This section describes the functional behavior of the recommendations as implemented for the chosen pilot sites, and documents the process that was used to establish the functional behavior once the target recommendations were chosen.
Assumptions made in this process were based on team understanding of Epic and Cerner EHR capabilities, as well as team experience with clinical decision support (CDS) implementation in general, and opioid CDS specifically.
Opioids for chronic pain management are often not started with an intent for chronic use; such prescriptions often transition from intended acute pain management. Therefore, it is very difficult to identify a new opioid prescription a priori as for chronic pain management.
Indications and associated diagnoses for opioid prescriptions are often not available within the EHR.
End dates for opioid prescriptions are often not reliable.
The initial focus is on Epic and Cerner EHR platforms.
The technical framework to be used for EHR integration is CDS Hooks, with the FHIR Clinical Reasoning Module to represent the behavioral artifacts.
Widely scaling this type of CDS is still bleeding-edge clinical informatics work, so an iterative and agile approach is recommended.
Both Epic and Cerner will be capable of supporting the provision of the following information:
The degree to which standard vocabularies are used by Epic and Cerner to provide this information must be verified at each implementation site, and to the extent that they are not uniform, terminology mappings will need to be established.
The first issue to establish is how to determine what constitutes an opioid. The CDC guideline uses NDC codes derived from SAS calculations intended for use against prescription data. Prescription data can include ordered medication and/or dispensed medication, depending on availability at the local site.
To avoid excess alerting, the algorithm uses the following exclusions:
To identify opioids, the IG provides a value set containing RxNorm Semantic Clinical Drug (SCD) and Semantic Brand Drug (SBD) concepts which include ingredient(s), strength, route and dose form. The team provides an enumerated (extensional) value set. However, a value set with that approach quickly becomes outdated as RxNorm updates frequently, or the value set must be updated and made available frequently. Therefore, the project team also includes an RxNorm query to retrieve all opioids in the extended-release/long-acting category and in the immediate-release category. Thus, each implementer can update their value sets and run the expressions at frequencies appropriate to the organization. The challenge with using queries is that RxNorm retires concepts for medications for which there is no input activity (e.g., the NDC code is no longer available). Thus, the output of the data retrieved will not include historical drugs for which codes have been retired. The impact of missing retired codes for MME calculations is likely to be minimal to none since most opioid prescriptions have short durations of 30 days or less and codes should be relatively recent.
Medication self-administration data are not available in the ambulatory setting, nor would such data be reliable. Therefore, the most proximal information about a patient’s medication usage is medication dispensed data. Such dispensed data is increasingly accessible from claims data available to insurers and used to evaluate provider medication utilization rates.1 Researchers can access population-based dispensed data from government sources.2 Some provider organizations also have access to dispensed data for patients treated by their clinicians. The project team considered calculating morphine milligram equivalents (MMEs) available to a patient based on medication dispensed data. However, since dispensed data are not available to all provider organizations and since the pilot site did not have such access, the guidance in this implementation guide addresses medication order (prescription) data. Using prescription data presented an issue regarding comprehensiveness of the data set. While dispensed data are generally available as structured data, evaluation of medication order data at the pilot organization showed 20% of opioid prescriptions used free-text (unstructured) instructions with over 10,000 unique patterns. Eliminating such prescriptions from the analysis could introduce disproportionate impact of clinical decision support interventions. The team built a parsing algorithm to create structured information from 80% of the free-text instructions. Clinical review of the parsing results by two, independent clinicians confirmed their accuracy. Therefore, the team included the parsed data in the MME calculations.
For geographic locations with pharmacies that participate in Surescripts, implementation sites have the option to use dispense data to perform the MME calculation. However, coverage varies by region, and even for regions with coverage, licensing costs may be required and must be considered as a factor when determining whether to use dispense data.
Calculation of daily dosage is based on the prescribed (order) or dispensed data based on expected use at the implementation site. The patient may be taking medication less often than prescribed or expected. For example, medication ordered every 4 hours as needed (PRN) leads to a calculation of six doses within a 24 hour period.
The daily dosage calculation is different based on whether dispsensed data is used:
Order Data (total daily dose based on the sig)
Expected use: Total daily dose = quantity per use * strength * daily frequency
Example: Order for two 24 mg tabs every 4 hours = 2 * 24 mg * 6 = 300 mg per day
Dispensed Data (dose not use sig)
Expected use: Total daily dose = (number dispensed * strength) / duration (days)
Example: Dispensed 60 pills, 25 mg each, for 30 days = (60 * 25) / 30, or 50 mg per day
Actual use may be less per day for a longer time period for either calculation, but the MME calculation addresses the prescribed or dispensed data. It is also possible that patients received dispensed medication and failed to ingest some or all of it. The extent to which differences exist between actual use and ordered or dispensed data requires access to reconciled medication data and may be confounded by trust issues.
