Skip to main content

Table 2 Estimated effect of industry marketing for opioid products on physicians’ opioid prescribing rate using g-computation model adjusting for time-varying confounders

From: Bias amplification in the g-computation algorithm for time-varying treatments: a case study of industry payments and prescription of opioid products

Receipt of industry payments for opioids in 2016

Receipt of industry payments for opioids in 2017

Number of physicians

Adjusted mean difference (95% CI) in opioid prescribing rate in 2018

Model 1

Model 2

No

No

221,787

Ref

Ref

Yes

No

10,826

+ 11.32% (9.87 to 12.78)

+ 7.21% (3.95 to 10.47)

No

Yes

5773

7.43% (5.51 to 9.35)

+ 3.63% (1.54 to 5.71)

Yes

Yes

12,558

+ 15.96% (15.24 to 16.69)

+ 13.47% (12.20 to 14.73)

  1. Model 1 includes physician characteristics (years in practice, sex, specialty, the medical school graduated), patients’ characteristics (average age of beneficiaries, proportion of male beneficiaries, average hierarchical condition category score of beneficiaries), and opioid prescribing rate in 2016. Model 2 included receipt of industry marketing for non-opioids in 2016 in addition to covariates in Model 1. Both models adjusted for time-varying confounder (i.e., opioid prescribing rate in 2017) using the g-computation algorithm. The 95% CIs were estimated by repeating the analyses on 200 bootstrapped samples