Adaptive interventions for opioid prescription management and consumption monitoring

Abstract

Objectives

While opioid addiction, treatment, and recovery are receiving attention, not much has been done on adaptive interventions to prevent opioid use disorder (OUD). To address this, we identify opioid prescription and opioid consumption as promising targets for adaptive interventions and present a design framework.

Materials and Methods

Using the framework, we designed Smart Prescription Management (SPM) and Smart Consumption Monitoring (SCM) interventions. The interventions are evaluated using analytical modeling and secondary data on doctor shopping, opioid overdose, prescription quality, and cost components.

Results

SPM was most effective (30–90% improvement, for example, prescriptions reduced from 18 to 1.8 per patient) for extensive doctor shopping and reduced overdose events and mortality. Opioid adherence was improved and the likelihood of addiction declined (10–30%) as the response rate to SCM was increased. There is the potential for significant incentives ($2267–$3237) to be offered for addressing severe OUD.

Discussion

The framework and designed interventions adapt to changing needs and conditions of the patients to become an important part of global efforts in preventing OUD. To the best of our knowledge, this is the first paper on adaptive interventions for preventing OUD by addressing both prescription and consumption.

Conclusion

SPM and SCM improved opioid prescription and consumption while reducing the risk of opioid addiction. These interventions will assist in better prescription decisions and in managing opioid consumption leading to desirable outcomes. The interventions can be extended to other substance use disorders and to study complex scenarios of prescription and nonprescription opioids in clinical studies.

INTRODUCTION

An increased incidence of chronic pain, life stressors, and access to opioids along with comorbidities and past substance use^1,2 are leading to Opioid Epidemic (OE). Approximately 3 million people in the United States currently suffer from opioid use disorder (OUD)³ and the opioid overdose deaths in the United States were estimated to be 107 622 during 2021.⁴ OUD, an established medical condition, involves the intentional use of opioids beyond pain relief.^5,6 This results in significant harm to patients,⁷ major healthcare expenses, and a decline in the quality of life depending on mild, moderate, or severe disorder.⁸ Most of the OUD patients will require expensive in-patient treatment and may experience opioid overdose leading to life-threatening complications.^9,10

Prescription problems (inaccurate or unsuitable or multiple) can contribute to onset of OUD.¹¹ Consumption problems (both irregular and/or excessive) can contribute to OUD overtime.¹² An unsuitable prescription can encourage the patient for poor consumption or abuse of opioids. We have identified key factors in opioid prescription and consumption.^6,12–17 These are categorized into patient, system, and healthcare professional (HP) factors, which individually as well as a group, impact opioid prescription, opioid consumption, and outcomes (Figure 1).

Figure 1.

Contextual representation of opioid prescription and consumption factors.

Patient factors, including past substance abuse, stressors, condition (pain, surgery, or terminal illness), and comorbidities (mental illnesses), influence opioid risk score in prescription decision.^18,19 The patient condition can lead to opioid prescription and consumption before surgery which is known to have higher chance of OUD over time.^12,15 Doctor shopping is influenced by patient behavior,²⁰ drug regulatory environment, lack of reliability, accessibility or limited use of Prescription Drug Monitoring Program (PDMP), and nonspecialists writing prescription.^16–18,21

The system factors include PDMP¹⁶ with primarily intrastate coverage, PMPInterconnect with limited interstate coverage,²² and EHR with access to comprehensive history and comorbidities. National coverage of prescription monitoring as well as increased use of prescription data will improve the prescription quality. Regulations and access to illicit opioids, treated here as system factors, will affect opioid prescription and consumption, respectively.^16,23

The HP factors, including opioid training, prescription guidelines, and use of nonspecialists for opioid prescription, can affect prescription quality and the number of prescriptions a patient may acquire (doctor shopping).^21,24 HP factors, specifically prescription guidelines along with regulations can change the opioid context used in the prescription decision. Using the current literature and prior experience in substance abuse research, we identify and relate contextual factors for opioid prescription and consumption. As additional factors, contextual role, and relationships are identified, Figure 1 can be easily expanded.

There is a critical need for adaptive interventions that can act proactively before patients develop OUD.^7,25,26 These interventions can help in managing both chronic and acute components^1,2,7,27 of OUD. We focus on adaptive interventions defined as “interventions that rely on information from multiple sources, generate the context of use, and then use the contextualized information in personalized decision making and actions.”^28–31 Although there has been considerable progress in addressing opioid addiction, treatment, and recovery, not much has been done on interventions to prevent OUD. Existing studies only focus on either prescription³² or consumption.^33,34To address this, we have identified and combined opioid prescription and opioid consumption as the most promising targets for adaptive interventions to prevent OUD. We have designed the theory-based adaptive interventions followed by the development of an analytical model and use of secondary data from multiple sources to evaluate the interventions.

OBJECTIVE

The extant literature lacks in frameworks, guidelines, or methods for designing adaptive interventions for OUD. To address this, we focus on: (1) How to design adaptive interventions for opioid prescription and consumption? (2) To what extent, adaptive interventions improve opioid prescription and consumption? and (3) To what extent, incentives can be offered for better prescription and consumption outcomes? We developed a framework to design adaptive interventions, including Smart Prescription Management (SPM) to support HPs and Smart Consumption Monitoring (SCM) to support patients. The integration of SPM with EHR, although challenging, will lead to improvement in decision-making by HPs^35,36 by reducing cognitive load and decision time.³⁷Figure 2 presents the comparison of existing prescription management^5,38 and consumption monitoring^25,39 approaches with the proposed adaptive interventions. SPM and SCM add several functions for improving prescription management and consumption monitoring. Although SPM or SCM can be used alone, the best option is to use them together as prescription and consumption can interact and can jointly improve the patient outcome.

