Lacking alternative resources, many people with substance use disorders (SUDs) visit emergency departments (EDs), especially after an overdose (OD). This gives care providers an opportunity to provide referrals to treatment for those who need it. But EDs may lack the time and resources to select the patients at highest risk—those who could benefit from intervention. This is especially true given the increase in the number of patients with at least one risk factor, and the recent doubling in opioid-related ED visits (2005 to 2014), many of them linked to drug overdoses.
A team from Johns Hopkins set out to develop risk models to help identify the highest-risk patients. Their models, using comprehensive statewide databases, link recent ED records with recent death records. This makes it possible for care providers to identify factors associated with risk of death from opioid OD. The authors published their study online September 9, 2019, and in Annals of Emergency Medicine, January 2020.
Using records from Maryland, a state with a high rate of OD deaths, the study linked ED hospital claims for adult visits for nonfatal OD encounters (2014 to 2015) with medical examiner death records (2015 to 2016). Working with a variety of combinations of OD and substance use histories, investigators calculated the probability of OD death in hypothetical patients.
Meeting the criteria for substance-related ED encounters were 139,252 patients. The approximate breakdown by diagnosis, with overlaps:
- 26%: OUD
- 85%: SUD other than OUD
- 13%: a nonfatal drug OD
The drugs involved in nonfatal OD were opioids, 45%; benzodiazepines, 17%; cocaine, 6%; alcohol, 4%; all other, 43%
Results: Characteristics of Subjects
Patients with substance-related ED encounters: 139,252
Subsequent opioid OD deaths: 963
Case fatality rate: 69.2 per 10,000 patients
Patients in the first category—those with ED encounters related to substance use—had an exceedingly high risk of OD death: six times greater than the rate for the 1,452 adult patients using the ED for other reasons
The next step was to determine which of the 139,252 patients were most likely to be at risk of OD death. Checking possible risk factors against hospital records, the team identified three components that had a strong statistical association with death:
- An opioid use disorder and an additional SUD
- Three or more previous nonfatal ODs
- A previous nonfatal OD involving heroin
Underlying chronic pain and mental health diagnoses also were associated with higher risk of OD death, but not as strongly as the three leading categories.
Conditions and Diagnoses That Increase Risk of OD
It was possible to pinpoint the risk more precisely by looking for certain conditions and diagnostic patterns. For example, the more ODs the patient had, the higher the risk of OD death.
Likelihood of OD Death
Likelihood, by diagnosis, from most likely to least likely:
OUD & SUD OUD Other SUD No OUD & No Other SUD
Risk of OD death also varied within each of the four categories above: OUD & SUD, OUD, Other SUD, and No OUD & No Other SUD. Risk was highest if the patient had two previous ODs. Risk was somewhat lower with one previous OD, but remained high.
This finding emphasized the importance of considering a patient’s complete history when evaluating risk. Checking for previous encounters enabled caregivers to best evaluate the degree of risk.
It was clear that, using the data, the investigators could calculate the probability of OD death for hypothetical patients with various combinations of OD and substance use histories. In other words, the team now had a way to select the appropriate ED patients—the ones who would benefit most from intervention.
- Alcohol and cocaine were often involved in opioid OD deaths
- Fentanyl played an increasingly important role
- The more ODs a patient had during the study period, the higher the risk of later OD death
- Risk of OD death varied with the drug used in earlier nonfatal OD events; heroin was associated with the highest risk
- In two-thirds of OD deaths, the patient had received inpatient care during the study period
Some limitations the authors listed are inherent in the design of the study. Diagnostic codes in claims records were used; some risk factors or other conditions may have been missing or misclassified. Also, the authors lacked information on whether patients were enrolled in treatment programs, patients’ possible medication use during treatment, and any criminal justice involvement. These factors could play a role in OD death.
Illicit fentanyl couldn’t be distinguished from other synthetic opioids, so the authors were not able to determine fentanyl’s role in OD death.
The study didn’t take into account the role and resources of the specific hospital or hospitals caring for patients.
Findings in Maryland, a state with high OD rates, may not hold true for other regions.
The study results show that patients receiving care in EDs for substance-related conditions differ in OD risk; the ways in which they differ can be assessed; and assessment identifies those most likely to benefit from intervention, including those “at extremely high risk of overdose death.”
The team emphasized the importance of two responsibilities of caregivers:
- “Incorporating routine data from patient records to assess risk of future negative outcomes.” These are primary targets for linking with lifesaving care.
- Considering the full medical history when assessing risk, “rather than relying on single-event information.”
Specific interventions the authors mentioned for reducing fatal opioid ODs were naloxone use and medication-assisted treatment.
In closing, the team noted that because EDs are likely to “continue to act as a critical interface” with patients with SUDs, hospital systems must be able to respond to—and intervene in the care of—patients with SUDs.
“Integrating data currently available from hospital records, such as those presented in this article, may be one simple but effective step in this direction,” they noted.
§ § §
Comments After writing this article, we wondered how EDs would use the information to predict risk. Say, for example, a new OD patient in the ED has diagnoses of both OUD and SUD, but no previous ODs. Another new OD patient has a diagnosis of OUD only—but has already had two ODs.
Having read and assimilated the article, highly experienced staff at a busy ED might quickly process the information in the article, use it to identify the second patient as being at greater risk, and refer appropriately.
But could less-experienced staff start with the patients’ history and go through a similar thought process?
We asked the study’s lead author for her thoughts.
Dr. Krawczyk agreed that, in an ideal world, everyone who walks through an ED with these conditions would be provided with all available resources. “But this study is showing that having just a few pieces of information can help find those at especially high risk,” she added. “In theory, a busy ED could use relatively simple algorithms to rank people by ‘highest risk of subsequent fatal high risk.’”
She went on to explain that many types of predictive technologies are used, for example, to generate a risk score or an alert that could flag especially high-risk patients and prompt providers to take action. In reality, implementing such technologies isn’t easy, “partly because physicians are already overwhelmed with alerts.” So the next step—and this is something Dr. Krawczyk hopes to work on—is to find an effective way to get this information to people who can use it, facilitating the appropriate referral, to naloxone or treatment initiation.
Reference Krawczyk N, Eisenberg M, Schneider KE. Predictors of overdose death among high-risk emergency department patients with substance-related encounters: A data linkage cohort study [Epub before print Sep 9, 2019]. Ann Emerg Med. 2020; Jan;75(1):1-12. doi: 10.1016/j.annemergmed.2019.07.014