OHDSI Home | Forums | Wiki | Github

OHDSI Face-to-Face at Columbia May2-3: Community study-a-thon

General
#21

Hi everyone! Hope it’s not too late, but wanted to toss out another 2 ideas about cardiovascular health based on recent guidelines

  1. The 2016 ESC/EAS (Europe) and 2017 AACE (Endocrinology, American) guidelines both note that patients with elevated triglycerides may have substantially increased ASCVD (Artherosclerotic Cardiovascular Disease) risk, even those who are already on statins for LDL lowering. They recommend considering fibrates for these patients to improve CV outcomes but there’s no randomized control trial that has proven this and just a few studies that support this. Therefore, the study would be:
    To compare the risk of ASCVD between patients with diagnosis of hyperlipidemia (HLD) on statins and elevated triglycerides who are treated with fibrates and patients with HLD on statins and elevated triglycerides not on fibrates. Estimate population-level effect of exposure on the hazards of outcome (ASCVD) from 1 day after exposure start to 0 days after exposure end (on treatment, or ITT)

  2. I’ve heard rumors from multiple sources that new ACC/AHA guidelines will be coming out at the upcoming ACC conference, and these new guidelines will go back to cholesterol targets instead of just grouping people into risk categories and doing high or low-intensity statin (as in the 2013 ACC/AHA guidelines). Essentially that numbers matter, not just the risk category. So another idea for study related to this would be:
    To compare the risk of ASCVD between patients with diagnosis of HLD on statins with ideal cholesterol levels and patients with HLD on statins with elevated cholesterol. Could also stratify further across different groups of cholesterol levels to see how risk changes (i.e. for every 50mg/dL) but that’s like running multiple studies. Estimate population-level effect of exposure on the hazards of outcome (ASCVD) from 30 day after exposure to statin (about how long it takes the level to drop on a statin) to 0 days after exposure end (on treatment)

#22

Hi,
Quick question - what does C/D stands for? Sorry didn’t get that one. Thanks

OHDSI F2F NYC 2018 - Open sharing tools for collaboration
#23

Hi all,

Here is my research question:

Research Question: To determine the comparative risk developing malignancy in patients with moderate to severe rheumatoid arthritis treated with tofacitinib and tumor necrosis factor inhibitors (TNFi).
Rationale: There is a significant concern for developing malignancy in patients treated with TNFi, other non-TNFi biologics, and targeted small molecules. To date, only limited evidence to guide clinical practice.
Patient population: new users of 1) tofacitinib or 2) TNFi for moderate to severe rheumatoid arthritis
Outcome: Time to develop any type of malignancy.
Model: Cox regression model

#24

Thanks @rchen for posting this question. Canagliflozin is a Janssen product and I and others on my team have done a good bit of work on to examine this product’s real-world safety and effectiveness. In fact, we’ve recently completed a study that addresses the amputation risk question that you identified. The study was registered on clinicaltrials.gov (https://clinicaltrials.gov/ct2/show/NCT03492580). We have also publicly posted the full protocol and all associated source code in the OHDSI github repository (https://github.com/OHDSI/StudyProtocols/tree/master/AhasHfBkleAmputation). The results are under review, and we plan to share our experience with this study as soon as the results can be made available. So, while I encourage anyone interested in this topic to reach out to me and would be happy to see this study executed across other sites across the OHDSI network, it would be ‘cheating’ for us to pick this study idea for the F2F meeting, since we want to take an idea from scratch all the way through design, protocol development, code implement, and network study execution in the two days while we are together, and too much of the work has already been done.

#25

Thanks @Evan_Minty for posting this question. Yes, I agree aspirin exposure may be difficult across many data partners. I also wonder how we would work through thinking about prevalent exposure status in this case, since it sounds like the T/C cohorts would be indexed off the bladder cancer diagnosis. Nonetheless, it sounds clinically important, so it seems worth considering the question as we vote for our study choice for the F2F.

#26

Thanks @Vojtech_Huser for posting this question. I hadn’t seen this trial, and it seems potentially quite important from a public health perspective that an antibiotic that is so commonly used could potentially increase mortality, cardiovascular death, and cerebrovascular disease. It’s also a nice example of a randomized trial that was performed which potentially could be valuable to replicate in real-world data, where the exposed population in clinical practice may be much broader than the subjects enrolled in the trial. The trial is only 4400 patients, whereas I expect we have millions of exposures to clarithromycin and amoxicillin across our network. I think it is noteworthy that the effect sizes are so small (HR=1.1 could easily be within the range of systematic error that we may estimate through empirical calibration). We also have the concern about the adequacy of capture of mortality across our network, and we likely have variable follow-up with a much smaller proportion being observed for 10 years. All that said, seems a worthy topic to consider in our community vote when picking a topic. Thanks!

