Factors associated with healthcare professionals’ use of stigmatizing language in individuals with diabetes in pregnancy

Abstract

Background

Clinicians’ language can influence patient perception and engagement in care. Use of stigmatizing language, including terms such as “noncompliant”, “refused”, “failed”, and “unmotivated”, has been found to contribute to distress among individuals with diabetes in pregnancy. However, it is unclear how often healthcare professionals use stigmatizing language in the electronic health record (EHR) and the clinical factors associated with its use among birthing individuals with diabetes.

Methods

A retrospective cohort study was conducted from 2018 to 2019, focusing on individuals with singleton, non-anomalous pregnancies diagnosed with pre-existing or gestational diabetes. We utilized an academic-community obstetric outcomes database, employing a Natural Language Processing (NLP) algorithm to identify stigmatizing terms in electronic health records, including variations of terms such as “failed,” “refused,” “unwilling,” “unmotivated,” “noncompliant,” and “non-adherent.” We then compared the clinical characteristics, outcomes, and glycemic markers of individuals with and without documented stigmatizing language. Unadjusted and adjusted logistic regression models assessed the association of clinical factors on the utilization of stigmatizing language.

Results

Out of 1,433 birthing individuals who met criteria, 128 (8.9%) exhibited stigmatizing language in their records. The most common terms were related to “noncompliant” (47%) and “refused” (42%), primarily used by physicians and nurses. Birthing individuals of color were about four times more likely to have stigmatizing language used in their EHR compared to White individuals (aOR 3.87, 95% CI 2.55–5.99, p < 0.001). Those speaking languages other than English and Spanish were three times more likely to have stigmatizing terms documented (aOR 3.18, 95% CI 1.98–5.01, p < 0.001). Finally, individuals using public insurance were nearly five times more likely to receive stigmatizing descriptions compared to those with private insurance (aOR 4.95, 95% CI 3.26–7.73, p < 0.001). No differences were seen in NICU admission after controlling for diabetes type, and few objective markers of glycemia were identified in the EHR.

Conclusion

Birthing individuals of color, non-English/non-Spanish speakers, and those on public insurance diagnosed with diabetes in pregnancy were significantly more likely to have stigmatizing language documented within their medical records. Clinicians’ use of stigmatizing language appears disconnected from glycemic concerns.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-026-08891-y.

Introduction

Language, both written and verbal, is how healthcare professionals can relay pertinent information and priorities to patients and their subsequent care teams. However, the specific terms and phrases used can convey negative meaning, potentially impacting an individual’s healthcare experiences and outcomes [1]. Stigmatizing language – terms that imply judgment or blame – can perpetuate socially constructed power dynamics, resulting in biased care [2]. These terms have been linked to increased psychological distress, diminished self-care, and adverse clinical outcomes [3, 4]. Furthermore, stigmatizing language has historically been linked to certain conditions, particularly diabetes, and has been known to be directed more toward black and brown individuals [4–6].

Given that health care professionals’ use of language can impact how individuals with diabetes engage with care, multiples organizations have published consensus statements encouraging use of strength-based and person-first language [3, 7–9]. These documents recommend against using stigmatizing terminology including terms related to “compliance”, “adherence”, “refusal,” and “failure”. Despite this, these terms continue to be used in clinical practice, potentially impacting how individuals with diabetes engage with care and leading to worse outcomes.

While several studies have examined the use of stigmatizing language for non-pregnant individuals with co-existing diabetes, [1, 4, 5] it is not known how often this language is used within birthing populations. Gestational and pregestational diabetes pose significant risks to maternal and neonatal well-being [10, 11]. The prevalence of diabetes, while increasing across all racial and ethnic groups in the United States, has risen more profoundly within minoritized communities [12]. Using Natural Language Processing (NLP), we examined the factors associated with stigmatizing language and the association with maternal and neonatal health outcomes. We hypothesized that stigmatizing language would be more prevalent in the medical records of marginalized birthing individuals and that it would be associated with increased rates of NICU admission.

