At the expertly-curated ADA Symposium on the Use of Real-World Data to Improve the Prevention and Care of Diabetes-Related Outcomes, attendees (n=~30) were immersed in multiple informative presentations and riveting discussion. The sessions ranged from sweeping overviews of the history of real-world data (RWD), to complex regulatory processes, to the nitty-gritty of RWD methodology. We were particularly struck by the diversity in use cases for RWD and by the passion in the room.
FDA’s Dr. Peter Stein took a fairly conservative stance on the use of RWD in diabetes regulatory decisions. He acknowledged the benefits of RWD, especially in helping with post-market surveillance studies, but ultimately believes RCTs will remain the gold standard in determining safety and efficacy. During the subsequent discussion, Dr. Stein touched on the potential use of RWD in helping to establish cardiovascular safety and on alternative endpoints in diabetes trials. From our view, it goes without saying that RCTs are extremely important, but we are increasingly unsure of their generalizability to broader patient populations outside of narrow RCT inclusion/exclusion criteria. We would like to see more RCTs compared to established, “real world” compounds like SFUs; as we understand it, FDA often suggests this class be avoided as a comparator and requires a placebo for phase 3 studies. One design that was not discussed during the meeting is Mendelian Randomization, which has become an increasingly popular method to attempt to overcome lack of randomization.
While much of the discussion focused on the benefits of RWD – particularly its utility in better understanding the broader and frequently underrepresented populations – there are clearly some limitations and some changes that can be made to improve them. The need to standardize RWD was of paramount concern, as was establishing guidelines around data definitions, data linkage, and data sharing. These seem like addressable problems to us. In the highlights below, speakers shared several examples of RWD applications in diabetes – read on for Dr. Kasia Lipska’s views on underestimating hypoglycemia incidence, an Onduo poster with the very first outcomes from the virtual diabetes clinic launch (-1.3% A1c drop in individuals with baseline A1c ≥8%), and a Kaiser Permanente analysis of the negative impacts of out-of-pocket costs.
The goal of the meeting, co-hosted by Drs. Edward Gregg of the CDC and Kamlesh Khunti of the University of Leicester, was to inform an ADA consensus publication on the use of RWD in diabetes. The authors aim to publish the manuscript in Diabetes Care in June-July 2019 and wasted no time, meeting right after the conference wrapped up to determine writing assignments. In one of our favorite sessions, Drs. Gregg and Khunti laid out the goals for the consensus statement and then opened up the floor for discussion. We were especially interested to hear attendees’ views on leveraging industry partnerships to enable better data linkage and to examine outcomes in underrepresented populations. We were also intrigued by a philosophical debate on the tough balance guidance must strike: while on the one hand, guidance serves to ensure best practices, on the other, it may stifle creativity and innovation necessary for solving tricky problems.
Greetings from Washington, DC where our team has been immersed in the ADA Symposium on the Use of Real-World Data to Improve the Prevention and Care of Diabetes-Related Outcomes. Leaders in academia, industry, and regulatory gathered to define the current and future utilization of real-world data in diabetes. Co-hosted by Drs. Edward Gregg and Kamlesh Khunti, the ultimate goal of the meeting was to launch the preparation of an ADA Consensus Statement on real-world data – in fact, the writing committee met just after the sessions wrapped up! See below for our top takeaways from the meeting.
- Top Highlights
- 1. FDA’s Dr. Peter Stein Takes Reserved Stance on RWE/RWD in Regulatory Decisions; “Show Us That [Time-In-Range] Improves Patient Lives [With] Validated Instruments… We Want to See a Validated Endpoint With a Validated Tool”
- 2. Dr. Lipska on Underestimating Hypoglycemia Incidence (by ~95%!), Barriers to Real-World Data Collection in Diabetes, Rise in Amputations (+50% Since 2009)
- 3. Onduo Poster (n=133) Shows Significant 1.3% A1c Decrease in Type 2 Patients with Baseline A1c ≥8%, Modest A1c Changes When Baseline A1c <8%; Prospective Study “Being Initiated”
- 4. United Health Group Study Finds Minimal Cost-Savings from DPP – How Can We Make Prevention Cost-Effective?
- 5. Kaiser Permanente’s Dr. Andrew Karter on the Negative Impacts of Out-Of-Pocket Costs; Changes in Health Plan Benefits as “Natural Experiments”
- 6. Characterizing Cost-Effectiveness of “Excellent” Glycemic Control in Type 1s with Diabetes Devices; MDI-Only and Pump-Only Therapies are “Very Cost-Effective”
- 7. Survey of Type 1s Establishes Key Role for Oral Adjuncts; Patients Name Time-in-Range + Weight Loss as Most Important Attributes in a Therapy
- 8. Patients with Type 2 Diabetes on Basal-Only Therapy Are Willing to Do More than HCPs Expect; 80% Patients Frustrated with Slow Progress to Reach A1c Goals
- 9. Patient-Facing Portal to EHR Positively Impacts Healthcare Utilization in Patients with Diabetes; Mobile Patient Portal as a “Leveler” in Reducing Disparities
- 10. Dr. DuMouchel Describes Gaps in FDA’s Database of Drug-Related Adverse Events, Emphasizes Vital Role for RWE in Assessing Drug Safety
- 11. Dr. Wharam Outlines Negative Consequences of High-Deductible Health Insurance: Delayed Primary Care, Increased ED Visits
- 12. Prof. Martin White Gives Overview of UK Public Health Policies to Fight Diabetes/Obesity Epidemic
- 13. FDA-Funded Harvard Medical School Study to Replicate 30 RCTs with RWE; Aims to Identify Space for RWE in Regulatory Decision Making
- Open Discussion to Inform the ADA Consensus Publication
Deputy Director of the FDA’s Center for Drug Evaluation and Research (CDER) Office of New Drugs Dr. Peter Stein provided a fairly reserved perspective on the use of real-world data (RWD) and real-world evidence (RWE) in the drug regulatory landscape. With certain exceptions for oncology and rare diseases, Dr. Stein asserted that there is a “limited use” for RWE in drug regulatory decisions, but noted that it is “certainly an area that we are looking into.” He emphasized that studies collecting RWD/RWE ask fundamentally different questions than RCTs (namely, effectiveness vs. efficacy). Still, he cited several benefits to RWD including: (i) analysis over a much broader and more diverse patient experience; (ii) massive sample sizes that may reveal infrequent events and drug-drug interactions; and (iii) lower resource intensity. In fact, he believes RWD elements can be incorporated over the entire spectrum of drug development, from informing site selection and trial feasibility in RCTs to supplementing pragmatic RCTs with claims and EHR data. Dr. Stein clarified that the use of RWE for post-approval safety indications is “common” and that Sentinel, which provides access to a large claims database, is increasingly being used by the FDA to replace post-market safety studies. However, according to Dr. Stein, the statutory and regulatory framework for FDA approval is “not likely to change soon.” To this end, he maintained that double-blind RCTs are the “gold standard for robust determining of efficacy and safety,” which is understandably of primary concern in regulatory decision-making. While Dr. Stein acknowledged the importance of understanding broader treatment effects made possible by RWD/RWE, evaluating internal validity (i.e., does this drug have the effect it purports to?) is the FDA’s main goal. Despite these reservations, Dr. Stein ended on a positive note, allowing for the possibility that improvements in analytic and design methods may eventually overcome the current limitations of RWD. We would love to hear more about why oncology gets the pass to use RWE – why Rare Disease may get a pass is more understandable given the small size of these populations.
