Opioid Overdose Risk in Patients Returning to the Emergency Department for Pain

Abstract

Objective

Using the Risk Index for Overdose or Serious Opioid-induced Respiratory Depression (CIP-RIOSORD) in patients returning to the emergency department (ED) for pain and discharged with an opioid prescription, we assessed overall opioid overdose risk and compared risk in opioid naive patients to those who are non-opioid naive.

Design

This was a secondary analysis from a prospective observational study of patients ≥ 18 years old returning to the ED within 30 days. Data were collected from patient interviews and chart reviews. Patients were categorized as Group 1 (not using prescription opioids) or Group 2 (consuming prescription opioids). Statistical analyses were performed using Fisher’s exact and Wilcoxon’s rank sum tests. Risk class and probability of overdose was determined using Risk Index for Overdose or Serious Opioid-induced Respiratory Depression (CIP-RIOSORD).

Results

Of the 389 enrollees who returned to the ED due to pain within 30 days of an initial visit, 67 (17%) were prescribed opioids. The majority of these patients were in Group 1 (60%). Both Group 1 (n = 40) and Group 2 (n = 27) held an average CIP-RIOSORD risk class of 3. Race significantly differed between groups; the majority of Group 1 self-identified as African American (80%) (P = .0267). There were no differences in age, gender, or CIP-RIOSORD risk class between groups. However, Group 2 had nearly double the number of predictive factors (median = 1.93) as Group 1 (median = 1.18) (P = .0267).

Conclusions

A substantial proportion of patients (25%) were high risk for opioid overdose. CIP-RIOSORD may prove beneficial in risk stratification of patients discharged with prescription opioids from the ED.

Introduction

Over 168 million opioid prescriptions were dispensed within the United States in 2018 [1]. Opioids have accounted for most drug-overdose related deaths in recent years and remains a significant cause of death in the United States. Although rates of opioid prescribing and prescription opioid-related deaths declined between 2017 and 2018, there was a 9.7% relative increase in rates of emergency department (ED) visits for opioid-related nonfatal overdose between 2018 and 2019 [2–4]. In 2018, 32% of opioid overdose related deaths were attributable to prescription opioids [2, 3].

The Risk Index for Overdose or Serious Opioid-induced Respiratory Depression (CIP-RIOSORD) is a validated tool that calculates a risk class (highest risk class) [5] for a patient’s probability of experiencing an opioid overdose within the next 6 months [6]. The tool was initially developed using the Veteran’s Health Administration patient population and then modified based on a commercially insured health plan (CIP) claims database. Its development focused on risk factors present within the past 6 months in patients who had developed prescription opioid-induced respiratory depression (OIRD) as compared to those without this adverse event [7]. Predictive factors to assess an individual’s risk class and estimated predicted probability of overdose within the following 6 months included substance use disorders, bipolar disorder, schizophrenia, stroke, cerebrovascular disease, significant kidney disease, heart failure, nonmalignant pancreatic disease, chronic pulmonary disease, recurrent headache, current maximum prescribed opioid dose, and use of specific medications (i.e., fentanyl, morphine, benzodiazepine, antidepressants) [7]. CIP-RIOSORD attributes point values to each of these predictive factors which are summed to derive a risk score (for example, a healthcare visit for substance use disorder is assigned 25 points). The calculated risk score is equated to a specific risk class, each of which is associated with an average predicted probability of OIRD within the next 6 months (Table 1).

Table 1.

CIP-RIOSORD risk class, risk score, and predicted probability of serious opioid-induced respiratory depression [6].

Risk Class	Risk Score (points)	Average Predicted Probability of an overdose/OIRD in the next 6 months
1	0–4	1.9%
2	5–7	4.8%
3	8–9	6.8%
4	10–17	15.1%
5	18–25	29.8%
6	26–41	55.1%
7	> 42	83.4%

Since its development, RIOSORD has been successfully applied in several studies. The VHA-RIOSORD tool has been independently validated by Metcalfe et al. as a tool with a strong concordance between ability to predict those at risk for opioid adverse events versus the actual incidence of such [6]. It has been used in other studies to target veterans with a RIOSORD risk class of ≥4 in order to assess naloxone [7]. Other studies have used RIOSORD to track outcomes post-implementation of a clinical decision support tool in the management of chronic prescription opioid use [8]. However, we are unaware of any studies to date that have applied this tool specifically within the ED population.

