Words Matter: Stigmatizing language in medical records of individuals electing medication for opioid use disorder

Abstract

Background:

There has been a concerted effort to minimize the use of stigmatizing language in healthcare settings; however, its prevalence in health record documentation remains high. This study examined clinicians’ stigmatizing and patients’ self-stigmatizing language in medical records for patients electing medication for opioid use disorder in two settings, office-based opioid treatment and opioid treatment programs.

Methods:

In this retrospective cohort study, data were extracted from health records for a distributed provider network serving publicly funded patients in Texas. Patients seeking medication treatment for opioid use disorder between December 2020 and November 2021 were included. Using a natural language processing algorithm to review open text fields in the health records, we identified the prevalence of stigmatizing language identified in the National Institute on Drug Abuse’s “Words Matter” tool.

Results:

The analytic sample included 1,391 patients (63.8% White, 27.5% Hispanic, 40.5% female, M_age= 40.8 years [SD=10.3]). Of the 6,099 notes analyzed, 81.1% (n=4944) contained stigmatizing language and 7.1% (n=430) contained self-stigmatizing language. In multivariate models, stigmatizing language was associated with age, race, rurality, education, income, opioid use, and treatment setting. Self-stigmatizing language was associated with treatment setting.

Conclusions:

Findings suggest that a patient’s sociodemographic background influences the stigmatizing language their clinicians use while treatment setting influences both provider and self-stigmatizing language. Additional research is needed to explore whether differences in stigmatizing language contributes to differential treatment outcomes seen in different opioid treatment settings.

1. Introduction

Stigma is the negative attitudes and beliefs that people implicitly or explicitly have about a group of people based on certain characteristics of the group. Although there have been efforts to reduce stigma surrounding mental health conditions with some positive results seen for depression (Pescosolido et al., 2021), it remains prevalent in addiction. Stigma comes in many forms, including societal and self-stigma (Andersen et al., 2022). It is one of the many challenges individuals with opioid use disorder (OUD) face, and it impacts the prescribing and receipt of medication for opioid use disorder (MOUD). (Cernasev et al., 2021; Cheetham et al., 2022; Stone et al., 2021; Wakeman and Rich, 2018; Witte et al., 2021).

Societal stigma includes the negative attitudes some people have around individuals with OUD that are based on individual characteristics of the patients themselves, their substance use, and their decision to seek MOUD (Bradford et al., 2024; Cheetham et al., 2022; Witte et al., 2021). Stigma is also commonly expressed in written form, including in medical documentation (Bradford et al., 2024; Hirshman et al., 2025). The presence of stigmatizing language in medical records can ultimately lead to discriminatory behaviors by other providers that affect the treatment OUD patients receive (Cheetham et al., 2022). This written language has the potential to propagate the stigma associated with OUD and individuals taking MOUD.

Self-stigma involves patients with OUD internalizing these negative societal attitudes, as well as their own negative attitudes about themselves. Internalizing stigma is common, especially among those with OUD. This leads these individuals to believe that they, their addiction, and their decision to seek MOUD makes them “less than” when comparing themselves to others. Self-stigma leads to depression, psychological distress, low self-esteem, low self-efficacy, and decreased help-seeking, ultimately impeding recovery efforts (Chang et al., 2019; da Silveira et al., 2018; Hammarlund et al., 2018; Saffari et al., 2022). Combating self-stigma will go hand in hand with combating societal stigma.

The National Institute on Drug Abuse (NIDA) released its “Words Matter” tool (NIDA, 2024) that advises health care professionals and individuals in the community to avoid specific words/phrases and provides alternative preferred language (e.g., instead of “addict,” use “person with a substance use disorder”). This is recommended for both verbal and written communication surrounding substance use. The premise behind this tool is to change the public perception around substance use disorders and those who struggle with substances to ensure that they are treated with respect and to change the culture of treatment (e.g., prescribers offering MOUD).

Reducing stigma around OUD and medications for OUD is necessary to optimize treatment delivery, adherence, and success (Witte et al., 2021). Despite knowledge of stigma and its potential to impact OUD treatment, it remains prevalent in healthcare documentation. Treatment programs offer MOUD care for some of the most vulnerable patients with OUD and the use of stigmatizing language has not been fully explored in these settings. The objective of this study was to examine predictors of stigmatizing and self-stigmatizing language documented in medical records of patients receiving MOUD. We hypothesized that patient-level characteristics and clinical treatment setting would be associated with stigmatizing and self-stigmatizing language detected in patient medical records.