Methods

Study setting

We completed a retrospective cohort analysis of individuals with perinatal diabetes who gave birth between January 2018 and December 2019 within a six-hospital community-academic health system. This time frame aligns with changes in American College of Obstetricians and Gynecologists (ACOG) practice guidelines, [10, 11] publication of consensus statements on use of language in diabetes, [3, 7] and precedes the COVID-19 pandemic. Individuals with diabetes in pregnancy were identified using ICD-10 codes within the University of Minnesota Obstetrics Measures (UMOMs) Database, a clinical outcomes database containing over 100,000 births by over 70,000 pregnant individuals from 2011 to 2022 [13]. This database links to information included within the Electronic Health Record (EHR) and provides clinician notes, flow sheets, and scanned prenatal records from outpatient offices. Individuals without diabetes, with multifetal gestation, with fetal anomalies, or with chromosomal abnormalities were excluded from the analysis. This study received IRB approval (STUDY00017388). The IRB waived need for consent.

We identified individuals with preexisting and gestational diabetes using ICD-10 codes 42 weeks prior to birth and extending to 10 weeks post-partum. Manual verification of diabetes type was then done by reviewing diagnoses indicated in documentation during admission for birth. Preexisting diabetes (Type 1 and Type 2 Diabetes) was categorized as diabetes diagnosed before pregnancy or by standard criteria before 15 weeks of gestation [10]. In contrast, gestational diabetes was diagnosed using a two-step process in alignment with the ACOG recommendations [11]. Individuals who did not require pharmacologic therapies (i.e., insulin, metformin, or glyburide) for glucose management were categorized as having diet-controlled gestational diabetes (GDMA1). Those who did require pharmacologic therapies were classified with medication-managed gestational diabetes (GDMA2). Manual verification of diabetes type was done by reviewing diagnoses indicated in documentation during admission for birth.

Identifying “Stigmatizing Language” cohorts

Using Natural Language Processing-Patient Information Extraction for Research (NLP-PIER), a secure system for searching, processing, and indexing clinical notes in the EHR at the University of Minnesota, we identified clinical notes containing pre-specified terms suggestive of stigmatizing language [14]. NLP describes a set of techniques used to convert written passages of texts into datasets that can be analyzed by statistical and machine learning models [15]. For this study, pre-specified terms were selected based on terminology with potential negative connotations, as determined by the American Diabetes Association [3]. Specifically, we identified the terms “fail”, “refuse”, “unwilling”, “unmotivated”, “noncompliant”, and “non-adherent”, along with variations and synonyms, including, but not limited to, “failure”, “failed”, “refusal”, “noncompliance”, and “non-adherence”.

Once identified by the NLP algorithm, pre-specified terms alongside the fifty preceding and following characters were extracted (henceforth termed “snippets”). These snippets, patient identifiers, and note details (i.e., provider type and date of entry) were compiled into a separate database. Once processed, two investigators (S.W. and A.G.) reviewed the remaining snippets to define further the context of the language used. For the term “failed,” several variations were identified, including phrases such as “pump failure,” which describe mechanical errors with insulin pumps. These were excluded from the analysis as they were found to be descriptors of an individual’s medical device rather than the individual. Additionally, several variations for “failed” were identified in the context of “failed glucose tolerance tests.” These were ultimately excluded from the final analysis because they represented common language typically used to describe an abnormal test and were not thought to reflect a clinician’s assessment of the individual’s approach to diabetes management.

The final groups compared included those “with stigmatizing language” and those “without stigmatizing language” based on whether specific stigmatizing terms or their variants were identified within the EHR. Healthcare professionals using the language included physicians, mid-level professionals (such as certified nurse midwives, physician assistants, and advanced practice providers), diabetes educators, and others. Physicians included licensed practitioners with Doctor of Medicine (M.D.) and Doctor of Osteopathic Medicine (D.O.) degrees. Mid-level healthcare professionals included healthcare workers with post-secondary school training who could perform some tasks similar to those of physicians, such as certified midwives, nurse practitioners, and physician assistants. Within this healthcare system, diabetes educators are registered dietitians and nurses with advanced education in diabetes management, serving as the primary point of contact for initiating diabetes management in pregnant individuals.

