Frequency of and Factors Associated With Nonmedical Opioid Use Behavior Among Patients With Cancer Receiving Opioids for Cancer Pain

Key Points

Question

What are the predictors for and frequency of the presence of nonmedical opioid use (NMOU) behavior in patients with cancer?

Findings

In this prognostic study of 1554 patients with cancer who were taking opioids for pain, 299 patients (19%) developed NMOU behavior. Single or divorced status, higher score on the Screener and Opioid Assessment for Patients with Pain tool, higher levels of pain severity, and daily opioid dose were independently associated with NMOU behavior.

Meaning

In the era of the opioid crisis, this information may assist clinicians in determining which patients with cancer may be at a higher risk of developing NMOU behavior; these patients will require a much closer follow-up for optimal opioid use in cancer pain management.

This prognostic study uses data from patients at the MD Anderson Cancer Center to assess predictors of nonmedical opioid use behaviors among patients with cancer receiving pain management treatment.

Abstract

Importance

One of the main aims of research on nonmedical opioid use (NMOU) is to reduce the frequency of NMOU behaviors through interventions such as universal screening, reduced opioid exposure, and more intense follow-up of patients with elevated risk. The absence of data on the frequency of NMOU behavior is the major barrier to conducting research on NMOU.

Objective

To determine the overall frequency of and the independent predictors for NMOU behavior.

Design, Setting, and Participants

In this prognostic study, 3615 patients with cancer were referred to the supportive care center at MD Anderson Cancer Center from March 18, 2016, to June 6, 2018. Patients were eligible for inclusion if they had cancer and were taking opioids for cancer pain for at least 1 week. Patients were excluded if they had no follow-up within 3 months of initial consultation, did not complete the appropriate questionnaire, or did not have scheduled opioid treatments. After exclusion, a total of 1554 consecutive patients were assessed for NMOU behavior using established diagnostic criteria. All patients were assessed using the Edmonton Symptom Assessment Scale, the Screener and Opioid Assessment for Patients with Pain (SOAPP), and the Cut Down, Annoyed, Guilty, Eye Opener–Adapted to Include Drugs (CAGE-AID) survey. Data were analyzed from January 6 to September 25, 2020.

Results

A total of 1554 patients (median [interquartile range (IQR)] age, 61 [IQR, 52-69] years; 816 women [52.5%]; 1124 White patients [72.3%]) were evaluable for the study, and 299 patients (19.2%) had 1 or more NMOU behaviors. The median (IQR) number of NMOU behaviors per patient was 1 (IQR, 1-3). A total of 576 of 745 NMOU behaviors (77%) occurred by the first 2 follow-up visits. The most frequent NMOU behavior was unscheduled clinic visits for inappropriate refills (218 of 745 [29%]). Eighty-eight of 299 patients (29.4%) scored 7 or higher on SOAPP, and 48 (16.6%) scored at least 2 out of 4 points on the CAGE-AID survey. Results from the multivariate model suggest that marital status (single, hazard ratio [HR], 1.58; 95% CI, 1.15-2.18; P = .005; divorced, HR, 1.43; 95% CI, 1.01-2.03; P = .04), SOAPP score (positive vs negative, HR, 1.35; 95% CI, 1.04-1.74; P = .02), morphine equivalent daily dose (MEDD) (HR, 1.003; 95% CI, 1.002-1.004; P < .001), and Edmonton Symptom Assessment Scale pain level (HR, 1.11; 95% CI, 1.06-1.16; P < .001) were independently associated with the presence of NMOU behavior. In recursive partition analysis, single marital status, MEDD greater than 50 mg, and SOAPP scores greater than 7 were associated with a higher risk (56%) for the presence of NMOU behavior.

Conclusions and Relevance

This prognostic study of patients with cancer taking opioids for cancer pain found that 19% of patients developed NMOU behavior within a median duration of 8 weeks after initial supportive care clinic consultation. Marital status (single or divorced), SOAPP score greater than 7, higher levels of pain severity, and MEDD level were independently associated with NMOU behavior. This information will assist clinicians and investigators designing clinical and research programs in this important field.