During the discussion, former ADA Chief Scientific Officer Dr. Robert Ratner referenced the FDA’s recent advisory committee on CVOTs and asked Dr. Stein for the Agency’s perspective on the use of RWE to support secondary outcomes, provided safety and efficacy have already been proven. Dr. Stein stipulated that the risk of confounding in RWE is higher when looking at hazard ratios of 1.2-1.3 (the 2008 guidance requirements). Still, he noted that there are “ways of doing pragmatic trials to answer these questions.” He was quick to add that such trials still require the “evolution of scientific methodology and analysis,” but that “maybe in a few years, the answer would be yes to pragmatic trials.”
Dr. Stein listed the specific challenges and limitations of RWD/RWE for regulatory decisions: (i) assuring accurate patient selection and comparable exposure/outcome risk between groups; (ii) assuring the “study drug” and comparison groups are balanced for risks of the outcome; and (iii) assuring comparable patient management, as patients’ disease status, risk factors, and/or medications may lead to differential co-interventions and diagnostic intensity.
Tandem’s Ms. Molly McElwee Malloy raised a question regarding the burden of proof needed for diabetes regulatory decisions, given that “A1c is no longer the benchmark that proves superiority.” She further elaborated, explaining: “We all know A1c doesn’t tell you what my blood sugar is doing every day, and we know time-in-range improves quality of life – it just makes sense.” We agree – a 5% improvement in Time in Range to patients translates into an hour a day better blood glucose – it’s hard to argue with this. Dr. Stein emphasized the importance of distinguishing between endpoints that are valuable in informing physicians vs. those useful for regulatory decision-making. He acknowledged that “evidence generated from those endpoints that inform practice is very helpful,” but maintained that “the FDA doesn’t regulate the practice of medicine, it regulates the label.” We would hope these would be related! He acknowledged that the FDA “tends to be skeptical.” Still, he noted that the Agency is “involved in discussions with many organizations about these matters” and left the audience with this call to action: “Show us that [time-in-range] improves patient lives [with] validated instruments… we want to see a validated endpoint with a validated tool.” We were not sure exactly what this means. Notably, CDRH’s Dr. Courtney Lias gave her valuable take on validating surrogate outcomes at DTM a couple weeks ago.
Selected Questions and Answers
Dr. Robert Ratner (Georgetown University, Washington, DC): The FDA had a meeting four weeks ago revisiting the 2008 guidance on the requirement for CVOTs. The ultimate vote was 10-9, clearly the panel was not convinced one way or the other. Based on the conversation we’re having, what direction do you think the FDA is moving in in terms of utilizing RWE for these secondary outcomes once drug is proven safe and effective?
Dr. Peter Stein: I think what we struggle with is where RWE can play a role. Certainly from a safety perspective – I already feel that is an important use of RWE. Here there’s a challenge – think about when you have an observational database analysis talking about fairly robust evidence where you have a higher hazard ratio – it can still be confounded but you’re more confident. The risk of confounding with a hazard ratio of 1.2-1.3 is higher. Could pragmatic trials answer those questions? Yes. There are ways of doing pragmatic trials to answer these questions that can focus on what we’re trying to ask and not the broader issues. It will still require the evolution of scientific methodology of analysis. I don’t want to say it won’t happen in the future because the progress amazing. Maybe in a few years the answer would be ‘yes pragmatic trials.’
Ms. Molly McElwee Malloy (Tandem, San Diego, CA): Achieving A1c is no longer the benchmark that proves superiority – there’s glycemic variability, time-in-range – how do you think about the burden of proof needed? We all know A1c doesn’t tell you what my blood sugar is doing every day and we know time-in-range improves quality of life. It just makes sense. But what about getting time-in-range to serve as approval for regulatory when you already have A1c – how do you suggest people go about that from a research perspective?
Dr. Stein: I think we should distinguish how those endpoints can be very valuable to inform physicians vs. whether they are ready for regulatory decision making. Basically, we care about saying that a drug has the efficacy that it purports to have – that’s the standard we have for approval. Typically, we use A1c for the approval of glucose-lowering drugs. When it is approved, we recognize the ways in which it will impact patients are not all necessarily defined in an RCT – this may be relevant for physicians. The evidence generated from those endpoints that inform practice is very helpful. But the FDA doesn’t regulate the practice of medicine, it regulates the label. Where there’s an interest in adding those outcomes to labeling, how do you know they are beneficial? We tend to be skeptical, we want to see a body of evidence. Show us that it improves patient lives, give us validated instruments that demonstrate that. We take a simplistic straightforward approach – we look for outcomes that impact how patients feel, function, or survive. We want to see a validated endpoint with validated tool. We’re involved in discussions with many organizations about these matters, we’re very open. The standards we have for validated instruments and meaningful difference are fairly rigorous. It’s a great question, and I think it could take a whole other three hours on how to proceed in this area. (Editor’s note – we wish they would have!)
Yale’s Dr. Kasia Lipska highlighted a major challenge in hypoglycemia surveillance – because we rely on healthcare utilization records (ED visits, hospital admissions), we miss the ~95% of severe hypo that is treated outside the medical system (and it may well be more than this). A recent study published in JAMA Internal Medicine (n=13,359) found that while 11.7% of participants self-reported a severe hypoglycemia episode, only 0.8% of those individuals visited an emergency department or hospital. Interestingly, 20% of the participants who made it to the ED/hospital for hypoglycemia did not self-report an event, which suggests that patient/caregiver reporting of hypoglycemia requiring assistance is not completely reliable. Dr. Lipska captured the overlooked hypos as the portion of an iceberg under water. In her estimate, only ~5% of severe hypos requiring assistance show up in an ED. Between 27%-87% of diabetes patients who call emergency medical services (EMS) for hypoglycemia are transported to a hospital emergency department. Dr. Lipska attributed the wide variability (27%-87% is quite imprecise) to differing state and county protocols for how EMTs should respond to hypo in the field; during Q&A, one audience member pointed out that in some places, EMTs aren’t reimbursed by Medicare/Medicaid for treating hypoglycemia unless they bring the patient to a hospital, which is fee-for-service reimbursement at its finest (and almost certainly leads to unnecessary hospital bills in some cases). Dr. Lipska estimated that 40%-50% of patients who turn up in the ED for hypo are admitted to the hospital. While relying on healthcare utilization claims is efficient in surveillance for MI or DKA, since patients are almost always hospitalized for these diabetes complications, it is wholly inadequate in measuring true hypoglycemia incidence in the population. As Dr. Lipska put it, “we need to get at the bottom of the iceberg,” and she offered several pathways to improved hypoglycemia surveillance. Patient-reported outcomes are key, even though the recent JAMA study tells us that self-report on its own is not accurate. Data-sharing through digital platforms, wearables, and social media presents another opportunity for population-wide hypoglycemia monitoring, and CGM allows for better capture of non-severe hypo. Notably, Dr. Lipska underscored that patients are more than willing to share their data for research purposes, assuming appropriate privacy protections are in place. She discussed a Research!America poll from 2013, emphasizing that >70% of patients surveyed said they were willing to share health data so that researchers could better understand diseases, new treatments and cures could be accelerated, and HCPs could improve patient care. Shedding light on the submerged portion of this iceberg could do wonders for pushing the field toward therapies and devices that expose patients to far lower risk of hypoglycemia.
