Dataset Information

Ensemble learning to predict opioid-related overdose using statewide prescription drug monitoring program and hospital discharge data in the state of Tennessee.

ABSTRACT:

Objective

To develop and validate algorithms for predicting 30-day fatal and nonfatal opioid-related overdose using statewide data sources including prescription drug monitoring program data, Hospital Discharge Data System data, and Tennessee (TN) vital records. Current overdose prevention efforts in TN rely on descriptive and retrospective analyses without prognostication.

Materials and methods

Study data included 3 041 668 TN patients with 71 479 191 controlled substance prescriptions from 2012 to 2017. Statewide data and socioeconomic indicators were used to train, ensemble, and calibrate 10 nonparametric "weak learner" models. Validation was performed using area under the receiver operating curve (AUROC), area under the precision recall curve, risk concentration, and Spiegelhalter z-test statistic.

Results

Within 30 days, 2574 fatal overdoses occurred after 4912 prescriptions (0.0069%) and 8455 nonfatal overdoses occurred after 19 460 prescriptions (0.027%). Discrimination and calibration improved after ensembling (AUROC: 0.79-0.83; Spiegelhalter P value: 0-.12). Risk concentration captured 47-52% of cases in the top quantiles of predicted probabilities.

Discussion

Partitioning and ensembling enabled all study data to be used given computational limits and helped mediate case imbalance. Predicting risk at the prescription level can aggregate risk to the patient, provider, pharmacy, county, and regional levels. Implementing these models into Tennessee Department of Health systems might enable more granular risk quantification. Prospective validation with more recent data is needed.

Conclusion

Predicting opioid-related overdose risk at statewide scales remains difficult and models like these, which required a partnership between an academic institution and state health agency to develop, may complement traditional epidemiological methods of risk identification and inform public health decisions.

SUBMITTER: Ripperger M

PROVIDER: S-EPMC8714265 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Objective This study was aimed to estimate the percentage of women taking opioids post-cesarean who could be detected in a state prescription monitoring program (PMP) and characteristics of those not able to be detected. Study Design This observational cohort study included all women with an Illinois address who delivered via cesarean section and used opioids within 24 hours prior to discharge at a tertiary care hospital between August 21, 2017 and March 1, 2018. The Illinois PMP was queried for presence of an opioid prescription filled within the first 3 months postpartum. Sociodemographic and clinical factors associated with an undetectable PMP record were evaluated in bivariable and multivariable logistic regression analyses. Results A total of 517 women underwent a cesarean delivery during the study period, of whom 344 (66.5%) met inclusion criteria. Of these women, 169 (49%) did not have a detectable PMP record of filling any outpatient postpartum prescription opioid. On bivariable and multivariable logistic regression analysis, year of delivery (2018 vs. 2017) was significantly associated with a higher incidence of detectable postpartum prescription opioid record in the PMP with increasing relative risk of detectable records in the second year of analyses ( n = 110/244 [45%] in 2017 vs. n = 59/100 [59%] in 2018, adjusted risk ratio [aRR] = 1.32, 95% confidence interval [CI]: 1.06-1.64, p = 0.013). No other sociodemographic or clinical characteristics was significantly associated. Conclusion Nearly half of women who underwent a cesarean section and who were administered opioids 24 hours prior to discharge did not have a detectable postpartum opioid prescription in the PMP. While identification of prescription filling improved with time, many of women were not detectable in the PMP system. These data call into question the accuracy of PMPs in identifying prescription opioid filling patterns in the postpartum setting.

Project description:ImportanceThe opioid epidemic continues to be a public health crisis in the US.ObjectiveTo assess the patient factors and early time-varying prescription-related factors associated with opioid-related fatal or nonfatal overdose.Design, setting, and participantsThis cohort study evaluated opioid-naive adult patients in Oregon using data from the Oregon Comprehensive Opioid Risk Registry, which links all payer claims data to other health data sets in the state of Oregon. The observational, population-based sample filled a first (index) opioid prescription in 2015 and was followed up until December 31, 2018. Data analyses were performed from March 1, 2020, to June 15, 2021.ExposuresOverdose after the index opioid prescription.Main outcomes and measuresThe outcome was an overdose event. The sample was followed up to identify fatal or nonfatal opioid overdoses. Patient and prescription characteristics were identified. Prescription characteristics in the first 6 months after the index prescription were modeled as cumulative, time-dependent measures that were updated monthly through the sixth month of follow-up. A time-dependent Cox proportional hazards regression model was used to assess patient and prescription characteristics that were associated with an increased risk for overdose events.ResultsThe cohort comprised 236 921 patients (133 839 women [56.5%]), of whom 667 (0.3%) experienced opioid overdose. Risk of overdose was highest among individuals 75 years or older (adjusted hazard ratio [aHR], 3.22; 95% CI, 1.94-5.36) compared with those aged 35 to 44 years; men (aHR, 1.29; 95% CI, 1.10-1.51); those who were dually eligible for Medicaid and Medicare Advantage (aHR, 4.37; 95% CI, 3.09-6.18), had Medicaid (aHR, 3.77; 95% CI, 2.97-4.80), or had Medicare Advantage (aHR, 2.18; 95% CI, 1.44-3.31) compared with those with commercial insurance; those with comorbid substance use disorder (aHR, 2.74; 95% CI, 2.15-3.50), with depression (aHR, 1.26; 95% CI, 1.03-1.55), or with 1 to 2 comorbidities (aHR, 1.32; 95% CI, 1.08-1.62) or 3 or more comorbidities (aHR, 1.90; 95% CI, 1.42-2.53) compared with none. Patients were at an increased overdose risk if they filled oxycodone (aHR, 1.70; 95% CI, 1.04-2.77) or tramadol (aHR, 2.80; 95% CI, 1.34-5.84) compared with codeine; used benzodiazepines (aHR, 1.06; 95% CI, 1.01-1.11); used concurrent opioids and benzodiazepines (aHR, 2.11; 95% CI, 1.70-2.62); or filled opioids from 3 or more pharmacies over 6 months (aHR, 1.38; 95% CI, 1.09-1.75).Conclusions and relevanceThis cohort study used a comprehensive data set to identify patient and prescription-related risk factors that were associated with opioid overdose. These findings may guide opioid counseling and monitoring, the development of clinical decision-making tools, and opioid prevention and treatment resources for individuals who are at greatest risk for opioid overdose.