Introduction

Cancer pain is one of the most frequent and debilitating symptoms in patients with cancer.^1,2,3,4,5 Opioids are the mainstay for cancer pain management. Currently, the use of opioids for cancer pain is suboptimally managed despite their need. Various contributing factors play a role in suboptimal use of opioids; these include limited education, variable access due to opioid shortages, and the need for opioid preauthorization, which is the result of the ongoing opioid epidemic in the US. The US government enacted restrictive legislation on opioid access, which indirectly affected clinical cancer pain management.^1,2,3,4,5 Nonmedical opioid use (NMOU), ie, the use of opioids without a prescription or in ways other than medically prescribed, is common and may significantly affect cancer pain management. Nonmedical opioid use is trending to be one of the main concerns in pain management in the opioid crisis in patients both with and without cancer.^{6,7,8,9,10,11,12,13,14} In patients without cancer, NMOU is associated with negative health outcomes and overdose deaths.^15,16,17 However, in oncology, the best practices for identifying and optimally managing NMOU are not yet clearly defined.^12,18,19 Prior studies by our team and others were also able to establish the risk factors for NMOU in patients with cancer, but not all the patients with risk factors will be diagnosed with having NMOU behaviors, and some patients with no previous risk factors will engage in NMOU behaviors.^20,21 Nonmedical opioid use behaviors may have long-term, devastating consequences for patients with cancer, including increased morbidity and mortality due to the possible development of substance use disorder, poor management of their cancer from lack of preventive and adherence practices to cancer treatments, and from the comorbidities (ie, mood disorders, insomnia) that accompany this condition.^22,23,24 Hence, there is a need to develop optimal guidelines to screen and manage NMOU in patients with cancer.^25,26 Nonmedical opioid use manifests in a range of behaviors, including early refill requests (self-escalation of opioids), resistance to prescription changes, opioid sharing between friends and family, requests for specific drugs, obtaining opioids from multiple physicians, using nonprescribed drugs, using opioids for nonprescribed indications (eg, anxiety, insomnia), family members expressing concern over the patient’s opioid use, discrepancy in pill counts, concurrent abuse of illicit drugs, impaired functioning per family or staff, prescription forgery, “lost or stolen” medications, and the selling or stealing of prescription drugs.¹⁹ One of the main aims of research on NMOU is to reduce the frequency of NMOU behaviors with interventions such as universal screening, reduced opioid exposure, and more intense follow-up of patients with elevated risk. Unfortunately, the absence of data on the frequency of NMOU behavior is the major barrier to conducting research on interventions to reduce NMOU. Additionally, there are no studies, to our knowledge, evaluating the independent predictors for NMOU behavior. Therefore, the aims of this study were to determine the overall frequency of NMOU behavior and to examine the independent predictors for NMOU behavior.

Methods

The institutional review board of the University of Texas MD Anderson Cancer Center approved this prognostic study and waived the requirement for informed consent from all the patients because this study involved the retrospective evaluation of previously stored patient information. This study followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline.

Patients were included in the study if they met the following eligibility criteria: (1) they had a diagnosis of cancer and had their initial outpatient supportive care consultations between February 12, 2016, and July 15, 2018; (2) they were taking prescribed opioids and completed the Screener and Opioid Assessment for Patients with Pain (SOAPP) assessment during their initial outpatient supportive care consultation; and (3) they had at least 1 outpatient supportive care follow-up visit within 3 months of the initial outpatient supportive care consultation. Patients were excluded if they had no follow-up within 3 months of initial consultation, did not complete the appropriate questionnaire, or did not have scheduled opioid treatments.

We reviewed the patients’ characteristics, including, age, sex, race/ethnicity, NMOU behavior (eBox in the Supplement), Eastern Cooperative Oncology Group performance status, and the scores of the tools that are routinely used in clinical care at the supportive care clinic of the University of Texas MD Anderson Cancer Center. These tools include the Edmonton Symptom Assessment Scale (ESAS), SOAPP, and the Cut Down, Annoyed, Guilty, Eye Opener–Adapted to Include Drugs (CAGE-AID) survey.

At the supportive care clinic, it is a standard procedure to conduct universal screening for higher risk factors for NMOU and NMOU behaviors during every encounter. All faculty and nurses on the team are fully trained and have been implementing this practice for more than 5 years.^12,27 In this study, we reviewed the routine documentation of each supportive care clinic encounter made by the physician and nurse for the presence of any of 14 observed behaviors of patients with cancer that might suggest NMOU (eBox in the Supplement).^{19,28,29,30,31} The total score of the 14 items of the NMOU assessment was summed on a numerical scale ranging from 0 to 14.