Dr. Lipska showed that ~40% of hypoglycemia episodes caused by insulin lead to hospitalization, while ~51% of severe lows caused by oral anti-hyperglycemic agents lead to hospitalization (we suspect sulfonylureas are at the top of this list). The silver lining here is that as new technologies help people dose insulin in a safer manner and novel diabetes therapies lower hypoglycemia risk, we should see fewer hospitalizations for the complication – though of course this depends on significant real-world uptake of the advanced drugs. If a patient is taking older insulin like NPH or an SFU, they may well be less likely to use CGMof any kind or even frequent enough or accurate SMBG. According to a Diabetes Care paper published last year, sulfonylureas are still the most common second-line diabetes drug prescribed to US adults with type 2; meanwhile, only 4% of patients are ever prescribed an SGLT-2 inhibitor and only 7% are ever prescribed a GLP-1 agonist, even though these newer classes come with much lower hypoglycemia risk. We must expand access to safer, more effective therapies in order to stimulate meaningful change in population-level hypoglycemia incidence. Novo Nordisk’s next-generation basal insulin Tresiba (insulin degludec) significantly lowers hypoglycemia risk compared to standard of care Lantus (Sanofi’s insulin glargine), and we know that wider use of next-gen insulins (Sanofi’s Toujeo in addition to Novo Nordisk’s Tresiba) will bring down hypo incidence in the population as well – it patients can gain access.
Dr. Lipska called into question the cost-effectiveness of new diabetes therapies (including DPP-4s and GLP-1s) in a 2016 Diabetes Care paper. At a high level, the study found that between 2006-2013, mean A1c and mean hypoglycemia incidence in the diabetes patient population were flat, despite increased use of DPP-4 inhibitors (0.5% to 15% of people with type 2 diabetes) and GLP-1 agonists (3% to 5% of type 2s). Sub-group analysis wasn’t done, though that was an obvious miss in our view and at the time she said they did not have time to do the additional analysis on the subgroups. Dr. Lipska acknowledged at the time that data sources on hypoglycemia were limited: The study relied only on ED/hospital admissions, missing ~95% of severe lows treated outside of medical care and overlooking all non-severe hypoglycemia. We imagine that newer diabetes therapies will affect non-severe hypo, severe hypo requiring assistance, and productivity even more than they would cut into hospitalizations. We’re very curious to see how the numbers in Dr. Lipska’s 2016 paper would change if there was better infrastructure to measure these less severe forms of low blood glucose – and we hope for more studies like this as CGM becomes more common. Moreover, we think it’s important to note that GLP-1 use was still extremely low in 2013 (5%), and SGLT-2 inhibitors were essentially non-existent (J&J’s Invokana was the first SGLT-2 approved by FDA, in late 1Q13). We’re optimistic that the dissemination of advanced products like GLP-1s, SGLT-2s, and next-gen insulins (Tresiba, Toujeo) will move the needle on population-level hypoglycemia incidence with more time and we hope very much that these relatively simple analyses can be completed.
Dr. Lipska also presented important and disturbing data on racial disparities in hypoglycemia incidence. From 1999-2011, hospital admissions for hypoglycemia were four-fold higher among Black vs. white Medicare beneficiaries. What accounts for this disparity? That remains an open question, but Dr. Lipska mentioned during Q&A that studies are ongoing to identify potential mediators.
She briefly touched on surveillance of other acute diabetes complications (MI, stroke, end-stage renal disease), and noted a 50% rise in amputations between 2009-2015. Amputation incidence was declining until 2009, when the reverse trend began. The audience was clearly struck by this finding, but Dr. Lipska explained that we’re not sure what’s behind the uptick. Importantly, the climb in amputation incidence started before SGLT-2 inhibitors were available (a recent BMJ publication linked SGLT-2s as a class to higher amputation risk vs. GLP-1s, so we think this is a crucial point to make). During the subsequent panel discussion, CDC’s Dr. Edward Gregg speculated that increased amputation rate among people with diabetes might be due to a growth in the number of patients with very poor A1c, but he emphasized that we can’t rule out measurement bias, either. [He noted at EASD 2017 that a rise in amputations was driven by men and amputations of the toe.] We speculate that it could also have to do with people living longer with diabetes, and thus being more susceptible to development of complications. Other diabetes thought leaders have previously underscored to us that amputation is a soft endpoint, meaning the decision to amputate involves input from providers/patients, and it can’t be adjudicated as rigorously as a MACE event.
Select Questions and Answers
Q: You showed us the good news about decreasing trends of multiple different diabetes complications, followed by the bad news of increased risk of amputation in recent years. Were you linking hypoglycemia as one of the explanations for amputations? Have temporal trends in obesity been looked at to explain the increase in amputations?
Dr. Lipska: I was not trying to link hypoglycemia as a potential causal explanation for amputations. Those are two separate things. Of course, it’s possible that they are somewhat related – for example, we know that hypoglycemia increases risk for CV events – but I wasn’t trying to tell that story in this talk (editor’s note – it is unclear whether she sees a connection).
In terms of what explains the recent rise in amputations, we can only conjecture at this point. I don’t think we have the data to provide a concrete answer. If you look at medication use, things have by and large gotten better. We’ve seen tighter glycemic control at the beginning of care and relaxation thereafter – whether this has something to do with amputations, we don’t know. I don’t know the answer. But I think it’s really important to do surveillance and then launch studies to get at what’s driving population trends, so we can fix things. I haven’t looked at BMI/ amputations.
Q: I noticed there was no data for Hispanics on your hypoglycemia admissions slide. Is that data difficult to tease out?
Dr. Lipska: We were using categories used in Medicare data; we have separately looked at race in more detail than just white, black, other. We’re doing some studies now looking at potential mediators that may explain the disparities between races. I think it’s an important question.
Dr. Lawrence Blonde (Ochsner Medical Center, New Orleans, LA): There have been recent changes in the definition of hypoglycemia that I think are quite useful. The outcomes beyond A1c movement I think is quite helpful. CGM offers the opportunity in the not-too-distant future to get at non-severe hypoglycemia, and I think it’s extremely important to capture that. CGM should influence how real-world diabetes studies are done in the future. CGM deserves more focus.
Dr. Lipska: I agree. I’m using CGM in nursing homes to look at hypoglycemia in those patients, who may have difficulty even reporting their episodes given cognitive impairment.
Dr. Edward Gregg (CDC, Atlanta, GA): The amputation data illustrates the questions that stem from real-world surveillance. We are speculating three-four different domains that have could have caused the recent uptick: (i) change in clinical decision-making by vascular surgeons, (ii) plateaus in aspects of diabetes control, (iii) healthcare reform, and (iv) we can’t rule out measurement bias.
Q: CMS will not pay for EMS services in some areas if the patient is not transported to the hospital, is my understanding. Were you able to capture that?
Dr. Lipska: That’s a great question, and an excellent point. We’re able to see from our CMS data if patients arrived at the hospital by ambulance or not – 66% of them do arrive at the ED via ambulance. But there are differences in protocol and reimbursement schemes, and it’s important to look at how this all varies across states and counties.
Q: Is there an elephant in the room? It’s clear that the US has enormous capacity to do population research, but we’re missing a very important tool, and that is the unified healthcare identifier. Is the absence of our ability to link databases going to inhibit the US from doing the kind of research that we ought to be able to do?
Dr. Lipska: Yes. We go through so many antics trying to link people across datasets. Yes, I agree with you. I think we all agree.