#27

Thanks @christian_reich and Elad. I’ve heard from several folks in the community about interest in studying infections with immunosuppressive agents, including chemotherapies and biologics, so it seems like a good area to have some focus. I don’t know anything about brentuximab and I don’t have much exposure to the drug in the datasets I have access to (n<1000 patients), but maybe others in the community have more sample to work with. I would think we’d need to be careful about defining an appropriate comparator, since in some databases, particularly claims data, many antineoplastics are non-specifically coded as procedural administrations and we may not know the specific active ingredients. I know @rchen is doing some good detective work for OHDSI’s collaboration with NCI to try to characterize the boundaries for how cancer treatment exposure is captured, but I am not sure how much findings there will generalize across the rest of the OHDSI community.

#28

Thanks @rijnbeek, Guy, and Katia. This is a very well-written project specification, and while I don’t know much in the asthma space, you make a compelling argument for why understanding the effect of inhaled corticosteroids among patients on LAMA on mortality could have a large public health impact.

As with some of the other project ideas, one concern I have is the adequacy of capture of death information across many of our data partners. Particularly for this problem where any effect difference we observe is likely be small, and may be well within the systematic error due to outcome misclassification. Forgive my ignorance for a potentially basic clinical question, but for LAMA and ICS, are we comfortable that dispensing records from pharmacy systems or prescriptions written in EHRs are sufficient proxies for exposure (and inferring concomitant use) or would there be consider about patient’s non-adherence or switching/augmentation of asthma treatments could also introduce systematic error due to exposure misclassification?

In any case, seems like a good question to add to our ballot for voting…

#29

Thanks @aostropolets, so if I understand the question, you are basically asking: among patients with COPD, is starting and stopping ICS better or worse than never starting ICS at all for the risk of asthma exacerbations and mortality?

I am struggling to think about how we would define the index date for the target and comparator cohort in this case. In the target cohort, for many databases, we can infer when ICS exposure started based on dispensing or prescription written, but I suspect we will have much less precision around a withdrawal date (though we can see when dispensings discontinued or no more refills ordered). For the comparator cohort, if the COPD patients aren’t exposed to ICS, is there some other treatment we would expect to see that we could index off of (LAMA or LABA)? Is the COPD Grading C/D something that we can infer based on medication history?

Anyway, thanks for the submission, I’ll add it to our ballot to vote on.

#30

Thanks @abedtash_hamed. It’s an interesting use case to consider a ‘known side effect’ that may have a lower real-world risk than what was previously observed in trials, because clinicians are now imposing their own risk minimization strategies.

I think you raise an important type of problem that I haven’t really wrestled with before, namely how do you examine the short-term effects vs. long-term effects of a product. If there was some active comparator to use in the study, I could imagine specifying a fixed short-term window like ‘to compare risk of CV events amongst patients with migraine who are new users of triptans vs. patients with migraine who are new users of indomethacin, during the period from 1d from exposure start to 180d from exposure start’. And then, to examine a longer-term effect, a question could be specified like: ‘to compare risk of CV events amongst patients with migraine who are new users of triptans who remain continuously exposed for over 1 year vs. patients with migraine who are new users of indomethacin who remain continuously exposed for over 1 year, during the period from 1 year from exposure start until end of observation’. (note, in this second question, since we require the 1 year of continous exposure, we can’t use that time as part of our time-at-risk).

What I can’t quite work my head around is how to compare short-term exposure with long-term exposure. It’s not clear how to reconcile the time-at-risk, since length of exposure is determining the comparator. Perhaps that’s what we could sort out at the F2F, if this problem is selected in the voting.

#31

Thanks @jill_hardin, given the extremely high prevalence of use of levothyroxine, it seems like any effect of dosing could be quite important from a public health perspective.

One of the challenges that I don’t think we’ve overcome as a community is how to think about the time-varying nature of dosing. In the claims data I’m most familiar with, we can try to infer average daily dose from each pharmacy dispensing record using the strength of the product * quantity / days supply. But we need to think about how to handle successive dispensings and how to handle dosing during gaps and periods of overlap. We’d also need to think about how to handle patients who may switch from low-to-high dose or vice versa. Perhaps there’s enough people who start on and remain on low dose who can be compared with people who start and remain on high dose, but I don’t know one way or the other if ‘stable dose’ users are generalizable to the population at large.

Anyway, you’ve done a good job of laying out the problem and rationale, so it’ll be up to the community to vote .

#32

Thanks @sm2206, there’s been a lot of work examining the effects of proton pump inhibitors in various observational databases, and I agree that many of the studies I’ve seen in the literature have had some serious methodological shortcomings, so it’d be nice to see what evidence we could generate if we applied the OHDSI community best practices to the question. It would be interesting to see how a comparison of PPI vs. H2RAs using a large-scale propensity score matched new user design across the OHDSI network would stack up with studies using different covariate adjustment strategies, such as the recent paper by Lazurus et al.

