OHDSI Home | Forums | Wiki | Github

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


You’ve probably seen @MauraBeaton’s announcement that the next US OHDSI face-to-face will be hosted by Columbia University in New York City on May2-3. Several folks have been asking about what is planned for that 2-day event. Here’s your quick answer:

We’re going to do a study together!

@mvanzandt and others in the community had suggested that a good community activity would be to bring everyone together to focus on one specific common goal: to generate reliable evidence on one clinical problem. We think that’s a great idea, so we’re going to do exactly that!

More specifically, our plan over the next 3 months prior to the face-to-face is to solicit the community for good candidate clinical questions that require evidence that we can generate using a comparative cohort design. The question could be a safety surveillance question, like what @jon_duke answered last year : ‘does levetiracetam increase the risk of angioedema, relative to phenytoin?’ or it could be a comparative effectiveness question, like the study @estone96 is leading: ‘does alendronate reduce the risk of hip fracture more than raloxifene?’. The questions do not necessarily need to examine the effects of drugs, but rather can represent any exposure to any intervention that is observable in our observational databases. For those who have taken the population-level effect estimation tutorial, you’ve heard @msuchard , @schuemie and me promote a general framework to define these types of problems using 5 core elements: 1) a target exposure cohort, 2) a comparator cohort, 3) an outcome cohort, 4) a time-at-risk period, and 5) a model specification.

At the face-to-face meeting, those in attendance will select 1 clinical question to focus on for the duration of the session, based on the community feedback and data availability. Then, the real fun will start: once aligned on the common problem, we will break out into groups to tackle the specific components required to design and implement the study. One group will focus on defining the target and comparator cohorts, ensuring that we have an adequate capture of the exposures of interest and proper application of whatever inclusion criteria are necessary to define the appropriate study population. A second group will create and validate an outcome cohort definition. A third group will produce a list of negative control outcomes that will be used for empirical calibration. A fourth group will make the analytic design choices and prepare an R package that implements the study using the CohortMethod package. These activities will all be pulled together into an open community protocol…and that initial draft will need to be completed on Day 1…then we’ll break to have fun in NYC.

On Day 2, we will get the study package running across the OHDSI network. If you have data in the OMOP Common Data Model v5 and want to participate in a network study, you’ll be able to kick off the study in the morning and (hopefully) have results to share with the community by the afternoon. While the study is running, we will hold a community brainstorm to discuss what we as a community can do to improve the validity and efficiency of observational research. From there, we will review the study diagnostics from all the participating data partners, and with any luck, have a final protocol and study results to share with the world!

That’s right, our goal for this face-to-face is to go from idea to results in two days (or less)! While this may seem incomprehensible in some circles, I’m confident if we work together as one community, we will be impressed by how much we can truly accomplish.

So who should come to the OHDSI F2F @ Columbia? Well, we want you there if any of the following describes you:

  1. You are passionate about the clinical questions that are to be studied, and can contribute to the clinical understanding of the problem and the synthesis of the evidence we generate with what is already known.
  2. You want to contribute to the epidemiological design of a clinical study, and see yourself leading the definition of exposure/outcome cohorts based on your clinical understanding and knowledge of observational databases.
  3. You want to apply your statistical expertise to the design of a population-level effect estimation study, and plan to contribute to the implementation of a R package that parameterizes the CohortMethod and associated OHDSI tools into an end-to-end study.
  4. You have observational data in the OMOP Common Data Model v5 format, and want to execute a network study against your data and share aggregate summary results with the community to advance our collective evidence about the question of interest.

Columbia’s Department of Biomedical Informatics is a wonderful venue for holding this intellectual ‘sprint’, but alas, as with most of New York City, space is at a premium. So, we will have to limit the number of participants for the OHDSI F2F. @MauraBeaton will be sending out a notice to allow you to express interest if you’d like to register and we’ll do our best to accommodate as many as we can.

I’m looking forward to collaborating with you all to generate reliable evidence that can meaningfully improve health. Happy hacking!

We are about 2 months away from the OHDSI Community Face-to-face meeting, hosted by Columbia on May2-3. If you want to be contribute to join and participate in the fun, please add your name here: https://www.ohdsi.org/events/2018-ohdsi-face-to-face/

As I posted earlier on this thread, our plan for the F2F is to all come together to design, execute, and report out a network study (in 48 hours or less). To pull this ambitious goal off, we need to identify a compelling research question that can be adequately addressed using our OHDSI data work with a comparative cohort design.