Clinical characteristics and clinical outcomes

Race, language, and insurance status were data factors provided by individuals during their most recent healthcare encounter. Race was subsequently categorized as White, Black or African American, Asian, American Indian (AI)/Alaska Native (AN), Native Hawaiian (NH)/Other Pacific Islander (PI), or Multiple Races. Languages were classified as English, Spanish, or other. Other languages included Arabic, Chinese, French, Hindi, Hmong, Russian, and Vietnamese. Insurance status was categorized as public (provided by state or federal government agencies), private (provided by private entities), or uninsured. If no race or language was identified, the metrics for these individuals were categorized as missing. If no insurance status was identified, the individual was classified as uninsured. Parity was based on the obstetric history provided during the initial prenatal visit.

The primary outcomes were NICU admission and NICU length of stay (NICU LOS). These primary outcomes were chosen due to their association with multiple complications that may arise in pregnancies affected by diabetes including prematurity, hypoglycemia, respiratory distress, and shoulder dystocia. NICU admission was defined as any instance where an infant was admitted to the neonatal ICU in the immediate postpartum period after birth. Secondary outcomes included gestational age (GA) at birth, mode of birth, shoulder dystocia, neonatal birth weight, neonatal hypoglycemia, and respiratory distress syndrome (RDS). Neonatal weight was defined as birth weight measured in grams. Large for gestational age (LGA) status was defined as birth weight > 90th percentile for gestational age at birth, and small for gestational age (SGA) status was defined as birth weight < 10th percentile [16]. Neonatal hypoglycemia was defined as any glucose levels < 40 mg/dL over 48 h of life, as documented in the neonate’s EHR, consistent with our institution’s clinical treatment guidelines. RDS was determined based on ICD-10 codes identified in the neonate’s EHR. The mode of birth was delineated as vaginal, operative vaginal, or cesarean birth. Shoulder dystocia was defined by failure to deliver the anterior shoulder after delivery of the fetal head.

Recorded glucose levels two weeks before birth were manually identified in clinical notes within the EHR, including fasting, 1-hour postprandial, and 2-hour postprandial values. The percentage of abnormal glucose levels was quantified based on ACOG recommended targets (Fasting < 95 mg/dL, 1 h postprandial < 140 mg/dL, and 2 h postprandial < 120 mg/dL).^{10 11} Based on institutional policies for initiating or titrating insulin during pregnancy, we classified individuals with at least 50% of their glucose levels elevated above the normal range two weeks before birth as having abnormal glucose levels. Cases in which glucose ranges were reported in the EHR, but the percentage of abnormal values could not be precisely calculated, were reported as “not able to be assessed.” Hemoglobin A1c (A1c) was identified from the clinical record if obtained within three months prior to birth.

Statistical analysis

Student’s t-tests and Wilcoxon rank-sum tests were used to investigate the association between stigmatizing language and continuous variables. Chi-square Fisher’s exact tests were used for categorical variables. Unadjusted and adjusted logistic regression models were used to examine the effect of various factors on the use of stigmatizing language. Odds ratios (OR) and 95% confidence intervals (CI) were obtained. All reported p-values are two-sided, and a significance level of 0.05 was used. Statistical analyses were performed using R (version 4.1.2, R Core Team).

Results

A total of 1,599 births among individuals with diabetes were identified between 2018 and 2019. After exclusion of fetal and chromosomal abnormalities (N = 43) and those without a diagnosis of diabetes (N = 123), 1,433 patients met the final inclusion criteria. Of these, 128 (8.9%) individuals had stigmatizing language documented in their EHR, whereas 1305 (91.1%) did not have such language (Fig. 1). Upon analyzing clinical notes, the most frequently used terms were “noncompliant” (47.1%) and “refused” (42.1%) (Fig. 2A). Stigmatizing language was found most frequently in notes written by physicians (54.1%) and nurses (22.4%) (Fig. 2B), which was overrepresented relative to the total number of notes written by these clinicians.

Fig. 1.

Study Flow diagram of Individuals

Fig. 2.

Summary of Note Terminology. A Table depicting identification of clinical notes with stigmatizing language. B Distribution of notes by clinical role

When analyzing cohort characteristics, birthing individuals of color, non-English/Spanish speaking individuals, those on public insurance, and those with preexisting diabetes and using oral medications more often had stigmatizing language documented within their medical records (Table 1). When adjusting for diabetes type, birthing individuals of color were approximately four times more likely to have stigmatizing language used in the EHR than White individuals (aOR 3.87, 95% CI 2.55–5.99, p < 0.001; Table 2). Individuals whose primary language was not English or not Spanish were approximately three times more likely to have stigmatizing language than English-speaking individuals (aOR 3.18, 95% CI 1.98–5.01, p < 0.001; Table 2). Finally, those using public insurance were approximately five times more likely to be described using stigmatizing language than those with private insurance (aOR 4.95, 95% CI 3.26–7.73, p < 0.001; Table 2).

