Abstract
Objectives:
The language used to describe people with substance use disorder (SUD) impacts stigma and influences clinical decision making. This study evaluates the presence of stigmatizing language (SL) in clinical notes and detects patient- and provider-level differences.
Methods:
All free-text notes generated in a large health system for patients with substance-related diagnoses between December 2020 and November 2021 were included. A natural language processing algorithm using the National Institute on Drug Abuse’s “Words Matter” list was developed to identify use of SL in context.
Results:
There were 546,309 notes for 30,391 patients, of which 100,792 (18.4%) contained SL. 18,727 (61.6%) patients had at least one note with SL. The most common SL used were “abuse” and “substance abuse.” Nurses were least likely to use SL (4.1%) while physician assistants were most likely (46.9%). Male patients were more likely than female patients to have SL in their notes (adjusted Odds Ratio (aOR) 1.17 (95% confidence internal (CI) 1.11–1.23), younger patients aged 18–24 were less likely to have SL than patients 45–54 years (aOR 0.55, 95% CI 0.50–0.61), Asian patients were less likely to have SL than White patients (aOR 0.45, 95% 0.36–0.56), and Hispanic patients were less likely to have SL than non-Hispanic patients (aOR 0.88, 95% 0.80–0.98).
Conclusions:
The majority of patients with substance-related diagnoses had at least one note containing SL. There were also several patient characteristic disparities associated with patients having SL in their notes. The work suggests that more clinician interventions about use of SL is needed.
Keywords: stigma, substance use disorder, natural language processing
Introduction
The drug-related overdose death epidemic in the United States continues, with most recent data reporting over 100,000 lives lost per year, an over five-fold increase since 2000 (1). Several interventions are being actively pursued to address this crisis, including relaxation of regulatory barriers to increase the number of buprenorphine prescribers (2), implementation of overdose prevention sites in select states (3, 4), and increased access to naloxone (5). The United States Department of Health and Human Services (HHS) recently released its four pillars to prevent overdoses, including primary prevention, evidence-based treatment, recovery support, and harm reduction (6).
A key intervention under HHS’s harm reduction pilar is to “develop educational materials and programs to reduce stigma,” which is the focus of this research. Stigma is “a mark of shame or discredit” (7), and it can be particularly harmful to individuals experiencing substance use disorder (SUD) (8–11). Stigma can also reduce the desire of policymakers to allocate resources to SUD, the willingness of the public to support policy change, and impede initiatives to screen patients for SUD, and thus identify and effectively treat those who are suffering (12).
Stigma can be experienced and communicated in various forms, including in the language that medical personnel use when referring to individuals with SUD (13) and even in the way individuals with SUDs refer to themselves (self-stigma) (14). Just small differences, such as referring to an individual as “a substance abuser” versus “having a substance use disorder”, can lead to differential judgments about culpability and clinicians’ responses to the individual (15–18). Healthcare professionals should utilize person-first, non-stigmatizing language throughout all aspects of health (19). Specifically for SUD, agencies such as the National Institute on Drug Abuse (NIDA) have created guidance emphasizing the point that “Words Matter” and that there are terms to use and avoid when discussing substance use and addiction (20). These recommendations include using person-centered language, such as “a person with a substance use disorder” instead of “addict”, “user”, or “abuser”; employing non-stigmatizing terminology about recovery and treatment, such as “being in recovery” instead of “clean”; and “medication for opioid use disorder” instead of “opioid replacement therapy” or “medication assisted treatment.”
Stigmatizing language (SL) in general is pervasive in medical records (21), which impacts patients as well as healthcare team members. It applies not only to individuals with SUD but also to other conditions including chronic pain (22) and diabetes (23). In a large sample of individuals (not just those with SUD) who read at least one of their outpatient medical notes, over 10% reported feeling judged and/or offended (24). Recent work from a member of our research team used natural language processing (NLP), a technique in which a computer is programmed to process, analyze, and interpret language including contextual nuances, to detect the presence of stigmatizing language (SL) in over 48,000 notes (25). The study was focused on terms considered to be generally stigmatizing (i.e., not specific to SUD) but found that 2.5% of admission notes contained SL and that there was a higher incidence for non-Hispanic Black patients. In this study, we aimed to determine the incidence of SL specific to patients with substance-related diagnoses in our clinical notes and, as a secondary outcome, determine if there were any disparities associated with demographic characteristics of patients seen in our health system.
Methods
This was a cross-sectional, observational, retrospective study of hospital notes contained in the electronic health record (EHR) of Mass General Brigham (MGB), a large healthcare system in Massachusetts and New Hampshire. The work was initially conducted as part of an internal quality improvement process entitled United Against Racism (26). The work produced a deidentified dataset that was used for reporting results and was determined to be not human subject research by the MGB Human Research Committee.