So, I have a action for all of you in the community: if you had a good research question that you’d like to see all of OHDSI come together to tackle, please propose it on this thread!

To help you get started, all you need to do is fill in this simple ‘mad libs’-style template to specify your research question:

To compare the risk of [Insert the name of your outcome of interest here] between [Insert the name of your target exposure here] and [Insert the name of your comparator cohort here] , we will estimate the population-level effect of exposure on the [Insert the metric of your analysis model here: hazards for Cox/ odds for logistic / rate ratio for Poisson] of the outcome during the period from [Insert your time-at-risk start: e.g. 1 day after exposure start] to [Insert your time-at-risk end: e.g. 30 days after exposure end] .

Examples from prior OHDSI network studies for added motivation:

Perhaps you are intrigued by a potential safety concern that FDA has posted on its website, as was @jon_duke when he asked:

“To compare the risk of angioedema between new users of levetiracetam and new users of phenytoin , we will estimate the population-level effect of exposure on the hazards of the outcome during the period from 1 day after exposure start to 0 days after exposure end”.

Perhaps you are a clinician treating patients with a particular disease and wondering which treatment option has the best outcomes, as was @estone96 when he asked:

“To compare the risk of hip fracture between new users of alendronate and new users of raloxifene , we will estimate the population-level effect of exposure on the hazards of the outcome during the period from 1 day after exposure start to ***all time after exposure start (intent-to-treat)***”.

Perhaps you are a researcher who is fascinated by the heterogeniety we observed in the OHDSI treatment pathway study that @hripcsa led, and want to dig deeper into which treatment sequences actually yield better results, @rohitv initiated with his study:

“To compare the risk of HbA1c reduction, myocardial infarction, and eye disorders between patients who switch from metformin to sulfonylureas and patients who switch from metformin to DPP4-inhibitors , we will estimate the population-level effect of exposure on the hazards of the outcome during the period from 1 day after exposure start to 0 day after exposure end”.

So, please post your research question on this thread, fill out the mad libs, tell us your T/C/O/time-at-risk/model, and your study idea can be considered by the community under your leadership, and together we can generate the reliable evidence that all stakeholders deserve.

1 Like

To compare the risk of mortality at hospital discharge, 30, 60 and 90 days post discharge between Video Assisted Thoracoscopic Surgery (VATS) and Open resection in a population with known or suspected lung cancer, we will estimate the population-level effect of exposure by cox proportional hazards and simple Kaplan-Meiers survival of short and long-term mortality during the period from hospital discharge to 3 years.

Stephen Deppen

According to the CDC, “From 2000 to 2016, more than 600,000 people died from drug overdoses. On average, 115 Americans die every day from an opioid overdose. 1
We now know that overdoses from prescription opioids are a driving factor in the 16-year increase in opioid overdose deaths. The amount of prescription opioids sold to pharmacies, hospitals, and doctors’ offices nearly quadrupled from 1999 to 2010,3,4 yet there had not been an overall change in the amount of pain that Americans reported.5,6 Deaths from prescription opioids—drugs like oxycodone, hydrocodone, and methadone—have more than quadrupled since 1999.7”

In response to the opioid epidemic (declared a public health emergency) HHS plans to focus on 5 major points. One of them being the collection of “Better Data” described as “Strengthening our understanding of the epidemic through better public health surveillance’
Therefore, I propose this study:
Among post-op patient and users of prescription opioids (hydrocodone, oxycodone, morphine, and codeine) which patients develop opioid addiction in 1 year?
Target cohort: Post-op patients on prescription opioids
Outcome: Opioid addiction development
Time-at-risk: 1 year
Model: Logistic regression (Adjusted for age, race, ethnicity, gender …)

Thanks @deppen, that’s a very well defined comparative cohort analysis
question and seems like something quite useful to study. Mortality is
sometimes a tricky outcome in some administrative claims systems that only
have incomplete linkage to a death register, but I would guess that many
OHDSI partners should have sufficient data to address this question. It’s
also a nice example that reinforces, as @nigam has trumpeted many times,
that OHDSI is much more than ‘drug evaluation’, because medical
interventions can take any form: here, alternative surgical procedures
represent an important comparative effectiveness question to explore, and
could benefit all those patients with lung cancer who may be making the
difficult choice between these alternatives. Thanks for your
contribution! Keep 'em coming!

Thanks @PaulaSaroufim, I definitely agree that the OHDSI community can and
should contribute evidence to the ongoing opioid epidemic, as there appears
to be a lot of information that could be known but isn’t about the disease
natural history and the effects of these products long-term.