Table 1.

Comparison of demographics and baseline characteristics by stigmatizing language

Variable	Individuals with stigmatizing language (N = 128)	Individuals without stigmatizing language (N = 1305)	P-value1
Age, mean (SD)	33.7 (6.6)	33.0 (4.8)	0.115
Race2, n (%)			<0.001
White	39 (4.8%)	766 (95.2%)
Black	59 (25.3%)	174 (74.7%)
Asian	13 (5.4%)	226 (94.6%)
AI or AN	4 (16.0%)	21 (84.0%)
NH or PI	0 (0.0%)	4 (100.0%)
More than 1 race	3 (12.0%)	22 (88.0%)
Language, n (%)			<0.001
English	94 (7.8%)	1115 (92.2%)
Spanish	3 (4.4%)	65 (95.6%)
Other	31 (19.9%)	125 (80.1%)
Insurance type, n (%)			<0.001
Uninsured	5 (7.9%)	58 (92.1%)
Public	92 (16.4%)	468 (83.6%)
Private	31 (3.8%)	779 (96.2%)
Parity, n (%)			<0.001
Nulliparous	28 (5.4%)	495 (94.6%)
Multiparous	100 (11.0%)	810 (89.0%)
DM type, n (%)			<0.001
T1DM	10 (16.1%)	52 (83.9%)
T2DM	33 (22.0%)	117 (78.0%)
GDMA1	41 (6.3%)	609 (93.7%)
GDMA2	44 (7.7%)	527 (92.3%)
DM medication, n (%)			<0.001
None	46 (6.7%)	636 (93.3%)
Insulin	67 (10.0%)	604 (90.0%)
Oral medication	15 (18.8%)	65 (81.2%)

¹Student’s t-test were used for continuous variables. Chi-square or Fisher’s exact tests were used for categorical variables

²Race is missing for 10 individuals with stigmatizing language and for 92 individuals without stigmatizing language

Table 2.

Comparison of stigmatizing language by demographics

Variable	Unadjusted1		Adjusted2
	OR (95% CI)	P-value1	OR (95% CI)	P-value1
Race3
White	Reference	Reference	Reference	Reference
Persons of Color	3.47 (2.34, 5.23)	<0.001	3.87 (2.55, 5.99)	<0.001
Language
English	Reference	Reference	Reference	Reference
Spanish	0.55 (0.13, 1.51)	0.315	0.52 (0.12, 1.45)	0.276
Other	2.94 (1.86, 4.55)	<0.001	3.18 (1.98, 5.01)	<0.001
Insurance type
Uninsured	2.17 (0.72, 5.33)	0.123	2.14 (0.7, 5.34)	0.135
Public	4.94 (3.27, 7.65)	<0.001	4.95 (3.26, 7.73)	<0.001
Private	Reference	Reference	Reference	Reference

¹Unadjusted logistic regression models were used to investigate the effect of demographics on stigmatizing language

²Adjusted logistic regression models were used, adjusting for DM type

³Race is missing for 10 individuals with stigmatizing language and for 92 individuals without stigmatizing language

When assessing clinical outcomes, no differences were found in the mode of birth. However, individuals with stigmatizing language documented in the medical record gave birth earlier than their counterparts (median 38 weeks 0 days vs. 39 weeks 0 days, p < 0.001; Table 3). While there was no difference in median birth weight, individuals with stigmatizing language more often had a birth complicated by shoulder dystocia (p = 0.007) and large for gestational age neonates (p = 0.027; Table 3). Additionally, their infants were admitted to the NICU and experienced RDS more often (p = 0.047; Table 3), though no differences were found in rates of neonatal hypoglycemia. When adjusting for diabetes type, no significant association was seen between stigmatizing language and NICU admission, LGA neonate, or RDS. However, the risk of shoulder dystocia was approximately 4 times higher in individuals described with stigmatizing language (aOR 4.65, 95% CI 1.72–11.32, p = 0.001; Table 4).

