher proportion of patients retained in care, as compared to those from the SOC condition clinics | Clinic/medical records | Repeated measures logistic regression models (1) “Unadjusted” analysis: predictors include time, condition, and interaction between time and condition (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus time-varying measurements of self-efficacy, depression, and substance use as covariates |
Hypothesis 1.2 | Medication adherence (percentage adherence in the last 30 days) | Condition (MI vs. SOC) | At each follow-up timepoint, patients from the MI condition clinics will have higher medication adherence, as compared to those from the SOC condition clinics | Adherence self-report (Visual Analogue Scale) | Linear mixed models (1) “Unadjusted” analysis: predictors include time, condition, and interaction between time and condition (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus time-varying measurements of self-efficacy, depression, and substance use as covariates |
Hypothesis 1.3 | Viral suppression | Condition (MI vs. SOC) | At each follow-up timepoint, the MI condition clinics will have a higher proportion of virally suppressed patients, as compared to those from the SOC condition clinics | Viral load (medical records) | Repeated measures logistic regression models (1) “Unadjusted” analysis: predictors include time, condition, and interaction between time and condition (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus time-varying measurements of self-efficacy, depression, and substance use as covariates |
Hypothesis 1.4 | Medication persistence (time to treatment discontinuation – in months) | Condition (MI vs. SOC) | Throughout the 24-month study period, patients from the MI condition clinics will be more likely to maintain medication persistence, as compared to those from the SOC condition clinics | Pharmacy pickups (pharmacy records) | Frailty models (extension of the Cox regression model to include random effects) To investigate hazard functions of main outcomes by conditions, we will calculate the hazard ratio for discontinuation between the MI vs. SOC condition. (1) “Unadjusted” analysis: predictors include time, condition, and interaction between time and condition (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus time-varying measurements of self-efficacy, depression, and substance use as covariates |
Hypothesis 2.1 | Retention in care (no missed clinic visits vs. at least 1 missed clinic visit) | Physician implementation of MI strategies (count variable) | Within the MI condition, increased implementation of MI strategies by physicians will be associated with increased likelihood of their patients being retained in care | • Clinic/medical records • Patient visit videos (MITI coding) |
Repeated measures logistic regression models (1) “Unadjusted” analysis: predictors include time, MI implementation, and interaction between time and MI implementation (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus psychosocial covariates |
Hypothesis 2.2 | Medication adherence (percentage adherence in the last 30 days) | Physician implementation of MI strategies (count variable) | Within the MI condition, increased implementation of MI strategies by physicians will be associated with increased medication adherence by their patients | • Adherence self-report (Visual Analog Scale) • Patient visit videos (MITI coding) |
Linear mixed models (1) “Unadjusted” analysis: predictors include time, MI implementation, and interaction between time and MI implementation (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus psychosocial covariates |
Hypothesis 2.3 | Viral suppression | Physician implementation of MI strategies (count variable) | Within the MI condition, increased implementation of MI strategies by physicians will be associated with increased likelihood of their patients achieving viral suppression | • Viral load (medical records) • Patient visit videos (MITI coding) |
Repeated measures logistic regression models (1) “Unadjusted” analysis: predictors include time, MI implementation, and interaction between time and MI implementation (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus psychosocial covariates |
Hypothesis 2.4 | • Retention in care (no missed clinic visits vs. at least 1 missed clinic visit) • Medication adherence (percentage adherence in the last 30 days) • Viral suppression |
• Self-efficacy • Motivation • Patient-provider relationship satisfaction |
Within the MI condition, greater self-efficacy, motivation, and higher patient-physician relationship satisfaction will be associated with increased likelihood of being retained in care, maintaining medication adherence, and achieving viral suppression | • Clinic/medical records • Adherence self-report (Visual Analog Scale) • Viral load (medical records) • HIV-ASES • LifeWindows • Prerana Interview |
Repeated measures logistic regression models (for binary outcomes) and linear mixed models (for continuous outcome) (1) “Unadjusted” analysis: predictors include time, self-efficacy, motivation, and patient-provider relationship (2) “Adjusted” analysis: includes predictors from unadjusted analysis, plus psychosocial covariates. To identify the optimal predictors in the hypothesized models, a model including a full-factorial combination of all predictors will be fit and reduced to a final optimal model utilizing an appropriate model selection strategy (e.g., backwards elimination) |
Exploratory analysis 1 | Physician implementation of MI strategies (count variable) | Time | Explore how successfully implementation of MI is sustained over time | Patient visit videos (MITI coding) | Poisson regression |
Exploratory analyses 2.1–2.3 | • Retention in care (no missed clinic visits vs. at least 1 missed clinic visit) • Medication adherence (percent adherence in the last 30 days) • Viral suppression |
• Time • Patient-level skills (self-efficacy, motivation, patient-provider relationship) as potential mediators • Provider-level skills (MI implementation) as a potential mediator |
Exposure to the MI training/exposure to physicians trained in MI will result in increases in patient-level and provider-level skills over time, which will be related to subsequent improvements in patient outcomes (i.e., higher rates of retention in care, greater adherence, higher rates of viral suppression). | • Clinic/medical records • Adherence self-report (Visual Analogue Scale) • Viral load (medical records) • HIV-ASES • LifeWindows • Prerana Interview • Patient visit videos (MITI coding) |
Time-lagged path models Mediation model: examine skills targeted by the MI intervention (both patient-level and provider-level) as potential mediators of patient outcomes. 1. “Unadjusted” mediation analysis: predictors include primary predictor variables and physician characteristics 2. “Adjusted” mediation analysis: includes predictors from unadjusted analysis, plus psychosocial covariates |