The problem you outline sounds more like a patient-level prediction problem
than a population-level effect estimation question. That’s definitely
something that OHDSI can explore, and something that @jennareps and
@Rijnbeek could discuss with you about bringing that to an upcoming
Patient-Level Prediction workgroup call to initiate a OHDSI network study,
assuming there’s shared interest (I’m willing to be a large sum that there
will be :grinning: ).

For purposes of the face-to-face, we are hoping to identify a
population-level effect estimation question that can be addressed using the
comparative cohort design. So, is there a related question in the opioid
space that may make sense to explore? Perhaps something like:

To compare the risk of opioid addition between *short-term opioid
exposure (acute < 10 day ) *and long-term opioid exposure (chronic > 30 d),
we will estimate the population-level effect of exposure on the odds of
the outcome during the period from 1 day after exposure start to 365
days after exposure start

Thanks Patrick for the feedback! I would love to discuss with @jennareps and@Rijnbeek about the potential for a network study that addresses the opioid epidemic !

For the question proposed, I believe the comparative group can be NSAIDS users; the question would be:

To compare the risk of opioid addition between opioid users and NSAIDS users, we will estimate the population-level effect of exposure on the odds of the outcome during the period from 1 day after exposure start to 365 days after exposure start.

We would have to dive in more details to accurately define and characterize the NSAIDS users, opioids users and the model…

Another piece that is of interest to both policy folk and medical educators is the penetrance of minimally invasive surgery over time in various settings. We would be interested in what the rates are in other countries.

I also like this one. A couple of considerations to help us consider how to parameterize it:

  1. Current state of evidence. From a very quick look it looks like there’s at least a couple of meta-analyses, eg.

Notably, the 2013 CHEST guidelines do make a (2C) statement:

For patients with clinical stage I NSCLC, a minimally invasive approach such as video-assisted thoracic surgery (thoracoscopy) is preferred over a thoracotomy for anatomic pulmonary resection and is suggested in experienced centers (Grade 2C).

  1. Given that, can we parameterize equipoise. Those familiar with OHDSI instances could better address this, but I presume we would want to specify a target cohort of early stage lung cancer, and would want TNM / staging information in our definition? More advanced cases are unlikely candidates for VATS, but may get an open debulking?

If it is the case where every VATS case could have been done with an open approach (and it looks like there’s reasonable equipoise for other indications, like mediastinal node staging), would we be confident (esp if we don’t have staging info) that a propensity matched design would identify a reasonable control cohort even if not every open lobectomy is a candidate for VATS .

  1. Outcomes. To Patrick’s mention around death I would add that perioperative death in either of these approaches would generally be rare. Perhaps this argues for a network study if we are confident in the accuracy of death in the databases. It would be great if could could also look at other outcomes - arrhythmia, pneumonia, air leak, LOS, etc.

Out of curiosity, are any OHDSI sites also NSQIP and / or ERAS (enhanced recovery after surgery) study sites, and is that data fed into their ohdsi instance? Would help in our confidence that these outcomes are systematically captured in these sites.

I agree entirely with your refinement Evan. My only worry is that many will not have Cancer registry or IASLC Edition X *where X can be 6-8) staging data. In my work at the VA we have done the basics (the original proposal) without stage, and are just now adding stage to our analysis as we get the data. A truly accurate, clinically representative analysis should just be stage 1 disease and ideally propensity matched by either a combination comorbidity score or pre-op PFT (FEV1).

The open debulking would be very rare prior to chemo -rads treatment. (there is an ASCO trial investigating surgery post systemic treatment after theraputic response). What is more common, is you go in thinking stage 1 and then find N2 disease or worse after complete staging of the thorax. For a mortality outcome pathological stage 1 disease is the best population for analysis.

