Using new data to personalize stroke preventionPublished Mar 12 2018, 4:20 PM
By Peter Noseworthy, MD, with Xiaoxi Yao, PhD, Mayo Clinic
The irregular heartbeat known as atrial fibrillation (AF) is a common and serious heart disease impacting about 3 million Americans today. Patients with AF experience palpitations, shortness of breath and dizziness, which can make simple activities such as playing with grandchildren or walking upstairs challenging. They also face a five-fold higher than average risk of stroke, a complication many fear more than death. Fortunately a class of drugs called anticoagulants, or ‘blood thinners,’ can help prevent stroke.
For 50 years, the only blood thinning drug option had been warfarin, which can be a cumbersome treatment because it interacts with certain foods and requires frequent blood testing to measure its effect on blood clotting. But over the past decade, new non-vitamin K antagonist oral anticoagulants (NOACs) have emerged. Based on the results of highly controlled clinical trials, NOACs are at least as effective as warfarin in stopping strokes. Importantly for patients, these drugs are more convenient to take. Blood testing is not required and there may be fewer problems with ingesting various food and drugs.
Figure 1. Timeline of oral anticoagulant medication options
Does that mean that we are free to prescribe these new treatments to all our patients with AF? Not so fast. While NOACs may provide patients with a better quality of life, important questions remain on getting the right treatment to the right patient. That’s where Mayo Clinic’s work with OptumLabs comes into the picture.
As a cardiologist, I try to help my patients control their symptoms and avoid stroke so that they can enjoy long and fulfilling lives. Now, with at least five different anticoagulant drugs options, as well as several devices and procedures available, choosing the right treatment for an individual patient can be difficult. Increasing this difficulty is that none of the major clinical trials compared each new NOAC drug option to one another, and many of my patients are different than those who participated in the clinical trials.
Fortunately, the data collected by OptumLabs from routine clinical encounters have helped us answer many of these questions that can help millions of people with AF.
Comparing new drugs in the OptumLabs Data Warehouse
Over the past five years, my colleague Dr. Xiaoxi Yao and I have used the OptumLabs Data Warehouse, a curated data set that includes de-identified claims data from more than 125 million commercial and Medicare Advantage beneficiaries of diverse ages and races across the U.S., to publish nearly 30 papers on the topic. We’ve studied three critical anticoagulation areas: effectiveness and safety, appropriate dosing and kidney damage over time.
Effectiveness and safety
Past clinical trials showed that NOACs generally prevent stroke as effectively as warfarin, but they do not compare each type of NOAC drug to one another in terms of stroke prevention and other possible side effects. In our study published in CHEST, we looked at medical claims data from October 2010 to February 2015 and compared stroke and bleeding episodes among three groups of patients with AF based on the NOAC that they took.
It turned out the three drugs are pretty similar for preventing stroke, but have different safety profiles when it comes to bleeding risk. Apixaban users had the least major bleeding, for example, while rivaroxaban users had the most major bleeding.
Figure 2. NOAC head-to-head comparison for major bleeding risk
NOACs have to be dosed very carefully. If the dose is too high, then patients could have unexpected bleeding, and if the dose if too low, they may develop blood clots and be at risk for a stroke. A key to giving a patient the right dose is first to know how well their kidneys work. This is important since some patients with kidney damage, which can be common in people with AF, may need much lower doses than patients without kidney damage because they have an increased risk for bleeding. To see if doctors were properly giving a lower dose to people with kidney problems, we looked at 14,865 patients with AF from October 2010 to September 2015.
In our report published in the Journal of the American College of Cardiology (JACC), we found almost 1 in 5 patients overall weren’t taking the right dose. This means many patients received a dose that was potentially too high, putting them at risk of bleeding, or potentially too low, increasing the risk of a stroke.