Table 3.

Comparison of birth and neonatal outcomes by stigmatizing language

Variable	Individuals with stigmatizing language (N = 128)	Individuals without stigmatizing language (N = 1305)	P-value1
Mode of birth, n (%)			0.265
Vaginal	63 (49.2%)	687 (52.6%)
C-section	64 (50.0%)	581 (44.5%)
Assisted	1 (0.8%)	37 (2.8%)
Gestational age at birth, median (Q1, Q3)	38.0 (37.0, 39.0)	39.0 (37.6, 39.4)	<0.001
Birth weight2 (grams), median (Q1, Q3)	3288.6 (2945.0, 3765.0)	3339.9 (2980.2, 3657.2)	0.625
Birth weight categories2, n (%)			0.027
SGA	6 (4.9%)	73 (5.7%)
AGA	89 (72.4%)	1030 (80.5%)
LGA	28 (22.8%)	177 (13.8%)
Shoulder dystocia, n (%)	7 (5.5%)	20 (1.5%)	0.007
NICU admission, n (%)	35 (27.3%)	203 (15.6%)	<0.001
NICU LOS (days), median (Q1, Q3)	9.6 (5.2, 23.7)	10.0 (2.8, 18.9)	0.364
Neonatal hypoglycemia, n (%)	62 (48.4%)	555 (42.5%)	0.198
Respiratory Distress Syndrome, n (%)	22 (17.2%)	147 (11.3%)	0.047

¹Wilcoxon rank-sum tests were used for continuous variables. Chi-square or Fisher’s exact tests were used for categorical variables

²Birth weight is missing for five individuals with stigmatizing language and for 24 individuals without stigmatizing language. One additional individual without stigmatizing language was missing the birth weight category due to having a gestational age in the 20th week

Table 4.

Comparison of neonatal outcomes by stigmatizing language

Variable	Unadjusted1		Adjusted2
	OR (95% CI)	P-value	OR (95% CI)	P-value
NICU admission	2.04 (1.33, 3.07)	<0.001	1.31 (0.76, 2.19)	0.316
LGA	1.84 (1.15, 2.85)	0.008	1.58 (0.97, 2.49)	0.057
Shoulder dystocia	3.72 (1.43, 8.58)	0.003	4.65 (1.72, 11.32)	0.001
Respiratory Distress Syndrome	1.63 (0.98, 2.62)	0.049	1.06 (0.56, 1.9)	0.846

¹Unadjusted logistic regression models were used to investigate the effect of stigmatizing language on outcomes

²Adjusted logistic regression models were used, adjusting for DM type and GA at birth

Since clinicians may use stigmatizing terminology to describe concerns about suboptimal diabetes management, we assessed hemoglobin A1c levels and recorded glucose readings. Only 28.9% of individuals with stigmatizing language and 13.0% of individuals without stigmatizing language had A1c assessed within three months of birth, and the median was below 6.0% for both groups (Supplemental Table 1). While only a third of patients in each cohort had any documented glucose in the two weeks prior to birth, elevated glucose values were found more often in those without stigmatizing language (Supplemental Table 1).

Discussion

In this study, we find that 9% of individuals with diabetes in pregnancy have clinical documentation that contains stigmatizing terminology. The most frequent terms include those related to “refusal” and “noncompliance” and these terms were most used by physicians and nurses. Stigmatizing language was used more frequently in birthing individuals of color, those speaking languages other than English and Spanish, and those using public insurance. No differences were noted in rates of NICU admission or length of stay after adjusting for diabetes type, though more individuals with stigmatizing language experienced shoulder dystocia at the time of birth. Notably, very few objective markers of glucose management were identified in the medical records, although the median hemoglobin A1c was in the normal range for both groups, and those with stigmatizing language had more glucose values within the goal range. This suggests that the use of stigmatizing language may not be systematically linked with clinician concern for hyperglycemia. This is similar to recent work showing that clinicians often describe individuals as having “poor glucose control” without objective data [17].