Hi everyone, another topic of high interest/controversy in the medical community that I have been discussing recently with other clinician-researchers is around canagliflozin or other SGLT2 inhibitors (new diabetes medications) and amputation. The clinical trials showed a surprising association between canagliflozin and amputation, so it now has a black box warning. However, the same association and warning does not exist for the other SGLT2 inhibitors, and clinically, we were surprised that canagliflozin alone has the association (either there might be no real increased risk–many diabetics, especially poorly controlled diabetics, are at risk for amputation; or all SGLT2 inhibitors might have this association)

So the research question would be:
To compare the risk of amputation between patients with diabetes on canagliflozin and patients with diabetes on other diabetes medications (or compare to other SGLT2 inhibitors), we will estimate the population-level effect of exposure on the hazards of the outcome during the period from 1 day after exposure start to 0 days after exposure end (on treatment, but could also do intent to treat or other time windows)

I’ll start throwing a few out there as well. Background to this one (which would need a little more lit search):
Folks are probably accustomed to some amount of controversy in ASA use for primary prevention of MI.
The CHEST 2012 thrombosis guidelines gave a weak reccommendation to do this in folks > 50 ( Grade 2B). But interestingly, reductions in MI are closely balanced with major bleeds. What helped tip the balance was a small signal in lower cancer mortality, as far as I know largely driven by cancers of the proximal colon. These are likely detected earlier with some handy bleeding, but there might also be an anticancer effect to ASA (there are some larger studies done here I would need to flesh out further).

A recent study suggested that a similar detection effect might apply to bladder cancer. We don’t know if there’s an associated mortality benefit. Finding mortality benefits in other cancers might, interestingly affect thrombosis guidelines further. That’s a lot of people.

( as I write this, I realize capturing aspirin might be a problem? Crap.) Anyway:

To compare the risk of 1,5 year mortality between patients on aspirin (for at least 1 year) at the time of bladder cancer diagnosis and patients not on aspirin at the time of bladder cancer diagnosis, we will estimate the population-level effect of exposure on the hazards of the outcome during the period from 1 day after exposure start to 5 years after exposure end.

Could conceivably extend to other cancer types in a bigger study.

Here is my entry into the idea pool:

In patients with heart disease:
time-at-risk=6 years

Per this clincial trial
https://www.ncbi.nlm.nih.gov/pubmed/25602299 , in the first 6 years the HR is 1.1 (this applies to patients with coronary heart disease AND not on statins)
But FDA is now making a more general claim here (removing the not=on-statin condition): https://www.fda.gov/Safety/MedWatch/SafetyInformation/SafetyAlertsforHumanMedicalProducts/ucm597862.htm

I am proposing on behalf of Elad Sharon from the NIH:

To compare the risk of opportunistic infections (will provide list of those) between Patients with ALCL/Hodgkin treated with CD30 inhibitors (brentuximab) and Patients with ALCL/Hodgkin with other drug therapies (any antineoplastics or immunmodulators), we will estimate the population-level effect of exposure on the rate ratio of the outcome during the period from 1 day after exposure start to 90 days after exposure end.

Reason this is interesting: CD-30 inhibitors are fairly new and CD-30 might be involved in fighting opportunistic (as opposed the ordinary day-to-day infections you pick up from your sneazing and coughing neighbor on a long distant flight) infections. This could be a good experiment to support or reject this hypothesis.

See below the proposal from Guy Brussele from the Department of Respiratory Medicine, Ghent University Hospital Belgium and Katia Verhamme from the Department of Medical Informatics of the Erasmus MC:

Does use of inhaled long-acting muscarinic antagonists (LAMA) without concomitant use of inhaled corticosteroids in patients with asthma increase mortality risk?

Asthma is a chronic disease of the airways, associated with chronic airway inflammation, variable airflow limitation and variable symptoms of cough, shortness of breath and chest tightness. Viral respiratory tract infections, and exposure to allergens and pollutants can induce asthma attacks (i.e. asthma exacerbations), which can be life-threatening. In 2015, world-wide 400.000 patients with asthma died due to asthma (Global Burden of Disease Study, Lancet RM 2017). The mainstay of treatment of asthma are inhaled corticosteroids (ICS). In patients with moderate to severe asthma (Global Initiative for Asthma [GINA] steps 3-5), long-acting bronchodilators are added to the maintenance treatment with ICS: either long-acting beta2-agonists (LABA) and/or long-acting muscarinic antagonists (LAMA). In asthma, most clinical evidence of add-on therapy with LAMA to ICS (with or without LABA) is available for tiotropium (Spiriva [Handihaler or Respimat]), but other LAMAs encompass glycopyrronium, umeclidinium and aclidinium.

For LABA, it is well established that LABA monotherapy (without concomitant therapy with ICS) in patients with asthma is associated with an increased risk of asthma attacks and mortality. Therefore, a LABA should always be added to an ICS when treating patients with moderate-to-severe persistent asthma; ideally in one single fixed dose combination inhaler (ICS + LABA). For LAMA, it is not yet known whether LAMA monotherapy (without concomitant therapy with ICS) in patients with asthma is associated with an increased risk of asthma attacks and/or mortality. However, since both LABA and LAMA do not reduce the chronic airway inflammation in asthma, we put forward the hypothesis that in patients with asthma LAMA monotherapy – similar to LABA monotherapy – increases the risk of mortality. Moreover, since there are no fixed dose combination inhalers of LAMA and ICS, the risk of asthma patients using a LAMA without an ICS is substantial.