Q: What should be done?
Dr. Franklin Wharam (Harvard Medical School, Boston, MA): All I’ll say is, the environment might be getting even more challenging with the publicity around data breaches, as people realize that the data they thought was secure is absolutely not in many cases. So, I don’t know that we’re heading in the right direction there.
Comment: Well, I look forward to the day when we have a more integrated system. (Applause.)
Q: There’s an issue in the US over who owns the data. What is your opinion on this? Is it the patient? Should they be able to sell data directly to a company that wants to use it?
Dr. Lipska: That’s a great question – who owns the data? It’s a hot topic right now. There’s a larger movement to put data in the hands of patients, as you know. Harlan Krumholz at Yale has been a mover and a shaker in that field, trying to allow patients to have access to the data about them and then they can share or sell for research purposes. This can be an opportunity for us researchers to get more access to the data, because as I mentioned, patients are by and large willing to share their data for research (they just don’t know they have rights to their own data or access to it). In terms of logistics, how to get people access to their data, that’s a huge obstacle. But there are platforms being developed for exactly that.
An Onduo poster depicted the first outcomes data from the initial launch of the Onduo virtual diabetes clinic. Onduo, the $496 million Sanofi-Verily joint venture, launched its virtual clinic to Blue Cross Blue Shield Members with type 2 diabetes in South Carolina (January 2018), Arkansas (February 2018), and Georgia (March 2018). Preliminary findings derived from a limited cohort of program participants (n=133) found a significant A1c decrease of 1.3% in the pre-specified primary analysis cohort of patients with baseline A1c ≥8% (n=44). As expected by the investigators, members with a baseline A1c between 6%-7.9% (n=89) maintained A1c over the follow-up period. Baseline A1c values were collected by FDA cleared at-home tests or lab reports and defined as A1c values obtained between 90 days prior or 24 days after Onduo app login. Follow-up kits or lab reports were collected 80-180 days after baseline A1c. A secondary analysis of four pre-specified subgroups was conducted, demonstrating that (i) those with the highest baseline A1c (>9%) achieved a significant A1c decrease (-2.5%); (ii) those in the middle A1c groups (baseline: 7%-7.9%, 8%-9.0%) trended towards non-significant A1c declines; and (iii) those with the lowest baseline A1c (6%-6.9%) actually saw a significant, albeit modest, A1c increase (+0.3%). We’d be interested to see this last group’s raw glucose data, because we suspect they are spending less time in hypoglycemia; it’s also worth noting that 0.3% is within the measurement error of some A1c lab assays. Still, as the poster notes, these preliminary, real-world findings suggest that the Onduo clinic can help to maintain glycemic control in those with baseline A1c 6.0%-7.9%, while driving meaningful A1c declines in those with A1c ≥8%. Per the poster, a prospective clinical trial is “being initiated’ to confirm the results with “more complete endpoint ascertainment.” The investigators also acknowledged that eventually RCTs will be needed to compare the Onduo clinic to standard of care.
These preliminary results seem stronger than the four-week feasibility findings presented at ADA, which showed a 30 mg/dl decrease in average blood glucose in those with baseline average blood glucose >154 mg/dl, reflecting ~0.7% A1c decrease if sustained. Still, it’s difficult to directly compare the results, as a baseline blood glucose >154 mg/dl likely includes individuals with lower initial A1cs than the 8% baseline cutoff used in this new analysis.
The Onduo virtual diabetes clinic consists of an app, a care team/coaching, unlimited BGM supplies, and CGM for those who qualify. Members received diet and lifestyle coaching by mobile chat in accordance with an AADE curriculum, and, where indicated, participated in telemedicine-mediated medication management per ADA guidelines.
Northwestern’s Dr. Ronald Ackerman noted a unique challenge in scaling the Diabetes Prevention Program (DPP): There’s no doubt that intensive lifestyle intervention works (58% reduction in type 2 diabetes incidence over three years), but as of now, it’s impossible to identify the people who will benefit the most and make the program cost-effective for payers. Dr. Ackerman walked through results from a study conducted by United Health Group (UHG) in collaboration with the YMCA. UHG targeted 759 employers and covered costs for employees to participate in the YMCA’s DPP (n=1,725 after matching). On the positive side, there was a dose-response relationship between attendance and weight loss: Participants who attended at least 16 DPP sessions (n=1,801) lost a mean 5.4 lbs vs. -0.2 lbs for participants who attended only 1-3 sessions (n=677). Dr. Ackerman pointed out that amount of weight loss was correlated with delayed diabetes onset in the original DPPOS, and he underscored that even modest ~5% body weight loss yields significant metabolic benefits. To this end, 47% of UHG beneficiaries attending ≥16 DPP classes achieved ≥5% weight loss vs. 0.4% of the beneficiaries attending 1-3 classes. Despite this success on adherence and weight loss, there was no significant difference in overall healthcare expenditures between DPP participants and matched controls – UHG reimbursed a mean $211 per participant (paying based on attendance) but did not see cost-savings over two years. Dr. Ackerman cautioned that this was a relatively short duration of follow-up; he postulated that over three or four years, average costs for YDPP users would drop relative to matched controls. The larger dilemma facing an insurer, however, is that only ~2% of people with prediabetes will go on to develop type 2 within one-two years. Prediabetes itself doesn’t necessarily drive up healthcare costs in the near-term, so we shouldn’t expect major divergence in total health spending between DPP participants and controls when the intervention is delivered to a random sample of people with prediabetes. The obvious solution given the current landscape is to target DPP to the population at the highest risk of developing diabetes, if possible; however, there are at least two forces pushing payers toward covering everyone with prediabetes: (i) The ACA requires commercial plans to provide DPP at no cost; and (ii) offering DPP is a retention strategy.
Describing this challenge another way, Dr. Ackerman explained that two million US adults develop type 2 diabetes each year. If all of them participated in the DPP, one million cases could be prevented. From this viewpoint, widespread implementation of the DPP seems like a no-brainer, but there are 84 million Americans with prediabetes and zero methods to accurately predict which two million will develop diabetes. Delivering the DPP to the entire prediabetes population would run a bill of $126 billion. Selecting the two million individuals at highest-risk for type 2 diabetes would make the program cost-effective, but we currently don’t have the appropriate risk stratification tools, Dr. Ackerman explained. Further, 88% of people with prediabetes are unaware of their dysglycemia (according to the latest figures from CDC), which presents another obstacle to scaling up the DPP, or getting it to the highest-risk individuals to save the greatest amount of money for the healthcare system.
Of note, cost-effectiveness at the population level is distinct from health benefit at the individual level. Delivering the DPP to all US adults with prediabetes might be extraordinarily expensive, and perhaps 98% of these people would not have developed type 2 diabetes within one-two years anyway, but as Dr. Ackerman noted, weight loss via lifestyle intervention has meaningful metabolic benefits. Achieving ~5% body weight loss would improve individual health in ways other than diabetes prevention (e.g. reducing CV risk, probably reducing cancer risk), so these benefits should be considered by HCPs/payers and incorporated into cost-effectiveness models whenever possible.