Figure 2.

Infrastructure support for opioid prescription decision-making.

The adaptive interventions are operationalized and evaluated using analytical modeling and secondary data related to doctor shopping,^1,2 opioid overdose mortality (OOM)⁴⁰ and overdose events,⁴¹ prescription quality and monitoring accuracy,^17,42–44 and the cost components of interventions.^45–48 We employ multiple metrics that also address the need for objective measures for opioids.⁴⁹ Our analysis shows that SPM improves the quality of prescriptions and reduces doctor shopping. As part of consumption monitoring, SCM improves adherence. The side effects, where the patient consumes opioids with less than medically safe time gap, are also much lower for SCM reducing the future risk of opioid addiction. To the best of our knowledge, this is the first paper that focuses on adaptive interventions for preventing OUD by addressing both prescription and consumption.

MATERIALS AND METHODS

Literature review

We used the PRISMA approach⁵⁰ to conduct a systematic literature review for adaptive interventions for opioids followed by an additional literature search on opioid interventions and the available opioid apps. The terms employed for PRISMA were “(Adaptive OR Smart OR Context-Aware) AND (Intervention OR Framework OR Method) AND Opioid AND (Prescription OR Consumption OR Monitoring)” in the abstract search in Web of Science, EBSCOHost, JSTOR, PubMed, IEEE Xplore, and AIS eLibrary. In fall 2022, we found 77 articles (Figure 3). Thirty-three duplicate and redundant articles were excluded in the screening. Out of the remaining 44 articles, 38 articles were excluded as they did not focus on adaptive or smart or context-aware interventions for opioids. This resulted in selection and inclusion of 6 articles on adaptive or smart or context-aware interventions for opioids for detailed analysis (Table 1).

Figure 3.

PRISMA approach for systematic literature review.

Table 1.

Analyzing studies on adaptive interventions

Article	Theme	Contribution	Intervention target	Limitations
32	Smart prescription for inflammatory bowel disease (IBD) in emergency department	Multimodal quality improvement intervention	Healthcare professionals (prescription)	Retrospective cohort study No method for designing adaptive or smart interventions
33	Smartphone-based intervention for opioids	Managing buprenorphine pills (a maintenance opioid)	Patients (consumption)	Exploring patient interest for consumption monitoring No method for designing adaptive or smart interventions
34	Smart intervention for predicting/monitoring drug craving	Smartphone app to track heroin craving, cocaine craving, and stress	Patients (consumption)	Random Forest to predict drug craving 90 min into the future No method for designing adaptive or smart interventions
25	Mobile health-based interventions for the Opioid Epidemic	M-health as an intervention for the Opioid Epidemic	Healthcare professionals (prescription) and patients (consumption)	Preliminary intervention with no evaluation No method for designing adaptive or smart interventions
51	Two behavioral economics-informed prescribing interventions (nonopioid and tapering) using EHR	Adaptive intervention proposed for opioid prescription for new and existing patients	Healthcare professionals (prescription)	Preliminary interventions with proposed randomized control trial (RCT) No method for designing adaptive or smart interventions
52	Smart home sensors to monitor opioid withdrawal symptoms	Smart home technology can lead to adaptive interventions for prescribed opioids	Patients (consumption)	Preliminary evaluation to show promise No method for designing adaptive or smart interventions

For supplementing the above literature, we found additional articles in a follow-up literature search on prescription, consumption, or opioids. The prescription literature, not adaptive or smart or context-aware interventions, included state monitoring for prescriptions,⁵³ machine learning to analyze opioid prescription,⁵⁴ and opioid stewardship by pharmacists.⁵⁵ The consumption literature included monitoring for opioid-related adverse drug events for patients in a hospital,⁵⁶ opioid monitoring in the waste water for cities,⁵⁷ and different ways of opioid consumption but not using adaptive interventions for opioid consumption monitoring.⁵⁸ The literature on general opioids included emergency overdose treatment, mindfulness-based interventions for opioid addiction,⁵⁹ in-patient detoxification, outpatient treatment after detoxification, and emergency response for overdose and Samaritan response to opioid overdose.⁶⁰ In addition, an article addresses adaptive intervention for recovery⁶¹ but not for opioid prescription and consumption.

A recent article in a comprehensive review and classification of 153 opioid apps found 24 apps for prescription and 20 for reminder-monitoring-support.⁶² These apps were designed for either prescription or consumption and were not context-aware, which reflected in the limited reach and impact (downloads). This study recommended the use of context-aware operation for prescription and consumption.⁶²

Theoretical support

OUD offers complex and multifaceted challenges which cannot be supported by the generic theoretical approaches that have been utilized for healthcare in general.^63–66 To address this, we explore the underlying behavioral factors and theories to create a theoretical foundation for OUD. There are several factors^67,68 affecting the opioid use behavior, some within and some outside the patient’s control. The effort to identify these factors leads to 3 interacting dimensions (Figure 4) of the prescription and consumption behavior: patient factors, system factors, and HP factors.⁶⁹ As behavior contributes to the cause of current mortality and morbidity,⁷⁰ interventions to change opioid use behavior are essential.⁴⁸ As shown in Figure 4, we developed a generalized model based on behavior change techniques (BCTs) affecting opioid use behavior.⁷¹ When a prescribed opioid is self-administered, the choice rests with the patients and their motivation to take the opioid. An intervention is a mechanism to modify the opioid consumption behavior of the patient. The design of theory-based interventions for practice can lead to improved opioid prescription and consumption outcomes.

