Wow, what a difference a week makes! Last week, we announced we’d hold a OHDSI virtual study-a-thon on COVID-19 in lieu of the OHDSI EU Symposium, not knowing exactly how we’d do it or what we’d do and mildly anxious that we only had three weeks to prepare for the event. Ah, the good old days, when the biggest anxiety in life was meeting logistics…
In the last 7 days, I have been tremendously impressed by how the OHDSI community has rallied together to figure how how we can collaborate to collectively generate evidence to promote better health decisions and better care. We are still very much looking forward to the OHDSI virtual study-a-thon on Mar26-29, but it very clear that the work needed to support public health can’t and isn’t waiting to start for two weeks, and whatever we do during the study-a-thon isn’t going to be the end of our efforts to inform the COVID-19 pandemic response. Unfortunately, it looks likely that we are going to be on this journey together for quite awhile.
The good news is that it’s also become abundantly clear that we can (and will!) make meaningful contributions to the current crisis through responsible analysis across our international data network. There are a lot of open questions that real-world evidence is best equipped to answer, either because observational data is the best or only current source of information, or because retrospective analyses of these data is the most efficient way to generate insights to address the immediate urgency until prospective data collection and research can be completed.
I do want to reinforce the differentiated value that I see we can make as a community. Lots of groups around the world are working hard to do COVID-19 case detection, and there is quite a bit the world will learn from those cases. But much of what we will learn can’t be determined today, rather it will unfold over the next several months as the disease progresses and prospective data collection of the longitudinal experience of these patients is captured. A key strength of our OHDSI community is our ability to design and execute retrospective analysis of existing healthcare data across our international data network. Together, we have access to the historical experience from hundreds of millions of patients, and it is our responsibility to learn whatever we can from those past experiences to inform the current situation…now not later. So, for those of you on the analytics side of the house, I encourage all of you to not idly speculate: ‘what would I want to do if I eventually get access to data with COVID-19 cases?’ but rather take action on the question: ‘how can I learn from the data I currently have access to?’.
And for those of you with a data focus, the efforts you are putting in today to standardize your data, be it conversion of legacy warehouses or real-time data streams, will pay dividends. I am certain that it’ll remain the case that no one data source will have sufficient information to meet our needs, so the only viable path forward is to work together as a network and learn from our collective resources. The prospective data collection underway today will become the retrospective analysis opportunity tomorrow, and we need to be prepared on both the data and analytics side to realize the potential of what we can deliver for the public health good.
A quick update on community activities that I am aware of:
@SCYou is doing heroic work, leading efforts in Korea to use recent national claims data from HIRA, standardizing the HIRA to OMOP CDM and preparing to apply OHDSI analysis packages. Chan, you are a true inspiration to all of us. Whatever we can do to support you, you’ve got it.
@Daniel_Prieto has been in close contact with UK NHS to determine how OHDSI can support their needs. We are hopeful that we will be able to use more recent CPRD and HES data to inform our current activities.
@mvanzandt is leading efforts to get data refreshed across various Iqvia datasets so that the most recent data can be made available for distributed analysis. I have heard that others in the community are also hard at work trying to get whatever is the most recent data possible accessible in their OMOP CDM instance. Thank you for these efforts, they will make a difference.
@Christian_Reich, @Alexdavv , @Dymshyts and entire vocabulary team are hard at work critically reviewing the source codes and standard concepts, with particular focus on diagnosis of viral disease, symptoms and complications, respiratory procedures, and associated measurements. In addition to adding new standard concepts for COVID-19 diagnosis and tests, they will be revising concept_relationships to enable more accurate analysis. Christian will announce to community when a new vocab is released, which we will encourage everyone in the OHDSI community to download and refresh your ETL so that we’re all on the same page.
@jennareps, @Rijnbeek and I have begun designing patient-level prediction studies. Across the world, towns, states, and countries are all aiming to ‘flatten the curve’, by employing a series of public health measures aimed at delaying the communal transmission of COVID-19 such that the number of infected patients are sufficiently spread out over time that the capacity of our healthcare system can accommodate the needs of the ill. Certainly all people should be heeding the advice of their government and public health officials, useful information from US CDC is here: https://www.cdc.gov/coronavirus/2019-ncov/index.html. In situations where demand for health services exceed supply, prioritization tools based on disease severity become valuable and education for the public about why they shouldn’t seek unnecessary care can make a difference. We are proposing 3 patient-level prediction studies that can inform this discussion: 1) Amongst T: persons with an outpatient visit (GP, urgent care, ER) who have flu or flu-like symptoms, who are O: persons who are admitted to hospital with flu or pneumonia, in TAR: 0d-30d from outpatient visit? The goal of this prediction is to identify, based on the medical history prior to the first encounter, which patients are likely going to need hospitalization. 2) Amongst T: persons with an outpatient visit (GP or urgent care) who have flu or flu-like symptoms but do not have pneumonia and who are not admitted to hospital on the same day or next day, who are O: persons who are admitted to hospital with pneumonia, in TAR: 2d-30d from outpatient visit? The goal with this analysis is to help with assurance of the public that if they are sent home, the likelihood of something bad happening is low so they should follow doctor’s advice and CDC recommendations for self-care. 3) Amongst T: persons with inpatient admission with pneumonia due to viral origin, who are O1: persons requiring intensive care services or O2: persons who die, TAR: during the hospital stay? The goal here is to use medical history information prior to admission to help with triaging those who arrive at hospital to determine which cases are likely to be more severe.