Figure 3. Prevalence of inappropriate NOAC dosing
Kidney damage over time
Even when dosed properly, researchers have noticed that warfarin and NOACs may gradually harm kidneys over time. In addition to stroke and bleeding risk, kidney damage itself is an important outcome to understand in patients taking anticoagulants. Published in JACC, another one of our studies looked at this in 9,769 patients with AF between October 2010 and April 2016.
We found that within two years of being on an oral anticoagulant, kidney damage was fairly common. NOACs as a group, however, were associated with less kidney damage than warfarin.
Figure 4. Kidney safety and outcomes associated with various oral anticoagulants
Putting the story together for patients
Keeping all of this in mind when prescribing anticoagulants for patients can be hard. So our team is working on a decision tool that considers the data from our research, as well as the research of others, to help doctors choose the safest, and most effective treatment plan for our patients.
Here’s an example of how our learnings can support appropriate anticoagulation choice for two very different hypothetical patients.
By combining large “real world” data sets with the information we’ve gained from clinical trials, I believe we are several steps closer to being able to precisely identify the right treatment for each individual patient.
Advancing the use of real-world data to help speed treatments to patientsPublished Feb 12 2018, 12:08 PM
By William Crown, PhD, OptumLabs
For more than 20 years, I used medical claims data to compare the effects of medications on patient outcomes and cost. In the back of my mind (and sometimes the front) I was always worried about what I might be missing since claims data contained limited clinical information. For example, I knew that it would be helpful to know the body mass index, cholesterol levels and blood pressure of patients when studying heart failure drugs. And, when studying breast cancer treatment, it would be critical to control for cancer stage. But none of this information was available in claims data alone.
This changed in 2009 with the HiTECH Act, which mandated meaningful use of electronic medical record (EMR) data for health care providers. The HiTECH Act transformed the data landscape almost overnight by making it possible to broaden our view into patient care and outcomes by linking patient-level EMR and claims data.
My previous concerns about possibly missing something important turned to curiosity: Could using real-world data such as linked patient claims and EMR records simulate a randomized controlled trial (RCT)? Could important evidence be produced in months, rather than years, to dramatically cut development costs and accelerate patient access to effective therapies?
OptumLabs is working to answer these questions.On the path to drug development, clinical trials typically take 6-7 years. Source: Pharmaceutical Manufacturers Association of America
What’s so great about randomized controlled trials?
Randomized controlled trials (RCTs) are used to obtain FDA approval to manufacture and sell a new drug. By randomly assigning patients to treatment and comparator groups, RCTs minimize bias from missing control variables. Randomization also reduces bias intentionally or unintentionally introduced by the investigator or sponsor in the study.
While these trials use strict research protocols to help ensure the collection of high quality data, their idealized environment often doesn’t reflect reality. For example, people do not always take their medication as prescribed for a variety of reasons. Data shows that people with chronic conditions like diabetes, hyperlipidemia, or hypertension don't take their medication 25-50 percent of the time.1 From the perspective of medication adherence alone, the very features of the research protocol that help to strengthen the internal validity of RCT results also may limit their generalizability to the real world.
The times, they are changing
Recent FDA regulatory mandates acknowledge this limitation of RCTs and a growing interest in using real-world evidence (RWE) — or, the output of research using real-world data — for approving new uses of an already approved drug and monitoring data to protect patient safety. Reauthorization of the Prescription Drug User Fee Act (PDUFA VI2) and the 21st Century Cures Act3 now require that the FDA
- Conduct a public workshop on regulatory uses of RWE by the end of FY 2018
- Fund a pilot and set methodology for RWE by the end of FY 2019
- Publish draft guidance by end of FY 2021
Researchers have already been generating RWE via claims analyses to gather safety data on approved products (e.g., FDA's Sentinel Initiative).4 This has helped establish research designs and methodologies for balancing comparator treatments across large numbers of observed attributes in the data. As a result, well-designed studies utilizing RWE are beginning to look more and more like trusted RCTs.
