OHDSI Home | Forums | Wiki | Github

HowOften: Community contributions wanted


As we discussed on the 20June2023 and 15August2023 community calls, @hripcsa and I would like to encourage our community to think big and collaborate together in a effort toward large-scale incidence characterization. HowOften is be a community-wide study to define a broad set of target cohorts T that’ll serve as denominators, and another broad set of outcome cohorts O that’ll serve as numerators. And for a defined list of time-at-risk windows (e.g. 30d, 1yr, all-time), stratified by age/sex/index year, we will compute the incidence of O in T for all T-O combinations within each database in our participating network, and then meta-analyze the results to produce composite summaries.

As with all OHDSI network studies, we will use GitHub to share study materials, including protocol and source code, which should be based where possible off of existing HADES packages. And we intend to make the full resultset publicly available through an interactive website, likely initially taking advantage of the RShiny modules built by the HADES team as part of the Strategus workflow. As we’ve seen with prior OHDSI work, background incidence rates can be used for a wide range of clinical applications, including providing disease natural history, providing context for pharmacovigilance by quantifying the magnitude of risk for known effects, and reporting digital quality measures (see @bnhamlin 's talk here).

At the OHDSI2023 Global Symposium, we are planning a community-wide workshop over 2 days to focus on HowOften, whereby we will review the final specification, help data partners with configuring and executing the study package, deploy the full resultset, and then collaboratively review the results from both methodological and clinical perspective and collaboratively develop interactive visualizations to further explore the results.

So what do we need from you right now? Please contribute phenotypes that we can use as target or outcome cohorts in the analysis. HowOften will be based on phenotypes that have been developed and evaluated by our community and submitted to the OHDSI Phenotype Library. The Library currently had a good number of phenotypes developed by the community so far (review here), but most phenotypes have still not undergone the peer review process and many other phenotypes that we’d probably be interested in haven’t been started. @Gowtham_Rao went through the OHDSI Phenotype Library submission process on last week’s community call, and the Phenotype WG is a good place to go to get community support. We’ve set a hard deadline of 15Sept2023 to get your phenotypes into the Library so that it can be considered for inclusion in HowOften.

I am really hoping that our clinical workgroups can contribute (Oncology - @agolozar , Psychiatry - @Andrew , Vision Care, Pregnancy/Maternal health - @acallahan ) but also think this a good opportunity for all workgroups to consider what topics they’d like to see covered. On last week’s call, we brainstormed on various cohort opportunities to consider, and several valuable phenotypes were identified on polleverywhere, including Wet AMD, autosomal dominant polycystic kidney disease, non-active uveitis, antibiotic-resistant infections, diabetic ketoacidosis, PASC. I hope others will follow @Jill_Hardin 's lead in stimulating discussions on the forums about further phenotyping opportunities for specific clinical problems of interest.

Thanks in advance for all your community contributions. We look forward to collaborating on HowOften with all of you!

Thank you, @Patrick_Ryan. I look forward to running the contributed phenotypes and testing out this additional dimension of large scale, namely asking the community members in parallel to contribute what is important to measure.

Everyone, I wanted to thank @tonysun for making an invaluable contribution to the OHDSI Phenotype Library. @tonysun’s cohort is now available as part of version 3.17.0 of the OHDSI Phenotype Library with the ID 722.

We are eagerly looking forward to more contributions from others in the community!

A draft of the HowOften protocol has been uploaded to the HowOften github Documents directory (https://github.com/ohdsi-studies/HowOften/tree/master/Documents). The draft is open for feedback, and we’ll revise with an extra document containing Ts and Os once they come in from the September 15 deadline. Note that the protocol includes both sides of the study: the Ts and Os that we carry out as part of the OHDSI Symposium, and the refreshment of the original all-by-all drug-effect incidence rate study. Thanks all.

Proposed studies for HowOften:
1) What is the rate of subjects with inflammatory bowel disease (IBD), plaque psoriasis, or psoriatic arthritis receiving concomitant biological treatment?
T = Subjects with IBD, plaque psoriasis, or psoriatic arthritis (3 cohorts)
O = exposure to biologic agents from 2 classes of indicated treatment simultaneously, using >= 30 overlap as criteria for defining simultaneous (as opposed to switch-over). Classes used will be TNF -alpha inhibitors, IL12/23 inhibitors, and !L 12 inhibitors.
Importance: the rate of dual use of biologics has not been estimated previously and dual use may prove to be more effective than single use.