Clinical research question
Is use of inhaled long-acting muscarinic antagonists (LAMA) without concomitant use of inhaled corticosteroids (ICS) in patients with asthma associated with an increased mortality risk?

In patients with asthma,
A) is use of monotherapy of inhaled LAMA without concomitant use of ICS associated with an increased risk of mortality compared with use of LAMA and ICS?
B) Is use of inhaled LAMA (with or without LABA; with or without Leukotriene Receptor Antagonist [LTRA]) without concomitant use of ICS associated with an increased risk of mortality compared with use of LAMA and ICS (with or without LABA; with or without Leukotriene Receptor Antagonist [LTRA])?

5- core elements of the comparative cohort design:
1) The target exposure cohort:
Patients with asthma (age: 12 – 50 years) exposed to LAMA without ICS;
Either (A) monotherapy with LAMA, or (B) therapy with LAMA and (LABA and/or LTRA).
2) The comparator cohort:
Patients with asthma (age: 12 – 50 years) exposed to LAMA and ICS;
Either (A) therapy with LAMA and ICS only, or (B) therapy with LAMA and ICS and (LABA and/or LTRA).
3) The outcome (cohort):
Mortality (all-cause mortality);
eventually in a second stage: respiratory mortality (asthma-related mortality).
4) A time-at-risk period:
From the first day of treatment with LAMA or LAMA and ICS till end of cohort study, end of follow-up or death.
5) A model specification:
Matching of subjects of the comparator cohort with subjects in the target exposure cohort on age, gender and propensity score (e.g. smoking, asthma severity, asthma control, previous asthma exacerbations, previous asthma hospitalizations …).

The choice of the age range (12 - 50 years) is based upon the following rationale:

  1. The LAMA tiotropium is indicated for the (add-on) treatment of
    asthma from the age of 12 years onwards;
  2. LAMA monotherapy
    (without ICS) and LAMA + LABA combination therapy (loose or fixed
    combination inhalers; with or without ICS) are indicated for the
    maintenance treatment of COPD, which mainly affects subjects older
    than 50 years.
1 Like

Looove this topic! Indeed, GINA protocol says that LAMA shouldn’t be used otherwise than as add-on therapy, which is reasonable from the medical perspective. I haven’t found any reliable research on LAMA monotherapy, which I believe is due to an inappropriate risk for patients who are prescribed LAMA only. So, eventually, I’m quite excited to investigate it!
Something that I want to mention: why don’t we add asthma exacerbations as another criterion?And why can’t we exclude patients with COPD to extend the upper age limit?

Getting back on track: I want to propose a relative research question:

To compare the risk of asthma exacerbations (and now I believe I should add mortality here) between patients with Chronic Obstructive Pulmonary Disease (COPD) C/D and inhaled corticosteroids withdrawal and patients with COPD C/D who have never used ICS, we will estimate the population-level effect of exposure on the rate ratio of the outcome during the period from 1 day after last exposure start to 180 days after exposure end.

There have been several trials focused on ICS withdrawal although they have discrepancies in the definition of outcome, treatment used, eligibility criteria etc. Additionally, patients with no use of ICS weren’t included in any of these studies.

Adding a reference:


Sudden cardiac death, acute myocardial infarction, and stroke are rare but serious cardiovascular events among users of serotonin 1b/1d agonists (triptans). Since triptans are known to have such adverse effects, physicians evaluate patients for cardiovascular risk factors prior prescribing these medications. As the result, lower reports of cardiovascular conditions are being reported among users of triptans, but it is not known how much triptans may contribute to serious cardiovascular events.


We intend to evaluate both short-term and long-term effect of triptans to cause serious cardiovascular events compared with new users.

Research question:

To compare the risk of serious cardiovascular events (sudden cardiac death, acute myocardial infarction, and stroke) between new users of triptans and patients who continuously used triptans for 1 year (short-term) and more than 1 year (long-term), we will estimate the population level effect of exposure on the hazards of the outcome during the period from 1 day after exposure start to 0 days after exposure end.