Our objectives were to use the CIP-RIOSORD tool in patients returning to the ED for pain who were subsequently discharged with an opioid prescription to: (1) determine their opioid overdose risk; and (2) compare opioid overdose risk in patients who are opioid naive to those who are non-opioid naive.

Methods

Study Setting and Enrollment

Study enrollment occurred in the ED of an urban safety-net hospital system with an annual patient volume over 70,000. The average patient presenting to our ED is 50 years old, 51% are African American, 50% male, and 25% are on Medicare. This study was a secondary descriptive analysis of data collected from an IRB approved prospective observational study of 389 patients ≥ 18 years of age returning to the ED within 30 days of an initial (index) visit (Figure 1). Full study details have been published previously [9]. Briefly, all community-dwelling adult patients able to provide consent were eligible for inclusion. Patients not meeting these criteria and those instructed to return to the ED within a set time frame were excluded (e.g., scheduled wound checks or return for suture removal). Patients were enrolled using systematic-time block sampling to mimic ED utilization rates by ED shift times based on historical data from our institution. This analysis only includes data from the ED visits in which a patient was discharged with a prescription opioid during the study period (n = 67).

Figure 1.

Enrollment schematic.

Variables Collected

Sociodemographic variables, comorbidities, medication history, and ED returns within 30 days after discharge were collected from patient interviews and chart review. Risk class and probability of overdose was determined for each patient using the CIP-RIOSORD tool [10]. Patients who reported not using prescription opioids at home and were discharged with an opioid prescription were classified as Group 1 (opioid naive). Patients reporting use of prescription opioids at home and who were discharged with an opioid prescription were classified as Group 2 (non-opioid naive).

Data Analysis

Descriptive summaries included counts and percentages for categorical variables and average, medians, and quartiles for numeric variables. Statistical analyses were performed using Fisher’s exact and Wilcoxon’s rank sum tests. All analyses were performed in SAS® for Windows Version 9.4.

Results

Population Characteristics

There were 389 patients who returned to the ED within 30 days of their index visit during the study period (Figure 1). Of these, 54% (212) had a pain related diagnosis at their index visit and 13.4% (52) were discharged from this visit with an opioid prescription. During the study period, there were a total of 67 visits in which a prescription opioid was issued. The average age was 46 (range 19– 81) and 20% (17) of patients were over age 55. Most patients were male (38, 57%), African American (41, 61%), did not report a history of chronic pain (44, 66%), and reported not using prescription opioids prior to the ED visit (Group 1) (40, 60%). Two patients in Group 2 reported that they were currently in a pain contract at the time of their ED revisit. The average number of CIP-RIOSORD predictive factors was 1 and the average CIP-RIOSORD risk class was 3. Over 25% of patients had a risk class ≥5, corresponding to a 29.8% (29.7 – 30 confidence interval) to 83.4% (83.2–83.7) average predicted probability of a serious opioid-induced respiratory event within the next 6 months.

Tables 2 and 3 display the characteristics of Groups 1 and 2. Of the 67 visits, 40 (60%) were by patients in Group 1. Race significantly differed between the two groups (P < .001). Most patients in Group 1 were African American (32, 80%) and in Group 2 were Caucasian (17, 63%). There were no differences in age, gender, CIP-RIOSORD risk scores, or CIP-RIOSORD risk class between groups. However, there was a significant difference in the number of CIP-RIOSORD predictive factors between groups, with Group 2 having nearly double the number of predictive factors (median = 1.93) as Group 1 (median = 1.18) (P = .0267) (Table 2).

Table 2.

Characteristics and comparison of study population by demographics and CIP-RIOSORD predictive factors (as categorical variables)