2) What is the rate of exposure to treatments specific for pulmonary arterial hypertension (PAH) in subjects with a diagnosis of PAH and prior left heart disease (LHD) or a diagnosis of LHD and prior PAH?
T = Subjects with diagnoses for both PAH and LHD with the index date being the occurrence of the latter of the 2 conditions.
O = exposure to any one of of the classes of treatment for PAH: Endothelin receptor antagonists, Phosphodiesterase 5 inhibitors and guanylate cyclase stimulators, or Prostacyclin analogues and prostacyclin receptor agonists.
Importance: the use of PAH treatment in subjects with LHD may cause significant adverse reactions depending on the severity of the LHD.

3) What is the rate of exposure to pulmonary endarterectomy (PEA) or Balloon Pulmonary Angioplasty (BPA) in subjects diagnosed with chronic thromboembolic pulmonary hypertension (CTEPH) - WHO pulmonary hypertension group 4?
T = Subjects with a diagnosis of CTEPH.
O = exposure to either PEA or BPA
Importance: PEA or BPA, definitive treatments for CTEPH, appear to be underutilized, especially in the US.

4) What is the rate of subjects diagnosed with one WHO pulmonary hypertension (PH) group later diagnosed with a second type of PH?
T = Subjects with a first diagnosis of one type of PH, WHO groups 1-4.
O = diagnosis of a second form of PH
Importance: Pulmonary hypertension has several different etiologies. Treatments for the different forms of PH differ considerably and treatment for one group may produce severe negative outcomes if used in another group. Subjects diagnosed with one form of PH while actually having another form may not be treated correctly. There is currently much debate over whether subjects may have multiple forms of the condition. This analysis will help to determine the prevalence of either misdiagnosis or miscoding of the condition.

1 Like

Proposed studies for HowOften:

1) What is the incidence of patients diagnosed with a primary adenocarcinoma of the colon or rectum (colorectal cancer) to recive potential curative surgery or oncological therapy?
T: Subjects diagnosed with colorectal cancer (Using cohorts specifying mismatch repair status / microsatelite instability and anatomical location in the colon / rectum / either (6 cohorts in total))
O: Patients reciving a antineoplastic drug specific for colorectal cancer, radiotherapy or a segemental or local resection of the large intestine.
Importance: Creating a real world investigation for how patients diagnosed with colroectal cancer are treated.

2) For patients with colorectal cancer who are neither operated curatively or reciving oncological treatment what is the mortality rate?
T: Subjects Subjects diagnosed with colorectal cancer not reciving antineoplastic therapy or curative intended surgery (Using cohorts specifying mismatch repair status / microsatelite instability and anatomical location in the colon / rectum / either (6 cohorts in total))
O: Death
Importance: Creating real world evidence of the prognosis of patients who have no curative treatment option, but can recive oncological treatment. Investigating if this differ by molecular subtype (MMR) or anatomocial location.

3) Among patients diagnosed with colorectal cancer who recive oncological treatment, but no curateive intended surgery, what is the mortality rate?
T: Subjects diagnosed with colorectal cancer, reciving antineoplastic therapy but no curative intended surgery (Using cohorts specifying mismatch repair status / microsatelite instability and anatomical location in the colon / rectum / either (6 cohorts in total))
O: Death
Importance: Creating real world evidence of the prognosis of patients who do not recive curative surgery, but get oncological treatment and investigating of this differ by molecular subtype or anatomical location.

4) Among patients diagnosed with colorectal cancer who recive a potential curative intended surgery, what is the incidence rate of complications, recurrence of disease, oncological therapy or death?
T: Subjects diagnosed with colorectal cancer, reciving a potential curative surgery (Using cohorts specifying mismatch repair status / microsatelite instability and anatomical location in the colon / rectum / either (6 cohorts in total))
Local recurrence,
Distant metastasis,
Exposure to antineoplastic drugs
Complications(Anastomotic leak, wound dehenicence, bowel obstruction, wound infection, intraabdominal abscesces, sepsis, UVI, pneumonia, heat failure, artrial fibilation, myocardial infarction, kidney failure, DVT, pulmonary embolism, postoperative hemorrhage, aspiration and respiratory insufficience)

Importance: Creating real world evidence of the prognosis of patients who is undergoing potential curative surgery for colorectal cancer and investigating of this differ by molecular subtype or anatomical location.

