In a prior post, I discussed an article titled "Medicaid status is independently predictive of increased complications, readmission, and mortality following primary total shoulder arthroplasty." The authors claimed to show that Medicaid status was an independent predictor of adverse outcomes using a national administrative claims database. To interrogate that conclusion, I submitted the article and my blog comments to ChatGPT and asked for a refined critique. Here’s what we arrived at:
Is Medicaid Really the Risk? Are we asking the right question?
Recent work using national databases has reported that patients insured by Medicaid are at increased risk for complications, readmission, and mortality following elective total shoulder arthroplasty (TSA). These findings have led to the conclusion that "Medicaid status is independently predictive" of adverse outcomes. But is Medicaid status truly the cause of poorer outcomes, or is it a proxy for other, more fundamental patient characteristics and circumstances?
Let’s start with a thought experiment: Imagine a 63-year-old widow living alone with diabetes, osteoporosis, and shoulder arthritis. She qualifies for Medicaid due to low income and limited assets. Her shoulder arthroplasty is scheduled in two months. Now imagine she marries a childhood friend who provides her with different health insurance. Her comorbidities, living situation, and health literacy remain unchanged. Should we expect her surgical risk to suddenly drop?
Of course not. The administrative label of her insurer has changed, but her biological and social risk factors have not. Yet the implication of the aforementioned study is that the change in payer alone would significantly reduce her odds of a complication. This framing fails what I would call the "sniff test."
Propensity Matching Pitfalls
The study used propensity score matching (PSM) to compare outcomes between patients on Medicaid and those not on Medicaid. However, matching included only age, sex, and discharge weight. Critically, key confounders such as comorbidity burden (Charlson Comorbidity Index) were excluded from the matching algorithm. The groups remained imbalanced across multiple preoperative risk factors: congestive heart failure, chronic lung disease, anemia, renal failure, substance use, and more. These are not minor oversights—they are core determinants of perioperative risk.
More broadly, Medicaid enrollment is intertwined with social determinants of health: poverty, underemployment, poor nutrition, limited access to preventive care, and systemic racism. These powerful drivers of outcome often go unmeasured in administrative data. To ignore them is to risk mistaking correlation for causation.
This misattribution reflects a well-known cognitive bias: the fundamental attribution error. As Daniel Kahneman explains in Thinking, Fast and Slow, we often over-attribute outcomes to individual traits while neglecting situational forces. Medicaid status is being treated as a standalone risk factor, rather than a marker of cumulative disadvantage.
The authors did apply multivariate regression after matching, but residual differences in baseline health status suggest significant causes of confounding were unaddressed. The assertion that "Medicaid status was independently predictive" risks directing attention toward an insurance label—a variable that is neither biologically meaningful nor directly actionable for most patients—instead of toward modifiable drivers of inequity.
Instead, we should ask: Do patients insured by Medicaid fare worse after adjusting for what we can measure? If so, it tells us that we have incompletely identified the barriers these patients face in accessing and benefitting from coordinated, high-quality care.
Clinically speaking, we do not treat Medicaid status. We treat people. Let us not confuse an administrative variable with the medical and social factors that determine the outcome of treatment.
Revised Abstract (from ChatGPT)
Background: Prior studies report worse surgical outcomes in Medicaid-insured patients, but the extent to which these reflect insurance status versus underlying comorbidity and social disadvantage remains unclear. This study assessed the association between Medicaid insurance and short-term adverse outcomes after elective total shoulder arthroplasty (TSA) using a national administrative database.
Methods: We conducted a retrospective cohort study using the Nationwide Readmissions Database (NRD) from 2016 to 2020. Patients undergoing elective primary TSA were identified by ICD-10-PCS codes. Medicaid patients were matched 1:1 to non-Medicaid patients on age, sex, and discharge weight using nearest-neighbor propensity score matching. Covariate balance was assessed descriptively but not using standardized mean differences. Logistic regression adjusted for residual covariates including the Charlson-Deyo Comorbidity Index (CCI), select diagnoses, and geographic factors. Outcomes included 180-day complications, readmissions, mortality, and other surgical and medical complications.
Results: 15,448 patients on Medicaid and 15,374 patients not on Medicaid were analyzed. Patients on Medicaid had higher CCI scores, greater comorbidity burden, and lower income. Adjusted analyses showed increased odds of any complication (OR 1.2), readmission (OR 1.2), and mortality (OR 1.4) in the Medicaid group. Significant associations also emerged for dislocation, pneumonia, sepsis, and other events.
Conclusion: Medicaid insurance was associated with increased adverse outcomes after TSA, even after adjustment for the variables that were identified. These differences must be due to social and clinical risks that were not identified in this study. Medicaid should be viewed not as an intrinsic risk factor, but as a proxy for structural disadvantage. Policy and care improvements must focus on optimizing comorbidity management, care access, and coordination for all patients, independent of insurance status, i.e. patients on other insurance programs are subject to the same risk factors as those covered by Medicaid. We should focus on identifying and managing these social, economic and medical risk factors rather than the type of insurance.
I'd like to illuminate this last point with a chart from the paper
Simple Guidelines for Propensity Score Matching (PSM)
Define the Exposure Clearly
Must be binary (e.g., Medicaid vs. non-Medicaid insurance).
Clearly identify the treated and comparison groups.
Choose Confounders Thoughtfully
Include only variables measured before treatment.
Choose variables that affect both treatment and outcomes.
Avoid variables on the causal pathway (e.g., complications).
Estimate the Propensity Score
Use logistic regression.
Include relevant confounders.
Match Patients
Use 1:1 nearest neighbor with a caliper (e.g., 0.2 × SD of logit).
Prefer matching without replacement.
Exclude unmatched cases.
Check Balance
Use standardized mean differences (SMD < 0.1 ideal).
Visualize using Love plots.
Analyze Outcomes Appropriately
Use matched methods (e.g., conditional logistic regression).
Don’t assume matched pairs are independent.
Report Transparently
Describe covariates used.
State matching method and caliper.
Report group sizes pre/post-matching.
Show covariate balance.
Detail outcome analysis.
A Fun Illustration: Osprey vs. Bald Eagle
If we were to use PSM to compare the hunting success of an osprey and a bald eagle:
Outcome: Hunting success rate = successful captures / attempts
Exposure: Species (osprey vs. eagle)
Confounders to Match:
Habitat (river, lake, ocean)
Wind/weather
Time of day/season
Prey density
Flight altitude
Visibility conditions
Not Confounders:
Bird color
Vocalization
Rarity or popularity
Endangered status
As with surgery, outcomes depend not just on species — or insurance — but on context, environment, and opportunity.
Let us not confuse administrative categories for biological truths. Statistical adjustment is no substitute for structural insight.
You can support cutting edge shoulder research that is leading to better care for patients with shoulder problems, click on this link
Follow on twitter/X: https://x.com/RickMatsen
Follow on facebook: https://www.facebook.com/shoulder.arthritis
Follow on LinkedIn: https://www.linkedin.com/in/rick-matsen-88b1a8133/
Here are some videos that are of shoulder interest
Shoulder arthritis - what you need to know (see this link).
How to x-ray the shoulder (see this link).
The ream and run procedure (see this link).
The total shoulder arthroplasty (see this link).
The cuff tear arthropathy arthroplasty (see this link).
The reverse total shoulder arthroplasty (see this link).
The smooth and move procedure for irreparable rotator cuff tears (see this link)
Shoulder rehabilitation exercises (see this link).