Figure 4.

A theoretical model for opioid prescription and consumption.

We examine the OUD problem through the lens of health behavior change theories⁷² to predict “when” and “what” intervention (cues to action/maintenance) is required to improve the opioid use behavior of patients, thereby making the intervention “adaptive intervention.” We also identify Effective Opioid Adherence (EOA) as a “Goal” for the patient. The simplest of BCTs that find prevalence in system design are interventions, a component of both the Stages-of-Change Model (SoCM)⁷³ and the Health Belief Model (HBM).⁷⁴ SoCM assists the patient in developing action plans for adhering to the prescribed opioids and helps in setting gradual goals to maintain the dosing regimen. HBM provides “cues to action” to activate the “readiness” of the patient to receive intervention while monitoring the prescribed opioids. Furthermore, Theory of Planned Behavior (TPB)⁷⁵ and Theory of Reasoned Action (TRA)⁷⁶ support setting goals and steering of the patient toward the goal. TPB and TRA emphasize on the perceived likelihood of performing the behavior which is the prerequisite to identify how much effort a patient is planning to exert to reach the goal. On the other hand, Cognitive Load Theory (CLT)⁷⁷ supports HPs to set cues for action for better decision-making and TPB and TRA support HPs to set goals for the patients. The patient’s opioid use behavior is a dynamic state,⁴⁸ due to the interventions, and continuously feeds to the knowledge base of interacting dimensions using the decision support system.

The specific requirements for adaptive interventions for opioid prescription and consumption are:

• Requirement 1 (HPs): Decision support for prescription, contextual information for decision making, opioid guidelines and HP training, incentives, and desirable outcomes.

• Requirement 2 (patients): Opioid adherence, opioid informational support, support for cognitive behavior therapy, and reducing risks for addiction, overdose, and mortality.

The proposed design framework will lead to interventions that are dynamic, motivate HP and patients to continue the intervention, and implement multiple smart functions including prescription support, consumption monitoring, and context awareness.

The design framework

The design of adaptive interventions starts with the identification of the system factors, patient factors, and HP factors (Figure 5). The interventions, single or composite, can be implemented as a suite of m-health apps. In addition to analytics, the designed interventions are personalized to improve suitability to different patients with mild, moderate, and severe cases of OUD and reduce the overall cost. In addition to being personalized, predictive, preventive, and participatory,^78,79 the designed interventions can be powerful (effective interventions). Personalized refers to matching the user needs and is performed through the use of context awareness as shown in Figure 5. Predictive implies that the integration of information from multiple sources in interventions can lead to predictive outcomes for the patients. As shown in Figure 5, both prescription management and consumption monitoring are integrating information from multiple sources for better decision-making. Preventive implies that by carefully managing prescription followed by consumption monitoring can prevent undesirable outcomes such as addiction. As shown in Figure 5, a better prescription management would match patients’ needs to the opioid prescription considering the addiction risk. In addition, consumption monitoring would prevent excessive opioid consumption by monitoring and incentives (Figure 5). Participatory implies HPs and patients are involved in the intervention. More specifically, HPs play a role in prescription management and patients are key participants in consumption monitoring managed by HP. Interventions are considered powerful (effective) as they help the patients deal with pain and other conditions without increasing the risk of opioid addiction with or without incentives. This condition is met by interventions being personalized, predictive, preventive, and participatory.^78,79

Figure 5.

The design framework for adaptive interventions.

Toward the adaptive interventions, the use of context awareness leads to improved decisions by HPs.⁴⁷ The context information is presented to assist and not replace medical decision-making by HPs.^80,81 Therefore, if the context is incorrect, the HP requests additional information.⁸² The context is implemented as facts and rules derived automatically or modified by HPs. This allows extraction of high-level context from low-level contexts such as dose consumed and not consumed from smart medication box and missing information. The interventions are implemented using both simple and sophisticated m-health apps, sensors, mobile devices, and smart medication boxes.

As part of the design framework, we developed several metrics to measure risk and prescription accuracy, opioid adherence, and the likelihood of future addiction. The goal is to keep opioid adherence between 80% and 100%: less than 80% may not be effective⁸³ leading to the need for stronger opioids. Taking opioids doses too closely is an indicator of potential abuse and future addiction. Table 2 shows the specific metrics, first 3 for prescription and next 3 for consumption.

Table 2.