Language is a powerful means of transmitting critical information in the clinical setting. Historically, patient access to clinical records was limited; however, passage of the 21st Century Cures Act [18] in the US ensured an individual’s right to EHR access potentially amplifying the impact of stigmatizing language in clinical documentation. The terms we use can carry implicit or explicit biases, which can potentially impact health professionals’ perceptions and patient care [2, 5]. Prior reports identify stigmatizing language more often in the EHR among individuals of color and potentially represent a source of bias in these populations [6, 19]. Individuals with diabetes describe the use of negative and stigmatizing words by health care professionals as having a negative impact on their experience with diabetes [20]. Recent qualitative work in individuals with gestational and preexisting diabetes identifies negative and stigmatizing provider interactions as a significant source of diagnosis-related distress [21, 22]. This highlights that the language healthcare professionals use can dramatically impact the patient experience and contribute to health outcomes [1].

We identified that stigmatizing language was used most often when caring for individuals from marginalized populations, without clear data suggesting these individuals have worse markers of glycemia. Within birthing populations, systemic barriers to care, particularly for low-income communities or individuals of color, can make it difficult for patients to attend appointments, access or use insulin, or adequately monitor their blood glucose levels [23–25]. What may be seen as ‘noncompliance’ by clinicians may actually be a byproduct of a system that fails to provide the necessary support for individuals to fully participate in their care. This context may be completely erased when utilizing these terms. While this may not have a measurable impact on all perinatal health outcomes in this study, stigmatizing language could potentially impact how individuals interact with the health system and patient-clinician relationships, which could, in turn, affect perinatal health outcomes.

Organizations such as the American Diabetes Association have provided guidance to limit the use of stigmatizing language in diabetes care [7, 8, 26]. For example, instead of stating an individual is “noncompliant”, clinicians can note the frequency with which individuals take their medications and potential barriers they experience. Adopting these language recommendations has the potential to improve the experience of individuals managing diabetes in pregnancy and contribute to more equitable healthcare delivery.

Several strengths of this study are noted. Use of NLP algorithms and manual verification of clinical contexts allows our team to systematically identify the use of stigmatizing language and its variants in EHR documentation throughout pregnancy. Additionally, we examined this language within a large community-academic health system with over 10,000 births per year across six hospitals in urban, suburban, and rural settings, providing care from multiple provider types, which enhances generalizability.

A limitation of this study is that language was only identified if it was found within the clinical notes of the EHR; therefore, we may have missed some individuals if the language was used only in scanned outpatient documents. Given this limitation, we are not able to provide details of stigmatizing language use in specific subgroups. Additionally, in establishing our final cohorts, we did not include those who only had terminology referencing “failed glucose tolerance tests.” Of note, 442 of 1433 (30.8%) in the total cohort had “failed glucose test” language in their clinical notes, highlighting the frequency with which this terminology is used. This terminology was felt to characterize test results, not an individual-specific characteristic; thus, it was not used in establishing final cohorts. However, it does highlight that the frequency with which individuals may see this stigmatizing language in the medical record is higher than the 9% we report.

Conclusion

Our study finds that healthcare professionals use stigmatizing language more often in birthing individuals of color, non-English/non-Spanish speakers, and those of public insurance. Limited objective data supported differences in glucose management in these populations. This highlights opportunities for health care professionals and health care systems to reassess the use of language in birthing populations.

Authors’ contributions

A.G. and S.A.W. were involved in the conception, design, and conduct of the study and the analysis and interpretation of the results. I.B.O. assisted with NLP analysis. A.G., R.M.M., A.Y.L., S.M., J.M., and O.O.A., verified and collected data from the EHR. K.T. completed all statistical analyses. S.K.N., A.A.B. and J.K. were involved with study design and data interpretation. A.G. wrote the first draft of the manuscript, and all authors edited, reviewed, and approved the final version of the manuscript. S.A.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Funding

Research reported in this publication was supported by NIH grant P30 CA77598 utilizing the Biostatistics Core shared resource of the Masonic Cancer Center, University of Minnesota, and by the National Center for Advancing Translational Sciences of the National Institutes of Health Award Number UM1TR004405. The content is solely the authors’ responsibility and does not necessarily represent the official views of the National Institutes of Health.

This work was presented in the form of two abstracts at the Society for Maternal Fetal Medicine 2025 Pregnancy Meeting in Denver, Colorado on January 30, 2025.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Study was reviewed by University of Minnesota IRB (STUDY00017388, Wernimont PI). The need for consent was waived by the Institutional Review Board (IRB).

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.