Dr. Ackerman’s talk was a hard-hitting reminder that prevention research is difficult (given the long duration of follow-up needed before benefits emerge) and investment in prevention is more difficult still. If payers like UHG doesn’t see cost-savings over one-two years, it’s hard to incentivize reimbursement for prevention programs, especially since the average US adult switches insurance plans after two years (i.e. UHG would pay for the DPP but wouldn’t reap the rewards). Moreover, Dr. Ackerman’s remarks suggested that even if we had universal coverage of the DPP, it wouldn’t be cost effective for the health system without (i) reliable risk stratification or (ii) lower-cost translations of the intensive lifestyle intervention (e.g. digital platforms, group delivery, etc.). The latter seems like a more realistic solution in the short term, and we’re keeping our fingers crossed that Medicare will expand its DPP coverage to digital programs (a large pilot to evaluate virtual DPPs is underway). In the meantime, we hope to see continued research into risk stratification of the prediabetes population – how do we separate the two million who will develop type 2 diabetes from the pool of 84 million with prediabetes? As for the issue of reaping the rewards of a preventive service, we wonder if there could be a role for “shared churn savings”; i.e., if a payer inherits a beneficiary in good health or who underwent a DPP, then they pay a portion of the cost savings to the individual’s former insurer. This pay-it-forward practice would not be easy to establish and we’re not aware of anyone implementing it, though it would certainly help to incentivize investment in preventive care.
Select Questions and Answers
Q: Doesn’t United Health Group have about 50 million members? So, it would seem that DPP uptake is pretty low. How does that impact your results?
Dr. Ackerman: The people we showed here were within the employer groups where United was trying to do this initial screening and referral linking (759 employers). During the same period of time, United implemented broader coverage strategies, and 29,000 of its beneficiaries participated overall. This was probably people who had opportunistic screening from their healthcare provider. As time goes on, we’re going to see more opportunistic screening that occurs in healthcare visits, makes people aware of their prediabetes, and maybe encourages them to enroll in the DPP.
Kaiser Permanente’s Dr. Andrew Karter shared fascinating patient behavior insights gleaned from changes in health plan benefits. As he explained, such changes create “a natural experiment,” revealing far more than cross-sectional data. In one example, Dr. Karter explained how Kaiser used to charge a copayment for test strip refills through 1999, but following a California bill mandating that plans cover supplies beginning in 2000, Kaiser switched to provide all members free strips between 2000-2001. Not surprisingly, the data leading up to 2000 showed a strong distinction in SMBG use between those who paid the most for strips vs. those who paid the least – in fact, there was nearly a 15% spread in SMBG activity. While Dr. Karter expected these copayment cohorts to converge when strips were provided at no cost, he instead saw no significant increase in utilization. Dr. Karter found these results “pretty surprising,” suggesting that perhaps a “legacy of entrenched habits,” in which testing behaviors were shaped due to long-term exposure to out-of-pocket (OOP) costs, were to blame. [This is another reason CMS was smart to begin covering CGM for its members – if fingersticks are not a part of their daily routine, it’ll be difficult to get them to start once they turn 65.] Kaiser later introduced a 20% co-insurance for certain members, resulting in a greater, albeit not clinically relevant, SMBG utilization decline as compared to the non-significant increase that had been observed when strips were free. As Dr. Karter explained, this phenomenon is known as loss aversion – when a stronger reaction is elicited due to loss than in response to an equivalent gain. Ultimately, Dr. Karter cautioned that OOP costs may undermine health system quality goals.
OOP cost is a greater impediment to therapy initiation than it is for long-term persistence. Dr. Karter presented data from over 200,000 patients with diabetes who were newly prescribed >800,000 cardiometabolic medications. The results indicated that an OOP cost ≥$20 was associated with double (!) the rate of never initiating a new prescription (7% vs. 3%) as compared to $0 OOP cost. While inadequate secondary adherence (≥20% of the time without medication supply) was shown to be greater when OOP cost was ≥$20 vs. $0 (39% vs. 32%), there was no significant impact of OOP cost on 24-month persistence. When evaluating the impact of extra OOP costs associated with adding a new medication, the results showed that younger patients (<45 years-old) were most sensitive to the extra OOP cost. Dr. Karter found these results “concerning,” as it’s especially important for this younger demographic to initiate and adhere to therapies. Indeed, data continues to surface indicating that health systems seeking to save money by bringing patient “skin in the game” may actually be costing themselves more in the long run.
Dr. Karter provided an example of the increased care law – “the availability of good healthcare tends to vary inversely with the need for it in the population served” – in action. High deductible health plans increase OOP cost, particularly in lower socioeconomic status (SES) patients, while value-based insurance design (VBID) eliminates OOP cost for certain high-value services. Certain health plans, Dr. Karter explained, are combining the two in an attempt to offset the negative impact of deductible plans on the adherence to essential medications. The data revealed that for patients initiating a deductible plan, adherence for cardiometabolic medications declined for those without VBID and remained stable for those with VBID – a positive indication that adding VBID is protective against a decrease in adherence. However, when stratified by SES (higher vs. lower), VBID was shown to offset the negative effect on adherence only among higher SES patients – adherence declined in lower SES patients regardless of the addition of VBID. Essentially, said Dr. Karter, VBID increased SES adherence disparities. As he put it, innovations, structural changes, and new models of care can at first increase health inequities because advantaged populations typically gain access and benefit from these changes first.
University of Michigan’s Dr. William Herman presented a cost-effectiveness analysis of type 1 diabetes therapies using contemporary data on the costs of treatment options coupled with 30-year DCCT/EDIC results on the incidence of diabetes complications. In this study, Dr. Herman aimed to capture the clinical, quality of life, and cost consequences of 30 years of “excellent” (defined as A1c ~7%) vs. “poor” (defined as A1c ~9%) glycemic control. His analysis relied on several assumptions, notably that (i) “intensive therapy” can be equally characterized by the use of MDI-only, pump-only, or pump with CGM and confers an A1c ~7%; and (ii) that “conventional therapy” (i.e. syringe use) confers an A1c ~9%. Given these assumptions (see below for attendees’ thoughts on their validity), Dr. Herman performed a subgroup analysis of DCCT conventional therapy participants who maintained A1c >8.8% vs. those who received intensive therapy and maintained A1c <7.2% over 30 years of the combined DCCT/EDIC follow-up. He highlighted significant differences in end-stage renal disease, amputation, myocardial infarction, coronary artery bypass surgery, congestive heart failure, and stroke between the two groups – all unsurprisingly favoring intensive therapy. Each of these complications were then translated into health utility scores for the EDIC population and ultimately used to calculate the event and ongoing costs of diabetes complications, comorbidities, and death derived from the literature. Dr. Herman found a $91,000 gap in the cumulative per person undiscounted costs accrued by type 1 diabetes patients with poor vs. excellent glycemic control over 30 years. Factoring in estimated costs of MDI-only, pump-only, and pump with CGM therapies (see below), Dr. Herman estimated the incremental cost-effectiveness ratio (ICER) for each option. He deemed MDI-only ($3,835/QALY gained) and pump-only ($52,654/QALY gained) therapies “very cost-effective” compared to conventional therapies, but characterized pump with CGM as “maybe too expensive,” as it clocked in at $266,457/QALY gained – well above the acceptable threshold of $100,000/QALY gained. MDI with CGM was not presented. Dr. Herman asserted that pump + CGM could serve as a “reasonable alternative” for those struggling to achieve A1c goals with MDI or pump alone, but should not be recommended as a standard primary treatment for everyone with type 1 diabetes. Still, Dr. Herman acknowledged that these intensive therapies may confer better clinical outcomes, quality of life, and other metrics not measured in the study, which could make these options more cost-effective.