The metrics for evaluating adaptive interventions

Metric	Notation in analytical model	Description
Opioid Risk Score	ORS	Represents the level of risk using patient history (substance use and any comorbidities)
Average number of prescriptions	N AvgPresc	Compares the decision effectiveness of interventions in number of prescriptions generated under varying levels of doctor shopping
Prescription quality	PQ	Using data from EHR, PDMP, and other source of information, it represents the accuracy of opioid prescription process
Effective Opioid Adherence	EOA	Represents the fraction of opioid doses taken on time
Abnormal Opioid Adherence	AOA	Probability that opioid doses are taken too closely
Future Opioid Abuse	FOA	Predicts if the patient will develop OUD in near future using f1(AOA)+f2(1−EOA)

Analytical model

In this section, we develop an analytical model for evaluating adaptive interventions for OUD. The model is appropriate⁸⁴ for studying OUD as chronic illness where opioids are used over an extended period. Prescription quantity and quality, and opioid consumption can be derived by the model. The model validation⁸⁵ is performed using verification, sensitivity analysis, and evaluation. Verification was done by step-by-step checking and debugging of the model.^84,85 The known relationships were utilized and integrated in the model.^86–91 Using the guidelines⁸⁵ and multiple methods,⁹² nominal range sensitivity was employed. We utilized secondary data for ensuring the adequacy or robustness.^84,85 The input data and results from various sources^86–91 were utilized for accuracy measurement. More specifically, we utilized the hospitalization rate and the emergency room visit rate⁹³; length of hospital stays, the daily cost of hospital stays, the cost of outpatient visits, and frequency of outpatient visits⁸⁸; cost of emergency room visits⁸⁶; cost and fraction of brand name medications and generic medications.⁸⁷ We adjusted the data to reflect current costs.⁹⁴ So, our results on opioid prescription, consumption, and healthcare cost are validated using published data,^86–91 while results on cost of interventions^45–48 are extrapolated based on known relationships and available data from multiple studies. We made 2 key assumptions to develop a tractable and accurate model.⁸⁴ These include: (1) patients are in independent living and can take opioids by themselves and (2) the intervention cost can be amortized over multiple patients.

Table 3 presents lemmas as the foundations of the analytical model and propositions as comparative results. We derive lemmas on a specific aspect of an intervention, such as prescription quality for SPM or Future Opioid Abuse for SCM. The corresponding proofs of lemmas and propositions are presented in Appendix A.

Table 3.

