Socioeconomic status and patient outcomes. A potentially flawed study is gaining traction in Orthopedics!

Hi all,

I am a current orthopedic surgery resident who also serves as a reviewer for one of the major joint replacement journals in orthopedics ( Journal of Arthroplasty; i.e JOA). I have some experience as a data scientist. JOA recently published a retrospective manuscript which concluded that genetic predisposition and NOT socioeconomic status was associated with likelihood for the development of joint infection after a knee replacement. Interestingly, the authors found patients on medicaid had higher odds of developing joint infection ( OR: 1.39; p=0.013). This to me seems like an issue of collinearity. Nonetheless, it is creating alot of buzz in the joint replacement literature, and may spur some changes to standard of care.

From reading the article, it seems like the authors fail to account for collinearity, particulary between Medicaid status and median house hold income. I am prepared to write a letter to the Editor to rebut the authors conclusions.

I will like to get everyone’s opinions on it. The title of the article is " Socioeconomic Status May Not Be a Risk Factor for Periprosthetic Joint Infection" by DeKeyser et al.

I can send the manuscript to those who cannot get access to it, if you email me at

Warm regards


The contribution I can make to this discussion is limited because I don’t have subject matter knowledge on this topic but here are some thoughts:

  • Collinearity leads to larger standard errors, confidence intervals and p-values but does not bias the estimates themselves. As the study features a rather large sample size, I don’t think collinearity is a big problem in this case. Also, collinearity is usually discussed in the framework of continuous variables but the authors included Medicaid and Household Median Income as categorical variables.
  • The authors provide little explanation of how they chose their adjustment variables. The principled way in my opinion would be to draw a DAG (directed acyclic graph) of the hypothesized causal relationships and decide on the adjustment variables based on that.
  • The authors make the (extremely prevalent) mistake of interpreting p-values >0.05 as “no difference” or “no effects”. From the abstract:

There was no difference in risk between the groups associated with education level or median household income (all, P > .05).

Yikes! As an example to counter that claim (see Table 2): The upper confidence limit for a household median income (10%-<25%) reaches up to 1.33 so the data are compatible at the 95\% level with an 33\% increase in risk (also with a recuction of 11\%)!

  • The authors also categorize continuous variables needlessly, the BMI for example. It would be better to model those using splines.
  • The authors include “missing” as a separate category in their models which is not ideal, as far as I know. There are better ways to deal with missing data, such as multiple imputation.

In view of these points, I would interpret the results of this study with caution.

1 Like

I don’t have access to the article, but the relationship between median household income (presumably by zipcode) and the Medicaid status of a specific patient is complicated.

In most states, being low income is a part of the requirement for gaining Medicaid coverage, so in most categories, the Medicaid patients will be low-income. But the relationship in the other direction is different, most low-income patients are not going to be on Medicaid, unless they’re mothers or children (and knee replacements for those age groups are probably fairly uncommon.)

However, Medicaid as a PAYER typically comes with reimbursement policies that have the practical effect of limiting access to certain types of care, relative to the access enjoyed by patients with other coverage: very low reimbursement for PCPs, essentially no access to dental care due to dismal rates, and probably (depends on the state) low payment rates and caps on the number of reimburseable services for PT, home visits, etc.

So Medicaid as a proxy indicator of poor access to POST-surgical care seems very plausible to me.

(Edit: I mention dental because it’s my understanding that the presence of dental caries is one of the leading reasons for ruling out patients as transplant candidates, as that materially increases the risk of infection. I don’t know if it’s also a known risk factor for knee replacement surgeries.)