Dexcom’s Dr. Claudia Graham pointed out during the discussion that Dr. Herman’s analysis is “trying to simplify something that is not easy to simplify.” She explained that CGM outcomes can vary substantially by device type – a factor that is “often overlooked” in such cost-analyses and that could have affected the CGM + pump results. To this end, Dr. Graham claimed that she has seen certain cost effectiveness studies demonstrating an ICER <$100,000/QALY for combined CGM and pump therapy. Dr. Graham also suggested analyzing the cost-effectiveness of CGM alone, now that 25% of type 1s on CGM are MDI users. We agree that this would be a valuable segment to evaluate – a recent analysis of Dexcom’s DIaMonD study estimated the ICER for the G4 vs. SMBG to be $98,108/QALY for the population of type 1 MDIs with A1c ≥7.5% (cost effective by most country’s standards).
Moreover, on Dr. Graham’s point regarding simplification, we’re not sure it makes sense to lump together MDI-only, pump-only, and pump + CGM therapies into one broader “intensive therapy” category based off a single A1c achievement alone. By doing so, the reduced vascular events were identical for all three interventions, while the costs differed. As Dr. Herman himself acknowledged, at the very least there likely exist differences in hypoglycemia, quality of life, and other more immediate measurements between these therapy types, which would certainly play a role in a broader definition of cost-effectiveness. While we absolutely appreciate the attempt to characterize the cost-effectiveness of diabetes devices, we wonder if this approach might have been too generalized.
A Sanofi-sponsored poster displayed findings from a survey of people with type 1 diabetes, highlighting the need for oral adjunct therapies (namely, Sanofi/Lexicon’s SGLT-1/2 dual inhibitor sotagliflozin). The survey was sent to 2,084 adults with type 1 diabetes through dQ&A’s patient panel. Of the 1,313 respondents (63% response rate), 63% ranked time-in-range as very important to them, and yet only 17% were satisfied with how available therapies address time-in-range. Similarly, 56% of respondents felt that prevention of weight gain is very important, and only 17% were satisfied with how current treatments address this. Insulin, of course, is notoriously associated with weight gain. Oral adjunct therapies for type 1 diabetes – including sotagliflozin as well as AZ’s SGLT-2 inhibitor dapagliflozin and Lilly/BI’s SGLT-2 empagliflozin – significantly improve patients’ time-in-range and stimulate weight loss. At the ATTD-sponsored consensus meeting on managing DKA risk with these agents, Dr. Satish Garg highlighted weight loss as the most meaningful benefit to his patients with type 1 who start an SGLT regimen, while Dr. John Buse emphasized the greater quality of life that comes with spending more time in-range. In the survey commissioned by Sanofi, patients indeed ranked (i) weight loss and (ii) improved time-in-range as the most important attributes in a drug. The analysis concluded that sotagliflozin offers both. The candidate was filed with FDA for a type 1 indication in March 2018, under intended brand name Zynquista. A decision is expected by March 22, 2019, after an anticipated Advisory Committee meeting. AZ is also planning to submit SGLT-2 inhibitor dapagliflozin (Farxiga) to FDA by year-end; the company has already filed dapa for a type 1 indication in Europe and Japan. Lilly/BI just presented phase 3 data on empagliflozin (Jardiance) in type 1 diabetes at EASD 2018 – read our full coverage here. To be sure, thought leaders are enthusiastic about how SGLT inhibitors can improve type 1 diabetes care, fulfilling key unmet needs (this was a major theme at both ADA and EASD this year). FDA and other regulators will surely express concern over the DKA risk associated with the class (when used in type 1), and while this is a very real safety issue, we continue to believe that it can be managed with careful monitoring and thoughtful patient selection. We’ll be curious to see how manufacturers respond to DKA concerns at the impending Advisory Committee meeting – this has not been scheduled but is almost guaranteed, as we understand it.
The poster also listed the highest-priority unmet needs reported by patients with type 1 diabetes. These were, in order: simple and predictable diabetes management, A1c reduced or maintained at target, daytime blood glucose between 70-180 mg/dl, reduced mental effort needed to manage diabetes, overnight blood glucose between 70-130 mg/dl, prevention of weight gain, fasting blood glucose between 70-180 mg/dl, flexibility with exercise and diet, postprandial blood glucose <180 mg/dl, and prevention of hypoglycemic events.
A Sanofi-funded study found that patients with type 2 diabetes are more frustrated regarding their slow treatment progress than HCPs are aware of and are willing to do more to reach their A1c goals than HCPs expect. dQ&A surveys were completed by 305 patients and 240 HCPs (n=160 PCPs, n=80 endocrinologists). Patients were ≥18 years-old, on basal insulin, and not currently taking prandial insulin or other injectable anti-diabetes medications. Notably, 37% of patients reported being “very willing” to do more to reach their A1c goal faster, while only 16% of HCPs thought this to be true. Likewise, a whopping 46% of patients indicated that they would be willing to visit their HCP more often, whereas just 17% of HCPs believed this of their patients. Similar patterns were observed for willingness to try a different injectable and willingness to make multiple therapy changes. In perhaps one of the most jarring disparities, 80% of patients indicated frustration regarding their slow progress in reaching A1c goals, while only 18% of HCPs believe this to be true of their patients. Despite this meaningful difference, 61% of both patients and HCPs (perceived patient desire) wanted to reach A1c goals faster. Patients were less optimistic regarding how quickly they could achieve their A1c goals – 17% of patients (vs. 33% HCPs) believed glycemic control could be obtained in <6 months, and 28% of patients (vs. 20% HCPs) believed this would take >1 year. We wonder how patients can be empowered to share these feelings with their providers – we expect that HCPs might suggest more aggressive treatment options if made aware of their patients’ frustration and willingness to do more. If nothing else, this poster calls attention to the tremendous gap in HCP-patient understanding and communication.
Maintaining A1c over the long-term was the only priority that both patients (62%) and HCPs (45%) listed as their top three treatment priorities; patients also cited staying healthy (44%) and avoiding weight gain (43%), while HCPs included avoiding side effects (55%) and affordability (53%).
The investigators flagged over-basalization as a concern confirmed in the results. HCPs indicated 100 U as the mean maximum daily dose acceptable, and 12% of HCPs reported that they would even consider prescribing ≥200 U. We wonder if HCPs would be less likely to increase basal doses (and more likely to add a second therapy) if they were aware of the high proportion of patients “very willing” to try a different injectable (45%) or to make multiple changes (40%).
Kaiser Permanente’s Dr. Mary Reed presented positive preliminary data showing the positive impact of patient portals on healthcare utilization in patients with diabetes. Upon implementation of the patient-facing portal to the EHR, patients with diabetes saw higher office visit rates, lower emergency department visit rates, and reduced preventive hospitalizations – this last outcome was especially prevalent in patients with diabetes and other complex conditions. Moreover, the vast majority of patients indicated that the portal was convenient, facilitated receiving faster answers, helped prepare them for clinician visits, and integrated information with other care. Roughly one-third of patients reported that portal use improved their health and only 17% reported having concerns regarding online privacy. Upon addition of a mobile patient portal, the frequency of portal use increased, suggesting an additive effect. Importantly, Dr. Reed noted “reassuring patterns” in mobile patient portal use across demographics. Whereas she might have expected the technology to enhance existing disparities in healthcare access, Dr. Reed explained that white patients were actually the least likely to use the mobile patient portal compared to other races. Moreover, those residing in low SES neighborhoods were more likely to use the mobile patient portal than those in high SES neighborhoods, and patients with lower adherence were more likely to access the mobile patient portal than those with higher adherence. Given these findings, Dr. Reed believes this technology may help reach populations that are traditionally more difficult to engage. In fact, she suggested the tool might even be considered “some kind of leveler” that could prove “useful in narrowing disparities.” We look forward to the full publication of these results and would be very interested to know what aspects of the patient portal were responsible for the improved outcomes.
