Abstract
This cohort study assesses the association between stigmatizing language, demographic characteristics, and errors in the diagnostic process among hospitalized adults.
Stigmatizing language (SL) is widespread throughout medical documentation.1 It is more likely to be found in the records of Black patients,2,3 patients with public insurance,2 and patients with certain comorbidities.3 We investigated associations between SL, errors in the diagnostic process, and demographics for hospitalized patients.
Methods
This multicenter, retrospective cohort study was conducted as part of the Utility of Predictive Systems for Diagnostic Errors (UPSIDE) study.4 Using a structured adjudication tool, UPSIDE assessed the presence of diagnostic errors and diagnostic process errors among patients who died while hospitalized (with the exception of people who died <48 hours of admission for trauma, burn, or out-of-hospital arrest) or were transferred to the intensive care unit more than 48 hours after admission at 29 hospitals from January to December 2019. Diagnostic errors were defined as missed opportunities to make a correct or timely diagnosis. Diagnostic process errors were diagnostic failure points occurring across the Diagnostic Error Evaluation and Research categories.5 Race, ethnicity, and other demographic variables were collected administratively at study sites. Full UPSIDE methods are described elsewhere.6 This study was approved by the University of California San Francisco institutional review board; informed consent was waived because of use of retrospective data.
As a secondary aim, reviewers identified the presence of SL throughout physician, nursing, and ancillary staff notes. Stigmatizing language was defined as containing one of the following features: questioning of patient credibility, racial or social class stereotyping, expressions of disapproval toward patients, and descriptions of difficult patients.1
Univariate analysis between SL, diagnostic errors, and demographics was performed using χ2 testing with Rao-Scott second-order correction, taking into account the sampling design of the UPSIDE study.6 We used generalized estimating equations to fit logistic regression models with clustering by hospital, sampling weights, exchangeable working correlations, and robust SEs to calculate unadjusted and adjusted odds ratios (ORs) quantifying the associations between SL and the presence of diagnostic process errors. Analyses were performed in R, version 4.3.2 and Stata, version 17.0. One-sided P < .05 indicated significance.
Results
After excluding 81 hospital admissions with missing data, 2347 were included in our subanalysis. Diagnostic errors were identified in 536 (23.2%) and SL found in 131 (5.1%). Presence of SL varied among sites (median, 4.0%; IQR, 1.1%-8.1%).
The prevalence of SL in documentation was higher among patients with diagnostic errors (8.2%) than among those without diagnostic errors (4.1%) (P = .01). Stigmatizing language was more common among Black patients (9.6%) than among Asian (3.9%) or White (3.8%) (P = .002) patients and among patients with housing instability (15.1% vs 4.9%; P = .02) (Table 1). In adjusted multivariate analysis of diagnostic process errors, SL was associated with delays in care at presentation (OR, 1.9; 95% CI, 1.3-2.9) and communication with patients and caregivers (OR, 3.8; 95% CI, 1.2-12.0) (Table 2).
Table 1. Prevalence of Stigmatizing Language by Presence of Diagnostic Error and Demographic Variables.
| Characteristic | Stigmatizing language, unweighted No. (weighted %) | P valuea | |
|---|---|---|---|
| Absent (n = 2216) | Present (n = 131) | ||
| Diagnostic error | |||
| Absent | 1723 (95.9) | 88 (4.1) | .01 |
| Present | 493 (91.8) | 43 (8.2) | |
| Sex | |||
| Female | 1016 (95.8) | 58 (4.2) | .24 |
| Male | 1200 (94.3) | 73 (5.7) | |
| Raceb | |||
| Asian | 109 (96.1) | 7 (3.9) | .002 |
| Black | 379 (90.4) | 40 (9.6) | |
| White | 1528 (96.2) | 68 (3.8) | |
| Other, NAc | 200 (92.5) | 16 (7.5) | |
| Ethnicityb | |||
| Hispanic | 123 (95.8) | 5 (4.2) | .74 |
| Non-Hispanic | 1965 (95.0) | 113 (5.0) | |
| Unknown | 128 (92.9) | 13 (7.1) | |
| Housing status | |||
| Stable | 2175 (95.1) | 121 (4.9) | .02 |
| Instability | 41 (84.9) | 10 (15.1) | |
| Language | |||
| English | 1933 (95.1) | 115 (4.9) | .07 |
| Not English | 191 (91.5) | 14 (8.5) | |
| Do not know | 92 (99.2) | 2 (0.8) | |
| Substance use disorder | |||
| Absent | 1920 (95.3) | 89 (4.7) | .13 |
| Present | 296 (93.0) | 42 (7.0) | |
Abbreviation: NA, not available.