Lemmas and propositions for prescription management and consumption monitoring

Lemmas	Propositions
Lemma 1: For basic prescription management (BPM) with no PDMP access, doctor shopping is expressed as the average number of prescriptions obtained as $N_{\mathrm{AvgPresc}}={\sum }_{I=1}^{N}I\times \left(\genfrac{}{}{0pt}{}{N}{I}\right){\mathrm{PP}}^I{(1-\mathrm{PP})}^{N-I}$ where N is the number of HPs tried by the patient and PP is the prescription probability for BPM.	Proposition 1: SPM will lead to lower level of doctor shopping than BPM or MPM. Furthermore, SPM will lead to fewer opioid overdose events and opioid overdose mortality.
Lemma 2: For prescription management using monitored prescription management (MPM), doctor shopping is expressed as $N_{\mathrm{AvgPresc}}={\sum }_{I=1}^{N}I\times \left(\genfrac{}{}{0pt}{}{N}{I}\right){\mathrm{PP}}^I{(1-\mathrm{PP})}^{N-I}$ where PP is the prescription probability for MPM.
Lemma 3: For Smart Prescription Management (SPM), doctor shopping is expressed as $N_{\mathrm{AvgPresc}}={\sum }_{I=1}^{N}I\times \left(\genfrac{}{}{0pt}{}{N}{I}\right){\mathrm{PP}}^I{(1-\mathrm{PP})}^{N-I}$ where PP is the prescription probability for SPM.
Lemma 4: For BPM with no PDMP access, prescription quality is PQ = 1−(ORSEST-HIGH−ORSEST-LOW)	Proposition 2: SPM will lead to higher quality of prescriptions than basic prescription management or monitored prescription management.
Lemma 5: For prescription management using MPM, prescription quality is expressed as PQ = 1−(ORSEST-HIGH−ORSEST-LOW)
Lemma 6: For Smart Prescription Management (SPM), the prescription quality can be given as PQ = 1−(ORSEST-HIGH−ORSEST-LOW)
Lemma 7: The AOA with no interventions (BCM) is $(1-{\mathrm{e}}^{-\mathrm{\lambda }{T}_{\mathrm{MIN}}})$, where λ is the opioid dosing rate and TMIN is safe time (minimum time between opioid doses for medical safety), and the Effective Opioid Adherence (EOA) with no interventions is given by $\mathrm{EO}{\mathrm{A}}_{\mathrm{BCM}}=\left({\mathrm{e}}^{-\mathrm{\lambda }{T}_{\mathrm{MIN}}}\right)\mathrm{ }\mathrm{x}\mathrm{ }(1-{\mathrm{e}}^{-\mathrm{\lambda }{(T_{\mathrm{MAX}}-T}_{\mathrm{MIN}})})$.	Proposition 3: Observed intervention will generate higher AOA than SCM, and furthermore, SCM will achieve higher EOA than observed interventions.
Lemma 8: For observed interventions (OCM), AOA can be given as $\mathrm{ }\left(1-{\mathrm{e}}^{-\mathrm{\lambda }{T}_{\mathrm{MIN}}}\right)+\left({\mathrm{e}}^{-\mathrm{\lambda }{T}_{\mathrm{MIN}}}\right)\mathrm{ }\times \mathrm{ }\frac{T_{\mathrm{MIN}}}{T_{\mathrm{MAX}}}\times P_{\mathrm{MI}}$, where POCM represents the probability of taking an opioid dose in response to observed intervention, and Effective Opioid Adherence (EOA) is given by $\mathrm{ }\mathrm{EO}{\mathrm{A}}_{\mathrm{OCM}}=\left(\left({\mathrm{e}}^{-\mathrm{\lambda }{T}_{\mathrm{MIN}}}\right)\times \left\{1-\frac{T_{\mathrm{MIN}}}{T_{\mathrm{MAX}}}\times P_{\mathrm{OCM}}\right\}\right)\times \mathrm{ }\left[\right\{1-{\mathrm{e}}^{-\mathrm{\lambda }{(T_{\mathrm{MAX}}-T}_{\mathrm{MIN}})}\}+\{{\mathrm{e}}^{-\mathrm{\lambda }{(T_{\mathrm{MAX}}-T}_{\mathrm{MIN}})}\times \frac{T_{\mathrm{MAX}}-T_{\mathrm{MIN}}}{T_{\mathrm{MAX}}}\mathrm{ }\times P_{\mathrm{OCM}}\left\}\right]$, where POCM is the probability of opioid dose consumption for observed intervention.
Lemma 9: For Smart Consumption Monitoring (SCM), AOA can be given as $\left(1-{\mathrm{e}}^{-\mathrm{\lambda }{\mathrm{T}}_{\mathrm{MIN}}}\right)$ and EOA is given as $\mathrm{EO}{\mathrm{A}}_{\mathrm{SCM}}=\mathrm{ }{\mathrm{e}}^{-\mathrm{\lambda }{\mathrm{T}}_{\mathrm{MIN}}}\mathrm{ }\mathrm{\times }\mathrm{ }\left[\right\{1-{\mathrm{e}}^{-\mathrm{\lambda }{({\mathrm{T}}_{\mathrm{MAX}}-\mathrm{T}}_{\mathrm{MIN}})}\}+\{{\mathrm{e}}^{-\mathrm{\lambda }{({\mathrm{T}}_{\mathrm{MAX}}-\mathrm{T}}_{\mathrm{MIN}})}\times \left(1-{\left(1-{\mathrm{P}}_{\mathrm{SCM}}\right)}^{\mathrm{M}}\right)\left\}\right]$
Lemma 10: Total cost of an intervention (TCI) can be given as $\mathrm{TCI}={\sum }_{I1=1}^{\mathrm{T}{\mathrm{P}}_{\mathrm{HOUR}}}{\mathrm{CP}}_{\mathrm{HOUR}-\mathrm{I}1}+{\sum }_{I2=1}^{\mathrm{T}{\mathrm{F}}_{\mathrm{HOUR}}}{\mathrm{CP}}_{\mathrm{HOUR}-\mathrm{I}2}+\left(\frac{\mathrm{Cos}{\mathrm{t}}_{\mathrm{FIX}}}{\mathrm{NP}\mathrm{*}\mathrm{NYRINTV}}\right)+\left(\frac{\mathrm{Cos}{\mathrm{t}}_{\mathrm{VAR}}}{\mathrm{NP}}\right)$ Where TPHOUR is the total time spent per year by healthcare professionals and CPHOUR-I1 is the cost of I1th hour for healthcare professionals. TFHOUR and CFHOUR-I2 represent the same factors for a family member.	Proposition 4: For certain effectiveness, the smart consumption and prescription interventions can lead to incentives for the patients and healthcare professionals and are cost effective.
Lemma 11: Healthcare savings due to proposed intervention can be expressed as $\mathrm{H}{\mathrm{C}}_{\mathrm{Savings}-\mathrm{I}}=\left[\mathrm{H}{\mathrm{C}}_{\mathrm{CostNoINTV}-\mathrm{I}}-\mathrm{H}{\mathrm{C}}_{\mathrm{CostINTV}-\mathrm{I}}\right]\mathrm{ }[1-{\left(D\right)}^{\mathrm{I}}]$

RESULTS

SPM, designed for HPs, uses Opioid Risk Score (ORS) based on patients’ history and comorbidities from EHR, prescription data from PDMP, and consumption data from sensors, smart medication box, and mobile apps as part of adaptive operations. In addition, data from insurance payers, community opioid usage, and/or social media can be included in ORS. SCM, designed for patients, includes context-aware reminders, sensing of opioid consumption, and a suitable action from HPs as part of adaptive operations.

Smart Prescription Management

In SPM, decision-making for opioid prescriptions is improved by utilizing the information from multiple sources as part of adaptive intervention. This includes information from PDMP and PMPInterconnect^22,95 from 42 of 51 US states, evidence-based medicine,⁹⁶ and nonopioids for high-risk patients.⁹⁷ Cognitive load^37,98,99 is managed by integrating multiple informational items to Opioid Context (Figure 6A). Use of an objective measure, ORS, facilitates objective decision-making by overcoming biases, pressure from patients and/or their families, and any incentives for HPs. ORS is integrated into the EHR for improving the quality of prescription decisions. Accuracy, reliability, and completeness of PDMP, EHR, CDS, context manager, and context analyzer can affect the accuracy of decision-making for prescription and consumption (Figure 6A). A prescription management app (SPM-App) implements the smart functions of SPM (Figure 6B).

