Review of Factors, Methods, and Outcome Definition in Designing Opioid Abuse Predictive Models

A deep learning method to detect opioid prescription and opioid use disorder from electronic health records

OBJECTIVE

As the opioid epidemic continues across the United States, methods are needed to accurately and quickly identify patients at risk for opioid use disorder (OUD). The purpose of this study is to develop two predictive algorithms: one to predict opioid prescription and one to predict OUD.

MATERIALS AND METHODS

We developed an informatics algorithm that trains two deep learning models over patient Electronic Health Records (EHRs) using the MIMIC-III database. We utilize both the structured and unstructured parts of the EHR and show that it is possible to predict both challenging outcomes.

RESULTS

Our deep learning models incorporate elements from EHRs to predict opioid prescription with an F1-score of 0.88 ± 0.003 and an AUC-ROC of 0.93 ± 0.002. We also constructed a model to predict OUD diagnosis achieving an F1-score of 0.82 ± 0.05 and AUC-ROC of 0.94 ± 0.008.

DISCUSSION

Our model for OUD prediction outperformed prior algorithms for specificity, F1 score and AUC-ROC while achieving equivalent sensitivity. This demonstrates the importance of a) deep learning approaches in predicting OUD and b) incorporating both structured and unstructured data for this prediction task. No prediction models for opioid prescription as an outcome were found in the literature and therefore our model is the first to predict opioid prescribing behavior.

CONCLUSION

Algorithms such as those described in this paper will become increasingly important to understand the drivers underlying this national epidemic.

Self-Reported Pain Tolerance and Opioid Pain Medication Use After Foot and Ankle Surgery

BACKGROUND

Orthopaedic surgeons must consider their postoperative pain management strategies to minimize harm from prescription opioid use. Patients often reference their pain threshold to predict how they will tolerate surgical pain and the need for postoperative analgesia, but the direct relationship between these factors has not yet been studied. The purpose of this study was to determine the relationship between patients' self-reported pain tolerance and prescription opioid usage after foot and ankle surgery.

METHODS

This is a retrospective follow-up of a prospective cohort study of adult patients who underwent outpatient foot and ankle surgeries. Patient and procedural demographics, opioid pills dispensed, and opioid pills consumed by the first postoperative visit were obtained. Patients were contacted at a mean of 13.1 ± 4.0 months postoperatively and asked to respond to the qualitative statement "Pain doesn't bother me as much as it does most people." Patients were also asked their quantitative pain threshold (0-100), with 0 being "very pain intolerant" and 100 being a "very high pain tolerance," as well other questions regarding past surgical and narcotic consumption history.

RESULTS

Of the 700 survey respondents, the average age was 50.9 years and 34.7% were male. Bivariate analysis determined that predictors of lower postoperative opioid consumption included higher quantitative (P = .047) and qualitative (P = .005) pain tolerance scores. Multivariate analysis for the entire cohort demonstrated that higher qualitative pain threshold was associated with lower postoperative opioid consumption (P = .005) but this did not meet statistical significance as an independent predictor of the top quartile of pill consumers.

CONCLUSION

Assessment of both qualitative and quantitative score of patients' pain threshold prior to surgery may assist the surgeon in tailoring postoperative pain control. Additionally, asking this question can create an opportunity for educating patients regarding responsible utilization of narcotic medication.

LEVELS OF EVIDENCE

Level III.

Diagnostic and Predictive Capacity of the Spanish Versions of the Opioid Risk Tool and the Screener and Opioid Assessment for Patients with Pain-Revised: A Preliminary Investigation in a Sample of People with Noncancer Chronic Pain

INTRODUCTION

Accurate assessment of the risk of opioid abuse and misuse in people with noncancer chronic pain is crucial for their prevention. This study aimed to provide preliminary evidence of the diagnostic and predictive capacity of the Spanish versions of the Opioid Risk Tool (ORT) and the Screener and Opioid Assessment for Patients with Pain-Revised (SOAPP-R).