Cohort Identification
Our cohort included patients who had inpatient and/or outpatient visit(s) in our system during the one-year period between December 1, 2020 and November 30, 2021 and whose encounter diagnosis contained at least one substance-related diagnosis code. The following International Classification of Diseases, Tenth Revision (ICD-10) codes, which capture substance use disorders and overdose of both drugs and alcohol, were included: F10, F11, F12, F13, F14, F15, F16, F18, F19, T40, T40.5, T40.7, T40.8, T40.9, T40.99, T42.3, T42.4, T43.60, T43.62, T43.63, T43.64, T43.69, T51.0, and X45. We then extracted free-text clinical notes of these patients by including common note types used by a wide range of providers, including physicians, advanced-practice providers, nurses, social workers and others who interact with patients and document the encounter: history and physicals, consults, discharge summaries, emergency department (ED) notes, group therapy notes, transfer/sign-off notes, progress notes, ED nurse triage notes, nursing notes, assessment/plan notes, nursing summaries, plan of care and opioid notes (a special note type created by our facility which lists the indication for opioid prescriptions, the primary prescriber, and SUD-related risk factors). Additionally, for each note, we included patient-reported demographic data, including age, gender, race and ethnicity.
Natural Language Processing Method
For SL, we included all keywords on NIDA’s guidance document (Appendix 1) (20) except for “habit,” for which an initial screen of phrases containing that word almost never indicated its use in a stigmatizing fashion (e.g., “the patient has a habit of walking each morning”). We randomly selected 10,000 clinical notes with stigmatizing keywords to develop our rule-based NLP algorithm. We used MTERMS NLP ecosystem, which extracts information from the notes and determine the semantic meaning based on the context (27). The algorithm identifies sentences containing a stigmatizing keyword(s) and determines whether the sentence contains stigmatization meaning or not based on the context of that language in the sentence. If a clinical note had at least one stigmatizing sentence, it was regarded as a stigmatizing note. The data preprocessing pipeline and NLP algorithm were built with Python3.
We then randomly selected 402 sentences that contained at least one stigmatizing keyword to evaluate the performance of our NLP algorithm. Among these sentences, 201 sentences were sampled from the notes that were determined to be stigmatizing notes based on our algorithm. Another 201 sentences were sampled from the non-stigmatizing notes determined by our algorithm. Manual reviews were performed by four researchers who were not blinded to the algorithm’s assignment. To determine agreement between the reviewers, we first selected 102 sentences to be reviewed by two reviewers. Any inconsistencies were discussed by the research team to obtain consensus. After that, the remaining 300 sentences were reviewed by one reviewer and discussed with the research team for consensus if the reviewer was unsure.
The nuance of determining the presence of stigma or not is more complex than just measuring the incidence of the potentially stigmatizing words. Below are a few examples: 1) the use of the word “alcoholic” is discouraged, yet there are uses of the word which are deemed clinically acceptable, such as “alcoholic cirrhosis” or “alcoholic hepatitis”. In these cases, “alcoholic” was deemed stigma only if it was used as a noun, e.g., “the patient is an alcoholic”; 2) use of the word “abuse” is discouraged, yet it can be acceptable in some cases. If “abuse” was used in “drug abuse” or “alcohol abuse” it was included as stigma, but if it was “emotional abuse” or “physical abuse”, it was not; and 3) use of the word “user” is discouraged, yet the word is commonly used in other contexts, such as “her user login name is.” Therefore, we included the word as stigmatizing if it followed a drug name, such as “he is a marijuana user” or “she is a heroin user.” Finally, we generated a confusion matrix and calculated the accuracy, sensitivity (recall), specificity, precision (positive predictive value (PPV)) and F1 score (the harmonic mean of the precision and recall) to measure the performance of our NLP algorithm.
Data Analysis
We then applied the developed NLP algorithm to all the collected clinical notes for our cohort. A note with at least one stigmatizing sentence was classified as a stigmatizing note. The Chi-square test was used for categorical variables and the independent t-test was used for continuous variables. Levene’s test was used to assess the equality of variances. A p-value less than 0.05 was statistically significant. Descriptive analysis was used on the stigmatizing keywords to count the word frequency and probability of keywords being considered stigmatizing when they occurred in a sentence. The count and percentage of SL were also calculated for different note types. Additionally, logistic regression with univariate and multivariate analysis was used to identify the covariates associated with the SL. All statistical analyses were performed by using R (version 4.0.3) or Python3.
Results
Our final developed NLP algorithm demonstrated adequate sentence-level performance, including: accuracy 0.968, sensitivity/recall 0.945, specificity 0.983, precision/PPV 0.975 and F1 score 0.960. For the one-year study period, there were 30,391 patients seen in our health system who had both a substance-related encounter ICD-10 code and at least one clinical note recorded. The mean age was 47.7 (SD 17.4) years, 37.5% (n=11,348) were female, 81.2% (n=24,655) were White, and 9.3% (n=2,828) reported Hispanic ethnicity. Table 1 shows patient demographic characteristics, both in total and stratified by having at least one stigma-containing note or not. Overall, our algorithm determined that 61.6% (n=18,727) of patients had at least one note containing SL. Black patients (64.7%) and patients with Medicaid insurance (70.1%) had disproportionately higher percentages.