Oracle’s Dr. William DuMouchel touched on several challenges to evaluating the safety of diabetes drugs in the real world, including polypharmacy and missing information. The vast majority of diabetes patients take more than one prescription medication, whether sequentially or simultaneously, so when someone presents with an adverse event, it’s difficult to determine which agent or which drug-drug interaction is to blame. (As an aside, we’ve heard that healthcare providers in the US are reluctant to prescribe combination therapies because of the potential hurdle in teasing apart adverse effects; the irony is that fixed-dose and fixed-ratio combination products offer a milder side-effect profile compared to monotherapy.) FDA manages a database of adverse events for all drugs on the market, and the agency has collected >10 million reports to-date for >5,000 products, according to Dr. DuMouchel. He called this method of data collection “spontaneous reporting,” because individuals choose if/when to mention a treatment-related adverse event to FDA. This data is analyzed against a mock control group of all other patients who have reported an event to the system for any therapy other than the one in question. Suffice it to say, this analysis is imperfect. Dr. DuMouchel pointed out that there’s no way to know whether a patient is reporting a side-effect after one dose of a medicine or after a full year of treatment – that is, FDA cannot through this method alone match adverse events to length or amount of exposure. Other important information is usually missing as well, such as what other therapies the patient was taking concurrently. Dr. DuMouchel proposed multivariate Bayesian logistic regression as a way to account for some of these factors statistically. Toward the end of his talk, he raised an interesting question around granularity: Should 10 issues be lumped together, or each considered as their own separate safety concern? In our view, the diabetes field would certainly benefit from greater consensus on this question, and others like it. As an example, DKA events were categorized differently in Lexicon’s inTandem program for sotagliflozin in type 1 vs. AZ’s DEPICT program for dapagliflozin in type 1, which was the subject of an NEJM letter published earlier this year. Ultimately, our sense is that DKA is a class effect of SGLT inhibitors used in type 1 diabetes, but regardless, it would be helpful for industry, regulators, clinicians, and patients alike if there was standardization of what counts as a DKA event. On a similar note, the debate continues over whether amputations are of concern when prescribing SGLT-2 inhibitors in type 2 diabetes care. J&J’s Invokana (canagliflozin) has a black box warning for lower limb amputations, but no other SGLT-2 product does in the US, since CANVAS was the only outcomes trial to show a significant imbalance in amputations between study drug and placebo. That said, some thought leaders have pointed out that amputations were measured and adjudicated differently (i.e. more rigorously) in CANVAS vs. EMPA-REG OUTCOME (for Lilly/BI’s Jardiance). This all goes to show that the diabetes field needs strong real-world evidence to address key safety questions (as well as just standardization within big trials); Dr. DuMouchel’s presentation was in fact quite timely, as these controversies rage.
In an overview of natural experiments to evaluate healthcare reform, Harvard’s Dr. Franklin Wharam mentioned several adverse impacts of high-deductible health plans (HDHPs) on people with diabetes. Switching to a HDHP leads diabetes patients to delay their first outpatient visit for an acute complication. Dr. Wharam described how low-income patients who switch to a HDHP show an increase in emergency department visits for acute diabetes complications, but this trend doesn’t carry over to high-income patients, which confirms that seeking care is related to the relative out-of-pocket cost for the individual. He also shared new data – to be published next week – showing that after switching to a HDHP, diabetes patients delay seeking care for a major sign or symptom of disease by a mean 1.5 months, delay major diagnostic tests by a mean 1.9 months, and delay major treatment by 3.1 months (all p<0.05). It wasn’t completely clear what comprises a “major sign or symptom” or “major treatment,” and for those details, we look forward to the full publication. We’d expect high-deductible insurance to deter people with diabetes (or really, any chronic disease) from important clinical care. But surprising or not, it’s important to have this data on the books to inform ongoing healthcare reform. Dr. Wharam noted that the NEXT-D2 project at UCLA is investigating the impact of the ACA and Medicaid expansion on diabetes outcomes, and we’re eager for these insights.
Public health expert Prof. Martin White discussed two policy initiatives in the UK to combat diabetes/obesity: (i) a soda tax and (ii) industry-led efforts to reduce junk food at grocery store checkout. The soda tax was announced in March 2016 but wasn’t implemented until April 2018, giving manufacturers two years to reformulate their drinks to have <5 g/100 ml added sugar OR <8 g/100 ml of sugar (there are two tiers at different tax levels). According to Prof. White, some beverage companies are taking the opportunity to make high-sugar drinks somewhat healthier, but they also diversified their portfolios and introduced new products, some of which are in the “exempt category” – milk-based drinks, pure fruit juices, and alcoholic beverages (these can still have >5 g/100 ml or >8 g/100 ml added sugar and not be taxed). Prof. White emphasized that it’s difficult to predict how industry will respond to top-down public health policy, as the UK soda tax example shows. Nonetheless, we’re pleased to see the UK join Mexico and a handful of US cities in imposing an excise tax on sugar-sweetened beverages. These policies have shown early efficacy in reducing soda consumption and increasing clean water consumption; longer-term, these taxes are likely to make an impact on the diabetes/obesity epidemic, based on what history tells us happened with cigarette taxes/smoking. Prof. White advocated that “we need a population approach to change the trajectory of diabetes globally,” and SSB taxes could be an important piece of that. Duke’s Dr. Kelly Brownell has expressed optimism on this front: “Soda companies are losing the tax fight all around the world,” he wrote to us last year. “It’s just a matter of time until these taxes are very common” (the taxation wave is not without its setbacks, however and we don’t think the pre-emption work was expected). Turning to his second example, Prof. White described high-profile advocacy campaigns in the UK to “dump the junk,” or to remove junk food from grocery store checkout counters (sometimes replacing it with healthy food options). He suggested that this was mainly motivated by consumer demand (mothers were having a hard time saying no to children who wanted candy at checkout. Overall, one month after supermarkets had introduced a checkout food policy, last-second purchases of unhealthy foods decreased by 17%, and this effect was sustained out to 12 months (15% decrease from baseline). Prof. White called this “quite a nice result.” Thus far, supermarkets have volunteered to “dump the junk,” and the UK government has proposed an official regulation on junk food in checkout aisles as part of its initiative to curtail childhood obesity.
Harvard Medical School’s Dr. Jessica Franklin described her team’s ongoing efforts to replicate 30 RCTs using RWE. Dr. Franklin believes there are important merits to comparing RCTs and RWD, claiming that this kind of analysis is the only way to assess the success of the entire research process and can shed light on the types of questions that are best answered with RWD. As part of an FDA-funded project, her team has identified 40 replicable RCTs, including six large CVOTs, with the ultimate goal of replicating 30 (she expects to eliminate some due to feasibility). Currently, Dr. Franklin is just starting the implementation process. She underscored that the final decision to undertake an RWE study is informed only by the study power and patient characteristics, not the RCT results. A similar protocol will be used for all replications, and Dr. Franklin hopes to implement each study in multiple databases whenever possible. By replicating RCTs with RWE, Dr. Franklin aims to not only determine which design/analytic choices make healthcare database studies interpretable for casual conclusions, but also to prepare for potential prospective analyses of ongoing RCTs. Hopefully, Dr. Franklin’s work will help to build confidence surrounding RWE. As Dr. Franklin emphasized during the discussion, her research assumes the RCT as the reference and is not intended to call into question the validity of previous RCTs; rather, her purpose is to identify sources, standards, and situations in which RWE can prove suitable for regulatory decision making.