Figure 6.

(A) Smart Prescription Management (SPM) and Smart Consumption Monitoring (SCM). (B) Mobile app implementation of SPM. (C) Mobile app implementation of SCM.

Smart Consumption Monitoring

SCM involves opioid dose reminders, consumption monitoring, and suitable action by HPs (Figure 6A). The context-aware reminders, generated only if the patient has not taken and it is still safe to take the prescribed dose, are facilitated by communicating with smart medication box, EHR, and self-reporting through mobile health apps. For certain opioids, the reminders include a message to not take the opioid until the next reminder. The consumption monitoring is done by sensors, devices, and m-health apps connected with the EHRs. The integrated information is analyzed, and HP will intervene if opioid doses are taken too closely or more frequently than prescribed. This “monitor and intervene” by HPs is part of adaptive operation (proactive), is less expensive, and efficiently utilizes resources.

The analysis of consumption patterns can also predict long-term outcomes using machine learning and AI-based tools. This information is communicated to the patient as a form of a text message or displayed on the mobile app including any suggestions from an HP. SCM includes the processing of opioid consumption using: (1) explicit instructions on observing dose consumption and deriving EOA, AOA, and FOA, (2) known patterns of both desirable and undesirable for opioid adherence, and (3) the algorithm receiving hints or instructions on what else to look for to detect any abnormal pattern. A consumption monitoring app (SCM-App) implements the adaptive functions of SCM (Figure 6C).

The algorithms for SPM and SCM are presented in Table 4. The algorithms are designed to cover all major steps from Figure 6A. A step can be expanded to accommodate new functions in SPM and SCM in future. For example, SCM can be extended by integrating point-of-care using sensors to detect illicit drug consumption,¹⁰⁰ and prescription management and consumption monitoring can be extended by integrating blockchain based models for managing doctor shopping.¹⁰¹

Table 4.

The algorithms for SPM and SCM

Smart Prescription Management (SPM)
Step P1: Initialization of all systems
Run Initial-system
Download Prescription Guidelines-Evidence-Based Medicine, State Regulations for Opioid Contextual Information
Estimate HP-Expertise-Level
If HP-Expertise-Level<Threshold-Low Suggest HP-Training-Opioid
Step P2: Information Integration and ORS
If Patient-Condition (Pain: Objective-&-Subjective, Surgeries≥Threshold-Condition)
Receive EHR History (Substance Use, Past Opioid Use, Surgeries, Pain Conditions)
Check for Comorbidities (Mental Illness)
Else Consider Nonopioid-Prescription Go to Step P4
If EHR-Status=Not-Available
Use Last-Known-Values
Add Additional Risk Factors
Request PDMP Information (PDMP, PMPInterconnect)
Else Opioid-Prescription=NO, Nonopioid-Prescription=YES
If (Doctor-Shopping-Level≥Threshold-Num) OR (Payment=Cash) OR (Previous-Prescribers=Nonspecialist)
Opioid-Prescription=NO
Offer Edu-Intervention, CBT and/or Incentives-Treatment
Else Go to Step P4
Extract and Integrate [Past Substance Use, Comorbidities, Opioid Risk Factors (Prior MAT)]
ORS=Max [PEHR (WUSEHISTORY×PUSEHISTORY+WCOMORBIDITY×PCOMORBIDITY+WCONTEXT×POPIOIDCONTEXT), (PPDMPACCESS×PPDMPENTRY)]
If Stressors>Threshold-High Add Stress-Weight to ORS
Step P3: Prescription Decision
If PDMP=Not-Available
PDMP-Entry=Last-Known-Value
If (ORS<Risk-Threshold) Opioid-Prescription=YES
Else Opioid-Prescription=NO
Step P4: Postprocessing
Notify PDMP and PMPInterconnect
Prescribe Companion Intervention/App
Update Prescription Accuracy/Errors, Prescription Quality, Update HP-Expertise-Level
Offer Incentives-HP (Educational-Training, Continuing Medical Education)
Smart Consumption Monitoring (SCM)
Step C1: Initialization of all systems
Run Initial-system
Receive Monitoring-Criteria, Action-Thresholds, Desirable Consumption Patterns (DCPs), and Undesirable Consumption Patterns (UCPs)
Step C2: Reminders and Monitoring
Send Contextual Reminders (Dose, Timing, Gap)
Monitor Opioid-Consumption (Smart Med Box, Sensor, Mobile App)
Integrate Data-Other-Sources (Multiple SMB, Other Meds: Prescribed or illicit)
Step C3: FOA Computation and Opioid Context
Compute EOA and AOA
FOA=W×(1−EOA)+(1−W)×AOA
Update Opioid Context (FOA, Consumption, Data-Other-Sources)
Step C4: Adherence Decision
If (FOA>Threshold) OR (UCP=YES)
Notify HP, Receive and Execute HP-Instruction
If HP-Instruction=CBT/Edu-INTV
Provide CBT/Edu-INTV to the patient
Else Incentives-Good-Adherence
Step C5: Postmonitoring-Processing

Notify Caregiver/Family and Update Consumption and Context.