Research Question:
Is the risk of osteoporosis or fracture increased in postmenopausal hypothyroid patients using high strength levothyroxine compared to patients using low strength levothyroxine?
Long-term use of levothyroxine has been associated with decreased bone mineral density, particularly in postmenopausal females on greater than replacement doses or in women receiving suppressive doses. Levothyroxine is used in the treatment of hypothyroidism and patients should be given the minimum dose necessary for desired clinical and biochemical response to limit risks for osteoporosis.

From: BMJ 2011; 342 doi: https://doi.org/10.1136/bmj.d2238

“Hypothyroidism is common in older people, particularly women,1 and over 20% of older people receive levothyroxine replacement long term.2 With normal ageing, thyroid hormone production, secretion, and degradation decreases,3 4 5 and therefore older people with hypothyroidism have lower requirements for levothyroxine replacement than younger people.3 5 Most people with hypothyroidism are diagnosed in early or middle adulthood,6 thus most will have been treated for many years by the time they reach older age. Although regular monitoring of levothyroxine doses is indicated,7 8 evidence suggests that the dose often remains unchanged as people age,9 10 and over 20% of older adults are overtreated,11 12 13 14 leading to iatrogenic hyperthyroidism.
Chronic hyperthyroidism may increase the risk of fractures, particularly in older people and postmenopausal women who already have a higher risk of osteoporosis and fractures.13 15 16 17 Studies have found that higher compared with lower doses of levothyroxine replacement18 19 20 and subclinical hyperthyroidism21 are associated with a lower bone density and bone quality, as measured by ultrasonography.22 An excess of thyroid hormone can also affect neuromuscular function and muscle strength23 and increase the risk of arrhythmias24 25 and falls,15which can raise the risk of fractures independent of bone density. Previous studies of the association between levothyroxine and fractures have had mixed results,15 26 27 28 29 30 largely because of small sample sizes and the inclusion of younger, lower risk populations. This problem has not been dealt with adequately in older women, and older people in general, who are at higher risk of fractures,15 26 31 more likely to be treated with levothyroxine,11.”

1) Target Cohort (T) : On treatment post menopausal (age >=50 years) patients with hypothyroidism exposed to high strength (From above publication defined as: >=0.044 mg/day) levothyroxine.
2) Comparator Cohort ©: On treatment post menopausal (age >=50 years) patients with hypothyroidism exposed to low strength (From above publication defined as: <0.044 mg/day) levothyroxine.
3) Outcome Cohort (O): Osteoporosis or Fracture
4) Model Type: Cox
5) Time at Risk Start and End: From the first day of treatment with high or low strength levothyroxine until the end of cohort study, end of follow-up or death.
6) Methods to adjust for bias: Use propensity score matching or stratification with or without trimming, use negative controls
tagging @Frank as he may share an interest in this topic

1 Like

There is a reported association of developing CKD with exposure to proton pump inhibitors in several small cohorts in the literature. This is particularly concerning given the widespread use of these drugs and the rising prevalence of CKD (defined as an eGFR < 60 mL/min) that currently affects 1 in 7 adults in the United States - the reasons for which are not clear at present. Proton pump inhibitors are commonly prescribed drugs but are frequently given to individuals who also have significant comorbidities such as diabetes and obesity (also known risk factors for the development of CKD). A large observational cohort study that is either able to detect a change in the GFR over time (calculated from serum creatinine, age, gender and race) or the development of new onset CKD, defined as an estimated GFR < 60mL/min, following the initiation of PPI use (after adjustment for major comorbidities) would result in significant change in clinical practice worldwide.
Currently small cohort studies have been plagued with inadequate adjustment, inability to create propensity matched control cohorts. As a result, there has yet to be a significant change in clinical practice. The results of a large adequately powered, propensity score matched study or a study that excludes individuals with comorbidities commonly associated with CKD would have an immediate impact on patient care.
Target Cohort – patients aged age 18 years and older with chronic exposure to proton pump inhibitors who do not have evidence of chronic kidney disease (defined as an estimated GFR of < 60mL/min) at the time of initiation of therapy
Comparator group: Patients aged 18 years and older with no exposure to proton pump inhibitors
Outcome: Incidence of chronic kidney disease
Model: Cox model for time to event analysis (development of CKD) with stratification by drug exposure duration. Regression for assessing rate of decline in GFR associated with exposure to PPI
Adjustment: propensity score using age, diabetes, hypertension, African American race and eGFR at the time of entry into the study.
Potential exclusions: patients exposed to known nephrotoxic agents - specifically chronic NSAID use.