Health system pharmacists help achieve the Triple Aim-improving the experience of care, improving the health of populations, and reducing per capita costs of health care”-by optimizing medication use and ensuring the safe and effective delivery of pharmacotherapy to patients.1
Improved medication use, achieved through the delivery of comprehensive medication management (CMM) by pharmacists,2,3
has been shown to lead to patient satisfaction and trust, reduced medical costs, and improvement in population health.4-6
Thus, pharmacists serve an integral role within the interdisciplinary team to deliver high quality and value health care.
The effective integration of pharmacists into the health care workflow to provide CMM for a primary care population can be challenging. There are far more patients who could benefit from pharmacist review and intervention than there are clinical pharmacists available to perform this work. Furthermore, widely accepted standards or automated mechanisms are lacking to identify patients who would most likely benefit from a pharmacist consultation. Tools to prioritize patients for pharmacist review do exist, such as the clinical pharmacy priority (CP2) score, medication user self-evaluation (MUSE) tool, or assessment of risk tool (ART),7-9
but many require manual chart review by a pharmacist or data collection via patient survey.
Other criteria, such as medication therapy management (MTM) eligibility criteria from the Centers for Medicare & Medicaid Services (CMS) Medicare Part D program (ie, ≥ 2-3 chronic diseases, taking ≥ 3-8 Part D drugs, annual costs for Part D drugs ≥ $4,044—in 2020),10
can be used to identify patients for a pharmacist program but may result in a list of patients that is very large, broad, and not always actionable. Therefore, using these measures to create a priority list for pharmacists often requires a manual, time-consuming process to apply additional criteria such as laboratory data, health care utilization, or other clinical indicators.11-13
This process likely varies across providers and settings and may not identify those with the greatest need or patients most likely to benefit from intervention.
The purpose of this study was to describe the process for identifying CMM-related medication-regimen characteristics and develop a tool (referred to internally as the medication management score [MMS]) to identify and prioritize patients in primary care or population-based settings using medical claims data.
IDENTIFICATION OF PATIENTS FOR CMM
In August 2019, we assembled a panel of subject matter experts in outpatient medication use to identify characteristics of patients who would most benefit from CMM. This panel included 5 clinical pharmacists practicing in ambulatory care or adult internal medicine settings within the Johns Hopkins Health System—2 practicing in acute care internal medicine, 1 practicing in ambulatory care internal medicine, 1 practicing in managed care, and 1 serving as a director of clinical services. All 5 completed residency training in either ambulatory care, internal medicine, or pharmacotherapy and held at least 1 board certification (pharmacotherapy specialist, ambulatory care pharmacist, and/or geriatrics pharmacist), and 3 had additional degrees (MBA or MPH).
The meeting was conducted by 2 individuals, including 1 investigator and 1 qualitative researcher, and recorded transcripts were reviewed by the project team. The goals of the panel meeting were to solicit participant ideas for definitions of medications at higher risk of adverse drug events (ADEs) and characteristics of ideal candidates for CMM.
We conducted 2 exercises: an unprompted panel discussion and a review of 5 patient cases supplemented with information related to the patient’s acute care utilization, recent medical costs, use of opioids, and predicted risk for acute care utilization from the Johns Hopkins Adjusted Clinical Group (ACG) system, which is a risk adjustment tool that uses medical and/or pharmacy claims as inputs and has been validated against health care costs and utilization.14,15
DATASET FOR INITIAL EVALUATION OF NOVEL MEDICATION RISK MARKERS
To assess the prevalence of risk markers developed at the medication level, we used a commercially available National Drug Code (NDC) database (ie, Cerner Multum Lexicon Plus 2019, Cerner Multum), which is a comprehensive database of all prescription drug products available in the US drug market. We considered medications at the NDC or active ingredient levels, such as for measures related to cost and dose form or for measures related to risk of ADEs. We evaluated medication data available as of January 1, 2014, to match our patient-level dataset.
We retrospectively reviewed QuintilesIMS patient-level administrative claims, derived from participating health plans across the United States. Our QuintilesIMS claims data included commercial insurance plans for individuals aged 18-63 years. In addition to demographic and enrollment information, the QuintilesIMS database includes diagnosis codes from International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM), procedure codes from Current Procedural Terminology, Fourth Edition (CPT-4), or the Healthcare Procedural Coding System (HCPCS); cost information from inpatient, outpatient, and emergency department (ED) settings; and pharmacy claims that included information on medication documented as NDC numbers, quantity, days supply, and cost.