2. Method

2.1. Patients and Procedures

The study population included patients with a diagnosis of OUD who were 200% below the federal poverty limit, uninsured or underinsured, at least 18 years old, and a resident of Texas. Priority populations were granted immediate access to treatment, including pregnant women, individuals who were injecting drugs, individuals at high risk for overdose, and individuals referred by the Department of Family Protective Services.

All patients received MOUD in a statewide network of office-based opioid treatment (OBOTs) programs and opioid treatment programs (OTPs). Buprenorphine and naltrexone are commonly provided in OBOTs, such as primary care clinics and outpatient substance use clinics. Federal regulations only allow methadone to be dispensed at OTPs, and they are also permitted to provide other medications for OUD (i.e. buprenorphine and naltrexone). Methadone is administered in-person and under the direct supervision of the clinic staff, and take-home doses are tightly controlled and only allowed after patients demonstrate adherence to and stability in treatment. With no strict federal regulations present in OBOTs, several OBOT clinics allow for and encourage the use of telemedicine. Clinicians practicing in OBOTs will typically provide weekly or monthly prescriptions, and urine drug screenings are performed but not required at a specific frequency. OTPs usually begin with daily observed dosing for methadone patients, and they require frequent drug testing. Clinic staff treating patients were required to attend and periodically present a case at the monthly virtual educational sessions organized by our staff. Clinicians also had training in evidence-based interventions, such as motivational enhancement therapy, motivational interviewing, overdose prevention, and seeking safety. For further description of the statewide network and population, see Marino et al., 2024.

This study used data from progress notes, substance use assessments, screenings, referrals, and other clinical documentation that were contained in the health records. The types of healthcare providers that submitted documentation were counselor interns, licensed clinical drug counselors, licensed medical social workers, licensed practicing clinicians, medical doctors, recovery coaches, registered nurses, and advanced practice registered nurses.

All patients with at least one note in their medical record from December 2020 to November 2021 were included. This study was deemed exempt from IRB review, and patients were not required to provide informed consent.

2.2. Measures

Sociodemographic variables analyzed were age, sex, race, ethnicity, rurality, and education. These were obtained from the medical record. Clinicians recorded comorbid mental health and substance use disorders. Days of opioid use at the beginning of treatment in the past 30 days and intravenous drug use history were self-reported by patients and documented by clinicians.

2.3. Data Analysis

Stigmatizing and self-stigmatizing language was based on NIDA’s “Words Matter” tool (NIDA, 2024), which was used to create a lexicon of stigmatizing words and phrases for addiction. To systematically and accurately identify stigmatizing language occurring in the large volume of unstructured clinical text, we leveraged natural language processing techniques. We used tidy text to preprocess the free-text data by cleaning and transforming the text data into tokens. To process the new structured data, we applied predetermined rules-based pattern matching to detect and classify language, which allowed us to create a binary code. This binary code classified whether notes and words/phrases contained stigmatizing language that matched with the stigmatizing language lexicon (yes/no). In addition, we checked all words and phrases and classified each as being contained in quotations or not. This allowed us to categorize stigmatizing language as being from patients directly (self-stigmatizing) or clinicians.

Our approach relies heavily on pattern recognition. The use of this system allows us to in, a repeatable fashion, screen and identify obvious or repeated language patterns by simplifying these large volumes of complex narratives. To increase reliability in successfully identifying contextually accurate words and phrases, we also assessed 50 samples of flagged instances to verify if they represented stigmatizing or self-stigmatizing language.

Our primary outcome (language frequency) was a count variable. (Gardner et al., 1995). Both the stigmatizing and self-stigmatizing language count data showed significant overdispersion based on the dispersion parameters (c = 1.6050 and 2.6342) warranting the use of a negative binomial model (Gardner et al, 1995). To examine the frequency and predictors of stigmatizing and self-stigmatizing language among various patient groups and opioid treatment settings (i.e., OBOT, OTP), we used multivariate regression models (Anvari et al., 2022; Burgess et al., 2021; Han et al., 2024; Weiner et al., 2023). Analyses were conducted in R statistical software.

