Adjusted odds of hospital admission or transfer were only 6% higher for interpreter-requested pediatric patients, a reversed trend from prior gaps

Description #

The adjusted odds of hospital admission or transfer to another facility were 6.0% higher for interpreter-requested (LEP) patients than for no-interpreter patients (adjusted OR 1.06, 95% CI: 1.01–1.11; p=0.014), corresponding to a small effect (ES < 0.2) the authors judged not clinically meaningful (Table 2). This is a reversed trend relative to a 2004 study at the same hospital (CHOA Egleston), where LEP patients had higher admission rates.

"Adjusted odds of hospital admission or transfer to another facility in interpreter requested patients were 6.0% higher than the adjusted odds in no interpreter requested patients (aOR: 1.06, 95% CI: 1.01, 1.11). This difference also corresponded to a small effect (ES < 0.2)." (Greenky, 2019, p. 4)

"In comparison to a 2004 study completed at CHOA's Egleston Hospital, our study showed that there was no clinically significant difference in admission/transfer rate between the two groups." (Greenky, 2019, p. 5)

Methods Context #

What? #ⓘ

The observable: hospital disposition, principally analyzed as a two-level outcome of admission/transfer versus discharge, recorded per encounter from the EPIC EMR.

"Hospital disposition was described as a four-level characteristic (ICU, floor, transfer, and discharge) but principally analyzed as two-level variable (admission or transfer versus discharge)." (Greenky, 2019, p. 2)

How? #ⓘ

Retrospective cohort study; hospital admission/transfer modeled with logistic regression adjusted for age at baseline, insurance status, means of arrival, and maximum acuity, reported as adjusted odds ratios with 95% CIs; effect sizes (Cohen's d) prioritized over p-values.

"crude and adjusted associations between interpreter categories and the study outcomes were modeled using... logistic regression for change in acuity, ED readmission within 7 days, and hospital disposition." (Greenky, 2019, p. 2)

Who? #ⓘ

All patients aged 0–18 presenting to three CHOA pediatric EDs in 2016 (152,945 patients / 232,787 encounters); LEP defined by an interpreter request during the encounter; dead-on-arrival and unlisted-language patients excluded.

"This was a retrospective cohort study that looked at all patients aged 0-18 years that arrived in the three CHOA EDs (Hughes Spalding Hospital, Egleston Hospital, and Scottish Rite Hospital) between January 1, 2016, and December 31, 2016." (Greenky, 2019, p. 2)

Other Notes #

Unlike the LOS, readmission, and acuity outcomes, this admission/transfer comparison was statistically significant (p=0.014), but the effect size remained small; the authors emphasize the contrast with the historically larger admission-rate gap reported in 2004 at the same institution.

Caveats #

• Single highly-resourced academic pediatric system limits generalizability of the reversed-trend findings The study was conducted entirely within a single, highly-resourced academic pediatric system (Children's Healthcare of Atlanta) with a mature, well-staffed interpreter program (37 interpreters, dedicated in-person Spanish service, video/phone backup). The authors caution that centers with fewer language-access resources may not show the same small/null differences between LEP and English-speaking patients, so the "reversed trend" of near-equivalent ED outcomes may not generalize.
• Interpreter-request EMR field is an imperfect proxy for LEP with substantial undocumented ascertainment LEP status and interpreter use were both operationalized from a single EMR "interpreter requested" field, which the authors acknowledge is an imperfect proxy. Only 83.3% of patients with a non-English primary language were documented as having requested an interpreter, leaving 16.7% whose interpreter use is unknown; there is no EMR field confirming an interpreter was actually offered or used, it is unclear who requested the interpreter (patient, family, or provider), and unclear who was LEP (patient, caregiver, or both). Patients who declined an interpreter but would have benefited are also misclassified. This measurement uncertainty could bias the small/null group differences in the ED outcome comparisons.