Our ability today to link multiple sources of real-world data can help strengthen RWE further. For example, at OptumLabs, we’re combining claims and clinical data. The EMR will tell us when the physician writes a prescription for a medication. But how do we know if the prescription was actually filled? That information is in claims data, not the clinical data. Details about how the patient responded to that medication, however, may only be found in clinical notes. It takes both pieces to see the whole picture, and linking these two sources helps remove biases that would arise from using either type of data alone.
Healthcare databases frequently used for real world evidence generation. Source: Franklin J, Schneeweiss S. When and How Can Real World Data Analyses Substitute for Randomized Controlled Trials? Clinical Pharmacology and Therapeutics 2017.
It’s an observational study — what could possibly go wrong?
There are contradictory findings when it comes to the reliability of RWE. A recent Cochrane Review concluded that observational studies generally have similar findings as RCTs in the same disease states.5 But substantial literature also shows wide variation in results from database studies within the same therapeutic areas.6
There are cases when an observational study and RCT may have very different initial results, but reach similar conclusions once the observational study design is adjusted to match that RCT. For example, the observational Nurses' Health Study showed that hormone replacement therapy (HRT) reduced the risk of heart disease among post-menopausal women.7 A decade later, results from a large RCT, the Women's Health Initiative, found the exact opposite.8
These seemingly different conclusions created huge anxiety among post-menopausal women being treated with HRT, as well as their physicians. However, it turns out that the difference in conclusions was not due to randomization at all. Rather, the risk of heart disease was highest in a two-year window after starting HRT and then declined over time. When the observational Nurses Health data was re-analyzed to control for this, the results were the same as the Women’s Health Initiative RCT.
The bottom line is that sometimes observational studies have been able to replicate the results of RCTs and sometimes not. To increase our confidence in the findings from observational studies, we need to understand why this is the case.
How can we raise confidence in appropriate use of observational studies?
OptumLabs is partnering with the Multiregional Clinical Trials Center at Brigham and Women’s hospital to explore this question. Observational Patient Evidence for Regulatory Science And uNderstanding Disease (OPERAND) is a collaboration of stakeholders from industry, academia and regulation to consider the principles behind, the methodology to draw upon, and the appropriate use of observational data in regulatory review and approval. We can then use these design elements to examine the circumstances under which real-world data — specifically, linked claims and EMR data — can be used to confirm previously published RCTs.
While individual RCTs have been replicated using observational database analyses before, we plan to replicate a large number of trials across several different therapeutic areas simultaneously (a novel feat). Our findings will complement the efforts of groups like Duke Margolis Center working on the FDA regulatory policy guidelines for RWE. Moreover, the ability to trust policy evaluations of benefit design, care organizational models, disease management programs or non-drug medical interventions all rest on the same principles being explored in OPERAND.
There will always be a place for RCTs in areas such as initial drug approvals. But with the vast amounts of high-quality data available to us today, we have a unique opportunity to explore how we can use this data to generate reliable RWE more quickly and at less cost — ultimately benefiting patients by accelerating access to effective treatments.
About the authors:
*William Crown, PhD, is chief scientific officer at OptumLabs
- Rowan C., Flory J., Gerhard T., et al. Agreement and Validity of Electronic Health Record Prescribing Data Relative to Pharmacy Claims Data: A Validation Study From a US Electronic Health Record Database. Pharmacoepidemiol Drug Saf. 2017. https://doi.org/10.1002/pds.4234.
- Anglemyer A, Horvath HT, Bero L Healthcare outcomes assessed with observational study designs compared with those assessed in randomized trials (Review). The Cochrane Collaborative: John Wiley & Sons, Ltd., 2014.
- Hernandez-Diaz S., Varas-Lorenzo C., Garcia Rodriquez LA. 2006. Non-steroidal anti-inflammatory drugs and the risk of acute myocardial infarction. Basic Clin Pharmacol Toxicol. 98(3):266-274. Kwok CS, Loke YK. 2010. Meta-analysis: the effects of proton pump inhibitors on cardiovascular events and mortality in patients receiving clopidogrel. Aliment Pharmacol. Ther. 31(8):810-823.