3. Results

Of the 1,391 patients included, 631 (45.4%) were seen in OBOTs and 760 (54.6%) were seen in OTPs. The sample was 63.8% White, 27.5% Hispanic, 40.5% female, and the average age was 40.8 years (SD=10.3).

There were 6,099 notes in the analysis. Within the notes, 81.1% (n=4944) contained stigmatizing language and 7.1% (n=430) contained self-stigmatizing language. The most common stigmatizing language was “abuse/abuser” and “clean.” The most common self-stigmatizing language was “clean” and “addict.” See word clouds displaying stigmatizing and self-stigmatizing language (Figures 1–3).

Figure 1.

Word cloud displaying stigmatizing language across providers and patients

Figure 3.

Word cloud displaying stigmatizing language used by providers

In a multivariate model, stigmatizing language was more likely to be in the medical record of patients who were older, rural compared to non-rural, and those with higher income. Stigmatizing language was less likely to be in the medical record of patients who were multiracial compared to White, had a high school or equivalent degree compared to those with less than a high school education, were using opioids at treatment initiation, and those who were seen in OTPs compared to OBOTs (Table 1).

Table 1.

Predicting stigmatizing language used by clinicians in medical documentation

Stigmatizing Language Model
Variable	Exp	Std error	p value
(Intercept)	4.35	0.25	< .001
Age	1.02	0.00	< .001
Sex	0.93	0.09	.44
Race
Native American/Native
Hawaiian and Pacific Islander	0.94	0.51	.25
Black or African American	1.27	0.19	.19
Multiracial - Unspecified	0.38	0.50	.04
Ethnicity	0.91	0.09	.35
Rurality	1.58	0.10	< .001
Education
Other/unknown	1.02	0.10	.99
Highschool or equivalent	0.82	0.10	.04
College and above	0.69	0.31	.24
Income	1.01	0.01	.01
Co-occurring Mental Health Disorder	1.35	0.27	.28
Co-occurring Substance Use Disorder	1.01	0.96	.99
Days of Opioid Use at Beginning of Treatment	0.99	0.00	.01
IV Drug Use	1.06	0.09	.54
Treatment Setting	0.67	0.09	< .001

Reference groups: female (sex), White (race), non-Hispanic (ethnicity), non-rural (rurality), less than high school (education), no (co-occurring mental health disorder), no (co-occurring substance use disorder), no (IV drug use), OBOT (treatment setting).

In a separate multivariate model, self-stigmatizing language was less likely to be in the medical record of those who were being seen in OTPs compared to OBOTs (Table 2). No sociodemographic or substance use variables were associated with self-stigmatizing language.

Table 2.

Predicting self-stigmatizing language in medical documentation

Self-Stigmatizing Language Model
Variable	Exp	Std error	p value
(Intercept)	1.75	0.60	.35
Age	1.02	0.01	.09
Sex	1.12	0.02	.63
Race
Native American/Native
Hawaiian and Pacific Islander	0.57	0.44	.19
Black or African American	1.15	0.49	.98
Multiracial - Unspecified	0.80	0.73	.76
Ethnicity	1.34	0.24	.22
Rurality	1.12	0.28	.67
Education
Other/Unknown	0.93	0.29	.80
Highschool or equivalent	0.93	0.24	.77
College and above	0.84	0.90	.85
Income	0.99	0.00	.86
Co-occurring Mental Health Disorder	1.30	0.52	.90
Co-occurring Substance Use Disorder	0.90	0.03	.70
Days of Opioid Use at Beginning of Treatment	0.99	0.00	.17
IV Drug Use	1.05	0.02	.98
Treatment Setting	0.53	0.02	.01

Reference groups: female (sex), White (race), non-Hispanic (ethnicity), non-rural (rurality), less than high school (education), no (co-occurring mental health disorder), no (co-occurring substance use disorder), no (IV drug use), OBOT (treatment setting).

4. Discussion

As a social phenomenon, stigma serves to diminish a group of individuals based on certain characteristics of the group. Our study suggests that the sociodemographic identity of our patients and their treatment setting impacted the use of stigmatizing language. Stigmatizing language was more likely to be in the medical record for those who were older, rural, and who had a higher income. Stigmatizing language was less likely to be in the medical record of those with a high school degree or equivalent compared to those with less than a high school degree and less likely for multiracial patients. Self-stigmatizing language was not associated with any of the sociodemographic or substance use characteristics. Stigmatizing language and self-stigmatizing language were both less likely in OTPs than OBOTs even after accounting for patient and substance use characteristics.

