I had a ton of feedback after publishing AI Won’t Save Healthcare earlier this week. Thank you for replying! The main counterargument to the article has been: Well, yeah, American healthcare generally sucks, but value-based care works. Value-based care, so far though, is mostly funded by government programs like Medicare and Medicaid, and I don’t think it’s antithetical to what I’m mainly saying: that companies are an inefficient way to do healthcare projects in the United States.
So my question is: If you could spend Google’s healthcare money and attention on anything other than doing this breast cancer study, and that would actually be beneficial sooner rather than later, what would it be? It could either be something that Google already does, or something completely different. It could be something that you want in your life healthcare-wise, or something you think everyone needs.
I have a ton, but my big one right now is creating some sort of program or social that connects new parents with kids of similar ages locally after their babies are born (and didn’t collect any of the data ;). Maybe by partnering with hospitals? The goal is to not have this be online, but in person. Having experienced this twice now, there is a very, very small official support network for new parents in America.
Re-build electronic health records into a system that prioritizes patient health data instead of serving billing. Let people have access to their records, and integrate it with third-party medical devices (including smart phones). No grand plan for how to make it work, but I think this would lay the foundation for a lot of valuable change. For one thing, it sounds like it would go nicely with value-based healthcare. For another, and maybe more importantly, pharma R&D companies are *starved* for good clinical data to help them understand real disease in real patients. If you make (properly anonymized) data available from natural experiments, you'll start getting better research, better long-term outcomes.
Currently in the US, patients' data is protected under HIPAA but most consumers still do not directly own their private health data. My visit to a primary care physician is "protected" and lives either on a physical document or their outdated electronic system. If I go to another provider, I will have to contact the old provider and have them integrate their records with the new provider, presumably with much lost in the way.
If Google could provide an encrypted system with an set of APIs for providers to access health records, perhaps built on blockchain, then I as a patient can truly own my data and have as much control over it as I have over my credit card transactions data or my emails. This would enable providers to have a much more holistic look at a person's health records and provide personalized care, thereby improving outcomes.
It's not sexy, but I'd continue trying to find ways to pay primary care docs for some of the important but frequently uncompensated work like end-of-life planning, monitoring chronic care patients via device data (and not expensive in-person visits), responding to patient portal messages, and paying for case management/care coordination. Or, paradoxically, keep throwing research money at smart programs focused on cutting American healthcare costs. Or maybe just add it to the funding that gives every American a baseline level of healthcare. If they want a private room, they can pay more, including through additional insurance, just like in France.
Related to the newsletter but not the question - Before transitioning to data science, I did a lot of BI/analytics work focused on the transition to value-based care. For example, rather than paying doctors for each visit by a diabetic patient under the fee-for-service (FFS) model - and thus seemingly incentivizing a doc to not keep great care of a diabetic so they appear in the office frequently - Medicare or Medicaid have a number of programs aimed at moving compensation to the quality of care for those patients. In this case, the quality metric would be hemoglobin A1c (HbA1c), the two- or three-month marker for blood glucose levels.
So someone somewhere writes a quality metric specification for a KPI to measure patient HbA1c control for a doc. How we define diabetics? Well, there's probably a standard list of diagnoses out there somewhere. But diagnoses can be documented in about 10 different places (problem list, medical history, encounter, hospital bill, professional (doc) bill, admission, etc.) in a medical record. Where do we look? What about the 30 women with gestational diabetes who were incorrectly given type-2 diabetes codes? The measure excludes patients who are on hospice. That might be documented in three or four areas with structured data, may be one of about 10,000 kinds of physician notes in free-text, or may not be documented anywhere at all. Those important HbA1c scores - well, most lab results are returned with a solid numeric score, but this one group of docs uses a lab that returns a number in free-text with inequality signs, and there's a weird old doc uses some lab that sometimes just returns "normal" or "abnormal." You get the idea.
Basically, you give 20 good analytics developers the specs to a quality measure, I guarantee you're going to get at least 18 output metrics, including some that vary significantly. The amount of time it takes to write good SQL for a measure (and set up a monitoring/quality improvement program around it) and corral the workflows so everyone is following steps (meaning getting everything captured in structured data) is considerable. Frequently, the docs who were failing the measure we're following the gist of it but just not documenting it in structured data. The analytics developers that can do this reliably can now command salaries >= data science. And the whole process just adds to the administrative bloat that is unique to the American healthcare system.
Having worked with a ton of docs doing this work, I am slightly less cynical about their motives Vicki hinted at in parts of her newsletter this week. American docs make a lot more money than they do anywhere else, which many justify by focusing on the cost of medical school here vs. elsewhere and the very low (per hour) pay during residency and fellowship. If you offered them free medical school but European doc pay, I guarantee 90% would stick with the American system. But I've seen 20 instances of a doc doing something for a patient for free or something that's minimally reimbursed for every one time I've seen a selfish move.
