Future data analysis

In observational studies like PRIMIER, there is no treatment randomization. Therefore, to minimize the effects of potential stable moderators (e.g., sex), channeling bias, and/or time- varying confounders, a Marginal Structural Model (MSM) approach will be utilized to analyze the final PRIMIER database. An MSM analysis is a weighted repeated measures approach using IM modality as a time-varying covariate as well as accounting for baseline characteristics. Weights produce a pseudo-population with a balance in both time-invariant and time-varying covariates, allowing for causal treatment comparisons using standard repeated measures models. The weighting will also be adjusted to account for missing data, providing validity under missing at random or missing completely at random. In order to incorporate adjustment for patients with missing visits, the same weight approach is used. However, instead of using a flag to designate IM modality, a flag denoting whether the patient remained in the study is used. The final weight for each patient’s observation is computed by multiplying the IM modality selection weights and the censoring weights. There are four major components which need to be computed/assessed to perform a MSM analysis: 1) weight estimates for each subject visit adjusting for IM modality, 2) weight estimates for each subject visit adjusting for study discontinuation, 3) an a priori chosen vector of time-independent variables, such as baseline characteristics, 4) an a priori chosen IM modality or set of IM modalities which will be assigned as the “treatment of interest” and tracked throughout time. Adjusted mean outcomes by study time as well as resulting F-test p-values will be reported by outcome measure.

As our sample size increases and the proportion of those completing long term follow-up assessments grows, we will be able to evaluate the impact of IM on pain scores in subgroups including but not limited to: sex, BMI < 30 vs BMI ≥ 30, and <40 on the PROMIS Depression Subscale vs ≥40 on PROMIS Depression Subscale. The potential of the varying impacts of IM on subgroups will be assessed by including interaction terms, i.e., subgroup by IM, in the MSM model. In the case that the interaction terms are statistically significant, separate MSM analyses will be performed within each subgroup. Also, the weight estimates for each subject visit will be computed adjusting for dose of modality or modalities of interest (e.g., acupuncture or IM physician visit) instead of IM modality in the MSM analysis. Accounting for modality dosing by visit over the course of the study may potentially create more robust weight estimates for each subject by providing more information than just a binary response variable.

Sensitivity analysis to assess the impact of retention rate will be performed on pre-specified subgroups of subjects who completed:

75 % of surveys over the study period

50 % of surveys over the study period

25 % of surveys over the study period

All analyses will be conducted using SAS (Cary, NC).