The Eastern Cooperative Oncology Group scale was used to assess patients’ performance status and assess how the disease affects the patients’ activities of daily living.³² This is accomplished using a 5-point scale (0 = fully active without restriction to 5 = dead).

The ESAS is a validated symptom assessment tool used to assess the severity of cancer-related symptoms on a 0 to 10 numerical scale (0, no symptoms; 10, worst possible symptoms). In addition, the ESAS is one of the most common symptom assessment tools used in patients with cancer.³³ Prior studies found that the ESAS had a strong correlation with other symptom assessment tools, such as the Hospital Anxiety and Depression Scale, and the Patient Health Questionnaire-9.^34,35,36 Pain, fatigue, nausea, depression, anxiety, drowsiness, shortness of breath, appetite, feelings of well-being, sleep, financial distress, and spiritual pain are assessed using this tool.³⁷

The SOAPP-14 is a validated tool used to assess the risk for inappropriate opioid use or NMOU. It consists of 14 items regarding antisocial behavior, substance abuse history, doctor-patient relationship, medication-related behaviors, and psychiatric and neurobiologic need for medicine. It is a 5-point Likert scale, and the choices are 0 (never), 1 (seldom), 2 (sometimes), 3 (often), and 4 (very often).^{18,31,38,39,40,41} The possible score range is 0 to 56; a score of 7 or greater suggests elevated risk for NMOU.^21,42,43

Finally, CAGE-AID is used to assess alcoholism and illicit drug use. Patient scores from 2 to 4 are considered positive for alcoholism and indicate an increased risk for severe symptom distress and potential NMOU and chemical coping in patients with cancer.^{44,45,46,47,48,49} The CAGE-AID consists of a 4-item questionnaire.^50,51

Statistical Analysis

Data were summarized using standard descriptive statistics including means, SDs, medians, and interquartile ranges (IQRs) for continuous variables; frequency and proportion were used to describe categorical variables. The association between categorical variables was examined using the χ² test or Fisher exact test when appropriate. The Wilcoxon rank-sum test was used to examine the differences in continuous variables between the presence and absence of NMOU behavior among patients. Univariate and multivariate Cox regression models were applied to assess the effect of variables of interest on the presence of NMOU behavior during follow-up visits. Significant variables identified using the multivariate model were further analyzed by recursive partitioning analysis to identify the best probability for the presence of NMOU behaviors in specific groups of patients. Recursive partitioning analysis is a method used in nonparametric regression to classify study participants into groups and predict values of outcomes based on values of predictor factors.⁵² We used the recursive partitioning analysis to create a classification tree to illustrate a non–model-based association between the presence of NMOU behaviors and the key predictor variables, including marital status, opioid daily doses, ESAS pain, and SOAPP status. In our classification tree of the presence of NMOU behaviors, each terminal node represented a homogeneous partition of the patients according to a predictor variable that was identified using the automatic stepwise variable selection technique. Our classification tree displayed the association between the presence of NMOU behaviors and the 4 predictor factors in an intuitive manner.

As there were no studies, to our knowledge, evaluating the presence of NMOU in patients with cancer, we used a convenience sample, and missing data were handled using the complete case analysis method. P values were 2 sided, and P < .05 was considered significant. All computations were carried out in SAS, version 9.4 (SAS Institute), and analyses took place from January 6 to September 25, 2020.

Results

A total of 3588 of 3615 patients (99%) who were referred to the supportive care clinic completed the SOAPP questionnaire (Figure 1). After exclusion criteria were applied, 1554 patients (median [IQR] age, 61 [52-69] years; 816 women [52.5%]; 1124 White patients [72.3%]) were evaluable for the study (eTable in the Supplement).

Figure 1. Flowchart.

NMOU indicates nonmedical opioid use; SOAPP, Screener and Opioid Assessment for Patients with Pain.

The median number of follow-up visits was 2 (IQR, 1-3), and the median time a patient was followed up after initial supportive care center consultation was 56 days (IQR, 28-77 days). A total of 328 patients (21%) exhibited a SOAPP score of 7 or greater (suggestive of elevated risk for NMOU), and 179 patients (11.5%) had CAGE-AID scores of 2 or greater.