The combination of the questions could create a strong pool of references for clinical trajectories for patients diagnosed with colorectal cancer. The use of both colon and rectal cancer as two different cohorts might highligt some differences in the overall patient population. The inclusion of the molecular subtypes by MMR status might create crediable evidence about the perioperative risk of these patients and the magnitude of patients who cannot recive curative surgery, but who might be treated with drugs such as immune checkpoint inhibitors.

(Unfortunatley we haven’t been able to upload all the phenotypes due to a cap on forum topics per day)

1 Like

This makes me very happy. @awrosen , you were so productive today that the Forums told you to slow down :slight_smile: A nice mark of productivity :slight_smile:

Thank you for your thoughtful engagement in HowOften, this is really great!

One of the original motivating use cases for the HowOften large-scale incidence characterization effort was to summarize the frequency of adverse events amongst patients exposed to drugs. Drug product labels list a range of potential adverse reactions that were observed in clinical trials or in post-marketing experience, but rarely provide a population incidence for how often we would expect to observe the ADR (let alone an estimate of the strength of association). As @hripcsa presented at prior OHDSI events, just knowing the magnitude of occurrence can help support clinical decision-making: if an event is extremely rare (even if potentially serious), then it may weigh less into the benefit-risk calculus than other more prevalently occurring events because the expected risk is so low. In OHDSI’s subsequent work characterizing background incidence rates for the COVID adverse events of special interest, we learned that incidence estimates are quite sensitive to a range of factors, including age, sex, calendar time, indexing event, and of course, database.

One aspect we haven’t yet explored is the impact of the drug indication on the incidence of adverse events. Given that some drugs can be indicated for multiple different target diseases, it seems plausible that the incidence of events could be impacted by the indication, but its unclear the magnitude of that impact or whether it falls within the general uncertainty already observed after we stratify by age and sex. As we look to expand HowOften to explore all exposures and all outcomes of interest, it is important to determine whether we also need to nest each exposure within their indication (which requires custom phenotyping for each exposure), or whether the exposure-outcome incidence will be a sufficiently reliable piece of evidence that we can generate at scale.

To gain greater insight into this question, we propose to use the current HowOften community effort to evaluate the impact of indication nesting.

We propose to use a sample of drug classes that have been used in prior OHDSI studies and estimate the incidence of a range of outcomes (from the OHDSI phenotype library, we can use the COVID AESI, Designated Medical Events, and other LEGEND outcomes already submitted). For each drug class, we will introduce a target phenotype for new users of the class (e.g. first occurrence of drug exposure with at least 365d prior observation), and we will also create a target phenotype for each drug class nested within its indications (e.g. add an inclusion criteria that requires observing the indicated disease in the 365d prior to or on exposure start). These cohorts will then allow for a head-to-head comparison between the overall and nested populations, so that we can estimate incidence sensitivity.

The drug classes we propose to include for this evaluation are:

Thiazide diuretics (ingredients: hydrochlorothiazide, chlorthalidone, metolazone, indapamide): 1) hypertension

dihydropyridine calcium channel blockers (dCCB) (ingredients: amlodipine, nifedipine, felodipine, nicardipine, nisoldipine, isradipine): selected indications: 1) hypertension (note also used for CAD)

Beta blockers (ingredients: metoprolol, atenolol, carvedilol, propranolol, labetalol, bisoprolol, nebivolol, nadalol, betaxolol, acebutolol, pindolol, penbutolol): selected indications: 1) hypertension, 2) heart failure, 3) acute myocardial infarction

SGLT2 inhibitor (ingredients: empagliflozin, dapagliflozin, canagliflozin, ertugliflozin) - selected indications: 1) Type 2 diabetes mellitus, 2) heart failure

GLP-1 receptor antagonists (GLP1RA) (ingredients: dulaglutide, liraglutide, semaglutide, exenatide, lixisenatide, albiglutide) : 1) Type 2 diabetes mellitus, 2) obesity

DPP-4 inhibitors (ingredients: sitagliptin, linagliptin, saxagliptin, alogliptin, vildagliptin): 1) Type 2 diabetes mellitus

Tumor Necrosis Factor alpha (TNFa) inhibitors (ingredients: adalimumab, infliximab, etanercept, certolizumab pegol, golimumab)- selected indications: 1) Rheumatoid arthritis, 2) Psoriatic Arthritis, 3) Crohns disease, 4) Ulcerative colitis, 5) Psoriasis

JAK inhibitors (ingredients: tofacitinib, ruxolitinib, baricitinib): selected indications: 1) Rheumatoid arthritis, 2) Ulcerative colitis