- Stampfer MJ et al. Postmenopausal Estrogen Therapy and Cardiovascular Disease: Ten-Year Follow-Up from the Nurses’ Health Study. N. Engl. J. Med. 325, 756-762 (1991).
- Rossouw JE et al. Risks and Benefits of Estrogen Plus Progestin in Healthy Postmenopausal Women: Principal Results from the Women’s Health Initiative Randomized Controlled Trial. JAMA 288, 321-333 (2002).
To address the opioid crisis, build a comprehensive national frameworkPublished Jan 25 2018, 6:51 PM
By Darshak Sanghavi, MD; Aylin Altan, PhD; Christopher Hane, PhD; Paul Bleicher, MD, PhD, OptumLabs
Among the myriad of challenges around the U.S. opioid epidemic is the lack of consistent, data-driven ways for the health system to measure and respond to it. That’s why OptumLabs collaborated with a panel of national clinical and public health experts to develop a comprehensive framework of 29 claims-based measures that are organized by four opioid-related domains: prevention, pain management, opioid use disorder treatment, and maternal and child health. These metrics have been shared with diverse stakeholders in various forms, and have garnered significant interest for their potential to guide a more holistic approach to evaluating and improving efforts to tackle this public health crisis. We are excited to share them with you! – Editor
This following piece first appeared in the Health Affairs Blog on December 18, 2017.
The annual rate of opioid-related deaths in the United States will surpass the historic peak annual death rates from motor vehicle accidents, HIV infections, and firearms to become the leading cause of death for people less than 50 years of age. The National Vital Statistics System reported roughly 64,000 deaths from drug overdoses in 2016 and year-to-year relative rate increases of more than 20 percent. Responding to this epidemic with such dramatic increases, the Department of Health and Human Services declared the national opioid crisis a public health emergency on October 26, in conjunction with President Donald Trump’s pronouncement on the same day.
The current opioid crisis came after substantial increases in per-capita frequency and dosages of prescription opioids beginning in the late-1990s. Researchers hypothesize the increases were driven in part by efforts to recognize pain as the fifth vital sign as codified by the Joint Commission in 2001, pharmaceutical marketing practices, and lack of recognition of the risks of dependence and addiction with prolonged use. According to the Centers for Disease Control and Prevention (CDC), when averaged nationally, roughly one prescription for opioids is written for every person in the United States annually.
Current gaps and opportunities in quality metrics
The epidemic nature of the opioid crisis and its immense complexity has exposed important gaps in the ability of the health care system to respond in a data-driven manner. Typically, federally-endorsed performance measures are developed to track and address key health care processes or outcomes targeted for improvement using blunt payment incentives, which understandably require a high degree of evidence and testing prior to widespread deployment. The development process is not typically geared to rapid-cycle quality improvement programs, which can be deployed via education, quality improvement, comparison reporting, and value-based payment incentives as has been demonstrated in the private sector. As a result, formal federal measure development has not kept pace with the rapid cadence of the opioid epidemic.
For example, the Pharmacy Quality Alliance developed three opioid misuse measures in 2015 (targeted towards identifying high frequency/dosage prescribers or towards “doctor-shopping” patients) and these measures did not complete the endorsement process of the National Quality Forum (NQF) until 2017. Broad adoption by federal payment programs may take several more years, during which time contributing prescribing trends could continue. Over a year ago, the CDC released guidelines on the appropriate dosing of opioids for pain, and the former U.S. Surgeon General mailed a letter to 2.3 million clinicians on the same topic, but no endorsed quality measure to assist with this recommendation has been made widely available.