Variable	Category	Group 1 (n = 40, 60%)	Group 2 (n = 27, 40%)	Overall (n = 67)	P-value
Race	African American	32 (80)	9 (33)	41 (61)	<.001
	Caucasian	8 (20)	17 (63)	25 (37)
	Other	0 (0)	1 (4)	1 (1)
Gender	Female	18 (45)	11 (41)	29 (43)	.840
	Male	22 (55)	16 (59)	38 (57)
Mental illness*	Yes	9 (23)	11 (41)	20 (30)	.173
	No	31 (78)	16 (59)	47 (70)
Chronic pain	Yes	10 (25)	13 (48)	23 (34)	.068
	No	30 (75)	14 (52)	44 (66)
CIP-RIOSORD Predictive Factors
Substance abuse	Yes	6 (15)	1 (4)	7 (10)	.228
	No	34 (85)	26 (96)	60 (90)
Antidepressant use	Yes	13 (33)	14 (54)	27 (41)	.124
	No	27 (68)	12 (46)	39 (59)
Benzodiazepine use	Yes	6 (15)	5 (19)	11 (16)	.745
	No	34 (85)	22 (81)	56 (84)
COPD	Yes	6 (15)	8 (30)	14 (21)	.221
	No	34 (85)	19 (70)	53 (79)
Heart failure	Yes	0 (0)	1 (4)	1 (1)	.403
	No	40 (100)	26 (96)	66 (99)
Kidney disease	Yes	3 (8)	2 (7)	5 (7)	1.000
	No	37 (93)	25 (93)	62 (93)
Pancreatic disease	Yes	4 (10)	2 (7)	6 (9)	1.000
	No	36 (90)	25 (93)	61 (91)
Recurrent headaches	Yes	0 (0)	2 (7)	2 (3)	.159
	No	40 (100)	25 (93)	65 (97)
Schizophrenia/bipolar	Yes	6 (15)	4 (15)	10 (15)		1.000
	No	34 (85)	23 (85)	57 (85)
Stroke	Yes	3 (8)	2 (7)	5 (7)		1.000
	No	37 (93)	25 (93)	62 (93)

All tests done using Fisher's exact test. RIOSORD = Risk Index for Overdose or Serious Opioid-induced Respiratory Depression.

^*Other than schizophrenia or bipolar disorder.

Table 3.

Characteristics of Study Population (as continuous variables)

Variable	Group	N	Mean	SD	Min	1st Quartile	Median	3rd Quartile	Max	P-value
Age (years)	1	40	46.45	15.52	19	33	49	59	81	.7635
	2	27	45.48	11.57	25	33	51	54	61
Number of predictive factors*	1	40	1.18	1.38	0	0	1	2	5	.0267
	2	27	1.93	1.47	0	1	2	3	5
CIP-RIOSORD Risk Class	1	40	2.93	2.13	0	1	3	5	7	.3906
	2	27	3.26	1.81	1	1	4	4	6
CIP-RIOSORD Risk Score	1	40	11.95	16.31	0	0	8	18	61	.4099
	2	27	12.15	10.42	0	0	13	17	33

All tests done using Wilcoxon’s rank sum test. RIOSORD = Risk Index for Overdose or Serious Opioid-induced Respiratory Depression; SD = standard deviation.

^*Number of CIP-RIOSORD predictive factors.

Twenty-two (55%) patients in Group 1 met at least one CIP-RIOSORD predictive factor (range 1–5) and 27.5% (11) of patients had a risk class ≥ 5. The most common predictive factor in Group 1 was antidepressant use (33%) followed by history of bipolar disorder or schizophrenia (15%), substance use disorder (15%), benzodiazepine use (15%), and chronic obstructive pulmonary disorder (COPD) history (15%). In Group 2, antidepressant use was also the most common predictive factor (54%), followed by a history of COPD (30%) and benzodiazepine use (19%). Slightly more than 22% (6) of patients in Group 2 had a risk class ≥ 5.

Discussion

Our findings illustrate a considerable proportion of ED patients (25%) were discharged with a prescription opioid despite having a high-predicted probability of a serious opioid related respiratory event within the next 6 months (risk class ≥ 5). The average risk class for both groups was 3 (Group 1, median 3; Group 2, median 4), translating to a 6.8% average predicted probability of an overdose within the next 6 months. Interestingly, the proportion of patients discharged with an opioid prescription who had a risk class ≥ 5 was higher in patients newly prescribed opioids (Group 1) (11, 27.5%) compared to those who were not opioid-naive (Group 2) (6, 22%).

Patients newly prescribed opioids shared similar predictive risk factors (antidepressant use, history of COPD, and benzodiazepine use) as those who were not opioid naive. There was a significant difference in the number of CIP-RIOSORD predictive factors between groups, with Group 2 having nearly double the number of predictive factors (median = 1.93) than Group 1 (median = 1.18) (P = .0267) (Table 3). This finding is interesting given that there was no significant difference in overall risk scores or class between the groups. Thus, despite patients in Group 2 having more predictive factors on average than those in Group 1, the average risk score and risk class was similar between the groups. This finding may be explained by Group 1 tending to have more of the “heavier” weighted predictive factors (such as schizophrenia, bipolar disorder or substance use disorder) compared to Group 2. This may also explain why Group 1 had a higher proportion of patients in the higher risk classes (risk class ≥ 5).