We used 2014 data to construct baseline covariates and claims-based pharmacy markers. We identified 2,034,769 enrollees with continuous medical and pharmacy insurance enrollment in 2014 and 2015 and restricted our sample to 1,541,873 (75.78%) individuals with complete data who were aged between 18 and 63 years in 2014.
COMPARISON OF CHARACTERISTICS OF POPULATIONS IDENTIFIED WITH NOVEL MARKERS AND MTM ELIGIBILITY CRITERIA
We compared patients with novel risk factors to others identified by more traditional criteria, such as MTM eligibility criteria used by CMS, to provide an initial look at populations identified by these measures. We further validated these new measures against health care costs and utilization.16,17
We used 2014 QuintilesIMS data to define our outcomes and used MTM eligibility criteria from 2014 supplied by CMS for comparison.18
Since MTM eligibility criteria are provided from Medicare as ranges, we compared our markers (alone and in combination with minimum count of chronic conditions and annual pharmacy costs) to 2 sets of criteria (ie, the broadest and the narrowest): (1) “broad” with at least 2 chronic conditions, at least 2 medications, and more than $3,017 annual medication costs and (2) “narrow”with at least 3 chronic conditions, at least 8 medications, and at least $3,017 annual medication costs.
We evaluated several outcomes in this analysis, including total costs, medical costs, pharmacy costs, count of inpatient admissions, count of ED visits, and medication persistency (ie, absence of clinically relevant gaps in therapy). These outcomes were part of Johns Hopkins ACG system output for these patients.
To identify clinically relevant gaps in therapy, we used the definition included in the ACG system version 12.0. Once the number of days between the end of one prescription and the beginning of the next prescription for the same medication exceeded a critical threshold, the definition of a “gap in therapy” was met. Based on perceived consequences of therapy gaps for different conditions, the ACG system defined a relevant gap as (1) 15 days for chronic conditions that are expected to be particularly sensitive to gaps in therapy (ie, HIV, bipolar, schizophrenia, heart failure, ischemic heart disease, diabetes, depression, and seizure disorder) and (2) 30 days for other chronic conditions not expected to be as sensitive to gaps in therapy.
Identifying and prioritizing patients for clinical pharmacy services in primary care is challenging, especially when using manual chart review, and we believe efficiencies can be gained using claims data. The key contribution of our project is the development of 3 novel claims-based markers for potential use in support of primary care/population-based CMM programs. This study provides an overview of the rationale, components, specifications, and properties of the markers. As noted, our marker development process included qualitative input from subject matter experts—a panel of practicing adult primary care clinical pharmacists at our academic medical center. In addition, we conducted an in-depth empirical exploration using medical and pharmacy claims from insured populations to assess medical costs and utilization in populations identified using these markers.
Our newly developed markers provided improved patient identification and prioritization compared with MTM eligibility criteria in our sample population. Compared with the MTM eligible subgroups and the overall population, persons meeting criteria similar to the MTM eligibility criteria (both broad and narrow, including a minimum number of chronic conditions and annual pharmacy costs) and who take medications defined by our MMS markers have increased total costs, medical costs, pharmacy costs, inpatient admissions, ED visits, and gaps in medication adherence.
Although MTM eligibility criteria were not designed to identify a subpopulation of patients for pharmacist review, we propose that our MMS markers can be used to identify patients for CMM. Patients who have 2 or more MMS markers are more likely to have higher costs and health care utilization than patients meeting MTM eligibility criteria. We found similar proportions of patients with gaps in adherence when comparing patients who have MMS markers and who meet MTM eligibility criteria. We also found some MMS subgroups without a minimum medication regimen size that have fewer adherence gaps than narrow MTM eligibility criteria.
This project is part of a larger effort that will eventually operationalize these and similar markers as part of ACG risk adjustment software. We plan to use the MMS markers in combination with other data from the ACG system, such as patient demographics, clusters of chronic conditions, adherence measures, or other clinical targets, and anticipate this to improve the specificity of these tools for identifying patients within a claims database and prioritizing work for pharmacists providing CMM. We also intend to explore various other thresholds for our markers, such as top 5% cost or other combinations of complexity scores as we continue to develop these tools.