IL-23 inhibitors (ingredients: guselkumab, risankizumab, tildrazumab): selected indications: 1) Psoriasis

Fluoroquinolone systemic (ingredients: ciprofloxacin, levofloxacin, ofloxacin, moxifloxacin, gatifloxacin, nadifloxacin, besifloxacin, norfloxacin, lomefloxacin, gemifloxacin): selected indications: 1) Urinary tract infection, 2) pneumonia

Cephalosporin systemetic (ingredients: cephalexin, ceftriaxone, cefazolin, cefdinir, cefuroxime, cefprozil, cefadroxil, cefoxitin, cefditoren, xefpodoxime, cefaclor, ceftazidime, cefizime, cefotetan, cephradine): selected indications: 1) Urinary tract infection, 2) pneumonia

Trimethoprim systemetic (ingredients: trimethoprim): selected indications: 1) Urinary tract infection, 2) pneumonia

Proposed studies for HowOften (sorry for being 48 hours past the deadline):
**1) What is the incidence rate of Alzheimer by age, gender and calendar time ?
T = the standardized data base background population described in the HowOften protocol
O = Incidence occurrence of Alzheimer- .

**2) What is the incidence rate of Alzheimer by gender and calendar time among patients with mild cognitive impairment or dementia
T = patients with mild cognitive impairment or dementia
O = Incidence occurrence of Alzheimer-

In as much as adverse events are of interest in the pharmacology space, it would be of great interest to look at adverse events after surgical procedures across our network of databases. The Surgery and Periop WG met with regards to how often and have put some target surgery cohorts together (based on previous work including the PROTEUS study thon & subsequent iteration, and prior work completed by WG member Brian Bucher). I have been a rate limiting factor with some long clinical weeks as of late.

Q: What are the incidence rates of post operative adverse events across major non cardiac surgery as a whole, and groups of representative surgeries?

T’s: Non-Emergent Major Non-Cardiac Surgery; AAA repair, Lower Extremity Bypass, Carotid Endarterectomy, Lung Resection, Esophagectomy, Pancreatectomy, Colectomy, Cystectomy, Nephrectomy, Coronary Artery Arter Bypass Graft Surgery, Aortic or Mitral Valve Repair or Replacement.

Outcomes were previously specified in Covid19 ATLAS – did a lot of prepatory work there for a characterization study. An appeal to anyone in the community who might be in the know: is there a viable path to get that JSON back? I have a list of cohort IDs.

Otherwise on review, I see many O’s of interest are otherwise already represented in the phenotype library.

I do think it’s of interest to have extended clean windows (or first time in history criteria) to help ensure we aren’t counting recapitulated historical problems. I suspect this is true for adverse events in the drug domain as well. Can I presume there’s a ‘study wide’ approach being taken there? Or would we want to configure those ‘clean windows’ at the cohort level?

O’s: MACE, CHF (any), CHF(first in history), post op Afib (first in history), Myocardial Infarction (first in history), Deep vein thrombosis (First in history), Pulmonary Embolism (first in history), Pneumonia (30d clean window), UTI (30d clean window), Sepsis (30d clean window), c difficile infection (first in history), blood transfusion (1 yr clean window), AKI (30d clean window), hemodialysis (first in history). The prolific @awrosen has also specified many outcomes of shared interest.
TAR: within 2 weeks, 30d, 60d, 1 yr post-surgery.

‘Nested indication is a great idea, and something I’d be interested in exploring further in better defining surgical cohorts in the future (e.g. colectomy for Cancer, as opposed to colectomy for perforated diverticulitis).

Could a corollary to that idea be a ‘nested comorbidity’? Eg. Diabetes and MNCS, CHF and MNCS, etc. If there is capacity to explore this (or, to take it further with ‘nested CHADS2 score’), it would form a great adjunct, and I’d invest in specifying it further.

For a second study, I’d propose we look closer into post operative atrial fibrillation and risk of stroke. This has long been an area of clinical uncertainty (with respect to recommendations around anticoagulation). Stroke risk is elevated in these patients, but may not be elevated to the same degree across different surgical contexts. There is a planned RCT in this area, and a large-scale study of incidence rates would help set the stage for that.

Q: In patients who undergo [Surgery type] and experience [any/broad ; paroxysmal/narrow] Atrial fibrillation (first episode) within 2 weeks of surgery, what is the (30d, 1yr, 2yr) risk of ischemic stroke?