A total of 299 of 1554 patients (19%) had at least 1 NMOU behavior. The median number of NMOU behaviors per patient was 1 (IQR, 1-3). Most (576 of 745 [77%]) of the NMOU behavior occurred by the first 2 follow-up visits (Table 1). The most frequent NMOU behavior was an unscheduled clinic visit for inappropriate refills (218 of 745 [29%]). Eighty-eight patients 29.4%) scored 7 or higher on SOAPP, and 48 (16.6%) scored at least 2 out of 4 points on CAGE-AID (Table 2).

Table 1. Percentage of Patients With Specific Nonmedical Opioid Use (NMOU) Behavior and Frequency of NMOU Behaviors.

NMOU	Frequency, No. (%)
	Patients with NMOU behavior (n = 299)	NMOU behaviors
		Total (n = 745)	First follow-up (n = 399)	Second follow-up (n = 177)
Behavior (ranked in descending order by the frequencies of behaviors)
1. Frequent unscheduled clinic visits or phone calls for refills	147 (49.2)	218 (29.3)	133 (33.3)	44 (24.9)
2. Self-escalation of opioid dose for excessive increase in the opioid dosage not consistent with patient’s pain syndrome	50 (16.7)	112 (15.0)	54 (13.5)	27 (15.3)
8. Use of nonprescribed restricted medications or illicit drugs	38 (12.7)	94 (12.6)	39 (9.8)	23 (13.0)
10. Reports of impaired functioning in daily activities due to opioid use	45 (15.1)	83 (11.1)	47 (11.8)	24 (13.6)
7. Resistance to changes in the opioid regimen even when clinically indicated	19 (6.4)	54 (7.2)	20 (5.0)	14 (7.9)
6. Request for specific opioid	21 (7.0)	46 (6.2)	22 (5.5)	14 (7.9)
5. Seeking opioids from multiple physicians (“doctor shopping”)	30 (10.0)	43 (5.8)	31 (7.8)	6 (3.4)
3. Reports of lost or stolen opioid prescription/medication	18 (6.0)	27 (3.6)	16 (4.0)	5 (2.8)
4. Frequent emergency department visits for opioids	12 (4.0)	22 (3.0)	12 (3.0)	6 (3.4)
11. Family member expressing concern over the patient’s opioid use	11 (3.7)	21 (2.8)	11 (2.8)	7 (4.0)
9. Requesting opioids for their euphoric effects or for symptoms such as anxiety or insomnia	5 (1.7)	13 (1.7)	6 (1.5)	5 (2.8)
14. Obtaining opioids from nonmedical sources, stealing, selling	6 (2.0)	10 (1.3)	7 (1.8)	2 (1.1)
12. Reports of hoarding drugs	1 (0.3)	1 (0.1)	1 (0.3)	0 (0.0)
13. Reports of stealing or selling prescription drugs	1 (0.3)	1 (0.1)	0 (0.0)	0 (0.0)
Total	299	745 (100)	399 (54)	177 (24)

Table 2. Frequency of SOAPP and CAGE-AID in Cancer Patients With and Without the Presence of NMOU Behavior.

Assessment	NMOU behavior, No. (%)		P valuea
	Positive (n = 299)	Negative (n = 1255)
SOAPP	88 (29.4)	240 (19.1)	<.001
CAGE-AID	48 (16.1)	131 (10.41)	.006

Abbreviations: CAGE-AID, Cut Down-Annoyed-Guilty-Eye Opener–Adapted to Include Drugs Assessment; NMOU, nonmedical opioid use; SOAPP, Screener and Opioid Assessment for Patients With Pain tool.

^aχ² test or the Fisher exact test.

Results of the multivariate Cox regression model showed marital status (divorced vs married, hazard ratio [HR], 1.43; 95% CI, 1.01-2.03; P = .04; single vs married, HR, 1.58; 95% CI, 1.15-2.18; P = .005), SOAPP score (positive vs negative, HR, 1.35; 95% CI, 1.04-1.74; P = .02), daily opioid dose or morphine equivalent daily dose (MEDD) (HR, 1.003; 95% CI, 1.002-1.004; P < .001), and severity of ESAS pain (HR, 1.11; 95% CI, 1.06-1.16; P < .001) were independently associated with the presence of NMOU behavior during follow-up visits (Table 3). Figure 2 shows good prediction capability and fitness for the final multivariate Cox regression model using a time-dependent area under curve score (0.697).