Despite both groups sharing a similar prevalence of chronic pain, race significantly differed between the two groups (P < .001) with the majority of Group 1 self-identifying as African American and the majority in Group 2 as Caucasian. This finding is consistent with literature showing disparities in opioid administration in African American populations [11]. There are several potential explanations that may account for the apparent racial differences between our study groups, such as differences in the type of pain conditions present within the groups, patient access to prescriptions (i.e., lack of a primary care physician) and ability to fill prescriptions received, and clinician bias.

There are several important limitations to our study. First, this was a secondary analysis on data that was collected previously from a study conducted at a single center. As the primary study involved only patients presenting within one institution (an institution predominantly serving an urban, underserved patient population), our results may not be applicable across other sociodemographically diverse emergency department patient populations. Although the objectives for the primary study were not to assess probability of a serious opioid adverse event, all data components of the CIP-RIOSORD risk index were collected and verified through review of the electronic medical record and by patient interview. Another important limitation is that we did not follow patients for 6 months after receipt of their ED opioid prescription to assess prescription fill rates/consumption or the actual incidence of an opioid adverse event after ED discharge. However, since the CIP-RIOSORD tool has been validated we do not think our findings are diminished by this limitation. Although, if we had followed patients for 6 months, it would have been interesting to compare the CIP-RIOSORD predicted probability to actual rates as well as the number of days the opioid was prescribed for, as most ED opioid prescriptions are written only for a few day’s supply.

Emergency department prescription opioid rates are on the decline given changes in recent opioid legislation and prescribing patterns spurred by the opioid epidemic [12]. This decline, however, varies by geographic region and patient characteristics [4, 13, 14]. Despite the national decline in overall prescription rates, we found a substantial proportion of patients (25%) were high risk for a prescription opioid overdose within 6 months. With pain accounting for up to 70% of emergency department visits, opioids will continue to play an important role in the management of certain emergent pain conditions [15, 16]. Therefore, it is essential that opioids, when indicated, are prescribed appropriately and safely. CIP-RIOSORD may be an effective risk stratification tool for use in the emergency department setting to reduce incidence of opioid overdose. How to best integrate the tool within the emergency department workflow, such as integration within the electronic medical record, should be further investigated.

 

Conflict of interests/Disclosures: Drs. Sophia Sheikh, Colleen Kalynych, and Phyllis Hendry both received support from the Florida Medical Malpractice Joint Underwriter’s Association Dr. Alvin E. Smith Safety of Healthcare Services Grant. Dr. Sophia Sheikh was also supported by the NIH/NIA-funded Jacksonville Aging Studies Center (JAX-ASCENT; R33AG05654). Dr. Ashley Norse was the recipient of a Dean’s Fund for Research Award from the University of Florida College of Medicine- Jacksonville. The remaining authors have no conflicts of interest to disclose.

Funding: This work was supported by a Dean’s Fund for Research Award from the University of Florida College of Medicine- Jacksonville; a Florida Medical Malpractice Joint Underwriter’s Association Dr. Alvin E. Smith Safety of Healthcare Services Grant; the NIH/NIA-funded Jacksonville Aging Studies Center (JAX-ASCENT; R33AG05654); and in part by the NIH National Center for Advancing Translational Sciences (NCATS) grant UL1 TR000064.

Abstract presentations covering portions of study results were presented at the North American Congress of Clinical Toxicology Annual Meeting on September 10–14, 2020 (virtual meeting); Society of Academic Emergency Medicine Annual Meeting on May 15, 2019, in Las Vegas, Nevada; and The Gerontological Society of America Annual Meeting held November 14–18, 2018 in Boston, Massachusetts.

Author Contribution Statement: S.S., A.B.N., C.K., and P.H. conceived the study, designed the trial, and obtained research funding. S.S., A.B.N., C.K., and P.H. supervised the conduct of the trial and data collection. M.H. and D.H. undertook recruitment of patients and along with D.J. and E.E. managed the data, including quality control. C.S. provided statistical advice on study design and analyzed the data. S.S. and D.J. drafted the manuscript, and all authors contributed substantially to its revision. S.S. takes responsibility for the paper as a whole.