Dr. Franklin explained that “one of the most challenging parts of this project so far” was reaching a definition for agreement between RCTs and RWE. Ultimately, her team decided to consider two definitions: (i) Regulatory agreement, which requires the RWD study to come to the same conclusion as the RCT based on statistical significance of the effect estimate; and (ii) Estimate agreement, which requires the RWD effect estimate to reside within the 95% CI from the RCT.
Dr. Franklin explained that several prior literature reviews have attempted to compare published RWD analyses with RCTs but there are some issues with this approach, including: (i) possible publication bias (i.e., investigators are more likely to publish when there is a significant effect); (ii) the RCTs and RWD focus on different populations; (iii) the RWD studies are of varying methodological quality; and (iv) if RCT results are published first, RWD studies that conflict the results may be suppressed.
Open Discussion to Inform the ADA Consensus Publication
Goals for The ADA Consensus Statement on RWD; Open Discussion Touches on Industry Partnerships, Applications of RWD in Diabetes, and Pragmatic Trials
In the final session of the meeting, co-hosts Dr. Edward Gregg (CDC, Atlanta, GA) and Dr. Kamlesh Khunti (University of Leicester, UK) summarized the goals for the ADA consensus statement on use of real-world data and opened up the floor for discussion. Dr. Khunti explained that the consensus statement aims to provide methodological considerations for assessing real-word evidence (RWE), including best practices for (i) analyzing RWD to inform study design; (ii) RWD interpretation; and (iii) application of RWD results. Furthermore, the statement intends to inform policies for grant applications and study proposals surrounding RWD collection and to develop a standardized format and requirement for scholarly publications when considering the manuscript review of studies incorporating RWE. Finally, the authors intend to establish a framework for engaging regulatory bodies in the utilization and interpretation of RWE. Dr. Khunti shared a detailed manuscript timeline that kicks off immediately – writing assignments and the manuscript outline were finalized right after the meeting wraps up! Dr. Khunti expects section drafts of the manuscript to be completed by January 15, 2019 and ultimately anticipates publication in Diabetes Care in June-July 2019. Dr. Gregg emphasized that, while the statement will focus on the methodology and structure surrounding use of RWD, specific issues related to diabetes will be integrated, including considerations for the comparative effectiveness of therapies in reducing diabetes morbidity and enabling primary prevention. Meeting participants engaged in a lively discussion – see below for some of the key themes that emerged.
On the intended consensus statement audience: University of Chicago’s Dr. Elbert Huang raised the important question of the need to identify the intended consumer of the consensus statement. As he put it, “if I were a young investigator, this would be a one-stop shop for all the standards and methods related to the use of RWD.” Dr. Gregg explained that the writing group is interested in reaching scientists “involved in the spectrum from clinical effectiveness to public health.” Dr. Huang agreed, and added that perhaps more general methodologists, even those not directly in the diabetes space, would also benefit from the standards. Another participant cautioned against getting too technical – she rightfully reminded the audience of the “broader audience” involved, including clinicians and policy makers – and suggested a section on how data scientists might teach non-experts to best interpret RWD literature. On a similar note, an audience member mentioned how clinicians are reading fewer journals and that perhaps a series of podcasts distilling the information would prove to be an effective means of disseminating the standards – we’d love to see this as an addition. A Glooko employee reminded the audience that industry looks at Diabetes Care as a “leading publication,” proposing that the statement could be used to “nudge folks in the clinic and industry” into action. We love the idea of the statement as an opportunity to prompt industry members towards data standardization and interoperability. Indeed, PCORI’s Dr. Joe Selby advocated for the role of researchers to “push public consciousness” about the need to support data linkage, specifically by educating health systems on the “attractiveness” of linking data.
On areas that were not discussed: Although the meeting’s agenda covered significant ground, the audience members had plenty of thoughts on issues that were not discussed. For example, how might industry partner with academia to advance the field? One participant provided the example of CGM, explaining that it affords “so much more richness of data,” but that there still exists a need for standardized data entry into the EHR. He also urged the audience to consider how industry partnerships might be encouraged to study populations not typically included in RCT samples. To this end, Allscripts Analytics’ Dr. Fatima Paruk asserted that “industry is almost certainly willing to collaborate there,” but that more guidance from the FDA surrounding data governance is a necessary first step. The opportunity to explore outcomes in traditionally underrepresented populations was one of the widely touted benefits of RWD – given the heterogeneity of the diabetes population, we think it’s extremely important to emphasize this approach. We were also interested to hear an audience member note the promise of RWD to consider healthcare resource utilization – as he pointed out, this area is “very hard” to examine in RCTs and “extremely important.” One participant suggested the use of RWD datasets in not only learning more about treatments, but also in better understanding the mechanisms of diabetes and perhaps identifying subgroups of patients most likely to respond positively to certain medications (as we’ve recently seen from Sanofi, Novo Nordisk, and others).
On using RWE to address questions in diabetes care and prevention: Many participants focused on the opportunity to leverage RWD for the better understanding of underrepresented populations and social determinants of health. One attendee suggested using RWD as a “platform for tailored interventions,” particularly in establishing community-building efforts among vulnerable populations. Another member proposed that RWD might be used for head-to-head studies, as traditional RCTs tend not to include active comparators (of course, there is also a push for studies to include active comparators to better reflect standard of care). Importantly, an attendee urged Drs. Gregg and Khunti to consider a caveat: While guidance is critical in promoting better practice and ensuring high-quality science, it can also unintentionally constrain the creativity necessary for addressing “tricky questions.” Dr. Gregg was quick to agree, responding: “We want to encourage innovation and creation, but also want to establish some recommendations that people can point to.” As he explained, there are likely “irresponsible analyses” of RWD that threaten its credibility – it’s therefore of the utmost importance that standards exist by which journal editors can review RWD consistently.
On the challenges surrounding pragmatic trials: PCORI’s Dr. Joe Selby asserted that “the idea of pragmatic trials as part of RWE generation is very important.” Still, he underscored the struggle to find sites “actually willing to promote” pragmatic trials, admitting that many physicians understandably “don’t want to be hassled.” Dr. Khunti agreed, claiming that “the biggest barrier” to pragmatic trials for HCPs is the timely consent process. To this end Dr. Selby argued that some simple questions can be answered through alternative recruiting methods leveraging the EHR and patient portals. More broadly, Dr. Selby advocated to change the methods by which pragmatic trials are conducted, suggesting that having a health system responsible for boosting recruitment “could make a huge difference.” He further explained that a model in which the health system contributes RWD for studies might provide “more ground” for a pragmatic, randomized trial. Dr. Gregg summarized the issue nicely: In one scenario, studies are conducted based on available big data sources where power is not the problem, but pulling out textured data is difficult; in the other, pragmatic trials can answer very specific, critical questions but are limited by low enrollment. Dr. Selby asserted that by involving the health system, these scenarios would “merge naturally.”
-- by Payal Marathe, Maeve Serino, Brian Levine, and Kelly Close