Evaluation of intervention

In this section, we discuss the results using secondary data, lemmas, and propositions. We estimate the performance of adaptive interventions under different scenarios. Using the minimal (N = 2), moderate (N = 3–5), and extensive (N = 6–30) doctors visited for opioid prescription by 15% of the opioid patients (10% extensive, 30% moderate, and 60% minimal), we constructed scenario 1 as baseline with no intervention.^1,2 Scenarios 2–4 correspond to BPM, MPM, and SPM. Figure 7A shows the average number of prescriptions based on the level of doctor shopping. The prescriptions are dependent on prescription probability and number of doctors seen by the patient. SPM improves the opioid prescriptions for doctor shopping (minimal, moderate, and extensive) and is most effective for extensive doctor shopping cases as shown by decline from 18 to 1.8 prescriptions per patient (Figure 7A).

Figure 7.

(A) The impact of interventions on average number of prescriptions due to doctor shopping. (B) Average number of prescriptions based on levels of doctor shopping and interventions. (C) Average number of prescriptions, overdose events, emergency room visits, and mortality for BPM, MPM, and SPM.

Since baseline data suffers due to lower level of self-reporting influenced by social desirability status,^102,103 we create more realistic scenarios by varying doctor shopping (minimal, moderate, and extensive) (Figure 7B). The highest decline in prescriptions occur when extensive shoppers form most cases (60%) (Figure 7B). This validates our Proposition 1. Furthermore, SPM cannot stop multiple prescriptions, it reduces the extra prescriptions for high-risk patients. Although BPM and MPM offer similar performance, best performance is still achieved by SPM.

Similar performance for average number of opioid prescriptions is obtained for SPM (Figure 7C). For every 100K of the US population, 51 400 opioid prescriptions, 62 opioid emergency room visits (OERV), and 5.2 OOM are observed⁴⁰ and opioid overdose events range from 227.5 to 266.7 per year.⁴¹ This is used to construct scenario 1 as the baseline (current situation). Then, we evaluated different interventions (scenarios 2–4) in Figure 7C in terms of average number of prescriptions, opioid overdose events, OERV, and OOM. The SPM has the most improvement compared to basic and monitored interventions.

Furthermore, we analyzed the impact of EHR accuracy on prescription quality for BPM, MPM, and SPM. The results with varying EHR accuracy (Table 5) are validated with published results with prescription accuracy from 59%¹⁷ to 75%⁴⁴ improving to 76%¹⁷ to 97%⁴⁴ with interventions. The monitoring accuracy was 63–96% but with adaptive interventions it improved to 90–96%.^42,43

Table 5.

Impact of EHR accuracy on prescription quality

EHR accuracy	Prescription quality
	BPM (EHR only)	MPM (EHR and PDMP)	SPM (EHR, PDMP, and context)
0.6	0.600	0.800	0.900
0.7	0.700	0.850	0.925
0.8	0.800	0.900	0.950
0.9	0.900	0.950	0.975
1.0	1.000	1.000	1.000

Next, we evaluate SCM by deriving EOA, AOA, and FOA. As shown in Figure 8A, AOA increases for basic consumption monitoring (BCM), observed consumption monitoring (OCM), and SCM as the chance of taking 2 opioid doses goes up in a larger time window. EOA declines as the time window is reduced. FOA increases for all cases, as the increase in AOA has more impact than the decrease in EOA toward the value of FOA. EOA, AOA, and FOA for variable effective time for opioids are shown in Figure 8B, where AOA reduces for BCM, OCM, and SCM. EOA increases as expected as the time window for EOA is increased. FOA declines for all cases, as the decrease in AOA has more impact than the increase in EOA. EOA, AOA, and FOA for the variable probability of taking a dose in response to an intervention are shown in Figure 8C. As the probability is increased, AOA increases slightly for BCM, OCM, and both versions of SCM. EOA increases significantly as the response rate to intervention is increased. FOA slightly declines, as the increase in AOA has less impact than the increase in EOA toward the value of FOA. We conclude that for increased safe time (the minimum time gap between 2 opioid doses), FOA increases, while for increased effective time, FOA declines.

Figure 8.

(A) Abnormal Opioid Adherence (AOA), Effective Opioid Adherence (EOA), and Future Opioid Abuse (FOA) under varying safe time. (B) AOA, EOA, and FOA under varying effective time. (C) AOA, EOA, and FOA under varying probability of response to intervention.

Next, the cost components of various interventions along with the secondary data used^45–47 in estimating the total cost of different interventions⁴⁸ are shown in Table 6. The data have been modified to reflect annual cost and then 7% annual increase in healthcare cost⁹⁴ to bring it to 2022 estimates. If the real cost of ICU/hospitalization is higher, our model will produce more favorable results for savings and larger incentives with 2 interventions. The cost for different scenarios and maximum incentives with interventions (Figure 9) can help decision-makers when to use SPM only, SCM only, or both SPM and SCM together. We observed that significant incentives ($2267–$3237) can be offered for severe OUD levels due to large potential healthcare savings (Figure 9).

Table 6.

