Primary language other than English independently predicted lower odds of signing consent (OR 0.74) for non-industry studies

Description #

On multivariable logistic regression adjusting for age, gender, race, ethnicity, histology and study type, having a primary language other than English independently predicted lower odds of signing consent for a non-industry-sponsored study (relative to an industry-sponsored study) compared with English-primary patients: OR 0.74 (95% CI 0.63–0.94, P = 0.005) (Table 1). The association survived adjustment for potential confounders.

"patients with a primary language other than English (OR 0.74, 95% CI 0.63 to 0.94, P = 0.005) and limited English proficiency (OR 0.74, 95% CI 0.58 to 0.95, P = 0.021) had lower odds of signing consent documents for non-industry sponsored studies than patients with English as their primary language." (Maria, 2023, p. 858)

Methods Context #

What? #ⓘ

The observable: the adjusted odds that a consent event for a patient with a primary language other than English occurred in a non-industry- (vs industry-) sponsored study.

"Multivariable analysis for patients with a primary language other than English signing consent documents" (Maria, 2023, Table 1)

How? #ⓘ

Multivariable GEE logistic regression clustered by patient, adjusting for prospectively identified covariates.

"After adjusting for age at consent, gender, race, ethnicity, histology and study type (observational versus interventional), patients with a primary language other than English (OR 0.74, 95% CI 0.63 to 0.94, P = 0.005) ... had lower odds of signing consent documents for non-industry sponsored studies." (Maria, 2023, p. 858)

Who? #ⓘ

All eligible consent events at one cancer centre, 2013–2018; English-primary patients are the reference category.

"English primary — Reference" (Maria, 2023, Table 1)

Other Notes #

A sensitivity model nesting patients within study (Extended Data Table 8) gave a consistent estimate (OR 0.79, 95% CI 0.65–0.96, P = 0.019); adding Medi-Cal status (Extended Data Table 7) gave OR 0.78 (0.63–0.98, P = 0.033).

Caveats #

• Retrospective single-centre EHR-based cohort that cannot establish causation (Maria 2023) The associations come from a retrospective, single-centre analysis built on electronic health record and clinical-trials-database data at one academic cancer centre. A retrospective design cannot prove causation; the single-centre setting limits generalizability (sensitivities around patient health information, study-related data, and regulatory differences make cross-centre replication difficult); and key variables were captured retrospectively and may be inaccurate — Medi-Cal insurance status is dynamic and may not reflect status at the consent event, and language information may not be documented accurately in the EHR. The authors argue that consistent associations across analyses support the hypothesis, but the estimates remain observational.