One challenge I anticipate is exposure misclassification, since both PPI and H2RA are available over-the-counter and may not be fully captured in many of our data partners. So, there could be some tension between using the most recent and timely data vs. using the time period where prescription coverage offered most exposures. I suspect this could be worked out if this question is selected in voting.

#33

Thanks @rchen. I appreciate the vantage point of looking the clinical guidelines as a source of inspiration for where we can make important contributions. It seems there’s areas of opportunities for determining which recommendations may exist which are not fully supported by high-quality evidence (or could be further augmented with real-world observational studies), and also I’ve seen several areas in guidelines where recommendations cannot be made because of the lack of availability of any evidence to support alternatives. It seems the effect of fibrates to supplement statin use in reducing cardiovascular risk should be better understood if its going to be part of recommendations for how to handle specific subpopulations with uncontrolled hyperlipidemia. I’ll add this to our ballot so we can see how the rest of the community feels about it.

#34

Thanks @BridgetWang. I’m very excited by this question. There is an ongoing RCT that is scheduled to report out next year to address this question (https://clinicaltrials.gov/ct2/show/NCT02092467), so OHDSI has an opportunity to ‘predict’ the RCT results using our observational data network, which has important methodological value for demonstrating the potential role of real-world data to augment, and in some circumstances, replace the need for RCTs, as has been proposed for consideration by the FDA and as part of the 21st Century Cures Act. The trial involves an active comparator, so we can define target and comparator cohorts for these drugs quite easily. Also, both trial outcomes (cancer, MACE) should be observable in most claims and EHR systems, as are most of the study inclusion criteria. There seems to be a lot of uncertainty about the safety of biologics in RA even after a lot of small clinical trials, so examining these outcomes in the real-world clinical practice across the OHDSI network could be an important contribution to the field.

Let’s see if the community is as excited as I am to pursue this question, I’ll add it to the ballot.

1 Like
#35

Agree. Thanks @Patrick_Ryan or the link to the Clinical trial. Just wanted to share a thought on why I think this will be a cool study.

From the molecular perspective, the TNF is one of the most notorious genes. Its promotor region on chromosome 6 has a plethora of variations leading to all sorts of disorder. In addition, a recent population level (kind of) exome sequencing of ~60,706 humans suggest variations in the enhancer regions of TNF that might contribute to dysregulation at the transcriptional level http://exac.broadinstitute.org/gene/ENSG00000232810. Our understanding of variations is generally gleaned by sequencing the patients with diseases - but not by how well or poorly they response (or would have responded) to a treatment.

OHDSI network study (across the massive datasets we have), therefore, might shed some light on the nature of patients according to their response to TNF inhibitors. Since we’ll know the features (comorbidities, drug etc) associated with such patients who do or do not respond to TNF inhibitors, therefore it might guide us (in future) on what type of patients one should sequence to look for variations and disease biology.

#37

Thanks @Patrick_Ryan . I am inclined to agree with you with the one caveat that its unusual for individuals to have a prolonged OTC exposure to PPIs without ever having it prescribed at some point. This might impact measurement of total exposure but there are ways we could overcome this. Looking at an era effect before they were available OTC might be one potential option.
The other thing that I want to emphasize when considering studies, is the potential impact for change in practice. CKD now affects 1 in 7 adults in the United States - and is increasing in prevalence across the globe. The etiology of the overwhelming majority of cases is not identified. Further PPI use is becoming increasingly more prevalent worldwide and being able to quantify a relationship here would have important public health implications globally. There is perhaps a subset of patients with a significantly higher risk where PPI avoidance would be particularly important.

#38

In considering the F-2-F choices, with the 1st proposal we would ideally need pathological stage (to reduce variation we may consider only looking at stage 1 or localized disease). At the VA that data is not currently in the OMOP CDM but we could link back to it from the CDW. Doing this at the meeting; however, would likely be problematic due to the VA’s data security (at least for me). I’m wondering who else have cancer stage in their OMOP data.

Stephen

#39

@deppen We’re working on integrating NLP-derived TNM staging data from pathology reports right now.

#40

Response from Guy:

Thank you for raising this question.
As for all maintenance treatments for chronic diseases, also in asthma patients, non-adherence (non-compliance) is an important problem.
Therefore, dispensing records from pharmacy systems or prescriptions written in EHRs are only proxies for exposure (no guarantee of exposure, and appropriate use / correct inhalation).
Dispensing records are somewhat better than prescriptions.

#41

Additional response from Katia:

With regard to adequacy of capture of death information: We are interested in all cause mortality and not necessarily death due to asthma (exacerbation) as this is indeed not available in all databases;
With regard to exposure and potential non-adherence or switching: There is the potential of underreporting of exposure as we only use prescription or dispensing data however this would not necessarily imply that it would be differential between LAMA (without ICS) vs LAMA+ICS. Also we could try to control for eventual bias in adherence by calculating adherence at time of study start (for instance MPR in previous year) and match on MPR.

t