Nidan,
“Here's the question. which method provides the best patient safety and the most effective methods for trial information gathering robustness? Which presents the lowest ultimate risk to the regulatory body decision making process? (400 ptsx24/7x# of tests(n=20)) or old school (200 ptsx # tests x 24 data points for the years test) .
Now, which method-protocol presents the most trial data and consequently provides the most robust data set for lower risk decisions? Continuous real time AI monitoring or periodic manually assessed data?
That, IMO, Is what is on the table now and BTW the changes will allow full/better trials in a fraction of the time.
Is a system for trails capable of monitoring 400 patients continuously better from a decision risk standpoint than current trail methods?”
Interesting point you raise and the discussion will raise many more questions than answers, I’m sure.
Going forward over time, 24/7 will enrich the data. In one or three trials, with little historic data in humans using our MoA, not a game changer unless miraculous (miracles usually don’t need 24/7 data, usually appear to the naked eye).
This could be the longest, most in depth answer, so, in order to keep it manageable, let’s try to look at it from the perspective of information vs intelligence. (The term: Artificial Intelligence, intelligence means information which is useful. Just as in the military, you don’t necessarily want the middle names of every soldier you are fighting against - that won’t help you defeat their overall strategy to any large degree.)
Data is great. Huge fan here. Must be known how to derive what are the most useful parts of that data.
Your intuition that “wiring” our trial patients for data collection is spot on as an ongoing base for the next(!) set of trials and collecting data in such a manner will help going forward. If that is what you are asking - the answer is, we will gather a lot of information about THESE patients. Might not be sufficient but will happen and can be used in later trials of the same indications but could differ due to the individual make up of patients. See examples below.
I sincerely doubt that the FDA could have collected 24/7 info on test subjects to reveal in real time that Round Up (weed killer) will put you at higher degree of risk for cancer/certain maladies 20 years after continued use. Or, 20 years after constant temporary use - say, one summer? (Maybe some day, if we can see potential cancer risk biomarkers 20 years ahead at the molecular level, then, yes.) Real time data would not have given us that. This is why so many people in the general population need to use a drug/med/device for years before we see the lawsuit commercials. The correlations take a long time to tease out over population groups and narrow the connection. (Disclaimer: unless the drug is killing trial participants or causing alarming risk right before your eyes during trialing, in which case...stop.)
Think of a couple of ways which machine learning (this is AI new terminology, btw) can and would benefit the most:
1. Data gathering
If our participants wear the monitor devices, we have access to a ton of data as you suggest. Some will be meaningful, some will be noise. We can sometimes do things with the noise but it can be so isolated as to be useless. Analogies: when family comes to visit, most patients might register more favorable cognition and mood data points in their ERP’s. That may have no correlation to the drug whatsoever. When specific individuals visit, data can become more favorable - this can scale to each individual. If Aunt Linda came to see you, your data went up. If she visits Susie, next door to you, Susie’s data does not go up, or, it may go down. That’s not useful, but definitely can be collected. The same can be said for “bath day” or drinking ginger ale - some folks will improve slightly briefly, some will decline, some will remain unchanged. This data would all be gathered, then have to be sifted through for meaningful correlations, and, as you can see, much of it will not correlate to the drug in isolation. Perhaps, visits from Aunt Linda also improve the placebo group and general population as well - people like visits. Others, less gregarious, will not enjoy these days...having data does not mean we will KNOW more. (Although, we will surely not know less.)
2. Predictive value
The best use researchers can make of this data is historical baseline for future predictions, not necessarily real time. Go figure. If I want to beat Tom Brady and the Patriots, the best use of my time may be studying OLD FOOTAGE of him playing various offensives and defensives, including mine and several similar to mine. Trying to beat TB without some historical data to use as predictive is a tough mark to hit. Real time play against him may allow you to adjust your game in the second half, however, in answer to your question, so might good old fashioned coaching!
If you gather enough data about the patriots (and this is team specific, might not hold true for all QB’s...think about that - as patient specific, but not for all AD sufferers enough for FDA to make a better decision in real time about a single cohort), with enough Patriot data, a machine may do analytics and correlate that, “With 90% certainty, when the team is inside the 30 following a long completion, if Brady hands off on first down for no gain, the chances of that drive ending in a TD go up! If this happens most of the time, a machine can analyze more of these correlations than a team of coaching staff and players in very short time. Sports analytics is now a field. The computer tells you what your percentages are to best confound TB’s chances in such a situation and many many many others. (We’ll leave the football’s PSI constant for this example)
Most people/fans/players/coaches may intuit that stopping a rush on first down is a good thing - bias! The computer will demonstrate that you will watch the man put up points. Counterintuitive but bettors are relying more on the hidden correlations “edges” which computers can determine have a higher percentage of likelihood.
We can and should add to a database to gather these for their predictive value.
3. Some conditions do not lend themselves to an advantage by “looking under the lid” constantly or leaving the lid off
Tumors grow and spread. In some cases, checking cancer constantly yields nothing of greater value than checking cancer progress (growth) every 2 weeks. That’s not to say that you can’t check every day, you just may not see anything because cancers can go through periods of activity and dormancy. Watching every day is akin to knowing the middle names of your enemies - better to know the overall battle strategy by observing troop movement when troops move, which isn’t usually every second. There is downtime built into nature - best way to use the downtime is to study what you should do in the event that the troops (condition) move N, S, E, or W. Then, when movement occurs, you are prepared with your best Go strategy.
Speaking of “noise”, our best intel told us we could take Iwo Jima in less than a week. That island was undermined with tunnels and had lookouts and snipers in cockpits of crashed planes...took a month and a half and fearsome more casualties. My greatest problem with “too much information” is that if it is not well established and mined for accurate “edges” - correlations, one goes down many rabbit holes leading to nothing productive. (Early on, we will get more “junk” and dead ends than nice, testable leads.)
In real time, some symptoms could potentially be alleviated. But, that can happen with old style trials and observant caregivers as well. Additionally, there is the chance that the data demonstrates that patients improve significantly right out of the gate. Great! But, we wouldn’t need the monitors to show that. If our trial participants all began reciting pi out to 100 digits after 3 days, everyone around them would know.
Anything less obvious than amazing benefit or epic unfavorable events, will probably have to be wait and see, study the trend, and weigh at the end as in trials of old. Going forward with all of the new data gained would be our hope for allowing a streamlined future. Would love to say this is enough right now but can’t justify that one or three trials will merit the change. - could not beat Patriots after watching one game. Seems improbable. Black Jack card counters can only shift the odds in their favor after playing the same strategy for many(!) hands, sometimes over weeks you will definitely come out ahead, but that’s with constant adherence to the system. And, then they ban you anyway. But, you get to keep your winnings along with all the point assigning knowledge you now have which is of no use going forward but took many college summers to learn! Might be better off just getting an old fashioned j-o-b ?
The old saying: it’s not what you don’t know that gets you, it’s what you know for certain that just isn’t so.
The monitors will help going forward to shift the odds in our favor over time. That’s the best we can expect at this time.
Please keep your sharp “edge”,
Bio
News 
Market Data 
Discover 