Table 1:
Patient demographic characteristics, in total and stratified by presence of at least one stigma note.
| Characteristics | Total SUD Cohort | With stigma note | Without stigma note | % of having stigma notes | p valueb |
|---|---|---|---|---|---|
| (n=30,391) | (n=18,727 pts) | (n=11,664 pts) | |||
| Number notes/patient | 18.0 (SD 44.3) | 25.3 (SD 54.3) | 6.2 (SD 12.2) | n/a | <0.001 |
| Number encounters/patient | 1.93 (SD 2.63) | 2.20 (SD 3.10) | 1.50 (SD 1.54) | n/a | <0.001 |
| - Subset number of hospital encounters/patient | 1.81 (SD 2.83) | 2.06 (SD 3.24) | 1.10 (SD 0.52) | n/a | <0.001 |
| - Subset number of office visits/patient | 1.83 (SD 2.15) | 1.98 (SD 2.41) | 1.63 (SD 1.72) | n/a | <0.001 |
| Age (years) | 47.7 (SD 17.4) | 47.1 (SD 16.1) | 48.6 (SD 19.2) | n/a | <0.001 |
| Female | 11348 (37.3%) | 6667 (35.6%) | 4681 (40.1%) | 58.8% | <0.001 |
| Racea | <0.001 | ||||
| White | 24655 (81.1%) | 15429 (82.4%) | 9226 (79.1%) | 62.6% | |
| Black | 2040 (6.7%) | 1320 (7.0%) | 720 (6.2%) | 64.7% | |
| Asian | 357 (1.2%) | 142 (0.8%) | 215 (1.8%) | 39.8% | |
| Native | 68 (0.2%) | 38 (0.2%) | 30 (0.3%) | 55.9% | |
| Mixed | 324 (1.1%) | 196 (1.0%) | 128 (1.1%) | 60.5% | |
| Unknown | 2947 (9.7%) | 1602 (8.6%) | 1345 (11.5%) | 54.4% | |
| Ethnicity, Hispanica | 2828 (9.3%) | 1678 (9.0%) | 1150 (9.9%) | 59.3% | 0.009 |
| Insurance | <0.001 | ||||
| - Medicaid | 7960 (26.2%) | 5580 (29.8%) | 2380 (20.4%) | 70.1% | |
| - Medicare | 5698 (18.7%) | 3379 (18.0%) | 2319 (19.9%) | 59.3% | |
| - Commercial | 8143 (26.8%) | 4628 (24.7%) | 3515 (30.1%) | 56.8% | |
| - Uninsured | 880 (2.9%) | 434 (2.3%) | 446 (3.8%) | 49.3% | |
| - Other/Unknown | 7710 (25.4%) | 4706 (25.1%) | 3004 (25.8%) | 61.0% | |
| Marital status | <0.001 | ||||
| - Single | 16305 (53.7%) | 10502 (56.1%) | 5803 (49.8%) | 64.4% | |
| - Married/Partnered | 8841 (29.1%) | 4928 (26.3%) | 3913 (33.5%) | 55.7% | |
| - Divorced | 3247 (10.7%) | 2223 (11.9%) | 1024 (8.8%) | 68.5% | |
| - Widow/Unknown | 1998 (6.6%) | 1074 (5.7%) | 924 (7.9%) | 53.8% | |
| Education Level | <0.001 | ||||
| - College and above | 7678 (25.3%) | 4431 (23.7%) | 3247 (27.8%) | 57.7% | |
| - High school or equivalent | 12839 (42.2%) | 8387 (44.8%) | 4452 (38.2%) | 65.3% | |
| - Didn’t complete high school | 2820 (9.3%%) | 1825 (9.7%) | 995 (8.5%) | 64.7% | |
| - Unknown | 7054 (23.2) | 4084 (21.8%) | 2970 (25.5%) | 57.9% | |
| Veteran status | <0.001 | ||||
| - Yes | 1441 (4.7%) | 917 (4.9%) | 524 (4.5%) | 63.6% | |
| - No | 24659 (81.1%) | 15384 (82.1%) | 9275 (79.6%) | 62.4% | |
| - Unknown | 4291 (14.1%) | 2426 (13.0%) | 1865 (16.0%) | 56.5% |
SUD = Substance use disorder, n/a = not applicable
Values reported as mean (standard deviation (SD)) or n (%)
Self-reported
p values calculated with Student’s T-test for continuous variables and Chi-squared for categorical variables
Table 2 indicates the incidence of stigmatizing words in included notes, as well as the probability of being stigmatizing in the context of the sentence based upon our developed NLP algorithm. The most used stigmatizing words/phrases were “abuse” (n=184,257) and “substance abuse” (n=57,093), for which 95.5% and 100%, respectively, were determined to be stigmatizing. Words on the NIDA list that are commonly used but infrequently stigmatizing in the context of clinical notes are “alcoholic” (10.4% stigmatizing), for example when used as an adjective, “drunk” (3.7% stigmatizing), for example when used as a verb, and “dirty” (10.2% stigmatizing), for example used to describe the bed linens or room rather than the individual or toxicology results. These terms highlight the importance of considering the context in which these words are used.