@schuemie and @Daniel_Prieto have begun designing population-level estimation studies. Based on feedback we’ve gathered from clinicians across the world, it seems several areas are attempting use of various agents as treatment for COVID-19-positive patients, such as hydroxychloroquine, antivirals like protease inhibitors and remdesivir, and immunosuppressants like tocilizumab, despite little immediate COVID-19 efficacy evidence and uncertainty about the long-term safety of these medicines in these off-target populations. Prophylactic use of these medicines for asymptomatic patients who were potentially exposed to COVID-19 is being discussed. These developments further increase the need to more comprehensively understand the real-world effects of these products overall, within subpopulations with existing viral disease, and among subgroups at higher risk for COVID-19 complications. Comparative cohort and self-controlled analyses will be developed to examine causal effects of exposure on incidence of viral disease and associated complications, as well as safety outcomes.
We recognize the need to design characterization studies. FDA has encouraged us to prioritize work to delineate the natural history of the disease where possible, but also believe that developing and evaluating phenotypes for the constellation of diagnostic patterns observed during prior viral outbreaks, including the past several flu seasons, would be useful as foundational research which can be utilized not only for the current outbreak purposes but also for other similar viral diseases research and surveillance in general. @Christian_Reich and I are coordinating the development and evaluation of phenotype algorithms to identify persons with viral diseases and associated symptoms and complications, as well as anticipated treatments, which can be used as the cohort inputs into our characterization, estimation, and prediction studies. The quality of our analyses are largely depending on the quality of our phenotypes, so it is important that we use clinical expertise, data domain knowledge, and empirical tools to produce definitions that we can consistently apply across the community, and not end up with an array of different half-baked definitions that will add noise to our evidence generating process. We will be using the new CohortDiagnostics package to support our iterative development, and PheValuator to estimate measurement error where appropriate. We will use atlas.ohdsi.org as a landing place for all finalized conceptsets and cohort definitions (thanks @lee_evans for your help in administrating that!).
How can you help?
If you are interested in participating in the study-a-thon, but haven’t yet signed up, please do so here.
If you are one of the 103 people who have already signed up for the study-a-thon, thank you! If you marked that your expertise is in literature review and evidence synthesis, then you can expect to receive a private email from me enlisting your support on some preparatory research. So the whole community knows what I’ll be asking, there are already multiple prediction models that have created in and around the space of community-acquired pneumonia, but they vary substantially by their target population, outcome definition, decision context, what information is needed as predictors, and the extent to which they were developed or evaluated on observational data like we have within our community. We need to summarize this prior knowledge to inform our future prediction designs. Also, past work has been published in defining and validating phenotypes for viral disease, symptoms, and complications, but it has not been compiled together to examine the heterogeneity of approaches and determine a best practice moving forward. We need to summarize what is known about the safety and effectiveness of treatments under consideration for COVID-19 to find the specific gaps that we can fill. If we can do an OHDSI-community-crowd-sourced literature review on these topics, it will lay a solid foundation for our future research together. We need leaders to coordinate the community effort, so that’ll be my ask.
If you have access to a patient-level data in OMOP CDM and haven’t yet done so, please run the OHDSI concept prevalence study, led by @aostropolets . We are using this aggregate summary information to directly inform the phenotype development process to ensure that as we define conceptsets, we are representing all concepts that are actually captured across the community as completely as possible. It has already proven invaluable to figure out the different ways that flu-like symptoms are showing up around the world, and will only increase in value as more and more of you participate.
If you have access to a patient-level data formatted in the OMOP CDM and can execute analyses to support the cause, please sign up here. All analyses will be prepared as R packages, which you’ll be able to build and execute in RStudio, and which will produce aggregate summary statistics (no patient-level data) which can be shared to the study coordinate for synthesis across the participating data partners. If you haven’t set up your local environment yet to run an OHDSI network study, useful resources are available in the Book of OHDSI, and Anna’s OHDSI concept prevalence study study would be a great first study to cut your teeth on.