Can AI turn a hospital discharge summary into a start of care packet?

Short answer: Sam Nadler demos a prototype that turns a CHF discharge summary into a start of care packet with risk factors, a four week schedule and matched staff. Anand Chandrasekaran says it captured the workflows well, but real intake is not real time.

Watch this answer from Episode 24

Watch this answer from Episode 24 (starts at 2:23) · Episode 24 notes and transcript

What was said on Built This Week

From Episode 24: This AI Fills Healthcare Shifts in Minutes, starting at 2:23.

Sam Nadler [2:22]: Amazing. Well, just to jump in, we built something with Arya in mind. And in our call, we discussed potentially building something for the start of care intake. And I'm not from the healthcare space, so this could be a little bit off, but I wanted to create a few different demos. Our first one is a complex CHF. So you would get something like this discharge summary from the hospital. Again, I don't know if this is what it actually looks like. Let's use AI to generate the start of care packet that would help inform from this, I would say, not messy, but maybe unstructured and somewhat coded discharge summary. So as you go through, you can see in this discharge plan, it, has this two w three, then one w four. Maybe for a health care audience, that that's obvious, but it's not so obvious for me. They have certain, home situations, for home care, whether they have, pets or stairs or they live alone. I can imagine all that could be important to understand who's the right person to help with this patient. But anyway, we trigger the start of care packet, and as you can see, it gives, exactly what's happening, congestive heart failure, risk factors. So you have the recent c h CHF exasperation. You have the clinical strategy.

Sam Nadler [3:52]: I thought this little projected four week schedule was cool. You have the diffs the different disciplines, how often they need to be there in the first week. So week one, the s in, I'm not even sure what that stands for, but I guess some sort of wound care nurse would need to be there every day for the first week, five days for the second week, etcetera. And you can see kind of what that schedule immediately matches theoretical staff. So you have Sarah Chin, oh, skilled nurse, that makes sense, who's, has wound care competencies. She speaks Spanish. Wow. Let's go ahead and send Sarah a auto drafted outreach to see if she can join, and this obviously doesn't work, but that's what we would do. It kinda drafts these, and you could theoretically engage with these different medical personnel quickly. Start of care details, you have the address, you have the level of urgency, some home contacts, which I mentioned before.

Sam Nadler [4:54]: They have four steps to enter home, elderly wife, at least someone's available, but maybe she it's will be unable to lift the patients. You got some missing info, gender's missing, maybe, one of the teammates of Arya would follow-up with this, maybe some are more important than others, and then you have some workflow actions that you could just go ahead and click through as you get through this patient's start of care. And then lastly, as just a little fun feature, I created a get well card functionality where if we click it and it's supposed to be punny and funny. Maybe that's not the right, moment to send, but you got that get well soon, Arthur, from Arya Health. Arthur sending you heartfelt wishes for a speedy recovery. We're rooting for you to feel hearty and strong, ready to take those steps back home soon. So this is The start of care packet. Again, I'm not from the health care space. Maybe it's off base. Maybe it's somewhat aligned, but would love to hear your thoughts.

Arun Kalaiselvan [6:02]: Well, I love that you and Medieval are on this, first off. That's fantastic. I actually think this is a pretty good starting point for the start of care workflow. Anand does I haven't seen Anand nod along so much So to new products.

Anand Chandrasekaran [6:19]: It's interesting that you captured some of the workflows really well. But what's most interesting in this is something that we are hoping to do in, the fourth version is already captured in this. Yeah. And I really like that part about it. Like so I mean, it's it's it's difficult to get into the SOC stuff and then be like, you cannot just, like, own the process immediately. The eventual goal is to own the process, right, from the first. Right? But, yeah, this captures that very well. I really like that. There is a lot of things that happens in this that takes we wish it is real time, but it is not real time. That's the only thing I think is off. That's all. I think I really like this kind of a summary that, like, this is cool and I really like it.

Arun Kalaiselvan [7:03]: Yeah. I definitely want to Yeah. Yeah, give shoot us a copy. It got all the titles right. That was cool. I The skilled nursing part, the occupational therapy, the physical therapy, and the home health aide. It got some details pretty spot on.

Anand Chandrasekaran [7:17]: I wouldn't believe if you tell me that you're not from the health care space, I wouldn't believe you from this. If I see this, I would be like, yeah. You got it right. So that's that's nice.

More questions from this episode

Go deeper

New episode every week

Watch us build with AI, live.

Real builds, founder interviews and the AI news that matters. Join 43K+ builders on YouTube.

▶ Subscribe on YouTubeBrowse all episodes →