Table 2:
Statistics of stigmatizing keywords and phrases found in clinical notes.
| Stigmatizing keywords | Total count | Stigmatizing usage count | Non-Stigmatizing usage count | Probability of stigmatizing usage (%) |
|---|---|---|---|---|
| abuse | 184,257 | 175,919 | 8,338 | 95.5% |
| substance abuse | 57,093 | 57,093 | 0 | 100% |
| alcoholic | 46,264 | 4,813 | 41,451 | 10.4% |
| user | 30,261 | 9,581 | 20,680 | 31.7% |
| clean | 15,873 | 2,976 | 12,897 | 18.7% |
| polysubstance dependence | 2,814 | 2,814 | 0 | 100% |
| drunk | 2,192 | 81 | 2,111 | 3.7% |
| replacement therapy | 1,895 | 679 | 1,216 | 35.8% |
| medication assisted treatment | 1,112 | 1,112 | 0 | 100% |
| addicted | 854 | 854 | 0 | 100% |
| dirty | 785 | 80 | 705 | 10.2% |
| addict | 651 | 651 | 0 | 100% |
| substance dependence | 569 | 569 | 0 | 100% |
| abuser | 216 | 181 | 35 | 83.8% |
| junkie | 55 | 55 | 0 | 100% |
| drug abuser | 41 | 41 | 0 | 100% |
| substance abuser | 21 | 21 | 0 | 100% |
| opioid replacement therapy | 9 | 9 | 0 | 100% |
| former addict | 4 | 4 | 0 | 100% |
| reformed addict | 0 | 0 | 0 | N/A |
| addicted baby | 0 | 0 | 0 | N/A |
| opioid substitution | 0 | 0 | 0 | N/A |
Table 3 describes the note types included in the analysis. Overall, there were 546,309 notes for our 30,391 patients, of which 100,792 (18.4%) contained stigma. The most common note type was “progress note,” for which 15.4% contained SL. History and Physical notes were most likely to contain SL (59.1%). In general, nursing notes, including ED triage notes (5.6%), nursing summaries (1.7%) and nursing notes (2.4%) had lower incidences of stigma. This finding was corroborated in Table 4, which lists SL by encounter type (hospital vs. office visit) and by the provider type of the note’s author. Hospital encounters were more likely to contain SL than were office visits (19.8% vs. 14.0%, p<0.001). About a quarter of physician and nurse practitioner notes (25.8% and 24.2%, respectively) were stigmatizing, physician assistants had a higher proportion of SL (46.9%) and nurses had a lower rate (4.1%).
Table 3:
Distribution of note types.
| Note Type | Total notes | Note character length, mean (SD) | Number of Stigma Notes | Number of non-Stigma Notes | % of stigma notes |
|---|---|---|---|---|---|
| n=546,309 | N=100,792 | n=445,517 | % of note type | ||
| Progress Notes | 234,559 | 3,941 (6,004) | 36,221 | 198,338 | 15.4% |
| Assessment & Plan Note | 96,156 | 359 (596) | 2,090 | 94,066 | 2.2% |
| aED Notes | 71,051 | 4,984 (6,662) | 25,534 | 45,517 | 35.9% |
| Group Note | 62,037 | 1,284 (892) | 18,878 | 43,159 | 30.4% |
| Plan of Care | 27,702 | 1,692 (1,647) | 213 | 27,489 | 0.8% |
| Consults | 21,014 | 11,347 (10,876) | 9,960 | 11,054 | 47.4% |
| ED Triage Notes | 16,610 | 359 (265) | 922 | 15,688 | 5.6% |
| History and Physical | 6,919 | 15,629 (11,359) | 4,090 | 2,829 | 59.1% |
| Discharge Summary | 6,207 | 23,273 (18,696) | 2,654 | 3,553 | 42.8% |
| Nursing Summary | 2,185 | 787 (760) | 38 | 2,147 | 1.7% |
| Nursing Note | 987 | 782 (1,176) | 24 | 963 | 2.4% |
| Transfer / Sign Off Note | 878 | 8,323 (9,954) | 168 | 710 | 19.1% |
| Opioid Note | 4 | 1,020 (487) | 0 | 4 | 0.0% |
ED = Emergency Department
ED notes include: ED Observation Initial Note, ED Observation Progress/Update Note, ED Observation Disposition Note, ED Provider Notes, ED Progress/Update Note, ED Notes
Table 4:
Note-level characteristics, based on encounter type and note author role.
| Characteristics | Total SUD notes | Stigma notes | Non-stigma notes | % of stigma notes | p value |
|---|---|---|---|---|---|
| (n=546,309) | (n=100,792) | (n=445,517) | |||
| Note Info | |||||
| Encounter type | <0.001 | ||||
| - Hospital encounter | 416,520 (76.2) | 82,651 (82.0) | 333,869 (74.9) | 19.8% | |
| - Office visit | 129,789 (23.8) | 18,141 (18.0) | 111,648 (25.1) | 14.0% | |
| Provider Info | |||||
| Provider Type | <0.001 | ||||
| - MD (n=5,222) | 176,524 (32.3) | 45,501 (45.1) | 131,023 (29.4) | 25.8% | |
| - Nurse (n=7,474) | 129,306 (23.7) | 5,263 (5.2) | 124,043 (27.8) | 4.1% | |
| - Nurse Practitioner (n=786) | 30,214 (5.5) | 7,805 (7.7) | 22,409 (5.0) | 25.8% | |
| - Mental Health Worker (n=584) | 63,503 (11.6) | 10,218 (10.1) | 53,285 (12.0) | 16.1% | |
| - Social Worker (n=442) | 42,719 (7.8) | 10,345 (10.3) | 32,374 (7.3) | 24.2% | |
| - Physician Assistant (n=581) | 21,704 (4.0) | 10,172 (10.1) | 11,532 (2.6) | 46.9% | |
| - Unknown (n=511) | 33,191 (6.1) | 3,760 (3.7) | 29,431 (6.6) | 11.3% | |
| - Others (n=2,188) | 49,148 (9.0) | 7,728 (7.7) | 41,420 (9.3) | 15.7% |
Values are expressed as number of notes (percent).