The National Comittee for Quality Assurance (NCQA) Healthcare Effectiveness Data and Information Set (HEDIS) took a similar approach to determining cumulative medication use over time. The related electronic clinical quality measure (eCQM) CMS 128 addresses continuous use of anti-depressant medication management as 84 to 114 days (short term) and 180 to 231 days (long term) based on active medication and can use prescribed or dispensed data as they are available. The eCQM assumes the medication is dispensed and used as written.
Active medications in an EHR may have a start date and indicate the number of doses written; however, if not discontinued, the algorithm assumes the ordered medication remains active.
The logic assumes the prescription is filled on the day of the order.
If the prescription includes refills, the system assumes the patient requested all refills of the medication.
The team considered how to define the concepts opioid-tolerant and opioid-naïve. Guidelines for pain management in adult cancer pain defined these terms as:3
Opioid tolerant: patients who are taking, for 1 week or longer, at least:
Opioid naive: patients who do not meet the above definition of opioid tolerant, and who have not taken opioid doses at least as much as those listed above for 1 week or longer.
The NCCN v.2010 pain management algorithms also supply simpler definitions for these terms:
The team was concerned that an opioid-naïve patient by the definitions provided could have received opioid prescriptions and potentially have an MME of greater than 50. Informing a physician that a patient with an elevated MME is opioid naïve would lead to confusion. Therefore, the team decided to base recommendations on the MME calculation and to avoid categorizing patients as opioid tolerant or naïve.
The project team considered options to determine that a provider has evaluated benefits and harms of opioid use. The implementer needs to determine how the local clinicians capture such assessments and map the concept to the SNOMED-CT concept: Assessment of risk for opioid abuse (procedure) SCTID: 454281000124100
The team rejected use of the concept At risk for drug therapy problem (finding) (SCTID: 451691000124102) as it was too generic as it could address non-opioid medication therapy concerns. The team also considered searching for use of patient responses to pain scales as evidence that the clinician evaluated patients’ responses to opioid use. However, pain scales have many uses without concomitant use of opioids and merely evaluating pain is not sufficient to assess risk for opioid-related harms. Therefore, the guidance does not include pain scale evaluation as a method for determining the benefits of opioid use.
To address the recommendation statement “optimize other therapies and work with patients to taper opioids to lower dosages or to taper opioids,” implementers should determine what local activities represent such actions and map the activity to the SNOMED-CT concept: High risk drug monitoring (regime/therapy): SCTID: 268525008
The project team considered potential benefits of regularly scheduled queries to identify the cohort of patients who have been given new prescriptions for opioids or prescriptions for escalating doses. Further the analysis of the cohort could also determine those for whom assessment of risk or high-risk drug monitoring had not occurred at the expected intervals, triggering a task list for a clinician role to be determined by the practice. This non-synchronous approach is challenged by the fact that CDS Hooks requires an event such as opening a record or ordering a mediation as a trigger. Since opioid prescriptions have limited number of days supplied, patients on continued therapy have encounters with physicians; therefore, triggering on the entry of a new prescription order allows intervention in a synchronous manner and is equally effective to the asynchronous approach to identify patients requiring evaluation and monitoring.
The following two criteria must be satisfied for every patient for each recommendation:
TODO - the decision points for this criteria are not well defined.
The assessment to determine whether the patient appears to be at end of life consists of three elements:
This section describes rationale and justification for guideline recommendations that are not currently targeted in this implementation guide.
Nonpharmacologic therapy and nonopioid pharmacologic therapy are preferred for chronic pain. Clinicians should consider opioid therapy only if expected benefits for both pain and function are anticipated to outweigh risks to the patient. If opioids are used, they should be combined with nonpharmacologic therapy and nonopioid pharmacologic therapy, as appropriate. (recommendation category: A, evidence type: 3).
Justification for deferring Recommendation #1: The recommendation includes a large number of incompletely defined conditions and non-pharmacologic therapy. Creating definitions and specific recommendations for non-pharmacologic therapy (as orderables) will be complex. Alternatively, on selection of an opioid order, a simple method might merely provide a selectable option linking to a general description of such therapy, but the impact on outcomes would be low.
Before starting opioid therapy for chronic pain, clinicians should establish treatment goals with all patients, including realistic goals for pain and function, and should consider how opioid therapy will be discontinued if benefits do not outweigh risks. Clinicians should continue opioid therapy only if there is clinically meaningful improvement in pain and function that outweighs risks to patient safety. (recommendation category: A, evidence type: 4).
Justification for deferring Recommendation #2: EHRs have different mechanisms for enabling treatment goals, ranging from narrative elements of a visit record, structured care plans, or unstructured visit notes. Therefore, determining that a care goal is established is difficult from a standards perspective. Assuring that such goals are realistic is more problematic. Alternatively, on selection of an opioid order, a simple method is to merely provide guidance (or an option to view such guidance).
