TABLE III.
Surgery Type and 12-Month Outcomes: Multivariable, 2-Stage Instrumental Variable Estimation Results*
| Revision/Conversion | Dislocation | |||||
|---|---|---|---|---|---|---|
| Coefficient | Standard Error | P Value | Coefficient | Standard Error | P Value | |
| Surgery type: THA (vs. HA) | −0.008 | 0.033 | 0.81 | 0.17 | 0.03 | <0.001 |
| Other key covariates | ||||||
| Age, in years | ||||||
| 66-69 | Ref. | — | — | Ref. | — | — |
| 70-74 | −0.03 | 0.05 | 0.52 | −0.04 | 0.05 | 0.46 |
| 75-79 | −0.09 | 0.48 | 0.068 | −0.03 | 0.05 | 0.55 |
| 80-84 | −0.18 | 0.05 | <0.001 | −0.11 | 0.05 | 0.04 |
| 85-89 | −0.20 | 0.05 | <0.001 | −0.06 | 0.05 | 0.25 |
| ≥90 | −0.33 | 0.06 | <0.001 | −0.15 | 0.06 | 0.01 |
| Sex | ||||||
| Female | Ref. | — | — | Ref. | — | — |
| Male | 0.11 | 0.02 | <0.001 | −0.01 | 0.03 | 0.59 |
| Race/ethnicity | ||||||
| Non-Hispanic White | Ref. | — | — | Ref. | — | — |
| Black/African American | −0.25 | 0.08 | 0.002 | −0.13 | 0.08 | 0.09 |
| Hispanic | −0.10 | 0.07 | 0.16 | 0.01 | 0.07 | 0.85 |
| Other | −0.23 | 0.09 | 0.01 | −0.21 | 0.09 | 0.02 |
| Low-income status | ||||||
| No | Ref. | — | — | Ref. | — | — |
| Yes | −0.045 | 0.04 | 0.18 | 0.09 | 0.03 | 0.01 |
| No. of comorbidities | 0.02 | 0.004 | <0.001 | 0.03 | 0.004 | <0.001 |
| Cementation | −0.23 | 0.03 | <0.001 | −0.05 | 0.03 | 0.07 |
| Overall facility volume of THAs | −0.0003 | 0.0001 | 0.02 | −0.0004 | 0.0001 | <0.001 |
For each model, the dependent variable is a binary indicator given the value of 1 if the patient experienced the outcome (e.g., revision/conversion) and given the value of 0 otherwise. Both models also included a constant term and control for the Census region of the surgery facility, an indicator for unspecified cementation, and residuals from the first-stage (choice of surgery type) instrumental variables estimation. The estimation procedure for each outcome accounted for clustering (i.e., multiple observations within the same facility). Both models were estimated using a probabilistic probit specification to account for the nonlinear, binary nature of the dependent variables and the instrumental-variable assumption of the normality of the residuals in each estimation stage. The regression coefficients represent the adjusted marginal effect associated with a unit change in the indicator covariate, controlling for all other factors. Positive or negative coefficients indicate a higher or lower likelihood, respectively, of experiencing the outcome (e.g., revision/conversion within 12 months after the index arthroplasty surgery). Given the nonlinearity of the dependent variables, the probit regression coefficients do not represent the magnitude of the effect. To provide a sense of the magnitude of the effect of our key variable of interest, surgery type, we utilized these regression coefficients to calculate the predicted (adjusted) probability associated with each surgery type (THA versus HA) for each of the 2 outcomes, while holding all other factors constant at their actual values. These predicted probabilities are presented in the text.