METHODS

We used the Current Opioid Misuse Measure (COMM) as criterion measure to assess the capacity of each tool to identify patients misusing opioids at the time of the assessment. Eighteen months later, we used the COMM and the Drug Abuse Screening Test-10 (DAST-10) to assess their predictive capacity. In total, 147 people with noncancer chronic pain participated in the diagnostic study, and 42 in the predictive study.

RESULTS

Receiver operating curve analysis showed that the SOAPP-R had an excellent capacity to identify participants who were misusing opioids at the time of assessment (area under the curve [AUC] = 0.827). The diagnostic capacity of the ORT was close to acceptable (AUC = 0.649-0.669), whereas its predictive capacity was poor (AUC = 0.522-0.554). The predictive capacity of the SOAPP-R was close to acceptable regarding misuse (AUC = 0.672) and poor regarding abuse (AUC = 0.423).

CONCLUSION

In the setting of Spanish-speaking communities, clinicians should be cautious when using these instruments to make decisions on opioid administration. Further research is needed on the diagnostic and predictive capacity of the Spanish versions of both instruments.

Risk factors of chronic opioid use after surgical procedures in noncancer patients: A nationwide case-control study

BACKGROUND

Surgery is an indication for opioid prescription in noncancer patients, and chronic use of opioids is associated with overdose and abuse.

OBJECTIVES

We aimed to evaluate the prevalence and risk factors associated with chronic opioid use (COU) following surgery among noncancer patients.

DESIGN

A nationwide case-control study.

SETTING

Retrospective analysis of the annual national patient sample data from 2012 to 2018 in South Korea.

PATIENTS

Adults without cancer who had undergone surgery and received noninjectable opioids during hospital stay.

MAIN OUTCOME MEASURES

COU during 3 months following surgery.

RESULTS

A total of 15 543 participants were included, and the prevalence overall and in opioid-naïve users was 8.1 and 5.7%, respectively. Prior exposure patterns of opioids [intermittent user, adjusted odds ratio (aOR) 2.35; 95% CI, 2.00 to 2.77, and continuous user, aOR 8.58; 95% CI, 6.54 to 11.24] and concomitant use of benzodiazepine (in continuous user, aOR 18.60; 95% CI 11.70 to 29.55) were strongly associated with COU compared with naïve users. Morphine milligram equivalent, type of opioid strength at discharge and prescription of nonopioid analgesics at discharge were also associated with COU. Compared with minor surgery, knee (aOR 1.49; 95% CI 1.17 to 1.89), spine (aOR 1.65; 95% CI 1.33 to 2.06) and shoulder (aOR 2.54; 95% CI 1.97 to 3.27) procedures showed a significantly positive association with COU. Sensitivity analysis in opioid-naïve patients showed similar results.

CONCLUSION

About 8.1% of noncancer patients who had undergone surgery and were prescribed noninjectable opioids became chronic opioid users in Korea. Identified risk factors could be used to derive strategies for safe opioid use in noncancer patients in the future.

Long-term opioid therapy definitions and predictors: A systematic review

PURPOSE

This review sought to (a) describe definitions of long-term opioid therapy (LTOT) outcome measures, and (b) identify the predictors associated with the transition from short-term opioid use to LTOT for opioid-naïve individuals.

METHODS

We conducted a systematic review of the peer-reviewed literature (January 2007 to July 2018). We included studies examining opioid use for more than 30 days. We classified operationalization of LTOT based on criteria used in the definitions. We extracted LTOT predictors from multivariate models in studies of opioid-naïve individuals.

RESULTS

The search retrieved 5,221 studies, and 34 studies were included. We extracted 41 unique variations of LTOT definitions. About 36% of definitions required a cumulative duration of opioid use of 3 months. Only 17% of definitions considered consecutive observation periods, 27% used days' supply, and no definitions considered dose. We extracted 76 unique predictors of LTOT from seven studies of opioid-naïve patients. Common predictors included pre-existing comorbidities (21.1%), non-opioid prescription medication use (13.2%), substance use disorders (10.5%), and mental health disorders (10.5%).

CONCLUSIONS

Most LTOT definitions aligned with the chronic pain definition (pain more than 3 months), and used cumulative duration of opioid use as a criterion, although most did not account for consistent use. Definitions were varied and rarely accounted for prescription characteristics, such as days' supply. Predictors of LTOT were similar to known risk factors of opioid abuse, misuse, and overdose. As LTOT becomes a central component of quality improvement efforts, researchers should incorporate criteria to identify consistent opioid use to build the evidence for safe and appropriate use of prescription opioids.

