Review Question - QID 218767

In this study, the investigators are evaluating the association between preoperative functional status and the change in BMI after total joint arthroplasty while adjusting for various confounding variables. Because the outcome (change in BMI) is a continuous variable, the most appropriate analysis to use for reporting the association between BMI and preoperative functional status that accounts for confounding variables is a linear regression analysis.

Regression analysis is used in observational studies that aim to evaluate the relationship between 1 or more predictor/explanatory variables and 1 outcome variable. If the outcome variable is continuous, then linear regression is appropriate. A linear regression output will give coefficients for each predictor variable. These coefficients can be interpreted as the change in the outcome variable with each 1-unit increase in the predictor variable (if continuous) or with the simple presence of the predictor variable (if categorical). For example, if a linear regression analysis in this study revealed a coefficient of -0.2 for the preoperative PROMIS-PF score, this means that for every 1-unit increase in the PROMIS-PF score, there is a 0.2 kg/m^2 decrease in the BMI. Potential confounders are entered as predictor variables in the linear regression equation and their coefficients can be interpreted in a similar manner.

Incorrect Answers:
Answer 1: Analysis of variance (ANOVA) is a statistical test that is used to compare whether mean values differ between 3 or more cohorts. For example, if preoperative physical function had been divided into 3 levels (low, medium, high) and the researchers wanted to compare the mean change in BMI between these 3 levels of physical function, then an ANOVA test would be appropriate. Notably, the ANOVA test can only reveal whether the means differ from each other - pairwise comparisons must be made on a post-hoc basis. Further, ANOVA does not account for potential confounders.

Answers 3 and 4: Logistic regression analysis and odds ratios are used to evaluate associations in observational studies when the outcome variable is binary. For example, if the outcome variable had been dichotomized into whether there was a decrease in BMI or not, then a logistic regression analysis would be appropriate. An odds ratio can be calculated from logistic regression coefficients by exponentiating them. For example, if a logistic regression is performed using whether there has been a decrease in BMI as the outcome variable and preoperative PROMIS-PF score as a predictor variable, a coefficient of +0.1 for the PROMIS-PF score would mean that every 1-unit increase in the preoperative PROMIS-PF score is associated with an exp(0.1) = 1.1-fold increased adjusted odds of achieving a reduction in BMI with total joint arthroplasty.

Answer 5: Relative risk is used in observational studies to evaluate associations when the exposure (which is the predictor variable) and outcome variables are binary. It is calculated as the probability of the outcome occurring with the exposure divided by the probability of the outcome occurring without the exposure. For example, if the outcome variable had been dichotomized into whether there was a decrease in BMI or not and the predictor variable had been dichotomized into high versus low preoperative physical function, then a relative risk could be calculated.

Bullet Summary:
Regression analysis is used in observational studies to evaluate the association between an outcome variable and a predictor or explanatory variable, and enables researchers to adjust for the effects of confounding variables.