T’s: (Non-Emergent Major Non-Cardiac Surgery; AAA repair, Lower Extremity Bypass, Carotid Endarterectomy, Lung Resection, Esophagectomy, Pancreatectomy, Colectomy, Cystectomy, Nephrectomy, Coronary Artery Arter Bypass Graft Surgery, Aortic or Mitral Valve Repair or Replacement) & (any, or paroxysmal) Atrial Fibrillation (first episode) within 2 weeks of surgery

O’s: Ischemic stroke

TAR: 30d after surgery, between 30d and 1yr after surgery, Ischemic stroke between 1yr and 2yr.

Teaser: Nested CHADS2 = 0, nested CHADS2 = 1, … , nested CHADS2 =6.

1 Like

Thank you everyone for your many phenotype and incidence task contributions. We look forward to carrying them out during the symposium HowOften workshop. We are in the midst of processing them and have come up with a plan for the workshop. The tasks will be lumped into a couple of segments as follows.

  1. All submitted phenotypes that were received over the last month will be used as ‘outcome’ cohorts, we will estimate the incidence rate of each in the general population.

  2. For the submissions with explicit incidence tasks including target-outcome phenotype pairs, we will run all those pairs. Thank you Andreas, Azza, and Joel for these contributions.

  3. As posted earlier in this thread by Patrick, we will run an abbreviated version of the original HowOften study, looking at selected drug classes as targets (with and without nested indications) and a selected set of outcomes.

As the protocol describes, we will stratify the results by age, sex, and calendar year, and use 4 time-at-risk windows. We are eager to get results from as many data partners in our community would can contribute.

We look forward to seeing you at the symposium.

One more to add:
5) What are the rates of MACE and Serious/Opportunistic infections in those diagnosed with auto-immune disease?
T = Subjects with a first diagnosis of Crohns disease, ulcerative colitis, ankylosing spondylitis, rheumatoid arthritis, psoriasis, or psoriatic arthritis (6 cohorts)
O = diagnosis of MACE, serious infections, or serious/opportunistic infections.(3 cohorts)

Importance: Treatment of auto-immune disease often leave the patient open to adverse outcomes including MACE and infections. Health authorities are often looking for background rates of these in subjects with auto-immune disease.

Id like to participate as a data partner on this. However, I’m going for paternity leave for the next 3 months.
Is there information about the deadlines for participating.

  • Will it be a test study package to run before the symposium, so that we know that the codes work ?
  • Will be the final study package run and results collected during the symposium ? Or is there a possibility to run the study package and add the data in the future ?
  • What is the best channel to follow the updates on this questions ?


@awrosen did you submit a cohort for wound dehenicence? please point me to the submission of that cohort? thank you

hi Evan, I specified your quetions but was not able to find phenotypes in the library for : c difficile infection (first in history), blood transfusion (1 yr clean window)

Hi @Evan_Minty for your second question we will need those 24 cohorts that make the T’s as phenotype entries in the library. We will not be doing automated " overlap" logic as part of this study, so we need the T cohorts fully specified as inputs, any chance you can configure the “any, or paroxysmal) Atrial Fibrillation (first episode) within 2 weeks of surgery” as a nesting/inclusion criteria in your surgery cohorts and submit them as phenotypes ?

Thanks, Azza. Posted to this thread, and will send a table via email.

In reviewing the counts, I think paroxysmal atrial fibrillation is too narrow when applied to the individual surgical groups, so will consider any atrial fib.

Thanks very much for this Azza. Have posted them here:

Blood Transfusion
c difficile

Hi @Azza_Shoaibi, I apologize for me not consistently using the right phrasing, I meant Fascial dehiscence or evisceration (Phenotype submission - Fascial dehiscence and evisceration), which is a more severe complication. Again sorry for the confusion.

  • Updated OHDSI PhenotypeLibrary to version 3.24.0. Unless there is a good reason, I would like to hold the OHDSI PhenotypeLibrary at this release version till after the OHDSI 2023 symposium. This version includes all the community submissions and cohort definitions from selected OHDSI Studies (e.g. LEGEND)
  • Questions from @awrosen @Evan_Minty @jswerdel have been incorporated. @Jill_Hardin request in another has also been included (part of Analysis 1). I have my list of T & O that i have added for considerations.
  • Questions were categorized into three groups as shown here https://github.com/ohdsi-studies/HowOften/pull/1 .
  • Analysis 1: 1 base cohort and multiple outcomes
  • Analysis 2: combination of T and O and specified by researchers in community
  • Analysis 3: research question posed by George and Patrick
1 Like