Project description:BackgroundA large proportion of individuals who use heroin report initiating opioid use with prescription opioids. However, patterns of prescription opioid use preceding heroin-related overdose have not been described.ObjectiveTo describe prescription opioid use in the year preceding heroin overdose.DesignCase-control study comparing prescription opioid use with a heroin-involved overdose, non-heroin-involved opioid overdose, and non-overdose controls from 2015 to 2017.ParticipantsOregon Medicaid beneficiaries with linked administrative claims, vital statistics, and prescription drug monitoring program data.Main measuresOpioid, benzodiazepine, and other central nervous system depressant prescriptions preceding overdose; among individuals with one or more opioid prescription, we assessed morphine milligram equivalents per day, overlapping prescriptions, prescriptions from multiple prescribers, long-term use, and discontinuation of long-term use.Key resultsWe identified 1458 heroin-involved overdoses (191 fatal) and 2050 non-heroin-involved opioid overdoses (266 fatal). In the 365 days prior to their overdose, 45% of individuals with a heroin-involved overdose received at least one prescribed opioid compared with 78% of individuals who experienced a non-heroin-involved opioid overdose (p < 0.001). For both heroin- and non-heroin-involved overdose cases, the likelihood of receiving an opioid increased with age. Among heroin overdose cases with an opioid dispensed, the rate of multiple pharmacy use was the only high-risk opioid pattern that was greater than non-overdose controls (adjusted odds ratio 3.2; 95% confidence interval 1.48 to 6.95). Discontinuation of long-term opioid use was not common prior to heroin overdose and not higher than discontinuation rates among non-overdose controls.ConclusionsAlthough individuals with a heroin-involved overdose were less likely to receive prescribed opioids in the year preceding their overdose relative to non-heroin opioid overdose cases, prescription opioid use was relatively common and increased with age. Discontinuation of long-term prescription opioid use was not associated with heroin-involved overdose.

Project description:STUDY OBJECTIVE:We assess whether an automated prescription drug monitoring program intervention in emergency department (ED) settings is associated with reductions in opioid prescribing and quantities. METHODS:We performed a retrospective cohort study of ED visits by Medicaid beneficiaries. We assessed the staggered implementation (pre-post) of automated prescription drug monitoring program queries at 86 EDs in Washington State from January 1, 2013, to September 30, 2015. The outcomes included any opioid prescribed within 1 day of the index ED visit and total dispensed morphine milligram equivalents. The exposure was the automated prescription drug monitoring program query intervention. We assessed program effects stratified by previous high-risk opioid use. We performed multiple sensitivity analyses, including restriction to pain-related visits, restriction to visits with a confirmed prescription drug monitoring program query, and assessment of 6 specific opioid high-risk indicators. RESULTS:The study included 1,187,237 qualifying ED visits (898,162 preintervention; 289,075 postintervention). Compared with the preintervention period, automated prescription drug monitoring program queries were not significantly associated with reductions in the proportion of visits with opioid prescribing (5.8 per 1,000 encounters; 95% confidence interval [CI] -0.11 to 11.8) or the amount of prescribed morphine milligram equivalents (difference 2.66; 95% CI -0.15 to 5.48). There was no evidence of selective reduction in patients with previous high-risk opioid use (1.2 per 1,000 encounters, 95% CI -9.5 to 12.0; morphine milligram equivalents 1.22, 95% CI -3.39 to 5.82). The lack of a selective reduction in high-risk patients was robust to all sensitivity analyses. CONCLUSION:An automated prescription drug monitoring program query intervention was not associated with reductions in ED opioid prescribing or quantities, even in patients with previous high-risk opioid use.