χ2 Test with Rao-Scott second-order correction.
Race and ethnicity were collected from administrative data at study sites.
Included race reported as unknown, other, unavailable, or declined in the administrative data source.
Table 2. Unadjusted and Adjusted Multivariate Analysis on the Association Between Stigmatizing Language and Diagnostic Process Errors.
| Diagnostic process errora | Unadjusted OR (95% CI) | Adjusted OR (95% CI)b |
|---|---|---|
| Access and presentation | 2.0 (1.5-2.8) | 1.9 (1.3-2.9) |
| History taking | 2.4 (0.9-6.1) | 2.0 (0.6-6.7) |
| Physical examination | 1.7 (0.9-3.1) | 0.8 (0.4-1.9) |
| Testing | 1.6 (0.9-2.8) | 1.0 (0.6-1.7) |
| Patient follow-up and monitoring | 2.4 (1.4-4.1) | 2.0 (0.98-4.1) |
| Obtaining referrals | 1.1 (0.6-2.0) | 0.6 (0.4,1.03) |
| Teamwork | 2.1 (1.2-3.8) | 1.1 (0.6-2.1) |
| Communication with patient or caregiver | 3.7 (1.4-10.2) | 3.8 (1.2-12.0) |
| Assessment | 2.6 (1.3-5.1) | 1.7 (0.4-7.3) |
Abbreviation: OR, odds ratio.
One or more of the process errors in this dimension contributed to the potential for diagnostic error.
Adjusted for race, ethnicity, housing status, English as primary language, substance use disorder diagnosis, age, primary payer, smoking status, mechanical ventilation, intensive care unit admission, and altered mental status at presentation.
Discussion
We found that SL in patient documentation was associated with diagnostic error and multiple diagnostic process errors. The prevalence of SL was higher among Black patients and patients with housing instability.
Our study is limited by documentation and detection bias. Despite reviewer training, we observed variation in frequency of SL across sites, possibly due to the varied patient demographics and medical complexity at the hospitals included in the cohort; this limits extrapolations of our findings more broadly to all US hospitals. Also, we did not ask reviewers to identify the type of SL or at what point in the hospital admission the documentation occurred. Lastly, the low proportion of records with SL precluded multivariate assessment of the association between SL and overall diagnostic error.
One potential mechanism to explain the association between SL and clinician diagnostic processes is that SL may be indicative of clinician biases that interfere with data gathering, communication, and clinical reasoning. Further work is needed to explore mechanisms to explain our findings and understand how clinician use of SL can be reduced.
Data Sharing Statement
References
- 1.Park J, Saha S, Chee B, Taylor J, Beach MC. Physician use of stigmatizing language in patient medical records. JAMA Netw Open. 2021;4(7):e2117052-e2117052. doi: 10.1001/jamanetworkopen.2021.17052 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Sun M, Oliwa T, Peek ME, Tung EL. Negative patient descriptors: documenting racial bias in the electronic health record. Health Aff (Millwood). 2022;41(2):203-211. doi: 10.1377/hlthaff.2021.01423 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Himmelstein G, Bates D, Zhou L. Examination of Stigmatizing Language in the Electronic Health Record. JAMA Netw Open. 2022;5(1):e2144967. doi: 10.1001/jamanetworkopen.2021.44967 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Dalal AK, Schnipper JL, Raffel K, Ranji S, Lee T, Auerbach A. Identifying and classifying diagnostic errors in acute care across hospitals: early lessons from the Utility of Predictive Systems in Diagnostic Errors (UPSIDE) study. J Hosp Med. 2024;19(2):140-145. doi: 10.1002/jhm.13136 [DOI] [PubMed] [Google Scholar]
- 5.Schiff GD, Hasan O, Kim S, et al. Diagnostic error in medicine: analysis of 583 physician-reported errors. Arch Intern Med. 2009;169(20):1881-1887. doi: 10.1001/archinternmed.2009.333 [DOI] [PubMed] [Google Scholar]
- 6.Auerbach AD, Lee TM, Hubbard CC, et al. ; UPSIDE Research Group . Diagnostic errors in hospitalized adults who died or were transferred to intensive care. JAMA Intern Med. 2024;184(2):164-173. doi: 10.1001/jamainternmed.2023.7347 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Sharing Statement