Table 3. Univariate and Multivariate Analysis for the Identification of Patients at Risk for Presence of NMOU Behavior.

Covariates	Cox regression model
	Univariate		Multivariatea
	HR (95% CI)	P value	HR (95% CI)	P value
Sex
Men vs women	0.89 (0.71-1.11)	.89	NA	NA
Race
Asian vs White	1.25 (0.79-1.98)	.34	NA	NA
Black vs White	0.96 (0.67-1.39)	.85
Hispanic vs White	1.27 (0.82-1.95)	.28
Other vs White	1.88 (1.11-3.17)	.02b
Marital status
Divorced vs married	1.45 (1.03-2.05)	.03b	1.43 (1.01-2.03)	.04b
Other vs married	1.91 (0.9-4.07)	.09	2.02 (0.95-4.31)	.07
Single vs married	1.61 (1.17-2.2)	.003b	1.58 (1.15-2.18)	.01b
Widowed vs married	0.98 (0.58-1.66)	.94	1.03 (0.60-1.74)	.93
Smoking
Current vs never	2.09 (1.45-3.00)	<.001b	NA	NA
Former vs never	1.22 (0.96-1.56)	.11
ECOG Performance status
0 vs 4	0.82 (0.08-7.85)	.86	NA	NA
1 vs 4	1.55 (0.22-11.19)	.66
2 vs 4	2.05 (0.29-14.68)	.47
3 vs 4	1.99 (0.28-14.26)	.495
CAGE-AID status
Positive vs negative	1.5 (1.1-20.4)	.01b	NA	NA
SOAPP status
Positive vs negative	1.67 (1.3-2.14)	<.001b	1.35 (1.04-1.74)	.02b
Age	0.99 (0.98-1.00)	.02b	NA	NA
MEDDs, mg/d	1.00 (1.00-1.00)	<.001b	1.003 (1.002-1.004)	<.001b
ESAS
Pain	1.15 (1.1-1.2)	<.001b	1.11 (1.06-1.16)	<.001b
Fatigue	1.05 (1.01-1.1)	.02b	NA	NA
Nausea	1.02 (0.98-1.06)	.36
Depression	1.01 (0.97-1.05)	.78
Anxiety	1.03 (0.99-1.07)	.13
Drowsiness	1.05 (1.01-1.09)	.02b
Dyspnea	1.01 (0.97-1.05)	.59
Appetite	1.01 (0.98-1.05)	.48
Well-being	1.03 (0.99-1.08)	.12
Sleep	1.04 (1-1.09)	.03b
Financial distress	1.03 (1-1.07)	.06
Spiritual pain	1.03 (0.98-1.08)	.22

Abbreviations: CAGE-AID, Cut Down-Annoyed-Guilty-Eye Opener–Adapted to Include Drugs Assessment; ECOG, Eastern Cooperative Oncology Group; ESAS, Edmonton Symptom Assessment System; HR, hazard ratio; MEDDs, Morphine Equivalent Daily Doses; NMOU, nonmedical opioid use; SOAPP, Screener and Opioid Assessment for Patients with Pain tool.

^aThe time-dependent area under the curve value of the receiver operating characteristic curve of the multivariate Cox regression model was 0.697, thereby indicating a good fit.

^bThese P values indicate a statistically significant difference.

Figure 2. Time-Dependent Area Under the Curve for the Final Regression Model.

Based on 50 perturbed samples. Shading indicates 95% CIs. The solid gray line indicates the time-dependent AUC of the final regression model. AUC, indicates area under the curve.

Preliminary recursive partition analysis found that being single, having an MEDD greater than 50 mg at consultation, and having a SOAPP score of 7 or greater were associated with a higher risk (56%) for the presence of NMOU behavior (eFigure in the Supplement).

Discussion

To our knowledge, this study was the first of its kind to evaluate the frequency of NMOU behaviors in patients with cancer. Our results suggest that NMOU behaviors occur in 19% of patients with cancer who are receiving long-term opioid treatments for cancer pain at a supportive care clinic at a comprehensive cancer center. Furthermore, results suggest that the most common NMOU behavior was unscheduled clinic visits for inappropriate refills. Patients with cancer pain who were single or divorced, had a higher MEDD, or had a higher severity of cancer pain were at a higher risk for having an NMOU behavior.