Predictors of chronic prescription opioid use after orthopedic surgery: derivation of a clinical prediction rule

Background

Prescription opioid use at high doses or over extended periods of time is associated with adverse outcomes, including dependency and abuse. The aim of this study was to identify mediating variables that predict chronic opioid use, defined as three or more prescriptions after orthopedic surgery.

Methods

Individuals were ages between 18 and 50 years and undergoing arthroscopic hip surgery between 2004 and 2013. Two categories of chronic opioid use were calculated based on individuals (1) having three or more unique opioid prescriptions within 2 years and (2) still receiving opioid prescriptions > 1 year after surgery. Univariate elationships were identified for each predictor variable, then significant variables (P > 0.15) were entered into a multivariate logistic regression model to identify the most parsimonious group of predictor variables for each chronic opioid use classification. Likelihood ratios were derived from the most robust groups of variables.

Results

There were 1642 participants (mean age 32.5 years, SD 8.2, 54.1% male). Nine predictor variables met the criteria after bivariate analysis for potential inclusion in each multivariate model. Eight variables: socioeconomic status (from enlisted rank family), prior use of opioid medication, prior use of non-opioid pain medication, high health-seeking behavior before surgery, a preoperative diagnosis of insomnia, mental health disorder, or substance abuse were all predictive of chronic opioid use in the final model (seven variables for three or more opioid prescriptions; four variables for opioid use still at 1 year; all< 0.05). Post-test probability of having three or more opioid prescriptions was 93.7% if five of seven variables were present, and the probability of still using opioids after 1 year was 69.6% if three of four variables were present.

Conclusion

A combination of variables significantly predicted chronic opioid use in this cohort. Most of these variables were mediators, indicating that modifying them may be feasible, and the potential focus of interventions to decrease the risk of chronic opioid use, or at minimum better inform opioid prescribing decisions. This clinical prediction rule needs further validation.

Prospective Evaluation of Utilization Patterns and Prescribing Guidelines of Opioid Consumption Following Orthopedic Foot and Ankle Surgery

BACKGROUND

Overprescription of narcotic pain medication is a major culprit in the present opioid epidemic plaguing the United States. The current literature on lower extremity opioid usage has limitations and would benefit from additional study. The purpose of our study was to prospectively assess opioid consumption patterns following outpatient orthopedic foot and ankle procedures.

METHODS

Patients undergoing outpatient orthopedic foot and ankle procedures who met inclusion criteria had the following prospective information collected: patient demographics, preoperative health history, patient-reported outcomes, anesthesia type, procedure type, opioid prescription and consumption details. The morphine equivalent dose was calculated for each prescription and then converted to the equivalent of a 5-mg oxycodone "pill." Univariable analyses were performed to identify variables with a statistically robust association with opioid consumption for inclusion in a multivariable linear regression. A stepwise backward regression was then performed to identify independent predictors of opioid consumption. Postoperative opioid utilization was reported for 988 patients (mean age: 49 years).

RESULTS

Overall, patients consumed a median of 20 pills whereas the median number of pills prescribed was 40. This resulted in a utilization rate of 50% and 20 631 pills left unused. Independent factors associated with higher opioid consumption were anesthesia type ( P < .004), age <60 years ( P < .001), preoperative visual analog scale (VAS) pain report of >6 ( P = .008), and bony procedures ( P = .008); residual standard error 16.73 ( F_7,844=14.3, P < .001).

CONCLUSION

Our study found that patients who underwent orthopedic foot and ankle procedures were overprescribed narcotic medication by nearly twice the amount that was actually consumed. Although we identified 4 independent factors associated with opioid consumption, the large residual standard error suggests that there remains a substantial degree of unexplained variance of opioid consumption observed in the patient population. Physicians face a challenging task of setting appropriate protocols when balancing pain relief and generalizable guidelines.

LEVEL OF EVIDENCE

Level II, prospective observational cohort study.