P-values were calculated with Chi-squared analysis
Finally, Table 5 shows the multivariable analysis of patient-level factors associated with having at least one stigmatizing note. Notable findings include: male patients are more likely to have SL notes than female patients (adjusted odds ratio (aOR) 1.17), younger patients aged 18–24 years were less likely than patients aged 45–54 years (aOR 0.55) to have stigma in their notes, and those with Medicaid or Medicare insurance were more likely than those with commercial insurance to have SL notes (aOR 1.41 and 1.23, respectively). There was no difference in stigma between White and Black patients (aOR 0.94), although Asian patients were much less likely to have stigma (aOR 0.45). Likewise, patients who had Hispanic ethnicity were also less likely to have stigma compared with non-Hispanic individuals (aOR 0.88). Additionally, patients with a greater number of encounters were more likely to have SL notes, with the greatest number experienced by patients with ≥20 encounters (aOR 6.60), indicating that the more patients interacted with the health system, the greater the odds of having at least one note containing SL.
Table 5:
Multivariable analysis of patient factors associated with having at least one note containing stigmatizing language.
| Adjusted Odd Ratio | 95% confidence interval | p value | |
|---|---|---|---|
| Gender | |||
| Male | 1.17 | 1.11–1.23 | <0.001 |
| Female | 1 (reference) | ||
| Age | |||
| 18–24 | 0.55 | 0.50–0.61 | <0.001 |
| 25–34 | 0.91 | 0.83–0.99 | 0.032 |
| 35–44 | 0.98 | 0.90–1.07 | 0.715 |
| 45–54 | 1 (reference) | ||
| 55–64 | 0.86 | 0.79–0.93 | <0.001 |
| 65–74 | 0.62 | 0.56–0.68 | <0.001 |
| ≥ 75 | 0.37 | 0.32–0.42 | <0.001 |
| Race | |||
| White | 1 (reference) | ||
| Black | 0.94 | 0.85–1.04 | 0.228 |
| Asian | 0.45 | 0.36–0.56 | <0.001 |
| Mixed | 0.91 | 0.72–1.15 | 0.409 |
| Native | 0.75 | 0.46–1.23 | 0.249 |
| Other/unknown | 0.75 | 0.67–0.84 | <0.001 |
| Ethnicity | |||
| Hispanic | 0.88 | 0.80–0.98 | 0.020 |
| Non-Hispanic | 1 (reference) | ||
| Marital Status | |||
| Single | 1 (reference) | ||
| Married | 0.77 | 0.72–0.82 | <0.001 |
| Partnered | 0.89 | 0.72–1.10 | 0.257 |
| Divorced | 1.17 | 1.07–1.28 | <0.001 |
| Widow | 1.11 | 0.95–1.29 | 0.182 |
| Unknown | 0.69 | 0.60–0.80 | <0.001 |
| Education Level | |||
| College and above | 1 (reference) | ||
| High school or equivalent | 1.19 | 1.12–1.27 | <0.001 |
| Didn’t complete high school | 1.25 | 1.13–1.38 | <0.001 |
| Unknown | 1.03 | 0.95–1.11 | 0.49 |
| Insurance | |||
| Medicaid | 1.41 | 1.31–1.51 | <0.001 |
| Medicare | 1.23 | 1.13–1.34 | <0.001 |
| Commercial | 1 (reference) | ||
| Uninsured | 0.79 | 0.68–0.92 | 0.002 |
| Other/Unknown | 1.22 | 1.14–1.30 | <0.001 |
| Veteran status | |||
| Yes | 1.20 | 1.06–1.35 | 0.003 |
| No | 1 (reference) | ||
| Unknown | 1.08 | 0.99–1.16 | 0.072 |
| Number of encounters | |||
| < 2 | 1 (reference) | ||
| 2–4 | 2.06 | 1.90–2.24 | <0.001 |
| 5–9 | 2.46 | 2.10–2.89 | <0.001 |
| 10–19 | 2.96 | 2.24–3.97 | <0.001 |
| ≥ 20 | 6.60 | 3.28–15.74 | <0.001 |
Discussion
Our research demonstrates that the majority of patients with substance-related diagnoses in our health system had SL in their medical records. This is particularly concerning in the context provided by the current director of the White House Office of National Drug Control Policy, who cited stigma by health care providers as a significant barrier to patients receiving addiction treatment (28). The language used in medical records is becoming increasingly important as patients have access to some of their medical records through the 21st Century Cures Act (29). These “open notes” have created controversy for patients with behavioral health conditions (30). Patients with SUD who experience stigma in healthcare settings have negative attitudes towards seeking care and earned mistrust of the healthcare system, emphasizing the role stigma has as a driver of healthcare disparities (31).
The frequency of which SL was used varied in our findings. The most common phrases were “abuse” and “substance abuse”, representing about 90% of the instances we detected. These terms may remain in the medical vernacular given their presence in the ICD-10 nomenclature and that language norms change gradually over time. Our analysis captured only notes and not these diagnosis codes, but clinicians may view these terms as acceptable, especially since federal agencies (e.g. the National Institute on Drug Abuse, the National Institute on Alcohol Abuse and Alcoholism, and the Substance Abuse and Mental Health Services Administration) continue to use these terms in their titles and they appear in common screening tools, like the Drug Abuse Screening Test (DAST-10). Likewise, the previous version of the Diagnostic and Statistical Manual of Mental Disorders (DSM), fourth edition, also used “abuse” but was transitioned to “use disorder” in the newer fifth edition (32). Although not the focus of this research, it may be that certain stigmatizing words have variable effects on patients, and words that defined individuals that may provoke more visceral response, like “junkie” (31) and “addict” (33) were less common but still present in notes. Further research could determine how patients variably react to these words and phrases when they see them written in their medical records.