Formalizing a metric to understand social determinants of health in specific communities would be a great task for corporations to tackle. For example, using location data whether to understand whether or not an individual lives within a food desert (no access to affordable and/or healthy food), is something that directly impacts health, but is not well studied clinically due to a dearth of data.
Re-build electronic health records into a system that prioritizes patient health data instead of serving billing. Let people have access to their records, and integrate it with third-party medical devices (including smart phones). No grand plan for how to make it work, but I think this would lay the foundation for a lot of valuable change. For one thing, it sounds like it would go nicely with value-based healthcare. For another, and maybe more importantly, pharma R&D companies are *starved* for good clinical data to help them understand real disease in real patients. If you make (properly anonymized) data available from natural experiments, you'll start getting better research, better long-term outcomes.
Currently in the US, patients' data is protected under HIPAA but most consumers still do not directly own their private health data. My visit to a primary care physician is "protected" and lives either on a physical document or their outdated electronic system. If I go to another provider, I will have to contact the old provider and have them integrate their records with the new provider, presumably with much lost in the way.
If Google could provide an encrypted system with an set of APIs for providers to access health records, perhaps built on blockchain, then I as a patient can truly own my data and have as much control over it as I have over my credit card transactions data or my emails. This would enable providers to have a much more holistic look at a person's health records and provide personalized care, thereby improving outcomes.
It's not sexy, but I'd continue trying to find ways to pay primary care docs for some of the important but frequently uncompensated work like end-of-life planning, monitoring chronic care patients via device data (and not expensive in-person visits), responding to patient portal messages, and paying for case management/care coordination. Or, paradoxically, keep throwing research money at smart programs focused on cutting American healthcare costs. Or maybe just add it to the funding that gives every American a baseline level of healthcare. If they want a private room, they can pay more, including through additional insurance, just like in France.
Related to the newsletter but not the question - Before transitioning to data science, I did a lot of BI/analytics work focused on the transition to value-based care. For example, rather than paying doctors for each visit by a diabetic patient under the fee-for-service (FFS) model - and thus seemingly incentivizing a doc to not keep great care of a diabetic so they appear in the office frequently - Medicare or Medicaid have a number of programs aimed at moving compensation to the quality of care for those patients. In this case, the quality metric would be hemoglobin A1c (HbA1c), the two- or three-month marker for blood glucose levels.
So someone somewhere writes a quality metric specification for a KPI to measure patient HbA1c control for a doc. How we define diabetics? Well, there's probably a standard list of diagnoses out there somewhere. But diagnoses can be documented in about 10 different places (problem list, medical history, encounter, hospital bill, professional (doc) bill, admission, etc.) in a medical record. Where do we look? What about the 30 women with gestational diabetes who were incorrectly given type-2 diabetes codes? The measure excludes patients who are on hospice. That might be documented in three or four areas with structured data, may be one of about 10,000 kinds of physician notes in free-text, or may not be documented anywhere at all. Those important HbA1c scores - well, most lab results are returned with a solid numeric score, but this one group of docs uses a lab that returns a number in free-text with inequality signs, and there's a weird old doc uses some lab that sometimes just returns "normal" or "abnormal." You get the idea.
Basically, you give 20 good analytics developers the specs to a quality measure, I guarantee you're going to get at least 18 output metrics, including some that vary significantly. The amount of time it takes to write good SQL for a measure (and set up a monitoring/quality improvement program around it) and corral the workflows so everyone is following steps (meaning getting everything captured in structured data) is considerable. Frequently, the docs who were failing the measure we're following the gist of it but just not documenting it in structured data. The analytics developers that can do this reliably can now command salaries >= data science. And the whole process just adds to the administrative bloat that is unique to the American healthcare system.
Having worked with a ton of docs doing this work, I am slightly less cynical about their motives Vicki hinted at in parts of her newsletter this week. American docs make a lot more money than they do anywhere else, which many justify by focusing on the cost of medical school here vs. elsewhere and the very low (per hour) pay during residency and fellowship. If you offered them free medical school but European doc pay, I guarantee 90% would stick with the American system. But I've seen 20 instances of a doc doing something for a patient for free or something that's minimally reimbursed for every one time I've seen a selfish move.
I think you're describing https://www.nct.org.uk/ in the uk?
Formalizing a metric to understand social determinants of health in specific communities would be a great task for corporations to tackle. For example, using location data whether to understand whether or not an individual lives within a food desert (no access to affordable and/or healthy food), is something that directly impacts health, but is not well studied clinically due to a dearth of data.
Simple stuff: https://twitter.com/soobrosa/status/1214181484791304193 and a lot on communicating it.
Would non-profits like https://parentsasteachers.org/ or https://www.nursefamilypartnership.org/ fit the kind of program you're thinking of?