LEP status independently predicted increased 30-day VTE (aOR 2.10) after knee arthroplasty

Description #

On multivariable logistic regression adjusted for age, gender, BMI, ASA physical status, smoking status, and length of hospitalisation, limited English proficiency (LEP) remained an independent predictor of increased 30-day postoperative venous thromboembolism (VTE) after elective total knee arthroplasty: adjusted odds ratio 2.10 (95% CI 1.06–4.12, p = 0.03) relative to English-proficient patients (Figure 2).

"Multivariable logistic regressions adjusted for age, gender, BMI, ASA physical status, smoking status, and length of hospitalisation found that patients with LEP had higher rates of 30‐day postoperative DVT (aOR: 2.84, CI:1.19–6.79, p = 0.02) and 30‐day VTE (aOR: 2.10 CI:1.06–4.12, p = 0.03) when compared to their EP counterparts (Figure 2)." (Nguyen, 2023, p. 577)

Methods Context #

What? #ⓘ

The observable: the covariate-adjusted odds of a 30-day postoperative VTE for LEP relative to EP patients, modeled as a binary outcome.

"Multivariable logistic regression modelling was used to calculate adjusted odds ratios of having a postoperative complication." (Nguyen, 2023, p. 577)

How? #ⓘ

Multivariable logistic regression with an a-priori covariate set; race/ethnicity excluded from the model owing to collinearity with English proficiency.

"Calculated odds ratios were adjusted for covariates chosen a prior based on prior arthroplasty literature, including age, patient‐reported gender, BMI, ASA physical status, smoking status, and length of hospitalisation (Gronbeck et al., 2019; Ko et al., 2021). Race/ethnicity were excluded from the regression model due to high collinearity between race/ethnicity categories with English proficiency status." (Nguyen, 2023, p. 577)

Who? #ⓘ

3450 elective TKA cases at a single academic medical centre (2013–2021), 9.2% (316) with LEP.

"A total of 3450 TKA cases were performed between January 2013 and December 2021. 9.2% (316) of the patients were classified as having LEP." (Nguyen, 2023, p. 577)

Other Notes #

The lower CI bound (1.06) is close to 1, indicating a statistically significant but imprecise adjusted estimate; see the qualifying caveat on rare-event imprecision.

Caveats #

• Rare complication events yielded imprecise adjusted odds ratios with wide confidence intervals Postoperative complications were rare events (crude rates mostly 0.1%–3.8%) in a cohort with only 316 LEP patients, so the adjusted odds-ratio estimates are imprecise. This is visible in the wide confidence intervals — 30-day DVT aOR 2.84 (1.19–6.79) and, most strikingly, 7-day pneumonia aOR 2.77 (0.57–13.51), whose interval spans more than 20-fold and crosses 1. The significant thromboembolic estimates should be treated as directionally suggestive but imprecise, and the point estimates as fragile.
• All LEP patients grouped despite heterogeneity in language, provider capacity, and English proficiency level All patients with limited English proficiency were analysed as a single group, defined only by self-reporting a primary language other than English. This collapses substantial heterogeneity: the degree of patients' actual English proficiency varies, and nursing and provider language-concordance capacity differs across languages (e.g., Spanish vs Chinese vs Russian). Pooling these subgroups may dilute or distort the association between language discordance and complications, and prevents the study from identifying which patients face the greatest risk.
• Single-center retrospective EHR cohort cannot identify mediators or confounders of LEP-associated complications The study is a retrospective cohort drawn from a single academic medical centre, and the authors note two consequences. First, although the electronic health record is a reliable source for complication counts, the design cannot identify the specific mediators or clinical confounders that drive the LEP–complication association, so the results are correlational rather than mechanistic. Second, single-site data limit external validity; the authors call for multi-site replication, while noting that larger national databases lack primary-language data. These features mean the effect estimates may not generalise and cannot establish causal pathways.
• Race and ethnicity excluded from adjustment due to collinearity with LEP, leaving residual confounding Race and ethnicity were dropped from the multivariable models because they were highly collinear with English-proficiency status (LEP patients were predominantly AAPI and Hispanic/Latinx). As a result, the adjusted odds ratios cannot separate the effect of language proficiency from that of race/ethnicity, and residual confounding by these correlated social factors remains. The "independent" adjusted association between LEP and thromboembolic complications should therefore be read as language proficiency inseparable from its correlated racial/ethnic context, not as a language effect net of race.