ACT Next Steps

Use this page to understand more about how to use your ACT query results, limitations of the network, and tips and tricks for improving the accuracy of your queries.


After running an aggregate count search in ACT, what do you do next? This guide describes several different use cases for ACT, limitations of the network, and tips and tricks for improving the accuracy of your searches. A key takeaway should be that ACT is a tool to help you find which sites have patients or datasets that would help with your research, but you will likely need to reach out to collaborators at those sites to recruit those patients for a trial, perform chart reviews to confirm that findings in ACT are correct, or to establish data use agreements to access patient-level records.


Finding collaborators at sites is difficult. Each site has a main ACT contact person, though due to limited resources, they might not be able to respond to your questions. Many sites also have websites with searchable online profiles of their researchers. These are useful tools for locating a potential collaborator who might be interested in working with you. As you reach out to potential collaborators, keep in mind that they will have many questions. They will want to know details about your study design, whether you already have a funded project or if you are looking for help preparing a proposal, whether the project has IRB approval, who the sponsor is, how many patients you need, and more.


-->  Click here for a list of main contact people at each ACT site (coming soon)
-->  Click here for a list of researcher profile websites at each ACT site (coming soon)
The next section of this guide provides a brief summary of each ACT use case, followed by detailed information for each use case.

Summary of ACT Use Cases

Clinical Trial Feasibility:



  • Determine if there are enough patients to conduct a clinical trial
  • Understand population characteristics to ensure that your study criteria are capturing the relevant demographic groups
  • View how changes to your study criteria might impact your trial. Create enrollment projection tables for grant applications.


  • The actual number of patients you will be able to enroll is likely much smaller than the counts returned by ACT
  • Certain enrollment criteria might be difficult to translate into an ACT query

Clinical Trial Recruitment:



  • Identify which sites have patients that match your trial’s inclusion and exclusion criteria
  • Prioritize sites based on number of matching patients, geographic location, or population characteristics


  • You will need to find collaborators at each site to help you recruit patients
  • Sites vary in policies and workflows for recruiting patients

Using ACT for Research (this capability coming soon)



  • Generate hypotheses using real-world data from 150M+ patient records
  • Discover interesting patterns and trends in the data
  • Identify sites where further exploration of their data might yield important clinical findings


  • Real-world data are very messy
  • Understanding differences between sites’ coding practices, data quality and data completeness is essential
  • Confirming findings by analyzing patient-level data is important before publishing results

COVID-19 Cohort Discovery



  • ACT has placed special emphasis in keeping data on patients tested for COVID-19 as up to date as possible
  • The ACT concept list has the latest codes for COVID related laboratory tests, diagnoses, and medications, as well as search terms for clinical course and COVID disease severity


  • Sites vary in how often they can update their data and when they adopt new codes
  • Because the clinical course of COVID can last several weeks, data “censoring,” where there has not yet been enough time to see patients’ outcomes, affects a large portion of the population

Details: Clinical Trial Feasibility

Aggregate count searches in ACT can provide important data to help you design a clinical trial. With 150M+ patient records, representing almost half of the U.S. population, ACT can give you an upper bound on the number of subjects that you might be able to recruit for a trial. Note, though, there are many reasons why the actual number you can expect to recruit might be much lower:

  • The counts are based on patients who received care at ACT sites. These are not counts of people who expressed interest in participating in clinical trials.
  • The counts might include patients who have not received care at that site in many years. As a result, the site might not have accurate or up to date contact information for the patients.
  • Because patients receive care from multiple providers, the same patient might be counted by more than one site.
  • Trials often contain inclusion or exclusion criteria that are not captured by ACT, such as physical exam findings, family history, or participation in other clinical studies. Through chart review or other forms of screening, many patients counted by ACT will eventually be determined to be ineligible for the trial.