There were differences in the types of notes and note authors that were more likely to contain or use SL. History and physical notes, consult notes and discharge summaries had the highest proportion, and hospital encounter notes were more likely than office visit notes to have this language. The findings make the case for more in-hospital education about stigma, as providers who do not consistently or commonly work with SUD populations may be less attuned to the effect SL has on patients with substance-related diagnoses. Likewise, the note authors also had variable frequencies of using SL, with the highest being physician assistants (46.9% of notes). It is unclear why this group had the highest proportion, when the other advanced practice provider group (nurse practitioners) had a rate closer to that of physician authors. Previous work has described a physician assistant-specific educational intervention to reduce stigma, and this may be useful in our system (34).
Finally, a primary aim of this study was to determine if there were disparities in use of SL based on patient characteristics. In the logistic regression analysis, men were more likely to have SL than women. A study of patients with co-occurring substance use and mental health disorders found that women were more likely than men to stay in residential treatment (35). Likewise, a study in Australia found that men with SUD had lower odds of seeking help when compared to women (36). Future research could investigate if stigma plays a role in these differences. Unfortunately, we were unable to ascertain if patients were transgender, and we are sensitive to the tremendous stigma that transgender patients face (37). Another notable finding was that patients with more encounters had a higher probability of SL-containing notes. This finding could indicate that patients with more encounters have more severe disease and may be more prone to stigma, or simply that the more notes that are written about patients, the greater the possibility they will be exposed to SL at some point.
Although Black patients overall had disproportionately higher percentages of SL in their notes, in the adjusted analysis we did not detect differences between Black and White race which may indicate additional factors driving the use of SL. However, given the recent marked increase in opioid-related overdose deaths amongst Black individuals, further evaluation to understand intersectional biases and factors driving stigma are needed (38). We discovered that Asian individuals were much less likely to have SL in their medical records. The reason for this finding is unclear, but it may be indicative of the heterogeneous composition of the “Asian” race category and numerous factors such as family, immigration status, religion and language that affect how medical providers view their patients (39). Likewise, we found that patients with Hispanic ethnicity were less likely than non-Hispanic patients to have SL. This finding mirrors other recent work we did at our health system that discovered that Hispanic patients were more likely to be provided with a prescription for naloxone after treatment for opioid overdose than non-Hispanic patients (40). Other patient-level factors – such as being single or divorced compared to married, less educated compared to higher degrees of education, with Medicaid or Medicare insurance compared to commercial, and being a veteran – were all associated with increased presence of SL. These findings highlight the intersectionality between substance use, stigma and social determinants of health and should serve as important reminders to clinical staff to assess their biases and be aware of how words used may affect these populations.
The next steps in this work are to increase awareness about the use of SL with clinicians. Simply sharing the results of this study may encourage those who are writing notes to use caution with the words they use, both for patients experiencing substance-related diagnoses as well as those with other medical conditions. The work could take the form of additional clinician education or via audit and feedback of actual notes by colleagues. Finally, engaging patients to share their experiences with stigma might further influence clinician behavior.
Limitations
This study was based at a single health system and findings may not be the same elsewhere. As manual review of a year’s worth of clinical notes was not feasible, we relied on our developed NLP algorithm. The algorithm had adequate test characteristics but was not perfect. There may have been cases where a patient reported a stigmatizing word that was recorded in the medical record as a quote (e.g., “the patient reports that she is an ‘addict’”), which would be included as a phrase containing stigma. However, including even direct quotations of SL in notes may encourage self-stigma and might not be desirable, but this is controversial. We also did not exclude the phrase “Alcoholics Anonymous;” a post-hoc analysis revealed that there were 635 instances of this term in the records, and they were retained as SL in our analyses. Additionally, we discovered some notes which appeared to use a premade template that said “substance abuse evaluation” followed by free text from the note author. We elected to include those instances as they are internally created and able to be edited and still seen by other providers and patients and thus perpetuate stigma. A noteworthy limitation is around accuracy of race and ethnicity data in our electronic health record. This is a self-reported field that may be overlooked or unverified, which may have impacted the analysis of disparities. Likewise, the sex/gender variable limited our ability to study the effects of SL on transgender or non-binary patients. Finally, the NIDA “Words Matter” document may not be a gold standard regarding which terms are stigmatizing or problematic.
Conclusions
In this study of over 500,000 clinical notes for patients with a substance-related diagnosis treated in a large health system, we detected that about 62% of patients had at least one note containing SL, and 18% of all notes had SL. Furthermore, there were several notable patient characteristic disparities associated with patients having these words and phrases in their notes. The work suggests that more clinician education or other interventions about use of SL is needed.