The need for a comprehensive quality measurement framework examining the impact of the epidemic from many vantage points has particular timeliness, as the White House opioid commission recently recommended that the nation "invest only in those programs that achieve quantifiable goals and metrics."OptumLabs, a collaborative research and innovation center, embarked on a five-month program to develop a comprehensive framework of 29 claims-based measures for the opioid crisis. The measure set for this program benefitted from the input of a panel of national clinical and public health experts, including representatives from the CDC and the Substance Abuse and Mental Health Services Administration. Four key domains for the metrics were identified: prevention, appropriate acute and chronic pain treatment, Opioid Use Disorder (OUD) treatment, and maternal/child health. Where possible, measures were abstracted from publically available specifications which were refined as needed by expert coding teams. The measures were calculated using the OptumLabs Data Warehouse, which includes de-identified integrated pharmacy, medical claims, and enrollment data from a geographically diverse population of approximately 150 million United States residents currently or previously enrolled in commercial and Medicare Advantage programs. The measures are described in Exhibit 1 and technical specifications are described in a supplemental index.
Exhibit 1: Comprehensive opioid use quality measure framework and annual trends, 2016
# Measure 2016 Enrollees meeting inclusion/exclusion criteriaa 5,568,625 Prevention Primary outcome measures 1. New opioid fillers per 1000 enrollees 122 2. Initial opioid prescription compliant with CDC recommendations (composite)b 55.4% 3. New opioid fillers who avoid chronic use 97.9% 4. Prevalence of opioid overdose (OD) per 100,000 person-years 35.9 Secondary outcome measures 5. Initial opioid prescription is prescribed while patient is not exposed to benzodiazepines (component of primary measure #2) 91.1% 6. Initial prescription is not for methadone (component of primary measure #2) 100% 7. Initial opioid prescription is for short acting formulation (component of primary measure #2) 99.6% 8. Initial prescription is for <50MME/day (component of primary measure #2) 77.2% 9. Initial opioid prescription is for <=7 days supply (component of primary measure #2) 79.7% 10. No use of opioids for new low back pain patients 87.1% 11. No concurrent opioid and benzodiazepine use 78.0% 12. Appropriate contact with provider before second opioid prescription 54.0% Pain management Primary outcome measures 13. Chronic pain treatment with opioids is optimally managed (composite)c 9.4% 14. Avoidance of breakthrough post-surgical pain leading to ED visit and new opioid presription 95.3% Secondary outcome measures 15. Appropriate contact with provider among chronic opioid users (component of primary measure #13) 95.1% 16. No ED visit for breakthrough pain among chronic opioid users (component of primary measure #13) 85.3% 17. Evidence of non-opioid pharmacological treatment for pain among chronic opioid users (component of primary measure #13) 45.9% 18. Evidence of non-pharmacological therapy for pain among chronic opioid users (component of primary measure #13) 23.8% Opioid use disorder (OUD) treatment Primary outcome measures 19. Evidence of medication-assisted treatment (MAT) among patients with opioid use disorder (OUD) or OD 27.8% 20. Prevalence of OUD per 1000 person-years 8.0 Secondary outcome measures 21. Evidence of MAT following OD 10.8% 22. Evidence of naloxone fill among patients with OUD or OD 0.7% 23. No opioid prescription following any OUD or OD diagnosis 41.0% Maternal, infant & child health Primary outcome measures 24. Percentage of infants with NAS born to mothers on MAT 20.6% 25. Initial opioid prescription compliant with CDC recommendations for patients under 18y age (composite) 68.6% 26. Prevalence of OD per 100,000 person-years under 18y age 7.2 Secondary outcome measures 27. Cases per 1,000 live births of infants born with neonatal abstinence syndrome (NAS) 1.2 28. New opioid filler per 1,000 enrollees under 18y age 36 29. Prevalence of OUD per 1,000 person-years under 18y age 0.21
Source: OptumLabs Data Warehouse. Note: aIncludes commercial and Medicare Advantage health plan enrollees with two years of continuous enrollment in medical and pharmacy coverage, no evidence of active cancer treatments, and not in long-term or palliative care. Denominators vary by measure and are noted for composite measures. bNumber of new opioid users in 2016: 750,594. cNumber of chronic opioid users in 2016: 311,870.