Parameters and sources for healthcare cost

Input parameters	Data source	OUD average	The level of OUD
			Mild	Moderate	Severe
The emergency room visit rate	93	0.094/year	0.02	0.08	0.32
Cost of emergency room visits	86	$2634	$2634	$2634	$2634
The hospitalization rate	93	0.043/year	0.01	0.04	0.16
Length of hospital stay	88	4.35 days	2	4	8
The daily cost of hospital stays	88	$2038/day	$2038	$2038	$2038
The cost of outpatient visits	88	$561	$561	$561	$561
Cost of brand name medication	87	$8670	$8670	$8670	$8670
Cost of generic medication	87	$847	$847	$847	$847
Brand name medications	87	6%	6%	6%	6%
Generic medications	87	94%	94%	94%	94%
Total intervention cost (SPM)	48	$560
		$700
		$933
Total intervention cost (SCM)	48	$750
		$970
		$1190
Total intervention cost (SPM and SCM)	48	$1530
		$1670
		$1903

Figure 9.

Healthcare savings based on incentives for different interventions and levels of opioid use disorder (OUD).

DISCUSSION

This study addresses one of the complex and persistent healthcare challenges, OE. Since access to opioids and consumption are important and potentially interacting factors in preventing OUD, we focused on adaptive interventions for improving both opioid prescription and consumption. We present a detailed framework to design adaptive interventions for opioid prescription and consumption. The primary goal of SPM and SCM is to provide prescription to all patients who need opioids and smart monitoring to reduce their risk of addiction, overdose, and mortality. Further, the patients with minimal and moderate shopping can be provided with educational support and incentives, while those in extensive doctor shopping can be provided with treatment information and incentives for reducing the future risk of addiction. These interventions can be utilized separately or can work in conjunction to prevent OUD.

Our analysis shows, by improving the accuracy of prescription, SPM minimized the effect of doctor shopping by 30–90% in average number of prescriptions. By analyzing the consumption, SCM improved the opioid adherence and reduced the addiction risk by 10–30%. Using secondary data, we evaluated scenarios with incentives for patients for opioid adherence and treatment, if needed, and to the HPs for higher quality prescriptions. Even though we used secondary data from multiple sources to evaluate the interventions, one limitation is the use of variety of newer and older data sources due to limited data availability. To address this limitation, future research should use more recent data, as it become available, for prospective validation. This may change the absolute values in results, but we are confident that there will not be much deviation in patterns of results and general observations.

To the best of our knowledge, this is the first paper that focuses on adaptive interventions for preventing OUD by addressing both prescription and consumption. Toward the future work, based on the literature and the analysis presented in the study, the relationships among various factors are identified in Figure 10. The work could focus on high-level exploration, such as how patient factors affect consumption and outcomes, or more specific connection, such as how nonspecialist can improve the quality of opioid prescription using training and specific opioid guidelines.

Figure 10.

Possible relationships among opioid factors and stakeholders.

CONCLUSION

The major contributions include: (1) the framework to design adaptive interventions, (2) design of 2 adaptive interventions, and (3) validation and evaluation using analytical modeling and secondary data. This can assist HPs in better prescription decisions and patients in managing opioid consumption leading to desirable outcomes. SPM minimized the effect of doctor shopping by 30–90% (in one case reducing number of prescriptions from 18 to 1.8 per patient) in average number of prescriptions. SCM improved the opioid adherence and reduced the addiction risk by 10–30%. There is the potential for significant incentives ($2267–$3237 or 20–30% higher) to be offered for addressing severe OUD. Higher improvement can be achieved by combining SPM and SCM.

Supporting innovation in healthcare while preserving privacy and security of patients is challenging.¹⁰⁴ In this study, SCM needs to collect consumption information for decision-making. Patient consent will be required for SCM and if SPM and SCM are used together. For opioid monitoring, patient can offer single use, multiuse, or unrestricted use of monitored information. Patients and HP should have secure access to their mobile device for using apps and accessing PHI. The monitored information can be anonymized or deidentified for research purposes.¹⁰⁵ For SPM and SCM, patient’s monitoring and consumption information must be protected and encrypted before transmitting to the HPs. For interoperability, the mobile apps will securely interact with EHR, and will not store any PHI.¹⁰⁵ Any information sensed and processed by monitoring system will be deleted after transmission to avoid future disclosure of PHI.¹⁰⁶ The information need to be stored in an end-to-end encrypted platform and that can be part of the EHR and is under HPs control.¹⁰⁴ If the patient indicates, the monitored information can be integrated with Personal Health Record (PHR) and mobile app will not commercialize any patient data.¹⁰⁵

Some interesting questions for exploration are: (1) if the proposed interventions increase the workload of the HP initially and how SPM reduces it over time while improving the prescription quality and (2) trade-off between barriers for patients and the need to reduce the risk of addiction under SPM and SCM. To the best of our knowledge, this is the first paper on adaptive interventions for preventing OUD by addressing both prescription and consumption. The interventions can lead to major breakthroughs and transform healthcare for opioid globally. The work can be easily extended/adapted to current and future substance use for both nonopioids and opioids obtained from prescriptions or from other sources. The modeling and results have significant implications for HPs, patients, insurance companies, and health IT researchers for improving opioid prescription, consumption, and outcomes.

FUNDING

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

AUTHOR CONTRIBUTIONS

All authors made significant contributions to the conception, design, and development of the framework for SPM and SCM. All authors worked on developing the analytical model and evaluation using secondary data. In addition, they contributed to developing, finalizing, and approving the final version submitted for publication.

SUPPLEMENTARY MATERIAL

Supplementary material is available at Journal of the American Medical Informatics Association online.

CONFLICT OF INTEREST STATEMENT

None declared.

DATA AVAILABILITY

The data underlying this article are available in the article.