Supplementary Material
Conflicts of Interest and Source of Funding:
Dr. Weiner is funded by National Institutes of Health grants 5R01DA044167 and 5R01HS026753. Dr. Lo’s work was funded by Mass General Brigham’s United Against Racism initiative. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Dr. Weiner is on the acute pain committee of Vertex Pharmaceuticals, Inc, and also employed by Bicycle Health, Inc. Other authors report no relevant conflicts of interest.
Footnotes
A preprint server was not used.
This work has been submitted for presentation at the 54th Annual Conference of the American Society of Addiction Medicine, Washington DC, April 2023 (decision pending).
References
- 1.National Institute on Drug Abuse. Overdose Death Rates. Available at: https://nida.nih.gov/research-topics/trends-statistics/overdose-death-rates. Accessed December 17, 2022.
- 2.U.S. Department of Health & Human Services. HHS releases new buprenorphine practice guideline, expanding access to treatment or opioid use disorder. April 27, 2021. Available at: https://www.hhs.gov/about/news/2021/04/27/hhs-releases-new-buprenorphine-practice-guidelines-expanding-access-to-treatment-for-opioid-use-disorder.html. Accessed December 17, 2022.
- 3.Kral AH, Lambdin BH, Wenger LD, Davidson PJ. Evaluation of an Unsanctioned Safe Consumption Site in the United States. N Engl J Med. 2020. Aug 6;383(6):589–590. [DOI] [PubMed] [Google Scholar]
- 4.Wakeman SE. Another Senseless Death - The Case for Supervised Injection Facilities. N Engl J Med. 2017. Mar 16;376(11):1011–1013. [DOI] [PubMed] [Google Scholar]
- 5.Legislative Analysis and Public Policy Association. Model Expanded Access to Emergency Opioid Antagonists Act. November 16, 2021. Available at: https://legislativeanalysis.org/model-expanded-access-to-emergency-opioid-antagonists-act/. Accessed December 17, 2022.
- 6.U.S. Department of Health and Human Services. Overdose Prevention Strategy. Available at: https://www.hhs.gov/overdose-prevention/. Accessed December 17, 2022.
- 7.Merriam-Webster Online Dictionary. Available at: https://www.merriam-webster.com/dictionary/stigma. Accessed December 17, 2022.
- 8.Ahern J, Stuber J, Galea S. Stigma, discrimination and the health of illicit drug users. Drug Alcohol Depend. 2007. May 11;88(2–3):188–96. [DOI] [PubMed] [Google Scholar]
- 9.Luoma JB, Twohig MP, Waltz T, Hayes SC, Roget N, Padilla M, Fisher G. An investigation of stigma in individuals receiving treatment for substance abuse. Addict Behav. 2007. Jul;32(7):1331–46. [DOI] [PubMed] [Google Scholar]
- 10.Olsen Y, Sharfstein JM. Confronting the stigma of opioid use disorder--and its treatment. JAMA. 2014. Apr 9;311(14):1393–4. [DOI] [PubMed] [Google Scholar]
- 11.van Boekel LC, Brouwers EP, van Weeghel J, Garretsen HF. Stigma among health professionals towards patients with substance use disorders and its consequences for healthcare delivery: systematic review. Drug Alcohol Depend. 2013. Jul 1;131(1–2):23–35. [DOI] [PubMed] [Google Scholar]
- 12.Yang LH, Wong LY, Grivel MM, Hasin DS. Stigma and substance use disorders: an international phenomenon. Curr Opin Psychiatry. 2017. Sep;30(5):378–388. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Botticelli MP, Koh HK. Changing the Language of Addiction. JAMA. 2016. Oct 4;316(13):1361–1362. [DOI] [PubMed] [Google Scholar]
- 14.Matthews S, Dwyer R, Snoek A. Stigma and Self-Stigma in Addiction. J Bioeth Inq. 2017. Jun;14(2):275–286. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Kelly JF, Westerhoff CM. Does it matter how we refer to individuals with substance-related conditions? A randomized study of two commonly used terms. Int J Drug Policy. 2010. May;21(3):202–7. [DOI] [PubMed] [Google Scholar]
- 16.Ashford RD, Brown AM, McDaniel J, Curtis B. Biased labels: An experimental study of language and stigma among individuals in recovery and health professionals. Subst Use Misuse. 2019;54(8):1376–1384. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Ashford RD, Brown AM, Curtis B. “Abusing Addiction”: Our Language Still Isn’t Good Enough. Alcohol Treat Q. 2019;37(2):257–272. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Kelly JF, Richard S, Wakeman S. Language, Substance Use Disorders, and Policy: The Need to Reach Consensus on an “Addiction-ary”. Alcoholism Treatment Quarterly, 34:1, 116–123. [Google Scholar]
- 19.Harris, Felman K. A guide to the use of recovery-oriented language. 2012. Available at: https://assets.website-files.com/5a13853ae83c4100019201c8/5fd91448e81d685e8729a58a_Recovery-Oriented%20Language.pdf. Accessed December 17, 2022.
- 20.National Institute on Drug Abuse. Words Matter - Terms to Use and Avoid When Talking About Addiction. Available at: https://nida.nih.gov/nidamed-medical-health-professionals/health-professions-education/words-matter-terms-to-use-avoid-when-talking-about-addiction. Accessed December 17, 2022.