Results of comprehensive framework development
Related to the prevention rubric of Exhibit 1, data from 2016 shows that the rate of compliance with CDC-recommended prescribing for a first opioid fill was approximately 55% and the rates of new opioid fills was 122 per 1,000 enrollees. To better explore the distribution, the measures were computed at the county level and found to be nearly normally distributed (See Exhibit 2), with substantial and meaningful variation, suggesting that local factors may strongly impact prescribing volume and guideline-adherence. (Similar variation was seen among several measures; data not shown.)
Exhibit 2: Distribution of initial opioid prescriptions in compliance with CDC recommendations and rates of new prescriptions at the county level, 2016
Source: OptumLabs Data Warehouse.
In 2016, the year the CDC issued guidance for prescribing, there was striking geographic variation (See Exhibit 3) in appropriate prescribing with further variation in the specific component of guideline non-adherence. In some areas physicians tended to prescribe high doses for new prescriptions, while in other areas, they prescribed for longer durations suggesting the need for tailored interventions based on location. For example, in Massachusetts, there was overall average performance in the composite measure of CDC-compliant prescribing when compared with national benchmarks, and the opportunity for improvement appears to relate to excessive opioid dosage, rather than prescription duration (See Exhibit 4).
Exhibit 3: County-based performance variation in selected opioid use quality measures, 2016
Source: OptumLabs Data Warehouse. Note: Blue indicates better performance. Grey areas indicate counties with insufficient data to calculate measures. Histograms depict performance on x-axis, and number of counties on y-axis.
Exhibit 4: County-based performance variation in selected opioid use quality measures, Massachusetts, 2016
Source: OptumLabs Data Warehouse.
Encouragingly, in the pain rubric of Exhibit 1, post-operative pain appeared to be well-managed with 95 percent avoiding emergency department visits resulting in receipt of new opioids. Our claims based measures related to patients on chronic opioids showed moderate utilization of non-opioid drug treatment by these patients with about 45 percent receiving this treatment, and low use of non-drug treatments (roughly one in four patients). It should be noted that patients may be accessing some of these therapies outside the benefits of their current health insurance plans.
In the treatment rubric of Exhibit 1, rates of medication-assisted treatment (MAT) for OUD were approximately 28%, with a particular opportunity in those following overdose where rates of MAT were 11%. This is consistent with reported national trends. Additional data from our analysis suggests the relative prevalence of OUD increased 50 percent over the 2014-2016 time period, potentially due to both higher incidence and improved coding with ICD-10 adoption.
Moving forward with comprehensive framework
While traditionally developed performance metrics are essential, their value in managing acute crises is limited by an extended ratification process, a focus on one measure at a time, and delayed provider feedback. By offering a comprehensive diagnostic snapshot with benchmarked data on populations that can be attributed based on patients’ geography, payer, or provider, “just in time” comprehensive quality frameworks can enable public health agencies, integrated health care providers, and/or health plans to more rapidly benchmark their status and deploy and evaluate the impact of interventions.
Within our larger organization, for example, initiatives for safer prescribing via pharmacy benefit management, access to MAT via expanded provider networks, clinical and consumer education programs, and dozens of other programs are being pursued in various populations and geographic regions, and such a framework could help assess outcomes over time. To take one example, first-fill drug utilization rules were placed in effect in July 2017 by several hundred clients served by our organization, and within two months of implementation, initial review found relative reductions of 82 percent in excessive dosage and 65 percent reduction in excessive duration of new opioid prescriptions relative to CDC opioid best practice prescribing guidelines among this cohort. In this manner, “just in time” frameworks may permit a precision-medicine approach to the opioid epidemic based on a common set of data-driven metrics.
About the authors:
*Darshak Sanghavi, MD, is chief medical officer at OptumLabs
*Aylin Altan, PhD, is senior vice president of research at OptumLabs
*Christopher Hane, PhD, is vice president of data science at OptumLabs
*Paul Bleicher, MD, PhD, is chief executive officer at OptumLabs