In recent years, supportive and palliative care teams have been increasingly involved in taking care of patients with cancer and NMOU behaviors.¹² To our knowledge, there is limited published literature on NMOU in patients with cancer.¹⁹ Recent studies have highlighted the need for the development of optimal diagnostic and treatment strategies for management of NMOU in patients with cancer so as to improve the quality of cancer care.^{12,17,18,19,20,21,32} The results of this study are important as there are, to our knowledge, no published studies focused on the frequency of NMOU behaviors in patients with cancer.^19,28,31 Our results suggest that approximately 1 in 5 patients are at risk of NMOU behaviors. Frequency of NMOU behaviors in noncancer pain conditions varies widely based on the patient population, care setting, and criteria for diagnosis.^6,53,54,55 The frequency is higher in studies evaluating patients in chronic pain management programs than in retrospective studies reviewing patient databases, such as those using physician-reported International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes to assess opioid abuse and dependence.^56,57

In our study, results suggest that in patients with cancer, their marital status (single or divorced), SOAPP score with severe pain, and their current opioid dose (high level) were associated with NMOU behavior. Based on these findings, a universal screening, setting limits on opioid use by limiting supply, more intense follow-up with an interdisciplinary team to provide optimal use of medications for pain and symptom management, and the provision of counseling and support to patients and their family members may help prevent the development of these NMOU behaviors.^12,27,28 However, further studies in this patient population are needed.

Strengths and Limitations

More research is necessary to discover accurate tools for screening patients at risk for NMOU behaviors. Future studies should also evaluate whether to incorporate routine evaluation and documentation of NMOU behaviors as part of pain management in patients with cancer who are taking opioids, because routine screening, in addition to use of the SOAPP and CAGE-AID questionnaires for risk assessment, may fail to detect NMOU behaviors.²⁸ Our study results suggest that early detection of NMOU is possible, as most of the NMOU behaviors were seen within the first 2 follow-up visits, suggesting the need for early screening during follow-up. Based on the results of this research, it may be possible to clinically determine, through the combination of several independent predictors, those patients with cancer who are at a higher risk of NMOU behavior. These patients will require closer monitoring for optimal opioid use in cancer pain management.

There are many limitations of this study. The retrospective nature of the study allowed assessment of consecutive patient samples to determine the percentage of patients with NMOU behavior. The main limitation is that we were only able to include in the analysis variables already included in the electronic health record. Future prospective studies should allow for the measurement of variables that are not part of regular clinical care, such as urinary drug screens, and validated mood and quality-of-life tools. Urine drug testing might contribute to a better understanding of the frequency and severity of NMOU in the future. However, the most common NMOU behavior among patients with cancer is to take more than the prescribed dose of opioids. This behavior cannot be captured on a urine drug test, because either there will be no other drugs identified or the test will not show an absence of the prescribed opioids. There may be other useful information that a urine drug test can provide; therefore, it should be considered as one more contributory variable to the overall diagnosis of NMOU. In this study, we did not assess the Charlson Comorbidity Index, which could be a potential covariate for the final multivariate model for evaluating the risk of NMOU behavior. Other potential limitations may include residual confounding owing the nature of the study and lack of generalizability as the study data were obtained from a single tertiary cancer center.

Better and safer analgesics are urgently needed to replace the current opioids, which were developed between 60 and 220 years ago. These highly nonselective drugs not only bind to mu opioid receptors along the nociceptive pathway but also bind to other areas of the brain, where they lead to undesirable effects, including NMOU behavior.^12,58

Conclusions

In this prognostic study, 19% of patients with cancer developed NMOU behavior within a median duration of 8 weeks after initial supportive care center consultation. This finding highlights the need for continuous vigilance and careful use of these opioids in this population. Marital status, SOAPP score, pain severity, and high MEDD may be independent predictors for NMOU behavior. This information will assist clinicians and investigators designing clinical and research programs in this important field.

Supplement.

eBox. Nonmedical Opioid Use Behaviors

eFigure. Recursive Partitioning Classification and Regression Tree of Patients With the Presence of NMOU Behavior

eTable. Demographic and Clinical Characteristics by NMOU Behavior Status