Other prioritization tools in published literature, such as the CP2 score and MUSE,7,8
consider the number of a patient’s medications but do not consider specific medication-level characteristics. For example, a patient would receive a higher CP2 score based on the number of “active items on medication list” but without differentiation of these medications or medication classes. In addition, many of these scores do not consider whether prescribed medications are filled, generating prescription claims. Therefore, our new MMS markers can be viewed as an improvement of these existing tools.
Our MMS pharmacy management risk markers can be applied on a routine basis using available claims or other similar databases, in combination with other patient characteristics, to automatically identify and prioritize patients for CMM. This automation allows pharmacists on interdisciplinary care teams to focus their time providing care directly to the right patients—those who are most likely to have problems related to medication use that leads to increased cost and utilization.
This study has some limitations to consider. Our subject matter expert panel consisted of 5 clinical pharmacists practicing within our multicenter academic health system, a group that did not include pharmacists such as consultant pharmacists in the community who also deliver CMM in the primary care or population health setting. We opted to focus on subject matter experts within our health system and did not use qualitative research methods due to limited resources. In addition, we did not include physicians or other clinicians who interact with medications on this panel, although other providers were included on our project team.
Because our suggested MCS component B required prescription claims for its calculation, the prevalence of MCS component B and overall MCS scores cannot be assessed with only medication data from Multum as was the case with component A. Although our score for medication complexity (MCS) performs similarly to MRCI,17
comparison of our MCS against MRCI and validation against patient outcomes is needed before it is used in practice.
Our marker for costly medications relied on AWP, an imperfect measure of cost. Because of factors such as insurance coverage, formulary status, and deductibles, using AWP to accurately estimate a patient’s out-of-pocket costs was challenging. This marker may be most helpful to identify high-cost medication use where less expensive alternatives exist within the same medication class (eg, high-cost branded product vs a lower-cost generic product). Unlike the other markers described here, the costly medications marker using AWP likely had less applicability to health systems with closed formularies (where medications are selected using negotiated prices or rebates, with less flexibility for alternatives considered by our marker) or outside the United States, with differing levels of government subsidization or negotiation of medication costs.
The risks and severity of ADEs or other negative patient outcomes for our proposed risky medication subcategories are admittedly different. For example, although ADEs from combinations of anticholinergic medications and their sequelae are potentially harmful for patients, especially if older or frail,32
ADEs from anticoagulants can be directly fatal.33
The MedWise Risk Score considers medication-level characteristics,12,13
including a scheme for differentiating risky medications into domains, which is an improvement over an admittedly heterogenous group of risky medications identified by ISMP or other organizations. We differentiated our high-risk medications into subcategories; empirically weighting these subcategories is an area for future research. In addition, we did not consider interactions between medications and between medications and patient characteristics (eg, renal impairment), and this is also an area for future research.
Our commercial insurance claims database did not include patients aged less than 18 years and aged 65 years or older, with fee-for-service Medicare or Medicaid data. These populations can have different demographics, medication use, and health care utilization when compared with commercially insured adult patients.34,35
Moreover, the MTM eligibility criteria used for comparison was specifically intended for use in the Medicare population. The impact of our proposed markers on other populations, including patients with Medicare and Medicaid, is an area for future research.
Medical claims data using ICD-9-CM diagnostic codes are not very reliable for identifying the incidence of ADEs,36,37
so we opted not to include this as an outcome.
We elected to use medication persistency as our quality metric related to medication use. This decision was made pragmatically, since it is already assessed in the ACG software to which our work is related. Future efforts will be made to integrate measures of medication adherence, such as the proportion of days covered within an interval that a patient has drug supply according to their fill records.
Beyond the characteristics that we have identified (eg, complexity, risk, and cost), there are several patient characteristics highlighted by our expert panel that are not commonly available either in patient EHRs or claims data, such as social determinants of health and health literacy. These factors likely affect medication use and adherence and should be areas for future development.38
Finally, these markers and future tools that we aim to develop are intended for health systems or pharmacy departments with access to claims data and analytical capability.