- 21.Park J, Saha S, Chee B, Taylor J, Beach MC. Physician Use of Stigmatizing Language in Patient Medical Records. JAMA Netw Open. 2021. Jul 1;4(7):e2117052. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.De Ruddere L, Craig KD. Understanding stigma and chronic pain: a-state-of-the-art review. Pain. 2016. Aug;157(8):1607–1610. [DOI] [PubMed] [Google Scholar]
- 23.Liu NF, Brown AS, Folias AE, Younge MF, Guzman SJ, Close KL, Wood R. Stigma in People With Type 1 or Type 2 Diabetes. Clin Diabetes. 2017. Jan;35(1):27–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Fernández L, Fossa A, Dong Z, Delbanco T, Elmore J, Fitzgerald P, Harcourt K, Perez J, Walker J, DesRoches C. Words Matter: What Do Patients Find Judgmental or Offensive in Outpatient Notes? J Gen Intern Med. 2021. Sep;36(9):2571–2578. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Himmelstein G, Bates D, Zhou L. Examination of Stigmatizing Language in the Electronic Health Record. JAMA Netw Open. 2022. Jan 4;5(1):e2144967. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Kilbanksi A. Mass general Brigham president and CEO’s update to employees on racial injustice. Available at: https://www.massgeneralbrigham.org/newsroom/articles/mass-general-brigham-president-and-ceos-update-employees-racial-injustice. Accessed December 17, 2022.
- 27.Zhou L, Plasek JM, Mahoney LM, Karipineni N, Chang F, Yan X, Chang F, Dimaggio D, Goldman DS, Rocha RA. Using Medical Text Extraction, Reasoning and Mapping System (MTERMS) to process medication information in outpatient clinical notes. AMIA Annu Symp Proc. 2011;2011:1639–48. [PMC free article] [PubMed] [Google Scholar]
- 28.Facher L. Rahul Gupta, first physician to serve as druz czar, says stigma among doctors is key culprit in addiction crisis. Stat News, September 2, 2022. Available at: https://www.statnews.com/2022/09/02/rahul-gupta-drug-czar-says-stigma-among-doctors-adds-to-addiction-crisis/. Accessed December 17, 2022. [Google Scholar]
- 29.Delbanco T, Wachenheim D. Open Notes: New Federal Rules Promoting Open and Transparent Communication. Jt Comm J Qual Patient Saf. 2021. Apr;47(4):207–209. [DOI] [PubMed] [Google Scholar]
- 30.Schwarz J, Bärkås A, Blease C, Collins L, Hägglund M, Markham S, Hochwarter S. Sharing Clinical Notes and Electronic Health Records With People Affected by Mental Health Conditions: Scoping Review. JMIR Ment Health. 2021. Dec 14;8(12):e34170. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Muncan B, Walters SM, Ezell J, Ompad DC. “They look at us like junkies”: influences of drug use stigma on the healthcare engagement of people who inject drugs in New York City. Harm Reduct J. 2020. Jul 31;17(1):53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Recovery Research Institute. Comparison Between DSM-IV and DSM-5. Available at: https://www.recoveryanswers.org/resource/comparison-between-dsm-iv-dsm-5/. Accessed December 17, 2022.
- 33.Dean JC, Rud F. The drug addict and the stigma of addiction. Int J Addict. 1984. Dec;19(8):859–69 [DOI] [PubMed] [Google Scholar]
- 34.Crapanzano K, Vath RJ. Long-Term Effect of a Stigma-Reduction Educational Intervention for Physician Assistants. J Physician Assist Educ. 2017. Jun;28(2):92–95. [DOI] [PubMed] [Google Scholar]
- 35.Choi S, Adams SM, Morse SA, MacMaster S. Gender differences in treatment retention among individuals with co-occurring substance abuse and mental health disorders. Subst Use Misuse. 2015. Apr;50(5):653–63. [DOI] [PubMed] [Google Scholar]
- 36.Harris MG, Baxter AJ, Reavley N, Diminic S, Pirkis J, Whiteford HA. Gender-related patterns and determinants of recent help-seeking for past-year affective, anxiety and substance use disorders: findings from a national epidemiological survey. Epidemiol Psychiatr Sci. 2016. Dec;25(6):548–561. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Wolfe HL, Biello KB, Reisner SL, Mimiaga MJ, Cahill SR, Hughto JMW. Transgender-related discrimination and substance use, substance use disorder diagnosis and treatment history among transgender adults. Drug Alcohol Depend. 2021. Jun 1;223:108711. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Centers for Disease Control and Prevention. Overdose death rates increased significantly for Black, American Indian/Alaska Native people in 2020. July 19, 2022. Available at https://www.cdc.gov/media/releases/2022/s0719-overdose-rates-vs.html. Accessed December 17, 2022.
- 39.Chang JS, Sorensen JL, Masson CL, Shopshire MS, Hoffman K, McCarty D, Iguchi M. Structural factors affecting Asians and Pacific Islanders in community-based substance use treatment: Treatment provider perspectives. J Ethn Subst Abuse. 2017. Oct-Dec;16(4):479–494. [DOI] [PubMed] [Google Scholar]
- 40.Weiner SG, Carroll AD, Brisbon NM, Rodriguez CP, Covahey C, Stringfellow EJ, DiGennaro C, Jalali MS, Wakeman SE. Evaluating disparities in prescribing of naloxone after emergency department treatment of opioid overdose. J Subst Abuse Treat. 2022. Aug;139:108785. [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.