- From: Joseph Scheuhammer <clown@alum.mit.edu>
- Date: Wed, 11 Aug 2021 10:31:32 -0400
- To: public-prtbl-prsnl-prefs@w3.org
Here are the minutes from the 29 Jul 2021 meeting. We used a google doc for taking minutes. The participants had an opportunity to amend those minutes and correct any errors after the meeting. As a result, the sense of a back-and-forth dialogue was reduced and the amended minutes are closer to notes. Present: Jutta Treviranus, Alison Paprica, Joseph Scheuhammer. Minutes/Notes: Jutta: * several different ways to go about this * there are a confusing number of preference interfaces which all differ in the choices given and the language used to express preferences -- it is the wild west * the better approach is to standardize this and have one platform and one protocol * this includes a better (accessible, easy to understand) user interface for the user to express their preferences * another way: capture the data o what are the types of health data that we want to capture? o what are the abuses? o what are the uses of data related to health? * What is the health data we need to capture? * What are the dangers of sharing that data? Alison: * Complicated but not complex, from a tech perspective, it should be possible to have our information follow us and be respected and acted on by data collecting organizations (e.g., a set of flags on my data file that allow, prohibit, or set conditions around specific data uses) * Practically, aim should be to start with small scale implementation that spreads * The idea of essential vs non-essential data from marketing probably doesn’t apply to health data in the same way a. not binary; almost all data can be useful in some way b. Researchers and health system planners work with whatever they have and could do more if more data were accessible c. Interestingly, the data that people talk most about missing from existing health data repositories is income data not health data (income is a key social determinant of health) * A desired future state could therefore be one in which data users understand that the more sensitive data they request, the fewer people will allow access, hopefully resulting in users tailoring their requests to be more acceptable and within social licence to more people * In terms of categories of health data that people focus on, some ones to consider are a. Admin data - population-wide data obtained through mechanisms like OHIP billing that has a little bit of data about the whole population. By its nature we have the most information about hospital/institutional care and other acute services, and relatively little information about people who are well or who are living while managing chronic conditions like diabetes.. These data are far from perfect (e.g., might limit you to having information about only one condition per visit for a person that has multiple chronic conditions) but have the advantage of covering the entire population and being longitudinal b. Primary care data - i.e., data from your family doctor’s office. There is a push to get more of this information because it will allow us to understand what is happening BEFORE people need hospital or institutional care and develop and plan policies and programs that are pro-active. For example, instead of basic knowledge about someone with diabetes (e.g., whether they are taking government funded diabetes drugs, whether they are participating in wound care programs, whether they had to have an amputation), we can know much more, how outcomes are related to blood sugar levels, height, weight, smoking). There is variation in terms of what kinds of data people would work with and have access too. In some cases, datasets contain basic information about a small number of variables. Other datasets can include detailed doctor’s notes, but then extra processing is required to ensure identifying information isn’t disclosed, e.g., redact the name “Mrs. Parkinson” if it occurs in doctor’s notes, but leave in Parkinson’s as a condition. c. Wearables data - there is a lot of interest in this because (i) it would allow us to have continuous, or at least more frequent data, compared to what is captured when people see their family doctor every few months and (ii) with AI we can do things with this data that we expect will lead to new discoveries. Some private firms are already doing interesting things with wearables data and also a few researchers. Demand for this data is likely to increase and there is a good opportunity to set some guardrails around use now before the data are out all over the place (maybe this is one of the places that you start) 4. Genomics data - I didn’t mention this during the call, but many people think that a combination of genomics data with admin and primary care data will launch the next wave of genomics discovery. Basically the idea is that we’ve already learned a lot about genetic conditions that we can recognize immediately, e.g., hereditary cancers, but there are many other phenotypes that we don’t even know have a genetic origin. Large linked datasets that include genetic information and information about chronic conditions, longevity, physical and mental health, responsiveness to medications can open up the possibility of new discovered and treatment plans, particularly when AI/ML used to analyse. d. Deep clinical data - I didn’t mention this during the call, but the MIMIC database is an example. Essentially it focuses on consented (or waived consent in the case of MIMIC) data about patients that are in environments like the ICU where a ton of data is collected. Similar to wearables and genomics, ML is opening up new possibilities in terms of the ways these data can be analysed. e. Social determinants of health aka health-related data - _https://www.health.gov.on.ca/en/pro/programs/heia/resources.aspx_ <https://www.health.gov.on.ca/en/pro/programs/heia/resources.aspx> has good information about why information about income, gender, language, employment, education, geography etc. are important to understanding and supporting health and well-being e. Other data - just noting here that there are all kinds of data and the list above is not exhaustive. For example, data about prescription medications taken, home care services received, data about healthcare providers and institutions, and the list goes on., * Considerations for establishing and implementing an operability standard a. Do not give the false impression that data are only used when people consent (there are special cases when data are used without consent - see _https://theconversation.com/the-public-needs-to-know-why-health-data-are-used-without-consent-123669_ <https://theconversation.com/the-public-needs-to-know-why-health-data-are-used-without-consent-123669> b. In the health field, the first step before data sharing is removing or coding any identifying information which means that people using the data are studying groups, populations, sub-populations vs. individuals, you probably want to incorporate that practice (or at least have it be one of the options, see e) below) c. Consider the steps you’ll take to develop a standard that aligns with what members of the public are willing to do or specify - this is partly about having different user interfaces, but also meant to get at the fact that early adopters (who are also the people most likely to be involved in any pilot projects) may have more knowledge and more interest in the details that most other people d. Alison personally recommends that the work focuses on unpacking vague “third-party” clause in privacy agreements e. Overall the standard and way forward should aim to establish a virtuous cycle where members of the public can easily set out their preference (perhaps very high level categories to start (e.g., unlimited, unlimited as long as data are de-identified, non-commercial use only, commercial use with notification, etc.), with option to refine generally or for specific studies and the organizations that respect and align with social licence are the ones that flourish because they get more -- ;;;;joseph. 'The only reason for time is so that everything doesn't happen all at once.' - B. Banzai -
Received on Wednesday, 11 August 2021 14:32:06 UTC