Before starting and periodically during opioid therapy, clinicians should discuss with patients known risks and realistic benefits of opioid therapy and patient and clinician responsibilities for managing therapy. (recommendation category: A, evidence type: 3).
Justification for deferring Recommendation #3: This recommendation might include provision of general guidance about risks. Potentially, if defined with value sets or direct referenced codes, opioid-induced risks might be identified with entry of new conditions on the problem list, or population of the record with new laboratory results. The CDS can alert the provider (synchronously or asynchronously) when such a new risk occurs. The number of potential conditions to address is quite large and challenges are similar to those for Recommendation #1.
Long-term opioid use often begins with treatment of acute pain. When opioids are used for acute pain, clinicians should prescribe the lowest effective dose of immediate-release opioids and should prescribe no greater quantity than needed for the expected duration of pain severe enough to require opioids. Three days or less will often be sufficient; more than seven days will rarely be needed. (recommendation category: A, evidence type: 4).
Justification for deferring Recommendation #6: It may be possible to determine a new prescription with no prior treatment in the last x months. Determining acute from chronic pain as the cause for the prescription may be more problematic. CDS can recommend limiting therapy to three days or less with maximum of seven days and restricting orders to immediate-release opioids. However, determining the lowest effective dose based on the individual patient’s condition and expected duration of pain is not trivial and requires individual physician-patient interaction. For the latter item, it is possible to potentially link to the recommendation without trying to define the lowest effective dose by condition.
Clinicians should review the patient’s history of controlled substance prescriptions using state prescription drug monitoring program (PDMP) data to determine whether the patient is receiving opioid dosages or dangerous combinations that put him or her at high risk for overdose. Clinicians should review PDMP data when starting opioid therapy for chronic pain and periodically during opioid therapy for chronic pain, ranging from every prescription to every 3 months. (recommendation category: A, evidence type: 4).
Justification for deferring Recommendation #9: This recommendation may be more problematic unless there is a way to define input from a PDMP as structured data similar to that retrieved with a query to an immunization registry – and associate review and recommendations for ordering with such data. The work also assumes the information from the PDMP is incorporated as structured data in the EHR. Challenges to implementation with potential mitigation for future work:
a. PDMP data are compilations of dispensed data from pharmacies. Such data includes NDC codes, several of which are represented by a single RxNorm ingredient code. Therefore, data from PDMP require terminology mapping to RxNorm ingredient codes to incorporate into medication ordering workflow pathways.
b. Clinical sites generally receive PDMP data in unstructured form, e.g., PDF files. To enable re-use of these data, PDMP output requires standardization to meet needs for semantic interoperability and incorporation directly into clinical software workflow.
c. Note that calculations regarding maximum daily dose of opioids for clinical decision support may provide different results for “as needed” (PRN) medications based on the method for determining daily dose. CDS calculating the maximum daily dose from dosage multiplied by the maximum frequency provides the largest dose potentially consumed by a patient during any of the days covered by the prescription. CDS calculating the maximum daily dose from number of doses (i.e., supply) divided by the days supplied (days covered by the prescription) may lead to a lesser daily total.
Clinicians should offer or arrange evidence-based treatment (usually medication assisted treatment with buprenorphine or methadone in combination with behavioral therapies) for patients with opioid use disorder. (recommendation category: A, evidence type: 2).
Justification for deferring Recommendation #12: Potentially complex for implementation. I suspect this recommendation would require the ability to include a referral order to a clinic or recommendation to use methadone or buprenorphine. But the work may be complicated by what information is available in the EHR with respect to behavioral therapies. Ideally, acknowledgement of the referral and fulfillment (i.e., the patient has been seen and accepted into the program, or is currently receiving one of the therapies from the index physician) would be communicated back to the referring provider and available to the decision support.
1. Centers for Medicare and Medicaid Services. Medicare Provider Utilization and Payment Data: Part D Prescriber. Available at: https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Medicare-Provider-Charge-Data/Part-D-Prescriber.html. Accessed 19 July 2018.↩
2. Data.Gov. Data Catalog: Pharmacy datasets. Available at: https://catalog.data.gov/dataset?tags=pharmacy. Accessed 19 July 2018.↩
3. Stokowski LA. Adult cancer pain: part 2 – the latest guidelines for pain management. Medscape Oncology. 2010. Available at: https://www.medscape.com/viewarticle/733067_2. Accessed 19 July 2018.↩
© CDC 2016+. Opioid Prescribing Support Implementation Guide (v3.0.2-11200) generated on Tue, May 26, 2020 16:37+0000. QA Page
Links: Home |
Version History |
|
Propose a change 