Despite these limitations, there are many ways ACT can help you increase the chances that your trial will be successful:

  • Try modifying your inclusion or exclusion criteria. This can sometimes result in large changes in the number of matching patients.
  • Try searching for each of your inclusion/exclusion criteria one at a time. This will help you narrow down which ones are reducing the counts the most. It can also help you identify problems with your search. For example, sometimes different sites use different codes for the same concept. If some sites have many patients matching the code, while others report no patients, then consider additional codes that sites might be using.
  • Add variables, such as age, sex, and race, to your search to understand the demographic breakdowns of the cohort. This can help you ensure that you search criteria are capturing a diverse population.

Details: Clinical Trial Recruitment

The ACT Network tells you which sites have patients that match your search criteria. However, it cannot identify individual patients or help you contact the patients. To do that, you will need to find a local collaborator at each site where you want to recruit patients. Below are some important things to consider:

  • In some cases, collaborators will be able to work with the site’s IT or informatics group to locate your ACT search and identify the matching patients. However, at some sites, the collaborator will need to recreate your search in the site’s own local tools. Some sites limit access to these tools to qualified faculty, which might exclude postdoctoral researchers, students, and other groups.
  • Sites vary in policies regarding patient recruitment. Instead of directly contact patients, some sites require researchers to contact patients’ physicians, who in turn may reach out to their patients.
  • Finding collaborators at sites is difficult. Many sites have websites with searchable online profiles of their researchers. We provide links to these resources here (coming soon). These are useful tools for locating a potential collaborator who might be interested in working with you.
  • As you reach out to potential collaborators, keep in mind that they will have many questions. They will want to know details about your study design, whether you already have a funded project or if you are looking for help preparing a proposal, whether the project has IRB approval, who the sponsor is, how many patients you need, and more.

Details: Using ACT for Research

More information and guidelines for this capability coming soon.

Details: Using ACT for COVID Cohort Discovery

There are some special considerations when studying COVID:

  • Finding COVID patients. The top-level concept folder, “ACT COVID-19”, groups together the most common search concepts related to COVID. Inside this folder is “Diagnostic Lab Tests”, which contains the ACT search concept “ANY Nucleic Acid Lab Test Positive”. This captures any COVID positive test result, regardless of the type of test and specific coding used by sites. This is the simplest way to capture COVID patients. Similarly, “ANY Antibody Lab Test Positive” matches patients with any kind of positive COVID antibody test. Note that it is possible that some COVID patients do not have a positive test result, and instead might have a diagnosis (e.g., ICD-10 code U07.1) or other code that suggests they have COVID. These codes can be found in the “ACT COVID-19 > Diagnosis” folder.
  • False COVID test results. Keep in mind that COVID tests have false positives and false negatives, and that patients might be tested multiple times, with both positive and negative test results. Just because a patient has a negative result, it does not mean they never had COVID. Even if the tests were accurate, they could have tested negative at one point and then later became infected. Similarly, once COVID positive patients get better, they could subsequently have a negative test result. A consequence of this is that the counts of the positive and negative patients at a site will likely add up to a number greater than the number of patients who were tested, since the same patient can be counted in both the positive and negative results.
  • Frequency of COVID patient data updates. Make sure you understand when sites last updated their databases. ACT sites generally started seeing patients with COVID-19 around March, 2020. (1) Sites that last updated their data before March will have no COVID patients. (2) Sites that last updated their data in May will have data on “first wave” patients but not more recent COVID patients. At sites in states where rates are decreasing, there will be proportionally fewer missing patients than in states where rates are increasing. (3) Sites that last updated their data one week ago will have patients who were recently tested positive for COVID, but because the typical clinical course of ACT is 2-4 weeks, their outcomes are not yet known (“censored”).
  • Frequency of non-COVID patient data updates. Many sites are updating data on their COVID patients more often than their non-COVID patients. This is important to remember when searching for non-COVID control groups to compare to the COVID patients. For example, non-COVID controls might not have any recent data. This can make them appear to have fewer co-morbidities, be taking fewer medications, etc., than the COVID patients who have recent data.
  • Patients avoiding hospitals during the pandemic. Patients, in general, have been avoiding hospitals since the pandemic started. As a result, since March, 2020, patients appear to be “healthier” than usual. Control groups consisting of non-COVID patients will seem relatively healthy, even if their data are complete and up to date. To control for this, when comparing COVID patients to non-COVID patients (e.g., for co-morbidity differences), you can limit concepts in your search to before 3/1/2020.
  • Data censoring. To avoid data censoring, you might want to limit your search to patients who tested positive for COVID at least a month before the date a site last updated their data. A simple method is to run one search that limits the COVID positive test to an early cutoff date, such as test results before 5/1/2020. This will work for sites that have updated their data since 6/1/2020. However, it doesn’t solve the problem for sites that haven’t updated their data since May 2020 or earlier; and, it excludes many patients at sites that update their data very frequently. Another approach is to run multiple searches with different date cutoffs; and, for each site use the count that corresponds to the one search that is most appropriate given their last update date.
  • COVID codes changed over time. CDC coding guidelines from 2/20/2020 suggested coding COVID patients with ICD-10 diagnosis B97.29 (other coronavirus) in combination with other codes, including J12.89 (pneumonia), J20.8 (acute bronchitis), J22 (lower respiratory infection), or J80 (acute respiratory distress syndrome). However, on 3/24/2020, the new CDC guidelines switched to the newly introduced ICD-10 emergency codes U07.1 (“COVID-19, virus identified”) and U07.2 (“COVID-19, virus not identified”). The first few LOINC codes for COVID laboratory tests were introduced on 1/31/2020, but dozens of LOINC codes have been added since then. Keep this in mind, since searching for only the newer (or older) codes will miss the older (or newer) patients. Also, it can take a while for hospitals to switch to new codes, patients with a positive COVID test result might not also be given a COVID diagnosis code, and many hospitals use custom internal COVID test codes instead of LOINC.
  • Determining when COVID patients became sick. Typical COVID patients will begin having symptoms, which are likely not captured in their electronic health records. After a few days they are tested for COVID. This is the specimen date. However, the result might not be available until 1 or 2 days later. Early in the pandemic, this delay was often several days. ACT sites vary in whether dates associated with laboratory tests are the specimen date or the result date. Also, hospitals might receive COVID patients from another site and not have the original date when they were tested. Because of these different reasons, it is hard to pinpoint when a COVID patient was infected. If, for example, you search for patients who had a COVID positive test on 4/10/2020, it will match people in various stages of disease. One approach is to search, for example, patients who had an inpatient admission starting on 4/10/2020, with a COVID positive test result anytime in the surrounding week (i.e., between 4/3/2020 and 4/17/2020). Because admission date has a clearer meaning than COVID test date, this might be a better way of selecting a cohort that is better aligned on clinical course.
  • Determining when COVID patients died. Sites vary in how death is coded. You can search for patients that sites know have died using the “ACT Demographics > Vital Status > Known Deceased” concept. This might match inpatient deaths, but sites will often not know if patients died elsewhere. Some sites assign a date to this concept corresponding to when the patient died. However, this is highly variable. Only the year the patient died might be known, or the death date could correspond to when the site learned that the patient was deceased. For COVID patients who are known deceased, one way the date of death can be estimated is by searching for when they no longer have any data. For example, search for known deceased COVID patients with any diagnosis (the entire “Diagnoses ICD-10-CM” folder) starting 5/1/2020. Repeat the search, but starting a month later on 6/1/2020, which will likely result in a smaller count. The difference between the counts might suggest how many patients died that month.
  • Determining COVID severity. Direct markers of COVID severity can be problematic. Sites might not know if patients died at home or at another hospital. Procedure codes for critical care or ventilation are often not available until the patient is discharged. Because many hospitals have repurposed floors as temporary ICUs, it is hard for sites to identify ICU patients based on their location in the hospital. Another approach is to use “indirect” indicators of disease severity. For example, a laboratory test order for arterial blood gases, regardless of the result, might indicate that the patient is in the ICU. An organized list of direct and indirect markers of COVID severity can be found in “ACT COVID-19 > Course of Illness > Illness Severity”.