These findings point toward preconceived notions that older individuals and those with greater financial resources should be more knowledgeable about the harms associated with opioid use. They also suggest that predetermined factors making up a patient’s identity (e.g., age) influence stigmatizing language. Less stigmatizing language in the medical records of those with a high school or equivalent education is consistent with previous work (Kim et al., 2023). The finding that multiracial patients were less likely to have stigmatizing language in their charts may be surprising at first glance, but this is contextualized in the makeup of the state and communities from which these patients were drawn. Texas is a minority-majority state where the Hispanic/Latino and multiracial population is greater than the White population.

Although some have found that methadone is more stigmatized than other medications for OUD (Carl et al., 2023; Randall-Kosich et al., 2020), we found less stigmatizing language in OTPs, the only outpatient treatment setting that dispenses methadone, compared to OBOTs. This may be influenced by the unique makeup of the patients our clinics served (e.g., financially disadvantaged) in conjunction with the treatment setting culture. OTPs are heavily regulated, specialized treatment programs, while OBOTs are staffed by healthcare providers who are not required to go through specialty training to prescribe buprenorphine or naltrexone (Saloner et al., 2023). The training required to practice in OTPs may foster a more recovery-oriented culture and lead to lower levels of documented stigmatizing language. This culture may similarly influence a positive perception in the patients and contribute to the reduced self-stigma we detected in this clinical population.

Despite efforts to reduce it, stigmatizing language remains prevalent in MOUD treatment, and its impact can be profound. Using this language may directly affect a patient’s experience in treatment, both how clinicians perceive their patients and the treatment that they provide (Ashford et al., 2019). Individuals with OUD who express greater perceived stigma are more likely to experience lower quality of life (Singh et al., 2018). Because stigma is a product of societal norms, changes at this level are crucial to enhance quality of life for these patients (Cheetham et al., 2022). Fostering an environment among the healthcare providers who treat patients with OUD that ensures harmful attitudes and beliefs are not perpetuated is essential to optimize treatment.

Standardized training in stigma and its effect on treatment is a key and feasible first step. Alternatively, it is possible that interventions targeting the introduction of preferred, recovery-oriented language versus the removal of stigmatizing language may be an effective approach. Achieving stigma reduction and ensuring proper treatment will ultimately require educating the public at large in addition to healthcare providers.

Future research could create or integrate a more expansive lexicon of stigmatizing terms. Although we were not able to examine provider-level differences, future analyses might account for provider type, licensure level, and years of experience. This could inform targeted training interventions, which would allow us to explore whether language shifts after staff training initiatives on stigma reduction. Using natural language processing to link stigmatizing language to specific treatment outcomes (e.g., retention, continued opioid use, overdose, patient satisfaction) is the logical next step.

4.1. Limitations

This study has limitations to note. Data were from publicly funded OTPs and OBOTs in Texas, and this context may impact the counts of stigmatizing language attributed to clinicians and patients. All patients were below the poverty line and underinsured or uninsured whereas private or commercially insured populations might exhibit different language patterns. There were different types of providers entering notes in the medical records. The analysis included words/phrases from NIDA’s “Words Matter” resource and did not comprehensively capture all potentially stigmatizing language. Future studies could use annotation or text segmentation approaches to further differentiate patient versus clinician language. In addition, we used natural language processing to identify stigmatizing and self-stigmatizing language in medical records. These practices are exploratory and deserve careful evaluation, refinement, and replication across various settings.

4.2. Conclusions

Although efforts have been made to reduce stigma, there remains work to be done. This study reinforces the utility of natural language processing in healthcare and provides a model that can be harnessed as a foundation for future research. With this, we have set the stage for studies that can examine whether stigmatizing or self-stigmatizing language mediates the associations between individual or treatment-related factors and MOUD outcomes. Eliminating the normative use of stigmatizing language is a critical component of our collective mission to optimize treatment engagement and success.

Figure 2.

Word cloud displaying self-stigmatizing language used by patients

Highlights.

• Stigmatizing language is common in medical documentation

• Stigmatizing and self-stigmatizing language is found in opioid use treatment

• Patient characteristics affect language used in opioid use treatment

• Opioid treatment setting influences language used in medical documentation